problem solving, decision

making, and professional

judgment
This page intentionally left blank
problem solving, decision
making, and professional
judgment
a guide for lawyers and
policy makers

paul brest
linda hamilton krieger

1
1
Oxford University Press, Inc., publishes works that further Oxford University’s objective
of excellence in research, scholarship, and education.

Oxford New York

Auckland Cape Town Dar es Salaam Hong Kong Karachi Kuala Lumpur Madrid
Melbourne Mexico City Nairobi New Delhi Shanghai Taipei Toronto

With offices in
Argentina Austria Brazil Chile Czech Republic France Greece Guatemala Hungary
Italy Japan Poland Portugal Singapore South Korea Switzerland Thailand
Turkey Ukraine Vietnam

Copyright © 2010 Oxford University Press, Inc.

Published by Oxford University Press, Inc.
198 Madison Avenue, New York, New York 10016

Oxford is a registered trademark of Oxford University Press

Oxford University Press is a registered trademark of Oxford University Press, Inc.

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system,
or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording,
or otherwise, without the prior permission of Oxford University Press, Inc.

_____________________________________________

Library of Congress Cataloging-in-Publication Data

Brest, Paul
Problem solving, decision making, and professional judgment : a guide for lawyers and
policymakers / Paul Brest, Linda Hamilton Krieger. — 1st ed.
p. cm.
Includes bibliographical references and index.
ISBN 978–0–19–536632–7 (pbk. : alk. paper)
1. Practice of law—Decision making. I. Krieger, Linda Hamilton, 1954– II. Title.
K346.B74 2010
320.6—dc22
2009052053
_____________________________________________

1 2 3 4 5 6 7 8 9

Printed in the United States of America on acid-free paper

Note to Readers
This publication is designed to provide accurate and authoritative information in regard to the subject
matter covered. It is based upon sources believed to be accurate and reliable and is intended to be current
as of the time it was written. It is sold with the understanding that the publisher is not engaged in rendering
legal, accounting, or other professional services. If legal advice or other expert assistance is required, the
services of a competent professional person should be sought. Also, to confirm that the information has
not been affected or changed by recent developments, traditional legal research techniques should be used,
including checking primary sources where appropriate.

(Based on the Declaration of Principles jointly adopted by a Committee of the

American Bar Association and a Committee of Publishers and Associations.)

You may order this or any other Oxford University Press publication by
visiting the Oxford University Press website at www.oup.com
For Lee Ross, for his wisdom and generosity, his enormous contribution to social
psychology and the study of judgment and decision making, and his leadership in
demonstrating how insights from the empirical social sciences can be applied to
improve, enrich, and ennoble our societies and ourselves.
This page intentionally left blank
contents

Acknowledgments xxvii
Preface xxix

part 1 introduction to problem solving and decision making

chapter 1. Problem-Solving and Decision-Making Processes:
Deliberation, Intuition, and Expertise 3
1.1 A Day in the Life of a Problem-Solving Lawyer 4
1.2 A Day in the Life of a Problem-Solving Policy Maker 7
1.3 Defining Problem Solving and Decision Making 8
1.4 Deliberative and Intuitive Problem Solving and Decision
Making 11
1.4.1 Deliberative Processes 11
1.4.2 Divergent and Convergent Thinking 13
1.5 Intuitive Processes in Problem Solving and Decision Making 14
1.5.1 Recognition-Primed Decision Making 16
1.5.2 The Role of Schemas and Scripts 17
1.5.3 The Role of Affect 19
1.6 The Interaction of Intuition and Deliberation 21
1.6.1 The Two-Systems Model of Cognition 21
1.6.2 The Limits of Deliberation: Bounded Rationality 22
1.6.3 The Necessity and Limits of Intuition 25
1.7 Professional Problem Solving and the Nature of Expertise 25
1.8 Lawyers and Policy Makers as Experts 29
1.8.1 Lawyers’ Knowledge about the Law, Legal Institutions, and
Actors 30
1.8.2 Policy Makers’ Knowledge about Legislative and
Administrative Institutions and Processes 30
1.8.3 Expertise in Other Substantive Domains 31
1.8.4 Problem-Solving Expertise 31
chapter 2. Framing Problems, Identifying Objectives, and Identifying
Problem Causes 33
2.1 The Decision Context and Problem Frames 33
2.2 Problem Framing 34
2.3 Problem-Framing Pitfalls 36
viii contents

2.3.1 Framing by Solution 37

2.3.2 Symptoms vs. Problems 38
2.3.3 Automatic Problem Framing and the Pitfalls of Expert
Problem Frames 40
2.4 Identifying Interests and Objectives 42
2.4.1 Multidimensional Frames and the Process of Framing-by-
Objectives 42
2.4.2 Underspecification of Objectives: Causes, Consequences,
and Corrective Strategies 43
2.5 Identifying Problem Causes 45
2.5.1 Define the Problem 48
2.5.2 Specify the Problem’s “What,” “Where,” “Extent,” and
“When” 48
2.5.3 Spot Distinctions and Identify Changes Responsible for
the Distinction 50
2.5.4 Identify Hypotheses for Further Testing 51
2.5.5 Attributing Causation 51
2.5.6 Problem 52
2.6 Moving the World in the Desired Direction Through Strategic
Planning 53
chapter 3. Generating Alternatives: Creativity in Legal and Policy
Problem Solving 61
3.1 “Pickers” and “Choosers”: Problem-Solving Frames of Mind 61
3.2 Frameworks for Generating Solutions to Present Problems 63
3.2.1 Basic Principles 63
3.2.2 Different Kinds of Alternatives 63
3.2.3 Buying Information and Time 64
3.2.4 Divergent Thinking and the Generation of Alternatives 66
3.2.5 On Possibilities and Path Constraints 67
3.3 Creativity in Professional Problem Solving 67
3.3.1 An Opening Scenario 68
3.3.2 Creativity: Working Definitions 69
3.3.3 Creativity and Cross-Fertilization 70
3.3.4 Creativity, Expertise, and Critical Thinking 73
3.3.5 The Movement Between Convergent and Divergent
Thinking 74
3.3.6 A Confluence Approach to Creativity—And the Skills It
Requires 76
3.3.7 The Individual, the Domain, and the Field 77
3.4 Problems in the Future Tense: Scenario Planning 80
 contents ix

3.4.1 What Are Scenarios? How Are They Used? 81

3.4.2 Why Use Scenarios? 84
3.4.3 Key Principles of Scenario Planning 84
3.4.4 Drawbacks of Scenario Planning 85
3.5 Barriers to Creative Problem Solving—And Overcoming
Them 86
3.5.1 Cognitive Process Constraints 86
3.5.2 Social Constraints 87
3.5.3 Practical Constraints 88
3.5.4 Personal Constraints 89
3.6 Summary: Generating Better Alternatives 89
chapter 4. Choosing Among Alternatives 91
4.1 Specifying the Consequences of Alternatives 91
4.2 The Spectrum of Decision-Making Processes 93
4.3 Deliberative, Noncompensatory Decision-Making Strategies 96
4.4 Compensatory Procedures 98
4.4.1 Reason-Based Decision Strategies 99
4.4.2 Quantitative Decision Strategies 102
4.5 A Subjective Linear Model: Siting a Wastewater Treatment
Plant 102
4.5.1 Identify the Major Attributes Affecting the Decision 105
4.5.2 Identify the Alternatives or Options Facing the Decision
Maker 106
4.5.3 Evaluate (Objectively or Subjectively, as the Case May Be)
the Attributes for Each Alternative 107
4.5.4 Translate the Evaluations into Normalized Values 108
4.5.5 Aggregate the Evaluations for Each Alternative and
Compare the Alternatives 109
4.5.6 Perform a Reality Check 109
4.5.7 Weight the Attributes: A Possible Additional Step 109
4.5.8 Decision Opportunities 111
4.6 Summary: The Features of Different Decision-Making
Strategies 112

p art 2 making sense of an uncertain world

The Challenges of Detecting Correlation and Imputing
Causation 115
The Lens Model 117
The Roles of Intuition and Analysis in Empirical Judgment 119
The Role of Statistics in Law, Policy, and Citizenship 121
x contents

chapter 5. Introduction to Statistics and Probability 123

5.1 Probability 123
5.1.1 Random Variables 123
5.1.2 The Foundations of Probability 124
5.2 Probability Distributions 125
5.3 Population Distribution vs. Empirical Distribution: The Aim of
Statistics 126
5.3.1 The Aim of Statistics 127
5.4 The Population and Empirical Distribution at Terra Nueva 128
5.4.1 Contingency Tables 128
5.4.2 The (Known) Empirical Distribution 130
5.5 Estimating Probabilities 132
5.5.1 Sampling 132
5.6 Conditional Probability and Independence 133
5.6.1 Independence 134
5.7 Randomness, Independence, and Intuition 136
5.7.1 Illusory Correlation 136
5.7.2 The Illusion of Control 137
5.7.3 The Gambler’s Fallacy 139
5.8 Testing Hypotheses—and the Concept of Statistical
Significance 141
5.8.1 What Null Hypothesis Statistical Testing Is and Is
Not 146
5.8.2 Intuitions about Hypothesis Testing 147
5.9 Cancer Clusters, the Texas Sharpshooter, and Multiple
Testing 149
5.10 Glossary 152
chapter 6. Scores, Dollars, and Other Quantitative Variables 153
6.1 Continuous Variables 153
6.1.1 Central Tendency 153
6.1.2 Dispersion, Variance, and Standard Deviation 154
6.1.3 Normal Distributions 156
6.1.4 Estimating the Population Mean 158
6.1.5 Intuitions about Sample Size and Variability 161
6.1.6 Comparing Two Groups: Testing for a Relationship
Between One Quantitative Variable (salary) and One
Categorical Variable (sex) 162
6.2 Regression 165
6.2.1 Line of Best Fit 166
6.2.2 Correlation 169
6.2.3 Regression to the Mean 170
 contents xi

6.2.4 Intuitions about Regression 172

6.2.5 Determining How Much of the Variance in the Outcome
Can Be Explained by the Predictor 175
6.2.6 Statistical Significance and Hypothesis Testing 178
6.2.7 Multiple Regression 179
6.3 Glossary 182
6.4. Appendix 184
chapter 7. Interpreting Statistical Results and Evaluating Policy
Interventions 185
7.1 Measuring Impact: Correlation vs. Causation 185
7.2 Establishing Causation 188
7.2.1 Confounding Effects 188
7.2.2 Design Effects 189
7.3 Reducing Unwanted Effects Through Controls 191
7.3.1 Randomized Controlled Studies 191
7.3.2 Constructed Controls 192
7.3.3 Reflexive Controls 193
7.3.4 Statistical Controls 194
7.3.5 The Advantages and Drawbacks of Different Methods 194
7.4 Some Further Examples of Program Evaluation 197
7.4.1 The Connecticut Crackdown on Speeding: A Legal Policy
Case Study 201
7.5 The Significance of Statistical Significance 205
7.5.1 Effect Size 205
7.5.2 Meta-Analysis 205
7.6 Experimental Design and External Validity 206
7.6.1 Within-Subject vs. Between-Subjects Experiments 207
7.6.2 External Validity: Generalizability 208
chapter 8. Explaining and Predicting One-Time Events 211
8.1 Calculating Conditional Probabilities (A Review) 211
8.2 The Probability of Conjunctive Events 213
8.2.1 The Probability of Dependent Conjunctive Events 214
8.2.2 Conjunctive Independent Events 215
8.2.3 Intuitions about Conjunctions 216
8.3 The Conjunction Fallacy and the Representativeness Heuristic 217
8.3.1 Explanation Through Narrative 219
8.4 The Probability of Disjunctive Events 222
8.4.1 Intuitions about Disjunctive Events 225
8.5 Bayesian Statistics 225
8.5.1 Structuring the Taxi Problem as a Tree 227
xii contents

8.5.2 Solving the Taxi Problem Using Bayes’ Theorem 228

8.5.3 Diagnosticity 229
8.5.4 Bayes’ Theorem and Intuition: Ignoring the Base
Rate 233
8.6 Confusion of the Inverse, or the Prosecutor’s Fallacy 235
8.6.1 Countering Faulty Intuition by Presenting
Frequencies 237
8.7 Appendix 239
8.7.1 Odds Version of Bayes’ Theorem Applied to the Taxi
Problem 239
chapter 9. Biases in Perception and Memory 241
9.1 Introduction: Stages of Information Processing 241
9.2 Biases in Acquisition, Retention, and Retrieval 242
9.3 Biases in the Acquisition of Information 243
9.4 The Biasing Effect of Schemas, Expectations, Theories,
and Interests 244
9.4.1 They Saw a Game 245
9.4.2 The Hostile Media Phenomenon 246
9.4.3 Why Cases Don’t Settle When They Should 248
9.5 Distortions in Retention and Retrieval 249
9.6 The Availability Heuristic—And a Detour into Metacognition 252
9.6.1 How the Availability Heuristic Works;
Metacognition 254
9.6.2 Other Phenomena Resulting from Evaluation of
Metacognitive Experiences 255
9.6.3 Availability, Vividness, and Inference 257
9.6.4 Vividness and the Media 258
9.7 Egocentrism: A Particular Memory Bias in Acquisition and/
or Retrieval 261
9.7.1 Debiasing Egocentrism 262
9.7.2 The Recollection of One’s Own Past Beliefs, Feelings,
and Events 263
9.8 Conclusion: Naïve Realism 264
chapter 10. Biases in Processing and Judging Information 267
10.1 Anchoring, Adjustment, and Activation 267
10.1.1 Anchoring in Negotiations 268
10.1.2 Extreme and Untrustworthy Anchors 270
10.1.3 Cross-Modal Anchors 270
10.1.4 Explanations for the Phenomenon 270
10.1.5 Debiasing 272
 contents xiii

10.2 Hindsight Bias 272

10.2.1 Risk/Benefit Analysis in Hindsight 275
10.2.2 Debiasing Hindsight Bias 276
10.2.3 The Legal System’s Responses to
Hindsight Bias 276
10.3 Confirmation Bias and Other Resistance to Modifying
Prior Expectations and Theories 277
10.3.1 Motives Underlying Confirmation Bias 280
10.3.2 Motivated Skepticism 280
10.3.3 Biased Assimilation 282
10.3.4 Confirmation Bias and Positive Hypothesis
Testing 283
10.3.5 Confirmation Bias as an Instance of Anchoring 286
10.3.6 Belief Perseverance after Evidential Discrediting 287
10.3.7 The Emergence of Coherence in Making
Judgments 288
10.3.8 Debiasing Strategies 289
10.4 Overconfidence 289
10.5 Intuition, Analysis, and the Role of Experts Redux 293
10.5.1 Clinical vs. Statistical Prediction 293
10.5.2 The Role of Experts 298
10.6 Conclusion 301
chapter 11. The Social Perceiver: Processes and Problems in
Social Cognition 303
11.1 Introduction 303
11.2 The Biasing Effects of Social Schemas: An Overview 304
11.3 The Cognitive Mechanics of Social Schemas 305
11.3.1 Stereotypes and Stereotyping 305
11.3.2 This Is Your Brain on Stereotypes: The Mental
Representation of Social Knowledge 308
11.4 The Effects of Social Schemas on Social Perception and
Judgment 317
11.4.1 The Effects of Social Schemas on the Characterization
of Ambiguous Behavior 317
11.4.2 The Effects of Social Schemas on Memory 321
11.4.3 The Effects of Stereotypes on Causal Attribution and
Behavioral Prediction 323
11.5 On the Self-Fulfilling Nature of Social Stereotypes 324
11.5.1 The Behavioral Confirmation Effect 325
11.5.2 Stereotype Threat 329
xiv contents

11.6 Beyond Stereotypes: Dispositionism and Situationism in

Social Attribution 331
11.7 Debiasing Social Judgment 334
11.7.1 Dual Process Models of Social Cognition: High Road
vs. Low Road Thinking 334
11.7.2 How Malleable Are Implicit Stereotypes? 338

part 3 making decisions

chapter 12. Choices, Consequences, and Trade-offs 343
12.1 Subjective Utility 343
12.1.1 Sources of Utility 344
12.1.2 The (Dis)utility of the Decision Process—And “Leaky”
Decisions 349
12.1.3 The Effects of Looking Forward and Backward 354
12.1.4 How Much Better Is More? 360
12.1.5 Hedonic Adaptation 364
12.2 Criteria For Rational Decision Making: The Focus on
Consequences and Trade-offs 366
12.2.1 Two Concepts of Rational Decision Making 368
12.2.2 The Axioms of Subjective Expected Utility Theory 369
12.3 From Expected Utility to Expected Return: Cost-Benefit
Analysis 373
12.3.1 An Example: A CBA Analysis of a Workforce
Development Program 375
12.3.2 Cost-Effective Analysis (CEA) 376
12.3.3 Quantifying Reduced Risks in Mortality 376
12.3.4 Difficult-to-Measure Benefits 381
12.3.5 Ancillary Costs and Benefits 382
12.3.6 The Value and Limits of Cost-Benefit Analysis 383
chapter 13. Complexities of Decision Making: Decision Processes;
Relationships to our Future Selves 385
13.1 How Would One Know If a Person Made a Suboptimal
Decision? 385
13.2 The Choice of Decision-Making Strategies 386
13.2.1 Overview of the Decision-Making Process: Decision-
Making Bricolage and the Ad Hoc Construction of
Preferences 386
13.3 Value-Based and Reason-Based Decision-Making
Strategies 388
 contents xv

13.3.1 Value-Based Decision-Making Strategies: Difficulties in

Evaluating Certain Aspects of the Alternatives Being
Considered 388
13.3.2 Reason-Based Decision-Making Strategies 390
13.4 Predicting Our Future Well-Being 395
13.4.1 Impact Bias: Mispredicting Emotional Reactions to
Future Events 397
13.4.2 Projection Bias: Mispredicting Future Preferences 401
13.4.3 The Consequences of the Impact and Projection Biases
and the Possibility of Becoming More Accurate 403
13.5 Positive Illusions: I’m Above Average, and My Future Looks
Rosy 404
13.5.1 Overoptimism 405
13.5.2 The Above-Average Effect 408
13.5.3 Are Positive Illusions Always Detrimental? 408
13.6 Intertemporal Choice 410
13.6.1 Discount Rates 410
13.6.2 Discounted Utility, Dynamically Inconsistent
Preferences, and the Pervasive Devaluation of the
Future 411
chapter 14. Complexities of Decision Making Continued:
The Power of Frames 419
14.1 Prospect Theory, The Endowment Effect, and Status Quo
Bias 419
14.2 Regret Aversion 428
14.2.1 Imagining the Counterfactual Outcome 429
14.2.2 Action vs. Inaction 430
14.2.3 The Effects of Anticipating Feedback about What
Would Actually Have Happened 433
14.2.4 Regret and Suboptimal Learning Processes 434
14.3 Sunk Costs 434
14.4 Mental Accounting 437
14.4.1 Decoupling Decisions and Consequences: The Value of
Feedback 439
14.4.2 Budgeting 439
14.4.3 Self-Control, Gift-Giving, and Income Accounting 440
14.4.4 Budgeting Time 441
14.4.5 Conclusion: So What? 441
14.5 The Downsides of Choices or Options 442
14.5.1 Choice Overload 442
xvi contents

14.5.2 Option Devaluation 443

14.5.3 Context Dependence 450
14.6 Cost-Benefit Analysis (Continued): Valuing Environmental
and Other Public Goods 455
chapter 15. Decision Making under Risk 459
15.1 Expected Utility in Decision Making under Risk 459
15.2 Dependency Diagrams and Decision Trees 460
15.2.1 Dependency Diagrams 460
15.2.2 Decision Trees 462
15.2.3 A Problem: Newport Records 463
15.2.4 Of Garbage, Anchoring, and Decision Costs 472
15.3 Risk Tolerance and Utility Functions 473
15.4 The Framing of Risky Decisions: Prospect Theory Redux 476
15.4.1 Litigation vs. Settlement 477
15.4.2 Targets and Risk-Seeking and Unethical Behavior 481
15.4.3 Low-Probability Gains and Losses 482
15.4.4 The Certainty Effect 483
15.4.5 The Pseudocertainty Effect and Other Violations of the
Independence Axiom 485
15.5 Appendix A 488
15.5.1 How People’s Different Risk Attitudes Toward Gains
and Losses Can Violate the Dominance Axiom 488
15.6 Appendix B 489
15.6.1 The Certainty Effect and Violation of the Independence
Axiom 489
chapter 16. The Role of Affect in Risky Decisions 491
16.1 Risks as Feelings 491
16.1.1 Anxiety as the Shadow of Intelligence 493
16.2 The Availability Heuristic and the Role of Vividness 495
16.3 Other Determinants of Responses to Risks 497
16.3.1 The Relative Acceptability of Risks; High Anxiety and
Outrage 497
16.3.2 Familiarity 499
16.3.3 Voluntariness and Controllability 500
16.3.4 (Un)natural Origins 500
16.3.5 Omission and Action Biases 500
16.3.6 Betrayal Aversion 502
16.3.7 Judgments of Risk and Benefit 503
16.3.8 Cultural Determinants of Perceptions of Risk 504
16.4 Risk vs. Uncertainty (or Ambiguity) 505
 contents xvii

16.4.1 The Concept and Psychology of Uncertainty 505

16.4.2 The Principle of Continuity (or Exchange), Maximin,
and “Robust Satisficing” 508
16.5 Complexities of Evaluating Risks 511
16.5.1 Evaluability and Framing Redux 511
16.5.2 Cognitive and Affective Processing of Probabilities 513
16.5.3 Worry and Probability Neglect 514
16.5.4 Anchoring and Risk 519
16.6 Cost-Benefit Analysis Concluded 520
16.6.1 Assessing Risks 520
16.6.2 Not All Risks Are Equal 520
16.6.3 Discounting for Dread 521
16.6.4 The Uncertain Risks of Catastrophic Harm 521

conclusion: the lawyer as cognitive counselor

Pervasive Biases and Their Largely Ineffectual Remedies 523
The Lawyer as Cognitive Counselor 525

p art 4 guiding and influencing decisions

External Environmental Factors 535
Information and Persuasion 535
Incentives 536
Psychology 536
chapter 17. Social Influence 539
17.1 Conformity to Group Norms (Social Proof) 540
17.1.1 Death by Conformity 544
17.1.2 Social Proof and Professional and Corporate
Malfeasance 545
17.1.3 More Examples of Social Proof 546
17.1.4 Conformity Effects and the Emergence of Cascades 547
17.1.5 The Strategic Deployment of Social Proof 549
17.2 Obedience to Apparent Authority 550
17.2.1 The Impulse Toward Obedience: Legal and Policy
Implications 553
17.2.2 Defusing the Impulses Toward Obedience 554
17.3 Consistency and the Escalation of Commitment 556
17.3.1 The Foot in the Door 556
17.3.2 The Low-Ball Technique 557
17.3.3 The Irrational Escalation of Commitment 558
17.4 Liking 562
17.4.1 Familiarity 562
xviii contents

17.4.2 Similarity 563

17.4.3 Reciprocity and Cooperation 563
17.4.4 Compliments 564
17.4.5 Physical Attractiveness 564
17.5 Reciprocity 566
17.6 Conclusion 568
chapter 18. Influencing Behavior Through Cognition 571
18.1 Persuasion Through Argument 571
18.2 Framing Arguments in Terms of Gains and Losses 574
18.3 Linguistic Form 575
18.3.1 Agentive vs. Passive 575
18.3.2 Verbs and Causal Inference 576
18.3.3 The Power of Nouns 577
18.3.4 The Power of Metaphors 579
18.4 From Inclination to Action: Channel Factors 581
18.5 Priming 581
18.6 Influencing and Guiding Citizens’ Decision Making 584
18.6.1 Counseling Citizens about Risks: When Better
Information Doesn’t Necessarily Lead to Better
Choices 584
18.6.2 When Incentives Are Counterproductive 586
18.6.3 Framing Citizens’ Choices and Other Aspects of
Choice Architecture 588
chapter 19. Group Decision Making and the Effects of Accountability on
Decision Quality 595
19.1 Thinking about Group Decision Making 595
19.2 Aggregation Without Group Process: The Wisdom
of Crowds 597
19.3 Groupthink 600
19.3.1 Groupthink Symptoms and Decision-Making
Characteristics 601
19.3.2 Groupthink Antecedents 602
19.3.3 Empirical Perspectives on the Groupthink
Framework 603
19.4 The Common Knowledge Effect 604
19.4.1 Research on the Common Knowledge Effect: The
“Hidden Profile” Paradigm 606
19.4.2 Causal Accounts of the Common Knowledge Effect 608
19.4.3 Reducing the Bias in Favor of Shared Information: What
Works, What Doesn’t, What Makes Things Worse? 610
 contents xix

19.5 Group Polarization 612

19.5.1 What Causes Group Polarization? 614
19.5.2 Reducing the Tendency Toward Group
Polarization 616
19.6 Strategic Participation in Group Decision Making 617
19.7 The Susceptibility of Groups to Biases and Heuristics 618
19.8 Improving Group Decision Making 621
19.8.1 Preventing Groupthink 621
19.8.2 Preventing Premature Convergence: Devil’s Advocacy
and Dialectical Inquiry 623
19.8.3 The Stepladder Technique 624
19.8.4 Generating and Evaluating Alternatives with Nominal
Groups: The Delphi Technique 625
19.9 Avoiding “Predictable Surprises” 625
19.10 Optimizing Group Decision Making 627
19.11 The Effects of Accountability on Individual and Group
Decision-Making Performance 628
chapter 20. Conclusion: Learning from Experience 631
20.1 Learning in the Classroom: The Foundations of Problem-
Solving Expertise 631
20.2 Learning from Experience 632
20.2.1 The Focused Feedback of Mentorship 634
20.2.2 After Mentorship: Learning from Mistakes 635
Index 639
This page intentionally left blank
part 1 introduction to problem solving and decision making
chapter 1. Problem-Solving and Decision-Making Processes: Deliberation,
Intuition, and Expertise
Figure 1.1: The Problem Space
Figure 1.2: Navigating the Problem Space
Figure 1.3: Divergent and Convergent Thinking
Figure 1.4: Divergent and Convergent Elements in Formal Problem
Solving
Table 1.1: The Two-Systems Model of Information Processing
chapter 2. Framing Problems, Identifying Objectives, and Identifying
Problem Causes
Figure 2.1: Stakeholder Analysis
Figure 2.2: Ecological Scope
Figure 2.3: The Elements of Ecological Integrity
Figure 2.4: Expected Return
Figure 2.5: A Strategy, or Logic Model, to Achieve Ecological
Integrity
Table 2.1: The Problem at Terra Nueva
Table 2.2: The Problem at Terra Nueva Annotated
chapter 3. Generating Alternatives: Creativity in Legal and Policy
Problem Solving
Figure 3.1: The Nine Dot Problem
Figure 3.2: Bisociation
Figure 3.3: Domain, Field, and Individual
Figure 3.4: Scenarios for California’s Central Valley
Figure 3.5: Scenarios Relevant to Foster Care
Figure 3.6: The Ping Pong Ball in the Pipe
chapter 4. Choosing Among Alternatives
Figure 4.1: Edenville
Table 4.1: Darwin’s Pros and Cons of Marriage
Table 4.2: Wastewater Treatment Plant: Qualitative Description of
Attributes
Table 4.3: Wastewater Treatment Plant: Quantitative Description of
Attributes
Table 4.4: Wastewater Treatment Plant: Aggregate Evaluations
Table 4.5: Wastewater Treatment Plant: Weighted Attributes
Table 4.6: Wastewater Treatment Plant: Aggregate Weighted
Scores
Table 4.7: Characteristics of Different Decision-Making Strategies
xxii contents

p art 2 making sense of an uncertain world

Figure A: The Lens Model
chapter 5. Introduction to Statistics and Probability
Figure 5.1: (A) and (B) Probability Distribution and Relative
Frequencies for Die Toss
Table 5.1: Basic Contingency Table
Table 5.2: Contingency Table with Marginal Probabilities
Table 5.3: The (Unknowable) Population Distribution
Table 5.4: The Empirical Distribution in Numbers
Table 5.5: The Empirical Distribution in Relative Frequencies
Table 5.6: Type I and Type II Errors
Table 5.7: Expected Relative Frequencies
Table 5.8: Expected Absolute Frequencies
Table 5.9: Observed versus Expected Frequencies
Table 5.10: Observed minus Expected Frequencies
Table 5.11: Which Cells of a Contingency Table Are Important?
Table 5.12: The Necessity of the “No-No” Cell
chapter 6. Scores, Dollars, and Other Quantitative Variables
Figure 6.1: Examples of Dispersion
Table 6.1 (a): Deviation of High Salaries from the Mean
Table 6.1 (b): Deviation of Low Salaries from the Mean
Figure 6.2: The Population Distribution of Assistant Managers’
Salaries
Figure 6.3 (a) and (b): Distributions of Salaries Having the Same
Mean but Different Standard Deviations
Figure 6.4: Sampling Distribution of the Mean of 100 Salaries
Figure 6.5: Null Hypothesis
Figure 6.6: Alternative Hypothesis
Table 6.2: Estimates of Boy Babies Born in Large and Small
Hospitals
Figure 6.7: Scatterplot of School Spending and Reading Scores
Figure 6.8: Regression of Reading Scores on Spending
Figure 6.9: Regression of Reading Scores on Spending with
(Almost) No Correlation
Figure 6.10: Regression of Reading Scores on Spending with
Perfect Correlation
Figure 6.11: The Winner’s Curse
Figure 6.12: Some Examples of Correlation Strengths
 contents xxiii

chapter 7. Interpreting Statistical Results and Evaluating Policy Interventions

Figure 7.1: Correlation vs. Causation
Figure 7.2: A Causal Diagram of Ice Cream Sales and Drownings
Figure 7.3: A Causal Diagram Of Health Lifestyle, HRT, and
Heart Disease
Figure 7.4: Parachutes reduce the risk of injury after gravitational
challenges, but their effectiveness has not been proved with
randomized controlled studies
Figure 7.5: Decline in Fatal Car Accidents from 1956 to 1957
Figure 7.6: Fatal Car Accidents, 1951–1959
Figure 7.7: Fluctuations in Car Accidents in Connecticut and
Nearby States, 1951–1959
chapter 8. Explaining and Predicting One-Time Events
Table 8.1: Probability Table for Terra Nueva
Table 8.2: Generalized Probability Table
Table 8.3: Details, Probability, and Verisimilitude
Figure 8.1: Venn Diagram of Events A and B
Figure 8.2: Structuring the Taxi Problem as a Tree
Table 8.4: Probability Table for the Taxi Problem
Figure 8.3: “Way Too General Practicioner”
Figure 8.4: Structuring a Medical Diagnosis Problem as a Tree
chapter 9. Biases in Perception and Memory
Figure 9.1: Model of Information Processing
chapter 10. Biases in Processing and Judging Information
Table 10.1: Real Estate Appraisals
chapter 11. The Social Perceiver: Processes and Problems in Social
Cognition
Figure 11.1: Social Categorization and Person Perception
Figure 11.2: Schematic Associations: Young Black Male
Figure 11.3: Schematic Associations with Negative Prime
Figure 11.4: Schematic Associations with Positive Prime
Figure 11.5: Jones’s Model of the Behavioral Confirmation Process
Figure 11.6: A Dual-Process Model of Social Perception and
Judgment: High Road vs. Low Road Social Information
Processing
Table 11.1: The Heider/Weiner Matrix
chapter 12. Choices, Consequences, and Trade-offs
Figure 12.1: The Peak-End Phenomenon
Figure 12.2: Bernoulli’s Curve for Wealth and Utility
xxiv contents

chapter 13. Complexities of Decision Making: Decision Processes;

Relationships to Our Future Selves
Figure 13.1: Intensity and Time Course of Predicted versus
Experienced Emotion
Figure 13.2: Do you prefer $7500 now or $8000 in one year?
Figure 13.3: Present Bias
Table 13.1: Easy and Hard to Evaluate Attributes
Table 13.2: Parents’ Qualities in Child-Custody Dispute
chapter 14. Complexities of Decision Making Continued: The Power of Frames
Figure 14.1: The Prospect Theory Value Function
Figure 14.2: Number of Google Results for “I____ Have Kissed
Her (or Him)
chapter 15. Decision Making under Risk
Figure 15.1: Dependency Diagram for Carrying an Umbrella
Based on NWS Forecast
Figure 15.2: Dependency Diagram for Carrying an Umbrella
Based on NWS and Your Own Forecast
Figure 15.3: The Structure of a Decision Tree
Figure 15.4: The Decision Tree With Utilities and Probabilities
Figure 15.5: Dependency Diagram for Newport Records
Figure 15.6: Decision Tree for Newport Records
Table 15.1: Solving the Decision Tree
Table 15.2: Is the Written Disclaimer Binding?
Figure 15.7: Newport Records Decision Tree with Probabilities
Figure 15.8: Solving the Decision Tree in Newport Records with a
Tree Program
Figure 15.9: The Last Word on Expected Value Analysis
Figure 15.10 Dominance in Lotteries
Table 15.3: Did the Plaintiff Understand the Written Disclaimer?
Table 15.4: Solving the Newport Records Decision Tree by Hand
Table 15.5: The Certainty Effect
Table 15.6: The Pseudocertainty Effect
Table 15.7 Separating the Domains of Gains and Losses
Table 15.8 Allais’ Paradox
chapter 16. The Role of Affect in Risky Decisions
Figure 16.1: Risks as Feelings
Figure 16.2: Certainty Equivalent of a Lottery
Table 16.1: Aggravating and Mitigating Risk Factors
Table 16.2a: Ellsberg’s Paradox: Gamble A vs. B
Table 16.2b: Ellsberg’s Paradox: Gamble C vs. D
 contents xxv

chapter 17. Social Influence

Figure 17.1: Asch’s Comparative lines
Figure 17.2: Asch Subject (center) and Two Confederates
Figure 17.3: A Subject (left) Assists Experimenter in Strapping
Shocking Device onto “Learner” in Miligram Obedience
Experiment at Yale University
Figure 17.4: Milgram’s “Shock Panel”
chapter 18. Influencing Behavior Through Cognition
chapter 19. Group Decision Making and the Effects of Accountability on
Decision Quality
Figure 19.1: Divergent and Convergent Elements in Deliberative
Decision Making
Figure 19.2: Groupthink’s Convergent Process
This page intentionally left blank
acknowledgments

We began teaching a course in Problem Solving, Decision Making, and

Professional Judgment in 1994, when Paul was Dean of Stanford Law School
and Linda was a lecturer there.1 The main motivation for developing the course
was to provide law students with a foundation in the problem-solving skills that
are a core part of the repertoire of excellent lawyers but are for the most part
ignored by legal education. The development of the course was encouraged and
supported by Charles T. Munger and by the Radcliffe Institute for Advanced
Study at Harvard University, where Linda held a research fellowship during the
2004–2005 academic year.
We greatly appreciate the following contributions to this book. Iris Brest read
and edited the revisions of each chapter more times than she would care to recall.
Tom Griffiths wrote an initial draft of the statistics chapters; Brit Turnbull
reviewed and corrected our revisions to these chapters, and then essentially
wrote an entirely new version; Christopher Bryan, David Nussbaum, and Aner
Sela helped develop and teach the first version of the book to include materials
on public policy.2 Alicia Thesing drafted the wastewater treatment plant problem
in Chapter 4. Marc B. Victor provided great assistance with the decision tree
problem in Chapter 15.3 Rick Hanushek helped us develop the regression
example in Chapter 6. Donna Shestowsky wrote an early draft of the material on
social influence in Chapter 17. Jimmy Benjamin provided useful ideas for the
decision-making chapters. Caeli Higney added material on program evaluation

1. Linda subsequently accepted a position at Boalt Hall, where she served as

a professor of law from 1996 to 2009, and Paul became president of the William
and Flora Hewlett Foundation (though he continues to teach a course on prob-
lem solving and decision making at Stanford). Linda is now a professor of law at
the University of Hawai‘i and a Senior Research Fellow at the Center for the
Study of Law and Society at the University of California at Berkeley.
2. Chris contributed to the material on complexities of decision making
(Chapter 13) and developed much of the material on persuasion through argu-
ment and priming (Chapter 18). Aner developed the material on metacognition
in Chapter 9, and also worked on priming. Dave contributed to the material on
belief revision in Chapter 10.
3. Marc Victor is an attorney and president of Litigation Risk Analysis, Inc.,
http://www.litigationrisk.com. He has performed decision-tree analyses on
behalf of clients involved in complex, high-stakes litigation since the 1970s.
xxviii acknowledgments

to Chapter 7. Kelly Spann contributed greatly to the material on scenario plan-

ning in Chapter 3. Chris Baker, Molly Elgin, Lauren Finzer, Bria Long, Daniela
Rubio, and Tony Wang helped develop, edit, and cite-check various chapters.
Chris Baker read the manuscript from cover to cover, and sharpened the statis-
tics chapters and the portions of the teacher's manual accompanying them.
Kathryn Segovia did amazing work on the page proofs.
We also want to thank Mahzarin Banaji, Maya Bar-Hillel, Ivan Barkhorn,
Jonathan Baron, Max Bazerman, Jeremy Blumenthal, Jack Dovidio, Susan Fiske,
Daniel Gilbert, Brent Harris, Richard Heckman, Tony Greenwald, Mark Kelman,
Russell Korobkin, Varda Liberman, George Loewenstein, Sendhil Mullainathan,
Danny Oppenheimer, Richard Revesz, Lee Ross, Barry Schwartz, Donna
Shestowsky, Paul Slovic, Tom Steinbach, Jeff Strnad, George Taylor, Barbara
Tversky, James Uleman, and Em Warren, who generously commented on the
manuscript.
We are also grateful to Tom Bauch and Henry Hecht, who respectively co-
taught the course with Paul at Stanford and Linda at Boalt Hall, and who contrib-
uted many ideas which found themselves in the book; to Liz Green, who took
pleasure in finding quotidian as well as obscure journal articles and other read-
ings; to Eric Brown, who corrected all allusions to professional sports; and to
many Stanford and Boalt Hall law students who have been good-spirited experi-
mental subjects as the materials have developed.
The citations make evident our indebtedness to Amos Tversky, Daniel
Kahneman, and many social psychologists who developed and expanded the
lines of inquiry they began.
preface

Law schools prepare students for work as business dealmakers, litigators,

legal services lawyers, city attorneys, public interest advocates, and corporate
general counsel, among other careers. Schools of public policy prepare their
graduates for careers in state, local, national, and international governments and
in the nonprofit sector; their graduates carry out a vast range of activities includ-
ing making and implementing policy, advocating for policy changes, and influ-
encing the behavior of consumers, citizens, and businesses.
This book is concerned with a set of qualities and skills that we believe to be
important across the entire range of careers that lawyers and policy makers pur-
sue—skills that are also important in people’s everyday lives as citizens and con-
sumers. The qualities are sometimes defined in terms of judgment or practical
wisdom; the skills in terms of problem solving and decision making.

“Everyone complains about his memory; no one complains about his

judgment.”
—La Rochefoucauld

The title of this book uses the terms problem solving and decision making in
their conventional senses. It uses judgment in two quite different ways. In
common parlance, the term implies good judgment—the capacity to assess situ-
ations shrewdly and to draw sound conclusions.14 We hope that the book will
contribute to improving readers’ judgment in this sense. But we also draw heav-
ily on the field of social science known as “judgment and decision making”
(JDM), in which “judgment” refers mainly to the processes of empiricism—how
one ascertains facts and makes predictions about the physical and social world.
Much JDM research asks how people actually come to judgments and make
decisions; it focuses particularly on the systematic errors made by intuitive deci-
sion makers—all of us, much of the time. In addition to introducing basic ana-
lytic and quantitative tools of decision making, the book surveys the JDM
literature in the hope that understanding these errors can at least sometimes
help avoid them.

1. Anthony T. Kronman, The Lost Lawyer: Failing Ideals of the Legal Profession
72–73 (New York: Cambridge University Press, 1993).
xxx preface

The book is divided into four parts.

Part 1 is a once-over-lightly introduction to problem solving and decision
making.
Because good decision making depends on accurate empirical knowledge,
and because most important legal and policy decisions are based on probabilistic
knowledge rather than certainties, Part 2 introduces the basic concepts of statis-
tics and identifies the systematic errors that bedevil intuitive empiricists.
Part 3 is about making decisions. Analogous to Part 2, it introduces the con-
cepts and techniques of rational choice and then devotes considerable attention to
systematic ways that intuitive decision makers deviate from models of rationality.
Part 4 considers how people’s decisions and behavior can be guided or influ-
enced—for better or worse—by individuals or groups. It asks how lawyers and
policy makers can use the insights of social psychology and the JDM literature to
counsel clients and improve decisions by citizens and consumers.
The last chapter briefly summarizes the material covered in the course and
suggests how students can use the knowledge they have gained in the classroom
to continue developing professional expertise over the course of a lifetime.
This book was originally intended for a stand-alone course, taught together
with problems or “situational” case studies of the sort used in business and
public policy schools and in some law schools as well.25 The authors have used
the text for comprehensive courses in law schools and public policy programs.
As a stand-alone course, it should be of interest to students wishing to become
better problem solvers and decision makers in any aspect of their work—
including their everyday lives—as well as to those with an academic interest in
statistics, decision making, JDM, and behavioral economics.
But the book also readily lends itself to being used in parts in conjunction
with courses including ethics, counseling, and negotiation. Most clinical legal
instructors begin with Part 1 (Chapters 1–4) and then teach selected chapters.
Chapters 5–8 offer an introduction to probability and statistics that can be taught
as a stand-alone course for students (and by instructors) who have little appetite
for the quantitative. For someone interested in teaching the core insights of the
JDM literature, they are contained mainly in Chapters 9 and 10 and 12–16 with
some snippets in the preceding chapters on probability and statistics.
The book focuses on individual decision makers who, in their roles as lawyers
and policy makers, are required to take others’ interests into account. While
these professionals are often involved in negotiations and mediations, the book
does not encompass game theory or negotiation as such. We have bitten off a big
enough piece as it is, and hope that the book will provide a solid foundation for
approaching this next level of complexity.

2. See, e.g., http://www.law.stanford.edu/casestudies/.

part one
introduction to problem solving
and decision making
This page intentionally left blank
1. problem-solving and decision-making
processes
Deliberation, Intuition, and Expertise

A client with a problem consults a lawyer rather than, say, a psychologist, social
worker, or business advisor because he believes that his problem has a signifi-
cant legal dimension. But real-world problems seldom conform to the boundar-
ies that define and divide different disciplines, and it is a rare client who wants
his lawyer to confine herself strictly to “the law.” Rather, most clients expect their
lawyers to integrate legal considerations with other aspects of their problem.
Solutions are often constrained or facilitated by the law, but finding the best
solution—that is, a solution that addresses all of the client’s concerns—often
requires more than technical legal skills. Indeed, it often turns out that no
solution is ideal in all respects, and that analyzing trade-offs is itself an important
nonlegal problem-solving skill.
Reflecting this reality, an American Bar Association report on the ten “funda-
mental lawyering skills” that new lawyers should acquire places “problem solv-
ing” at the very top of the list—even before legal analysis.1 At their best, lawyers
serve as society’s general problem solvers, skilled in avoiding as well as resolving
disputes and in facilitating public and private ordering. They help clients
approach and solve problems flexibly and economically, not restricting them-
selves to the decision frames that “legal thinking” tends to impose on a client’s
needs. Good lawyers bring more to bear on a problem than legal knowledge and
lawyering skills. They bring creativity, common sense, practical wisdom, and
that most precious of all qualities, good judgment.
Designing and implementing public policy—whether done by lawyers or
people with other professional backgrounds—call for the same attributes. While
counseling and litigating focus on the individual client’s interests, policy making
is intrinsically concerned with many individuals and institutions with different
and often clashing interests. Understanding and accommodating competing,
even incommensurable, interests and designing policies that will change behav-
iors in desired ways are among the policy maker’s fundamental skills.
This chapter inquires into the nature of problem solving and decision making,
both in general and more particularly in lawyers’ work with individual clients
and policy makers’ work in government agencies and nonprofit organizations.
To illustrate problems in both domains, we begin with some vignettes from

1. Robert MacCrate, Legal Education and Professional Development—An

Educational Continuum (St. Paul, West Publishing, 1992).
4 problem solving, decision making, and professional judgment

a day in the professional life of two characters: Luis Trujillo, a partner at a mid-
sized law firm in Orange County, California; and Christine Lamm, the director
of a county environmental protection agency.

1.1 a day in the life of a problem-solving lawyer

It is an ordinary, if busy, work day for Luis Trujillo. On arriving at his office, he
finds on his calendar an initial consultation with a long-standing client at 9:00
a.m., a noon meeting with Karen Moore, a friend from law school, and an after-
noon conference with one of the firm’s associates to discuss the strategy in a
breach of contract action, which the associate has been handling under Trujillo’s
mentorship.
Trujillo’s nine o’clock meeting is with Jack Serrano, owner of Terra Nueva
Properties, a real estate development company that builds and manages low
and moderate income rental housing projects in the Orange County area.
Serrano takes great pride in his company’s reputation for providing family-
friendly, affordable housing. Until now, the company has enjoyed good rela-
tionships with its tenants and with local, state, and federal government
agencies. In all of his many years in business, Serrano has never been involved
in litigation.
Serrano arrives for his meeting with Trujillo in a state of obvious distress. He
is carrying a copy of the local newspaper, with a front-page story about a wave of
illnesses suffered by his tenants, allegedly the result of a polyurethane foam
product used to insulate the apartments. The article quotes tenants as saying that
the walls of the apartments smell bad, “like chemicals,” and it is accompanied by
a photo of tenants holding a piece of insulation at arm’s length. The article also
contains graphic descriptions of the tenants’ physical ailments and is accompa-
nied by yet another photo—this one of a lawyer and grim-faced residents of Terra
Nueva, captioned “Foam Insulation Syndrome Downs Local Residents—Tenants
to File Class Action Lawsuit.” The article quotes a report of a consumer organiza-
tion saying that similar outbreaks of “foam insulation syndrome” have occurred
elsewhere in the country.2 We return to Trujillo’s meeting with Jack Serrano
later in this and subsequent chapters.
Trujillo does pro bono work for the Los Angeles Volunteer Legal Services
Association (VLSA). After finishing his meeting with Serrano, he turns to a
memorandum from VLSA concerning mass firing of employees, without notice,
when a small manufacturing plant decided to move its operations to Mexico.

2. For an example of how life immitates hypothetical problems, see Leslie Wayne,
Chinese Drywall Linked to Corrosion, New York Times, November 23, 2009, http://www.
nytimes.com/2009/11/24/business/energy-environment/24drywall.html?scp=1&sq=chi
nese%20drywall&st=cse.
 problem-solving and decision-making processes 5

The terminations do not appear to violate either federal or state statutes, but
Trujillo has the germ of an idea of how to deal with this (to which we will return
in Chapter 3).
Trujillo’s musings are interrupted by a phone call from the front desk, alert-
ing him that Karen Moore has arrived and is headed back to his office. Moore is
a vice president for Big-Mart, a chain of discount department stores in the region.
Trujillo has helped negotiate many real estate contracts for Big-Mart, which has
grown quickly to have over thirty stores. Trujillo and Moore spend most of the
time discussing a complex deal involving a new location. But toward the end of
the lunch, Moore presents a quite different problem.
On Trujillo’s advice some years ago, Big-Mart has done regular internal audits
to ensure that it is in compliance with the law, rather than await regulatory
actions or litigation.3 She reports that the human resources director has taken an
extensive look at Big-Mart’s employment records, and has discovered an unset-
tling disparity between the salaries of male and female assistant managers. The
average male assistant manager makes $39,257 a year, while the average woman
makes $38,528—a disparity of $729.
Trujillo wonders whether the disparity might be due to other factors, such as
seniority or education. Resolving this question will require the statistical analysis
of Big-Mart’s employment data, something we will defer to Part 2 of the book.
Later in the day, Trujillo meets with Anna Wilkins, a recent law school gradu-
ate and new associate at the firm. Before turning to the main point of the meet-
ing, a breach of contract case, Trujillo mentions an incident in a trial in a tort
case—the first trial in which she ever acted as lead counsel. Anna had been about
to object to a question on the ground that it called for hearsay, when Trujillo
tugged at her sleeve and indicated that she should let it pass. Wilkins says
that she has since checked her recollection of the law. The response would
certainly have been inadmissible, and she wonders why Trujillo stopped her
from objecting. “You’re absolutely right on the law,” he says, “but we’re really
not contesting that particular factual issue. Moreover, we had been making quite
a few objections, and the judge was communicating her increasing irritation to
the jury.”
They then discuss the breach of contract case that Wilkins is handling for
the firm. The firm’s client, Clyde Evers, has sued Newport Records, a small
recording company. Newport refuses to pay for accounting software that Evers
customized and installed, saying that the software does not do what Evers said it
would do. The amount due is $600,000.

3. See Thomas D. Barton, Preventive Law and Problem Solving: Lawyering for
the Future (Lake Mary, FL: Vandeplas Publishing, 2009).
6 problem solving, decision making, and professional judgment

Yesterday, Wilkins received a phone call from Evers, who seemed upset that
nothing had happened since the case was filed quite a long time ago, and asked
her whether she couldn’t hasten its resolution.
Based on her knowledge of summary judgment from her Civil Procedure
class in law school, her reading of the contract (which disclaims any warranty of
performance), and her study of the relevant law concerning warranties, Wilkins
believes that Evers can win on summary judgment and proposes to file a motion
to that effect. After examining the case file, Trujillo introduces Wilkins to the
practical realities of summary judgment practice in the state courts.
Trujillo explains that even though a motion for summary judgment could
theoretically bring about a quick disposition of the case, it could have untoward
consequences. The judge before whom the motion is likely to be argued views
summary judgment—especially for plaintiffs—with considerable skepticism.
It is true that the contract disclaims any warranty of performance. But it
appears that Evers had made extravagant oral representations about what
the software would do. Even if those representations are not formally binding,
they may bias the judge further against summary judgment once he learns
of them.
Moreover, the law requires that warranty disclaimers be in a particular type-
face, which is somewhat different from the disclaimer in the contract with
Newport. The judge might regard the difference as inconsequential and grant
summary judgment; or he might have the jury determine whether or not the
defendant actually read and understood the disclaimer.
And there is yet another problem. The defendant claims that after a brief trial
period it stopped using Evers’ software and purchased an off-the-shelf product
instead. If the written disclaimer of warranty is ineffective, Newport may have a
claim against Evers for breach of warranty. But the statute of limitations on
this claim—which is much shorter than the statute of limitations governing
Evers’ claim—is about to run out. The defendant’s lawyer, a local sole practitioner
not known for high-quality work, probably hasn’t been focusing on the case; but
the motion may lead him to pay attention and file the claim.
What’s more, Trujillo explains, the cost of litigating the motion for summary
judgment will not be insubstantial. If the motion is denied, Evers’ costs would be
greatly increased. Even if the motion is granted, Newport Records will likely
appeal, with attendant costs and the possibility of still having to go to trial.
“Hmmm, I take your points,” says Wilkins. “We’ll just have to wait until the
case comes to trial.” But Trujillo responds, “Not so fast. Did Evers give you any
indication why he was upset that the case wasn’t progressing? Surely you dis-
cussed the time frame with him at an earlier point.” Wilkins replies that Evers
mentioned that he was anxious to have the funds from the judgment to invest in
a new venture.
Trujillo then asks Wilkins whether she can think of any available options
beside moving for summary judgment and waiting for trial, and eventually
 problem-solving and decision-making processes 7

they discuss the pros and cons of approaching Newport Records’ lawyer with a
settlement offer.

1.2 a day in the life of a problem-solving policy maker

Christine Lamm received a joint degree in law and public policy only ten years
ago. A deputy administrator in the county’s environmental protection agency,
she was catapulted into the role of acting administrator of the department upon
the sudden departure of her boss two years ago. Last year, with some misgivings
based on her lack of experience and her tendency to do everything “by the book”
in an overly deliberative manner that did not always take political realities into
account, the mayor formally appointed her head of the department. She serves
in that capacity “at the pleasure” of the mayor, meaning that she can be removed
by the mayor at will.
Lamm begins the day by continuing to work on a complex project involving
the siting of a wastewater treatment plant in Edenville. But the work is inter-
rupted (and will not be resumed until Chapter 4) by an urgent phone call from
Paula Henderson, the mayor’s chief of staff, about the Terra Nueva affair. The
mayor saw the same newspaper articles that brought Serrano to Luis Trujillo’s
office that morning, and he wants something done about the foam insulation
problem “ASAP.” Henderson asks Lamm to draft a set of proposed regulations
banning use of polyurethane foam insulation in new construction and renovation
projects in the county.
Lamm listens anxiously to Henderson’s request, mindful that her continued
employment turns on her ability to remain in the mayor’s good graces. But
Lamm doesn’t just have her job to worry about; she feels personally and profes-
sionally committed to approaching the foam insulation problem at Terra Nueva
in a manner consistent with principles of sound public policy making. Quickly
calculating how best to mediate between these two sets of concerns, Lamm
explains to Henderson that, under the state’s Administrative Procedures Act, any
regulatory initiative banning the foam insulation will have to be premised on
agency findings—arrived at using scientifically acceptable methods—that the
foam was in fact causing harm. After a rather pointed response, underscoring
the mayor’s desire to act decisively in response to requests from community
groups that had long supported him, Henderson agrees with Lamm’s suggestion
that she convene a working group to investigate the causal connection between
the insulation product and the symptoms experienced by Terra Nueva residents.
After getting off the phone with Henderson, Lamm decides to take a walk to
clear her head, and to give a maintenance worker access to her office to repair a
light above her desk. When she returns, she notices that her computer screen is
dark, and recalls that she had a number of word-processing and spreadsheet
documents open when she left. Grumbling to herself that the last thing she
8 problem solving, decision making, and professional judgment

needs on a stressful day like this is a bunch of lost work, she checks the electric
plug, thinking, “I’ll bet the maintenance guy accidentally unplugged it when he
moved my desk; that sort of thing has happened to me before.” She checks under
her desk where the electrical socket is located. “It’s plugged in, but it seems a bit
loose,” she thinks. “I’ll jiggle it.” In the process, the plug comes completely out
of the socket, and the background sound of the hard drive and fan, which she
had not noticed until then, suddenly goes silent. The computer had not been off
before, but it is now. On further inspection, Lamm discovers that the screen was
dark because the cable connecting the computer to the monitor had come loose.
Now she in fact has lost some work.

1.3 defining problem solving and decision making

As these vignettes suggest, in their day-to-day work, lawyers and public policy
makers are constantly working to solve problems, either alone or in collaboration
with others. The qualities they need to do this well are sometimes defined in
terms of judgment or practical wisdom, the skills in terms of problem solving
and decision making.
The academic and professional literatures provide a variety of definitions of
the term problem. For example, Charles Kepner and Benjamin Tregoe define a
problem as a situation where “something has gone wrong.”4 This definition cap-
tures at least two of the situations described above. “Something has gone wrong”
at Jack Serrano’s Terra Nueva apartments, and something “went wrong” with
Christine Lamm’s computer.
More broadly, law professor Gerald Lopez defines a “problem” as a situation
in which “the world we would like varies from the world as it is.”5 Correspondingly,
Lopez defines problem solving as “trying to move the world in the desired
direction.”6 This definition aptly describes Christine Lamm’s project for siting
the wastewater treatment plant. It also describes the situation presented by
VLSA’s potential case on behalf of the laid-off woodworkers. From the laid-off
workers’ perspective, “the world they would like” is one with a legally enforce-
able right to reasonable notice of pending layoffs, if not protection from the lay-
offs themselves. However, “the world as it is” apparently provides no such claim.
Crafting a novel legal theory, and then persuading a judge to apply it and provide
a remedy, represents an effort to “move the world in the desired direction.”
Problems also include situations where nothing has gone wrong yet, but
where there is reason to believe that if some action is not taken, something may

4. Charles H. Kepner and Benjamin B. Tregoe, The New Rational Manager viii
(Princeton: Princeton, 1981).
5. Gerald P. Lopez, Lay Lawyering 32 UCLA Law Review, 2 (1984).
6. Id.
 problem-solving and decision-making processes 9

go wrong in the future. Problem solving in these cases calls for the deployment
of strategies calculated to head off foreseeable future problems. Much of lawyers’
work in their roles as counselors, deal makers, estate planners, and legislative
drafters involves anticipating and avoiding problems that might arise. That’s
why Trujillo recommended that Big-Mart do a regular legal audit—the equiva-
lent of an annual physical exam. To take another example, the standard lease
agreement that Trujillo had drafted years ago for Jack Serrano tries to anticipate
various things that might “go wrong” with a particular tenancy—persistent loud
parties disturbing to neighbors, chronically late or unpaid rent, undesirable
subletting arrangements, to name a few. Part of Trujillo’s craft as a lawyer
involves his ability to anticipate problems of this sort and to work into the lease
agreement mechanisms through which Serrano can address them quickly,
effectively, and at the lowest possible cost. Unless anticipated ahead of time, a
“something may go wrong” problem can easily become a “something has gone
wrong” problem.
To accommodate problems of these various types, we adopt a more inclusive
definition of the term problem, similar to that suggested by Allen Newell and
Herbert Simon: a “problem” is any situation in which the state of affairs varies,
or may in the future vary, from the desired state, and where there is no obvious
way to reach the desired state.7 For example, we will see in Chapter 4 that there
is no single obvious solution to Christine Lamm’s problem of where to site the
wastewater treatment plant.
Newell and Simon define the conceptual area between the existing and desired
states of affairs as a problem space. To solve a problem is to navigate through the
problem space—through the virtual area between the actual or potential unsatis-
factory state and the desired state. We can represent this conception of a problem
and the problem solving process in the following way, as shown in Figure 1.1.

PRESENT The Problem Space DESIRED

STATE STATE

figure 1.1 the problem space.

7. Allen Newell and Herbert A. Simon, Human Problem Solving (Englewood

Cliffs, NJ: Prentice Hall, 1972).
10 problem solving, decision making, and professional judgment

Problem solving often requires solving a number of constituent subproblems.

For this reason, the problem space can be viewed as containing a number of
segmented paths, one or more of which leads from the initial state to the desired
state—if the problem is solvable at all. Each of the nodes located along these
paths represents a decision point—a point at which the problem solver must
choose between different available courses of action. In most situations, there is
more than one possible pathway through a problem space. Sometimes these
present themselves as distinct options at the outset. At other times, a single path-
way branches off into different directions partway through the traverse. Either
way, problem space navigation is almost always multinodal. Moving from one
node to the next requires making a decision, as the graphic in Figure 1.2
illustrates.

PRESENT DESIRED
STATE The Problem Space STATE

figure 1.2 navigating the problem space.

The various pathways through a problem space may not be equally satisfac-
tory. Some pathways that appear promising at the outset may ultimately prove to
be dead ends. Some are inferior because they demand the expenditure of exces-
sive resources, or create new problems even while solving the original one, or
because they compromise other objectives. Problem solving is thus subject to
what Newell and Simon term path constraints. A completely satisfactory solution
is a path that leads through the problem space and is consistent with all relevant
constraints.
The process described by Newell and Simon can be analogized to an expedi-
tion by explorers who must cross uncharted territory to get to their desired des-
tination. The explorers may take one path, only to have to backtrack after
discovering that it ends at a steep cliff. They may take another and encounter a
wide river. How to cross the river in effect poses an ancillary problem, which
must be solved to continue the journey. As for path constraints, the explorers
might be vegetarians, or might have mores that preclude them from traveling on
the Sabbath, which could hinder their progress toward the destination but nev-
ertheless must be taken into account.
If “problem solving” consists of “trying to move the world in the desired direc-
tion,” it must ultimately eventuate in a decision—a “commitment to a course of
 problem-solving and decision-making processes 11

action that is intended to produce a satisfying state of affairs.”8 In terms of the pre-
ceding discussion, decision making involves choosing a particular pathway across
the problem space that lies between the actual and desired states of affairs. The
“best” solution to a problem is one that satisfies, to the greatest extent possible, the
broadest range of objectives, including constraints, implicated by the problem.

1.4 deliberative and intuitive problem solving

and decision making

There are essentially two distinct, but complementary, approaches to problem

solving and decision making. One relies on analysis or deliberation, the other on
intuition. While intuition is pervasive, deliberation is relatively rare because,
among other things, it requires considerable cognitive energy. Deliberative deci-
sion making is informed by intuition at the same time as it corrects for the limi-
tations and biases of pure intuition. Intuition can also be informed by deliberation,
as happens in the development of expertise. While the processes of intuition are
largely opaque to the decision maker, deliberation is transparent. For this reason,
among others, we begin with deliberation.

1.4.1 Deliberative Processes

An ideal deliberative model of decision making consists of the following steps or
elements:
1. State, or “frame,” the problem to be solved;
2. Identify and prioritize the relevant values, interests, and objectives;
3. Identify and resolve major uncertainties concerning the cause of the problem;
4. Generate a range of plausible solutions or alternative courses of action;
5. Predict the consequences of the courses of action and assess their impact
on the relevant interests or objectives;
6. Select the course of action that optimizes the interests or objectives to be
served (i.e., make a decision);
7. Implement, observe, and learn from the outcome of the decision.
The process is recursive, beginning with the need to frame the problem in
terms of the interests involved and to consider the interests in the context of the
particular problem. After completing step 5, a decision maker would be wise to
review the earlier steps, not just because he may have accidentally omitted

8. J. Frank Yates, Elizabeth S. Veinott, and Andrea L. Patalano, Hard Decisions, Bad
Decisions: On Decision Quality and Decision Aiding, in Emerging Perspectives on
Judgment and Decision Research 13–63 (Sandra L. Schneider and James Shanteau
eds., New York: Cambridge University Press, 2003).
12 problem solving, decision making, and professional judgment

something, but because the concreteness of positing solutions can reframe

objectives and his conception of the overall problem.
1.4.1.a Framing the Problem Problem solvers sometimes go about solving
the wrong problem because they do not frame the issues adequately. They may
mistake symptoms of a problem for the problem itself, or define the problem too
narrowly, or define the problem in terms of a ready solution without taking account
of the objectives they are actually trying to achieve. Some lawyers might immedi-
ately frame Jack Serrano’s problem solely as defending against the potential class
action suit. By contrast, as we will see in Chapter 2, Trujillo will help Jack Serrano
consider the problem at Terra Nueva from a variety of different perspectives,
resisting the temptation to adopt the first problem frame that comes to mind.
1.4.1.b Identifying Interests and Objectives The German philosopher
Friedrich Nietzsche is reputed to have said, “To forget one’s purpose is the
commonest form of stupidity.”
The best frame for a problem is the one that incorporates the broadest possi-
ble range of purposes, interests, objectives, and values implicated by the situa-
tion. For this reason, the second step in deliberative problem solving entails a
thoroughgoing specification of all relevant interests and objectives, not just those
most readily brought to mind. For example, Jack Serrano obviously has an inter-
est in minimizing his legal liability to tenants of Terra Nueva. But he may have
other interests, such as his reputation in the community and his ongoing
relations with tenants. Because of Christine Lamm’s responsibilities as county
environmental administrator, concerns for the tenants’ health predominate, but
they must be balanced against other economic and social interests.
Anna Wilkins thought that a summary judgment motion was the solution to
Evers’s problem in Newport Records—getting the matter resolved quickly. But
this was a solution to a problem she did not fully understand, and could not
understand without asking her client why he was so anxious to have his case
resolved. In considering where to site the waste treatment plant, Christine Lamm
will also have to understand her various stakeholders’ interests with consider-
able specificity.
Sometimes, a client may come to a lawyer without a clear sense of his under-
lying objectives and interests, but with his mind set on a particular solution. In
these situations, a good lawyer will slow down the client and help him identify
all his interests and objectives before generating, let alone selecting among,
alternative courses of action.
1.4.1.c Diagnosing Causes While the causes of some problems are perfectly
clear, many others call for analysis or diagnosis. For example, in responding to
the situation at Terra Nueva, both Trujillo and Lamm must consider whether
Serrano’s tenants’ distress is really being caused by the foam insulation. It is
possible that their symptoms result from or are heightened by a different cause.
With respect to gender inequities at Big-Mart, Trujillo’s partner will need to
determine whether the difference in Big-Mart’s compensation of male and
 problem-solving and decision-making processes 13

female assistant managers is statistically significant and, if so, whether it is

attributable to gender or to some other variable. As Christine Lamm worked her
way through her “dead computer” problem, she had to determine or, in any
event, should have determined, whether the dark monitor screen resulted from
an unplugged power cord or from some other cause.
Just as a physician who misdiagnoses the underlying cause of a set of symp-
toms is likely to prescribe an unhelpful, or even harmful, treatment, so too may
a lawyer or policy maker take useless or counterproductive actions based on an
inadequate analysis of the facts. Section 2.5 introduces issues of empirical
analysis, which are then considered in detail in Part 2.
1.4.1.d Developing Alternatives The best problem frame is not necessarily the
first to come to mind, and this is true of potential solutions or courses of action
as well. Problem solving often benefits from a period of divergent thinking about
different possible solutions, rather than from rapid convergence on the first
seemingly attractive strategic option that sashays by. We examine the generation
of alternative courses of action in Chapter 3.
1.4.1.e Evaluating Alternatives Once a variety of potential solutions or courses
of action have been generated, a deliberative problem solver proceeds to evaluate
them. In this phase, the problem solver must predict the consequences of each plau-
sible option, and then assess the consequences in light of his client’s objectives.
We take a first look at evaluating alternatives in Chapter 4, and continue to
explore these issues throughout the rest of the book.
1.4.1.f Choosing and Implementing a Course of Action Eventually the
problem-solving process comes to a conclusion and a decision must be made.
Quite often, this requires making trade-offs among competing interests—a pro-
cess introduced in Chapter 4 and explored further in Part 3. As implementation
progresses, the selected solution is monitored, adjusted if necessary, and
reviewed to see what can be learned from the experience of its selection and
implementation.
***

1.4.2 Divergent and Convergent Thinking

The deliberative approach to problem solving combines elements of divergent
and convergent thinking. Divergent thinking expands the range of perspectives,
dimensions, and options related to a problem. Convergent thinking eliminates
possible alternatives through the application of critical analysis, thereby eventu-
ally reducing the number of options that remain open. Divergent thinking con-
ceives; convergent thinking critiques. Divergent thinking envisions; convergent
thinking troubleshoots, fine tunes, selects, and implements.
As the following figure suggests, you can think of divergent thinking as
two lines emerging from a common point and then moving away from each
other, and of convergent thinking as two lines coming together from different
directions on a single point, as shown in Figure 1.3.
14 problem solving, decision making, and professional judgment

Figure 1.4 illustrates how the model of deliberative problem solving just
described above combines elements of divergent and convergent thinking.
Early in the process, when a problem is being framed, when interests
and objectives are being identified, and when alternative solutions are being
generated, divergent thinking can bring a great deal of value to the problem-
solving endeavor. Divergent thinking enables us to conceptualize the problem
from a wide variety of perspectives, so as to permit consideration of the broadest
possible array of potential solutions. Divergent thinking helps identify the full
range of interests implicated by a particular decision. And divergent thinking
inspires innovation in coming up with solutions to the problem. Later in the
process, convergent thinking comes into play in analyzing causation, evaluating
options, choosing which course of action to implement, and implementing and
monitoring the choice.

1.5 intuitive processes in problem solving and decision

making

Intuition or know-how . . . is neither wild guessing nor supernatural inspiration, but

the sort of ability we use all the time as we go about our everyday tasks.
—Hubert and Stuart Dreyfus, Mind Over Machine9

“Most of a person’s everyday life is determined not by their conscious inten-

tions and deliberate choices but by mental processes that are put into motion by
features of the environment that operate outside of conscious awareness and
guidance.”10 Most of the time we solve problems without coming close to the
conscious, step-by-step analysis of the deliberative approach. In fact, attempting
to approach even a small fraction of the problems we encounter in a full, delib-
erative manner would bring our activities to a screeching halt. Out of necessity,
most of problem-solving is intuitive. In contrast with the deliberative model of
decision making, intuitive decisions rely on a process “that somehow produces
an answer, solution, or idea without the use of a conscious, logically defensible
step-by-step process.”11 Intuitive responses “are reached with little apparent

9. Hubert Dreyfus and Stuart Dreyfus, Mind Over Machine: The Power of
Human Intuitive Expertise in the Era of the Computer 29 (New York: Free Press,
1986).
10. John Bargh and Tanya Chartrand, The Unbearable Automaticity of Being, 54
American Psychologist 462 (1999).
11. Kenneth Hammond, Human Judgment and Social Policy: Irreducible
Uncertainty, Inevitable Error, Unavoidable Injustice 60 (New York: Oxford
University Press, 1996).
 problem-solving and decision-making processes 15

Divergent
thinking

Convergent
thinking

figure 1.3 divergent and convergent thinking.

Step 1: Define the

Problem

Step 2: Identify and Prioitize Objectives

Step 3: Clarify the Problem’s Causes

Step 4: Generate Alternative Solutions

Step 5: Evaluate Alternative Solutions Against Objectives

Step 6: Select a Solution

Step 7: Implement and Monitor

the Chosen Solution

figure 1.4 divergent and convergent elements in formal problem

solving.
16 problem solving, decision making, and professional judgment

effort, and typically without conscious awareness. They involve little or no con-
scious deliberation.”12

1.5.1 Recognition-Primed Decision Making

Experts employ intuition—often informed by deliberation and reflection on
past decisions—through a strategy that Gary Klein calls recognition-primed
decision making.13 According to this model, experience in the world gives us
myriad problem schemas—mental maps that allow us to immediately “size up” a
situation, see it as an example of a problem prototype, understand its meaning,
and know what action to take.14 If the problem does not have an obvious solution,
then rather than systematically generate and compare a number of options (as
in the deliberative model), we evaluate a plausible option through “mental
simulation”—by imagining ourselves carrying it out.15 If this does not seem to
lead to a good outcome, we evaluate another possible option.16
A problem schema often supplies the problem solver with a prefabricated
agenda for thought and action. It tells her what features of the problem situation
are important, what additional information, if any, is needed, and what action
she needs to take to reach a solution. The decision maker can then execute the
solution, often automatically, without conscious deliberation or thought. As this
description implies, many problem schemas have stock solutions, stored in the
mind as part of the schema itself.

The [chess] grand masters . . . use their intuition to recognize the promis-
ing moves that they should examine more closely. They shift to an analytic
mode by looking at the moves they will play out in the context of the
game, and rely on their ability to mentally simulate what will happen if
they play a move. In the course of these mental simulations, some of the
moves drop out because they are found to contain weaknesses. By the end
of the mental simulations, the grand masters are usually left with only a
single move they find playable.
— Gary Klein, Intuition at Work 75 (2003).

12. Robin Hogarth, Educating Intuition 14 (Chicago: University of Chicago Press,

2001).
13. Gary Klein, Sources of Power: How People Make Decisions 17 (Cambridge:
MIT Press, 1998).
14. Id. at 17, 89.
15. Id. at 20, 21.
16. In the process of solving the problem, we may even change our understanding of
the goal we are pursuing. See Klein, 122.
 problem-solving and decision-making processes 17

Recognition-primed decision making describes the mode of operation of, say,

an experienced firefighter deciding on how to make his way safely into a burning
building, or a doctor or nurse responding to a medical crisis. Klein describes the
response of two nurses working in a neonatal intensive care unit.17
Darlene had been working with premature infants for a long time; Linda was
an experienced nurse, but new to this unit. Linda was responsible for an infant
whose temperature had dropped slightly several times during her watch; each
time, she turned up the temperature in the isolette. She also noticed that the
place in the baby’s heel where a medical technician had taken a blood sample
was still bleeding a bit, and attributed it to a sloppy procedure. When Darlene
saw the baby, something “just looked funny.” After doing a quick physical, she
woke up the duty physician, saying that she thought the baby had sepsis. If they
had awaited the results of a blood culture before putting the baby on antibiotics,
it would probably have been too late. Darlene’s experience enabled her to see a
pattern which was not available to Linda—and also to respond. As Klein writes:
A “pattern” is a set of cues that usually chunk together, so that if you see a few
of the cues you can expect to find the others. When you notice a pattern you
may have a sense of familiarity—yes, I’ve seen that before! As we work in any
area, we accumulate experiences and build up a reservoir of recognized
patterns. The more patterns we learn, the easier it is to match a new situation
to one of the patterns in our reservoir . . .
Once we recognize a pattern, we gain a sense of situation: We know what
cues are going to be important and need to be monitored. We know what types
of goals we should be able to accomplish. We have a sense of what to expect
next. And the patterns include routines for responding. . . . If we see a situa-
tion as typical then we can recognize the typical way to react. 18

1.5.2 The Role of Schemas and Scripts

Schematic processing is not only the key to recognition-primed decision making,
but to navigating the everyday world. Every person, object, and situation we
encounter is unique, but to treat them as such would be impossible. Were we to
treat every experience as sui generis, we would fast be inundated by an unman-
ageable complexity that would overwhelm our cognitive capabilities. To function
at all, we must radically simplify our experience of the world.
In a classic article entitled On Perceptual Readiness, cognitive psychologist
Jerome Bruner observed that when we perceive a stimulus from our environment,
our first task is to fit that information into some existing knowledge structure

17. Gary Klein, A Case Study of Intuition, in Intuition at Work: Why Developing
Your Gut Instincts Will Make You Better at What You Do 3–9 (New York:
Doubleday, 2002).
18. Id. at 11, 12–13.
18 problem solving, decision making, and professional judgment

represented in memory.19 Perception is given meaning only when filtered through

and incorporated into preexisting cognitive elements, such as schemas.20
As people learn through experience, they organize their acquired knowledge
into an interconnected web of associations. A schema represents an individual’s
accumulated knowledge, beliefs, experiences, and affective orientations toward
the schematized construct. The schema activation process is automatic. We do
not see an object consisting of four upright legs topped by a square platform and
backed by a vertical plane; we see a “chair.” The “chair” schema is automatically
activated by the incoming visual information. This perception happens uninten-
tionally, outside of conscious awareness, through a process that is not readily
interrupted and that interferes little with other ongoing mental activity.21
A schema imposes meaning on the inherently ambiguous information
supplied by raw perception. Once a schema is activated, we implicitly expect
incoming information to be consistent with its elements.22 Schematic process-
ing is both inevitable and a pervasive source of errors of judgment. Our need to
impose order on the world leads us to see patterns even where they do not exist,
and schemas lie at the core of inaccurate stereotypes based on race, sex, and
other factors.
A script is a schema consisting of a sequence of social interactions in which
you are an actor or observer.23 Richard Nisbett and Lee Ross analogize a script to
a cartoon strip, in which each scene depicts a basic social action; for example,
a restaurant script might involve entering, ordering, eating, paying, and leaving.
“The importance of scripts . . . lies in the speed and ease with which they
make events (or secondhand accounts of events) readily comprehensible and
predictable. Their potential cost . . . is the possibility of erroneous interpreta-
tions, inaccurate expectations and inflexible modes of response.”24

19. Jerome Bruner, On Perceptual Readiness, 64 Psychological Review, 123–152 (1958).

20. Eleanor Rosch, Principles of Categorization (1978), in From Readings in Cognitive
Science: A Perspective from Psychology and Artificial Intelligence 312–22 (Allan
Collins and Edward E. Smith eds., San Mateo, CA: Morgan Kaufmann Publishers, 1988);
Ziva Kunda, Social Cognition (Cambridge: MIT Press, 1999); Richard E. Nisbett
and Lee Ross, Human Inference: Strategies and Shortcomings of Social Judgment
(Englewood Cliffs, NJ: Prentice Hall, 1980); Susan Fiske and Shelley Taylor, Social
Cognition (2d ed. Columbus, OH: McGraw-Hill, 1991); C. Neil Macrae et al.,
Stereotypes and Stereotyping (New York: Guilford Press, 1996). Fiske and Taylor
define a schema as a “cognitive structure that represents knowledge about a concept or
type of stimulus, including its attributes and the relations among those attributes.”
21. Laraine Winter, James Uleman, and Cathryn Cunniff, How Automatic Are Social
Judgments?, 49 Journal of Personality and Social Psychology 904–17 (1985).
22. Kunda, Social Cognition: Making Sense of Other People 18–21 (Cambridge,
MA: MIT Press, 1999).
23. See Roger C. Schank and Robert P. Abelson, Scripts, Plans, Goals, and
Understanding (Hillsdale, NJ: Erlbaum, 1977); Nisbett and Ross, supra at 34.
24. Nisbett and Ross, supra at 34–35.
 problem-solving and decision-making processes 19

Christine Lamm’s response to the computer problem provides an example of

schematic processing in everyday life. The sight of her dark computer screen
activated a “dead computer” schema, which she had developed as a result of
earlier experiences with computers and with other electrical appliances. Once
activated, the “dead computer” schema spontaneously supplied a causal theory
(electrical cord unplugged), a thumbnail narrative accompanying the causal
theory (cord unplugged by the window cleaner), and a solution: “plug in the cord.”
This intuitive process was virtually self-executing and required very little cogni-
tive exertion. In nontechnical terms, Lamm was just “sizing up the situation.”
Schemas and scripts play a role in a lawyer’s deciding whether to object to a
question put to a witness at trial. The process is more reflexive than reflective: A
lawyer instinctively calls out “objection!” and often only formulates a rationale
while rising to address the court. Anna Wilkins’ and Luis Trujillo’s different reac-
tions to the question that called for hearsay flowed from different schemas or
scripts. For Wilkins, the question triggered a hearsay schema, based on the rules
of evidence she had learned in law school. Trujillo intuitively placed the rules in
the broader script of a trial narrative that would be helpful to the client’s case.
Beyond these particular examples, lawyers and policy makers are constantly
called upon to engage in what Karl Weick has called “sensemaking.”25 They must
size up or interpret situations—for example, a complex financial transaction—
with a keen eye for anomalies, for what might go wrong. Recognizing patterns
without allowing oneself to become trapped by them is an essential part of
this task.

1.5.3 The Role of Affect

The heart has its reasons that reason does not understand.
—Blaise Pascal

Intuitive problem solving and decision making depends not only on the
essentially mental processes of recognizing patterns, but on affect as well. In
recent years, researchers have given increasing attention to the role of affect in
decision making—ranging from a “faint whisper of emotion to strong feelings
of fear and dread, to visceral drives such as hunger and sexual need.”26 Thomas
Hoving writes that the art historian, Bernard Berenson

25. Karl Weick, Sensemaking in Organizations: Foundations for

Organizational Science (Thousand Oaks, CA: Sage Publications, 1995).
26. Paul Slovic et al., The Affect Heuristic, in Heuristics and Biases: The Psychology
of Intuitive Judgment 397–420 (Thomas Gilovich, Dale Griffin, and Daniel Kahneman
eds., New York: Cambridge University Press, 2002). The authors define affect as “the
specific quality of ‘goodness’ or ‘badness’ (1) experienced, as a feeling state (with or without
consciousness) and (2) demarcating a positive or negative quality of a stimulus.”
20 problem solving, decision making, and professional judgment

sometimes distressed his colleagues with his inability to articulate how he

could see so clearly the tiny defects and inconsistencies in a particular work
that branded it either an unintelligent reworking, or a fake. In one case, in
fact, Berenson was able to say only that his stomach felt wrong. He had a curi-
ous ringing in his ears. He was struck by a momentary depression. Or he felt
woozy and off balance.27
American soldiers in Iraq and Afghanistan with expertise in detecting roadside
improvised explosive devices (IEDs) describe affective responses preceding con-
scious thought. On an unusually quiet morning in Mosul, a sergeant saw a car
with the windows rolled up and a couple of children inside, and warned a soldier
not to approach the car. “My body suddenly got cooler; you know, that danger
feeling.”28 A moment later, the car exploded. (See Section 10.5.2 for the sergeant’s
explanation of this intuition.)
In a laboratory experiment by Antonio Damasio and his colleagues, partici-
pants were allowed to gamble by choosing cards from several decks, some of
which had been stacked against them. Well before the participants could articu-
late why they disliked these decks, they showed physiological signs of tension
when choosing from them.29
Without emotions, our decision-making processes would be overwhelmed by
the burdens of cognition. “The action of biological drives, body states, and emo-
tions may be an indispensable foundation for rationality . . . Rationality is prob-
ably shaped and modulated by body signals, even as it performs the most sublime
distinctions and acts accordingly.”30 Emotions and reasoning exist in a delicate
balance, however, and (as we discuss in Chapter 16) emotions can sometimes
overwhelm reasoning to our detriment.
Affect plays an important role in many lawyers’ activities—not just under
pressure in the course of a trial, but in the relative calm of one’s office. Consider
Luis Trujillo’s trying to assess the candor of a witness, or someone he is negotiat-
ing with, or, indeed, his client. Or consider Christine Lamm’s dealing with upset
tenants at Terra Nueva and how her sense of her own professional responsibili-
ties is implicated in how she handles their problem. Among other things, the

27. Thomas Hoving, False Impressions: The Hunt for Big Time Art Fakes 19–20
(New York: Simon & Schuster, 1996). For a discussion of the role of intuition in the
discovery of another art fraud, see Malcolm Gladwell, Blink: The Power of Thinking
Without Thinking 3–17 (New York: Little Brown, 2005).
28. Benedict Carey, In Battle, Hunches Prove to be Valuable, New York Times, July 28,
2009, http://www.nytimes.com/2009/07/28/health/research/28brain.html?pagewanted=
2&_r=1&hp.
29. Antoine Bechara, Hanna Damasio, Daniel Tranel, Antonio R. Damasio, Deciding
Advantageously Before Knowing the Advantageous Strategy, 275 (5304) Science 1293–95
(Feb. 28, 1997).
30. Antonio Damasio, Descartes’ Error 200 (New York: G.P. Putnam’s Sons, 1994).
 problem-solving and decision-making processes 21

skilled professional must be able to differentiate between her own emotions and
those of clients and others. This capacity is a component of so-called emotional
intelligence.31 Like many of the other skills of intuition, these capacities are best
improved through reflective experience and practice.

1.6 the interaction of intuition and deliberation

1.6.1 The Two-Systems Model of Cognition

The two models of decision making described above exemplify what psycholo-
gists and cognitive scientists—following the pathbreaking work of Amos Tversky
and Daniel Kahneman—have described as a dual process or two-systems model of
cognition. “System 1 quickly proposes intuitive answers to judgment problems as
they arise, and System 2 monitors the quality of these proposals, which it may
endorse, correct, or override. The judgments that are eventually expressed are
intuitive if they retain the hypothesized initial proposal without much
modification.”32 Kahneman and Shane Frederick summarize the two systems in
this chart in Table 1.1.

table 1.1 the two-systems model of information processing

System 1 (intuitive) System 2 (reflective)

Process Characteristics
Automatic Controlled
Effortless Effortful
Associative Deductive
Rapid, parallel Slow, serial
Process opaque Self-aware
Skilled action Rule application
Content on which Processes Act
Affective Neutral
Causal properties Statistics
Concrete, specific Abstract
Prototypes Sets

31. See Daniel Goleman, Emotional Intelligence: Why It Can Matter More
than IQ (10th Anniversary Edition New York: Bantam, 2006); John D. Mayer, Peter
Salovey, and David Caruso, Models of Emotional Intelligence, in Handbook of Intelligence
396–420 (Robert Sternberg ed., 2d ed. New York: Cambridge University Press, 2000).
32. Daniel Kahneman and Shane Frederick, Representativeness Revisited: Attribute
Substitution in Intuitive Judgment, in Heuristics and Biases: The Psychology of
Intuitive Judgment, supra at 49.
22 problem solving, decision making, and professional judgment

Kahneman and Frederick suggest that “System 2 endorsements of intuitive

judgments are granted quite casually under normal circumstances.” They give
the example of this puzzle: “A bat and a ball cost $1.10 in total. The bat costs
$1 more than the ball. How much does the ball cost?”
Almost everyone we ask reports an initial tendency to answer “10 cents”
because the sum of $1.10 separates naturally into $1 and 10 cents, and
10 cents is about the right magnitude. Many people yield to this immediate
impulse. The surprisingly high rate of errors in this easy problem illustrates
how lightly System 1 monitors the output of System 2: people are not accus-
tomed to thinking hard, and are often content to trust a plausible judgment
that quickly comes to mind.
At very least, the two systems provide useful metaphors for the different ways
that people process information. Moreover, there is considerable evidence that
they map onto different physical parts of the brain. Especially when the problem
has any affective component, System 1 activities take place in the amygdala and
possibly the orbitofrontal cortex; System 2 activities take place in the lateral pre-
frontal cortex—an area of the brain that is associated with reasoning and that is
much larger in human beings than in any other species.33

1.6.2 The Limits of Deliberation: Bounded Rationality

Problem solving and decision making in professional contexts and in everyday
life call for a mixture of intuition and deliberation—of System 1 and System 2
processes. The predominance of intuitive decision making is an inevitable aspect
of the human condition of limited cognitive ability and time—what Herbert
Simon has called the condition of bounded rationality.34
We solve thousands of little problems and make thousands of decisions every
day. Given God-like cognitive powers and infinite time, we could apply a delib-
erative model to all of these actions. But under conditions of bounded rationality,
we rely mostly on intuitive decision making. Indeed, we do not think of most
events as decisions at all; they are just something we “do.” Even when we are
self-conscious about decision making and employ something resembling the
deliberative model, we seldom seek to optimize the outcomes of a decision. This
would require taking into account and ranking every criterion relevant to our

33. Samuel M. McClure et al., Conflict Monitoring in Cognition-Emotion Competition, in

The Handbook of Emotion Regulation 204–26 (James J. Gross ed., New York: Guilford,
2006); Matthew D. Lieberman, Reflexive and Reflective Judgment Processes: A Social Cognitive
Neuroscience Approach, in Social Judgments: Implicit and Explicit Processes 44–67
(Joseph P. Forgas et al. eds., New York: Cambridge University Press, 2003).
34. Herbert A. Simon, Models of Man: Social and Rational 204–05 (New York:
John Wiley & Sons, 1957); James G. March and Herbert A. Simon, Organizations
140–41 (New York: John Wiley & Sons, 1958).
 problem-solving and decision-making processes 23

satisfaction with the outcome. Rather, to use Simon’s evocative neologism, we

satisfice,35 opting for a reasonably good outcome—often the first that meets some
threshold of satisfaction—rather than devoting excessive cognitive energy to
seeking the very best.
In both professional and personal contexts, then, human decision making is
a continual intermixture of intuitive and deliberative processes, with the consid-
eration of all interests, options, and constraints seldom, if ever, being fully pur-
sued. When the stakes are high and the variables ascertainable, it often makes
sense to follow a relatively deliberative route. For example, it makes more sense
to articulate your criteria and engage in comparison shopping for an apartment
or automobile than for an ice cream cone. But sometimes, even though the
stakes are high, time is limited and one has no choice but to rely on recognition-
primed, intuitive decision making. As Justice Oliver Wendell Holmes famously
remarked in a self-defense case, “Detached reflection cannot be demanded in the
presence of an uplifted knife.”36 Consider the on-the-spot decisions demanded of
firefighters, police officers, and, for that matter, trial lawyers.
What Herbert Simon writes of judgment by managers applies as well to that
of lawyers and policy makers:
It is a fallacy to contrast “analytic” and “intuitive” styles of management.
Intuition and judgment—at least good judgment—are simply analyses frozen
into habit and into the capacity for rapid response through recognition. Every
manager needs to be able to analyze problems systematically (and with the aid
of the modern arsenal of analytical tools provided by management science
and operations research). Every manager needs also to be able to respond to
situations rapidly, a skill that requires the cultivation of intuition and judg-
ment over many years of experience and training. The effective manager
does not have the luxury of choosing between “analytic” and “intuitive”
approaches to problems. Behaving like a manager means having command of
the whole range of management skills and applying them as they become
appropriate.37
Kenneth Hammond and Berndt Brehmer have suggested that judgment takes
place on a cognitive continuum:
The analytical end of the continuum is characterized by a form of thinking
that is explicit, sequential, and recoverable. That is, it consists of a series of
steps that transform information according to certain rules . . . [which] can be
reported by the thinker . . . Intuitive thinking, on the other hand, is implicit,

35. Simon, supra note 34.

36. Brown v. United States, 256 U.S. 335, 343 (1921).
37. Herbert A. Simon, Making Management Decisions: The Role of Intuition and Emotion,
1 Academy of Management Executive 57 (Feb. 1987).
24 problem solving, decision making, and professional judgment

nonsequential, and nonrecoverable. Usually, the thinker can report no more

than the outcome of his thinking . . . Most instances of thinking will have
both intuitive and analytic components. We will refer to this composite as
quasi-rational thought.38
Most real-world problem solving and decision making is quasi-rational in this
sense. One’s initial “take” on almost any problem is essentially intuitive. Most
problems come to our attention as a result of some trigger, which activates one
or more schemas. As a result, we approach a problem with an intuitive grasp of
the interests involved; an initial, spontaneously generated hypothesis about its
cause; and a corresponding, also spontaneously generated, proposal for its
solution. The difference between deliberative and intuitive decision-making
processes lies in where our thought processes go from there—on how system-
atically and deliberatively we navigate through the problem space.
Consider Luis Trujillo’s approach to the question whether to seek summary
judgment in the Newport Records case. As an experienced litigator, Trujillo’s
summary judgment schema is embedded in a larger civil litigation script, which
includes settlement as well as pretrial or posttrial judgment as a means by which
clients’ cases are brought to successful conclusion.
Each of the factors Trujillo considers (e.g., the opposing counsel’s and judge’s
reactions to the motion) is a problem-solving exercise in itself, which he
approaches largely intuitively rather than through a sustained process of inferen-
tial reasoning. As Gary Blasi (from whom we borrowed the Newport Records
problem) notes, Trujillo does not retrieve all the irrelevant details of the past
summary judgment-related experiences, but rather the “action scripts” he
extracted from them. If these scripts had names, they might carry labels like
“passive-acting opposing counsel provoked to prepare case,” or “state court judge
denies legally meritorious summary judgment motion because of general nega-
tive attitude toward summary adjudication.”39 Trujillo’s consideration of the
consequences likely to follow from filing the motion also involves mental
simulation—imagining what happens if the judge denies or grants it.
Trujillo’s handling of the Newport Records situation also reflects elements of
a deliberative problem-solving approach. Note, for example, that he did not
simply accept the manner in which Wilkins structured the decision to be made:
“should we file a motion for summary judgment or wait for trial?” Rather, he
reframed the decision as: “given the client’s current objectives, what is the best
course of action to take at this time?” He explicitly assesses two alternative

38. Kenneth R. Hammond and Berndt Brehmer, Quasi-Rationality and Distrust:

Implications for International Conflict, in Human Judgment and Social Interactions
(Leon Rappoport and David A. Summers eds., New York: Holt, 1973).
39. Gary L. Blasi, What Lawyers Know: Lawyering Expertise, Cognitive Science, and the
Functions of Theory, 45 Journal of Legal Education 313, 355 (1995).
 problem-solving and decision-making processes 25

courses of action (moving for summary judgment and doing nothing), and
concludes that one is superior to the other. While he reaches this conclusion
intuitively and quickly, he confirms his hypothesis more systematically in the
discussion with Wilkins. Then, recognizing that neither solution meets all of the
client’s objectives, Trujillo broadens the problem frame—a strategy characteris-
tic of more deliberative problem-solving processes.
As Hubert and Stuart Dreyfus note, “when time permits and much is at stake,
detached deliberative rationality . . . can enhance the performance of even the
intuitive expert. . . . [S]uch deliberation tests and improves whole intuitions.”40
Note also—and this is an important element of interaction between lawyers and
their clients as well as associates—that the very fact of discourse with Wilkins
forces Trujillo to articulate intuitive judgments, a process which tends to move
decision making toward analysis and deliberation.

1.6.3 The Necessity and Limits of Intuition

If employing the deliberative model takes too much time—more, really, than we
could spend on most decisions—the intuitive model makes a virtue of “jumping
to conclusions.” It offers a degree of efficiency without which we could not
cope with the myriad problems and decisions we encounter daily in our personal
and work lives. Trying the most ready-to-mind solution first (such as checking a
computer’s power cord when the screen is blank) often proves successful and
costs far less in time and cognitive effort than taking a more deliberative
approach.
However, reliance on mental shortcuts may lead the intuitive decision maker
to overlook significant aspects of problems—as in Christine Lamm’s analysis of
and “solution” to her computer problem—and to consider an impoverished set
of potential solutions—as in Anna Wilkins’ selection of a summary judgment
motion as the obvious response to a client’s desire to have his case move quickly
to conclusion. Avoiding this error requires, among other things, taking the time
to acquire a better understanding of the situation and the client’s objectives,
and to consider the range of alternatives that might be employed to achieve them.
We return to the strengths and limitations of intuitive judgment throughout
the book.

1.7 professional problem solving and the nature of expertise

Analysis and intuition work together in the human mind. Although intuition is the
final fruit of skill acquisition, analytic thinking is necessary for beginners learning a

40. Hubert Dreyfus and Stuart Dreyfus, Mind Over Machine: The Power of
Human Intuitive Expertise in the Era of the Computer 40 (New York: Free Press,
1986).
26 problem solving, decision making, and professional judgment

new skill. It is also useful at the highest levels of expertise, where it can sharpen and
clarify intuitive insights.
—Hubert and Stuart Dreyfus, Mind Over Machine41

What is the relationship between expertise and the two models of problem
solving outlined above? Experts and others with proficiency in a subject often
rely heavily on intuition, but their intuitions are usually informed and comple-
mented by deliberation.
Before turning to expertise, consider the various nonprofessional activities in
which you have know-how, competence, or proficiency—for example, crossing the
street, driving a car, cooking, or playing or watching a sport. Although expertise
denotes a proficiency (whether in medicine, law, or poker) that sets someone
apart from most fellow citizens, we develop competence in these activities of
everyday life much as an expert does in a professional domain—through obser-
vation, education, mentorship, and feedback from experience. Recall, if you can,
the transition from novice to proficient actor—when you learned how to drive,
for example—as you moved from approaching each activity deliberatively,
often with conscious reference to rules or procedures, to (for the most part) just
“doing it.”
Both everyday competences and expertise are domain-specific. You can be an
expert chef and have no idea what’s going on in a football game. In Life on the
Mississippi, Mark Twain writes:
Give a man a tolerably fair memory to start with, and piloting will develop it
into a very colossus of capability. But only in the matters it is daily drilled in.
A time would come when the man’s facilities could not help noticing land-
marks and soundings, and his memory could not help holding on to them
with grip of a vise; but if you asked the same man at noon what he had had for
breakfast, it would be ten chances to one that he could not tell you. Astonishing
things can be done with the human memory if you will devote it faithfully to
one particular line of business.42
In what may be one of the earliest texts to connect judgment and expertise,
Twain goes on to describe other qualities that an expert must have:
A pilot must have a memory; but there are two higher qualities which he
must also have. He must have good and quick judgment and decision, and a
cool, calm courage that no peril can shake. Give a man the merest trifle of
pluck to start with, and by the time he has become a pilot he cannot be
unmanned by any danger a steamboat can get into; but one cannot quite say

41. Id. at xiv.

42. Mark Twain, Life on the Mississippi 116–17 (New York: Penguin Classics, 1883:
1962).
 problem-solving and decision-making processes 27

the same for judgment. Judgment is a matter of brains, and a man must start
with a good stock of that article or he will never succeed as a pilot.43
In a study of nurses’ development of expert judgment, Patricia Brenner quotes
this description of a novice nursing student:44
I gave instructions to the new graduate, very detailed and explicit instructions:
When you come in and first see the baby, you take the baby’s vital signs and
make the physical examination and you check the I.V. sites, and the ventilator
and make sure that it works, and you check the monitors and alarms. When I
would say this to them, they would do exactly what I told them to do, no
matter what else was going on. . . . They couldn’t choose which one was the
most important. . . . They couldn’t do for one baby the things that were most
important and then go to the other baby and do the things that were the most
important, and leave the things that weren’t as important until later on. . . .
If I said, you have to do these eight things, . . . they did these things, and
they didn’t care if their other kid was screaming his head off. When they did
realize, they would be like a mule between two piles of hay.
Hubert and Stuart Dreyfus note that the solution for the student is a set of
hierarchical decision-making rules, spelling out priorities. As she develops
expertise, however, these largely give way to an intuitive sense of the situation.
What does an expert professional, like Luis Trujillo, bring to problem solving
in a particular domain that a nonexpert would be unable to provide? Drawing on
Newell and Simon’s metaphor of problem solving as navigation across a virtual
problem space, an expert possesses special navigational knowledge or skill with
respect to problems of a particular type. Experts often know about pathways
through a problem space of which nonexperts are unaware. Though several
paths may look promising at the outset, an expert may be better able to predict
which ones will prove to be dead ends or worse.
Thus, a person might hire an expert to navigate a particular type of problem
space for the same reasons that a novice group of explorers would hire a guide
rather than set out entirely by themselves. An experienced guide has the ability to
“read” the terrain, finding likely routes and avoiding perils that would be invisible
to the novice. Indeed, in law, medicine, and some other domains, only a licensed
professional is permitted to guide another person across a problem space.
Consider Luis Trujillo’s expertise in serving as a guide for his client, Jack
Serrano, with respect to the complaints of the tenants at Terra Nueva. Trujillo

43. Id. at 118–19. We are grateful to James Shanteau, Mark Twain on Expertise, http://
mail.sjdm.org/pipermail/jdm-society/2007-August/003168.html, for this quotation.
44. Patricia Brenner, From Novice to Expert: Excellence and Power in Clinical
Nursing Practice 23 (Upper Saddle River, NJ: Pearson Higher Education, 1984), quoted
in Dreyfus and Dreyfus.
28 problem solving, decision making, and professional judgment

must know the substantive law and relevant procedural rules bearing on the
threatened suit whether or not it is filed. Such knowledge is necessary back-
ground to considering other means of resolving the problem. If Trujillo does not
know the law in this particular area, he has the expertise to research and learn it.
(This ability to do legal research is an area of expertise initially acquired at law
school and then honed on the job.)45
By the same token, in considering the siting of the wastewater treatment
plant, Christine Lamm will combine procedural knowledge (likely acquired in
public policy school) about how to structure decisions of this sort with substan-
tive knowledge (probably acquired on the job) of the factors relevant to the deci-
sion. And in addressing the Terra Nueva problem, it is possible that both Trujillo
and Lamm will need to acquire knowledge about medicine, epidemiology,
and the properties of polyurethane foam—to understand the connections
between exposure to the insulation and the claimed illnesses. They won’t have
time to—and won’t need to—acquire the extensive knowledge that would make
them either intuitive or sophisticated analytic problem solvers in these domains;
rather, they will have to rely on other experts.
Trujillo and Lamm and the scientific experts on whom they may call differ
from laypersons not merely in the quantity of domain-specific knowledge, but in
the quality of its cognitive organization. Experts possess large sets of particular
schemas, which describe the attributes of domain-relevant problems and con-
tain solutions to them.46 The development of professional expertise entails learn-
ing many increasingly nuanced expert schemas. Once learned, these schemas
enable the expert to recognize recurring patterns, rapidly draw inferences from
them, and execute effective responsive interventions.
Robin Hogarth asserts that the essential difference between experts and nov-
ices is that they process knowledge differently. Not surprisingly, his description
bears a strong resemblance to Gary Klein’s description of recognition-primed
decision making:47
First, experts acquire habits that help them process more information. They
learn to counteract limitations in short-term or working memory and are able
to “chunk” information more effectively. In other words, despite facing
normal limitations on memory, experts find ways of restructuring informa-
tion so that they can take more into account . . . For example, whereas a
medical student might see a patient’s symptoms as several different items of

45. See Gary Marchant and John Robinson, Is Knowing the Tax Code All It Takes to be a
Tax Expert? On the Development of Legal Expertise, in Tacit Knowledge in Professional
Practice 3–20 (Robert Sternberg and Joseph Horvath eds., Mahwah, NJ: Lawrence
Erlbaum, 1999).
46. Kurt Van Lehn, Problem Solving and Cognitive Skill Acquisition, in Foundations of
Cognitive Science 527, 545–46 (Michael I. Posner ed., Cambridge: MIT Press, 1989).
47. Hogarth, Educating Intuition, supra at 158.
 problem-solving and decision-making processes 29

information, the experienced physician may recognize the same information

as an instance of a pattern of symptoms, that is, a specific disease.
Second, experts and novices use different problem-solving strategies.
Novices tend first to identify a specific goal and then work backward through
the details of the problem to find a way to reach that goal. Experts, however,
tend first to take in the details of the problem they face, then determine
(through a process of recognizing similarities) the general framework that
best fits the data, and, finally, work forward from that framework to explore
possible solutions (goals). The process that novices follow (working backward)
is more deliberate; the process that experts follow (working forward) is more
intuitive. Not surprisingly, experts solve problems faster than novices.
In summary, professional expertise involves the development and deploy-
ment of an information-rich, schematically organized mental database, which
enables the expert to extract key elements from a complex informational array
and to match those elements with an existing expert schema stored in memory.
Once this match has occurred and the relevant schema has been activated, the
schema supplies the expert with a great deal of substantive and procedural
information. This information can then be used either intuitively, or through a
more deliberative approach, to identify problem-relevant goals and objectives,
generate alternative courses of action, accurately predict the consequences asso-
ciated with those courses of action, and select and then implement the course of
action most likely to realize the identified interests and goals.

1.8 lawyers and policy makers as experts

To understand just what sort of expertise lawyers and policy makers possess,
consider these examples of the kinds of tasks they are called upon to perform:
• An appellate lawyer seeks to influence the course of the law by persuading
a court that common-law doctrines prohibiting race discrimination by
common carriers should prohibit sexual orientation discrimination by
landlords.
• A public health administrator is asked by the mayor to develop a set of
initiatives to address the disproportionate rate of asthma-related deaths of
minority children in a large city in California.
• A company’s general counsel works with its president and chief operating
officer to develop a procedure for responding to sexual harassment claims
if any should arise; or responds to the crisis occasioned by a particular
accusation of harassment.
• A business lawyer identifies risks involved in a movie deal, including other
parties’ incentives to behave strategically, and structures the transaction so
as to mitigate the risks to his client.
30 problem solving, decision making, and professional judgment

• The executive director of a community foundation is asked by her board of

directors to develop a plan for grant making over the next five years to
reduce teen-age obesity in the community.
• An administrator at the U.S. Fish and Wildlife Service seeks to develop and
implement a plan to reintroduce wolves to an area of Montana in which
ranchers have traditionally opposed such initiatives.
• A lobbyist for a coalition of marine preservation groups seeks the enact-
ment of legislation banning the use of sonar in or near areas frequented by
humpback and gray whales.
• The Governor of American Samoa, an elected official, designs a response
to strikes at the two major U.S.-owned tuna packing plants in the territory.
The strikers demand that American Samoa no longer be exempted from
the minimum wage provisions of the Fair Labor Standards Act; the tuna
companies threaten to pull out of American Samoa if the workers don’t
return and the existing exemptions aren’t kept in place.
• An environmental lawyer representing a local government agency negoti-
ates an agreement with neighboring jurisdictions involving the quality of
drinking water in their common watershed.
Lawyers arguably bring three different kinds of professional expertise to these
diverse tasks: (1) knowledge about the law, legal institutions, and legal actors;
(2) knowledge about particular substantive domains; and (3) expertise in prob-
lem solving as such. Policy makers bring similar forms of expertise to their work,
though their process-related knowledge tends to focus more on legislative and
administrative domains rather than those involving judicial institutions.

1.8.1 Lawyers’ Knowledge About the Law, Legal Institutions, and Actors
In the first of these vignettes, the lawyer’s task of persuading an appellate court
to extend settled case law to a new area calls for creative problem solving with
respect to both analogical reasoning and advocacy. The task draws on the skills
of doctrinal analysis, legal research, writing, and advocacy—many of which play
a background role in the other tasks as well. The vignettes also highlight other
kinds of legal expertise, such as knowing how to influence judges, juries, admin-
istrative officers, and other actors in the legal system.

1.8.2 Policy Makers’ Knowledge about Legislative and Administrative

Institutions and Processes
In various other of these scenarios, public policy makers must use their
knowledge of legislative and administrative and regulatory processes to design
workable policies, and to navigate the often complex processes by which they
are adopted and implemented.
 problem-solving and decision-making processes 31

1.8.3 Expertise in Other Substantive Domains

In addition to knowledge of the legal and policy institutions, many of the tasks
outlined in the vignettes draw on experience beyond the domains of law and
public policy making as such. The general counsel relies on her sense of how
organizations function; and in her crisis prevention mode, she must know how
to deal with the press, the public, and investors as much as with legal actors. The
same can be said of the Governor of American Samoa. He must know how the
U.S. legislative process works, but he must also know how to mediate between
angry workers and indignant corporate representatives to resolve a conflict that,
if not suitably settled, may do serious harm to both the Samoan economy and its
social fabric. The environmental lawyer negotiating for the local government
agency must not only be familiar with state and federal water law; she must also
know a good deal about water quality standards, hydrology, and other relevant
areas of science and technology. The executive director of the community
foundation will need to learn a great deal about the behaviors that lead to teen
obesity and which local organizations have effective strategies to change those
behaviors.
Sometimes, the lawyer’s or policy maker’s knowledge merely shadows exper-
tise possessed by others, such as an organization’s chief operating officer,
accountants, or consulting scientists. Sometimes, however, the lawyer or policy
maker may have a broader perspective than his or her clients or constituents
do—simply by virtue of having encountered analogous problems in other
spheres. For example, over a professional lifetime, a good transactional lawyer
develops a sense for what can go wrong with a deal and how to guard against it.
A policy maker working with administrative agencies and legislatures develops
both general know-how about the processes and knowledge about particular
officials.

1.8.4 Problem-Solving Expertise

A fundamental premise of this book is that lawyers and policy makers can
develop problem-solving expertise applicable to a range of unfamiliar substan-
tive domains. Whether or not this can be tested empirically, you will at least have
some intuitions about this by the time you finish the book.
***
The materials that follow combine what might be termed positive and nega-
tive approaches to problem solving. On the positive side, they provide some of
the basic tools necessary for sound and creative problem solving. On the nega-
tive or cautionary side, they survey cognitive, social, and motivational phenom-
ena that distort people’s perceptions of reality, and impede their understanding
and pursuit of their own or their client’s goals.
This page intentionally left blank
2. framing problems, identifying objectives,
and identifying problem causes

A problem well put is half solved.

—John Dewey1

Chapter 1 introduced a model of problem solving that moves systematically

through a number of steps leading up to the selection of the particular course of
action that the decision makers believe is best-calculated to solve the problem at
hand. These steps include:
1. State, or “frame,” the problem to be solved;
2. Identify and prioritize the relevant values, interests, and objectives;
3. Identify and resolve major uncertainties concerning the cause of the
problem;
4. Generate a range of plausible solutions or alternative courses of action;
5. Predict the consequences of the courses of action and assess their impact
on the relevant interests or objectives.
6. Select the course of action that optimizes the interests or objectives to be
served (make a decision);
7. Implement, observe, and learn from the outcome of the decision.
In this chapter, we take a closer look at the first several steps, which reflect John
Dewey’s view that “a problem well put is half solved.”

2.1 the decision context and problem frames

The frame in which one views a problem or decision is a function of the decision
context,2 which is determined both by the values, interests, or objectives at stake
and by the authority of the decision maker. People may address the same prob-
lem in different decision contexts. For Frank Serrano and his lawyer, Luis
Trujillo, litigation determines the initial context for addressing the problem at
Terra Nueva, but the broad range of possible stakeholders and interests suggest
the possibility of different or broader decision contexts as well.

1. John Dewey, Logic: The Theory of Inquiry (New York: Holt, Rinehart, and
Winston, 1938).
2. Ralph Keeney, Value-Focused Thinking: A Path to Creative Decisionmaking
30 (Cambridge: Harvard University Press, 1992).
34 problem solving, decision making, and professional judgment

Christine Lamm’s decision context for Terra Nueva is circumscribed by the

scope of her official authority. For example, while Serrano might (in principle)
offer to address the tenants’ complaint by lowering their rent, this option is not
available to Lamm. Of course, the decision context for a public official is often
subject to interpretation or even to legal contest. The U.S. Environmental
Protection Agency determined that, while the Clean Air Act authorized the regu-
lation of conventional local pollutants, the agency lacked authority to treat carbon
dioxide as a pollutant. In Massachusetts v. EPA, the Supreme Court reversed this
decision.
Ralph Keeney gives an example of how decision contexts constrain the con-
sideration of potentially valuable solutions to problems:
Suppose that a utility company is required to spend $1 billion to reduce the
probability of a major accident at a nuclear power plant in the event of a large
earthquake. The main reason is to minimize radiation danger to the residents
of a nearby town. But suppose there is evidence that such an earthquake
would probably destroy the town. Indeed, it may be the case that parts of the
town would be destroyed by earthquakes or other events that would not
damage the nuclear plant with its current protection standards. An alternative
that used $200 million from the utility to improve safety in town and the
town’s ability to respond to disasters might be much better for the town’s
residents than the $1 billion spent on the plant. It would also lead to lower
utility rates for the company’s customers.3
In actuality, though, the limited jurisdictions of agencies—limitations that serve
other legitimate ends—are likely to frustrate such tradeoffs: the authority to regu-
late nuclear power plants and to take other means to ensure the town’s safety
probably reside in different agencies—perhaps even different governments.

2.2 problem framing

The major constraints on creative problem solving do not arise from jurisdic-
tional limitations on decision contexts, but from human psychology. Chapter 1
introduced the concept of “schemas,” which structure people’s perception of the
environment. We schematize problems in the same ways that we schematize
everything else. Problem schemas, also referred to as “problem frames,” serve
important cognitive functions. They enable us to perform a wide range of
problem-solving tasks intuitively, with little expenditure of cognitive resources.
But like all knowledge structures, problem frames have a potential downside

3. Id. at 205.
 framing problems and identifying causes 35

as well. They can cause us to over-simplify or otherwise misconstrue a problem,

leading to an inferior solution strategy.
Problem frames set boundaries on our thinking, defining what is “inside”
and what is “outside” the scope of the problem situation. They not only tell us
how to describe the problem, but also indirectly indicate what goals, objectives,
and solutions to consider.
A particular frame inevitably provides only one of a number of possible views
of reality and implicitly blocks the consideration of alternative perspectives with
other possible solutions. When you are viewing a situation through a particular
frame, though, it seems to provide a complete picture of reality. Indeed, the
frame is often invisible: You have the illusion that you’re seeing the world “just
as it is,” and it is difficult to imagine that there could be another way to view it.
To use another analogy, seeing a problem within a particular frame is like
walking down one of a number of paths that do not converge, with the result that
your understanding of the problem and hence the solutions to it become path-
dependent—just as some technological practices (video recording formats,
operating systems) can become path-dependent—and crowd out valuable, even
superior alternatives.
We have just used two different metaphors to illustrate the same phenome-
non. Does one or the other resonate more with you? Can you think of yet another
one? The metaphors we bring to a situation have a powerful framing effect, and
people’s framing of the same situation with different metaphors is a ready source
of misunderstanding and conflict. Do you view attending the university as a con-
sumer’s purchase of a commodity or as becoming a member of a community?
Do you view joining a law firm or a government agency as essentially transac-
tional or as joining a family? Do you view negotiation as a game, a war, or a col-
laboration? (Might professional negotiators get stuck in a particular frame
themselves?)
Frames make a difference to outcomes. There’s the old story of one monk
who asks the abbot, “May I smoke while I pray?” while another monk asks “May
I pray while I smoke?” Though we don’t have a reliable account of the abbot’s
responses, it’s a good guess that the second monk had a better chance of getting
permission. In an experiment by Varda Liberman, Steven M. Samuels, and Lee
Ross, American college students, Israeli pilots, and their flying instructors played
a Prisoner’s Dilemma–type game in which they had the choice of cooperating or
defecting. In all the groups of participants, those who were told that the exercise
was called the “Wall Street Game” were more likely to defect than those who
were told it was called the “Community Game.”4

4. Varda Liberman, Steven M. Samuels, and Lee Ross, The Name of the Game: Predictive
Power of Reputations versus Situational Labels in Determining Prisoner’s Dilemma Game
Moves, 30 Personality and Social Psychology Bulletin 1175–85 (2004).
36 problem solving, decision making, and professional judgment

Framing plays an important role in public discourse and, therefore, in public

policy making. As linguist George Lakoff notes, “People use frames—deep-seated
mental structures about how the world works—to understand facts. . . It is impos-
sible to think or communicate without activating frames, and so which frame is
activated is of crucial importance.”5 The application of this observation is com-
monplace in contemporary politics. Consider the use of “death tax” or “estate tax”
to refer to a tax on the transfer of wealth upon the death of wealthy Americans.
Or consider framing legislation designed to protect lesbians and gay men against
employment discrimination as “special rights for gays” or “guarantees of employ-
ment opportunity for all Americans.” These rhetorical tactics work because they
activate cognitively compelling frames, which predispose people toward particu-
lar preferences and judgments.
We sometimes frame problems in terms of particular time horizons, focusing
either on the near term or long term, but not on both. It often requires less cog-
nitive effort to think of the near term because one needs to do less “mental
simulation” and there are fewer uncertainties. Moreover, a short-term outcome
that is laden with either positive or negative affect is likely to be more salient
than one in the distance. On the other hand, one can be so focused on long-term
goals that one ignores more immediate considerations: Consider someone who
takes a clerkship with a known tyrant of a judge because of its (supposed) résumé
value.
In short, because they do so much cognitive “work,” problem frames are
enormously powerful. You cannot avoid viewing problems through frames, but
with effort you can become aware of how you are framing a situation and whether
there are alternatives. Understanding the origins of your frames is a good start.
Asking how other stakeholders in the situation might frame it can only broaden
your horizons and reduce possible conflict.
In any event, as the epigraph from John Dewey suggests, effective problem
solving begins with effective problem framing. We begin the chapter by explor-
ing common pitfalls involved in framing problems and suggest a variety of tech-
niques for developing better problem frames. We then work on specifying the
goals or objectives implicated in a problem and show their relationship to the
way the problem is framed. Finally, we outline the empirical task of identifying
the causes that may underlie a problem.

2.3 problem-framing pitfalls

People often solve the wrong problem. They may mistake symptoms of a problem
for the problem itself, define the problem too narrowly, or define the problem in

5. George Lakoff, Thinking Points: Communicating Our American Values

and Vision 10 (New York: Farrar, Staus & Giroux, 2006).
 framing problems and identifying causes 37

terms of one particularly salient, but not necessarily optimal, solution. Problem
framing can occur automatically with little or no conscious, considered thought.
Experts, who “over-learn” reoccurring problem frames as part of their profes-
sional training, are particularly susceptible to automatic problem framing.
The most common problem-framing errors can be divided into three broad
groups:
1. defining the problem in terms of one salient potential solution;
2. mistaking a salient symptom of the problem for the deeper problem itself;
and
3. defining a multidimensional problem unidimensionally, often as a result
of “expert,” or otherwise automatic, problem-framing processes.
We consider each of these in turn.

2.3.1 Framing by Solution

There’s an anecdote about a farmer who drives to a neighbor’s house to pick up
some bags of chicken feed. Finding no one home and the feed bags left neatly
stacked by the neighbor’s barn door, the farmer drives up beside the barn and
loads the feed into the truck bed. As he is about to get back into his truck’s cab,
he sees that his front left tire is flat—and then remembers that he has no jack in
the truck. Feeling no small measure of exasperation, the farmer begins the long
walk to the nearest gas station, without noticing that the barn’s hay-lift pulley
was perfectly positioned to lift the front of his truck.6 The farmer erred in fram-
ing his problem too narrowly. Specifically, he confused the problem (“How can
I lift my truck?”) with one particularly salient solution (“Find a jack!”).
Like the farmer in this story, clients and their lawyers are prone to “framing
by solution.” A client often comes to a lawyer with only a vague sense of his
underlying objectives, but with his mind set on a particular solution. Consider
our farmer again. Several years after the flat tire incident, he and his wife decide
that they want to retire. They ask their lawyer to transfer ownership of the farm
to their three children as equal partners. Specifically, they tell the lawyer that
they want her to draft them a grantor retained income trust (GRIT) that during
their lifetimes transfers ownership of the farm to their children. They have a
friend who did this, they explain, and they hear that the technique minimizes
estate tax liability, which is of great concern to family farmers.
The lawyer informs the couple that the tax laws have changed so that a GRIT
will not achieve these ends. Moreover, in the course of the consultation, she
learns that two of the children play different roles in running the farm, reflecting
their different interests and talents, and that the third child has moved to New
York City and has not been involved in the farm at all. The lawyer points out that

6. Jacob Getzels, Problem Finding and the Invention of Solutions, 9 Journal of Creative
Behavior 12, 15–16 (1975).
38 problem solving, decision making, and professional judgment

the trust and tax issues are relatively minor compared to questions about how
the children will participate harmoniously in running the farm and share in its
profits (or losses). She knows from experience with other family businesses that
whatever stability in family relations may exist while the parent is actively
running the enterprise often dissolves upon the parent’s retirement or death.
The clients’ problem frame—“how do we set up a GRIT?”—failed to capture
important dimensions of their actual problem: How should we structure our
estate plan to best provide for our children? Helping her clients develop a better
problem frame was an important aspect of the lawyer’s job.
The most powerful protection against shallow, solution-based problem frames
is the deceivingly simple question, “Why?” For example, by asking the clients
what they want to accomplish, the lawyer leads her clients to reframe their prob-
lem as, “How can I best provide for the needs of our family after our deaths?”
Once the problem frame is expanded past the false boundaries drawn by the
presumed solution of drafting a particular instrument, the utility of other inter-
ventions becomes apparent. By asking “why” until a client’s deepest practical
goals and objectives are recognized, a lawyer can assist her client in solving the
right problem.
Ralph Keeney provides a good example from the realm of public policy, involv-
ing the decision of how to transport hazardous material to a distant waste dump:
One objective may be to minimize the distance the material is transported by
trucks. The question should be asked, “Why is this objective important?” The
answer may be that shorter distances would reduce both the chances of acci-
dents and the costs of transportation. However, it may turn out that shorter
transportation routes go through major cities, exposing more people to the
hazardous material, and this may be recognized as undesirable. Again, for
each objective concerning traffic accidents, costs, and exposure, the question
should be asked, “Why is this important?” For accidents, the response may be
that with fewer accidents there would be fewer highway fatalities and less
exposure of the public to hazardous material. And the answer to why it is
important to minimize exposure may be to reduce the health impacts of the
hazardous material. To the question, “why is it important to reduce health
impacts?” the response may be that it is simply important. This indicates that
the objective concerning impacts on public health is a candidate to be a
fundamental objective in the decision context.7

2.3.2 Symptoms vs. Problems

Most people do not go through their lives looking for problems to solve. Solving
problems requires time, attention, money, and other resources that most citizens
and officials would rather expend in other ways. As a result, many of the problems

7. Keeney, supra at 66.

 framing problems and identifying causes 39

that lawyers and policy makers face are situations where something has “gone
wrong” rather than opportunities to keep things from going wrong or, better yet,
to improve the world. The uproar over Terra Nueva is the trigger that gets the
immediate attention of Luis Trujillo and Christine Lamm. Because the events
are vivid, consequential, and emotionally freighted, it is easy to experience the
trigger as the problem itself. But this is not necessarily so. The trigger may just
be the symptom of a deeper problem.
Of course, this does not mean that the symptom should be ignored. More
often than not, the issue that triggered the problem demands attention. The
mistake is in failing to identify the relationship between the trigger and the
deeper, often multidimensional, state of affairs that gave rise to the problem.
Consider the following example. In 1992, an undercover investigation by the
California Department of Consumer Affairs caught Sears Auto Centers system-
atically defrauding customers by selling them unnecessary repairs. A parallel
investigation in the State of New Jersey uncovered a similar pattern of fraudulent
sales activity by Sears Auto Centers located there. After initially denying that
anything improper had occurred, Sears eventually admitted that “mistakes had
occurred” and paid many millions of dollars to settle the two matters. If Sears’s
problem were defined as “resolving the enforcement actions at the lowest feasi-
ble cost,” the multimillion dollar settlements might well be viewed as a success-
ful solution.8
However, consumer advocates and eventually Sears itself defined the prob-
lem differently. The problem was not just the state enforcement actions or even
the fraudulent sales themselves. A deeper problem, and the cause of these symp-
toms, lay in Sears’s management practices, which encouraged dishonest behav-
ior. These practices included the imposition of mandatory repair-dollars-per-hour
quotas on mechanics, the use of a commission-based compensation system, and
the deployment of high-stakes contests—all designed to encourage employees to
maximize repair sales. Seen through this lens, the fraudulent behaviors and the
enforcement actions themselves were symptoms of a deeper problem involving
Sears’s management and compensation practices. Solving the problem required
changing these practices, as well as resolving the lawsuits.
As this example suggests, effective problem framing often requires careful
thinking about the causal antecedents that gave rise to a particular problem
trigger:
1. Generate an initial statement of the problem—an “initial frame
statement.”
2. Identify the “trigger” that instigated the problem-solving procedure in the
first place.

8. Clifton Brown, Sears Auto Centers, The Department of Accountancy, University

of Illinois at Urbana-Champaign (2000), http://www.business.uiuc.edu/ce%
2Dbrown/accy304spg01/Downloads/Sears%20Auto%20Centers.pdf.
40 problem solving, decision making, and professional judgment

3. Identify the situation or context in which the problem trigger occurred.

Assess the relationship between the trigger and this situation or context,
asking such questions as:
a. Is the trigger the whole problem, or is it part of a larger, deeper, or mul-
tidimensional problem?
b. What are the trigger’s causal antecedents, and which, if any, of them are
within my control?
c. If I solve the trigger problem, but do nothing else, what might happen?
d. What did my initial frame statement overlook?
4. Reframe the problem as necessary, incorporating additional aspects of the
deeper problem uncovered in these steps.

2.3.3 Automatic Problem Framing and the Pitfalls of Expert Problem Frames
Problem solving often occurs automatically, through the intuitive activation and
application of well-learned problem schemas. In these situations, some salient
aspect of the problem activates a stored representation of “problems-of-this-type”
residing in memory. Once activated, the schema directs the problem solver’s
attention to whatever information the schema contains. This may include a
ready-made problem frame, a set of plausible solutions, and an information
search blueprint, telling the schema-holder what type of unknown information
is relevant to solving the problem and how to go about obtaining and using that
information to select the most effective solution strategy.
Experts, including lawyers and specialized policy makers, are particularly sus-
ceptible to automatic problem framing. The process of developing professional
expertise entails learning sets of “expert frames.” These frames efficiently cap-
ture aspects of the situation that are relevant to the professional’s particular area
of expertise. But they are inevitably rather narrow, often failing to capture impor-
tant dimensions of the problem as it is actually experienced by clients, citizens,
and other stakeholders.
Lawyers tend to frame problems differently from the way their clients do
because they approach the problem with different schematic mind sets. The
lawyer hears the client’s story through the filters provided by the lawyer’s expert
schemas. These tell him which aspects of the client’s narrative are important and
which are not. They direct the lawyer to follow up on certain subjects, asking
additional questions and probing for detail, and to cut the client short when she
dwells too long on “irrelevant” matters. In this way, aspects of the situation that
the client subjectively experiences as important may become invisible to the
lawyer, and the entire matter may head in a direction that poorly serves the
client’s interests, broadly conceived.9

9. Conversely, the client may come to the lawyer wanting a legal solution, but having
a mistaken impression of the nature of applicable legal constraints. Because he is
 framing problems and identifying causes 41

Consider the different ways in which the problem at Frank Serrano’s Terra
Nueva apartments might be framed. If Serrano had consulted a stereotypic
“hardball” litigator, the lawyer might have conceptualized the problem simply as
being “sued” by the tenants and the solution as winning a decisive victory in
court.
But the lawsuit may be only one aspect of a multidimensional problem.
Serrano has been the subject of a front page news story that, he worries, may be
followed by others. With his reputation at stake, Serrano does not just have a
lawsuit problem, but a public relations problem as well—a problem that could be
exacerbated by a no-holds-barred litigation defense strategy. And what about
Serrano’s relationships with his company’s various stakeholders and constituen-
cies: tenants, community activists, financial backers, regulators, and his contrac-
tors and suppliers?
Though successfully defending Serrano against the suit will constitute an
important element of a successful outcome, these other aspects of the situation
may also be important to Serrano. If he is to do his job well, Trujillo must act as
a good counselor as well as a skilled litigator. He must also help Serrano identify
nonlegal aspects of the situation that he may experience as important problems
now or in the future.
A good lawyer understands that, even though it was the client’s identification
of the problem as a “legal” one that brought him to the office, a purely legal
frame may be too narrow. But a lawyer who reflects on his or her own profes-
sional development will be aware of the forces that induce such framing.
Professional expertise often works like a kind of zoom lens, focusing in on one
small portion of a broad landscape, revealing its features in great detail. This
focus—an intentional myopia—facilitates the accurate, efficient, and effective
identification of problems and the rapid deployment of interventions designed
to solve them. But it can also lead the expert to miss important features of the
larger picture, with significant collateral consequences for what lies beyond his
field of vision.
The solution to this dilemma is not to abandon the schemas that lie at the
core of the lawyer’s professional expertise. After all, it was because of this exper-
tise that the client walked in the door. Optimal problem framing in lawyer-client
collaborations requires a “both/and” rather than an “either/or” approach—an

unfamiliar with the law and lacks a robust set of expert legal schemas, the client may in
fact misframe his problem. The client may dwell on material that he erroneously believes
to be pertinent to the legal frame. He may frame his problem in terms of an ineffective or
unavailable solution. He may resist his lawyer’s efforts to direct the conversation toward
subject matters that the lawyer knows are relevant to the proper framing of the legal
problem, or he may be reluctant to provide in sufficient detail information that the
lawyer knows is important.
42 problem solving, decision making, and professional judgment

approach that enables the lawyer-client team to view the problem from a variety
of different perspectives.

2.4 identifying interests and objectives

The idealized model of problem solving (set out in Chapter 1 and at the begin-
ning of this chapter) first defines the nature of the problem and then identifies
client objectives. But the problem definition is inextricably bound up with the
those objectives. Since problems often implicate a number of different objec-
tives, the best problem frame makes room for consideration of all of them.
In the following discussion, we use objectives and goals synonymously to refer
to relatively concrete aspects of the client’s or other stakeholders’ desired out-
comes. Interests are somewhat more general or abstract than objectives and
somewhat more concrete than values. These concepts are not separated by bright
lines, but lie on a continuum. It is useful to bear in mind that someone’s actual
experience of a solution as a “success” or a “failure” will be driven by both con-
crete and more intangible factors, ranging from monetary gains or losses on one
side of the spectrum to fidelity to philosophical, religious, or spiritual values on
the other.

2.4.1 Multidimensional Frames and the Process of Framing-by-Objectives

To identify the entire range of interests and objectives affected by a particular
problem, it is helpful to view the situation from the standpoint of all its potential
stakeholders and constituencies. We refer to this process as a stakeholder analysis.
Identifying the interests of stakeholders other than himself does not imply that
an individual is or should be altruistic—he may or may not be. Rather, the anal-
ysis ensures the consideration of all external factors that could affect resolution
of the problem at the same time as it helps canvass the client’s own interests.
A problem’s trigger draws attention only to those objectives most closely asso-
ciated with it. In Serrano’s case, the trigger may be the threat of a lawsuit. The
most salient objective is avoiding or winning the lawsuit or, at least, minimizing
its financial costs. But looking at the problem from the perspectives of other
actors (as shown in Figure 2.1) conduces to a more complete articulation of the
interests at stake:
The stakeholder analysis will help Serrano and Trujillo systematically con-
sider what actions should be taken to:
1. minimize Serrano’s financial exposure, taking into account both potential
liability in the lawsuit and costs of defense;
2. minimize damage to Serrano’s reputation;
3. minimize damage to Serrano’s relationships with government regulators,
financial backers, present and potential business partners, current tenants,
and prospective tenants;
 framing problems and identifying causes 43

figure 2.1 stakeholder analysis.

4. minimize psychological stress on Serrano and others in his family and

business; and
5. “do right” by anyone who has actually been injured by materials used in the
homes.
Problem: Sketch a “stakeholder analysis” for Christine Lamm in addressing
the Terra Nueva matter.

2.4.2 Underspecification of Objectives: Causes, Consequences, and Corrective

Strategies
There are many reasons why decision makers may underspecify the interests
and objectives implicated by a particular problem. We have discussed the prob-
lem of “superficial” or “shallow” objectives, illustrated by the story of the farmer
with the flat tire. The farmer underspecified his objectives by treating one pos-
sible means of solving his problem as the end itself. We have also seen how
expert schemas—for example, that of the hardball litigator—can lead to ignoring
interests that do not come within the ambit of a professional’s conception of a
problem.
Social norms, fear of embarrassment, or other social constraints may also
interfere with the full specification of client goals and objectives. Jack Serrano
may be truly concerned about the possibility that residents in his housing units
are being physically harmed. He may be experiencing disturbing feelings of fear,
self-doubt, or shame. But something about the litigation context tends to sub-
merge litigants’ concern for others—particularly opposing parties. Perhaps
we automatically think of litigation as the “Wall Street Game,” or even as the “War
Game.” Many lawyers may feel that it is not their place to raise other-regarding
44 problem solving, decision making, and professional judgment

objectives and interests unless the client does so first. But unless the lawyer
provides an opening for the consideration of these issues, the client is apt to
avoid them too, implicitly assuming that they are irrelevant or inappropriate.
Hidden or submerged objectives do not disappear simply because they were
never explicitly identified. But if they are not identified, they are less likely to be
achieved. Under these circumstances, the client may feel dissatisfied with an
outcome, even if it achieved all overtly specified goals.
Stakeholder analysis suggests that the client may himself comprise a variety
of different “stakeholders.” A client’s various “selves” may have differing, even
competing interests in a particular problem situation. Serrano’s social self, his
emotional self, his financial self, and his ethical/religious self may all have
“stakes,” or interests, in the Terra Nueva situation. These interests may compete
and even conflict with each other. For example, if Serrano were only concerned
with mitigating legal liability, his lawyer might take steps that would damage his
relationship with tenants, suppliers, regulators, and others. The conflicts may be
essentially intrapersonal as well as strategic. For example, strong empathy with
the tenants or a desire to apologize may compromise the possibility of an ulti-
mate victory in the courts. Likewise, a legal victory might leave Serrano wracked
with guilt or shame.
In summary, the optimal framing of a problem requires a thoroughgoing
identification of all the important objectives implicated by the situation. Here is
a checklist for going about this:
1. Construct a list of stakeholders and then view the problem situation from
each stakeholder’s perspective.
2. Identify the objectives pertaining to the client’s relationship with each
stakeholder.
3. Imagine the best outcome for each stakeholder and identify exactly what it
is about that outcome you would like to achieve.
4. Imagine the worst outcome for each stakeholder and identify exactly what
it is about that outcome you want to avoid. (People often find it difficult to
think about worst-case scenarios.10)
5. Watch out for “shallow” objectives, resulting from the confusion of means
with ends. Deepen these by asking, “Why?” until you cannot go back any
further (without becoming unconstructively existential or just annoying).
6. Allow for the surfacing of objectives that may be difficult to discuss in the
particular social context in which decision making is occurring. If the client
is an individual, consider the whole person, and make the lawyer-client

10. Karen A. Cerulo, Never Saw it Coming: Cultural Challenges to Envisioning

the Worst (2006); Lee Clarke, Thinking About Worst-Case Thinking, 78 Sociological
Inquiry 154 (2008).
 framing problems and identifying causes 45

relationship as safe a space as possible for the discussion of sensitive, per-

sonal issues.
7. Stay open to the possibility that additional objectives will emerge as prob-
lem solving continues, particularly as various alternative courses of action
are being considered. If a particular course of action causes unexplained
discomfort, that may signal the presence of an overlooked interest, objec-
tive, or an as yet unidentified path constraint.
Given the constraints of bounded rationality (see Section 1.6.2), “human
beings have unstable, inconsistent, incompletely evoked, and imprecise goals at
least in part because human abilities limit preference orderliness.”11 An indi-
vidual’s objectives may only emerge with clarity during the process of deciding
how to pursue them.12 None of this diminishes the value of being as clear as
possible about one’s objectives from the outset. However, it does suggest that
as the problem-solving or decision-making process proceeds, one should be
open to new objectives coming into sight and to a revised understanding of their
relationships and relative priorities.

2.5 identifying problem causes

In many problem situations, solving a problem effectively requires accurately

diagnosing its causes.
Causal analysis often takes place at the intuitive end of the problem-solving
continuum—and often successfully so, notwithstanding Christine Lamm’s com-
puter fiasco. Developing accurate intuitions for analyzing causes is a crucial
aspect of acquiring expertise in a domain—though even within their specialties,
experts often follow up their intuitive causal hypotheses with more deliberative
diagnostic procedures. In any event, expertise is domain specific. The physician
with fabulous skills in diagnosing medical problems may have no intuitions
whatever about why her car stalls on the way to work.
Yet the nature of many lawyers’ and policy makers’ work requires them to
engage in problem solving in a wide variety of substantive domains, and this
requires developing a general sense of how to analyze causal relations. General
problem-solving skills include both a positive and what might be called a precau-
tionary component. On the positive side are structured analytical techniques
designed to assist causal reasoning. The precautionary components consist of
awareness of errors that the intuitive empiricist is prone to make. The remainder

11. James March, Bounded Rationality, Ambiguity, and the Engineering of Choice, 9 Bell
Journal of Economics 598 (1978).
12. John W. Payne, James R. Bettman, and Eric J. Johnson, The Adaptive Decision
Maker 10 (New York: Cambridge University Press, 1993).
46 problem solving, decision making, and professional judgment

of this section introduces one positive analytic technique developed by Charles

Kepner and Benjamin Tregoe13 and touches briefly on some of the errors of the
intuitive empiricist. We continue both subjects in Part 2, with an exploration of
statistical techniques that can assist in attributing causation and with a deeper
examination of the cognitive biases and heuristics that can impede sound
empirical judgments.
Recall the conversation between Christine Lamm’s and Paula Henderson, the
mayor’s chief of staff. Lamm has suggested putting together a working group of
public health professionals whose task it will be to determine whether the foam
insulation that the mayor proposes to ban is, in fact, making Terra Nueva’s ten-
ants sick. In the following scenario, Lamm is conducting an initial meeting with
Marsha Yamamoto, who works at the Division of Environmental and Occupational
Medicine at the County’s Department of Public Health, and Sam Cooke, a toxi-
cologist from the State Department of Occupational and Environmental Safety,
who has conducted preliminary interviews with Jack Serrano and various of his
employees and contractors. Lamm sets the meeting’s agenda.
LAMM: I’ve called this meeting to figure out what we know and what informa-
tion we are going to need to determine whether the foam insulation, or any-
thing else in the apartments, is actually making the tenants sick. Before I
initiate a formal rule-making process geared toward proposing a countywide
ban on the use of the material, I want to be sure that we have some reason to
believe that the insulation is actually causing harm. I know that the mayor is
concerned about the political angle here, but our concern is the science, not
the politics.
YAMAMOTO: Before we start, I think you should see this morning’s Gazette.
Poor Jack Serrano now has the United Electrical Workers after him as well.
That’s going to make the mayor even more edgy.
LAMM: Why in the world would the UEW be interested in this?
YAMAMOTO: You recall the new factory that United Semiconductor built a year
ago? Well, it turns out that many of their employees were relocated from the
company’s Valley View plant, and many moved into Serrano’s newest apart-
ments—into units 5 and 6. A couple of months ago, the union began trying
to organize the employees. It held a meeting in the community hall at Terra
Nueva. One of my assistants went to the meeting and saw their leaflets, which
said something about workers’ health. Maybe they think that showing con-
cern for the tenants at Terra Nueva will be helpful in organizing the factory.

13. Charles H. Kepner and Benjamin B. Tregoe, The New Rational Manager
(2d ed. Princeton: Princeton Research Press, 1997). For a similar approach, see Dean K.
Gano, Apollo Root Cause Analysis (2d ed. Yakima, WA: Apollonian Publications, 2003).
 framing problems and identifying causes 47

In any event, the article says that the UEW surveyed families in the new
buildings and that 40 percent have reported headaches, dizziness, or rashes
developing since they moved into their apartments. The union claims that
this is “overwhelming evidence” that the foam insulation is causing the sick-
ness. I should add that the article does not say whether the UEW had heard
complaints before it did the survey.
LAMM: By the way, didn’t I read in the documents we got from the mayor’s
office that all of the affected Terra Nueva residents live in units 5 and 6? What
can you tell me about those units?
YAMAMOTO: They were finished and people started moving into them a
year ago.
COOKE: I talked yesterday to one of Serrano’s people, who said that the manu-
facturer who supplied the insulation that they used for the older apartments
had raised his prices so much that the contractor bought the insulation for the
new units from another company. Apparently, it’s a completely new product.
Serrano was among the first developers in the county to use it. He told me
that it did have a sort of “chemical” smell when it was installed—lots of syn-
thetics do—but it wore off pretty quickly.
LAMM: Well, that’s interesting. Do you know if there are any other differences
in the materials used in the new apartments—for example, carpets or
paint?
COOKE: I’m pretty sure that there aren’t.
YAMAMOTO: Let me just say that if 40 percent of the Terra Nueva tenants are
sick, that’s a pretty strong argument in favor of there being a problem with
the insulation. I am also mindful that there have been similar reports of “foam
insulation syndrome” in other parts of the country.
COOKE: Well, hold on a minute, Marsha. My guess is that if you asked anyone
if they had had headaches, dizziness, or rashes at some point, 40 percent or
more would say yes. If there was something wrong with the foam insulation,
it should be making everyone in those units sick. And we don’t know anything
about the other units. I recall reading an article about people throughout
Europe getting sick from Perrier water after they learned that it contained a
tiny trace of some chemical. The company withdrew the so-called “contami-
nated” bottles, but the whole thing turned out to be mass hysteria. Clusters of
illnesses of this sort often turn out to be illusions.
In my view, there isn’t any solid evidence in the toxicology literature sup-
porting the theory that this particular product causes illness of this type.
Where I sit right now, particularly after doing some interviews last week, I
don’t think the problem is the foam insulation, or the union for that matter.
48 problem solving, decision making, and professional judgment

In my purely personal opinion, I wonder whether this whole thing may in fact
be a tactic by the tenants’ association. When I talked to Serrano last week, he
mentioned that, because his costs had risen, he raised rents at Terra Nueva a
couple of months ago. The new rents will go into effect the first of next year.
He told me that the tenants’ association presented tremendous opposition
to the rent hikes and that the association’s president is one of the named
plaintiffs in new lawsuit.
LAMM: Forty percent of the tenants may be a lot or a little. But we really don’t
know enough to come to any conclusions yet. We don’t even know whether
the apartments with the new product have the same insulation as the ones in
the reports of so-called “foam insulation syndrome”—whatever that is. These
are things we need to explore.
At this point, all we have is some different theories about what’s going on.
So let’s step back and walk through this systematically. I’m going to suggest
that we follow these steps:
1. Define the problem;
2. Specify the problem’s “what,” “where,” “when,” and “extent”;
3. Spot distinctions and identify changes responsible for the distinction;
4. Identify causal theories for further testing.

2.5.1 Define the Problem

LAMM (continued): Before we try to identify the cause of the problem, let’s begin
with the problem itself. To generalize, a “gone-wrong” problem involves a
deviation from some standard or norm. Here, the norm is health and the
deviation is the tenants’ sicknesses.
Initially, I had thought that the problem was “tenants experiencing head-
aches, dizziness, and rashes.” But we don’t want to jump to any conclusions.
So at this point it’s probably best to describe the problem as reports that ten-
ants are experiencing these symptoms. We don’t know what the tenants are
actually experiencing, or what questions are being asked to elicit information,
or how accurate the surveys are. Notice, by the way, that headaches, dizziness,
and rashes are pretty amorphous symptoms. We may need to come back to
that issue later. Anyway, let’s move on to step 2.

2.5.2 Specify the Problem’s “What,” “Where,” “Extent,” and “When”

LAMM (continued): The next step is to a get a clear sense both of what the prob-
lem is and what the problem is not. Let me draw a chart showing what we
know in these respects and also what we don’t know.
The what is whether there are reports of illnesses. The where, so far as we
know, are the new units. And the extent is 40 percent of the residents who
report having symptoms—40 percent of those surveyed. But we don’t know
 framing problems and identifying causes 49

table 2.1 the problem at terra nueva

Is Is not

What Reports of headaches,

dizziness, or rashes

Where Units 5 & 6 (new units) [What about the other units?]

Extent 40% of those surveyed in [60% did not report

those units symptoms?]

When Since 1 year Before

After new units built Before

After United Semiconductor Before

employees moved in

After rent increase Before

announced

Before union began After?

organizing?

how many tenants were surveyed, or who they were, or how the survey was
conducted—let alone anything about the health of tenants in the other units.
YAMAMOTO: What difference does that make if this many people are sick?
LAMM: That’s a good question, Marsha. It’s easy to get fixated on where the
problem is. But you can’t really begin to unpack the causes of the problem
until you know where it isn’t. Suppose, just for the sake of argument, that lots
of the tenants in the old units, with older insulation, also report the symp-
toms. There could still be a problem. But it wouldn’t be the problem that the
tenants, the mayor, and now half the county thinks it is—the new insulation.
That’s why we need to learn more about how the tenants in the other units are
doing—and also to be sure we know what kinds of insulation their apart-
ments have.
Just to finish my chart, let’s talk about the when of the reports. We really
don’t know when any of the tenants first experienced these symptoms, but for
the moment, let’s say it’s a year ago—it doesn’t seem as if it could have been
any earlier than that. That will give us a baseline for some other “whens”: the
symptoms were reported after the new units were built, after United
Semiconductor employees moved in, and after the rent increase was announced.
We don’t know whether the first reports were made before the union organiz-
ing campaign, although it seems likely they were—otherwise, why would the
union have conducted the survey; still, we need to check this out.
50 problem solving, decision making, and professional judgment

2.5.3 Spot Distinctions and Identify Changes Responsible for the Distinction
LAMM (continued): Now let me add a column about anything that seems
distinctive about what is and is not and about anything else noteworthy:

table 2.2 the problem at terra nueva annotated

Is Not Distinctive, noteworthy, questions,

notes

What Reports of headaches, Illness reports [Actual symptoms?]

dizziness, or rashes

Where Units 5 & 6 (new [What about (1) New insulation [though not
units) the other sure about whether some older
units?] units have the same insulation]
(2) plaintiffs live here

Extent 40% of those [60% did [How many people were surveyed?
surveyed in not report Who was surveyed? What about the
those units symptoms?] other units?]

When Since 1 year Before

After new units built Before New units

After United Before United Semi employees move in

Semiconductor
employees moved in

After rent increase Before Rent increase affected all tenants

announced (not just those in new units)

Before union After? Doesn’t seem causally related—but

organizing? check whether reports had been
made before survey

With respect to the what, the reports of illness obviously are distinctive.
With respect to the where, the fact that the known complaints come from
people living in the new units is distinctive. The particular insulation used in
their apartments may be distinctive, but we really don’t know for sure. The
whens seem most interesting. The complaints only began after the new units
were built, after the United Semiconductor employees moved in, and after
the announcement of the rent increase. But it is at least noteworthy that,
though the rent increase affects all the tenants, the only people complaining
of illness are in the new units. We need to check out if the symptoms were
reported before the union-organizing campaign.
YAMAMOTO: If so, that eliminates the union as the cause of the whole mess.
 framing problems and identifying causes 51

LAMM: Probably so—though we can’t discount the fact that the union may be
distorting or exaggerating the extent of the problem.

2.5.4 Identify Hypotheses for Further Testing

LAMM (continued): In any event, we end up with a number of possible ideas
about what’s going on at Terra Nueva. But let me suggest that we spend our
time and, I’m afraid, some of the county’s resources, focusing on one ques-
tion in particular: whether there is a causal connection between any materials
in the Terra Nueva apartments and the tenants’ reported headaches, dizzi-
ness, and rashes. To the extent we have a clear answer to this, it will affect any
other theories of what’s going on.
In addition to doing some research, including contacting the manufacturer
to see whether there have been other problems of this sort, we will need to
learn about any differences in insulation and other materials in the newer and
older units that could be related to the tenants’ sicknesses. And we need to
bring an epidemiologist to conduct a study that clearly specifies the symptoms
the tenants may be experiencing and collects data on those symptoms appear-
ing in tenants living in the new units and in those living in the old units.

2.5.5 Attributing Causation

The discussion about what’s actually going on at Terra Nueva illustrates a
number of issues relevant to the process of attributing causation, and we note
some others as well.
1. Think about what has changed with respect to any context or environment
that might explain the problem. When attempting to identify a problem’s
causes, specify the problem both with respect to what, where, and when it
is, and what, where, and when it might reasonably have been but is not.
Comparisons between, say, where and when the problem happened and
where and when else it might have happened but didn’t can be instructive.
2. Inferences of causation often depend on assessing probabilities. Yet when
assessing probabilities, one’s attention often (and often misleadingly)
focuses on particularly vivid descriptions of risks—a tendency encom-
passed by what the psychologists Amos Tversky and Daniel Kahneman call
the availability heuristic. In this case, individual tenants’ stories and front-
page newspaper reports of “foam insulation syndrome” occurring else-
where in the country provide a vivid explanation for the situation at Terra
Nueva. The statistical analysis that Christine Lamm and her colleagues will
undertake will be pallid by comparison. We will do the statistics in Chapter
5, and discuss the availability heuristic and some other phenomena that
tend to distort judgments of probability in Chapter 9.
3. People also tend to make judgments about causation based on the
degree to which they perceive a resemblance between the cause and effect.
52 problem solving, decision making, and professional judgment

For example, it is more intuitive that bad-smelling foam insulation,

rather than odorless material, would cause the symptoms reported by Terra
Nueva residents. This is one aspect of the phenomenon that Tversky and
Kahneman called the representativeness heuristic,14 which we will discuss
later as well (See Section 8.3).
4. People doing intuitive statistical analysis tend to focus on cases where the
putative cause (foam insulation) and effect (illness) occur together. But as
we will see in Part 2, one usually cannot draw valid statistical conclusions
without taking account of cases where the possible “cause” exists but the
“effect” does not occur (foam insulation but no illness), where the “cause”
is absent but the “effect” occurs (no foam insulation, but tenants feel ill),
and where neither occurs (no foam, no sickness). For instance, if it turned
out that 40 percent of the tenants in the foam-free units reported illness, it
would seriously undermine the theory that the foam caused the illness.
5. Be open to considering every plausible causal theory at the beginning of
the deliberative process, and don’t get fixated on any particular one. With
respect to each plausible theory, consider how to go about testing its
soundness. (Resist the common tendency to overvalue information that
would confirm a favored theory at the expense of information that could
disconfirm it.)
6. Correlation does not entail causation. The incidence of shark attacks is
positively correlated with ice-cream sales at the beach. Which causes which?
What might cause both?
7. One factor alone may not have caused the effect; it may result from the
confluence of several causes. Some of these may be background conditions
that were present before and after the particular effect, or present in differ-
ence places or circumstances where the effect did not occur. But they may
still be but-for causes of the effect.
8. Acquiring information to test causal theories is often expensive, and the
cost of obtaining the information itself must be balanced against the costs
of making decisions without it.

2.5.6 Problem
Toward the end of the discussion, Christine Lamm remarked, “There’s some-
thing bothering me that I can’t quite put my finger on.” It sounds as if she has
in mind yet another hypothesis for the sicknesses at Terra Nueva. What might it
be—and how would you go about testing it?

14. Daniel Kahneman and Shane Frederick, Representativeness Revisited: Attribute

Substitution in Intuitive Judgment, in HEURISTICS AND BIASES, The Psychology of Intuitive
Judgment 49 (Thomas Gilovich, Dale Griffin, and Daniel Kahneman eds., New York:
Cambridge University Press, 2002).
 framing problems and identifying causes 53

2.6 moving the world in the desired direction through

strategic planning

Terra Nueva and the examples given so far in this chapter demonstrate decision-
making strategies to use when something has gone wrong. But lawyers and
policy makers also have opportunities to be proactive in problem solving, to
move the world in a desired direction. Strategic planning—whether by state or
local governments, nonprofit organizations, or foundations—is a classic case of
forward-looking problem solving.
Here we provide an example of strategic planning by the William and Flora
Hewlett Foundation. Unlike governments, which often implement strategies
through their own agencies and personnel, grant-making foundations do so
through grants to nonprofit organizations. However, a foundation cannot know
what grants to make to what organizations until it has established its own goals
and determined strategies for achieving them.
Note to students: While the discussion of goals, barriers, outcomes, and targets fits
within the theme of Chapter 2, much of the rest of the materials concern the choice
among alternatives, and fits better with Chapter 4 or elsewhere. To present a coherent
story, we present all aspects of the strategic plan here rather than divide it among
several chapters.
Since its establishment in 1967, the Hewlett Foundation’s concerns have
included preserving the natural heritage of the American West. In 2009, the
Foundation’s Environment Program engaged in a strategic planning process for
this area of its grant-making.15 The Foundation was willing to devote approxi-
mately $125 million over five years to achieve its goals in the West, with the hope
that this would be matched by other donors interested in the region. The total
commitment, though fairly large by philanthropic standards, pales in compari-
son to the magnitude of the problems addressed. Hence the need for great clarity
and focus in goals and strategies.
Scope of work. The Foundation began by defining the geographic scope of its
work. Bounding its definition of the American West from “the uplift on the east-
ern edge of the Rocky Mountains to the Pacific coast” (as shown in Figure 2.2)
made it possible to establish clear goals for the region and analyze the barriers to
achieving them.
Goals. There are many possible goals related to protecting the natural heri-
tage of the American West, from preserving iconic landscapes to ensuring the
viability of farming and ranching communities. The Hewlett Foundation envi-
sioned “an ecologically vibrant West where the landscape is unspoiled and people
and wildlife thrive.” Since functioning natural systems underpin this vision, the

15. The planning was done by foundation staff members, led by Environment Program
Director Tom Steinbach, with the assistance of Ivan Barkhorn and staff from the Redstone
Strategy Group, LLC.
54 problem solving, decision making, and professional judgment

Anchorage
Anchorage

Calgary

Seattle

Denver

Los Angeles

figure 2.2 ecological scope.

Foundation made its ultimate goal ensuring the “ecological integrity” of the
West, meaning that natural systems would function close to the way they would
in the absence of human activity (see Figure 2.3).
Barriers. The next step in the Foundation’s strategic planning process was to
identify the barriers to achieving ecological integrity in the West. These barriers in
effect describe the problems the Foundation would need to solve to reach its goals.
The Foundation’s research and experience showed that current and emerging
human uses in the West, in addition to bringing economic benefits, can pose

Support sustainable human uses

Preserve core areas

Establish corridors

Conserve threatened ecosystems

Protect key habitat

Facilitate climate adaptation

figure 2.3 the elements of ecological integrity.

 framing problems and identifying causes 55

threats to ecological integrity. Leading threats include unplanned population

growth, resource extraction, and human-induced climate change. Rapidly
increasing population and per capita consumption in the West have led to the
use and development of formerly unoccupied lands. This has sometimes
fragmented habitats and blocked species’ migration routes, threatening their
survival. Agricultural and urban areas can also place increasing demands on
freshwater supplies, which may have negative impacts on species and ecosys-
tems. Current methods of resource extraction can create health and environ-
mental problems. In the face of these pressures, some policies critical to ensure
responsible stewardship of the West needed strengthening or improved enforce-
ment to ensure the ecological integrity of the region.
Outcomes and targets. Analysis of these problems led the Foundation to spec-
ify four key measurable targets16 in achieving its goal of ecological integrity in the
West:
1. Land conservation improved for 150 million acres.
2. Water and riparian conservation improved for 2400 river miles.
3. Fossil fuel development reduced on 85 million acres, and renewable use/
energy efficiency increased by 100 thousand gigawatt hours per year.
4. Lasting support for conservation created in each of the West’s four major
ecological zones.
Each of these key outcomes ties back to the Foundation’s overall goal and is
based on analyses of the barriers to achieving that goal. For example, the target
for improving land conservation emerged from the Foundation’s recognition of
two key problems:
• On private land, increasing population, tourism, energy development, and
division of large spaces into suburban “ranchettes” created challenges to
ecological integrity.
• Land management plans for the public land that makes up over 85 percent
of the West sometimes prioritized human use and resource extraction over
ecological integrity.
Setting goals, analyzing problems, and developing strategies to solve them
was an iterative process. Measurable targets were established to enable the
Foundation to track its progress and amend its strategies as necessary.
Throughout the planning process, program staff remained aware of the costs of
collecting information. They attempted to strike a balance between optimizing
grant-making decisions based on the best information available and working to
obtain better information.
Multiple stakeholders. Recall Serrano’s analysis of the stakeholders in the
Terra Nueva problem. The Hewlett Foundation’s strategic planning also took

16. These targets were specified in much greater detail than appears in this overview.
56 problem solving, decision making, and professional judgment

into account that it is only one of many stakeholders concerned with the American
West. The frame through which the Foundation’s Environment Program views
the problems facing the West sometimes differs from that of state and federal
policy makers, businesses, citizens, interest groups, and other organizations.
A collaborative approach to strategic planning drew on the knowledge and
resources of these stakeholders and set the groundwork for future collaboration
with them.
Advocacy. Foundations and the nonprofit organizations they support have
various tools for achieving their goals, including providing direct services
(supporting public parks, for example), basic and applied research, and policy
advocacy. Advocacy, typically based on applied research, was a major strategy for
achieving the goals of the Foundation and its grantees in Western conserva-
tion.17 for example, improving land conservation requires policy changes that
create new protected areas and improve management and planning of existing
protected lands. The Foundation’s grantees, aware of the high cost of purchasing
land, found it important to advocate for federal and state funding and incentives
for private conservation.
Understanding the many stakeholders in the West is essential to building
constituencies supportive of Western conservation. Although other stakehold-
ers’ priorities do not necessarily overlap completely with the Foundation’s, the
Foundation was able to identify opportunities for collaboration with other
funders and interest groups. For example, hunters and anglers share the
Foundation’s interest in expanding land protections that help ensure both more
enjoyable sporting and greater ecological integrity.
Stakeholders include governments as well as private interests. After eight
years of stalled policy progress at the national level during the Bush administra-
tion, the political environment became more favorable to conservation in 2009.
The Foundation’s advocacy strategies appropriately changed to respond to
the Obama administration’s conservation-minded Secretaries of Agriculture,
Energy, and Interior, and to greater interest in conservation at local levels.
Comparing alternative courses of action. The next step in the Environment
Program’s strategic planning was to generate and compare different possible
courses of action to achieve the outcomes it desired. The Foundation’s analysis
of the problem led it to embrace a combination of broad West-wide policy
changes and work in specific regions, such as the Alberta Tar Sands, where the
extraction of oil is particularly damaging to the environment.
Within this broad approach, the Foundation sought the optimal way to maxi-
mize the impact of its grants in improving the ecological integrity of the West.

17. Federal law provides constraints on certain activities defined as “attempts to influ-
ence legislation,” but the kind of advocacy in which the foundation and its grantees
engaged is fully compatable with those constraints.
 framing problems and identifying causes 57

Although we defer a full discussion of the concept of expected return until Part 3
of the book, we preview it here in order to complete this example of strategic
planning.
An expected return framework helps estimate the effectiveness of dollars
spent on particular grant-making strategies in achieving the Foundation’s goals.
Quantitative expected return analysis involved multiplying estimates of the
benefit an intervention would have if it succeeded by the likelihood that it will
succeed, and dividing by the costs of the intervention to the Foundation (see
Figure 2.4). The expected return for each promising project could thus be
compared, and the most cost-effective ones chosen.
Benefit is a monetizable measure of the progress a specific set of grants would
make toward the Foundation’s overall goal. In this case, the benefit was the
number of acres improved multiplied by what expert interviews and scientific
studies indicated would be the magnitude of improvement in ecological integrity
produced by the grant.
The likelihood of success is the estimated probability of a particular interven-
tion succeeding. Interviews with stakeholders and experts, and polls by the
League of Conservation Voters helped the Foundation calculate likelihoods of
success for particular policy changes in the various political jurisdictions encom-
passed by the West.
For the denominator, the Foundation considered the cost to the Foundation
of achieving the desired outcome. For a strategy to be cost-effective, the cost
must be less than the benefit times the likelihood of success—ideally a great deal
less. The higher the calculated expected return, the greater the Foundation’s
“bang for the buck.” The Foundation’s main costs were the amounts of its grants
and the staff resources needed to make and monitor the grants and facilitate
collaborations among stakeholders.
A rough comparison of the expected return of various strategies helped the
Foundation choose among different possible investments to achieve its conser-
vation goals. The expected return framework helped staff set clear targets and
make quantitative comparisons. Staff complemented the rough assumptions
necessary for the calculations with continual internal review and external feed-
back to select the combination of strategies that would most cost-effectively
achieve the Program’s desired outcomes.

Benefit Likelihood
(area-adjusted of success
Expected integrity improvement) × (%)
return =
Cost
($)

figure 2.4 expected return.

 58 problem solving, decision making, and professional judgment
Logic Model
Activities Outcomes Goal
Achieve West-wide policy changes that protect roadless areas
Land
Increase and target public funding for private conservation Land conservation is
Improve public land management through BLM, USFS, and state management plans improved for 150 M
Ensure climate protection policies include ecosystem conservation and adaptation priorities acres

Improve state policies for instream flow and riparian zones

Increase public funding for purchase of instream flow rights Water The
Improve watershed protection and instream flows through dam relicensing processes Water and riparian
Link surface water and groundwater regulations to reduce aquifer depletion
conservation is improved ecological
for 24 K river miles
Increase efficiency of municipal water use integrity of
the West is
Improve state and federal energy development regulations Energy ensured for
Increase ecologically sound state & federal approvals for renewable energy facilities & transmission Fossil fuel development is
reduced on 100 M acres,
wildlife
Restrict inputs and infrastructure for high carbon fuels
Increase state and federal requirements for renewable energy and efficiency
and renewables and and people
efficiency are increased by
Align utility economic incentives with increased energy efficiency investment 100,000 gigawatt hours per
Ensure national climate policy addresses high carbon fuels
year

Mobilize key constituencies to support conservation Broad-based Support

Lasting support for
Provide public feedback to decisionmakers
conservation created
Distribute analysis and research to educate decisionmakers in each of the West’s 4
major ecological zones

figure 2.5 a strategy, or logic model, to achieve ecological integrity.

 framing problems and identifying causes 59

The Strategic Plan, or Logic Model. Figure 2.5 shows the core of the strategic
plan, sometimes called a logic model, which connects the specific activities the
Foundation supports to its outcomes and targets, and ultimately to the goal of
ensuring the ecological integrity of the West. As the preceding discussion
indicates, a strategic plan is developed by working backward from the general
goal to more specific outcomes and then to the activities necessary to produce
those outcomes. Of course, a strategy is only a working hypothesis that must
remain flexible to seize high-return opportunities and respond to unanticipated
threats to critical habitat or migration corridors. Although details of the strategy
were modified in the light of changing circumstances within a year of its adop-
tion, it continues to provide the framework for the Foundation’s work in the
American West.
Targets, grantmaking, monitoring, and evaluation. Foundation staff deter-
mined outcome metrics for each of the activities and outcomes—some of which
are measurable on an annual basis, others of which may take a number of years
to assess.
Having identified the activities necessary to work toward the goal, Foundation
staff then identify the organizations that can carry out the activities. Staff mem-
bers engage in due diligence, make grants, monitor organizations’ performance,
and evaluate progress toward their goals.
We will discuss the need for feedback in decision making later in the book.
For now, suffice it to say that feedback is vital to any decision maker’s long-term
effectiveness. Monitoring and evaluation will help the Foundation respond to
changing conditions and problems as they arise. The Foundation’s explicit met-
rics and targets are helpful in measuring the contribution of each grant to achiev-
ing its overall goal. A rigorous evaluation of the program as a whole will document
successes and lessons learned for use in the next round of strategic planning.
Returning to the overall topic of this chapter, the Hewlett Foundation’s stra-
tegic plan for the West provides an illustration of how analyzing the problems to
be solved, examining relevant interests, and comparing the consequences of dif-
ferent courses of action can guide proactive attempts to move the world in a
certain desired direction.
This page intentionally left blank
3. generating alternatives
Creativity in Legal and Policy Problem Solving

In Chapters 1 and 2, we explored the first three steps in a formal model of prob-
lem solving. These steps included problem framing, the specification and pri-
oritization of relevant interests and objectives, and the identification of major
uncertainties concerning the problem’s causes. We also saw in Chapter 1 that
not all problems occur in the present tense. Some will emerge in a future that we
can anticipate and plan for more or less effectively. With any problem, present or
future, the first three steps in a formal approach are followed by a fourth, involv-
ing the generation of potential solutions. In this chapter, we consider this fourth
step in more detail, first in the context of present problems, then in the context
of problems that may emerge in the future. Finally, we consider barriers to the
effective generation of alternative solutions and suggest some possible ways to
overcome them.
More than any other aspect of problem solving, the generation of alternatives
is associated in popular consciousness with the notion of creativity, which we
explore in this chapter. As will become clear, however, creativity is implicated at
almost every step of the problem-solving process. At its core, creativity is a stance.
It is about being a “chooser,” not merely a “picker.”

3.1 “pickers” and “choosers”: problem-solving

frames of mind

In The Paradox of Choice, Barry Schwartz draws a distinction between “choosers”

and “pickers.” A chooser is someone who thinks actively before making a deci-
sion. She reflects on what is important to her in life, what is important about a
particular decision, and what the decision’s short- and long-term consequences
might be. As importantly, after surveying the most readily apparent solutions, a
chooser may conclude that perhaps none is satisfactory, and that if she wants a
better alternative, she will have to create it.1

1. Barry Schwartz, The Paradox of Choice: Why More Is Less 75 (New York:
HarperCollins, 2004).
62 problem solving, decision making, and professional judgment

“Pickers,” on the other hand, are “relatively passive selectors from whatever
is available. With a world of choices rushing by like a music video,” Schwartz
observes, “all a picker can do is grab this or that and hope for the best.”2
As Schwartz’s description suggests, what separates choosers from pickers is
the ability to pause and reflect about a problem before taking action. Generally
speaking, people are uncomfortable with an unsolved problem and, as a result,
may reflexively seize upon the first plausible solution that comes to mind.
A chooser can forbear from this sort of reflexive “picking” long enough to think
expansively about the values, goals, and objectives implicated by a problem, a
process we considered in Chapter 2. She is able to do two other things described
in Chapter 2: she can view the problem from multiple perspectives, and she can
think systematically about its causes.
Choosers are able to resist the urge to jump at the first solution that comes to
mind. This is more difficult than it might seem. As we saw in Chapter 1, many
decisions result from the spontaneous application of problem schemas. Schemas
supply a standard problem frame, a set of apparently relevant interests and objec-
tives, and one or two stock solutions. In this sort of recognition-primed problem
solving, only those solutions that are schematically associated with the problem
are even considered. Over time, as a person responds to a particular type of prob-
lem with a stock solution, that response can become deeply habituated, even if it
fails to yield optimal solutions. And if this response pattern is common to a
community of experts—lawyers or administrative officials, for instance—the
response can come to be viewed as normative.
For these reasons, striking out beyond a preset solution to a schematized
problem can be unsettling. The same is true of novel problems, for which no
standard solution has yet been developed. In such situations, “pickers” may feel
confused and anxious, which may increase the tendency to seize at the first plau-
sible solution that presents itself.
Choosers, on the other hand, are able to cope with the indeterminacy accom-
panying an unsolved problem. In the empty space between problem recognition
and the selection of an alternative solution, a chooser can think more systemati-
cally, and ultimately decide more successfully, about how best to frame the prob-
lem, about the goals, values, and interests it implicates, about its causes, and
about its potential solutions. It is this last subject—the generation of alternative
courses of action—to which our attention now turns.

2. Id. at 224.
 generating alternatives 63

3.2 frameworks for generating solutions to present problems

3.2.1 Basic Principles

In his work on “lateral thinking,”3 Edward deBono observes that, much of the
time, we “satisfice” in solving problems, that is, abandon the search for alterna-
tives as soon as the first plausible solution comes to mind. On the upside, this
saves cognitive energy and time. The downside, of course, is that the first solu-
tion that comes to mind may not be very good. Another downside is that if no
solution comes readily to mind we may prematurely conclude that the problem
is unsolvable and abandon the effort to solve it. Seeing no obvious way to reach
a desired goal, we simply conclude that, “It can’t be done.”
To avoid these pitfalls, we need to train ourselves to “leave the problem open”
long enough to generate a rich range of alternative solutions.4 In this respect, we
agree with law professor Tom Barton, who writes:
[P]roblems are likely to be solved better—more reliably, more durably, more
respectfully and with fewer side-effects—where a diversity of alternative pro-
cedures is available to approach the problem . . . The more dimensions of a
problem have been considered in advance of applying a solution, the more
comprehensive and better the solution is likely to be.5
Unfortunately, life provides no guarantee that problem solvers will generate a
rich set of alternative solutions, nor any assurance that those alternatives will
circumvent well-worn ruts. Here are some pedestrian ways to think about
increasing the likelihood that a full complement of alternatives will be identified.
We describe prospects for enhancing creativity later in the chapter.

3.2.2 Different Kinds of Alternatives

In Chapter 2, we considered the variety of interests and objectives that a person
might have in relation to a problem or decision. One way of stretching your

3. Edward DeBono, Lateral Thinking: Creativity Step by Step (New York: Harper
& Row, 1970).
4. One note of caution is warranted here. Both the quality of decisions and decision
makers’ subjective sense of satisfaction with their decisions can be impaired by the desire
to “leave doors open,” that is, to avoid committing to a particular course of action. For a
discussion of this problem, see Jiwoong Shin and Dan Ariely, Keeping Doors Open: The
Effect of Unavailability on Incentives to Keep Options Viable, 50 Mgmt. Sci. 575 (2004);
Daniel T. Gilbert and Jane E. J. Ebert, Decisions and Revisions: The Affective Forecasting of
Changeable Outcomes, 82 J. Personality & Soc. Psychol. 503 (2002).
5. Thomas D. Barton, Conceiving the Lawyer as Creative Problem Solver, 34 Cal. W. L.
Rev. 267, 269 (1998).
64 problem solving, decision making, and professional judgment

ability to generate different types of alternatives is to use these interests and

objectives to generate potential solutions of two sorts: 6
1. solutions that maximize the achievement of the most important among
many objectives, even at the expense of the others; and
2. solutions that simultaneously satisfy several important objectives, even if
no one solution is optimal for any one of them.
Taking the first step before the second helps keep your aspirations high and
may lead to solutions that achieve secondary as well as primary. As we describe
in greater detail in Chapter 12 (on trade-offs), it is not always possible to “have it
all,” but you cannot know unless you have thought about alternative solutions
long enough to try.

3.2.3 Buying Information and Time

An ideal solution solves an entire problem in all of its aspects. But many prob-
lems have no “silver bullet” solution that satisfies all the interests at stake. More
importantly, early in one’s thinking about a problem, one may not have suffi-
cient time, information, or “maneuvering room” to devise an optimal solution.
Sometimes, it seems, the world won’t wait for us to generate and evaluate a rich
set of alternatives. If we don’t decide quickly, the world may decide for us.
Under these circumstances, it is worth considering various types of
information-buying or time-buying alternatives. That “alternatives exclude”7 is
a fact of life. But one can sometimes delay the inevitable, and this can conduce
to better outcomes down the road. Here are some examples.
3.2.3.a Information-Gathering Alternatives Consider steps that will gather
essential information without (significantly) compromising ultimate solutions.
One productive way to gather more information is to study the people involved
in a problem. For example, product design engineers spend endless hours deter-
mining what will work for consumers. They map the users’ behavior, ask them
to keep journals, interview diverse groups of users, and seek out people who
have never seen a particular product before to ask for their reaction.8 They do this
with the understanding that a useful technological innovation is only one step
toward adoption.
Like engineers, good lawyers seek to understand what will actually assist
clients and be persuasive to legal decision makers. Lawyers will read a judge’s
opinions in order to better understand how he or she approaches issues. In
general, studying the people involved in a problem may provide cues as to what

6. See John Hammond, Ralph, Keeney, and Howard Raiffa, Smart Choices: A
Practical Guide to Making Better Decisions (Boston, MA: Harvard Business School
Press, 1999).
7. John Gardner, Grendel 159 (New York: Knopf, 1971).
8. Bruce Nussbaum, This is the IDEO Way, Business Week, May 17, 2008.
 generating alternatives 65

solutions should look like, and whether they will pass the approval of
gatekeepers.
Think of some information-gathering alternatives that Luis Trujillo or
Christine Lamm might devise after being confronted with the problem at Terra
Nueva discussed in the previous chapters.
3.2.3.b Time-Buying Alternatives Especially when faced with a stress-inducing
problem, many of us are inclined to follow the adage, “Don’t just sit there, do
something!” and follow a “fire, ready, aim” process. But, some problems take
care of themselves over time. How often have you made a doctor’s appointment,
only to have the symptoms disappear before entering the waiting room? Even
those problems that are likely to persist may require time for analysis, informa-
tion gathering, consultation, and the generation and evaluation of an alternative
course of action. In these circumstances, consider first alternatives that will buy
you time—again, without (significantly) compromising ultimate solutions. What
might Luis Trujillo advise Jack Serrano to do in this respect in the face of public
controversy and the threat (or actuality) of suit?
One important way that people keep their options open over time is, well, to
buy options—not financial options (to purchase securities at a particular price),
but real options, where the underlying asset is not a financial instrument.
William Hamilton writes:9
An essential characteristic of all options is asymmetry in the distribution of
returns—greater upside potential than downside exposure. This results from
opportunities to terminate the investment or otherwise limit negative out-
comes while taking future actions to exploit positive outcomes fully . . . In
concept, real options are quite similar to financial options because they both
create the flexibility, but not the requirement for further action, as the future
unfolds.
If you have ever made a non-refundable deposit on any purchase, you have
purchased a real option.10
Cass Sunstein suggests that certain government policies can be understood to
create societal options against irreversible harms. Sunstein writes:
[W]here a decision problem is characterized by (1) uncertainty about future
costs and benefits of the alternatives, (2) prospects for resolving or reducing
the uncertainty with the passage of time, and (3) irreversibility of one or more

9. See William F. Hamilton, Managing Real Options, in Wharton on Managing

Emerging Technologies 271 (George S. Day, Paul J. Schoemaker, and Robert E.
Gunther eds., 2000).
10. Joseph Grundfest and Peter Huang have analyzed the litigation process in terms
of the plaintiff’s buying a series of options to continue litigating. Real Options and the
Economic Analysis of Litigation: A Preliminary Inquiry (1996).
66 problem solving, decision making, and professional judgment

of the alternatives, an extra value, an option value, properly attaches to the

reversible alternative(s).”11
Richard Posner uses the option concept to describe prudential responses to
the problem posed by climate change. He argues that we cannot afford to wait to
mitigate catastrophic climate change; making cuts in greenhouse gas emissions
now “can be thought of as purchasing an option to enable global warming to be
stopped or slowed at some future time at a lower cost.”12

3.2.4 Divergent Thinking and the Generation of Alternatives

In Chapter 1, we introduced the concepts of “divergent” and “convergent” think-
ing, derived from the work of Joy Paul Guilford.13 Divergent thinking involves
the generation of multiple potential solutions to a problem. Convergent thinking
posits or evaluates those solutions.
Experts often approach problems in a “convergent” manner—that is, they focus
attention and effort on a relatively narrow range of frames, objectives and options.
Convergent thinking is an essential cognitive skill. But optimal problem solving
also includes divergent thinking, which expands the range of options, perspectives,
and elements that are considered. Divergent thinking conceives; convergent think-
ing critiques. Divergent thinking envisions; convergent thinking troubleshoots,
fine tunes, selects, and implements. Both are required for good decision making.
In Section 1.4.2, we suggested that you can think of divergent thinking as two
lines emerging from a common point and then moving away from each other,
and of convergent thinking as two lines closing in from different directions on a
single point. We also suggested that a fully deliberative problem-solving process
combines elements of both. Divergent thinking makes its greatest contribution
when a problem is being framed, when interests and objectives are being identi-
fied, and when alternative solutions are being generated for later evaluation. It
also plays an important role in generating alternative hypotheses about the
causes of a problem. Divergent thinking enables us to conceptualize a problem
from a wide variety of perspectives. It is generative.
It is useful to think about divergent thinking in relation to schematic information
processing, which we introduced in Section 1.5.2. In a sense, divergent thinking
takes the thinker beyond the particular schematic framework triggered by his ini-
tial take on a problem and helps him think across a variety of interconnected
schemas. For this reason, it is useful then to think of divergent thinking as

11. Cass Sunstein, Worst-Case Scenarios 179 ff (Cambridge: Harvard University

Press 2007), quoting Anthony C. Fisher, Uncertainty, Irreversibility, and the
Timing of Climate Change Policy 9 (2001), http://www.pewclimate.com/docUploads/
timing_fisher.pdf, quoted in Worst-Case Scenarios.
12. Richard Posner, Catastrophe: Risk and Response 161–62 (New York: Oxford
University Press, 2005), quoted in Worst-Case Scenarios.
13. Joy Paul Guilford, Creativity, 15 American Psychologist 444–54 (1950).
 generating alternatives 67

“panschematic,” combining elements of different but somehow related schemas

in nonhabituated ways. This sort of thinking does not always come easily.

3.2.5 On Possibilities and Path Constraints

Without lifting your writing implement from the paper, draw four or fewer
straight lines connecting all nine dots in Figure 3.1 below.14

figure 3.1 the nine dot problem.

How did you go about trying to solve the problem? Most people try drawing
lines that pass through the dots and remain within the imaginary square they
constitute—and then they give up. They implicitly assume that the problem has
a path constraint that, in fact, it does not have—the rule, “Don’t go outside the
dots.” Who knows where we learned this tacit rule—perhaps from childhood
games of “connect the dots,” from the injunction to “color inside the lines,” or
from the fact that we are used to seeing objects in (more or less literal) frames.
In any event, it then becomes a limiting frame for solving the problem.15
For present purposes, we use the nine-dot problem to illustrate an important
point about the generation of alternative courses of action. It is often easier to
assume the presence of path constraints that limit a problem’s potential solutions
than it is to recognize their absence. Indeed, the very process of structuring a prob-
lem may create path constrains. We will say more about this later in the chapter.

3.3 creativity in professional problem solving

Judge Richard Posner has bemoaned that “only in law is ‘innovation’ a

pejorative.”16 Georgetown University law professor Carrie Menkel-Meadow has

14. We are not sure who created the Nine Dot Problem. We found it in James L.
Adams, Conceptual Blockbusting: A Guide to Better Ideas 24–25 (Reading MA: Basic
Books, 3rd ed., 1986).
15. You can find the “classic” solution (and several other wicked ones as well) in James
Adams’ book (supra) or on any number of Web sites.
16. U.S. v. McKinney, 919 F.2d 405, 421 (7th Cir. 1990) (Posner, J., concurring).
68 problem solving, decision making, and professional judgment

asked, “Is it possible to speak of legal creativity or is the phrase itself an

oxymoron?”17 Not only is it possible, but it is commonplace in the professions,
ranging from law to policy making to medicine and engineering.

3.3.1 An Opening Scenario

Recall that our intrepid problem-solving lawyer, Luis Trujillo, does volunteer
work for the Los Angeles Volunteer Legal Services Association (VLSA). The
Association’s director has sent him a memorandum outlining a potential pro
bono employment case. After having been led to believe that their employer was
doing extremely well financially, and that their jobs were secure, twenty wood-
workers at a custom furniture plant in nearby Anaheim had suddenly been
terminated after the plant’s owners decided to move manufacturing operations
to Vietnam to take advantage of cheaper labor costs there. The plant’s employ-
ees, who were not unionized, had come to work one day, been gathered into a
meeting room, and told that they no longer had jobs. They got no advance notice,
no severance pay . . . nothing. The manager gave them their final paychecks,
including unpaid wages and vacation pay, told them to gather up their personal
property, and had them escorted off the premises by security guards.18
Had the company employed more people the terminations would have been
covered by the federal WARN Act, which requires advance notice of pending
mass layoffs, or an equivalent amount in severance pay. Absent special rules of
this sort, California’s employment at-will doctrine, codified in California Labor
Code Section 2922, presents a seemingly insurmountable obstacle to recovery of
any sort of monetary relief—even relief in the nature of severance benefits.
Section 2922 is generally interpreted by human resources professionals and
employment lawyers to mean that, absent an enforceable contract providing
otherwise (e.g., an individual contract or a collective bargaining agreement),
employment can be terminated at any time, for any reason, so long as the reason
is not itself illegal. As a technical legal matter, employees are not legally entitled
to the customary “two weeks’ notice” of popular imagination. Lawyers at the
VLSA, who frequently advise newly terminated workers about the “at-will” rule,
see no way around the problem. But before giving up, the director wants Trujillo’s
firm to consider whether there might be some novel legal theory under which
the laid-off workers might obtain a remedy.
Trujillo, whose practice focuses on real estate, construction, and commercial
matters, and who knows very little about employment law, takes a quick look
at the relevant Labor Code section. It provides, in relevant part, that “[a]n

17. Carrie Menkel-Meadow, Aha? Is Creativity Possible in Legal Problem Solving and
Teachable in Legal Education?, 6 Harv. Negotiation L. Rev. 97–144, at 125 (2001).
18. This scenario is based on an analogous case litigated by one of the authors in
the early 1980s. Carson v. Atari, Inc., California Superior Court, County of Santa Clara,
No. 53073 (1982).
 generating alternatives 69

employment, having no specified term, may be terminated at the will of either

party on notice to the other.” Trujillo is puzzled. He knows that, under the Uniform
Commercial Code, “notice” of termination of a contract involving continuing or
successive performance of no specified duration is interpreted as meaning “rea-
sonable” notice. He wonders why the rule should be any different for employment
contracts. Checking the annotations, and doing a quick Westlaw search, he finds,
to his amazement, only two dusty old commercial contract cases—one from 1902
and the other from 1905—that interpret “notice” to mean “reasonable notice.” But
with those two old cases as precedent, and with arguments by analogy, the theory
might work. Or, it might not. But in any event, Trujillo’s problem sets the stage
for thinking about creativity in legal problem solving—and beyond.

3.3.2 Creativity: Working Definitions

Howard Gardner, whose work on multiple intelligences has profoundly influ-
enced education theory and practice, defines creativity in the following way:
People are creative when they can solve problems, create products or raise
issues in a domain in a way that is initially novel but is eventually accepted in
one or more cultural settings . . . [A] work is creative if it stands out at first in
terms of its novelty but ultimately comes to be accepted within a domain. The
acid test of creativity is simple: In the wake of a putatively creative work, has
the domain subsequently been changed?19
Harvard Business School professor Theresa Amabile’s definition sounds
many of the same themes, when she writes that a product is considered creative
to the extent that experts in a domain consider it “both a novel response and an
appropriate, useful, correct, or valuable response to an open-ended task.”20
Notice what these two definitions have in common: they both acknowledge
that creative solutions are not only novel, they are also effective. So, as Yale
psychologist Robert Sternberg observes, “Creativity is the ability to produce work
that is novel (that is, original, unexpected), high in quality, and appropriate (that
is, useful, meets path constraints).”21
Along the same lines, Jerome Bruner notes that a creative solution to a prob-
lem tends to produce in the knowing observer a sense of “effective surprise.”22

19. Howard Gardner, Intelligence Reframed: Multiple Intelligences for the

21st Century 116–17 (New York: Basic Books, 1999).
20. Beth A. Hennessey and Theresa M. Amabile, The Conditions of Creativity, in The
Nature of Creativity: Contemporary Psychological Perspectives 11, 14 (Robert J.
Sternberg ed., New York: Cambridge University Press, 1988).
21. Robert J. Sternberg, Wisdom, Intelligence, and Creativity Synthesized 89
(New York: Cambridge University Press, 2003).
22. Jerome S. Bruner, On Knowing: Essays for the Left Hand 18 (Cambridge,
MA: Harvard University Press, 1962).
70 problem solving, decision making, and professional judgment

What makes the solution effective is that it solves the problem. What makes the
solution surprising is that it differs from the stock alternative(s) spontaneously
triggered by the relevant problem schema. Most of the time, creative solutions
grow out of some established schema or combination of schemas. They involve
old things combined in new ways, or new things combined in old ways. A break
with the past makes the solution novel, but it is often continuity with the past
that makes it effective.
In this regard, notice that Luis Trujillo’s idea about a potential solution to the
Anaheim woodworkers’ problem did not exactly come out of “legal left field.”
The concept of “reasonable notice” that he proposed to introduce into employ-
ment law had deep roots in commercial law. This illustrates a number of impor-
tant points about how creativity operates in professional fields.

3.3.3 Creativity and Cross-Fertilization

During the 1950s and 1960s, writer Arthur Koestler23 studied humor, literature,
and biology to develop insights into the nature of creativity. He used the following
joke to illustrate one of those insights:
Question: Where does a general keep his armies?
Answer: In his sleevies.
What makes the joke funny, Koestler suggested, is essentially the same attri-
bute that makes an idea creative: the unexpected juxtaposition of two self-
consistent, but habitually incompatible frames of reference. In the joke, the
words “general” and “armies” evoke a military frame of reference, which brings
with it a set of associations centering on adult masculinity, seriousness, and all
things martial. Laughter results when that frame unexpectedly intersects with
another that involves juvenile talk, body parts, and clothing.
Creativity, in Koestler’s view, involves a process of “bisociation,” in which two
self-consistent but habitually dissociated frames of reference are brought
together, so that a construct from one is used productively in another. The idea
can be illustrated by Figure 3.2, in which each of two frames of reference is rep-
resented as a discrete plane. Ordinarily, the planes are dissociated and do not
intersect. The creative act brings them together. The line at which the two planes
then intersect is the zone of bisociation, where creative ideas emerge.
Luis Trujillo’s way of thinking about the Anaheim woodworkers’ problem
illustrates Koestler’s insight. Whether or not employment and commercial
contracts might seem to represent closely associated frames of reference, legal
doctrines have developed in very different ways in the two domains. The legal
requirement of reasonable notice of termination is as foreign to employment
contract law as it is taken for granted in commercial law. Relatively few lawyers

23. Arthur Koestler, The Act of Creation (London: Hutchinson, 1964).

 generating alternatives 71

figure 3.2 bisociation.

practice in both areas. For these lawyers, employment law and commercial law
represent discrete areas of specialization involving “habitually incompatible
frames of reference.” In such situations, the solution to a problem that might
seem intractable to an employment lawyer habituated to a doctrinal frame-
work in which “notice” has not been equated with “reasonable notice,” might
appear quite obvious to a lawyer habituated to doctrines in commercial law, in
which it has.
Many of the legal innovations we recognize as “creative” derived from this
kind of cross-fertilization between separate doctrinal or disciplinary domains.
Consider, for example, the warranty of habitability in landlord tenant law. The
concept of a “warranty” was originally unknown to the common law of real prop-
erty. In fact, the principle of caveat emptor stood in direct opposition to the notion
that an interest in real property might come with an implied warranty of any
kind.24 To solve a persistent social problem occasioned by informational asym-
metries between landlords and tenants, disparate levels of market sophistica-
tion, and other sources of bargaining inequality, lawyers eventually persuaded
courts to take the warranty concept from contract law and graft it onto landlord
tenant law.25
A similar process, through which a construct taken from one legal domain
is introduced to solve a problem in another, is illustrated by Charles Reich’s

24. See, e.g., Clifron v. Montague, 40 W. Va. 207 (1985); Walsh v. Schmidt, 206 Mass.
405 (1919); and Daly v. Wise, 132 N.Y. 306 (1892) (applying the principle of caveat
emptor).
25. See, e.g., Green v. Superior Court, 10 Cal. 3d 616 (1974); Javins v. First Nat’l Realty
Corp., 138 App. D.C. 369 (1970); Lemle v. Breedon, 51 Haw. 426 (1969) (abrogating the
doctrine of caveat emptor and imposing an implied warranty of habitability in landlord
tenant transactions).
72 problem solving, decision making, and professional judgment

landmark 1964 Yale Law Journal article, The New Property.26 In The New Property,
Reich urged that statutory entitlements to government benefits, such as Aid to
Families with Dependent Children or Medicaid, should be viewed as a species of
personal property, of which a person could not be deprived without due process
of law. Any first year law student who has read Goldberg v. Kelly in Civil Procedure
has seen the fruits of this innovation: Reich’s idea provided the basis for decades
of work protecting social welfare beneficiaries from the arbitrary, capricious, or
otherwise unfair denial or deprivation of public benefits.
Sometimes, a legal innovation can result from the cross-fertilization of
a particular legal subject matter area with an idea from an entirely different
discipline, such as economics. Consider in this regard the First Amendment
metaphor of the “marketplace of ideas,” emerging from Justice Oliver Wendell
Holmes’ dissent in Abrams v. United States. Although Holmes did not use this
phrase himself, Justice William Brennan later drew on the opinion to write:
[T]he ultimate good desired is better reached by free trade in ideas—that the
best test of truth is the power of the thought to get itself accepted in the
competition of the market . . . 27
Here, a principle drawing its persuasive appeal and normative force from the
domain of laissez faire economics was applied to the completely different domain
of free speech.
Law, public policy, and business contain many additional examples of innova-
tive solutions to seemingly intractable legal problems. Consider these, many
of which were self-consciously devised by lawyers, social advocates, and
entrepreneurs:28
• elimination of the privity requirement (that a manufacturer was only liable
to the purchaser of a product) in product liability;
• recovery of compensatory damages for emotional distress;
• the creation of a common law, and later of a constitutional “right to
privacy”;
• joint custody;
• same-sex marriage;
• condominiums, cooperatives, the corporation;
• pornography as a violation of civil rights;
• sexual harassment as a violation of sex discrimination laws;
• truth and reconciliation commissions;
• the joint venture;
• the “poison pill” as a response to a threatened hostile takeover; and
• the political action committee (PAC) and its successors.

26. Charles A. Reich, The New Property, 73 Yale L.J. 733, 785 (1964).
27. Abrams v. United States, 250 U.S. 616, 631 (Holmes, J. dissenting).
28. Menkel-Meadow, supra at 125.
 generating alternatives 73

As Carrie Menkel-Meadow notes, many of these ideas have come about by the
creative reading or misreading of legal words and concepts—by “expanding,
aggregating, disaggregating, rearranging and altering existing ideas and con-
cepts, borrowing or translating ideas from one area of law to another or from
other disciplines, and . . . by use of redesign or architecture of words and con-
cepts to build both new legal theories at the abstract level and new institutions at
the practical level.”29
In making this observation, Menkel-Meadow’s view harmonizes with that of
psychologist Mihály Csikszentmihalyi, who suggests that creativity often involves
crossing the boundaries of different domains, combining elements of one with
elements of another to create something new. “An intellectual problem,” he sug-
gests, “is not restricted to a particular domain. Indeed, some of the most creative
breakthroughs occur when an idea that works well in one domain gets grafted
onto another and revitalizes it.”30 In legal problem solving, as in many other
domains, innovation entails a kind of conceptual blending, a process in which the
problem solver combines two or more organizing conceptual frames to create a
distinctly new frame that does not merely hybridize its constituent elements, but
also generates its own emergent structure.31

3.3.4 Creativity, Expertise, and Critical Thinking

Creative individuals take problems that others see in one way and recast them as
something different and new.32 The preceding discussion suggests that bringing
the perspectives from a diversity of subject areas and ways of thinking about
problems—e.g., law, economics, science, and engineering—may be an effective
way to generate a range of innovative solutions. Approaching a problem from
interdisciplinary perspectives supports the exercise of three cognitive skills or
techniques that Theresa Amabile suggests lie at the core of creativity:
1. the ability to “break set,” that is, to diverge from one’s habitual way of think-
ing about a particular problem;
2. the ability to “break script,” or diverge from group norms about how a
particular type of problem “should” be solved; and
3. the ability to view problems from a variety of vantage points.

29. Id.
30. Mihaly Csikszentmihalyi, Creativity: Flow and the Psychology of Discovery
and Invention 88 (HarperPerennial 1996).
31. George H. Taylor, Law and Creativity, in Philosophy and American Law 81
(Francis J. Mootz III ed., New York: Cambridge University Press, 2009); Gilles Fauconnier
and Mark Turner, The Way We Think: Conceptual Blending and the Mind’s Hidden
Complexities (New York: Basic Books, 2002).
32. Id. at 255.
74 problem solving, decision making, and professional judgment

Taken together, these three points caution against increasing levels of

specialization in the teaching and practice of many professional disciplines.
At the same time, for creativity to be effective, it depends on expertise within
a domain. From the standpoint of the creator, innovation is necessarily linked to
what has come before. And from the perspective of its users, a product or prac-
tice must be rooted in a domain in order for it to be understood or recognized as
useful. Thus, the creative problem solver must have expertise in the domain in
which the problem arises, because even the most radically novel product or prac-
tice tends to be grounded in one or more domains.33 For example, in our hypo-
thetical scenario, it is the fact that the “reasonable notice” idea is rooted in
commercial law that made it intelligible to Luis Trujillo. Those same roots will
increase the odds that a court will find the argument persuasive.
So professional creativity often requires mastery of a domain in question—in
Trujillo’s case, a knowledge of substantive law combined with skill in legal
analysis and research. Paradoxically, however, the fact that Trujillo was not a
specialist in employment law, but rather in commercial law and real estate, may
have helped him identify potential solutions to the no-notice problem that
might have evaded an employment law specialist. Thus, while domain knowl-
edge is essential, specialization can sometimes actually constrain the generation
of alternative solutions to commonly encountered problems. Not only are spe-
cialists likely to approach a problem with a pre-scripted set of expert frames and
solutions, they may lack the exposure to concepts, tools, or principles from other
fields that might be adapted to solve problems arising in their own.

3.3.5 The Movement Between Convergent and Divergent Thinking

Isolated moments of creativity have little effect unless they are part of an overall
process that includes convergent as well as divergent phases. Innovative ideas
are seldom self-executing; their execution often requires critical, strategic, and
analytical thinking—all of which require convergent thought. For a good prob-
lem solver, creative and critical thinking are transient mindsets that are deployed
synchronically during a problem-solving process.34 People move between these
phases in the blink of an eye—or less.
Though they occur in convergent or evaluative phases as well,35 moments of
inspiration and creative insight are most common in divergent phases, which
permit breaking away from familiar patterns. The divergent phase is rich with

33. Sharon Balin, Achieving Extraordinary Ends: An Essay on Creativity

(Dordrecht: Kluwer Academic, 1988).
34. Raymond S. Nickerson, Dimensions of Thinking: A Critique, in Dimensions of
Thinking and Cognitive Instruction: Implications for Educational Reform
(Vol. 1) 495–509 (Beau Fly Jones and Loma Idol eds., Hillsdale, NJ: Erlbaum, 1990).
35. Joy P. Guilford Transformation: Abilities or Functions, 17 Journal of Creative
Behavior 75–86 (1983).
 generating alternatives 75

metaphors going back to Greek mythology, in which one was possessed by a

Muse during moments of creativity. Indeed, the creative work was seen as flow-
ing from the Muse herself. The creator was merely a conduit through which the
Muse’s creation passed, and he was always at peril of her abrupt withdrawal.36
Later, the Muses were joined by the “daemon,” a spirit that lived in, and could
suddenly depart from, the writer’s pen. Wrote Rudyard Kipling, “When your
Daemon is in charge, do not think consciously. Drift, wait, and obey.”37
William James both captured and exaggerated the dichotomy of the phases
when he contrasted everyday thinking with creative states. In everyday
thinking:
The force of habit, the grip of convention, hold us down on the Trivial Plane;
we are unaware of our bondage because the bonds are invisible, their restrain-
ing action below the level of awareness. They are the collective standards of
value, codes of behavior, matrices with built-in axioms which determine the
rules of the game, and make most of us run, most of the time, in the grooves
of habit—reducing us to the status of skilled automata which Behaviorism
proclaims to be the only condition of man.38
In a creative state:
Instead of thoughts of concrete things patiently following one another in a
beaten track of habitual suggestion, we have the most rarefied abstractions
and discriminations, the most unheard of combination of elements, the sub-
tlest associations of analogy; in a word, we seem suddenly introduced into a
seething cauldron of ideas, where everything is fizzling and bobbing about in
a state of bewildering activity, where partnerships can be joined or loosened
in an instant, treadmill routine is unknown, and the unexpected seems
only law.39
The Jamesian dichotomy continues in much of the popular literature, which
regards critical thinking as dour, constrained, and mechanical, while creativity is
unconstrained, fanciful, undisciplined, and free-spirited.40 Alex Osborn, the

36. See generally Penelope Murray, Genius: The History of an Idea (Oxford, UK:
Blackwell, 1989).
37. Rudyard Kipling, Working Tools (1937), in The Creative Process: A Symposium
161–63 (Brewster Ghislin ed., Berkeley, CA: The University of California Press, 1985).
38. William James, Talks to Teachers on Psychology 64 (New York: Henry Holt,
1908).
39. William James, Great Men, Great Thoughts, and the Environment, 46 Atlantic
Monthly 441–59 at 456 (1880).
40. See, for example, Roger von Oech, A Whack on the Side of the Head: How
You Can Be More Creative (New York: Warner Books, 1983); Paul Sloane, The
Leader’s Guide to Lateral Thinking Skills: Unlocking the Creativity and
Innovation in You and Your Team (2d ed. Philadelphia, PA: Kogan Page, 2006).
76 problem solving, decision making, and professional judgment

advertising executive who created the technique of “brainstorming,” contrasted

the “creative mind” as an idea generator that works through free association with
the “judicial mind,” which functions logically and acts as a filter. In Osborn’s
view, creative people are creative because they are able to “turn off” their judicial
minds, allowing free association to occur.41 These stereotypes capture the spirit
of convergent and divergent thinking, and they may be helpful in freeing people
to let their minds roam. But effective creativity requires both modes of think-
ing.
A similar dichotomy is often drawn between expertise and creativity, with the
view that expertise carries with it preconceptions about how a problem should be
approached that stifle “out-of-the-box” thinking.42 Too much knowledge is said to
impede creativity.43 Yet, as we have mentioned, effective creativity often calls for
domain-specific knowledge that mainly experts possess. The synthesis of these
dichotomies requires acknowledging the value of expertise at the same time as it
counsels against overspecialization in the teaching and practice of law, public
policy, or any other discipline.

3.3.6 A Confluence Approach to Creativity—And the Skills It Requires

Most contemporary researchers view creativity as requiring a confluence of
skills, mindsets, and abilities. The following taxonomy draws on the findings of
Theresa Amabile and Robert Sternberg:44
Intellectual skills. These include the abilities to generate ideas that are novel,
high in quality, and task appropriate; to recognize which of one’s ideas are worth
pursuing and which are not; to “think about thinking,” particularly in relation to
how problems are defined; and to persuade others in the field to accept one’s
ideas.
Self-governance skills. These include planning, self-monitoring, and the evalu-
ation of one’s thought processes and products.
Intrinsic task motivation.45 Intrinsic motivation conduces to creativity, while
extrinsic motivation may inhibit it.46 Sternberg writes that “people rarely do truly

41. Alex F. Osborn, Applied Imagination (New York: Scribner, 1953).

42. Edward deBono, New Think: The Use of Lateral Thinking in the Generation
of New Ideas 228 (New York: Basic, 1968).
43. Robert W. Weisberg, Creativity and Knowledge: A Challenge to Theories, in Handbook
of Creativity 226–50, 226 (R. J. Sternberg ed., New York: Cambridge University Press,
1999).
44. Robert J. Sternberg and Linda A. O’Hara, Creativity and Intelligence, in Handbook
of Creativity, supra at 255.
45. Teresa M. Amabile, Creativity in Context 83–127 (Boulder, CO: Westview, 1996).
46. For a summary of this research, see Mary Ann Collins and Teresa M. Amabile,
Motivation and Creativity, in Handbook of Creativity, supra at 297. Mihalyi Csikszentmihalyi,
The Domain of Creativity (1990), in Theories of creativity 190–214 (Mark A. Runco & Robert
S. Albert eds., Newbury Park, CA: Sage). Howard Gardner, Creating Minds: An Anatomy
 generating alternatives 77

creative work in an area unless they really love what they are doing and focus on
the work rather than the potential rewards.”47
Individual traits. Sternberg notes that some people find it easier than others
“to transit back and forth between conventional and unconventional thinking.”48
Certain personality traits may be positively correlated with creative production.49
These include a preference for moderate risk-seeking, willingness to tolerate
ambiguity, persistence, and self-efficacy, as well as an attraction to complexity
and aesthetic orientation. Bucking the crowd to “break script” requires self-
confidence, independence, and persistence, as well as communicative skill.

3.3.7 The Individual, the Domain, and the Field

We agree with those theorists and researchers who argue that the acid test of
creativity is whether, in the wake of a putatively creative work, the domain into
which it is introduced has been changed.50 For creative ideas to succeed in a
domain, they must be accepted by “gatekeepers,” who are empowered to make
decisions about what should or should not be included in the domain.51 In the
art world, these gatekeepers include critics, publishers, gallery owners, buyers,
and funders. In law, they include clients, other lawyers, administrative officials,
legislators, and judges. In public policy, they include government officials,
legislators, citizens, and nonprofit organizations.
Referring to our story of Trujillo and the woodworkers, imagine how oppos-
ing counsel or a skeptical judge might respond to Trujillo’s attempt to graft the
Uniform Commercial Code’s construction of “notice” as “reasonable notice”
onto California’s at-will employment rule. Anyone familiar with the rough and
tumble of state court motion practice can easily envision Trujillo’s opposing

of Creativity Seen Through the Lives of Freud, Einstein, Picasso, Stravinsky, Eliot,
Graham, and Gandhi (New York: Basic, 1993).
47. Robert J. Sternberg, Wisdom, Intelligence, and Creativity Synthesized,
108 (New York: Cambridge University Press, 2003).
48. Id. at xv. By the same token, Joy Paul Guilford posits that what we define as “think-
ing” comprises five different mental operations (cognition, memory, divergent thinking,
convergent thinking, and evaluation), which act on four categories of content (figurative,
symbolic, semantic, behavioral) to produce intellectual products. Creativity depends on
cognitive flexibility—the thinker’s ability to break apart different configurations of preex-
isting products and combine them in original ways. See generally Joy Paul Guilford, The
Nature of Human Intelligence (New York: McGraw-Hill, 1969).
49. For a summary of the relevant research, see Timothy I. Lubart, Creativity, in
Thinking and Problem Solving 290–332 (Robert J. Sternberg ed., 1994); Robert J.
Sternberg and Timothy I. Lubart, Defying the Crowd: Cultivating Creativity in
a Culture of Conformity (New York: Free Press, 1995).
50. Howard Gardner, Intelligence Reframed: Multiple Intelligences for the
21st Century 116-17 (New York: Basic Books, 1999).
51. Emma Policastro and Howard Gardner, From Case Studies to Robust Generalizations:
An Approach to the Study of Creativity, in Handbook of Creativity, supra at 213–25.
78 problem solving, decision making, and professional judgment

counsel rising to the podium in response to Trujillo’s proffered interpretation

and remarking with a wry smile, “Well, Your Honor, the plaintiffs’ legal theory
in this case is . . . how shall I say? . . . ‘creative.’” It would not be a compliment.
The point is that for a creative idea to spark innovation, others in the domain
or field must be persuaded of its merit. Thus, successful creators are also
entrepreneurs, who are able not only to generate new ideas but to sell them to
others.52
Along these lines, Robert Sternberg has set out what he calls a propulsion
theory of creative contributions,53 which describes how creative ideas can propel
a field forward. On a continuum of impact, these include ideas that:
• move a field forward in the same direction it was going, but extend it into
new subject matter areas. Consider in this regard the reach of neo-classical
economics into law during the 1980s;
• move a field forward in the same direction in which it was heading, but
stretch further and faster than others in the field were ready to go. If even-
tually accepted, such ideas are said to have been “ahead of their time;”
• re-orient a field, so that it moves in a different direction than previously;
• move a field backward to an earlier point and proceed forward from there
in a new direction;
• propose that a field has reached an impasse, and should move in a different
direction “from an entirely new point in the multidimensional space of
contributions.”54
Consider these two significant innovations, the first of which has become an
integral part of legal doctrine and the second not:
• Joseph Sax’s work developing public trust doctrine in the law of takings.55
Prior to Sax’s contribution, property rights were conceived as implicating
only activities occurring within the property’s physical boundaries.
Accordingly, limiting those activities was viewed as a “taking” of a right
appertaining to ownership. Sax, however, urged courts to view property

52. For more on this point, see Robert J. Sternberg, Wisdom, Intelligence, and
Creativity Synthesized (2003), supra at 21, 101–09.
53. Robert J. Sternberg, A Propulsion Model of Types of Creative Contributions, 3 Review
of General Psychology, 83–100 (1999); Robert J. Sternberg, James C. Kaufman, and
Jean E. Pretz, The Creativity Conundrum: A Propulsion Model of Kinds of
Creative Contributions (New York: Psychology Press, 2002).
54. Robert J. Sternberg, Wisdom, Intelligence, and Creativity Synthesised,
supraat 138.
55. Joseph L. Sax, The Public Trust Doctrine in Natural Resource Law: Effective Judicial
Intervention, 68 Mich. L. Rev. 471 (1970); Joseph Sax, Takings and the Police Power, 74
Yale L. J. 36 (1964).
 generating alternatives 79

rights quite differently, in a way that recognized the interconnectedness

among different uses on various pieces of property. “Once property is seen
as an interdependent network of competing uses, rather than as a number
of independent and isolated entities,” Sax observed, “[m]uch of what was
formerly deemed a taking is better seen as an exercise of the police power
in vindication of what shall be called ‘public rights.’”56
• The concept of “market share” or “enterprise” liability was first applied in
product liability cases involving the antimiscarriage drug diethylsilberstol
(DES), which can cause vaginal cancer in female children. Prior to the
California Supreme Court’s decision in Sindell v. Abbott Laboratories,57
plaintiffs who were harmed by DES but did not know which of numerous
drug manufacturers had produced the particular medication their mothers
took were unable to obtain a remedy in tort, as they could not establish
causation. In Sindell, the court rejected this requirement and held that a
plaintiff could recover from any of the major manufacturers in proportion
to their share of the market for the drug during the relevant time period.
The entire concept of actual causation, derived from the common law of
tort, was replaced with the concept of market share, taken from economics
and business. The “market share” approach was subsequently rejected by
the same court, and by others as well.58
Mihály Csikszentmihalyi and other scholars have proposed a “systems
approach” to creativity, which emphasizes the relationship among an individual,
a performance domain, and a field. The domain is a related set of bodies of
knowledge—in law, for example, constitutions, judicial doctrines, legislative
enactments, administrative regulations, and rules of procedure. A field is the
people who exercise influence or control over the domain,
A domain supplies an individual working within it with a set of interrelated
rules, patterns, and representations—the “old,” if you will. The creative individ-
ual draws on information in a domain and transforms or extends it, putting it to
some new use. Creativity occurs when the individual generates an idea that mod-
ifies these rules, patterns, and representations in some way, creating “the new.”
However, most such ideas are quickly forgotten, much in the same way in which
most genetic mutations are “forgotten” by evolution. Novel ideas are not
incorporated into the domain unless they are accepted by the gatekeepers who

56. Joseph L. Sax, Takings, Private Property, and Public Rights, 81 Yale L. J. 149, 151
(1971).
57. Sindell v. Abbott Laboratories, 163 Cal. Rptr. 132, 607 P.2d 924 (1980).
58. Brown v. Superior Court, 751 P.2d 470 (Cal. 1988); Hymowitz v. Eli Lilly and Co.,
539 N.E.2d 1069 (N.Y. 1989).
80 problem solving, decision making, and professional judgment

constitute the field. Once the new ideas are accepted, the domain preserves and
transmits them to other individuals.59
These relationships are illustrated by Figure 3.3.60

Culture

Domain

Selects Transmits
Novelty Information

Produces
Novelty

Field Individual
Stimulates
SOCIETY Novelty PERSONAL
BACKGROUND

figure 3.3 domain, field, and individual.

The systems approach emphasizes that creativity cannot be understood as

existing independently from the environment in which it operates. Creativity
necessarily involves an interaction between the individual, the domain in which
she works, the field that evaluates her work products, and the broader culture in
which all three are embedded.
Indeed, since whether a novel idea becomes a creative contribution depends,
in large measure, on its reception in the field into which it is injected, then one
can speak of creative fields as well as creative individuals. It is interesting to spec-
ulate when and why particular fields, such as law and public policy, are more or
less open to creative ideas.

3.4 problems in the future tense: scenario planning

So far, we have focused on solving problems that occur pretty much in the
present. But individuals, corporations, nonprofit organizations, and govern-
ments are often called upon to act today to solve problems that lie in the future.
As in the case of our farmer’s estate planning problem in Chapter 2, the future

59. Mihály Csikszentmihalyi, Society, Culture, and Person: A Systems View of Creativity
(1988), in The Nature of Creativity 325–39 (Robert J. Sternberg ed., New York:
Cambridge University Press , 1988).
60. Figure 3.3 is from Mihály Csikszentmihalyi, Implications of a Systems Perspective for
the Study of Creativity, in Handbook of Creativity, supra at 13–35, 315.
 generating alternatives 81

may be as near as the next generation. In addressing problems of energy and

climate change, the future may involve generations not yet born.
One major barrier to solving such problems creatively is people’s tendency to
anchor their thinking in the present and make only incremental adjustments as
they contemplate the future. One tool for overcoming this barrier is scenario
planning.61

3.4.1 What Are Scenarios? How Are They Used?

Scenarios are alternative descriptions of different possible futures that help deci-
sion makers consider the implications of these future possibilities for planning
and decision making today. Scenario planning helps decision makers expand
their conceptions of what is possible and imagine a broader range of futures
than might otherwise be considered.62 It is a creative tool rather than a forecast-
ing instrument. The scenario planning firm, Global Business Network (GBN), a
member of the Monitor Group, suggests that the technique can be used as:63
• a decision tool – “future proofing” a portfolio of activities and proposed
actions;
• a prioritization tool – determining where and how to allocate finite resources;
• a testing tool (“wind tunnel”) – using multiple “settings” to strengthen an
existing strategy, innovation, initiative, or priority;
• an oversight tool – adding perspective and insight to other planning
processes;
• an integrative tool – applying judgment to complexity for making sense of
the world;
• a generative tool – producing innovative ideas, programs, products, and
services;
• a timing tool – reacting appropriately (i.e., neither overreacting nor under-
reacting);
• a scanning tool – monitoring for deeper shifts in the external environment;
• a proactive tool – combating reactive demands; taking affirmative steps to
prepare for the future;
• a conversation tool – talking about difficult issues in a safe (hypothetical) way.
Scenario planning has been used by governments, businesses, and entire
industries. Here we consider their use by policy makers to enable safe and

61. Much of our discussion is based on the work of the Global Business Network
(GBN), a member of the Monitor Group.
62. Diana Scearce and Katherine Fulton, What If? The Art of Scenario
Thinking for Nonprofits 11 (Global Business Network, 2004).
63. Erik Smith, Chapter 5: Using a Scenario Approach: From Business to Regional Futures,
in Engaging the Future: Forecasts, Scenarios, Plans and Projects 85 (Cambridge:
Lincoln Institute of Land Policy, 2007).
82 problem solving, decision making, and professional judgment

productive dialogues among diverse stakeholders considering challenging or

contentious policy issues.64
For example, in 2002, the Great Valley Center (GVC),65 a nonprofit organiza-
tion that supports community sustainability and regional progress in California’s
Central Valley, used scenarios to help engage citizens in thinking about the
future of this huge region that runs from Redding in the north to Bakersfield in
the south. Working with GBN, the Great Valley Center engaged citizens and
experts in exploring how development issues, including traffic congestion, access
to quality education, land use policy, agriculture, and regional industry, might
evolve over the coming twenty-five years. Four scenarios were developed, as
shown in Figure 3.4.6667
Improving

Toxic gold New eden

A world in which the San Joaquin Valley A world in which clean air, a diverse
willingly becomes the storage center for economy, high-tech agriculture, and a
California’s toxic waste. The move creates highly educated workforce characterize
new jobs and provides much-needed funding the region. Revamped public education
for schools and hospitals, but at a price. New and transportation systems thrive. Fresno
health hazards crop up. Image and quality of booms,becoming California’s third
life attract or repel investment. largest city.
Negative

Positive
External influences

Rosa’s world A tale of two valleys

A world in which social disinvestment A world in which unemployment is high.
Social conditions

cripples the Valley. Difficulty attracting Ethnic and class divisions deepen,
“clean” industry and talented workers stunts threatening the region. Increased poverty
regional growth and limits its attractiveness. among the “have-nots” leads to decreased
The public education system crumbles, and investment. Local tax revenues decline as its
civil unrest flares. State funding dwindles. residents migrate elsewhere. Nevertheless, a
Air and water quality worsen. Latino middle class slowly emerges.

Worsening

figure 3.4 scenarios for california’s central valley.

Global Business Network, a member of the Monitor Group. 67

Materials illustrating these scenarios were broadly disseminated through

electronic and print media.68 Citizens and students in this conservative and
increasingly multiethnic region discussed a number of issues, including its

64. Smith 85–93.

65. Great Valley Center & Casey Family Programs examples have been adapted and
paraphrased from Scearce, et al.
66. Scearce 73–76.
67. Id. 76.
68. Hosley, David, President, Great Valley Center, personal interview, Nov. 21, 2008.
 generating alternatives 83

racial and economic divisions, with the goal of catalyzing regional change.69
Although it is too early to discern the effects of the exercise, GVC was sufficiently
optimistic to launch a new scenario initiative, “The Great Valley in 2020,” includ-
ing publication of a workbook for use by nonprofits and other organizations to
guide their scenario-planning efforts.70
Scenarios were also employed in a 2002 strategic planning effort by Casey
Family Programs, a nonprofit organization dedicated to providing and improv-
ing foster care. The organization was struggling to clarify its strategy and mis-
sion. A then-rising stock market had dramatically increased Casey’s endowment,
inspiring the organization to consider what success for the organization would
look like ten years down the road. Two critical uncertainties were considered:
• “How will U.S. society and government evolve over the next 10 years? Will
citizens become more/less actively involved in social issues and supportive
of an interventionist government?
• What role will technology play in fostering community?”7172
Scenario workshops yielded the results shown in Figure 3.5.

Interventionist
government

Big Mother Families First

A world characterized by recession, A world in which U.S. and global
decreased funds, international instability, economies thrive. Families and
and a strain on social safety nets. communities are high on the U.S.
Technology used not for community- government agenda. Neighborhoods are
building, but for surveillance. wired. The private sector and foundations
join the government in promoting social
transformation.
Undermines

community
community

Promotes

Role of technology

Concrete Cocoon Netizens.comm

Role of government

A world in which the techno-economy A world of individualistic affluence, in

is strong, but the digital divide is wide. which universal online access creates
Privatization means a decline in efficient online education and commerce.
government social spending. The “Netizens” create strong communities
Internet isolates even more than TV. that are highly differentiated and difficult
School violence is on the rise. to serve.

Minimalist
government

figure 3.5 scenarios relevant to foster care.

Global Business Network, a member of the Monitor Group. 72

69. Scearce 73–76.

70. Hosley, David, President, Great Valley Center, personal interview, Nov. 21, 2008.
71. Scearce 67.
72. Id.
84 problem solving, decision making, and professional judgment

As it turned out, key aspects of the Big Mother scenario did in fact unfold. This
scenario encompassed international geopolitical crises, increasing nationalism,
and an economic downturn—none of which was generally expected in 2002.
Chiemi Davis, Casey’s senior director of strategic planning and advocacy, reports
that “because of the scenario work . . . we now consider ourselves much better
prepared for the current economic and cultural realities.”73 GBN notes that the
Casey Foundation “saw the potential threat in a bleak scenario, like Big Mother,
and prepared accordingly.” 74

3.4.2 Why Use Scenarios?

Many decision makers assume that the future will closely resemble the present.
Scenarios challenge this assumption by imagining how the future world might
plausibly be dramatically different, with the goal of assessing various decision
alternatives in possible futures. The scenarios are not grounded principally in a
continuation of past trends or data. Rather, they involve plausible visions of the
ways that relevant uncertainties might evolve in the future. Most of these uncer-
tainties are driven primarily by forces beyond the control of the decision maker.
Diana Scearce and Katherine Fulton of GBN note that the goal of scenario
planning is not to predict the future, but rather “to arrive at a deeper understand-
ing of the world in which . . . you operate, and to use that understanding to
inform your strategy and improve your ability to make better decisions today and
in the future.”75 Rather than being predictions, projections, or forecasts,76
scenarios are more like simulations or war games, allowing decision makers to
imagine the outcome of their choices in differing worlds. By challenging a “busi-
ness-as-usual” vision of the future, scenarios can help counterbalance a number
of biases described in later chapters, including confirmation bias and other
judgmental errors, overconfidence, dogmatism, belief perseverance, and
groupthink.

3.4.3 Key Principles of Scenario Planning

Three key principles form the foundation of scenario planning:77
• Taking the long view: Rather than considering potential near-term changes,
scenario planning is often used to imagine how large-scale, systemic
changes may affect the future that strategic plans are designed to address,
five, ten or even twenty years from now.

73. Id.
74. Id.
75. Id. 7.
76. George A. Goens, Beyond Data: The World of Scenario Planning. Constructing Plausible
Plots in a World that Defies Linear Logic, April 2001, http://www.aasa.org/publications/
saarticledetail.cfm?ItemNumber=3761.
77. Smith 83.
 generating alternatives 85

• Thinking from the outside in: Rather than focusing on factors under our
control, scenario planning explores how events beyond our control might
affect the future.
• Including multiple perspectives: Scenario planning helps overcome myopic
visions of the future, and forces policy makers to think more openly about
what the future might bring.
In developing a scenario, planners must determine an appropriate time frame
based on when the most important effects of the decision will play out. The
relevant horizon for policy decisions may be as long as several decades. Planners
must also identify key future uncertainties, including external drivers—a process
that can be aided by considering emerging social, technical, economic, environ-
mental, and political (STEEP) trends.
After prioritizing key uncertainties, participants can begin to brainstorm future
scenarios. This is conceptually challenging. An almost infinite variety of scenarios
are possible—after all, almost anything could happen in the future. To remain
useful, the exercise must be bounded. A useful set of scenarios generally will
have quite different implications for decision making (though it would be note-
worthy and important if radically different scenarios led to identical decisions for
an organization or polity). In selecting scenarios for consideration, planners seek
out scenarios that are:
• Plausible. The scenario must be believable.
• Relevant to the key strategic issues and decisions at hand. If the scenario
would not cause a decision maker to act differently compared to another
scenario, there is little use in considering it.
• Challenging to today’s conventional wisdom. It should make one think
about different possibilities and options.
• Divergent from each other. Together, the scenarios should “stretch” the
thinking about the future environment, so that the decisions take account
of a wider range of issues.
• Balanced. It is useful to ensure that a group of scenarios strike a good
psychological balance between challenges and opportunities, risks, and
upside potential.78
Once a compelling set of scenarios is identified, it is useful to identify “lead-
ing indicators” to serve as warning signs that a given scenario is unfolding.

3.4.4 Drawbacks of Scenario Planning

Whatever their benefits, the vividness and detail of scenarios can contribute to
judgmental biases—particularly those related to the availability heuristic—that

78. Id. 84.

86 problem solving, decision making, and professional judgment

mislead participants about the probability that a particular scenario will occur.79
Scenario planners work to weave together vivid storylines; their storytelling often
entails “unpacking” a scenario into its component parts to illustrate how a given
set of events could converge to result in such an outcome. But vivid stories about
how a particular event might unfold make its occurrence seem more likely than
it may actually be.80 And the unpacking process can contribute to a logical error
in which participants’ assessments of the probability of events that are mutually
exclusive and collectively exhaustive add up to more than 100 percent.81 Scenario-
thinking thus may impair predictive accuracy by inspiring forecasters to “attach
excessive probabilities to too many possibilities.”82

3.5 barriers to creative problem solving — and

overcoming them

Earlier in this chapter, we used the nine-dot problem as a way to illustrate one
common obstacle to the generation problem solutions—imagining nonexistent
path constraints.
Most people are not able to solve the nine-dot problem at all. By contrast, we
do come up with solutions to most of our personal and professional problems—
though we have a tendency to adopt the first plausible alternative that comes to
mind, so long as it seems “good enough.” Such “satisficing” often makes sense:
We have many little problems to solve in the course of a day. However, even
when a particular situation warrants giving more thought to possible solutions,
we tend not to systematically consider a range of alternatives. We are constrained
by cognitive, social and emotional, practical, and personal factors. We consider a
number of these here.

3.5.1 Cognitive Process Constraints

Because problem schemas often include a stock solution, the triggering of a
particular problem schema also tends to trigger a pre-scripted solution, which,
because of its close cognitive association with the problem trigger, comes “first
to mind” and strikes us as particularly promising or apt. We tend to adopt this
default solution, and, if we consider alternatives at all, they tend to be incremental,
altering the stock solution in small ways rather than significantly broadening the
range of potential solutions we consider.

79. See Philip E. Tetlock, Expert Political Judgment: How Good Is It? How Can
We Know? The Limits of Open Mindedness 194, 196 (Princeton and Oxford: Princeton
University Press, 2005); Section 9.6, infra.
80. Daniel Reisberg, Cognition: Exploring the Science of the Mind 386–87
(New York: W.W. Norton Press, 2001).
81. Id. 197.
82. Id. 201.
 generating alternatives 87

Specialization often exacerbates this tendency. As the quotation from William

James cited in Section 3.3.5 suggests, specialization of any kind tends to create
“grooves of habit.” As an expert encounters more and more problems of a par-
ticular type, he may become increasingly susceptible to the rote application of a
narrow range of solutions. With increasing expertise in his own field and with-
out exposure to others, he may become less adept at borrowing concepts from
different frameworks and adapting them to solve a problem within his own
field.

3.5.2 Social Constraints

In Conceptual Blockbusting, Stanford University engineering Professor James
Adams presents readers with the problem illustrated in Figure 3.6.

4''

figure 3.6 the ping pong ball in the pipe.

Imagine that a steel pipe is imbedded in the concrete floor of a bare room. The
inside diameter of the tube is .06 inches larger than the diameter of a ping pong
ball, and, in fact, a ping pong ball is resting on the concrete floor at the bottom of
the pipe. You are one of a group of six people in the room, in which you also find
a carpenter’s hammer, a chisel, a box of cereal, a file, a wire coat hanger, a monkey
wrench, a light bulb, and one hundred feet of clothesline. You must find a way to
get the ping pong ball out of the pipe without damaging the ball, the tube, or the
floor.83

83. Adams, supra at 54.

88 problem solving, decision making, and professional judgment

Despite its obvious effectiveness, Adams observes, few people presented with
this problem propose to solve it by having the people present pee into the pipe,
thus floating the ping pong ball to the top of the tube where it can easily
be retrieved. Few if any people would even think of this solution, let alone
propose it, because, as Adams notes, urination is a private activity, at least in our
society.84
Though somewhat atypical, Adams’ scenario illustrates another class of
impediments to the generation of a robust set of alternative solutions—social
conformity effects. We don’t want to make fools of ourselves or offend the sensi-
bilities of others in our social environment—and proposing untested ideas about
alternative approaches to a problem can be risky. Social norms tend to develop
around recurring problems, providing a relatively small set of “acceptable” can-
didate solutions. Fearing disapproval, or perhaps not wanting to offend some-
one—often a person in authority—who has already proposed a plausible, more
“traditional” solution, may keep us from even imagining, let alone suggesting a
new approach. Indeed, pursuing solutions that violate social norms may be
perceived as somehow “cheating,” even if the solution does not violate any
explicit rules applicable to the problem itself.
We do not know whether lawyers and policy makers are more or less suscep-
tible to social conformity constraints than anyone else. But lawyers are often
viewed—by themselves as well as by others—as conservative, risk-averse, and
precedent-bound. Deviation from the “tried and true” can easily slip into devia-
tion from the applicable “standard of practice.” Such deviations may, at least
under some circumstances, subject a lawyer not only to ridicule, but also to
malpractice liability or bar discipline. In like fashion, the public accountability of
policy makers and the constrained routines of bureaucrats often promote risk-
averse behavior and inhibit taking creative approaches to solving problems.
While some professional school courses—probably more in public policy
than in law—encourage students to be creative problem solvers, much of the
core curriculum trains students to be more comfortable criticizing ideas than
creating them. The tendency to criticize ideas prematurely inhibits the genera-
tion of a rich and varied array of potential solutions or alternative courses of
action. Risk aversion, fear of breaking the rules or making a mistake, and
an implicit belief that there is one right answer to every problem all combine
with our general discomfort with an unsolved problem to tend us toward quick
convergence on one, or at most a small set, of stock, safe solutions.

3.5.3 Practical Constraints

Deliberative, process-conscious problem-solving procedures take time. This
holds as true for the generation of alternative solutions as it does for the framing
of the client’s problem, the identification of goals and objectives, and the

84. By the way, based on informal empirical observations, men are not more likely to
solve this problem than women.
 generating alternatives 89

resolution of causal uncertainty. Often, people do not have as much undistracted

time as they need to work on any particular problem, and this may prevent even
the best-intentioned problem solver from generating a range of options before
selecting a solution for implementation.
Even when time is available, our ability to generate alternative solutions may
be constrained by knowledge deficits. Helping a client generate alternative solu-
tions requires knowledge about the particular practical context—whether in
business, government, the nonprofit sector, or the client’s personal or profes-
sional life—in which the problem has arisen. A field’s institutionalized practices
often provide both the building blocks for potential solutions and obstacles
around which a successful solution must navigate. Professionals must under-
stand these if they are to generate practical, rather than merely imaginative,
alternatives.

3.5.4 Personal Constraints

As with many other of their “multiple intelligences,”85 individuals differ in their
capacities for instrumental creativity. The differences may be rooted in differing
needs for certainty and tolerance of ambiguity, in different levels of comfort
with offering ideas that may be rejected, and simply in the capacity for
imagination.
There is considerable speculation on how these capacities develop, or do not,
and on how they can be improved. Without rejecting any possibilities—from
psychological counseling to solving lots of puzzles and brainteasers—we suspect
that repeatedly being challenged to solve practical problems in law and related
domains is likely to improve professionals’ capacities in this respect, whatever
their differing starting points. It also seems apparent that particular environ-
ments are more or less conducive to the risk-taking that accompanies exposing
one’s ideas to others. It falls to research not yet conducted to determine whether
law and public policy students can develop better creative problem-solving skills
through transactional case studies (of the sort taught in business schools), simu-
lated exercises, and reflective clinical practice.

3.6 summary: generating better alternatives

It also remains unclear whether the various techniques for enhancing creativity
hawked by any number of “how-to” books can systematically be employed to
improve the problem-solving abilities of lawyers and policy makers. But in sum-
mary, we offer a number of helpful ideas, gleaned from the trade literature and
the work of academic researchers.
Perhaps most significantly, if necessity is the mother of invention, overwork
makes for effective birth control, creating the ideational equivalent of China’s

85. See Howard Gardner, Intelligence Reframed, supra.

90 problem solving, decision making, and professional judgment

“one child” policy. The generation of a broad range of potential solutions to any
given problem appears to benefit from:
• time to ruminate and let problems percolate in the unconscious mind;
• permission to sit for a while with an unsolved problem. In the quest for
creativity, idleness may prove a virtue.
• awareness of the constraints that schematic expectancies and other pre-
scripted responses implicitly impose on our thinking;
• the ability to go beyond the frames and schemas automatically triggered by
a problem of a particular sort; and
• willingness to at least temporarily eschew evaluation, reserve judgment,
and postpone evaluative scrutiny of potential alternative solutions.
Here are a few practical guidelines for applying these general principles to
specific problem-solving situations:
1. Focus individually on each of the objectives identified earlier in the
problem-solving process and, with respect to each one, ask, “How could it
be achieved?”
2. Generate at least three ideas for each identified objective before moving on
to the next.
3. Question the constraints you have imposed on possible solutions to the
problem.
4. Think about analogous problems in other fields or disciplines and ask
whether there is some device from those fields or disciplines that might be
adapted to solve the problem.
5. Consult with people with knowledge sets other than your own to get their
ideas about potential solutions.
6. Generate a number of potential solutions before critiquing or choosing any
one of them.
We close the chapter with a problem, which you can use to stretch your think-
ing about the topics we have covered here.
Luis Trujillo and his client, Jack Serrano, do not yet have enough information
to have any confidence in the causes of the problems at Terra Nueva. Yet, they
need to act quickly. If the newspaper report is accurate, a class action suit has
already been filed against Serrano. Remember that defending the suit is not
necessarily Serrano’s only interest. He may also be concerned with his reputa-
tion; with his relationships with government regulators, financial backers, pres-
ent and potential business partners, current tenants, and prospective tenants;
with dealing with his and his family’s psychological stress; and with “doing
right” by anyone who has actually been injured by materials used in the homes.
Generate a list of actions that Serrano might take within the next days or
weeks to address these or any other interests you think he might have.
4. choosing among alternatives

Having thought through a variety of alternative strategies that may achieve your
objectives, it’s time to make the decision. Sometimes, the process up to this
point will lead inexorably to a single choice. Often, however, you may be faced
with competing interests or objectives or competing strategies for achieving
them—as Jack Serrano is, for example, in trying to decide how to respond to his
tenants’ complaints.

4.1 specifying the consequences of alternatives

The first step toward choosing among several plausible alternatives is to be clear
about the consequences of choosing each one. The alternatives facing Serrano
are fraught with uncertainties, including how others will react to any actions he
takes. Before tackling such complexities, consider how you might assess the
consequences of alternatives in a simpler case—perhaps one closer to your
own experience. Suppose that you are considering renting an apartment, and
that you have four major criteria:
• Cost. The lower the rent, the better.
• Size. You need at least two rooms, and three (one of which would serve as
a guest room) would be ideal.
• Proximity to the campus. The closer the better.
• Neighborhood. You’d like to live in a safe, attractive neighborhood.
You have now seen a number of apartments that seem plausible candidates.
How might you describe them in terms of these criteria?
Cost is the simplest and most readily quantifiable. When you think about
size, you realize it’s not just the number of rooms, but their dimensions. The
guest room is really tiny—all right for an occasional out-of-town friend, but your
mother might give you a hard time when she visits. (On the other hand, that may
keep her visits reasonably short.)
With respect to proximity, the apartment is not really within walking distance
of the campus, but there’s a bus stop just in front of your door; also it seems like
a fairly short bike ride. But does the bus go directly to campus, or do you have to
transfer? Would biking require you to ride on a busy commercial street with no
bike lanes? The neighborhood looks pretty nice during the daytime, but you
haven’t been there at night when the jazz nightclub across the street is open.
Also, now that you think about it, you haven’t seen a supermarket anywhere
nearby.
92 problem solving, decision making, and professional judgment

This little scenario suggests a number of general guidelines for specifying the
consequences of a potential decision.1
• Describe the consequences of the highest-ranked alternatives in terms of
each of your interests or objectives.
• Wherever possible, describe the consequences quantitatively; this will facil-
itate the comparison with other (existing or possible) alternatives. Make
sure that the quantity captures all the dimensions of an interest (e.g., size
of the rooms as well as number; commute time to campus as well as
distance.)
• Don’t disregard consequences that can’t be quantified, such as the quality
of the neighborhood or the safety of the bike ride. As Einstein famously
said: “Not everything that counts can be counted, and not everything that
can be counted counts.”
• Gather data about the consequences. If you’ve only seen the neighbor-
hood during the daytime or on weekends, go there on an evening or
weekday.
• Pretend that you have already chosen an alternative, and imagine a day
in your life from morning to night. Engage in such “mental simulation”
with a number of alternatives, not just your favorite. Otherwise, you may
become too committed to one choice and have difficulty imagining the
others.
• Be open to new interests or objectives coming to mind. Even when you
follow a deliberative model, the processes are recursive, with subsequent
steps reminding you of lacunae in earlier ones.
After engaging in the process with all plausible alternatives, one of them
may so obviously dominate others—that is, be better in some respects and at
least as good in all others—that the decision is inescapable. (Also, an alterna-
tive may be obviously dominated by others, so you can eliminate it from consid-
eration.) The interesting decisions, though, are ones where there is no one
obvious choice.
Note that this little exercise makes at least two implicit simplifying assump-
tions: (1) that there are no significant risks or uncertainties in how the future will
unfold, and (2) that you are pretty good at predicting how you will feel about the
decision once you actually begin living with it. In fact, real-world decision making
tends to be more complex in both of these respects. But we’ll leave that for part
3 of the book.

1. We are indebted to John Hammond, Ralph, Keeney, and Howard Raiffa, Smart
Choices: A Practical Guide to Making Better Decisions (1999) for some of the
guidelines.
 choosing among alternatives 93

4.2 the spectrum of decision-making processes 2

Underlying every decision is an implicit metadecision of how much effort to

devote to the decision and what processes to employ for making it. The objective
of the metadecision is to minimize the sum of the costs of making the decision
and the costs of suboptimal outcomes. Cass Sunstein and Edna Ullman-Margalit
catalog these decision-making costs, which include time, money, unpopularity,
anxiety, boredom, agitation, anticipated ex post regret or remorse, feelings of
responsibility for harm done to self or others, injury to self-perception, guilt, and
shame.3
Sunstein and Ullman-Margalit also catalog a mixture of decision processes, to
which we add a few of our own:
Intuition. Not only are most of the myriad decisions that we make every day
done intuitively, but intuition continues to play a role even when we attend to the
decision-making process. Intuition often gives us an affective or emotional
response to an alternative we are considering. Negative affect, including just an
uneasy feeling, provides a warning of a problem with the choice, even when the
problem is hidden from our consciousness. To fail to attend to this signal is to
invite a poor decision.
Intuition is a double-edged sword, however, and it can mislead. Suppose that
you are considering renting an apartment. One apartment seems pretty nice,
albeit expensive, and you are quite taken with the real estate agent who recom-
mended it to you. Though your feelings about the realtor may unconsciously
spill over to the apartment, they are unlikely to improve your satisfaction with
the apartment in the long run. Indeed, every salesperson worth her salt will use
techniques to make you feel positive toward her and hence toward the transac-
tional experience.4 We’ll have a lot more to say about the interplay of intuition
and deliberation in later chapters.5
Heuristics. In Section 1.6.1 we described the two-systems model of judgment
and decision making. System I heuristics, or mental shortcuts, allow people to
make myriad judgments about probability, causation, and other matters with
little or no cognitive effort. Although heuristics often lead to good outcomes,
they can also mislead. For example, in the Terra Nueva problem, a foul-smelling

2. We are indebted to Cass Sunstein and Edna Ullman-Margalit, Second-Order

Decisions, 110 Ethic 5–11 (October 1999) for stimulating our thinking about these issues
and categorizing decision-making strategies.
3. Id. 12.
4. See Robert Cialdini, Influence: Science and Practice ch. 16 (Boston: Pearson/
Allyn & Bacon, 2000).
5. For some general readings on the subject, see Better Than Conscious? Decision
Making, the Human Mind, and Implications for Institutions (Christoph Engel and
Wolf Singer eds., 2008); H. Plessner et al., Intuition in Judgment and Decision
Making (2008).
94 problem solving, decision making, and professional judgment

odor might evoke images or feelings of illness. Whether or not this is generally
accurate, it would be best to check it out with some conscious System II analysis
before making an important decision based on the linkage.
Routines. Routines are patterns of behavior that one follows regularly and
without thought: for example, swallowing vitamin pills with one’s morning
coffee or taking a particular route to work. Most bureaucracies and other organi-
zations depend on routinized ways of doing things—new employees are taught
to do it that way because “that’s the way it is done.” A routine may be the result
of past deliberation, but present actors do not necessarily advert to its rationale
and may not even know what it is.
Rules. A rule specifies the outcome of a decision given the existence of certain
facts. Rules have low decision-making costs and constrain discretion. Many gov-
ernment regulations take the form of rules: “Maximum speed limit, 65 miles per
hour” and “No bicycle riding in the playground.” Rules typically are both over-
and under-inclusive and therefore entail some individually suboptimal deci-
sions—traffic conditions may be such on a particular stretch of highway that
75 miles per hour is perfectly safe, but don’t waste your time arguing this to the
highway patrol officer or the judge. Self-imposed rules—for example, “never
have second helpings”—are often designed to avoid succumbing to temptations.
Even though a rule does not call for deliberation by the immediate decision
maker, it does require that someone has deliberated in advance to create
the rule.

Checklists

Checklists, of the sort used by airplane pilots, are similar to rules or presump-
tions. In 1935, the so-called “Flying Fortress” often crashed on take-off until the
military realized that this four-engine aircraft was so complex that pilots needed
to rely on a checklist rather than their intuition. Recent studies suggest that using
a checklist can make a huge difference in critical medical care. For example, by
checking off five simple steps in putting catheters into patients—from washing
hands with soap to putting a sterile dressing over the site—physicians and nurses
could dramatically reduce the number of infections. Atul Gawande, Annals of
Medicine: The Checklist, NEW YORKER, December 10, 2007.

Presumptions and rules of thumb. A presumption is a rule that can be rebutted

under special circumstances that cannot be specified in advance; in effect, it
shifts the burden of proof, requiring you to justify—to yourself or others—why
you are deviating from the rule. In personal life, you might have a presumption
against eating second helpings, which could be overcome if it would be rude to
 choosing among alternatives 95

your host. A common example from law practice is: “Never ask a witness a ques-
tion on cross-examination unless you know the answer yourself.” An environ-
mental regulation might prohibit discharging more than a certain amount of a
pollutant unless the business can prove that it isn’t feasible to keep the discharge
below the limit.
Presumptions of this sort permit quick, nondiscretionary decisions. They can
also allow for a gradual learning curve. People new to an organization or field of

Business Rules of Thumb

In the course of researching this chapter, we came across these examples from a
book on business rules of thumb, modestly subtitled, Words To Live By And Learn
From That Touch Every Aspect Of Your Career:

• Throw a tantrum within the first ten minutes of the negotiating (from a
sports consultant).
• Never hire a salesman you’d want your daughter to marry.
• Don’t order drinks served with a paper umbrella.

SETH GODIN AND CHIP CONLEY, BUSINESS RULES OF THUMB (Warner Books 1987).

practice may adhere to them strictly and loosen the constraints as they develop
more experience.
Like a rule, a presumption may be the result of someone’s deliberation before-
hand. Unlike a rule, a presumption calls for a degree of deliberation at the time
of its application—enough deliberation to determine whether the circumstances
justify rebutting the presumption.
Standards. A standard leaves it to the individual decision maker and those
who will review her judgment to apply a norm such as “reasonableness,” “exces-
siveness,” or “competency” to a particular situation. For example, in addition to
setting particular speed limits, California law provides that “No person shall
drive a vehicle upon a highway at a speed greater than is reasonable or prudent
having due regard for weather, visibility, the traffic on, and the surface and width
of, the highway.”6 Depending on whether the norms of reasonableness, etc., are
well defined by practice and precedent, a standard may require more or less
deliberation.
Small steps. A decision maker may proceed by taking small, reversible steps
and observing their consequences, rather than making a big decision in one fell
swoop. This is akin to the use of time-buying options, discussed in Section 3.2.3.

6. People v. Huffman, 88 Cal. App. 4th Supp. 1, 106 Cal. Rptr.2d 820 (2000).
96 problem solving, decision making, and professional judgment

Sunstein and Ullman-Margalit give the examples of Jane living with Robert
before deciding whether to marry him, and observing of the Anglo-American
common law tradition of deciding cases on as narrow a ground as possible.7
Delegation. Rather than make the decision yourself, you can delegate the
responsibility to another individual or agency. A client may ask her lawyer to
make a decision on her behalf; an administrative official may be required to
delegate a decision to an agency.
Random processes. In addition to providing a fair way to resolve some inter-
personal disputes, a random process like flipping a coin can quickly resolve
intrapersonal disputes when you are faced with two equally attractive options
and the choice doesn’t justify further analysis.

4.3 deliberative, noncompensatory decision-making strategies

Under a noncompensatory decision-making strategy, a strength in one attribute

(e.g., the rent is very low) does not compensate for a weakness in another (e.g.,
the apartment has only two rooms).
Elimination by aspects (EBA) is a paradigmatic noncompensatory procedure.
The decision maker identifies relevant attributes, assigning each a minimum
threshold. She then assesses the alternatives in terms of the first attribute, elim-
inating alternatives that do not meet the threshold. If several alternatives remain,
she evaluates them in terms of the next attribute, and so on. Ideally, but not
inevitably in practice, the attributes are ordered in terms of their importance.
EBA is cognitively pretty easy, and people make many decisions ranging from
choosing restaurants and vacation destinations to buying cars using variations of
EBA—often “quick and dirty” approximations.
Sometimes the threshold will follow inexorably from your values or needs—
for example, a vegetarian would eliminate meat dishes in a restaurant before
choosing among options for her entrée. But many thresholds are more matters
of preference—for example, fresh rather than farm-raised salmon. If no avail-
able alternative meets your thresholds, you would reduce some of them and go
through the procedure again.
What happens when several alternatives meet the threshold depends on your
rule for stopping the search. For example, one of us (Paul) has a simple strategy
for shopping for a suit. He has two attributes—whether he likes it and whether
it is available in his size. When he sees the first suit that meets these thresholds
he stops the search. To be sure, there may be even nicer suits in the store (let
alone the wonderful suits in other stores), but all decision making involves a
trade-off between the costs of decision making, including time and cognitive
effort on the one hand and accuracy on the other—and he does not like shopping
for clothes.

7. Sunstein and Ullman-Margalit, supra.

 choosing among alternatives 97

To understand how EBA works and see its limitations, let’s consider the hous-
ing example in greater detail. Suppose that you are considering thirteen rental
apartments, and that you rank the four major attributes in this order: cost, size,
proximity to the campus, and neighborhood, with cost being the most impor-
tant. You then determine a specific acceptability threshold for each attribute. For
example, you cannot spend more than $1500 a month, and you need at least
600 square feet of space. Under EBA, you eliminate, say, eight apartments that
have monthly rents greater than $1500. Of the five apartments remaining, you
eliminate two studios with 575 square feet of floor space.
Each of the three remaining apartments costs about $1300 a month and has
about 800 square feet of space. One is located in the more cosmopolitan down-
town center—an eight-minute commute to campus; the two others are virtually
adjacent to one another in a smaller, more suburban neighborhood north of the
town—also an eight-minute commute to campus. Though the atmospheres and
amenities of the two neighborhoods differ, each is pleasant, safe, and attractive
in its own way.
EBA helped whittle thirteen choices down to three, all of which are satisfac-
tory. But notice that the procedure can lead to a suboptimal outcome if the thresh-
olds for the first several attributes were based on preferences rather than absolute
necessities. Perhaps you eliminated an apartment that was a bit too expensive
but that was so great in all other respects that it would have justified stretching
your budget. This is an inherent limitation of any noncompensatory procedure.
Along similar lines, EBA may allow others to manipulate your rankings for
their own interests. Amos Tversky provides this parodic example from an actual
television commercial:
“There are more than two dozen companies in the San Francisco area which
offer training in computer programming.” The announcer puts some two
dozen eggs and one walnut on the table to represent the alternatives, and
continues: “Let us examine the facts. How many of these schools have on-line
computer facilities for training?” The announcer removes several eggs. “How
many of these schools have placement services that would help you find a
job?” The announcer removes some more eggs. “How many of these schools
are approved for veteran’s benefits?” This continues until the walnut alone
remains. The announcer cracks the nutshell, which reveals the name of the
company and concludes: “That’s all you need to know in a nutshell.”8
Suppose that you can attend the classes in person and would prefer to do so,
that you are seeking training to advance in your current job, that you are not
a veteran, etc. Perhaps the best instructors teach at places that lack all of the
attributes touted by the commercial. In such a case, the EBA methodology out-
lined above would lead you far astray from your ideal outcome. EBA is only

8. Amos Tversky, Elimination by Aspects: A Theory of Choice, 79 Psychological Review

281 (1972).
98 problem solving, decision making, and professional judgment

useful if the aspects to be assessed are important to your decision-making pro-

cess and if they are ordered in terms of their importance.910

Fast and Frugal Heuristics

Gerd Gigerenzer and his colleagues argue that, in making factual judgments,
people use “fast and frugal heuristics” similar to EBA. For example, in estimating
the relative size of two cities, if we recognize the name of one but not the other,
we will employ the “recognition heuristic” to conclude that if we recognize one
city but not another, the recognized city is larger. Gigerenzer asserts that, in gen-
eral, we tend to make judgments based on a single cue and only look to another
as a tie-breaker if it doesn’t produce a result. More generally, under the take the
best heuristic, the decision maker tries cues in order, one at a time, searching for
a cue that discriminates between the two objects in question. For example, “when
inferring two professors’ salaries, the [professorial] rank cue might be tried first.
If both professors are of the same rank (say both associate professors), then the
gender cue might be tried. If one of the professors is a woman and the other is a
man, then we might say that the gender cue ‘discriminates.’ Once a discriminat-
ing cue is found, it serves as the basis for an inference, and all other cues are
ignored. For instance, if gender discriminates between two professors, the infer-
ence is made that the male earns a higher salary, and no other information about
years of experience or highest degree earned is considered.”9
There is little doubt that people, in fact, often use simplifying decision proce-
dures in making factual judgments as well as decisions. But there is serious
controversy whether we employ only one cue at a time and, more fundamentally,
whether fast and frugal heuristics tend to produce accurate judgments.10

4.4 compensatory procedures

As we mentioned, noncompensatory decision procedures such as EBA can yield

outcomes that seem somewhat arbitrary. For example, if your highest priority in
voting for a candidate for public office is her position on gun control, then a
noncompensatory procedure may eliminate candidates who have the “wrong”

9. Gerd Gigerenzer, Peter M. Todd, and the ABC Research Group, Simple
Heuristics that Make Us Smart 99 (New York: Oxford University Press, 1999).
10. See, e.g., Daniel M. Oppenheimer, Not So Fast! (and Not So Frugal!): Rethinking the
Recognition Heuristic, 90 Cognition B1 (2003); Mark G. Kelman, The Heuristics
Debate: Its Nature and Its Implications for Law and Policy (Oxford University Press,
forthcoming 2010).
 choosing among alternatives 99

position on this issue but who hold positions that you agree with on many
other issues. Also, a noncompensatory procedure does not account well for
your intensity of preferences. Unless gun control matters to you more than all
other issues together, you may not be choosing the candidate who best reflects
your interests.
In fact, most decisions invite making trade-offs among competing interests
or objectives. For example, you might trade off the candidate’s position on gun
control with her position on other social and economic issues that you deem
important. In choosing the apartment, you might be willing to trade off the
rental price for proximity. One can approach making trade-offs by (1) giving
reasons for one’s choices or (2) assigning quantitative values to different attri-
butes of the things being chosen.

4.4.1 Reason-Based Decision Strategies

Consider Charles Darwin’s approach to trade-offs in these notes on a scrap of
paper with the heading, “This is the question,” as shown in Table 4.1.11

table 4.1 darwin’s pros and cons of marriage

MARRY Not MARRY

Children—(if it please God)—constant No children (no second life) no one to

companion (friend in old age) who will be care for one in old age. . . . Freedom to
interested in one, object to be beloved and go where one liked—Choice of Society
played with—better than a dog anyhow— and little of it. Conversation of clever
Home, and someone to take care of men at clubs—Not forced to visit
house—Charms of music and female relatives, and to bend in every trifle—to
chit-chat. These things are good for one’s have the expense and anxiety of
health. Forced to visit and receive relations children—perhaps quarreling. Loss of
but terrible loss of time. My God, it is time—cannot read in the evenings—
intolerable to think of spending one’s whole fatness and idleness—anxiety and
life, like a neuter bee, working, working and responsibility—less money for books,
nothing after all.—No, no won’t do.— etc.—if many children, forced to gain
Imagine living all one’s day solitarily in a one’s bread. (But then it is very bad for
smoky dirty London House.—Only picture one’s health to work too much.)Perhaps
to yourself a nice soft wife on a sofa with my wife won’t like London; then the
good fire, and books and music perhaps— sentence is banishment and degradation
compare this vision with the dingy reality with indolent idle fool—
of Grt. Marlboro’ St.

11. Charles Darwin, The Autobiography of Charles Darwin, 1809–1882, 232–33

(Nora Barlow ed., New York: Norton, 1969). Quoted in Gerd Gigerenzer, Peter M.
Todd, and the ABC Research Group, supra, at 7.
100 problem solving, decision making, and professional judgment

At the bottom of the first column, Darwin wrote: “Marry—Marry—Marry

Q.E.D.,” though he also noted, “There is many a happy slave.”12

The Importance that We (Sometimes) Attribute to Reasons

You are about to take the bar examination. If you pass the exam, you will celebrate
with a week’s vacation in Hawaii. If you fail the exam, you will console yourself
with a week’s vacation in Hawaii. Your travel agent says that you can lock in low
airline fares now, and that the fares are sure to rise if you wait. Will you book the
vacation now, or wait until you receive the results of the exam? In a study by Eldar
Shafir, Itamar Simonson, and Amos Tversky, most students chose to pay the
additional cost and wait. They wanted to choose the vacation for the “right”
reason, which couldn’t be known in advance.13

Darwin was applying a reason-based compensatory decision-making

strategy—giving reasons for the pros and cons of the decision without explicitly
assigning any weights to the criteria. In a letter to Joseph Priestly, Benjamin
Franklin develops a pro-and-con approach into a somewhat more quantitative
strategy:14
When . . . difficult cases occur, they are difficult, chiefly because while we have
them under consideration, all the reasons pro and con are not present to the
mind at the same time; but sometimes some set present themselves, and at
other times another, the first being out of sight. Hence the various purposes
or inclinations that alternately prevail, and the uncertainty that perplexes us.
To get over this, my way is to divide half a sheet of paper by a line into two
columns; writing over the one pro, and over the other con. Then during three
or four days’ consideration, I put down under the different heads short hints

12. Note, by the way, the role of images, scenarios, and affect in Darwin’s list-making.
Recall Antonio Damasio’s description of the role of affect in decision making in Section
1.5.3. As it turned out, Darwin married his cousin, Emma Wedgewood, in 1839, and
they had ten children, many of them notable in their own right. Twenty years later, he
published On the Origin of Species by Means of Natural Selection.
13. Eldar Shafir, Itamar Simonson, and Amos Tversky, Reason-Based Choice, 49
Cognition (1993).
14. Benjamin Franklin to Joseph Priestly, September 19, 1772. At the time, Priestly
was happily preaching and conducting scientific experiments. When offered the position
of librarian to Lord Shelburne, he asked Franklin for advice. Rather than advise on the
merits, Franklin offered a method of decision making.
 choosing among alternatives 101

of the different motives, that at different times occur to me, for or against the
measure.
When I have thus got them all together in one view, I endeavor to estimate
their respective weights; and where I find two, one on each side, that seem
equal, I strike them both out. If I find a reason pro equal to two reasons con,
I strike out the three. If I judge some two reasons con, equal to some three
reasons pro, I strike out the five; and thus proceeding I find at length where
the balance lies; and if, after a day or two of further consideration, nothing
new that is of importance occurs on either side, I come to a determination
accordingly.
And, though the weight of reasons cannot be taken with the precision of
algebraic quantities, yet when each is thus considered, separately and com-
paratively, and the whole lies before me, I think I can judge better, and am
less liable to make a rash step, and in fact I have found great advantage from
this kind of equation, in what may be called moral or prudential algebra.
Franklin’s procedure of canceling the pros and cons of a decision is a rough
approximation of quantifying the strength and importance of the attributes
involved in the decision. Franklin seemed to have in mind a yes-no decision
about whether to take a particular action. But you could extend his procedure to
deciding among multiple alternatives, such as the choice among apartments.
If you are choosing between two apartments, write down the pros and cons of
each and do the cancellation. If you are choosing among more than two, you
might begin with what appear to be relatively strong and weak candidates, then
see if you can eliminate the weak one, and put another in its place.15
Almost surely, Darwin and Franklin did not think their processes would
determine their decisions rather than inform their intuitions. But this is by
no means a useless project. Indeed, as we suggest below, even when one uses
a more quantitative process, it is always useful to perform an intuitive “reality
check” at the end.

4.4.2 Quantitative Decision Strategies

“Value-focused thinking involves starting at the best and working to make it a reality.
Alternative-focused thinking is starting with what is readily available and taking the best
of the lot.”16

15. In SMART CHOICES, supra, Hammond, Keeney, and Raiffa describe how one can
extend this procedure to handle multiple alternatives simultaneously.
16. Ralph L. Keeney, Value-Focused Thinking: A Path to Creative Decisionmaking
6 (Cambridge: Harvard University Press, 1992).
102 problem solving, decision making, and professional judgment

Consider these two policy decisions:17

1. A mayor must decide whether to approve a major new electric power gen-
erating station and, if so, where to locate it. There is need for more electric-
ity, but a new station would worsen the city’s air quality, and the mayor is
concerned with the effects that his actions will have on:
• the residents’ health
• their economic conditions
• their psychological state
• the economy of the city and the state
• carbon dioxide and other greenhouse gas emissions
• businesses
• local politics
2. A city is considering alternative approaches to addressing a growing prob-
lem of drug addiction, which is thought to be responsible for a growing
crime rate and the city’s deterioration. The city’s goals include:
• reduce the number of addicts
• reduce costs to the city
• reduce crimes against property and persons
• improve the quality of life and health of addicts
• improve the quality of life of the city’s residents
• reduce organized crime
• safeguard civil liberties and civil rights
Even if there were only two possible locations for the power plant and
two possible strategies for dealing with the drug addiction problem, the number
of interests involved would make it difficult to use Darwin’s or Franklin’s approach.
With multiple locations or strategies, the problems become very complex.

4.5 a subjective linear model: siting a wastewater

treatment plant

In fact, there are a variety of sophisticated quantitative techniques for addressing

the trade-offs among multiple objectives.18 They are rooted in Multi-Attribute
Utility Theory (MAUT) and are taught in decision analysis courses in business,
engineering, and public policy schools. Here, we shall introduce a rather simple
approach, sometimes called a subjective linear model. The model is subjective
because the weightings and ratings ultimately depend on the decision maker’s
preferences—whether as an individual acting on her own behalf or as an agent

17. Ralph L. Keeney and Howard Raiffa, Decisions with Multiple Objectives:
Preferences and Value Tradeoffs 1–2 (New York: Cambridge University Press, 1993).
18. See Keeney and Raiffa; Keeney, Value-Focused Thinking.
 choosing among alternatives 103

acting on behalf of a public, private, or nonprofit entity. It is linear because

preferences are aggregated simply through addition.
Although simple both in concept and use, the subjective linear model is quite
robust and will take you quite a way toward solving even complex problems.
We will illustrate it with an example.
The mayor of Edenville has asked Christine Lamm to staff a task force on
siting a wastewater treatment plant. The city’s current wastewater treatment
plant no longer meets the demands of its growing population. To further com-
plicate the problem, the majority of the population growth is occurring in the
northern outskirts of Edenville, far away from the city’s only current treatment
plant in south central Edenville. While the houses in this formerly rural area are
currently served by septic systems, the projected growth will require connections
to sewer lines.
Lamm considers the options broadly:
(1) Expand/upgrade the existing plant with new technology to make it more
efficient, and construct new pipelines to the north.
(2) Select a site for a new wastewater treatment plant, preferably in the north.
This problem pits several competing interests—environmental, economic, and
political—against each other.
The expansion/upgrade option may be less expensive and less contentious,
and will also solve the immediate problem by increasing capacity enough to
meet current needs with more efficient technology. However, the siting of a new
facility may better serve the city’s longer term needs if growth trends continue.
And, a new facility may be able to benefit more from new technology that allows
not only better treatment, but also reduced energy consumption. However, the
community’s only reference point is the current plant, a smelly, outdated eye-
sore, blamed for the blight in south central Edenville. If not handled carefully,
the proposal of a new facility is likely to stir up public opposition of the usual
NIMBY (not in my backyard) variety. Lamm, while cognizant of the political
expediency of the upgrade option, has an intuitive preference for a new facility
that would serve the growing community’s long term interests. Lamm would
also like to see if she can incorporate some of her mayor’s green initiatives into
the wastewater treatment plant solution.
Lamm decided to conduct the decision process in a highly transparent manner
with community outreach and public involvement. She believes that early and
regular public involvement is likely to result in the best outcome, to produce the
best project site and design, to encourage public buy-in from various stakeholders,
and to reduce the likelihood of litigation. She met with her staff to prepare a list
of objectives for the site selection to be discussed at the first public meeting.19

19. For the assertion that objectives-based thinking is superior to alternative-based

thinking, see Keeney 29–52.
104 problem solving, decision making, and professional judgment

1. Meet long-term capacity needs.

2. Incorporate green initiatives.
3. Provide truck access that minimizes traffic through residential neighbor-
hoods.
4. Site the plant in proximity to source and treated water release locations to
minimize pump stations.
5. Minimize costs.
She directed her staff to prepare a fact sheet on “Our Growing Need for Water
Solutions” that will explain the city’s current capacity crisis. She requested that
staff explain that the need for additional wastewater treatment capacity in north
Edenville is due to that community’s surging growth, which has strained the
current plant. Quality wastewater treatment is critical to protect public health
and the environment, including threatened and endangered species. A new or
improved plant is essential for further development in the north part of the city.
The existing wastewater treatment plant is burdened by high operating costs due
to overloading, failing equipment, and operating problems, resulting in exces-
sive energy use.
Lamm’s staff posted the fact sheet on the city’s Web site, placed the “Water
Solutions” issue on the next city council meeting’s agenda, and invited the public
to attend this important meeting. At the city council meeting, the staff presented
their preliminary list of objectives and submitted them for public comment.
Several community members presented additional concerns, which the staff
framed as community objectives.
Community objectives:
1. Maintain/improve high water quality.
2. Dedicate adequate land for odor containment and screening.
3. Reduce impacts of construction, traffic, noise.
4. Maintain property values.
The interested parties who spoke during the public comment period included
a community organizer on behalf of the south central residents to express her
community’s frustration with the odors and noise of the current plant and to
argue for a northern site as the only equitable location given that the population
growth is centered in the north. New northern residents spoke against the siting
of a new facility in the north due to concerns about property values, odors, and
compatibility with the residential community. (Residents were skeptical about a
city engineer’s statement that odors were contained by filtering air emissions
from the plant’s exhaust.)
Residents from the neighboring city to the north, Pleasantville, expressed
concern that siting a new facility at its border would negatively impact their com-
munity. Several environmentalists stressed the need for cleaner, greener tech-
nology at the plant, possible initiatives for reduction of wastewater through
 choosing among alternatives 105

recycled water, and the importance of exceeding clean water regulations to

ensure the best water quality for the community.
Lamm asked the staff to incorporate the public comments, accepted the addi-
tional (community) objectives, and directed the staff to post the “Shared
Objectives” list on the city’s Web site. She hoped that, by generating a list of
shared objectives, she would incorporate early community input while still
retaining the ultimate decision-making power.
In continuing with this example, we shall use the term alternatives to describe
the things to be chosen among (e.g., possible sites for wastewater treatment
plant). And we shall call the features (e.g., capacity, costs) that make an alterna-
tive more or less attractive attributes, criteria, or aspects. By definition, a decision
involves two or more alternatives. Most difficult decisions involve several
attributes.
Lamm’s decision procedure involves these steps:
1. Identify the major attributes affecting the decision.
2. Identify the alternatives or options facing the decision maker.
3. Evaluate (objectively or subjectively, as the case may be) the attributes for
each alternative.
4. Translate the evaluations into normalized values.
5. Aggregate the evaluations for each alternative and compare the
alternatives.
6. Perform a reality check.

4.5.1 Identify the Major Attributes Affecting the Decision

The attributes flow from Lamm’s objectives. She directs her staff to translate the
shared objectives into attributes for the site selection, and they produce the
following list:
1. Adequate size: Site size with sufficient area usable for long-term capacity
needs, odor containment, and screening.
2. Environmental impacts: Minimize real and perceived impacts on air (odors),
land, water, and endangered species.
3. Low traffic impacts: Truck access that minimizes traffic through residential
neighborhoods.
4. Low land costs: Minimize costs, inexpensive land, efficient planning, and
construction.
5. Low pipeline lengths and costs: Minimize costs of wastewater transportation,
including pumping stations; keep site proximate to growing sewage sources
and proximate to treated water release location; and minimize costs of
pipelines.
6. Compatibility: Ensure compatibility with surrounding land uses, and design
plant to integrate with neighboring uses.
106 problem solving, decision making, and professional judgment

Lamm decides to treat “greenness” as a decision opportunity to be discussed

later rather than as a specific attribute. In the final trade-off analysis, she will
look for opportunities to incorporate positive enhancements like habitat restora-
tion, hiking trails, and rooftop gardens.

4.5.2 Identify the Alternatives or Options Facing the Decision Maker

Lamm’s staff then evaluates the most promising of the available sites, consider-
ing demolition of existing structures, the presence of hazardous materials,
zoning regulations, impacts on adjacent property values, compatibility with
neighboring activities, downstream uses, wetlands, and endangered species.
Based on this analysis, the staff prepares a list of four alternatives, as shown in
Figure 4.1, including one alternative for the expansion/upgrade option:

figure 4.1 edenville.

 choosing among alternatives 107

• Site A. Airport site. This site, directly adjacent to the Edenville Municipal
Airport, is in the outer, western section of Edenville. This site would be
subject to height restrictions imposed by flight paths maintained by the
airport and would necessitate a contract with the airport. Initial cleanup of
land and pipelines to the north would be required.
• Site B. Northeast suburban site. This site is directly adjacent to a river that
feeds into a lake that is used for recreational activities. It is in the heart of
the quickly growing northern suburban land.
• Site C. Outermost north rural site. This site features a pristine piece of
large land in the northernmost section of Edenville. This area mostly con-
tains farmlands and is not yet impacted by development pressures.
• Site D. Expansion/upgrade of existing site. This major expansion option
would transport sewage from the growing population center in outer
Edenville to the existing, large wastewater treatment plant in south central
Edenville by laying pipelines through neighborhoods. The existing plant
would be greatly expanded and equipped with new technology to make it
more efficient.

4.5.3 Evaluate (Objectively or Subjectively, as the Case May Be) the Attributes
for Each Alternative
Lamm considers each alternative in turn:
• Site A. The airport area, which is not residential but industrial, presents a
site compatible with the plant. The flight paths would likely be easy to
accommodate. However, this site, while politically optimal, requires a large
economic investment in the cleanup of the land due to illegal dumping and
existing structures. Because of its location, this site would also require
costly pipelines to the north.
• Site B. This site is located optimally for proximity to influent sources, since
it is at the heart of the growing residential community. It will reduce the
number of pump stations involved. However, it is also greatly impacted by
development pressures, making any land expensive and limiting the size
of any facility. Similarly, the recreational character of the adjacent area
makes industrial use of this site problematic. Also, the downstream recre-
ational uses would require that the quality of the effluent from the waste
treatment plant be fairly high.
• Site C. This site provides the most space, with the capacity to screen the
plant. It sits high on hill, however, and would affect the view for the popula-
tion in north Edenville. Also, the site is the habitat of some small animals,
an issue which may be raised by environmental organizations.
• Site D. This expansion/upgrade option includes a lower capital cost for treat-
ment facilities and avoids the need to site new treatment facilities. This
option is attractive because most community members prefer the maximum
108 problem solving, decision making, and professional judgment

use of the existing wastewater treatment plant before siting a new one.
With investment in new technology, the upgraded plant could make use
of existing infrastructure for space-efficient increase of treatment capacity
and effluent quality. However, the disadvantages are the costly pipelines,
technology investment, size limits, and environmental justice issues.
Lamm first creates a chart using qualitative words to describe the values of the
attributes for each possible site as shown in Table 4.2.

table 4.2 wastewater treatment plant: qualitative description of attributes

Site Size Enviro Traffic Land Cost Pipeline Compatible

Impacts Impacts Lengths

A Good Very good Good Poor, due to Good Very good

cleanup

B Good Poor Poor Very poor Ideal Poor

C Very good Poor Poor Very good Very good Okay

D Poor Very good Good Very good Poor Okay

4.5.4 Translate the Evaluations into Normalized Values

Using a 100-point grade scale where the highest score is 100, she then translates
her evaluation of the attributes into rankings on this scale, as shown in Table 4.3.
Once she begins to use numbers, she makes finer gradations than when she
used words. For example, Lamm thinks that both the Site A and Site B provide
adequate sized land, but she gives the airport site a slight edge.

table 4.3 wastewater treatment plant: quantitative description of attributes

Site Size Enviro Traffic Land Cost Pipeline Compatible
Impacts Impacts Lengths

Score Score Score Score Score Score

A Good 80 Very 90 Good 85 Poor, due 60 Good 75 Very 90

good to cleanup good

B Good 75 Poor 60 Poor 60 Very poor 50 Ideal 100 Poor 60

C Very 100 Poor 50 Poor 50 Very good 95 Very 90 Okay 65

good good

D Poor 50 Very 85 Good 80 Very good 100 Poor 50 Okay 70

good
 choosing among alternatives 109

She then sums up the score of the attributes for each site, as shown in
Table 4.4

table 4.4 wastewater treatment plant: aggregate evaluations

Site Size Enviro Traffic Land Cost Pipeline Compatible Total

Impacts Impacts Lengths

A 80 90 85 60 75 90 480

B 75 60 60 50 100 60 405

C 100 50 50 95 90 65 450

D 50 85 80 100 50 70 435

4.5.5 Aggregate the Evaluations for Each Alternative and Compare the
Alternatives
Site A has the highest score, followed at a distance by Site C.

4.5.6 Perform a Reality Check

Finally, Lamm thinks again about each of the site options—especially the top
two scores—and asks whether the decision in favor of the Site A (airport site)
makes “gut sense.” After all, the procedure is only designed to aid her in making
and explaining or justifying an essentially subjective decision—to help her
clarify her thinking by disaggregating the factors involved in the decision.
Her procedure makes some simplifying assumptions. It assumes that the attri-
butes are essentially independent of each other.20 This is not always the case. For
example, the pipeline lengths might be related to environmental impacts.21

4.5.7 Weight the Attributes: A Possible Additional Step

The procedure also assumes that each of the attributes—size, environmental
impacts, etc.—is equally valuable. In fact, Lamm might believe that some of
these are more important than others and want to weight the decision process
accordingly. She could therefore add one more wrinkle to this procedure before
undertaking the reality check. She could rank the attributes in importance and

20. See, e.g., Paul Kleindorfer, Howard Kunreuther, and Paul Schoemaker,
Decision Sciences: An Integrative Perspective 137–39 (New York: Cambridge
University Press, 1993).
21. There are multi-attribute decision approaches that are designed to reflect the fact
that some of our preferences are nonlinear. Indeed, through a process known as conjoint
analysis, it is possible to assess an individual’s utility for two or more attributes and deter-
mine trade-offs without engaging in the rather abstract weighting process that Christine
Lamm engaged in. See Jonathan Baron, Thinking and Deciding 340 (3rd ed. New York:
Cambridge University Press, 2000).
110 problem solving, decision making, and professional judgment

assign percentages to each one so that percentages all add up to 100 percent,
with (say) these results:

Size 20%
Enviro Impacts 15%
Traffic Impacts 15%
Land Cost 5%
Pipeline Lengths 20%
Compatible 25%

At this point, Lamm would multiply each attribute’s normalized value by its
weight (step 4 above), as shown in Table 4.5.

table 4.5 wastewater treatment plant: weighted attributes

Site Size (20%) Enviro Traffic Land Cost Pipeline Compatible
Impacts Impacts (5%) Lengths (25%)
(15%) (15%) (20%)

Score Score Score Score Score Score Score Score Score Score Score Score
x x x x x x
weight weight weight weight weight weight

A 80 16 90 13.5 85 12.75 60 3 75 15 90 22.5

B 75 15 60 9 60 9 50 2.5 100 20 60 15

C 100 20 50 7.5 50 7.5 95 4.75 90 18 65 16.25

D 50 10 85 12.75 80 12 100 5 50 10 70 17.5

She would then sum up and compare the final scores for each firm, as shown
in Table 4.6.
table 4.6 wastewater treatment plant: aggregate weighted scores

Site Size (20%) Enviro Traffic Land Cost Pipeline Compatible Total
Impacts Impacts (5%) Lengths (25%)
(15%) (15%) (20%)

Score x Score x Score x Score x Score x Score x

weight weight weight weight weight weight

A 16 13.5 12.75 3 15 22.5 82.75

B 15 9 9 2.5 20 15 70.5

C 20 7.5 7.5 4.75 18 16.25 74

D 10 12.75 12 5 10 17.5 67.25

 choosing among alternatives 111

Even with this weighted adjustment, Site A and Site C still lead the choices,
confirming the earlier unweighted findings.
Is the additional step worthwhile? The calculations don’t take much effort,
but if you put yourself in Lamm’s shoes or, more accurately, in her head, you
may find it difficult to assign weights to the attributes and may be left feeling that
the weightings are pretty arbitrary. In fact, there are some useful strategies for
determining a decision maker’s weightings.22 But it turns out that a simple non-
weighted process works pretty well, even for complex decisions.23
In any event, a subjective linear model has clear advantages over noncompen-
satory procedures for making decisions with multiple alternatives and attributes.
Though it may be tedious, it is computationally simple. It is especially useful
when a decision must be based on a number of different people’s input, because
it tends to make the potential differences in their desires or values explicit.

4.5.8 Decision Opportunities24

Bearing in mind that the choices are now very close in value and that assessing
the risks associated with the initial project costs and likely delays at Site A is
extremely difficult, Lamm reviews her options in light of a consideration she
shelved earlier—making the wastewater treatment plan “green.” She always
thought that this would be a benefit. She could have included it in the formal
analysis, but left it out because it was not a high public priority. But she can now
use it as a tie-breaker.
It is clear that Site C provides the more green opportunities as compared to
Site A. Its large, pristine space, affords great potential for hiking trails and other
compatible recreational uses, rooftop gardens, community educational center,
and beautiful views, which should assuage the neighboring city’s fears. Lamm
enjoys the idea of turning an undesirable project into a landmark and commu-
nity asset. She decides to include several other green initiatives to help allay the
environmental concerns regarding development in Edenville’s valued open
space. She would like to impose green construction requirements, like solar
panels, on the building. She would also like to incorporate community-wide
wastewater reduction measures, such as incentives for Edenville residents to
install low-flow toilets, to help allay the environmental concerns of developing
the open space. With this vision in mind and her growing discomfort with the
likely hassles of Site A, Lamm now confidently pursues Site C as the preferred
option.

22. For example, under the process of conjoint analysis mentioned in a previous foot-
note, one would determine Lamm’s utility function by interrogating her about many
hypothetical trade-offs.
23. See Section 10.5.1; Robyn Dawes, The Robust Beauty of Improper Linear Models in
Decision Making, 39 American Psychologist 571–82 (1979).
24. For a description of decision opportunities, see Keeney, Value-Focused Thinking,
241–67.
112 problem solving, decision making, and professional judgment

4.6 summary: the features of different decision-making

strategies

Table 4.7 summarizes the salient features of some of the different decision-
making strategies outlined above. “Transparency” refers to how clearly one can
articulate and explain the basis for the decision—which are predicates for
discussing it with others and justifying it to them.25

table 4.7 characteristics of different decision-making strategies

Method What Sorts of Quality Effort Transparency
Decisions

Rules, presumptions Repetitive Varies Low High

Intuition, heuristics All Varies Low Very low
Noncompensatory All Moderate Moderate Moderate to high
Compensatory All High High High

Most decisions about which lawyers are asked to counsel clients are not of the
repetitive sort that would permit using rules. And although clients may some-
times ask a lawyer for her “gut reaction,” good counseling usually calls for a
transparent process in which the client’s interests and the extent to which alter-
natives satisfy them are articulated and discussed.
The range of public policy decisions is huge, with many being rule-based or
amenable to formal decision-making processes and others calling for on-the-
spot assessments that preclude formalization. The administrative law scholar
Thomas O. McGarity writes of two different ways of thinking about administra-
tive decision making:26
Although the goal of rational agency decisionmaking seems unexceptional,
its proponents had in mind a very ambitious agenda. They meant to interject
a new and very different way of thinking into a firmly entrenched bureau-
cratic culture. Following the conventional nomenclature, we may label this
new kind of thinking “comprehensive analytical rationality.” The term “com-
prehensive” suggests that this kind of thinking ideally explores all possible
routes to the solution of a problem. The term “analytical” implies that it

25. The table is based on J. Edward Russo and Paul J.J. Schoemaker, Winning
Decisions 155 (New York: Doubleday, 2002).
26. Thomas O. McGarity, Reinventing Rationality—The Role of Regulatory
Analysis in the Federal Bureaucracy 6–7 (New York: Cambridge University Press,
1991).
 choosing among alternatives 113

attempts to sort out, break down, and analyze (quantitatively, if possible) all of
the relevant components of a problem and its possible solutions. The term
“rationality” captures the pride that its proponents take in its objectivity and
the dispassion with which it debates the pros and cons of alternative solutions
without regard to whose ox is being gored. In practice, comprehensive
analytical rationality has been dominated by the paradigms of neoclassical
micro-economics.
This kind of rationality contrasts sharply with the thinking that has tradi-
tionally dominated the rulemaking process in most regulatory agencies,
which I shall refer to as “techno-bureaucratic rationality.” I use the term
techno-bureaucratic to distinguish the thinking that dominates highly
technical rulemaking activities that must grapple with highly complex (and
often unresolvable) issues of science, engineering, and public policy. Some of
the existing models of bureaucratic thinking, such as Lindblom’s perceptive
“muddling through” model, are relevant to bureaucratic programs that have
highly technical, scientific, and engineering components. I use the word
“rationality” because, unlike many students of regulation, I do not believe that
this kind of thinking is irrational per se. Techno-bureaucratic rationality is
a rationality built on a unique understanding of the regulatory universe that
is born out of frustrating hands-on experience with unanswerable questions
of extraordinary complexity. It is, in a sense, a “second best” rationality
that recognizes the limitations that inadequate data, unquantifiable values,
mixed societal goals, and political realities place on the capacity of structured
rational thinking, and it does the best that it can with what it has.
All things being equal—that is, given sufficient time, cognitive capacity, and
patience—value-based decision making is more transparent and seems more
likely to satisfy stakeholders’ long-run needs than its alternatives. But all things
are seldom equal. People appropriately tend to reserve full-fledged decision
value-based models (of which we have provided the most simple example) for a
very limited number of important decisions, where it is plausible to quantify or
at least ordinally rank the attributes of each alternative.
This page intentionally left blank
part two
making sense of an uncertain
world

Life is the art of drawing sufficient conclusions from insufficient premises.

—Samuel Butler

Most issues that lawyers and policy makers encounter involve uncertainties.
We may be uncertain about what happened in the past: what caused the tenants’
rashes and other symptoms at Terra Nueva? And we are inevitably uncertain
about what will happen in the future: what is the likelihood that a particular
approach to eradicating malaria will actually work? How much do we need to
reduce greenhouse gas emissions to prevent the earth’s temperature from rising
more than 2 degrees, and what will happen if it does rise more?
Part 2 is devoted to judgment in the empirical sense of how one ascertains or
predicts facts about the physical and social worlds in situations of uncertainty. It
focuses mainly on the past—on whether a particular event or phenomenon
occurred and on understanding its nature. Often—as in the case of assessing
liability for a tort or crime—the past is of legal interest in its own right. And
almost always understanding what occurred in the past is essential to making
forward-looking decisions. For example, the decision whether to ban the use of
a putatively toxic product will be based on evidence about what happened to
people who were (or were not) exposed to it in the past.

the challenges of detecting correlation and imputing

causation

Whether one’s interest is backward- or forward-looking, the central question is

usually one of causation. Sometimes, causation can be established directly
through observation of the event—for example, an eyewitness saw the defen-
dant’s car run into the plaintiff’s. In these situations, while there may be conflict-
ing evidence, the fundamental paradigm is a Newtonian model of cause and
effect. In many other cases, though, the evidence is essentially of a probabilistic
nature. For example, even in the absence of knowledge of the process by which
a supposed toxic substance causes illness, causation may be hypothesized based
116 problem solving, decision making, and professional judgment

on the correlation of many instances of the putative cause (exposure to a pesticide)

and effects (illnesses) and the exclusion of other possible causes. By the same
token, predicting the consequences of most decisions is a probabilistic enter-
prise that requires taking into account various uncertainties.
Correlation by no means entails causation, but correlation is usually a nec-
essary condition for inferring and imputing causation. Many if not most of
the empirical judgments made by lawyers, policy makers, and individuals in
their personal decisions involve detecting correlation between events in order
to make judgments about causation.1 On what data can they base their
conclusions?
The examples from Terra Nueva and Big-Mart to be considered in Chapters 5
and 6 involve fairly large datasets—large enough that we can use statistical tech-
niques to understand relationships between supposed independent variables or
predictors and dependent or outcome variables. By contrast, in our everyday pro-
fessional and personal lives we must often make assessments of correlation
based on relatively small samples in situations that do not provide very good
information or allow for generalization. Often, we depend on feedback from our
own experiences. Consider these situations:
• Multiple and not necessarily clearly defined outcome variables. Consider the
number of indicators of what constitutes a good employee or a good public
policy decision and the number and vagueness of the attributes of “good-
ness” in these contexts.
• Multiple and ambiguous predictors. Consider the number of factors that go
into an employer’s predicting whether candidates will perform well on the
job, or a trial lawyer’s predicting whether prospective jurors will be inclined
toward, or at least not biased against, her client.
• Feedback only on positive decisions. The employer and trial lawyer never see
the performance of the candidates they rejected, and thus never have the
opportunity to learn whether they would have been satisfactory.
• No feedback at all. One of the authors met a diagnostic radiologist at a party.
The radiologist examines X-rays and MRIs and then communicates his
conclusions to orthopedists. Asked (not entirely innocently) whether the
orthopedists then tell him whether his diagnoses were correct, he responded
that they often do not.
• Delayed feedback. Contrast the feedback available to a meteorologist who
forecasts the daily weather with that available to most employers, lawyers,
and policy makers.
• Self-fulfilling prophecies and treatment effects. Having hired the candidate, the
employer has a vested interest in having him succeed and may act to make

1. Richard Nisbett and Lee Ross, Human Inference: Strategies and

Shortcomings of Social Judgment (Englewood Cliffs, NJ: Prentice Hall, 1980)
 making sense of an uncertain world 117

that happen. Consider the not uncommon situated where a teacher believes
that a student will succeed or fail and, as a result of the “Pygmalion effect,”
helps make the prediction true. A famous series of experiments on the
determinants of worker productivity at the Western Electric Company’s
Hawthorne plant showed that productivity was increased by virtually
any intervention that made the workers feel important and valued—the
so-called Hawthorne effect.2

the lens model

The “lens” model, first proposed by Egon Brunswik, and later developed by
Kenneth Hammond, Robin Hogarth, and others,3 provides a particularly helpful
way of conceptualizing the tasks of empirical inquiry under conditions of uncer-
tainty. The model assumes that there is a “real world” out there—we’ll call it the
environment—and that it is the observer’s task to use available evidence, or indi-
cators, to discover it.4 We offer our own, slightly revised version as shown in
Figure A.

Environment Observer

Lenses

figure a the lens model.

2. Richard Herbert Franke and James D. Kaul, The Hawthorne Experiments: First
Statistical Interpretation, 43 American Sociological Review 623–643 (1978).
3. Kenneth R Hammond, Human Judgment and Social Policy: Irreducible
Uncertainty, Inevitable Error, Unavoidable Injustice (New York: Oxford University
Press, 2000). Robin Hogarth, Human Judgement and Social Choice (2d ed. New York:
John Wiley & Sons, 1987).
4. While this is inconstant with the wildest version of postmodernism, it is entirely
consistent with the notion that reality is “socially constructed”—that is, that there is no
one right or universal way to describe things, events, or phenomena. One need only look
at the contemporary debates over abortion to see two very different views about what it
means to be a “person,” with vastly different moral and legal implications for terminating
its existence. In the end, though, the question whether someone did in fact terminate its
existence (whatever the disagreement about the nature of the “it”) is a fact subject to
empirical study—a fact that can be known with some degree of certitude.
118 problem solving, decision making, and professional judgment

On the left side is the environment, or real world. One can think of the
lines emanating from the environment as indicators of its actual features or
characteristics. A fundamental characteristic of the environment is that it is
probabilistic—that is, the real world is accessible only through what Hammond
terms multiple fallible indicators: The relationship between the real world and
its indicators cannot be represented by strict functional rules, but only as
probabilities.5
On the right side is the observer’s perception of reality, or judgment, as he or
she seeks to integrate those multiple fallible indicators into a coherent under-
standing of the real world. But perception inevitably takes place through lenses,
depicted in the center—lenses that can be more or less distorted.
To use a simple perceptual example: You’re waiting on a corner to meet your
friend Karen for lunch. You see an object in motion on a sidewalk. The light
reflecting off the object (the left side of the lens diagram) shows its features, its
gait, and its apparel. As it gets closer, you discern a human being . . . a woman . . .
Karen. You wave to her. But as she gets still closer, you realize that it’s not Karen
after all, but a stranger. What happened? The characteristics were ambiguous—-
multiple fallible indicators. Moreover, your perception of those indicators was
not “neutral”—for example, it incorporated whatever schemas were active in
your mind at the time: you were supposed to meet Karen at this very time and
place and therefore expected that the person would be her.
In many legal applications, the person making the ultimate judgment—let’s
say a judge hearing a civil case—will not have direct access to the event but must
rely on reports of others’ perceptions. For example, suppose that a judge is trying
to determine whether Donna sideswiped Patricia’s car while it was parked at the
supermarket. Donna denies being at the scene. An eyewitness testifies that she
saw Donna graze the car and drive off. An expert testifies that paint scratched
onto Patricia’s car matches that of Donna’s. The lens model applies not only to
each witness’s recollection of what he or she observed, but also to the judge’s
perception of the statements of each witness. Consider the many ambiguities
in the evidence and the opportunities for distortion both in people’s initial per-
ception and in their recollection. (One might think of the adversary model,

5. Hogarth, supra at 10. For all practical purposes, it does not matter whether the
uncertainty is inherent in the environment or only reflects our limited access to the
environment (as it really is). In either case, we are saddled with the lack of certainty.
Hammond notes that the indicators can be fallible on two dimensions: the accuracy of
information or the accuracy of the cue or indicator reflecting the information. Using the
term “ecological” to refer to its relationship to the environment, or real world, he writes:
“Accuracy of information can be described in terms of its ecological reliability; the accuracy
of a cue or indicator is described as its ecological validity.” For example, while wind direc-
tion may have a high ecological validity for predicting rain, holding up one’s finger to the
wind is unreliable compared to a wind gauge. Hammond, supra at 170.
 making sense of an uncertain world 119

including rules of evidence and direct and cross examination, as designed to

help the judge identify those distortions.)
Finally, consider Christine Lamm’s task of determining whether the polyure-
thane foam caused the illnesses supposedly experienced by the tenants of Terra
Nueva Properties. Epidemiological data concerning the relationship between
exposure to the foam and illness are nothing but a bunch of fallible indicators,
which can only be effectively sorted out through statistical methods. Any conclu-
sion will be explicitly couched in term of the probability of a relationship
existing.

the roles of intuition and analysis in empirical judgment

The goal of factual judgment is to understand some aspect of the world as it

really is. We make thousands of such judgments a day, most of them uncon-
sciously and intuitively. Only on a highly selective basis do we engage in con-
scious analysis. When you are about to cross the street and see a car only in the
far distance, you don’t give it a second thought—perhaps not even a first
thought. Yet if the car is close to the intersection, you may consciously consider
whether you can safely cross before it arrives. In Chapter 1, we broadly consid-
ered the roles of intuition and analysis in professional decisions. Here we focus
particularly on their role in making empirical or factual judgments. In this
regard, it is helpful to consider the traditional stages of the scientific method.6
1. Observing phenomena,
2. speculating to form hypotheses,
3. testing the hypotheses, and
4. generalizing to create a rule or set of objective criteria, typically of the form:
if cause then effect.
To invoke Hans Reichenbach’s useful distinction, the first two stages concern
the process of scientific discovery while the latter two concern the process of sci-
entific justification.7 (Consider the analogy to the processes of divergent and con-
vergent thinking discussed in the preceding chapters.) As Richard Wasserstrom
put it:8

6. This description, though highly conceptualized and oversimplified, in effect sets

out the ground rules for empirical inquiry and proof. For a more nuanced and sophisti-
cated discussion, see Hugh G. Gauch, Jr., Scientific Method in Practice (Cambridge:
Cambridge University Press, 2002).
7. Hans Reichenbach, Experience and Prediction (Chicago: University of Chicago
Press, 1938).
8. Richard A. Wasserstrom, The Judicial Decision: Toward a Theory of
Justification 25–28 (Stanford, CA: Stanford University Press, 1961). Wasserstrom draws
120 problem solving, decision making, and professional judgment

One kind of question asks about the manner in which a decision or conclu-
sion was reached; the other inquires whether a given decision or conclusion
is justifiable. . . .
Consider: . . . A scientist who has discovered a vaccine which purportedly
provides complete immunization against cancer informs the scientific commu-
nity that he hit upon this particular chemical combination in the following
manner. He wrote down 1,000 possible chemical combinations on separate
pieces of paper, put them all into a big hat, and pulled out the pieces at random.
. . . The scientist has announced how he arrived at the conclusion that this
chemical formula might immunize against cancer, but of course he has not
answered the question of whether the vaccine will in fact immunize. . . .
Whether this formula is an effective vaccine is a quite different one.
Furthermore, if ex hypothesi the vaccine were effective, it certainly would not
be rejected because of the way in which the scientist selected it for testing. . .
The example is not purely hypothetical. Friedrich August Kekulé von
Stradonitz, who is credited with discovering the structure of benzene, said that it
came to him in a dream of a snake eating its own tail.
Intuition is critical to the second of the four stages of the process, and it also
plays roles in observation, hypothesis testing, and generalization.9
Part 2 considers essentially three questions. First, what are, in effect, “best
practices” for engaging in these processes in a formal, analytic way? Second, in
what ways do people systematically deviate from these practices, and with what
consequences? And finally, what, if any, useful generalizations can one make
about the roles of intuition and analysis in empirical judgment?

the analogy between these two processes and an appellate judge’s developing intuitions
about the correct decision in a case from justifying the decision. The former is analogous
to the speculative process of developing a hypothesis under the scientific method, while
the latter is analogous to the process of testing and generalizing a hypothesis. He writes
that “to insist—as many legal philosophers appear to have done—that a judicial opinion
is an accurate description of the decision process there employed if and only if it faithfully
describes the procedure of discovery is to help guarantee that the opinion will be found
wanting. But if the opinion is construed to be a report of the justificatory procedure
employed by the judge, then the not infrequent reliance on such things as rules of law and
rules of logic seems more plausible.”
9. The process is often recursive, with testing and generalizing leading to new observa-
tions and hypotheses. Hence Michael Polanyi’s description of a mathematician’s work as
an alternation between intuition and analysis. Michael Polanyi, Personal Knowledge:
Towards a Post-Critical Philosophy 130–31 (New York: Harper & Row, 1964).
 making sense of an uncertain world 121

the role of statistics in law, policy, and citizenship

Statistical thinking will one day be as important for

good citizenship as the ability to read and write.
—H.G. Wells10

I keep saying that the sexy job in the next 10 years will be statisticians. And I’m not
kidding.
—Hal Varian, chief economist, Google11

The “best practices” for testing empirical hypotheses are rooted in statistics.
Examples of statistics abound in legal disputes, some of which involve major
issues of public policy. The evidence in Anne Anderson v. W.R. Grace & Co., made
famous by A Civil Action, involved alleged correlations between drinking water
from contaminated wells and childhood leukemia in Woburn, Massachusetts.
Statistical evidence has played a central role in litigation concerning asbestos,
silicone breast implants, electromagnetic radiation from cell phones, and autism
putatively caused by the mercury preservative in childhood vaccines. In Castenada
v. Partida,12 the Supreme Court relied on statistical analysis of the correlation
between Mexican Americans in the population and on grand juries to conclude
that Mexican Americans were systematically excluded from grand jury service.
In Hazelwood Independent School District v. United States,13 the Court applied a
similar analysis to determine whether the school district discriminated in the
hiring of African American teachers.
Beyond their professional identities, lawyers and policy makers are citizens,
consumers, and community leaders. If nothing else, basic concepts of probabil-
ity and statistics will improve their understanding of newspaper articles on
health, the environment, and many other subjects and give them a healthy skep-
ticism about claims made by advocates, businesses, and governments.
This hardly entails that every lawyer, policy maker, and citizen must be a stat-
istician. But it does suggest that you should grasp the fundamental concepts of
probability and statistics—to understand how a statistician thinks about
the empirical world. Possessing this basic knowledge will alert you to issues
you might otherwise miss and, just as important, provide you with an
informed skepticism about causal claims. Such knowledge will help you know

10. Oscar Wilde went even further, to write that “it is the mark of an educated man to
be moved by statistics.” But see Joseph Stalin: “The death of one man is a tragedy; the
death of a million is a statistic.”
11. Steve Lohr, For Today’s Graduate, Just One Word: Statistics, New York Times,
Aug. 5, 2009.
12. 430 U.S. 482 (1977).
13. 33 U.S. 299 (1977).
122 problem solving, decision making, and professional judgment

when to seek expert advice, and will enable you to communicate with experts.
Moreover—referring back to the “two systems” of judgment described in
Section1.6.1—there is reason to believe that a grasp of probability and statistics
will reduce the “intuitive” statistician’s errors. The evidence indicates that “even
very brief” statistical training has “profound effects on people’s reasoning in their
everyday lives14 and mitigates the biases of unmediated System I processes.15
The following chapters are premised on the belief that understanding the
fundamental concepts is more important than being able to do complex calcula-
tions. Our goal, essentially, is to help you become a savvy consumer of statistics
and strengthen your empirical intuitions. If nothing else, the chapters should
demystify some statistical concepts and terms that you encounter in reading
scholarly articles, legal opinions, and newspapers.

14. Richard Nisbettt, David Krantz, Christopher Jepson, and Ziva Kunda, The Use of
Statistic Heuristics in Everyday Inductive Reasoning, in Thomas Gilovich, Dale Griffin, and
Daniel Kahneman, Heuristics and Biases: The Psychology of Intuitive Judgment
520, 528, 529 (New York: Cambridge University Press, 2002)
15. Franca Agnoli, “Development of Judgmental Heuristics and Logical Reasoning:
Training Counteracts the Representativeness Heuristic” 6 Cognitive Development 195 (1991);
Franca Agnoli & David H. Krantz, “Suppressing Natural Heuristics by Formal Instruction:
The Case of the Conjunction Fallacy,” 21 Cognitive Psych. 515 (1989). See also Peters, et al.,
Numeracy and Decision Making, 17 Psychological Science 407 (2006).
5. introduction to statistics and probability

This chapter introduces statistical thinking and its application to law and public
policy. At the end of the chapter, you can find a glossary with definitions of the
numerous terms we discuss. Our main example will involve the tenants’ com-
plaints of illnesses possibly caused by foam insulation in the apartments at Terra
Nueva (discussed in Chapter 1). Before turning to this example, though, we
provide a brief introduction to probability.

5.1 probability

A natural way of expressing beliefs about events that may or may not occur is
with ideas of chance. You might say that the odds of a certain candidate winning
an election are 3 to 1, or that he or she has a 75 percent chance of winning. In
making these statements, you are expressing your beliefs in terms of the proba-
bility (or chance) of a particular event. The language of probability is an effective
way to communicate the strength of beliefs in cases where we are not sure about
an outcome but have some intuition on which to base our thoughts. In these
situations of uncertainty, as well as situations in which the involvement of chance
is more explicit, we use probabilistic notions to express our knowledge and
beliefs.

5.1.1 Random Variables

A variable can take on either quantitative or categorical (qualitative) values.
Quantitative variables are characteristics like height, grade point average (GPA),
and number of years of education, which take on a range of numbers. Categorical
(or qualitative) variables are attributes, such as gender or occupation, that have two
or more distinct categories.
A simple example of a variable is the result of a coin toss, which can take on
the values “heads” and “tails.” A random variable is a variable whose value is
determined at least in part by chance. Chance, or randomness, comes from fac-
tors we do not know about or cannot determine. If one knew the way the coin
was facing to begin with, the amount, direction, and placement of the force on
the coin, and the relevant physics equations, one could likely determine on which
side the coin would land. In that case, the variable would be deterministic.
However, not being able to determine these things, we consider the outcome
random.
124 problem solving, decision making, and professional judgment

If we tossed a coin ten times, we might see four heads, five tails, and then
another heads: HHHHTTTTTH, for example. These results are called realized
values, realizations, or instances of the random variable coin toss.

5.1.2 The Foundations of Probability

An event is a possible outcome of a random variable. It can be one particular
realization or a set (range) of them: for example, a coin coming up heads or a
candidate winning an election. For an event A, we write the probability of A
occurring as P(A). Thus we would write the probability of seeing heads as a
result of a coin flip as P(heads). In the Terra Nueva example, we would write the
probability of a randomly-chosen resident having a rash as P(rash present).
The formal theory of probability adopts a simple standard: probabilities are
real numbers between 0 and 1. Thus odds of 3:1 in favor of an event and a 75
percent chance of its occurrence are both equivalent to the probability of that
event being 0.75.
The entire mathematical treatment of probability follows from three axioms:
1. For an event A, P(A) is a number between 0 and 1.
2. If A is certain to happen, then P(A) = 1. If it is certain not to happen, then
P(A) = 0.
3. If A and B are mutually exclusive events, then P(A or B) = P(A) + P(B).
The third axiom tells us how to calculate probabilities for events that cannot
occur together. For example, if we flip a coin we will get either heads or tails.
These are mutually exclusive events—we cannot get both heads and tails at the
same time from this one coin. So the third axiom entails: P(heads or tails) =
P(heads) + P(tails). Since we know that we are certain to have either heads or
tails,1 we can apply the second axiom and know that P(heads or tails) = 1. And we
can go on to work out that P(heads) + P(tails)=1, or equivalently, P(tails) =
1 - P(heads).
When two events, like heads and tails, are mutually exclusive and we are cer-
tain one or the other must occur, the events are called complements of each
other. The complement of an event A is often written “∼A” and read “not A.”
Using the same reasoning as above for a more general case, we can use the
axioms to show the probability of A not occurring:
P(~A) = 1 – P(A)
This is called the complement rule.

1. This example assumes that heads and tails are not only mutually exclusive but jointly
exhaustive: i.e., one and only one of them must occur. If we wanted to factor in the possi-
bility that a coin could land and remain on its edge, the formula would be somewhat more
complicated: P(heads or tails or edge) = 1. Therefore, P(tails) = 1 – P(heads) – P (edge).
 introduction to statistics and probability 125

5.1.2.a Odds We just said that “the odds of a candidate winning an election
are 3:1” is equivalent to “the candidate has a 75 percent chance (or a probability
of 0.75) of winning.” Odds of 3 to 1 means that the event is likely to occur about
3 times for every time it does not, that is, 3 out of 4 times, which can be repre-
sented as ¾ = 0.75. In terms of probabilities, the odds of an event are the ratio of
the probability that it will happen to the probability that it will not:
Odds of event A = P(A)/P(∼A).
We know from the complement rule that P(∼A) = 1 - P(A), so
Odds of event A = P(A)/(1 - P(A)).

In this case, the ratio of the probability of the event—P(win) = 0.75—to the
probability that it does not occur—P(lose) = 0.25—is 3:1.
5.1.2.b Meaning of Probabilities: Frequentism vs. Subjectivism What does it
mean to say, “The probability of the coin landing on heads is 0.5?” There are two
ways to define probabilities, frequentism and subjectivism. A frequentist would
define the probability of heads as the relative frequency of heads if the coin were
tossed infinitely many times. After many many tosses, 50 percent of them will
have landed on heads and 50 percent on tails, so each probability is 0.5.
On the other hand, a subjectivist would say that the probability of a coin land-
ing on heads represents how likely this event seems, possibly based on some
evidence. For example, she might examine the coin and conclude that there is no
evidence that it is weighted to make heads more likely than tails or vice versa, so
the two events must have equal probabilities.

5.2 probability distributions

From the axioms in 5.1.2, we saw that the probabilities of mutually exclusive and
jointly exhaustive events sum to 1. In flipping a coin, P(heads) + P(tails) = 1.
The same holds with more than two mutually exclusive events. When tossing a
die, we can consider the random variable of what number lands facing up, with
possible values 1, 2, 3, 4, 5, and 6. Then P(1) + P(2) + P(3) + P(4) + P(5) +
P(6) = 1.
With any random variable, a total probability of 1 is distributed among all pos-
sible values. If we represented the total probability of 1 with a dollar in coins,
those coins would be distributed over the possible values, with, 50c/ on heads and
50c/ on tails. In the die toss, 1/6, or about 17 cents, would be on each number
1–6. In these examples, the probability happens to be distributed evenly, but this
doesn’t have to be the case: imagine a die weighted so that the “6” came up more
than 1/6 of the time.
Like a set amount of money ($1) or a set number of die faces, there is a set
amount of probability (1), and it is distributed (not always evenly) over the mutu-
ally exclusive events. How it is distributed over the events is called the probability
126 problem solving, decision making, and professional judgment

distribution of a random variable. Every random variable has a probability distri-

bution. The probability distribution for the random variable coin flip is 0.5 on
heads and 0.5 on tails. The probability distribution for the random variable die
toss is 1/6 on each of the numbers 1 through 6 (see Figure 5.1(a)). When you are
about to toss a die, you do not know which side will come up on top, but you
might know the probability distribution.

5.3 population distribution vs. empirical distribution: the aim

of statistics

We can display the probability distribution of a discrete quantitative or ordered

categorical random variable in a histogram. The probability distribution of die
toss, in which each side of the die has an equal probability of 1/6 (about 0.17) of
facing up, is displayed in Figure 5.1(a).
Suppose we tossed a die 100 times and obtained the following frequencies,
the number of times each event occurs:

Side Up Frequency

1 11
2 19
3 22
4 19
5 11
6 18

Dividing these by the total number of tosses, 100, we obtain the relative fre-
quencies, which are displayed in Figure 5.1(b). The relative frequency of an event
is the proportion of total opportunities on which it actually occurs, i.e., the ratio
of the number of times the event occurs (its frequency) to the total number of
times it could have occurred. In this case, for example, the die came up a 4 in 19
times out of a possible 100, so the relative frequency of the event that it comes up
a 4 is 19/100 = 0.19.
The difference between Figure 5.1(a) and 5.1(b) is key to statistics. Figure
5.1(a) represents the probability distribution, which is the same for every die
toss. Figure 5.1(b), on the other hand, results from just one experiment of one
hundred tosses. Each of these one hundred realizations came from one die toss,
with the probability distribution shown in Figure 5.1(a). For another experiment
of one hundred tosses, the result would be slightly different, because it depends
on chance.
 introduction to statistics and probability 127

0.00 0.05 0.10 0.15 0.20 0.25 Die Toss Probability Distribution Die Toss Relative Frequencies

0.00 0.05 0.10 0.15 0.20 0.25

Relative frequency
Probability

1 2 3 4 5 6 1 2 3 4 5 6
(a) Number facing up (b) Number facing up

figure 5.1 (a) and (b) probability distribution and relative

frequencies for die toss.

The distribution in Figure 5.1(b)—the relative frequencies in our sample—is

called the empirical distribution, because it is based on what we observed in the
sample.
Imagine the empirical distribution of a sample size of one—that is, the rela-
tive frequency distribution from just one toss. It would have one bar of height 1.0
and no other bars, very different from the probability distribution. The larger the
number of tosses, the closer Figure 5.1(b) will look to Figure 5.1(a). After infi-
nitely many tosses, Figure 5.1(b) will look just like Figure 5.1(a). In fact, in the
frequentist interpretation of probability, that is how Figure 5.1(a) is defined.2
Therefore, we can call Figure 5.1(a) the population distribution, the relative
frequencies among infinitely many tosses. Population distribution refers to the
same thing as probability distribution. However, the term “population distribu-
tion” emphasizes the frequentist view that each realization is picked from a
population.

5.3.1 The Aim of Statistics

In the case of a die toss, we happen to know the population distribution, since it
is reasonable to assume equal probabilities for the six sides. In most real-world
situations, however, the population distribution is unknown. In fact, it is what
we are looking for. The aim of statistics is to work backward—using only the
observed sample, such as what gave us Figure 5.1(b), to infer properties of the
population distribution, such as Figure 5.1(a).

2. Subjectivist might disagree with this frequentist interpretation of Figure 5.1(a).

Instead, they might say that the distribution is uniform due to the symmetry of the die, an
argument that avoids imagining impossible experiments of infinitely many tosses and
instead emphasizes the current state of one’s knowledge.
128 problem solving, decision making, and professional judgment

It is worth pausing for a moment to emphasize this point. Recall the lens
model discussed in the introduction to Part 2: you can never know the real world
directly, but can only estimate it through multiple fallible indicators. In this
sense, the die toss random variable is completely atypical because (assuming that
it’s a perfect die) we know that its population distribution is 1/6 for each possible
value.
In summary then:
• Population distribution: The true distribution of the population will never
be known. We estimate its characteristics by studying our sample, but
these estimates will probably be a little bit off the truth—a different random
sample of the same size will usually have resulted in slightly different esti-
mates.
• Empirical distribution: The distribution of events within our random
sample from the population is called the empirical distribution.

5.4 the population and empirical distribution at terra nueva

Recall the Terra Nueva problem discussed in Chapter 1. The survey to determine
the incidence of rashes included 453 residents. We know the empirical distribu-
tion of the incidence of rashes of those who live in apartments with and without
foam insulation. But these are fallible indicators of the population distribution,
which cannot be known directly.3 The task of statistics is to inform us how good
an estimate of the population we can make based on the sample. The purpose of
statistical inference is not to determine whether the foam insulation caused a
rash in these particular 453 individuals; it is to use the sample to infer whether
foam insulation is related to rash in the general population of Terra Nueva.

5.4.1 Contingency Tables

The Terra Nueva problem involves two categorical variables: We will use insula-
tion to represent the variable indicating whether an apartment has the foam
insulation. The insulation variable takes on the two values: “foam” and “not
foam.” We will use rash to represent the variable indicating whether the resi-
dents of an apartment have a rash. The rash variable takes on the values “absent”
and “present.” As a shorthand, we’ll just use foam to mean foam = present and
rash to mean rash = present.

3. In this particular instance, it might have been feasible to sample the entire actual
population of Terra Nueva. However, in most real-world situations, from environmental
hazards to public opinion polls, this isn’t possible or cost-effective. In any event, as a con-
ceptual matter, statistics assumes that the actual population is infinite—all possible past,
future, and present tenants of Terra Nueva.
 introduction to statistics and probability 129

The typical format for depicting two categorical variables is called a contin-
gency table because, as we will see later, it can be used to determine whether one
variable is contingent on the other. In Table 5.1, in each of four cells is a
frequency, the number of members of the sample that fall into that cell’s joint
categories.

table 5.1 basic contingency table

Rash

Present Absent

Insulation Foam foam & rash foam & no rash

Not Foam not foam & rash not foam & no rash

This is a two-by-two contingency table. In other circumstances, we might

have used three or more categories for rash—none, mild, and severe. (The cate-
gories are not inherent in a quality but are chosen based on what is relevant to
the problem at hand.) In any event, the categories are mutually exclusive and col-
lectively exhaustive (MECE): each apartment can either have foam or not, and
each tenant can either have a rash or not—and there are no other possibilities.
We can populate each cell with the probability that any two of the variables
occur together. The contingency table in Table 5.2 also shows the marginal prob-
ability—“marginal,” because it is shown in the margin of the confluence of each
variable; for example, the top right margin shows the probability that an apart-
ment has foam; the bottom left margin shows the probability that a tenant of that
apartment has a rash. Based on the principles discussed in Section 5.1.2, the
total probability (bottom right) must sum to 1.0.

table 5.2 contingency table with marginal probabilities

Rash

Present Absent Marginal

P(foam &
Insulation Foam P(foam & rash) P(foam)
no rash)
P(not foam P(not foam P(not
Not Foam
& rash) & no rash) foam)
Marginal P(rash) P(no rash) 1.0
130 problem solving, decision making, and professional judgment

5.4.1.a The (Unknowable) Population Distribution The Terra Nueva complex

consists of 465 units total (12 vacant). 279 (60 percent) of the units have foam
insulation and 186 (40 percent) do not. With many families, including children,
the complex has over 1000 residents, of whom we have surveyed 453.
Let’s suppose that when a new tenant comes to Terra Nueva, she chooses
or is assigned to an apartment without anyone’s taking into account the insula-
tion type. So for each tenant, the variable insulation has a probability distribution
of 0.6 on “foam” and 0.4 on “not foam.” That is, there’s a 60 percent chance
she is placed in a foam apartment and a 40 percent chance she’s placed in a
no-foam apartment. As just mentioned, these are the marginal distributions of
the variable insulation, and appear in the right-most column of Table 5.3:

table 5.3 the (unknowable) population distribution

Probabilities Rash

Present Absent Marginal

Insulation Foam 0.25 0.35 0.60

Not Foam 0.15 0.25 0.40

Marginal 0.40 0.60 1.00

We cannot know the actual population distribution of rash, but—only because

your authors have invented the example—we can tell you that, among the popu-
lation under consideration, a person moving into Terra Nueva has a 40 percent
chance of developing a rash. Therefore, the marginal distribution of rash is 0.6
on “absent” and 0.4 on “present,” as displayed in the bottom row of the probabil-
ity table.
Again, only because we have constructed the microcosm of Terra Nueva,
we can tell you that the probability that a randomly-chosen resident both inhabits
an apartment with foam insulation and develops a rash— P(rash & foam)—is
0.25. (Even our omniscience has its limits: we can’t tell you whether the rash is
caused by the foam, but only that we observe both events happening together
25 percent of the time.) Because the marginal distributions have already been
determined, the other three bold probabilities in the table are set to make each
row and column add up to the correct marginal probability. The four bold
numbers comprise the joint distribution of the two variables insulation and rash.
Just as each pair of marginal probabilities sums to 1, the four joint probabilities
sum to 1.
 introduction to statistics and probability 131

5.4.1.b The (Known) Empirical Distribution

Usually, the probabilities of having a rash with or without foam (Table 5.3) are
unknown. They cannot be discerned directly but only based on the sample of the
453 residents of Terra Nueva who were surveyed. The population (of all 1000
residents) is equivalent to the infinitely many die tosses it would take to obtain
an empirical distribution like Figure 5.1(a); the sample (of 453 surveyed resi-
dents) is analogous to the sample of 100 die tosses it took to obtain the empirical
distribution of Figure 5.1(b). Suppose that the survey of Terra Nueva’s residents
has yielded the following information (Table 5.4):

table 5.4 the empirical distribution in numbers

Observed Frequencies Rash

Present Absent Total

Insulation Foam 120 149 269

Not Foam 63 121 184

Total 183 270 453

In some situations, one might want to display the data as proportions of the
total—that is, as relative frequencies. Table 5.5 displays the same information as
Table 5.4, but in relative frequencies. (Note that the format of the table resembles
that of the (unknowable) probability distribution of the population from which it
sampled. Table 5.3 is analogous to Figure 5.1(a), while Table 5.5 is analogous to
Figure 5.1(b).)
(A different sample of tenants would almost surely have produced slightly
different numbers than the sample represented in Tables 5.4 and 5.5).

table 5.5 the empirical distribution in relative frequencies

Observed Relative Frequencies: Rash

Present Absent Total

Insulation Foam 0.26 0.33 0.59

Not Foam 0.14 0.27 0.41
Total 0.40 0.60 1.00
132 problem solving, decision making, and professional judgment

5.5 estimating probabilities

Our goal is to use the frequencies we observe in this sample of residents to esti-
mate the probabilities, i.e., the relative frequencies in the total population. For
example, in our sample of 453 residents, we see that 120 had both a rash and
foam insulation, giving a relative frequency of 0.26. But a sample is always a fal-
lible representation of the population. That is, an estimate is almost inevitably
somewhat off from the true probability. How can we know how closely (or not)
the relative frequencies of the sample approximate it?
We begin with a discussion of the concept of sampling, and then turn to
the concepts of conditional probability and independence. Finally we dis-
cuss the matter of hypothesis testing—and in particular ascertaining the
independence (or not) of having a rash and living in a foam-insulated
apartment.

5.5.1 Sampling
Like pollsters before an election, lawyers and policy makers often are interested
in the characteristics of a large population but lack the time or resources to mea-
sure or survey every member of that population. This leads us to choose a smaller
random sample from the population and take the relevant measurements of
those in the sample.
Often, investigators take a simple random sample, which means that each
population member has an equal chance of being in the sample. From the infor-
mation on the sample, we use statistics to infer a conclusion about the entire
population.

Population -> Sampling -> Sample -> Inference -> Conclusions about population.

For pollsters, the population of interest consists of all voters in the region
of interest. A simple poll might survey a random sample of, say, 1000 likely
voters from the population of all voters in the region, then use statistical
inference to predict the outcome of the election. However, if the sample was
not chosen at random and therefore is not representative of the population as
a whole, the conclusions may be incorrect. This is a problem called selec-
tion bias or sampling bias; examples are in the “Surveys” box. (In reality,
pollsters do not always use simple random sampling and often instead
use complex formulas based on knowledge about party registration, past
elections, etc.)
You may have noticed that such predictions come with a “margin of error.”
That’s because the sample does not exactly reflect the entire population from
which it is drawn. In the next chapter, we’ll discuss the characteristics of the
sample that determine its margin of error.
 introduction to statistics and probability 133

Surveys

Efforts to obtain random samples, especially in surveys, are subject to practical

pitfalls. Three presidential polls provide a good illustration. The 1936 Literary Digest
poll that erroneously predicted that Alfred Landon would defeat Franklin Roosevelt
by a large margin was biased in two ways: it had a selection bias by using lists of
automobile owners and telephone subscribers to poll voters; but in 1936 only the
relatively affluent had cars and phones. It also had a nonresponse bias: as it turned
out, Landon voters were more likely to return the questionnaire. The 1948 mispre-
diction that Thomas Dewey would defeat Harry Truman was based partly on the
fact that more of the interviewers were Republicans than Democrats and that,
consciously or not, they made their leaning known to the respondents, who told
them what they wanted to hear. Exit polls that erroneously favored John Kerry over
George W. Bush in some precincts in the 2004 presidential election may have
been due to (youthful) Kerry voters’ greater willingness to respond to the inter-
viewers, which in turn may have been due to the interviewers’ relative youth.

HANS ZEISEL AND DAVID KAYE, PROVE IT WITH FIGURES: EMPIRICAL METHODS IN LAW AND
LITIGATION 103–04 (New York: Springer-Verlag, 1997); Evaluation of Edison/
Mitofsky Election System 2004 prepared by Edison Media Research and Mitofsky
International for the National Election Pool (NEP), Jan. 19, 2005, http://www.
exit-poll.net/faq.html. See generally Fritz Scheuren, “What is a Survey,” http://
www.whatisasurvey.info/.

5.6 conditional probability and independence

In Section 5.4.1, we discussed marginal and joint probabilities. Recall that mar-
ginal probabilities are those (in the margins of the contingency table) that give
the probability for a single variable, regardless of what the other variable equals,
such as P(foam) = 0.6 and P(rash) = 0.4, while joint probabilities are those in the
center of the table, such as P(foam & rash) = 0.25.
In this section, we introduce conditional probability. A conditional probability
is the probability that event A will occur if we know that B has occurred. It is
written
P(A|B)
and is stated: “the probability of A given B.”
For example, suppose you are flying from San Francisco to Cedar Rapids via
Chicago. We can talk about the probability of your catching the flight from
134 problem solving, decision making, and professional judgment

Chicago to Cedar Rapids given that your flight from San Francisco leaves on
time—or given that it is thirty minutes late.
Conditional probabilities tell us the probability distribution of one variable
when another variable is known. In the Terra Nueva case, one might want to
know the probability of a tenant’s having a rash, given that she lives in an
apartment with foam insulation. We represent this conditional probability
formally by
P(rash| foam)
As noted above, the “|” is read “given,” so this is stated as: “the probability that
the tenant has a rash, given that she lives in an apartment with foam insulation.”
Although this probability does not appear in Table 5.3, it can be calculated
from those that are. Because we know the insulation is foam, we focus only on
the top row. The conditional probability is that fraction of the total probability of
the top row (0.60) for which a rash is present. That is,
P(rash| foam) = 0.25 / 0.60 = 0.42
One way to think about this is in terms of frequencies: Imagine that there are
1000 people total. Six hundred of them have foam insulation. And of those 600,
250 have a rash. So, of the group with foam insulation, the fraction with a rash is
250/600 = 0.42—that is, 42 percent of those with foam insulation have a rash.
More generally, we can write this calculation as:
P(rash present| foam) = P(foam & rash)/P(foam)
Is P(rash|foam) the same as P(foam|rash)? Or to put the question more gener-
ally, is P(A|B) the same as P(B|A)? Absolutely not! Consider the difference
between P(pregnant|woman) and P(woman|pregnant).

Bush Voters and Moral Values

Correction: The Mail of January 3rd contained the incorrect statistic that
four-fifths of Bush voters identified moral values as the most important
factor in their decision. In fact, four-fifths of those identifying moral values
as the most important factor of their decision were Bush voters. New
Yorker, Feb. 14–21, 2005, 38.

5.6.1 Independence
Conditional probabilities are used to express information about how knowledge
of one variable influences our knowledge of another. Often, we want to know if
 introduction to statistics and probability 135

two variables are, or are not, related to each other. In the language of probability
theory, we can ask whether or not two variables are independent.
Formally, we say that two events are independent of each other if P(A|B) =
P(A)—which also implies that P(B|A) = P(B). This means that the probability of
A happening does not change, regardless of whether B happens or not, and vice
versa. This matches up well with the intuitive notion of independence: If we say
that two things are independent, we usually mean that they have no effect on one
another. In the example above, is the probability of your catching the flight from
Chicago to Cedar Rapids independent of the fact that your flight from San
Francisco left on time? Is it independent of the fact that the flight number is the
same as the month and day of your birth?
In the Terra Nueva case, if insulation and rash were independent, then being
in an apartment with foam would not make it any more or less likely that a resi-
dent would develop a rash. In terms of conditional probabilities, the probability
of having a rash given that one lives in a foam-insulated apartment would be
identical to the probability of having a rash given that one lives in any apartment,
or P(rash|foam) = P(rash). Based on the calculations in the previous subsection,
does this seem to be true?4 We’ll find out in a while.
In gambling, events are supposed to be statistically independent. For exam-
ple, subsequent spins of a roulette wheel should give numbers that are com-
pletely unrelated to previous spins. If this were not the case, a player could use
the result of previous spins to improve his or her chances of winning, to the
detriment of the casino. By the same token, subsequent tosses of a coin are also
independent: the information that the last toss produced a head tells you nothing
about the expected outcome of the next toss.
A convenient consequence of independence is that when two variables are
independent, the probability that an event involving one and an event involving
the other both occur is the product of the marginal probabilities. If the first spin
and the second spin of a roulette wheel are independent, then the probability
that the ball lands on red on the first and on black on the second is
P(first spin = red & second spin = black) = P(first spin = red) * P(second spin
= black)
Similarly, if getting a rash were independent of one’s type of insulation, then
the joint probability that a Terra Nueva resident has foam insulation and a rash
would be
P(rash & foam) = P(rash) * P(foam)

4. Without knowing a tenant’s insulation type, her chance of having a rash is the mar-
ginal probability, 0.40. When we learn that her apartment has foam insulation, however,
the chance increases to 0.42. So insulation and rash are not independent.
136 problem solving, decision making, and professional judgment

In our case, in particular, this would be 0.6*0.4 = 0.24. If rash and insulation
were independent, and we selected a Terra Nueva resident at random, there
would be a 24 percent chance that the resident had both foam insulation and
a rash.
In general, if events A and B are independent,
P(A & B) = P(A) * P(B)
This is called the multiplication rule for independent events.
Problem: When rolling a pair of dice, what is the probability of rolling a
double six?
The idea of independence is essential to statistics. The existence of a depen-
dency between two variables means that those variables are related. Much of the
practice of statistics involves attempting to find relationships among variables,
and to quantify the strength of those relationships. These tasks correspond to
testing whether variables are independent and measuring the extent to which
they deviate from independence.

5.7 randomness, independence, and intuition

Schemas are an essential part of the human situation. “We are predisposed to see order,
pattern, and meaning in the world, and we find randomness, chaos, and meaning-
lessness unsatisfying. Human nature abhors a lack of predictability and the absence of
meaning.
—Thomas Gilovich.5

Much psychological research has examined how good people are at detecting
relationships or their absence. It turns out that people often neglect the implica-
tions of statistical independence when evaluating the probability of events being
produced by a random process.

5.7.1 Illusory Correlation

As a consequence of our schematic perception and desire to impose order on
reality, we sometimes see the correlations that we expect to see even where they
do not exist—especially when there is a plausible explanation for the correlation.
For example, a study of arthritis patients showed that they were virtually certain
that their pain was correlated with the weather, when the actual correlation was
close to zero.6 In another experiment, done at a time when many psychologists

5. Thomas Gilovich, How We Know What Isn’t So: The Fallibility of Human
Reason in Everyday Life, 9 (New York: The Free Press, 1991).
6. Donald A. Redelmeier and Amos Tversky, On the Belief that Arthritis Pain Is Related
to the Weather, 93 Proceedings of the National Academy of Sciences 2895 (1999).
 introduction to statistics and probability 137

diagnosed psychiatric disorders by interpreting a patient’s drawings in a “Draw-

a-Person” Test, Lauren J. Chapman and Jean P. Chapman surveyed clinicians and
found considerable agreement about what characteristics of the drawing were
characteristic of particular disorders—for example, strongly dependent people
emphasize the mouth, paranoids emphasize the eyes, and impotent men draw
men with broad shoulders. Psychologists’ beliefs in these correlations were strong,
despite robust evidence that the correlations were no stronger than random.7
Richard Nisbett and Lee Ross observe that in the contest between expecta-
tions or theory and evidence or data, “expectations based on a priori theories or
on semantic connotations overwhelm the influence of data that do not coincide
with expectations or even refute them.”8

5.7.2 The Illusion of Control

The winner of the Spanish lottery, who bought a lottery ticket ending in 48, when asked
why he chose that number, explained that for 7 nights in a row he dreamed of the number
7, and since 7 times 7 equal 48, he knew that would be the lucky number.9

Related to illusory correlation, people also manifest what Ellen Langer has
termed the illusion of control—acting as if they exercise control over events that
are in fact matters of chance.10 For example, some gamblers throw the dice softly
to get low numbers and hard to get high numbers; people playing a game of
chance will bet more against an opponent who looks like a “schnook” than one
who seems competent; and people who choose a lottery ticket demand a higher
price to sell it than those given a lottery ticket at random. In some of these cases,
a misunderstanding of chance may be combined with a desire or need for per-
sonal efficacy—the drive to master one’s environment begins early in infancy,
and the sense of not being in control can be stressful and debilitating—or with a
tendency to believe that if we desired an outcome and it occurred, we must have
been responsible for bringing it about.11

7. Lauren J. Chapman and Jean P. Chapman, Genesis of Popular But Erroneous Psycho-
Diagnostic Observations, 72 Journal of Abnormal Psychology 193–204 (1967); Lauren J.
Chapman and Jean P. Chapman, Illusory Correlation as an Obstacle to the Use of Valid
Diagnostic Signs, 74 Journal of Abnormal Psychology 271–80 (1969).
8. Richard Nisbett and Lee Ross, Human Inference: Strategies and
Shortcomings of Social Judgment 97 (Englewood Cliffs, NJ: Prentice Hall, 1980).
9. Stanley Meisler (1977), cited in Reid Hastie and Robyn Dawes, Rational Choice
in an Uncertain World 154 (Thousand Oaks, CA: Sage Publishing, 2001).
10. Ellen J. Langer, The Illusion of Control, 32 Journal of Personality and Social
Psychology 311 (1975).
11. Ellen. J. Langer and Jane Roth, Heads I Win, Tails It’s Chance: The Illusion of Control
as a Function of the Sequence of Outcomes in a Purely Chance Task, 34 Journal of Personality
and Social Psychology 191–98 (1975); Ellen J. Langer, The Illusion of Control, in Judgment
138 problem solving, decision making, and professional judgment

The illusion of control is a special case of illusory correlation—involving the

correlation between one’s own actions and an outcome. The foundations for
the illusion of control lie in people’s tenuous understanding of probability and
the difficulties of assessing correlations between putative causes and effects,
especially when the environment does not lend itself to accurate feedback. The
illusion may arise from the apparent connection between intending an outcome
and having the outcome occur—an instance of both the availability heuristic
(one’s own action is readily brought to mind) and the representative heuristic
(the similarity between the intention and the outcome).12
There is ample experimental evidence that even bright, well-educated people
engage in “magical thinking” in which, for example:
One grumbles bitterly about Grandma just before she falls and breaks a hip
or expressed anger toward the greedy landlord the day before he is arrested
for tax evasion. Without a shred of evidence about who caused their troubles,
one may still feel implicated. . . . [T]he sense may range from an outright
belief in personal responsibility for the bad outcome to a nagging feeling of
responsibility that persists despite the rational belief that one is not actually
responsible. In any case, the inference involves an erroneous perception of
causality.13

There are thirty-nine categories of work that are prohibited on the Sabbath.
Category No. 27, Kindling a Fire, also rules out the “kindling” of anything electri-
cal, including a television. It was Friday, May 6, 1994: Game Four of the Stanley
Cup conference semifinals between the Rangers and the Washington Capitals
was scheduled for that Saturday afternoon. I had decided to switch on the televi-
sion Friday afternoon—before Sabbath began, at sundown—and just leave it on
until Sabbath ended, twenty-five hours later. This wasn’t, technically, “being in
the spirit of the Sabbath,” but it wasn’t technically a sin, and the Rangers were

Under Uncertainty: Heuristics and Biases (Daniel Kahneman, Paul Slovic, and Amos
Tversky eds., New York: Cambridge University Press, 1982). See also Paul Presson and
Victor Benassi, Illusion of Control: A Meta-Analytic Review, 11 Journal of Social Behavior
and Personality (1996); Shelley Taylor, Positive Illusions 26 (1989).
12. See Suzanne Thompson, Wade Armstrong, and Craig Thomas, Illusions of Control,
Underestimations, and Accuracy: A Control Heuristic Explanation, 123 Psychology Bulletin
113 (1998); Nisbett and Ross, supra at 136–37. In Section 13.5, we will discuss research
indicating that positive illusions, such as the illusion of control, are adaptive because they
increase motivation and persistence. http://www.answers.com/topic/illusion-of-control.
13. Emily Pronin et al., Everyday Magical Powers: The Role of Apparent Mental Causation
in the Overestimation of Personal Influence, 91 Journal of Personality and Social
Psychology 218 (2006).
 introduction to statistics and probability 139

very likely nine victories away from winning the Stanley Cup for the first time in
fifty-four years.
. . . I switched on the television, turned down the volume, and draped a bath
towel over the screen to hide from the neighbors the flickering blue light of our
moral weakness.
“Do you really think that if you turn on the TV on Sabbath God will make the
Rangers lose?” Orli asked.
Her naiveté astounded me.
“I don’t think He will. I know He will.”

Shalom Auslander, Personal History, NEW YORKER, Jan. 15, 2007, PG38.

5.7.3 The Gambler’s Fallacy

The gambler’s fallacy is the belief that, after a run of several of the same event
(same color, number, side of a coin or die, etc.), that event becomes less likely
than otherwise on the following trial. For example, an empirical study of roulette
players in a casino showed that the longer the streak of outcomes of a particular
type (e.g., black), the more likely a player was to bet against the streak.14 Similarly,
people tend not to bet on a recent winning lottery number.
Underlying the gambler’s fallacy is the belief that, after seeing a run of tails in
a sequence, the next toss is more likely to result in a head: it does not reflect the
fact that the outcomes of the bets, or the tosses of the coin, are independent. For
independent events, the probabilities of different outcomes do not change as a
consequence of the results seen so far.
The gambler’s fallacy can be seen as a misapplication of the frequentist inter-
pretation of probability. The fact that a fair coin produces heads with probability
0.5 does not mean that for every finite number of tosses, half of the outcomes are
guaranteed to be heads. Those who commit the gambler’s fallacy explain it with
the idea that after a number of tails, the coin must come up heads in order
to “balance out” the sequence. The phenomenon is often attributed to the
representativeness heuristic, mentioned briefly at the end of Chapter 2 and
described in more detail in Section 8.3. In this case, sequences containing
approximately equal numbers of heads and tails are representative of those
produced by a fair coin. Seeing a head after a sequence of tails makes the

14. Rachel Croson and James Sundali, The Gambler’s Fallacy and the Hot Hand:
Empirical Data from the Casinos, 30 Journal of Risk and Uncertainty 195 (2005).
140 problem solving, decision making, and professional judgment

sequence more representative of a fair coin than seeing another tail, so people
give it a higher probability.15
The other side of the gambler’s fallacy coin, so to speak, is the belief that an
individual is sometimes on a winning streak, so that random sequences are
interpreted as other than random. For example, the roulette study mentioned
above indicated that players made more “outside” bets (bets on red/black, odd/
even, or groups of numbers) when they had previously won on outside bets and
more inside bets (bets on particular numbers) when they had previously won on
inside bets.16 The notion that athletes exhibit periods of unusually good play—
the “hot hand,” “streak shooting,” or being “in the zone”—may be an example of
this fallacy.
There is a considerable debate about whether and when players actually have
a hot hand. The psychologists Thomas Gilovich, Robert Vallone, and Amos
Tversky recorded a professional basketball teams’ statistics for an entire season
and concluded that, in fact, players’ shots were independent.17 The researchers
did not argue that an individual can never be “in the zone”; only that their study
indicated that claims of this sort were unfounded. Sports fans, including schol-
ars of judgment and decision making as well as our own students, strongly resist
this finding. For example, an article on Bowlers’ Hot Hands criticized the basket-
ball study on the ground that it did not control for several confounding influ-
ences and presented evidence that most bowlers actually have a higher proportion
of strikes after bowling a certain number of consecutive strikes than after con-
secutive nonstrikes.18
Our apparently poor intuitions about randomness may also help provide
insight into another psychological phenomenon—the idea of meaningful coinci-
dence. Seemingly unlikely events catch our attention and often appear in newspa-
pers. Scott Plous gives an example of two men sharing the same surname, the
same model of car and even exactly the same car keys, both being in a shopping
center on the same time on April Fool’s Day, and managing to confuse their cars.19

15. This argument was made by Amos Tversky and Daniel Kahneman, Belief in the
Law of Small Numbers, 76 Psychological Bulletin 105–10 (1971).
16. An inside bet is made within the 36 numbers of the roulette table, for example, on
a single number or on two or four adjacent numbers. An outside bet is, say, red or black,
odd or first 12. The correlations were small but statistically significant at p < 0.05.
17. Thomas Gilovich, Robert Vallone, and Amos Tversky, The Hot Hand in Basketball:
On the Misperception of Random Sequences, 17 Journal of Personality and Social
Psychology 295–314 (1985).
18. Reid Dorsey-Palmateer and Gary Smith, Bowlers’ Hot Hands, 58 American
Statistician 38 (2004).
19. Scott Plous, The Psychology of Judgment and Decision Making 154 (New
York: McGraw Hill, 1993).
 introduction to statistics and probability 141

Similar chains of strange co-occurrences can be found discussed in many books,

and the Internet provides several sites dedicated to recording coincidences.20
By not noticing the multitude of uninteresting events that take place, we
underestimate the number of opportunities that have been available for an inter-
esting (and unlikely) coincidence to occur. A one-in-a-million chance will often
arise when given a million opportunities, and if we do not notice those opportu-
nities we can be surprised when it does arise.
Similar sorts of phenomena account for many of the strange coincidences that
we encounter. Statisticians Persi Diaconis and Fred Mosteller, who have investi-
gated coincidences in detail, discuss a newspaper article about a woman who
won the New Jersey lottery twice within four months. This seems very unlikely,
but Diaconis and Mosteller estimate the probability of something like this hap-
pening to somebody in the United States to be as high as one in thirty.21

5.8 testing hypotheses — and the concept of statistical

significance

Hypothesis testing allows us to determine whether an apparent empirical rela-

tionship is large enough to make us confident that it reflects a real relationship
in the population, not just in the random sample we chose. Hypothesis testing
allows us to use the sample we have to reach a conclusion about the relationship
between the two variables in the whole population. It involves:
• hypothesizing that there is no real underlying relationship between the two
variables;
• calculating a measure of how unlikely our sample data would be if, in fact,
there were no relationship; and then
• determining whether the data are extreme enough to justify rejecting the
null hypothesis and concluding that there is a relationship.
Step 1) State the two hypotheses of the test:
• The null hypothesis is the default hypothesis—typically that there is no
relationship—that we would believe in the absence of sufficient evidence
against it. We need evidence to reject the null hypothesis. Typically, it states
that there is no relationship between two variables. In the Terra Nueva case,
the null hypothesis is that insulation and rash are independent.
• The alternative hypothesis is the scenario that we might suspect is the
case but require evidence to conclude that it is. Typically, it states that

20. For example, see http://go.to/coincidences/.

21. Persi Diaconis and Frederick Mosteller, Methods of Studying Coincidences, 84
Journal of the American Statistics Association 853–61 (1989).
142 problem solving, decision making, and professional judgment

two variables are related. In the Terra Nueva case, the alternative hypothe-
sis is that insulation and rash are not independent but rather are related
such that rash is more likely given the presence of insulation.
We test the hypotheses by comparing the observed relative frequencies (Table
5.5) to the relative frequencies that we would expect if insulation and rash were
independent. We can calculate the latter using the multiplication rule, as shown
below.
The question is, how far apart do those two tables have to be to convince us
that the null hypothesis is false?
The two tables will almost certainly be somewhat different. But is the differ-
ence small enough to be chalked up to chance or is the difference statistically
significant? A hypothesis test allows us to reach a conclusion about the entire
population based only on our sample data.
Step 2) Set the level of the test.
Since we have only a sample with which to test our hypothesis, it is possible
that we will make an error. In fact, we can distinguish two kinds of errors that
can result from performing a hypothesis test. There are two possible states of the
world: a relationship between two variables exists, or it does not. And there are
two possible conclusions we can draw: that a relationship exists, or that it does
not. In the case where no relationship exists, but we conclude that one does, we
have made a Type I error. In the case where a relationship exists but we conclude
that it does not, we have made a Type II error.
A good way to think about the difference between Type I and Type II errors is
in the context of medical diagnosis. A medical test can give two different kinds
of erroneous results: it can return a positive result when the patient does not
have the condition tested for (a false positive), or it can return a negative result
when a patient actually has the condition tested for (a miss). Both of these errors
can have important consequences for the patients involved, and medical research-
ers try to develop tests that avoid making them. In statistics, false positives are
called Type I errors and misses are called Type II errors. (See Table 5.6.)
table 5.6 type i and type ii errors

Test suggests Test does not suggest

relationship (rejects null) relationship (does not reject null)

Relationship exists Hit Miss (Type II error)

(alternative is true)
No relationship (null is False positive Correct negative
true) (Type I error)

In hypothesis testing in science, policy, and law, it is typical to err on

the side of avoiding Type I errors. There are at least two motivations for this.
 introduction to statistics and probability 143

First, it is important to make sure that false findings of relationships do not

readily make their way into the scientific literature. Second, the probability of
making a Type I error is easy to quantify. For a Type I error to occur, the null
hypothesis must be true. Since the null hypothesis is more clearly defined than
the alternative (which could be any number of possible relationships), it is easier
to calculate probabilities of events occurring under the null hypothesis.
The significance level of a hypothesis test is the highest probability of Type I
error that is acceptable. Often, the level—sometimes referred to as alpha level
and denoted by the Greek letter α (alpha)—is chosen to be 0.05. So, even if the
null hypothesis is true, we expect to reject it one out of twenty times. For exam-
ple, if we conduct twenty well-designed experiments to determine whether a
supposedly toxic substance in fact causes harm, on average one of them will sug-
gest that it causes harm even if the substance is in fact harmless.
The value α = 0.05 is arbitrary, but it is accepted by convention in many disci-
plines. In some cases, where we want the probability of a Type I error to be
reduced, it is appropriate to use a more stringent alpha level, such as α = 0.01,
0.001, or 0.0001. One context in which people are very careful with the alpha
level is where they are conducting many statistical tests at once. We pointed out
above that, even if no relationship existed, about 1 in 20 tests would conclude
statistical significance at the α = 0.05 level. This is called the problem of multiple
testing or multiple comparisons. To reduce the chance of making at least one
Type I error among several tests, one could make the α level more stringent (i.e.,
smaller) for each test or, better yet, use an approach based on Bayes’ Theorem
(discussed in Section 8.5).
The choice of the alpha level also has implications for the probability of a Type
II error. All things being equal, the probability of making a Type II error will
increase as α decreases—if you are more stringent about the values of a statistic
that you will allow to count as evidence against the null hypothesis, it becomes
more difficult to reject the null hypothesis, and the probability of a Type II error
rises accordingly. The choice of an alpha level reflects a compromise between
producing a lot of Type I errors and a lot of Type II errors.
You might think of the different burdens of proof in civil and criminal cases
as different balances between Type I and Type II errors. If the alternative hypoth-
esis is that the defendant is liable, the civil standard requiring a “preponderance”
of the evidence manifests a very slight preference against Type I errors, and the
criminal standard of “beyond a reasonable doubt” manifests a very strong prefer-
ence against Type I errors, erring on the side of letting the guilty go free rather
than convicting an innocent person.
In the Terra Nueva problem we will set the level to the customary α = 0.05.
Step 3) Suppose that the null hypothesis is true.
The main idea of a statistical hypothesis test is to find how extreme the sample
data would be if the null hypothesis were true.
144 problem solving, decision making, and professional judgment

The null hypothesis is that insulation and rash are independent. Supposing
independence is true, and using the observed marginal relative frequencies
as estimates of the marginal probabilities, the multiplication rule gives us the
estimate:
P(foam & rash) = P(foam) * P(rash) = 0.59 * 0.40 = 0.24
We can do the same calculation for the other combinations of insulation and
rash: foam &not rash, not foam & rash, and not foam & not rash, and present them
in a new table of the relative frequencies expected under the null hypothesis. (See
Table 5.7.)

table 5.7 expected relative frequencies

Expected Relative Frequencies: Rash

Present Absent Total
Insulation Foam 0.24 0.35 0.59
Not Foam 0.16 0.25 0.41

Total 0.40 0.60 1.00

To express the expected distribution in terms of (absolute) frequencies, we

multiply the relative frequencies by the total number of residents in our sample,
453. This gives us the following frequencies expected under the null hypothesis
of independence, as shown in Table 5.8.

table 5.8 expected absolute frequencies

Expected Frequencies Rash

Present Absent Total
Insulation Foam 109 160 269

Not Foam 74 110 184

Total 183 270 453

Step 4) Calculate how far the data are from what would be expected under the
null hypothesis.
 introduction to statistics and probability 145

Let us compare the observed frequencies with the frequencies expected under
the null hypothesis of independence, as shown in Table 5.9.

table 5.9 observed versus expected frequencies

Observed Rash

Present Absent
Insulation Foam 120 149
Not Foam 63 121

Expected Rash

Present Absent
Insulation Foam 109 160
Not Foam 74 110

In our sample, the upper left and lower right cells contain frequencies (of 120
and 121) higher than their expected frequencies (of 109 and 110). Consequently,
in the lower left and upper right cells, the observed frequencies are lower than
what is expected under independence. The further from the expected frequen-
cies the observed frequencies are, the further from independent the variables
appear to be. Here are the differences in frequencies between observed and
expected, as shown in Table 5.10.

table 5.10 observed minus expected frequencies

Obs–Exp Rash

Present Absent

Insulation Foam 11 –11

Not Foam –11 11

A hypothesis test determines whether these differences between observed

and expected frequencies are large enough to indicate a relationship between
insulation and rash.
146 problem solving, decision making, and professional judgment

In a 2-by-2 table, the four differences will always have the same magnitude, with
the cells along one diagonal negative and the cells along the other diagonal posi-
tive. This is because the four marginal frequencies are the same for both observed
and expected tables, so each row and column of Table 5.10 has to sum to 0.
So the difference from what’s expected under independence is 11 in each
cell, not 0. But is 11 large enough to allow us to conclude a relationship in the
population?
To determine whether the data are extreme enough to justify the conclusion
that there is a relationship, we calculate the probability, if independence were
true, that the difference would be 11 or greater. This probability is called the
p-value of the test. If the p-value is small (specifically, smaller than our chosen
significance level of 0.05), then these data are quite extreme to be due only to
chance—quite improbable if independence were true—and we conclude that a
relationship exists, i.e., that the data are statistically significant.
The p-value can be calculated by a computer program or found in a table.
While we’re not going to go into the details here, it is calculated using the chi-
squared test (pronounced “kai squared”) for a relationship between two qualita-
tive variables.
For our data, with a sample of 453 and a difference in each cell of 11, the
p-value is 0.03. That is, if insulation and rash were independent, there would be
a 3 percent chance that the difference in each cell were 11 or greater. This is quite
improbable—0.03 is less than the level we set of 0.05—allowing us to conclude
that a relationship between insulation and rash exists.
Step 5) State your conclusion.
Because this result would be very unlikely if insulation and rash were inde-
pendence true, we are justified in rejecting the null hypothesis and concluding
that there likely is a relationship between insulation type and the absence or
presence of a rash.

5.8.1 What Null Hypothesis Statistical Testing Is and Is Not

We have just reviewed the technique called null hypothesis statistical testing
(NHST) as applied to two qualitative variables (foam insulation and rash). We
will examine its application to other sorts of variables in the next chapter. It is
important to emphasize that NHST asks the question: “given that the null
hypothesis (H0) is true, what is the probability of getting these (or more extreme
data),” and not: “given these data, what is the probability that the null hypothesis
is true.”22 P (data| H0) is as radically different from P(H0|data) as P(pregnant|woman)
and P(woman|pregnant).23 Even experts sometimes get this wrong, and for an

22. See Jacob Cohen, The Earth is Round (p <.05), 49 Am. Psychol. 997 (1994).
23. In Section 8, we will examine Bayes Theorem, which in principle can convert P
(data| H0) into P(H0|data). But it requires assigning a value P(H0), the probability of the
null hypothesis prior to the experiment, which is often not known.
 introduction to statistics and probability 147

understandable reason: we really want to know the answer to the second ques-
tion, which tells us how big a deal we can make of the data.
In general, this does not pose a problem when the results of a study turn out
to be statistically significant. The concern, rather, is that NHST may produce too
many misses (Type II errors), where a real effect goes unnoticed. We will return
to this issue in Section 7.6.

5.8.2 Intuitions about Hypothesis Testing

An expert on organizational behavior is conducting a study of what factors con-
duce to a company’s long-term success. He selects fifty companies that have thrived
over many decades and identifies a number of factors—involving mission, organi-
zational structure, leadership, and the like—that are present in all or most of them.
The approach is sometimes called selection on the dependent variable because it
only examines cases where the particular outcome (successful companies) that
presumably depends on other factors (e.g., leadership) occurs. Is it sound?
Consider the broader question of which cells of a 2-by-2 contingency table
(such as that involved in the Terra Nueva case) you need to test a hypothesis of
independence. (See Table 5.11.)

table 5.11 which cells of a contingency table are important?

Rash

Present Absent
Insulation Foam A B
Not Foam C D

• Cell A. To what extent do the putative cause and effect happen together?
This can be phrased either as (1) when the putative cause occurs, to what
extent does the effect also occur?, or (2) when the effect occurs, to what
extent did the putative cause occur? Most people find it intuitively obvious
that this cell is necessary.
• Cell B. To what extent does the putative cause occur when the effect does
not? Consider the possibility that many Terra Nueva residents who live in
apartments with foam insulation do not have a rash.
• Cell C. To what extent does the effect occur even when the putative cause is
absent? Consider the possibility that many Terra Nueva residents have a
rash even though they do not live in apartments with foam insulation.
• Cell D. To what extent did neither the putative cause nor the effect occur?
Though less obvious, this cell is needed as well, as we describe immedi-
ately below.
148 problem solving, decision making, and professional judgment

Consider these two possible scenarios, in which cells A, B, and C are identical,
as shown in Table 5.12.

table 5.12 the necessity of the “no-no” cell

Rash

Present Absent

Foam 100 12
Insulation
Not Foam 20 0

Rash

Present Absent

Foam 100 12
Insulation
Not Foam 20 100

In the first scenario, 100 percent of the 20 residents of apartments without

foam have rashes, compared to 89 percent of residents of apartments with foam
(100 out of 112)—so it appears that living in an apartment with foam is safer
than living in an apartment without foam. In the second, only 17 percent (20 out
of 120) of residents of apartments without foam have rashes.
Which cells were missing in the study of successful companies—and how
does this affect your confidence in its conclusions?

The Sports Illustrated Cover Jinx

A week after he was featured on the cover of the very first issue of Sports
Illustrated, Major League Baseball player Eddie Mathews, suffered an
injury that forced him to miss seven games, giving rise to the notion that
being on the cover “jinxed” an athlete. In 2002, the editors reviewed all
2456 magazine covers to find that 37 percent of the featured athletes suf-
fered “measurable and fairly immediate” negative consequences. Alexander
Wolf, That Old Black Magic, Sports Illustrated, Jan. 21, 2002, 50–61.
How would you determine whether there was actually a Sports
Illustrated cover jinx?
 introduction to statistics and probability 149

5.9 cancer clusters, the texas sharpshooter, and multiple

testing

A seemingly high incidence of a disease among those exposed to a possible caus-

ative agent often gives rise to the victims’ belief—and often to a class action suit
alleging—that the agent is the cause of the disease. This was the case with the
association of childhood leukemia with contaminated drinking water in Woburn,
Massachusetts (the subject of the book and film, A Civil Action); the association
of respiratory disease with contaminated drinking water in Hinkley, California
(the subject of the film Erin Brockovich); the association of connective tissue
diseases with silicone breast implants; and the association of brain cancer
with electromagnetic radiation from electric power lines, microwave ovens, and
cellular telephones.
The first indication of an environmentally caused disease has often been a
pattern of instances of the disease coupled with the putative disease-causing
agent. For example, a Turkish physician noticed many cases of myeloid leuke-
mia among hatters. He investigated this and correctly deduced that it was because
of their heavy use of benzene as a cleaning fluid. Similar patterns link the inhala-
tion of asbestos with malignant mesothelioma; mothers’ use of DES during
pregnancy with their daughters’ vaginal cancer; and smoking with lung cancer.
If this is so, ask the British epidemiologists Ravi Maheswaran and Anthony
Staines, why are scientists so skeptical about cancer clusters (regions with unusu-
ally high cancer rates at some time)?24 They answer:25
[B]ecause, in practice, the yield has been very disappointing given the resources
and energy expended. In addition, it is fraught with technical difficulties.
The poor yield is well documented. The Centers for Disease Control in
Atlanta investigated 108 cancer clusters over 22 years but found no clear cause
for any of them.
A major problem with cluster investigation is the choice of boundaries
used to define the cluster. The suspicion of a cluster usually begins with the
identification of a group of cases of cancer. The boundaries defining this
cluster, for example the geographical area and time frame, are determined
later. This process defines the population from which the cancer cases arose.
The tighter the boundaries are drawn, the higher the cancer rate will be. This
has been described as the “Texas sharpshooter” procedure, where the sharp-
shooter first empties his gun into the barn door and then carefully draws the

24. Laypersons may have a different view than experts. See Phil Brown, Popular
Epistemology and Toxic Waste Contamination: Law and Professional Ways of Knowing, 33
Health and Social Behaviour 267 (1992).
25. Ravi Maheswaran and Anthony Staines, Cancer Clusters and Their Origins, 7
Chemistry and Industry 254–256 (1997).
150 problem solving, decision making, and professional judgment

target around the bullet holes.

For example, three cases of leukemia over a year in a small town are not
surprising. But what if these three cases can be linked in some way, say the
subjects live in the same neighborhood, go to the same church or used to go
to school together? Although this can seem alarming, it is almost always pos-
sible to define boundaries that will create a disease cluster—or that will make
a cluster disappear. It is not possible to do a conventional statistical analysis
in such a situation.
Maheswaran and Staines also discuss the problem of multiple comparisons
(which we introduced in Section 5.8):
A further statistical problem is that of multiple comparisons. If, for example,
rates for several cancers are examined, perhaps in several age groups, it is
likely that a “statistically significant” elevated rate will be found. It isn’t just
researchers searching through large databases who do this. Members of the
public or media may also inadvertently make multiple comparisons if they
are on the lookout for something unusual. Indeed, this is the origin of many
reported cancer clusters . . .
Our readiness to see clusters of diseases is another instance of our intuitions
about randomness discussed in Section 5.7—the intuition that in eight tosses of
a fair coin HHTHTHTT is more likely than the sequence HHHHHHHH. Here
is a randomly generated list of 48 0’s and 1’s:
0101100010010011101110101111000101110111111110101
Note that there is a run of eight 1’s, completely by chance. This is analogous
to a cancer cluster with no environmental cause specific to the geographical
area. To people in that area, it looks alarming, but actually, such variation occurs
by chance.
Suppose that the average annual incidence of a particular form of cancer is 1
percent, and suppose that a neighborhood has 1000 residents. On average, we
would expect 10 residents of the neighborhood to be diagnosed in a year. The
population of the United States is 293,000,000, so one can draw borders to
define 293,000 neighborhoods of 1000 residents each. Simply as a matter of
chance, some neighborhoods will have zero cases of cancer and some will
have 50.
When conducting many tests, we should expect to see at least a few statisti-
cally significant results by chance, and should correct our methods to take this
into account. If we are not keeping track of all of the possible tests we could
have conducted, our statistical analyses will be skewed in favor false positive
results (Type I errors). If we use appropriately conservative statistical tests,
keeping track of the number of tests we have conducted, and still find statisti-
cally significant results, then it seems that further inquiry is warranted. The
further inquiry might be of an epidemiological nature—for example, looking at
 introduction to statistics and probability 151

the cancer rates in other places with similar drinking water contamination. Or
it might involve controlled experiments with laboratory animals—also essen-
tially statistical. Finally, it might involve learning how the chemical alters genes
to produce the cancer.
Multiple testing, along with various biases, can affect the power, and even
validity, of studies that have statistically significant results. In the provocatively
titled, Why Most Published Research Findings Are False,26 John Ioannidis notes
that the hotter a scientific field (with more scientific teams involved), the less
likely the research findings are to be true. He argues that the possibilities for
improvement lie mainly in taking account of the prestudy odds of the truth of
the relationship that the study is testing. We will consider the notion of prior
probability in the introduction to Bayesian statistics in Section 8.5. But consider
these possible determinants:
• data from previous experimental studies;
• data from previous epidemiological or other observational studies;

Legal Problem

A plaintiffs’ law firm that specializes in toxic and related torts has a particular
interest in cancers putatively caused by strong electromagnetic fields. It has been
monitoring one hundred neighborhoods near large transformer stations. One of
those neighborhoods recently reported what appears to be a high incidence of
brain tumors in children. After informing several of the unfortunate families of
their legal rights and being retained by them, the firm undertook a well-designed
survey, which showed a statistically significant incidence of brain tumors. Does
this raise any concerns for you as a statistically astute observer of the legal
system?

• knowledge about the mechanics of causality—e.g., the effects of a particu-

lar chemical on cell division;
• general social, economic, or scientific theories—e.g., the general effects of
incentives and deterrents;
• an expert’s informed intuition;
• common (uninformed) intuitions.

26. John P. A. Ioannidis, Why Most Published Research Findings Are False, 2 PLOS MED
e124 (2005).
152 problem solving, decision making, and professional judgment

5.10 glossary

Conditional probability, The probability that A will occur given than B has
P(A|B) occurred

Independence Events A and B are independent when P(A|B) =

P(A) and P(B|A) = P(B)

Random sample A subset of a population, chosen by a random

mechanism

Simple random sample A subset of a population, chosen by a random

mechanism in which each population member
has an equal chance of being chosen

Sampling bias (selection bias) The amount by which an estimate is off from the
population parameter it is estimating due to the
way the sample was chosen from the population

Statistical inference Making conclusions about a population based

upon a sample from it

Variable A characteristic that can take on more than one

value

Quantitative variable A variable that takes on numerical values

Qualitative or categorical A variable that can be any of a number of

variable categories

Random variable A variable whose value is determined by chance

6. scores, dollars, and other
quantitative variables

6.1 continuous variables

The Terra Nueva problem involved two qualitative variables—foam insulation

and rash. Many of the issues that concern policy makers and lawyers involve
quantitative variables as well. Sometimes, policy makers may just need a picture
of a population—for example, the age distribution of residents of the United
States. Sometimes, they may want to compare two different populations—for
example, the income of men and women holding the same job. This would
involve one qualitative variable (sex) and one quantitative variable (income). And
policy makers often are concerned about the relationship between two quantita-
tive variables, say the relationship between per-pupil expenditure and academic
achievement.
We introduce the statistical concepts underlying such studies by trying to
learn whether Big-Mart is discriminating against female assistant managers
by paying them lower wages than their male counterparts. We will look at the
relationship between wage and gender at the Big-Mart stores, as well as other
relevant information, to help determine this issue.

6.1.1 Central Tendency

Suppose that you were head of Big-Mart’s human resources division, and were
asked for information about all assistant managers’ salaries. If you could com-
municate only a single number, what would it be? Most likely you would provide
information relating to the central tendency of a sample of salaries. This means
a number that indicates a salary that is somehow typical of the population. There
are three commonly used descriptive statistics that describe the central tendency
of quantitative distributions—the mean, the median, and the mode.
The mean is the average. So if you pulled five assistant managers’ files at
random, and their salaries were $36,857, $37,532, $39,774, $40,830, $42,797, the
mean salary of these employees would be (36,857 + 37,532 + 39,774 + 40,830 +
42,797)/5 = $39,558.
The mean is a good way to indicate the central tendency of a set of data, and
is the statistic most often used to describe quantitative variables. However, it is
sensitive to outliers. An outlier is an observation that is very different from the
other observations. Outliers can be the result of data entry errors, or they may
reflect an aspect of the population unrelated to your concern. Consider the assis-
tant manager in our small sample with a salary of $42,797. If the clerk doing data
154 problem solving, decision making, and professional judgment

entry mistook the four for a nine in that survey answer, the mean would change
from $39,558 to $49,558, a $10,000 increase just as a result of the change in
that single salary. Or his salary might be atypically high because he was the
boss’s son.
In cases in which it is important to avoid the effects of outliers, the median is
a better measure of central tendency. The median is the number that falls in the
middle of the observations, when they are ordered by magnitude. (If there are an
even number of observations, the median is the average of the two observations
that fall in the middle when the scores are ordered.) So, for our five salaries, the
median is $39,774. You can see why the median will not be strongly affected by
outliers—the values of the extreme scores are not involved in the calculation.
Thus, the change in one salary that would lead to a $10,000 increase in the mean
would have no effect on the median. The median is often used in reports of
housing prices, since, in a neighborhood of mostly homes worth less than
$500,000, just one multimillion dollar home could increase the mean by
thousands.
In some cases it is useful to consider the mode of a set of data. The mode is
the value that occurs the most times. The term has the same origin as “mode” in
fashion—the most popular number.

6.1.2 Dispersion, Variance, and Standard Deviation

So, if you were allowed to report only one piece of information about a set of
salaries, some measure of central tendency would be a good candidate. What
would be your choice for a second piece of information? Central tendency neglects
an important aspect of a set of numbers, its dispersion. For example, the set of
five salaries given above have a mean of $39,558. But the same mean might have
come from a set of five in which three employees earn $19,550 and the other
two earn $69,570, or from a set of five in which all five earn $39,558. Dispersion
is how spread out the data are. The former set has large dispersion, while the
latter has small (actually zero) dispersion. Paying attention to dispersion lets us
discriminate between bunched and spread data.
Figure 6.1 shows two sets of data. Which has greater dispersion?

(a) 35,000 37,500 40,000 42,500 45,000

(b) 35,000 37,500 40,000 42,500 45,000

figure 6.1 examples of dispersion.
 scores, dollars, and other quantitative variables 155

When the mean is used to measure central tendency, the standard deviation
(SD) is usually used to measure the dispersion. The standard deviation of a pop-
ulation is an indication of the amount by which the numbers differ from the
mean.
How do we determine the standard deviation? We want to know how far the
five salaries deviate from the mean salary. Since we do not know the population
mean,1 we estimate it with the mean of our sample of five, $39,558. For example,
by subtracting $39,558 from each of our five salaries, on the high side we have
what is shown in Table 6.1(a).
table 6.1(a) deviation of high salaries from the mean

Salary 39,774 42,797 40,830

Deviation 216 3239 1272

And on the low side, we have what is shown in Table 6.1(b).

table 6.1( b) deviation of low salaries from the mean

Salary 36,857 37,532

Deviation -2701 -2026

Because by definition the mean is in the middle, the average of the deviations
themselves will always be zero. To get a useful measure of the amount of disper-
sion in the sample we need to make the numbers positive and give extra weight
to those observations further from the mean. To do so, we square each observa-
tion’s deviation from the mean and then take the average of all those deviations,
which is 5,888,960.2
This gives the variance, which is a calculation of how far the data points
deviate from the mean. But because we squared each deviation, the value of the

1. See Section 5.3.

2. Actually, we take a kind of average in which, instead of dividing by the number of
data points (denoted N), we divide by one less than that (N-1), in this case 4. (2162 + 32392
+ 12722 + (-2701) 2 + (-2026) 2)/4 = 5,888,960. Why N-1? We want to average the squared
deviations of the five salaries from the population mean, but, since it is unknown, we are
using the sample mean instead. The sample mean is calculated from our same sample of
five salaries, so it will be especially close to these particular five salaries. That is, because
it is calculated from this particular sample, the sample mean is closer to the sample values
than is the population mean. Therefore, the deviations we calculate will be smaller than
the deviations we really want, the deviations from the population mean. In order to make
up for this, we increase the estimate slightly by decreasing the denominator of the average,
from N to N-1.
156 problem solving, decision making, and professional judgment

variance is entirely incommensurate with salaries. Therefore, we take the square

root of the variance to get the standard deviation, in this case √5,888,960 =
$2,427.
A small standard deviation indicates that scores vary little around the mean,
while a large standard deviation means that they vary a lot.
6.1.2.a A Sample of One Hundred Salaries with a Mean of $38,627 For rea-
sons we will come to below, a sample of only five assistant managers’ salaries is
usually too small to do useful statistical analyses. Therefore, we based the rest of
the chapter on a sample of one hundred salaries, described in the Appendix
(Section 6.4). Whereas our sample of five salaries had a mean of $39,558, the
sample of one hundred has a mean of $38,627. The standard deviation for the
large sample turns out to be $1,800. If you wonder why the samples might have
different means, stay tuned.

6.1.3 Normal Distributions

In Section 5.2, we introduced the concept of probability distributions for qualita-
tive variables. Quantitative variables also have probability distributions. While
qualitative variable have discrete amounts of probability (e.g., 0.5 on heads, 0.5
on tails), the probability distribution of a quantitative variable is a continuous
function of the possible values of the variable.
A common probability distribution of quantitative variables is the normal dis-
tribution, whose graph is the bell curve. Height, weight, and many other natural
phenomena tend to be characterized by a normal distribution.
Although not inevitable, it is not unusual for the salaries of a group of employ-
ees to be normally distributed. If you were to take the sample of one hundred
assistant managers and plot their salaries on the vertical axis, you would end
up with approximately a bell curve. This would be the empirical distribution
of salary (analogous to Figure 5.1(b)). If you knew all the Big-Mart assistant
managers’ salaries, you would have the population distribution of salary
(analogous to Figure 5.1(a)), which would even more closely approximate the
normal distribution. Figure 6.2 shows a normal distribution of the salaries of all
Big-Mart assistant managers.
One important quality of a normal distribution is that its mean, median, and
mode are all the same. To find the mean, draw a vertical line down from its peak
to the horizontal axis.
Another property of a normal distribution is the 68/95/99.7 rule: 68 percent
of the population fall within a range that is one standard deviation to the left and
one standard deviation to the right of the mean, 95 percent fall within two stan-
dard deviations to either side, and 99.7 percent fall within three standard devia-
tions to either side. This is true, regardless of how large or small the standard
deviation is—that is, regardless of the population’s dispersion.
 scores, dollars, and other quantitative variables 157

Probability Distribution of Salary

0.00020
0.00015
0.00010
Density
0.00005
0.00000

32000 34000 36000 38000 40000 42000 44000 46000

Salary

figure 6.2 the population distribution of assistant managers’

salaries.

So if the salaries of Big-Mart assistant managers are normally distributed, and

you sample a salary at random, the probability is about 0.68 that it will be within
1 SD, or $1,820, of the mean, and 0.95 that it will be within 2 SD, or $3,640.
A competitor of Big-Mart is the Bulls-eye chain of stores. As a check on
your understanding of the idea of standard deviation, consider Figure 6.3.
Panel (a) displays the probability distribution of salary among Big-Mart assistant

Distribution of Salary - Big - Mart Distribution of Salary - Bullseye

0.00000 0.00005 0.00010 0.00015 0.00020

0.00000 0.00005 0.00010 0.00015 0.00020

Density

Density

25000 30000 35000 40000 45000 50000 55000 25000 30000 35000 40000 45000 50000 55000
(a) Salary (b) Salary

figure 6.3 (a) and (b) distributions of salaries having the same mean
but different standard deviations.
158 problem solving, decision making, and professional judgment

managers (the same as Figure 6.2 but graphed on a different horizontal scale),
and panel (b) displays the probability distribution of salary among Bulls-eye
assistant managers. Bulls-eye assistant managers have the same mean salary as
do Big-Mart assistant managers. Which population distribution has the greater
standard deviation?
The standard deviation tells how useful the mean of the sample of salaries is
in estimating the mean of the population of assistant managers’ salaries. If we
could look at a second number (after the sample mean) describing the distribu-
tion of salaries, the standard deviation would be a good choice.
You can visualize the bell shape of a normally distributed curve as an upside-
down bowl. Since the inside of the bowl is facing down, we say that there the
curve is concave downward. On the left and right sides of the curve, it is the
shape of a part of a right-side-up bowl. The inside of the bowl faces up, and this
part of the curve is called concave upward. An inflection point of a curve is a
point at which the curve switches from being concave up to concave down or vice
versa. In a normal distribution, there are two inflection points, and they are each
exactly one SD away from the mean. Using this fact is a good way to get a rough
idea of the SD when you have only a graph of the probability distribution.

6.1.4 Estimating the Population Mean

As in Terra Nueva, we have data on only a sample of the total population. The
mean of the sample is our best estimate of the mean of the population. But how
precise is the estimate?
The statistic describing the precision of an estimate is its standard error (SE).
The SE of the sample mean is the SD of its sampling distribution. To understand
this new term, consider:
Recall that the sample mean of the one hundred Big-Mart salaries is $38,627.
We want an idea of how far off this estimate is from the true population
mean. Suppose that we now take ninety-nine more samples of one hundred
managers each, and calculate the mean of each of these hundred samples. The
distribution of these hundred means—which are likely not to be all the same—is
the sampling distribution. This distribution is always normal distribution (even
if the samples themselves are not), and it will always have a smaller SD than
most of the samples on which it is based. The distribution might look like
Figure 6.4.
The mean of means, the peak of the sampling distribution, is not likely to be
hugely different from the mean of our first sample. But the standard error—the
SD of the sampling distribution—will be much smaller than the SD of this one
sample.
Why so? When you take a mean of each sample, you are getting rid of some
of the variation. Accordingly, the variation of the means of a bunch of samples is
likely to be smaller. Another way to think about it is to ask yourself which would
be easier to guess: the salary of the next assistant manager chosen at random, or
 scores, dollars, and other quantitative variables 159

Sampling Distribution of Mean of 100 Salaries

0.00020
0.00015
Density
0.00010
0.00005
0.00000

25000 30000 35000 40000 45000 50000 55000

Sample Mean of 100 Salaries

figure 6.4 sampling distribution of the mean of 100 salaries.

the mean of the next one hundred salaries chosen at random? In the mean of
100, an extreme score is likely to be canceled out when combined with the other
99. On the other hand, it is quite possible that one randomly chosen salary will
be extreme.
So how far from the population mean is any sample mean likely to be? Rather
than take a hundred different samples, you can just use this formula to calculate
its standard error (SE)—

SD
SE =
N

—where SD is the standard deviation of the sample and N is the size of the
sample. The smaller the SE, the better an estimate our sample mean is of the
mean of the population.
Looking at the denominator, you’ll see that the larger the sample, the smaller
the standard error—though the SE only decreases by the square root of the size
of the sample. As we obtain more and more observations, the distribution of
the sample mean becomes tighter and tighter about the population mean, the
standard error decreases, and the precision with which a sample mean estimates
the population mean increases.
160 problem solving, decision making, and professional judgment

Why is the mean of a larger sample a more precise estimate of the population
mean than the mean of a small sample? Let’s go back to the discussion of cate-
gorical variables early in the previous chapter. Suppose that you didn’t know
anything about the mechanics of coin tosses or dice rolls: how much informa-
tion does one toss of a coin that lands on heads, or one roll of a die that lands on
four, provide about the chances of coins coming up heads or dice landing on
four? What about two tosses or rolls? What about a thousand? So, too, the mean
of the sample of many salaries provides a better estimate than the mean of a
sample of one, or even five.
Looking at the numerator, you’ll see that the smaller the standard deviation of
the sample, the smaller the SE. Recall that the standard deviation of a sample is
a measure of the dispersion of the items in the sample, in this case the disper-
sion of salaries. Which is more dispersed: a sample of the population of Big-Mart
assistant managers or a sample of Big-Mart employees (including everyone from
cleaning staff to executives)? And which population mean would you be better
able to estimate from their respective samples of one hundred: the population of
Big-Mart assistant managers or the population of all Big-Mart employees?
Notice that the formula for determining SE does not take account of the size
of the population from which the sample is taken. Our confidence in the rela-
tionship between the sample mean and the population mean is based solely on
the size and standard deviation of the sample and not on the proportion of the
sample to the population as a whole. If this seems counterintuitive, consider: as
long as a pot of soup is well stirred, one sip is all you need to taste the soup—no
matter what the size of the pot.
In our example, using the formula above, the SE of the sample mean is
$1820/√100 = $182.
Another common way to indicate our confidence in the estimate is to report
a range, called a confidence interval, around it that is likely (often with 95 percent
probability) to cover the true mean. (The confidence interval is the same as the
margin of error in a public opinion poll.)
Because the sampling distribution of the sample mean is normally distrib-
uted, we are 68 percent confident that the sample mean is within one SE of the
population mean, and 95 percent confident that it is within two SEs. The smaller
the SE, the smaller the confidence interval.
Thus, using our estimated SE of $182, we can be about 95 percent confident
that the mean score for the population is within $364 of our estimate:
$38,627 ± 2 * $182,
i.e., between $38,263 and $38,991. In other words, the estimate of $38,627 has
a margin of error of 2 * $182 = $364.3

3. More precisely, there is a 95 percent chance that the sample mean is within 1.96
standard errors of the population mean. A more precise estimate of the 95 percent
confidence interval for the population mean is therefore $38,627 ± 1.96 * $182.
 scores, dollars, and other quantitative variables 161

Sample vs. Population Mean

Though it is generally true that one never knows the population mean and must
estimate it from the sample, Sears, Roebuck, and Co. vs. City of Inglewood (Los
Angeles Superior Court, 1955) provides a rare counterexample. The City of
Inglewood imposed a half-percent sales tax on sales made by stores to resi-
dents living within the city limits. Sears sued to recover taxes erroneously
imposed on nonresidents. Sears retained a statistician, who randomly sampled
33 out of the 826 business days during the period at issue, and examined sales
slips from each of those 33 days. Based on the mean overpayments on these
days, the expert estimated that the total overpayment was $28,250 with a 95
percent confidence interval between $24,000 and $32,400. The judge refused to
accept the estimate and required Sears to prove its refund claim on each indi-
vidual transaction rather than on the sample information. Subsequently, a
complete audit of all transactions over the entire 826 days was performed and
the actual refund amount was determined to be $26,750.22—well within the
confidence interval.
R. Clay Sprowls, The Admissibility of Sample Data Into a Court of Law: A Case
History, 4 UCLA L. REV. 222 (1957).

6.1.5 Intuitions about Sample Size and Variability

As the preceding discussion indicates, estimates vary less when we have more
observations. We just relied on your intuitions to show this. But studies by
Daniel Kahneman and Amos Tversky indicate that this is not intuitively obvious
to many people. Kahneman and Tversky4 gave the following problem to Stanford
undergraduates:
A certain town is served by two hospitals. In the larger hospital about 45
babies are born each day, and in the smaller hospital about 15 babies are born
each day. As you know, about 50% of all babies are boys. The exact percentage
of baby boys, however, varies from day to day. Sometimes it may be higher
than 50%, sometimes lower. For a period of 1 year, each hospital recorded the
days on which more than 60% of the babies born were boys. Which hospital
do you think recorded more such days?
Take a moment to write down your answer before reading on.
An extreme outcome is more likely to occur in a small sample, because an
average pulled from a small sample has the potential to be much more affected by
an unusual pattern of births and consequently has higher variance. You can think

4. Daniel Kahneman and Amos Tversky, Subjective Probability: A Judgment of

Representativeness, 3 Cognitive Psychology 430–454 (1972).
162 problem solving, decision making, and professional judgment

about this in the terms that we presented above: for each day, we compute an
estimate of the probability of a boy being born. This estimate is in fact the mean
of a set of observations, where each observation corresponds to a birth and is
either 1 (if a boy was born) or 0 (if a girl was born). The standard error of the mean
of 45 observations is less than that of the mean of 15 observations, so the confi-
dence interval on the percentage of boys born is tighter for the large hospital. Hence
we would expect more days with 60 percent boys from the smaller hospital.
The students’ responses are shown in Table 6.2. Only 20 percent of the
respondents answered correctly:

table 6.2 estimates of boy babies born in large and small hospitals
Response Frequency

Larger hospital 12
Smaller hospital 10
About the same (within 5%) 28

Kahneman and Tversky found several other circumstances in which people

fail to comprehend the relationship between sample size and variance, and
explained this failure in terms of the representativeness heuristic (discussed in
Section 8.3), in which people intuitively assign a probability to an event according
to its similarity to the processes that are supposed to produce it. A heuristic is an
unconscious mental shortcut that substitutes for conscious analysis. In this case,
if the process is known to produce boys and girls with equal probabilities, the
heuristic suggests that any random sample of births will produce an equal
number.
The availability heuristic (described in Section 9.6) can also affect people’s
intuitive inferences from small samples. In an experiment, students were sig-
nificantly more influenced by personally hearing two or three fellow students
evaluate a particular course than by the mean course evaluations based on a
sample of “dozens of previous students.”5

6.1.6 Comparing Two Groups: Testing for a Relationship Between One

Quantitative Variable (salary) and One Categorical Variable (sex)
So far we have examined a sample of one hundred assistant managers at
Big-Mart to ask how confident we could be that the mean of their salaries reflects
the mean of the population of all assistant managers. Here we build on this
scaffolding to compare two groups of assistant managers: men and women.

5. Eugene Borgida and Richard E. Nisbett, The Differential Impact of Abstract vs.
Concrete Information on Decisions, 7 J. Applied Social Psychology 258 (1977).
 scores, dollars, and other quantitative variables 163

In particular, we would like to know whether male and female assistant

managers at Big-Mart have different salaries, on average.
The human resources director of Big-Mart was motivated to ask this question
when, in reviewing the sample of one hundred assistant managers, which
includes forty-nine men and fifty-one women, she noticed that the groups had
different mean salaries:
mean salary of males: $39,257,
mean salary of females: $38,528.
Thus, in our sample, on average, male assistant managers make $729 more
than female assistant managers.
While males have a slightly higher mean in our sample, we need a statistical
test to determine whether this is just due to the chance involved in sampling
from the population, or whether there is a real difference in the population.
Thus, we need to address the question whether, in the whole population of Big-
Mart assistant managers, there is a relationship between salary and sex—whether
we can say with sufficient confidence that the curves for the population of male
and female assistant managers also diverge.
We engage in the process of hypothesis testing that we used to determine
whether there was a relationship between foam insulation and rashes in Terra
Nueva. We’ll go through the same five steps as before.
Step 1) State the hypotheses of the test:
• The null hypothesis indicates that there is no relationship between two
variables—that the two curves have the same mean (Figure 6.5). It is what
we set out to reject.

Null Hypothesis
0.00020
Density
0.00010
0.00000

36000 38000 40000 42000

Salary

figure 6.5 null hypotheses.

164 problem solving, decision making, and professional judgment

• The alternative hypothesis is that there is a relationship between two vari-

ables—that the two curves have different means. Figure 6.6 shows one of
myriad possible alternative hypotheses. (One rationale for null hypothesis
testing is that there is only one null hypothesis.)

Alternative Hypothesis
0.00014

Females Males
Density
0.00008
0.00002

36000 38000 40000 42000

Salary

figure 6.6 alternative hypotheses.

In our example, the null hypothesis is that there is no relationship between

male and female assistant managers’ salaries.
Step 2) Set the level of the test. Let’s set it at α = 0.05, meaning that we are
willing to allow a 5 percent chance that we conclude statistical significance even
if no relationship exists.
Step 3) Suppose that the null hypothesis is true. The main idea of a statistical
hypothesis test is to find how extreme the sample data would be if the null
hypothesis were true.
Step 4) Calculate how extreme the sample data are. How likely or unlikely
would it be to see a difference in mean salaries of at least $729 in a random
sample if that difference were $0 in the population? We calculate this probability
with the t-test (using statistical software).6 While we won’t show the calculation
here, the probability (the p-value) of having data this extreme is 0.03. In other

6. The t-test considers the difference of $729 relative to the variance of the data, and
compares it to the t-distribution, a probability distribution like the normal distribution but
that varies according to the size of the sample.
 scores, dollars, and other quantitative variables 165

words, under the null hypothesis of no salary difference in the population, there
would be a 3 percent chance that, in samples like ours, the mean salaries of the
two groups would differ by at least $729.
Step 5) State your conclusion. If the p-value is less than the alpha level, we
reject the null hypothesis in favor of the alternative. In our case, the p-value is
0.03. This means that, under the null hypothesis of no relationship between sex
and salary, only 3 percent of samples would be more extreme than ours. Since
this is less than 5 percent (the level of Type I error we were willing to tolerate) we
can reject the null hypothesis. We conclude that there is a relationship between
sex and salary.
So we have concluded that there is a statistically significant difference between
the salaries of male and female assistant managers at Big-Mart. But can we con-
clude that Big-Mart is discriminating based on sex in setting salaries? Perhaps
there is another variable that explains the difference. We explore this possibility
in Section 6.2.7.b below.

6.2 regression

In Section 5.6, we encountered the definition of independence between vari-

ables.
• If variables are independent, the value of one of them provides no informa-
tion about the value of the other. For example (so far as we know), the days
of the week on which people are born provides no information about their
height or weight.
• If variables are not independent, differences in one correspond to differ-
ences in the other. For example, height and weight are related: taller indi-
viduals tend to be heavier.
Much of statistics is concerned with whether, how, and how much variables
are related.
In this section, we look at the relationship between two quantitative variables.
We use an example from an important realm of public policy: education. In par-
ticular, we will look at the relationship between a state’s spending on public edu-
cation and student achievement as measured by scores on the NAEP (National
Assessment of Educational Progress) test.7

7. Each state’s NAEP scores aggregated from individual student scores, using sam-
pling methods discussed in the preceding chapters. Tests are administered to a random
sample of students in schools that themselves are chosen to be representative of schools
and students in their states selected to participate. NAEP typically selects 3000 students in
approximately 100 schools in each state for each grade and subject.
166 problem solving, decision making, and professional judgment

235

MA
205 210 215 220 225 230
Average 4th Grade Reading Scores

NH VT
VA DE CT
ND MT MN ME
CO
WA WY PA NJ
SD OH NY
ID
UT IAKS NE
MO WI
KY
FL MD
TX IN MI
NCAR OR IL RI
GA WV
OKTN SC
AK
LA HI
AL
AZ NV NM
CA
MS
200

5 6 7 8 9 10 11 12 13 14 15
Per-Pupil Spending in $1000s

figure 6.7 scatterplot of school spending and reading scores.

6.2.1 Line of Best Fit

A good way to gauge the relationship between two quantitative variables, such as
per-pupil spending and achievement, is to produce a scatterplot. This is a graph
with one variable along each axis, with each instance of data represented by a
point. Figure 6.7 shows the relationship between states’ per-pupil spending in
public education and fourth grade reading achievement. The horizontal (X) axis
measures per-pupil spending, ranging from $4,000 to $12,000. The vertical (Y)
axis measures NAEP reading test scores, on a scale that ranges from 200 to 235.
In this scatterplot, each point represents one state. For example, looking at
the leftmost datapoint, Utah spends $5475 per pupil and its average fourth
graders score is 221.3. Moving to the right, Idaho, Oklahoma, Arizona, and
Mississippi spend in the mid to high $6000s per pupil, and have average reading
scores ranging from 222 to 204.
Looking at the scatterplot is a good way to get an idea of the relationship
between two quantitative variables. In this section we will use statistics to assess
linear relationships. If two variables vary linearly with each other, the rate of
change of one compared to the other is constant. For example, if achievement and
spending are linearly related, then, if spending increases by a certain amount,
achievement will increase by some constant amount. If the variables are linearly
related, the points cluster around a straight line, and the tighter their clustering
the stronger the relationship.
 scores, dollars, and other quantitative variables 167

A large number of relationships relevant to law and policy are approximately

linear—at least in the range of interest. For example, income may increase lin-
early with years of higher education, and the number of crimes in a particular
neighborhood may decrease linearly as the number of police patrols in the neigh-
borhood increases.
When studying two quantitative variables, we often consider one the predic-
tor or independent variable and the other the outcome, response, or dependent

Nonlinear Relationships

Nonlinear relationships are also common. For example, the number and severity
of traffic accidents increases with the speed at which vehicles are traveling, but
the increase is geometric (i.e., based on a power of the speed)—rather than arith-
metic. In this case, the relationship would not be shown by a line, but by an
upward-sloping curve. Or consider the relationship between room temperature
and productivity in the workplace. A scatterplot of points with horizontal position
indicating temperature and vertical position indicating productivity might have
a ∩ (inverted U) shape, with low productivity values at both extremes and high
productivity in the middle.
In cases where the relationships between the variables are monotonic—that is,
the slope of the curve does not change direction—the variables sometimes can
be transformed to make their relationship linear. For example, the radius of a
circle and its area are not linearly related, but the radius is linearly related to the
square root of the area. Other functions commonly used to linearize a relation-
ship are the square, the logarithm, and the reciprocal.

variable. When constructing a scatterplot, it is conventional to put the predictor

on the horizontal (or x) axis and the outcome on the vertical (or y) axis.
In our example, we would like to predict students’ achievement when we
know the state’s per-pupil spending, so spending is the predictor and achievement
is the outcome. If these two variables are linearly related, we can describe their
relationship with the line of best fit, also called the regression line. This is the
line that gives the best prediction of the dependent variable when the indepen-
dent variable is known. Figure 6.8 shows the line of best fit for the relationship
between spending and achievement.
To predict a state’s level of student achievement when we know its per-pupil
spending, we would find the point on the line directly above the known value of
spending, and its height would be our guess of achievement. For example, if we
168 problem solving, decision making, and professional judgment

235

MA
230
Average 4th Grade Reading Scores

NH VT
VA DE CT
225

ND MT MN ME
CO
WA WY PA NJ
SD OH NY
ID
UT IAKS NE
MO WI
220

KY
FL MD
TX IN MI
NCAR OR IL RI
215

GA WV
OKTN SC
AK
210

LA HI
AL
AZ NV NM
CA
205

MS
200

5 6 7 8 9 10 11 12 13 14 15
Per-Pupil Spending in $1000s

figure 6.8 regression of reading scores on spending.

know that a state spends $10,000 per pupil, our best guess is that its students
would have an average NAEP score of about 220.
Often the response (such as achievement) is represented by the letter y and the
predictor (such as spending) by x. The line of best fit is then represented by the
regression equation, which is also the general equation for a line:
y = a + bx.
Because we can use this line to predict the response when we know the pre-
dictor, it is also called the prediction equation. In our example:
Predicted achievement = a + b * spending.

• a is the intercept—the height at which the line of best fit intercepts, or

crosses, the vertical axis at $0 per pupil. This is not shown in Figures 6.7
and 6.8; it is off the chart to the left. In this purely hypothetical case, a is
205—a state’s average predicted achievement score if the predictor, per-
pupil expenditure, were $0. (Since $0 is outside the range of the data, we
would not venture to make that prediction; although the relationship
between spending achievement appears linear in the range of the data we
do not have evidence that it is linear elsewhere, and it is dangerous to make
extrapolations outside the range of the actual data.)
 scores, dollars, and other quantitative variables 169

• b is the slope of the line—the amount the line goes up for each unit it goes
right. b is also called the coefficient of x. 1.5 is the slope of this regression
line. With an increase in spending of $1000, the predicted average score
increases by 1.5.
• x is the predictor, in this case spending.
• y is the predicted outcome, in this case achievement.
While x and y vary from state to state, a and b represent constant numbers.
If b is positive, the line slopes up, so the outcome increases when the predictor
increases. If b is negative, the line slopes down, so the outcome decreases when the
predictor increases. a and b can be calculated by any statistical computer program.
If the slope (b) equaled zero, we would have a horizontal line, which would
correspond to no relationship between the predictor and the outcome. In our
example, no matter how much was spent, we would not change our prediction of
achievement.
In our example, if spending increases by $1000, the predicted outcome y (achieve-
ment) increases by 1.5 points on the NAEP test. So our regression equation is
Predicted achievement = 205 + 1.5 * spending (in thousands).
For example, if a state spends $8000 per pupil, our best guess is that its fourth
graders will on average score 217 on the reading test. You can observe this by
looking at the regression line in Figure 6.7.
The difference between a data point’s observed outcome and its predicted
outcome is called the error of the prediction. In Figure 6.7, we can see that
California spends $8,417 per pupil, and its students score 206.5 on average. But
the line of best fit predicts that students in a state with $8,417 of spending per
pupil will achieve scores of 217.7 on average. The difference of 11.2 is the error
of that observation.
The line of best fit is the line that minimizes the sum of the squared errors
over all observations in the sample. (Squaring each error ensures that it is posi-
tive and weights those observations that are further away.)

6.2.2 Correlation
Another way of saying that two variables are linearly related is to say that they are
correlated. The correlation coefficient (or correlation), denoted r, of two variables
indicates the strength of the linear relationship.
r has the same sign (positive, negative, or zero) as b, so, like b, it is positive
when the line of best fit slopes up (when the outcome increases with an increase
in the predictor) and is negative when the line slopes down (when the outcome
decreases with an increase in the predictor). However, the magnitudes of b and
r are generally not equal.
• b is the amount by which y increases for a one-unit increase in x. Two
states that differ in spending by $1000 differ in predicted achievement by b.
170 problem solving, decision making, and professional judgment

• r is the number of standard deviations that y increases for a one SD increase

in x. In this case r = 0.46.
Recall the discussion of SD earlier in Section 6.1.2. We won’t do the calcula-
tions here, but in this example, one SD in spending = $1,934, and one SD in test
scores is 6.45. So two states that differ in spending by one SD ($1,934) differ in
predicted achievement by r SDs, or 3.0.
r is always between -1 and 1. Unlike b, r is not affected by the scale of the vari-
ables. If spending were measured in dollars instead of thousands of dollars,
b would be 1000 times smaller, but r is the same no matter what the scale.
The correlation coefficient thus reflects two aspects of a linear relationship:
direction and strength.
Direction: A positive correlation coefficient indicates a positive linear relation-
ship, in which the two variables increase and decrease together. A negative
correlation coefficient indicates a negative linear relationship, in which, as
one variable increases, the other variable decreases. In this case, the regres-
sion line would slope downward from left to right and we say that the variables
are inversely or negatively correlated. (What is an example of two inversely
correlated variables?)
Strength: The correlation coefficient (r) also describes the strength of this
relationship: for a weak relationship, the correlation coefficient is small (close
to 0), while a stronger relationship will be reflected in a larger (closer to 1 or
-1) correlation coefficient. When two variables have a large correlation
coefficient, we say that they are highly correlated.
Let’s look at the extremes of r in the context of the NAEP reading score exam-
ple. Recall that b = 0 whenever r = 0. So if spending and achievement were not
correlated, the data points would be scattered randomly on the scatterplot, and
the line of best fit would be a horizontal line at the average achievement among
all fifty states, i.e., 217, as shown approximately in Figure 6.9. (Samples being
what they are, our effort to randomize did not quite succeed. What does
Figure 6.9 actually suggest about the correlation?
This means that per-pupil spending gives us no information at all about read-
ing achievement.
At the other extreme, suppose r = 1. Then we would say that spending and
achievement were perfectly correlated. As is evident in Figure 6.10, if you know a
state’s per-pupil expenditure, you can predict exactly its fourth grade reading
scores.

6.2.3 Regression to the Mean

The name regression was coined by Francis Galton, who developed the concept
while exploring the relationship between the heights of fathers and sons. Galton,
a cousin of Charles Darwin, was keenly interested in the nature of heredity and
explored its implications in a number of domains. When examining the heights
 scores, dollars, and other quantitative variables 171

235 NE OK
PA
MD DE
UT
FL
230
Average 4th Grade Reading Scores

RI
AK MA
LA CT
NM SC
VT
225

NJ
IL KS
NC
NV SD
220

CO AR
AZ
WV VA
NH
215

KY IN
MN TN
WA ND OR
WI
210

HI TX
NY OH
AL WY
MI MT
IA
205

CA MO
ID
GA
MS
200

ME

5 6 7 8 9 10 11 12 13 14 15
Per-Pupil Spending in $1000s

figure 6.9 regression of reading scores on spending with (almost)

no correlation.
235

NJ
NY
CT
VT
RI
DE MA
ME
AK
230
Average 4th Grade Reading Scores

MDPA
WY
WI
OHMINH
IL
225

HI WV
INNE
MN
VA
MT
KS
GA
220

IA CA

NMOR
MO
CO
215

WA
ND
SCLA
TX
SD
210

KY
NC FLAR
AL
NVTN
205

MS
ID

OK
200

AZ
UT

5 6 7 8 9 10 11 12 13 14 15
Per-Pupil Spending in $1000s

figure 6.10 regression of reading scores on spending with perfect

correlation.
172 problem solving, decision making, and professional judgment

of men in successive generations, Galton found that while tall fathers tended to
have reasonably tall sons, the sons were seldom as tall as their fathers. Likewise,
very short fathers had reasonably short sons, but the sons were unlikely to be as
short as their fathers. Because it seemed like the children’s heights were moving
away from the heights of their fathers, and towards the mean height, Galton
dubbed this phenomenon regression to the mean.
6.2.3.a Regression and Chance Galton came to understand that this process
is not due to a characteristic unique to heredity, but to the involvement of chance.
Regression to the mean is a consequence of the fact that some chance is involved
in determining the height of an individual, i.e., that r is not equal to 1 or -1. By
chance in this context, we mean anything that is not explained by the predictor
(father’s height), for example, mother’s height or nutrition—that is, chance com-
prises both factors that we don’t know about and factors that we know about but
are not considering.
To see how the involvement of chance produces regression to the mean, con-
sider a case in which chance alone determines the outcome. Imagine that you
telephone two numbers selected from the phone book at random, and ask the
people at the other end to tell you their height. The first person you speak to
reports his height as 6’2”. Do you expect the next person to be of similar height,
or closer to the average height of people in the phone book?
When chance is the only process involved, it is intuitive that an extreme
event—like a very tall person—is likely to be followed by something less extreme.
Moderate outcomes are just more probable. But the same phenomenon holds
even when there is a relationship between pairs of events, as there is between the
heights of fathers and sons. While heredity contributes to determining a son’s
height, so do several other factors. Having a tall father pulls the son’s height
toward the tall end, but chance pulls it toward the middle. Consequently, regres-
sion to the mean takes place.
Just as factors other than a father’s height determine the height of his son, so do
many factors besides a state’s per-pupil expenditures determine fourth grade read-
ing scores. Consider that states may differ in families’ average education and aver-
age income, and that these factors may also play a role in student achievement.

6.2.4 Intuitions about Regression

Regression to the mean is not intuitively obvious, and even people trained in
statistics can fail to recognize and interpret it. Kahneman and Tversky8 gave the
following problem to graduate students in psychology:
The instructors in a flight school adopted a policy of consistent positive rein-
forcement recommended by psychologists. They verbally reinforced each

8. Daniel Kahneman and Amos Tversky, On the Psychology of Prediction, 80

Psychological Review 237–251 (1973).
 scores, dollars, and other quantitative variables 173

successful execution of a flight maneuver. After some spending with this

training approach, the instructors claimed that contrary to psychological doc-
trine, high praise for good execution of complex maneuvers typically resulted
in a decrement of performance on the next try. What should the psychologist
say in response?
Because performance on flight maneuvers is variable—especially among
novice pilots—a pilot who does exceptionally well on one maneuver is likely to
do less well on the next, regardless of the instructor’s response. By the same
token, a pilot who does exceptionally poorly on one maneuver is likely to do
better on the next. The quality of any particular student’s flight maneuvers may
look like a normal distribution, with his own mean score defining the center,
and exceptionally good and bad performances lying in the tails.
Though trained in statistics, none of the graduate students attributed the
phenomenon to regression to the mean. Rather, they thought that verbal rein-
forcement might not work for pilots or that positive reinforcement led to the
pilots’ overconfidence, and some doubted that the flight instructors had accu-
rately perceived the phenomenon.
More generally, people tend to engage in nonregressive predictions. An early
experiment by Kahneman and Tversky showed that subjects’ predictions of a
student’s grade point average (GPA) based on (a) the student’s grades in per-
centile form, (b) her success on a mental concentration task, and (c) a measure
of her sense of humor—variables that ranged from highly predictive to entirely
unpredictive—showed almost no regressiveness in any of the conditions.9
In this and similar studies, the subjects were given information that they
believed (whether or not extravagantly) to be diagnostic, i.e., predictive. When
diagnostic predictors are combined with predictors that subjects believe to be
nondiagnostic, this has the effect of diluting the diagnostic predictors and
producing more regressive predictions. For example, the strength of the predic-
tion that a social work client who had sadomasochistic thoughts was likely to
be a child abuser diminished when subjects were also informed that the client
“fixes up cars in his spare time.” This is not due to any enlightenment about the
regression phenomenon, but more likely to the operation of the representative-
ness heuristic, which made a person who fixes cars as well as having sadomas-
ochistic thoughts seem less typical of a child abuser than one who only has
sadomasochistic thoughts.10 (Other statistical and logical errors caused by the
representativeness heuristic are discussed in Section 8.3.)

9. Richard E. Nisbett and Lee Ross, Human Inference: Strategies and

Shortcomings of Social Judgment (Englewood Cliffs, NJ: Prentice-Hall, 1980).
10. Richard E. Nisbett, Henri Zukier, and Ronald E. Lemley, The Dilution Effect:
Nondiagnostic Information Weakens the Implications of Diagnostic Information, 13 Cognitive
Psychology 248 (1981). Nisbett and Ross, 154–55.
174 problem solving, decision making, and professional judgment

The winner’s curse refers to the phenomenon that, in an auction with many
bidders and high variance in their estimates of the value of the commodity, the
winning bidder will overestimate the value of the commodity and pay too much
for it. This may occur, for example, when oil companies bid for exploration
rights, or baseball teams bid for free agents, or companies compete to acquire
another company.
Richard Harrison and Max Bazerman link the winner’s curse to regression to
the mean.11 Imagine two normally distributed curves, one showing bidders’ esti-
mates of the value of the auctioned item and the other showing their actual bids.
Assume that the midpoint of the estimate curve is the actual value of the item.
The average bid will be lower than the average estimate, since people will bid less
rather than more than their estimate of the value of the item. The winning bid
will, by definition, be above the mean of bids, and will reflect one of the highest
estimations of the item’s value.This concept is illustrated in Figure 6.11.12

Bids (B) Estimates (E)

– –
B E
Variables Assumptions
–
E = Estimates 1. True value ≈ E
B = Bid 2. True value will be equal for
D = (Amount of discounting) all bidders
=E–B

figure 6.11 the winner’s curse 12 copyright © 1995, reprinted by

permission.

11. J. Ricardo Harrison and Max H. Bazerman, Regression to the Mean, Expectation
Inflation, and the Winner’s Curse in Organizational Contexts, in Negotiation as a Social
Process 69 (Roderick Moreland Kramer and David M. Messick eds., 1995). They draw on
an earlier paper by Bazerman and William E. Samuelson, I Won the Auction But Don’t
Want the Prize, 27 Journal of Conflict Resolution 618–634 (1983).
12. Id.
 scores, dollars, and other quantitative variables 175

By much the same psychology that people engage in nonregressive

predictions, bidders at an auction tend not to adjust their estimates
downward. And, in the same way, it leads to inflated expectations and systematic
disappointment.
Later chapters of the book will discuss biases—in addition to the representa-
tiveness heuristic— that may contribute to inflated expectations and to the pos-
sibilities for debiasing.

6.2.5 Determining How Much of the Variance in the Outcome Can Be

Explained by the Predictor
To recapitulate what we have learned about regression thus far: Suppose we
wanted to predict the average reading achievement of a particular state’s fourth
graders. If we knew nothing about that state’s per-pupil spending, our best pre-
diction would be the mean achievement of all fifty states’ averages. However, if
we knew its per-pupil spending, and had determined a linear relationship
between spending and achievement, we could use this information to improve our
prediction. Instead of predicting the mean achievement, represented by a hori-
zontal line on the scatterplot, we would predict the achievement of the point on
the line of best fit directly above the state’s spending. Since this uses additional
relevant information, it is likely to be a better prediction, closer to the state’s true
achievement.
In addition to knowing the strength of the correlation between the two vari-
ables, you may be interested in knowing the explanatory power of the predictor,
that is, how much of variation of the data can the predictor variable account for.
For example, how much of the variation in the height of sons can be accounted
for by their fathers’ height? How much of the variation in states’ fourth grade
reading scores can be accounted for by state spending? The more we can
account for, the better we understand this piece of the world and the better
we—as individuals or policy makers—can use the information for decision
making.
Recall the discussion of variance and standard deviation in Section 6.1.2: The
deviation of a typical data point from its prediction is measured by the sample’s
variance (or, alternatively, by its standard deviation, which is the square root of
the variance). We calculated the total variance of the sample. The total variance
is also called the variance around the mean, because it is the average of the
squared errors, where the error of an observation is its distance from the sample
mean.
While the line of best fit provides the best prediction of how a change in one
variable will affect another, there are many errors—points that are not on the
line of best fit.
Based on the data, we can estimate the variance around the line of best fit.
Instead of averaging the squared differences between state achievements and
176 problem solving, decision making, and professional judgment

the mean achievement, we average the squared errors, where the errors are the
vertical distances of the observations from the line of best fit; that is, the differ-
ences between each sample member’s (state’s) observed outcome and its pre-
dicted outcome based on its spending (see Figure 6.8). These errors are generally
smaller than the errors from the sample mean, so the average of their squares
will be smaller. In other words, the variance around the line of best fit is smaller
than the variance around the mean. This is because the regression line captures
some of the systematic changes in one variable as a function of the other.
The difference between these two variances is referred to as the variance
accounted for by or explained by the predictor, here spending. In other words, the
total variance of the outcome consists of two parts:

Total varianceof outcome Variance of outcome Variance of outcome

= +
(around its mean) explained by predictor due to chance

This is called the analysis of variance (ANOVA) equation. The better the pre-
dictor, the larger the proportion of the total variance it explains. This proportion
is therefore used as a measure of the usefulness of a predictor and is called the
coefficient of determination:

Variance explained by predicttor

Coefficient of determination =
Total variance of outcome around its mean

The coefficient of determination answers the question: by what percentage

does taking the predictor into account decrease the variance in the outcome?
It turns out that the coefficient of determination is r2—the square of the cor-
relation coefficient discussed above. r2 is the percentage by which knowing the
value of the predictor decreases the variance of the outcome. The greater r2, the
more the predictor accounts for or explains the outcome.
If the variance around the line of best fit is small compared to the total vari-
ance, then r2 is large, signifying a strong linear relationship. At the extreme, an
r2 of 1 (r of 1 or –1) means that all of the variance in the outcome can be explained
by the predictor: knowing the value of the predictor, you will know exactly the
value of the outcome. A small r2, on the other hand, signifies a weak linear rela-
tionship: using the line of best fit instead of the outcome mean does not decrease
the variance by much. An r2 of 0 means that taking the predictor into account
does not decrease the variance, or improve the prediction, at all.
In our example, the correlation between a state’s per-pupil spending and its
students’ fourth grade reading achievement is 0.46. Thus, spending accounts for
about 0.462 = 0.21, that is, 21 percent of the variance in achievement.
 scores, dollars, and other quantitative variables 177

r r2
Aspirin and reduced risk of heart attack 0.02 0.0004
Chemotherapy and surviving breast cancer 0.03 0.0009
Calcium intake and bone mass in post-menopausal women 0.08 0.0064
Ever smoking and subsequent incidence of lung cancer
within 25 years 0.08 0.0064
Alcohol use during pregnancy and subsequent premature
birth 0.09 0.0081
Effect of nonsteriodal anti-flammatory drugs (e.g., ibuprofen)
on pain reduction 0.14 0.0196
Gender and weight for U.S. adults 0.26 0.0676
ECT for depression and subsequent improvement 0.29 0.0841
Sleeping pills and short-term improvement in chronic
insomnia 0.3 0.09
Age and episodic memory 0.33 0.1089
Elevation above sea level and lower daily temperatures in
the U.S. 0.34 0.1156
Viagra and improved male sexual functioning 0.38 0.1444
Age and reasoning 0.4 0.16
Weight and height for U.S. adults 0.44 0.1936
Age and speed 0.52 0.2704
Gender and arm length 0.55 0.3025
Nearness to equator and daily temperature in the U.S. 0.6 0.36
Gender and height for U.S. adults 0.67 0.4489

Timothy Salthouse, “Cognitive Research and Aging” (workshop on neuroeconomics and

aging, Stanford University, Stanford, CA, March 31, 2006).

figure 6.12 some examples of correlation strengths.

Does an r2 of 0.21 strike you as large or small? Correlations considerably

smaller than this are not at all unusual in the realms of medicine, science, and
social science, and can be quite meaningful. A correlation of 0.23 (r2 = 0.05)
between administration of the antiretroviral drug AZT and the survival of AIDS
patients represented a reduction of deaths from 61.5 percent to 38.5 percent, and
led to the termination of the clinical trials so that the control group could receive
the drug as well.13
In your own personal life, would you take seriously the fact that eating certain
foods, or smoking, accounts for 21 percent of the variance in your chance of
developing heart disease?

13. Paul Slovic and Ellen Peters, The Importance of Worldviews in Risk Perception, 3
Risk Decision and Policy 165–170 (1998), citing Robert Rosenthal, How Are We Doing in
Soft Psychology, 45 American Psychologist 775–7 (1990).
178 problem solving, decision making, and professional judgment

6.2.6 Statistical Significance and Hypothesis Testing

The NAEP reading scores are based on a sample of pupils in a sample of schools,
and it is possible that a correlation between spending and achievement in the
sample does not indicate a correlation in the population. It is possible that the
correlation in the sample is a result of chance; so we need a way to test whether
the relationship we see in the sample is statistically significant. The approach to
hypothesis testing is not essentially different from that discussed in the previous
chapter.
Step 1) State the two hypotheses of the test.
Recall from Section 6.2.1 that we represent the line of best fit with the regres-
sion equation—

y = a + bx

—where x represents the predictor (spending), y the outcome (achievement), and

a the intercept of the line and b its slope. A slope of 0, or equivalently a correla-
tion coefficient r of 0, would indicate that there was no linear relationship
between x and y. This gives us our null hypothesis: b = 0. Our alternative hypoth-
esis is: b ≠ 0.
The slope of the line of best fit of our sample was 1.5, not 0. But if we had
randomly chosen a different set of students from different schools, our b might
have been 0 or even negative. To learn whether b is significantly different from 0,
we do a hypothesis test.
Step 2) Set the level of the test.
We set the level to the customary α = 0.05.
Step 3) Suppose that the null hypothesis is true.
The value b = 1.5 per $1000 spent is specific to our sample. Suppose in the
population, b = 0, i.e., there is no true linear relationship between spending and
average score. In the sample, we still would observe a slope not exactly 0, because
of chance. We want to know: how likely or unlikely would it be to observe (in the
sample) a slope as large as b = 1.5?
Step 4) Calculate how extreme the sample data are, compared to what we
would expect. How likely or unlikely would the observed measurements be if the
null hypothesis were true?
In this example, if the true (population) value of b were 0 (no linear relation-
ship), the probability that the sample value of b is 1.5 or greater (or –1.5 or less)
is 0.001. This is calculated (with statistical software) using the t-test for a regres-
sion line slope, which considers the observed b (1.5) relative to the estimated
variance and compares it to the t distribution.
Step 5) State your conclusion.
Since the above probability is quite small (less than our set level of 0.05), we
are justified in rejecting the null hypothesis, and we conclude that there is a
 scores, dollars, and other quantitative variables 179

linear relationship between a state’s per-pupil spending and its fourth graders’
reading achievement.

6.2.7 Multiple Regression

Thus far, we examined linear regression of a quantitative outcome on one quan-
titative predictor. This is called simple linear regression. But if a single variable
accounts for only some portion of the variance of the data, perhaps we can
improve our prediction by taking account of other variables. In Galton’s study,
for example, we might improve our prediction of sons’ heights by taking account
not only their fathers’ height, but their mothers’ as well.
When we consider more than one variable that might affect the outcome, we
use multiple regression. The predictive variables collectively are called the
explanatory variables. With two explanatory variables, for example, the multiple
regression equation is
y = a + b1x1 + b2x2.
In our example, the regression equation is of the form
Predicted height = a + (b1 * father’s height) + (b2 * mother’s height).
As in simple regression, the coefficient of a variable in multiple regression is
the amount by which the predicted outcome would increase if that variable were
to increase by 1. But in multiple regression, it is that amount, assuming that all of
the other explanatory variables remain constant.
In our effort to predict reading achievement, we might regress on another
variable in addition to spending:
Predicted achievement = a + (b1 * spending) + (b2 * another variable).
What might you hypothesize as other variables that could account for a part
of the variance of average reading achievement among the states? As its name
implies, multiple regression analysis can look at any number of variables at the
same time.
6.2.7.a Using Multiple Regression Analysis in Employment Discrimination
Cases Multiple regression analysis is a common tool in determining whether an
employer is discriminating based on sex, race, or another forbidden characteris-
tic. Recall that in Section 6.16, we concluded that there was a statistically signifi-
cant difference between the salaries of male and female assistant managers at
Big-Mart. But perhaps something other than gender accounts for the difference.
For example, (1) salaries may be based on management experience, and (2) male
assistant managers may, on average, be more experienced than females.
Regression analysis can test both of these hypotheses.
The same personnel files that contain employees’ salaries also have informa-
tion about their experience. A simple regression of salary on number of months
180 problem solving, decision making, and professional judgment

of management experience would determine whether the salaries of the sample

of one hundred assistant managers correlated with their experience, and we
would then test whether the correlation was sufficiently strong to justify reject-
ing the null hypothesis that there is no correlation in the population of assistant
managers as a whole.
Let us suppose that there is a statistically significant correlation between the
employees’ months of management experience and their salaries. So now we
need to know whether the difference between the sample of male and female
employees’ salaries remains statistically significant when one controls for differ-
ences in their experience. This would involve doing a multiple regression of
salary on both gender and experience, obtaining a regression equation of the
form
y = a + b1x1 + b2x2,
or
Predicted salary = a + (b1 * gender) + (b2 * experience).
The problem is that gender is a categorical variable. We cannot plug “male”
and “female” in for x1. We need to use numbers.
The solution is to use the numbers 0 and 1. It is arbitrary which category is
assigned which number; let’s say we represent females by 0 and males by 1. Our
variable is then a dummy, or indicator variable, a variable that can take on only
the values 0 and 1, according to a condition. Here the condition is whether the
applicant is male. An indicator variable is represented by an I followed by the
condition in parentheses. Our new variable is
x1 = I(male) = 1 if the individual is male
= 0 if the individual is female
Using this indicator variable for x1, our regression equation is
Predicted salary = a + b1 * I(male) + b2 * experience.
Then, for a female with (say) 20 months of experience, we predict her salary
to be
Predicted salary = a + b1 * 0 + b2 * 20 = a + b2 * 20.
For a male with 20 months of experience, we predict his salary to be
Predicted salary = a + b1 * 1 + b2 * 20 = a + b1 + b2 * 20.
So a male and a female of equal experience are predicted to differ in salary
by b1 dollars. If b1 turns out to be positive, the male has higher predicted salary.
If it is negative, the female has higher predicted salary. Finally, if b1 = 0, there
is no difference in predicted salary between males and females when we have
controlled for experience.
 scores, dollars, and other quantitative variables 181

6.2.7.b Using Multiple Regression in Other Domains of Public Policy Multiple

regression analysis plays an important role in many areas of policy. Here are just
a few examples:
• In the famous Supreme Court case of McCleskey v. Kemp, the appellant,
who was sentenced to death, claimed that there was discrimination in
imposing the death penalty based on the race of the crime’s victim. David
Baldus, a law professor at the University of Iowa, ran a regression on a
large number of death penalty cases. The indicator of receiving the death
penalty14 was regressed on the indicator of the victim being Black. As covari-
ates, he included race of defendant, method of killing, motive, and defendant’s
criminal history. The victim’s race turned out to be significant in the direc-
tion suggesting that those who killed Whites were more likely to be sen-
tenced to death than were those who killed Blacks.15
• An analysis by John Donohue and Steven Levitt suggests that the nation-
wide legalization of abortion in 1973 led to a reduction in crime rates two
decades later—presumably the result of fewer unwanted and poorly cared-
for children. Among other things, the scholars looked at crime data in
states where abortion was legal before Roe v. Wade, correlations between a
state’s abortion and crime rates, and correlations between age and
crime.16
• In More Guns, Less Crime: Understanding Crime and Gun-Control Laws
(1998), John R. Lott used regression analysis to argue that citizens’ posses-
sion of guns reduced the amount of violent crime—presumably through
the deterrent effect on those contemplating assaults. Ian Ayres, John
Donohue, and Steven Levitt have argued that the analysis does not support
this conclusion and, indeed, that the data show the opposite.17 (The dispute
also illustrates the relative opaqueness of econometric analyses as com-
pared to, say, randomized controlled studies, and a layperson’s difficulty in
following technical arguments about the assumptions underlying the
former.
• Lawrence A. Greenfeld, whom President Bush named in 2001 to lead the
Bureau of Justice Statistics, lost his job in 2005 after refusing to suppress

14. Often, when the outcome is categorical, the more advanced method of logistic
regression is used instead of ordinary regression.
15. See Adam Liptak, Death Penalty Found More Likely When Victim Is White, New
York Times, Jan. 8, 2003.
16. John J. Donohue III and Steven D. Levitt, The Impact of Legalized Abortion on Crime,
116 Quarterly Journal of Economics 136–44 (2001).
17. See Ian Ayres and John J. Donohue III, Shooting Down the “More Guns Less Crime”
Hypothesis, 55 Stanford Law Review 1193 (2001); Steven D. Levitt and Stephen J. Dubner,
Freakonomics 130–134 (New York: HarperCollins Publishers, 2005).
182 problem solving, decision making, and professional judgment

an analysis that showed that, although White, Black, and Hispanic drivers
are stopped by police at about the same rate, Hispanic drivers were searched
or had their vehicles searched by the police 11.4 percent of the time and
Blacks 10.2 percent of the time, compared with 3.5 percent for White driv-
ers; and Blacks and Hispanics were also subjected to force or the threat of
force more often than Whites.18
• Many Supreme Court Justices have urged Congress to increase the salaries
of federal judges, arguing (in Justice Alito’s words) that “eroding judicial
salaries will lead, sooner or later, to less capable judges and ultimately to
inferior adjudication.” In A Skeptical Response to the Judicial Salary Debate,
Stephen Choi, G. Mitu Gulati, and Eric Posner used multiple regression
analyses to find the absence of any correlation between salaries and perfor-
mance in the highest courts of the fifty states. For proxies for performance,
the scholars measured effort in terms of the number of pages of opinions
published, skill by citations by out-of-state judges and law journals, and
independence by disagreement with fellow judges with the same political
party affiliation.19
While multiple regression analysis has great power, it is subject to many of
the same limitations as simple regression analysis: multiple regression can mea-
sure the strength only of relationships that are linear or that can be “linearized”;
it is sensitive to outliers, and (unless the explanatory variables are carefully
manipulated in an experiment, instead of just observed) it cannot prove the exis-
tence of causal relationships. As we will discuss in Section 7.2.1, there could be
a so-called confounding variable that you didn’t measure that influences the
ones you did measure. To eliminate confounding variables, one must conduct
randomized controlled experiments (discussed in Sections 7.3.1 and 7.3.5).

6.3 glossary
Central tendency Mean (average), Median (middle), Mode (most common)
measures
Coefficient of The amount of total variance explained by the predictor
determination (r2)
Confidence A range of numbers in a sample that are, say, 95 percent likely
Interval to encompass the population mean
Correlation A linear relationship between two variables
Correlation The strength of the linear relationship between two variables
coefficient (r)

18. Eric Lichtblau, Profiling Report Leads to a Demotion, New York Times, Aug. 24, 2005.
19. Http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1077295.
 scores, dollars, and other quantitative variables 183

Dispersion How spread out the data are. (See Variance and Standard
Deviation)
Empirical The distribution of the sample
distribution
Independence Where the value of the so-called predictor or dependent
variable provides no information about the outcome,
response, or independent variable
Line of best fit A straight line that best describes the relationship between the
predictor and response variables
Multiple regression Linear regression analysis that takes into account more than
analysis one predictor variable
Normal A specific bell-shaped curve with the mean, median, and
distribution mode all at the center point
Regression A technique for determining correlation between two or more
analysis continuous variables
Regression to the The tendency of extreme predictor variables to produce
mean responses that are less extreme (i.e., closer to the mean)
Sampling The distribution of the sample mean
distribution
Scatterplot A collection of datapoints showing the predictor and outcome
values for each item
Standard deviation A measure of the dispersion or variance of a sample. Square
(SD) root of variance
Standard error (SE) The standard deviation of the sampling distribution
Variance Squared deviation of data points from the mean
184 problem solving, decision making, and professional judgment

6.4 appendix
big-mart employee salaries by months of experience and sex

ID MOE SEX SAL ID MOE SEX SAL ID MOE SEX SAL

1 67 M 39435 35 40 F 39492 69 37 F 40253

2 60 F 39028 36 46 F 39589 70 55 M 38728
3 46 M 39055 37 60 F 40966 71 43 F 38662
4 82 M 39345 38 51 M 38010 72 21 F 36672
5 31 M 39889 39 38 F 36843 73 67 F 39497
6 47 F 37224 40 84 F 43140 74 50 M 37684
7 60 M 40468 41 69 M 40097 75 70 M 42339
8 49 M 39304 42 54 F 39636 76 27 M 36193
9 27 M 36988 43 75 F 41645 77 38 M 39632
10 47 F 38453 44 46 M 40843 78 65 F 41166
11 4 F 35855 45 42 F 36688 79 39 M 37685
12 45 F 36917 46 38 F 36254 80 14 F 35413
13 29 F 34965 47 33 F 37486 81 39 F 35719
14 43 F 35391 48 65 M 40923 82 48 F 37141
15 44 M 38901 49 48 F 37572 83 61 M 41842
16 61 M 38581 50 20 M 37740 84 44 F 38788
17 49 M 38544 51 81 M 40460 85 74 F 37753
18 55 F 37720 52 26 F 38531 86 29 M 38877
19 44 F 39473 53 58 F 37724 87 54 M 38677
20 63 F 39147 54 46 F 39188 88 59 F 40387
21 31 F 36349 55 53 M 39628 89 60 F 37749
22 64 F 40668 56 45 M 40427 90 62 M 38093
23 60 M 41882 57 52 M 36545 91 46 M 35393
24 59 F 41261 58 42 M 38764 92 54 M 38157
25 51 M 38867 59 36 F 37931 93 44 M 39142
26 66 F 40186 60 63 M 40507 94 13 F 37532
27 53 F 40777 61 54 M 39292 95 74 M 39670
28 38 F 38910 62 58 M 38789 96 25 M 38273
29 46 M 36527 63 70 M 42989 97 40 F 37270
30 74 M 41168 64 64 M 39730 98 64 M 40559
31 47 F 39965 65 45 M 39462 99 63 F 38539
32 60 M 41698 66 30 F 38294
33 73 F 40113 67 64 F 39655
34 46 F 39335 68 82 M 40697
7. interpreting statistical results and
evaluating policy interventions

The primary use of statistics in science, policy, and law is to throw light upon the
causal relationships that might exist between two or more variables. In this chap-
ter we consider some of the ways in which statistics interacts with scientific
reasoning, as well as some psychological findings that suggest caution in moving
from statistical to causal conclusions. Much of the chapter focuses on evaluating
the outcomes of policy interventions, and we also consider efforts to understand
underlying social phenomena of concern to policy makers.
Recall that the introduction to Part 2 introduced the “lens” model, under
which the real world can only be perceived through multiple fallible indicators.
This chapter is concerned largely with identifying, separating, and giving appro-
priate weight to those indicators in order to establish the existence (or nonexis-
tence) of causal relationships. One way in which this is done is through evaluation
activities aimed at assessing the conceptualization, design, implementation and
utility of social intervention programs.1
There are a number of different ways in which we use evaluation in policy
making. Oftentimes, policy makers will oversee the design of a particular social
program or select from among different types of interventions that use different
assessment designs. Thus, it is important for policy makers to understand poten-
tial design flaws and the advantages and disadvantages of different types of
program evaluation. More importantly, perhaps, is the ability for a policy maker
to understand how to analyze impact assessments, compare different evalua-
tions, think critically about the utility of such evaluations and make judgments
about the inferences that can be drawn from them.

7.1 measuring impact: correlation vs. causation

Policy makers often design, support, and evaluate health and social programs
designed to improve people’s lives. How can they learn whether the programs
actually achieve their intended effects?
The critical issue in impact evaluation is whether or not a program produces
more of an effect or outcome than would have occurred either without the inter-
vention or with an alternative intervention. Common among all approaches to

1. Peter H. Rossi and Howard E. Freeman, Evaluation: A Systematic Approach

19 (Beverly Hills: Sage Publications, 1985).
186 problem solving, decision making, and professional judgment

figure 7.1 correlation vs. causation http://xkcd.com/552/ .

Source: Reprinted by permission of xkcd.

evaluating impact are the goals of establishing (1) that a particular intervention
is correlated with the outcome, and (2) that the intervention causes the outcome—
for example, establishing that the rate of malaria infections in a village goes
down (1) when residents use bed nets and (2) because they use bed nets.
Imagine that we conducted a study of drowning deaths in Florida and discov-
ered that the number of drowning deaths in a given month was highly correlated
with ice cream sales for that month.2 We could develop a number of possible

A man rode the Fifth Avenue bus to work every day. And every day, when the bus
stopped at Forty-Second Street, he opened the window and threw the sports sec-
tion of the newspaper into a litter basket. After many years, the bus driver asked
why he did this, and the man replied: “I do it to keep the elephants off Fifth
Avenue.” “But there aren’t any elephants on Fifth Avenue,” said the bus driver, to
which the man responded: “See, it works!”

explanations for this correlation. (a) It could be that eating ice cream causes
drowning, with more people developing cramps in the water and thus not being
able to swim. (b) It could also be that drowning causes people to buy ice cream,
as friends and family of the victims seek to console themselves by eating. (c) Or
it could be that there is some other variable that causes both drowning and ice
cream consumption—for example, the average temperature in a given month.

2. This example comes from Ewart Thomas, to whom the authors are also grateful for
less direct influences.
 interpreting statistical results and evaluating policy interventions 187

Further testing could tease these explanations apart, but the observation of a cor-
relation is consistent with all three causal scenarios.
A causal diagram represents the causal process that results in a correlation of
interest. The three scenarios listed above as possible explanations of the correla-
tion between drowning deaths and ice cream sales are diagramed in Figure 7.2.
A solid arrow indicates a hypothesized causal relationship in the arrow’s direc-
tion. A dashed line, on the other hand, emphasizes that, while a correlation exists
between two variables, we are not hypothesizing any causation. When one vari-
able, such as temperature, is related to two other variables (drownings and ice
cream sales), a correlation appears between those two variables. But correlation
does not imply causation.

Drownings

Ice cream Ice cream High

Drownings Drownings
sales sales temperature
Ice cream
sales
(a) (b) (c)

figure 7.2 a causal diagram of ice cream sales and drownings.

Scenarios (a) and (b) of Figure 7.2 show alternative causal hypotheses.
Scenario (c) implies that temperature may be a confounding or lurking variable.
Confounding is the situation in which a variable you are not studying (here tem-
perature) is related to both the variables you are studying (here ice cream sales and
drownings). Data show a correlation between the variables you are interested in,
but a third lurking variable may actually be causing both.
Consider, for example, a health survey of 21,101 Finns that indicated that pet
owners were more likely to suffer from illnesses that included high blood pres-
sure, high cholesterol, ulcers, depression, kidney disease, and being overweight.
Did owning a pet cause these diseases? Did having these diseases cause people
to own pets? Alternatively:
Pets seem to be part of the lives of older people who have settled down and
experienced an increase in the number of illnesses, whereas young healthy
single people have no time, need, or possibility for a pet. Associations of pet
ownership with disease indicators were largely explained with socio-
demographic factors. Characteristics were those that bring forward the back-
ground of poor health in epidemiological investigations: male gender, low level
of education, life without a couple relationship, and poor social standing.3

3. Leena K. Koivusilta and Ansa Ojanlatva, To Have or Not To Have a Pet for Better
Health?, 1 PLoS ONE e109 (2006), doi:10.1371/journal.pone.0000109 (2006).
188 problem solving, decision making, and professional judgment

7.2 establishing causation

Establishing what the outcome actually was and its correlation with the interven-
tion is often hard enough. For example, in attempting to evaluate the outcome of
the bed net program, it may be difficult to gather the data necessary to establish
a baseline of malaria infections and then to track the number of infections over
time. Establishing causation is usually more difficult, because even if there is a
change, it may be due to factors other than the intervention being studied (intro-
duction of bed nets): perhaps the weather was drier than usual, thus producing
fewer mosquitoes, or perhaps other antimalaria interventions were introduced at
the same time.
A gross outcome is all the changes that occur from time point A, before pro-
gram implementation, and time point B, after implementation.
Gross Outcome = Effects of intervention (net outcome) + Effects of extrane-
ous and confounding factors + Design effects.4
Implementation of bed net program
A---------------------------------------------------------------------> B
(no bed nets) (bed nets fully distributed)
In the above intervention, the gross outcome would consist of all changes that
occurred during the implementation period, including the changes in malaria
infection rates attributable to the program, as well as changes resulting from
other factors, such as a change in the weather or other variables that could have
also reduced malarial infection rates.
What we ultimately care about is the net outcome, or the effects that are attrib-
utable to the intervention itself. As we see above,
Net Outcome = Gross Outcome – Effects of extraneous and confounding
factors – Design effects.
Therefore, it is important to identify and, if possible, measure the effects that
are not related to the intervention that may influence the gross outcome.

7.2.1 Confounding Effects

There are a number of different types of confounding variables that could influ-
ence the relationship between two given variables of interest. Such confounding
factors could either (a) be the actual cause of the correlation, or (b) influence the
causation such that the gross outcome must be reduced by the effect attributable
to the interfering factor.
The different (general) kinds of confounding variables that policy makers and
evaluation designers should be aware of include:

4. Rossi and Freeman, supra at 191.

 interpreting statistical results and evaluating policy interventions 189

• endogenous change: environmental effects such as temperature;

• secular trends: long-term trends in the community of interest such as a
decline in birthrates or increased employment;
• interfering events: a short-term event such as an earthquake or a war; and
• maturational trends: especially relevant when evaluating change in young
children.
Additionally, one must be aware of uncontrolled selection biases, meaning that
some people may be more likely than others to participate in the program under
evaluation, which will in turn influence the outcome. Common uncontrolled
selection biases include self-selection and differential attrition.
Self-selection is especially relevant in voluntary programs, where some people
may be more likely to volunteer for the given treatment than others. Take for
instance a charter school such as the well-known KIPP (Knowledge is Power
Program) program (an evaluation of KIPP is described in Section 7.4). KIPP
admits students randomly from a pool of those who wish to attend. But to be in
the pool, students or parents must be motivated to sign up. Such motivation may
make these children significantly different from their counterparts in public
schools whose parents did not try to enroll them in KIPP. Thus, when compar-
ing the educational outcomes between KIPP and non-KIPP students, one must
take into account these possible self-selection biases.
There are similar processes at work that lead to differential attrition in pro-
gram participation. It is very rare that all program participants complete treat-
ment, and individuals who drop out of the program may be significantly different
than those who continue participating. For instance, children who find the KIPP
program too demanding and unrewarding may be the most likely to drop out.
Thus, the consequence of attrition is often that those who may have needed the
program the least and were more likely to have changed on their own are the
ones who are most likely to remain in the program.5 One can easily imagine how
such attrition can skew evaluation results if not properly accounted for.
Different types of controls and evaluation designs are aimed at reducing and
accounting for the impact of these confounding variables. As we will see in our
discussion below, some are better than others at achieving this goal.

7.2.2 Design Effects

Design effects are effects that are attributable to the evaluation itself or flaws in the
evaluation design. These include errors that could spring from the measurement
tool itself, as well as errors resulting from the simple process of measuring.
Measurement reliability refers to how prone to error the given measuring
instrument is. In other words, its describes the extent to which an instrument
produces results that vary from one administration to another when applied to

5. Id. 194.
190 problem solving, decision making, and professional judgment

the same object. Some measurements, such as the measurement of height and
weight through standard instruments, have a high degree of reliability. Others,
such as surveys that rely on subjects’ responses to written or oral questions, are
more prone to unreliability because of the inherent ambiguity in language and
interviewer differences in their administration of the survey. In most cases, it is
not possible to eliminate unreliability completely, although it is possible to make
adjustments in results that take unreliability into account.
Even if a measuring procedure produces consistent results and is therefore
deemed reliable, it will not be valid if it does not measure what it is intended to
measure.
Measurement validity depends on whether there is a consensus that the mea-
sure is valid among the appropriate stakeholders, such as members of the scien-
tific community. For example, the effectiveness of the bed net program could be
measured by a number of alternative measures, including (1) self-reported
malarial infection rates, (2) hospital visits related to malaria, or (3) professional
surveys of malarial infection rates. A good outcome measure is generally one
that may be feasibly employed, given the constraints of time and budget, and
that is directly related to the goals of the program.
The process of conducting an evaluation in and of itself can also influence the
outcome of the evaluation. This can happen through what is known as the
Hawthorne or placebo effect. The Hawthorne Effect stems from a famous experi-
ment, in which efforts to discover whether changes in factory lighting intensity
would improve worker productivity disclosed that any change had that effect.6
Researchers concluded that the effect was the result of the experiment itself—
workers interpreted their participation in the experiment as a sign that the firm
was interested in their personal welfare, and thus were motivated to work harder.
In medical experiments, the Hawthorne Effect is known as the placebo effect, and
refers to the fact that subjects may be as much affected by the knowledge that they
are receiving treatment as by the treatment itself. For example, in some experi-
ments testing the effectiveness of pain relievers, the control groups receiving
sugar pills (unbeknownst to them) reported a decrease in pain levels.
The Pygmalion Effect refers to the fact that expectations can also influence
outcomes. An experiment by Robert Rosenthal and Lenore Jacobson found that
if teachers were led to expect enhanced performance from some children rather
than others, then those children did indeed perform better.7 The experiment
confirmed the researchers’ hypothesis that biased expectancies could essentially
affect reality and create self-fulfilling prophecies as a result. Policy makers
and evaluators should be especially aware of such effects where treatment is

6. Fritz J. Roethlisberger and William J. Dickson, Management and the

Worker (Cambridge, MA: Harvard University Press, 1939).
7. Robert Rosenthal and Lenore Jacobson, Pygmalion in the Classroom
(expanded edition New York: Irvington, 1992).
 interpreting statistical results and evaluating policy interventions 191

delivered in a non-neutral manner, such as by an individual whose personal

biases could influence the way in which treatment is administered.

7.3 reducing unwanted effects through controls

Confidence in the results of evaluations and other empirical studies not only
requires identifying various confounding factors and potential design flaws, but
using controls. Here we discuss four important kinds of controls:
• Randomized controls—the “gold standard” of evaluation, in which partici-
pants are randomly assigned either to the treatment (or experimental)
group—those who receive the intervention—or to a control group.
• Constructed controls—a major form of quasi-experimental evaluation, in
which the treatment and control groups are constructed to be similar to
each other.
• Reflexive controls—in which one looks at the same population before and
after the intervention.
• Statistical controls—in which multiple regression analysis is used to con-
trol for suspected confounding factors.
The approaches other than randomized controls are sometimes called obser-
vational studies, because the experimenters merely observe the populations they
are studying.

7.3.1 Randomized Controlled Studies

Randomized controlled studies are employed by randomly assigning program par-
ticipants to either a treatment group or a control group. Here it is important to
ensure that every person in the target population has the same chance as any
other person of being selected into either group. The goal is not that the control
and treatment groups be perfectly comparable. Rather, the goal is complete ran-
domization such that because the experimental and control groups differ from
one another only by chance, whatever processes or confounding factors that
may be competing with the treatment to produce outcomes are present in both
groups to the same extent, except for chance variations. Thus, because we are
comparing the treatment group to a control group that has been exposed to the
same confounding factors, we do not need to worry about how such effects
have potentially skewed our results.
After collecting data on the two groups’ outcomes, you compare the measures
to see if there are sufficiently large—statistically significant—differences between
the two. As we saw in Chapters 5 and 6, statistical significance is a function of
the magnitude of the differences and the size of the groups. Too small a sample
may make it difficult to draw any conclusions one way or the other. But if you
find statistically significant differences, you can conclude that the intervention
192 problem solving, decision making, and professional judgment

had an impact because the random assignment controlled for other factors. The
main drawbacks to randomized controlled studies are that they are expensive to
perform and difficult to replicate.
The U.S. Food and Drug Administration requires that most new pharmaceu-
tical products be subject to randomized controlled studies before being intro-
duced into the market. Participants in clinical trials receive either the new drug
(the treatment) or a control (a placebo, or conventional treatment if there is one),
and researchers analyze whether the health of those receiving the new drug
improves more—or less—than the health of those in the control group.
Randomized controlled studies are not mandated for social interventions,
and they are seldom done due to time and expense constraints. But when feasi-
ble, they can provide essential insights into the actual effects of social programs.8
In 2005, the U.S. Department of Health and Human Services published the
National Head Start Impact Study, a randomized controlled study analyzing the
impact of the Head Start program on a variety of measures of children’s social
and cognitive development.9 The study employed random controls and found
that Head Start had limited effects on the social and cognitive development of
participating children. This conclusion contradicted many previous Head
Start studies, which had relied on constructed or statistical controls.

7.3.2 Constructed Controls

When using constructed controls a group of untreated or differently treated targets
are selected by nonrandom methods to be as comparable as possible in crucial
respects to the targets constituting the intervention group. This process is often
known as matching. The success of this control depends on how closely the con-
structed group resembles the intervention group in all essential characteristics.
For instance, if we were using a constructed control for an evaluation of a prekin-
dergarten program we would try to select a nontreatment group that matched
our prekindergarteners in terms of age, race, socio-economic status, language
spoken at home, parental education level, and any other factors that we thought
might affect early childhood learning.
However, it is important to note that some characteristics—such as motiva-
tion and parental attitudes about education—would be difficult to “match,” and
that not controlling for such factors could influence the outcome. In a 1969

8. The Coalition for Evidence-Based Policy maintains an excellent Web site on social
programs that work: http://www.evidencebasedprograms.org.
9. U.S. Department of Health and Human Services, Administration for Children and
Families, “Head Start Impact Study: First Year Findings: (Washington, DC: U.S.
Department of Health and Human Services, 2005), available at http://www.acf.hhs.gov/
programs/opre/hs/impact_study/.
 interpreting statistical results and evaluating policy interventions 193

evaluation of the Head Start program,10 first-graders who had participated in

Head Start were compared with first-graders of comparable backgrounds in
the same or nearby schools who had not participated. The study was widely
criticized, however, because it did not control for differences in environmental
circumstances.
The use of matching techniques has increased greatly in the last decade, with
some researchers going as far as to argue that matching “mimic[s]
randomization.”11 However, economists are more skeptical about the use of
techniques, finding that matching performs well only under specific data
conditions.12 The critiques of the use of matching in observational and quasi-
experimental studies are discussed below.

7.3.3 Reflexive Controls

With reflexive controls, or a “before and after” approach, targets are compared to
themselves, before the intervention took place. Such a technique may be required
in large, full-coverage programs in which it is impossible to define randomized
or constructed controls or locate nonparticipants. This may also be an appropri-
ate technique where it is reasonable to believe that the targets remain identical
in relevant ways before and after the intervention. Reflexive control designs,
however, are highly vulnerable to incorrect estimation of net effects because they
do not control for confounding factors such as maturational effects, secular
trends, and interfering events.
The so-called “broken windows” theory of crime reduction provides a famous
example of the problem of before-after studies: New York’s zero-tolerance policy
for vandalism and other petty crimes in the 1990s was claimed to have resulted
in a significant reduction of all crimes, including felonies. But it turned out that
crime rates declined simultaneously, and to approximately the same extent, in
other U.S. cities that had not adopted this approach. The decline might have
resulted from the waning of the crack epidemic, or the increase in the number
of drug offenders in prisons (and therefore not on the streets), or the general

10. Victor G. Cicerelli et al., The Impact of Head Start : An Evaluation of the
Effects of Head Start on Children’s Cognitive and Affective Development.
(Athens: Westinghouse Learning Corporation and Ohio University, 1969).
11. Debbie L. Hahs-Vaughn and Anthony J. Onwuegbuzie, Estimating and Using
Propensity Score Analysis with Complex Samples, 75 Journal of Experimental Education
31 (2006).
12. Multiple studies conclude that matching only performs well in data that satisfy
three criteria: 1) treatment and comparison groups come from identical sources, 2) indi-
viduals in both groups live in the same geographical area, and 3) “the data contain a rich
set of variables that affect both” the outcome and the probability of being treated. Jeffrey
Smith, and Petra Todd, Does Matching Overcome LaLonde’s Critique of Nonexperimental
Estimators?, 125 Journal of Econometrics 309 (2005).
194 problem solving, decision making, and professional judgment

demographic decline in the number of young males.13 Here, simple observation

of the correlation between implementation and declining crime rates was not
enough to establish causation.

7.3.4 Statistical Controls

Rather than matching participants with nonparticipants, a researcher can also
use statistical controls to adjust for statistical differences between the two groups.
This can be done in cross-sectional surveys, using the techniques of multiple
regression analysis to control for potentially confounding differences among
individuals. For example, in the study (mentioned in Section 6.2.7.a) that con-
cluded that killers of white victims were more likely to receive the death penalty
than killers of Black victims, statistical techniques were used to control for the
method of killing, motive, and defendant’s criminal history.

7.3.5 The Advantages and Drawbacks of Different Methods

As noted above, observational designs are vulnerable to confounding. For exam-
ple, a large cross-sectional study in the 1990s found that hormone replacement
therapy (HRT) for postmenopausal women reduced the risk of heart disease.14
But a later randomized controlled trial found that HRT actually increased the
risk of heart attacks.15
The later study randomly assigned 16,608 postmenopausal women to groups
of roughly equal size and gave one group the hormone combination and the
other a placebo. It turned out that the original studies had not controlled for the
fact that women on HRT “were also more likely to see a doctor (which is how
they were put on hormone therapy in the first place), and probably more likely to
exercise and to eat a healthful diet, than women who were not taking the drug.”16
Thus, the relationship wasn’t that taking hormones made women healthy, it was
that healthy women took hormones.
Even when the investigator controls for every known potential confounder,
there always could be more confounders that the investigator did not think of or
that were difficult or expensive to measure. It has been suggested that, in the
studies that found a decreased heart attack risk in subjects on HRT, the culprit
confounder was a variable representing lifestyle—including diet and exercise,

13. The literature on this theory is nicely summarized in Wikipedia, s.v. “Fixing Broken
Windows,” http://en.wikipedia.org/wiki/Fixing_Broken_Windows (last visited June 30,
2008).
14. Jane Brody, New Therapy for Menopause Reduces Risks, New York Times, Nov. 18,
1994.
15. Gina Kolata with Melody Petersen, Hormone Replacement Study: A Shock to the
Medical System, New York Times, July 10, 2002.
16. Susan M. Love, Preventive Medicine, Properly Practiced, New York Times, July 16,
2002.
 interpreting statistical results and evaluating policy interventions 195

frequency of doctor visits, and other less tangible aspects of attention to one’s
health. Variables like these, even if thought of, may be too vaguely defined to
control for completely. The suspected confounding in these studies is repre-
sented in Figure 7.3.

Take HRT

Healthy lifestyle,
regular doctor
visits
Low risk of
heart
disease

figure 7.3 a causal diagram of health lifestyle, hrt, and heart

disease.

When subjects are randomly assigned to treatment (exposure) or placebo (no

exposure) groups, effects of confounders disappear. Because exposure to HRT is
randomly assigned, it cannot be correlated with any other characteristics of the
subjects. The two groups are almost exactly the same in their characteristics;
there is very little chance that a systematic difference (such as diet or exercise
behavior) exists between them. (Some known confounders, such as age, are often
purposely balanced between the groups, to avoid the possibility of a difference
occurring by chance in the randomization.)
Even where the available data set is extensive, with a “rich” set of covariates and
samples lending itself to very precise matching between the experimental and
control groups, selection bias can still sneak in. Kevin Arceneaux, Alan Gerber,
and Donald Green demonstrate this by comparing estimates generated by
matching to an experimental benchmark in a voter mobilization experiment.17
Based on a large-scale randomized field experiment, the authors developed an
experimental benchmark measurement of the effect of get-out-the-vote phone
calls on people’s voting behavior. Using random assignment, the average
treatment effect on receiving a get-out-the-vote phone call on voter turnout
was estimated to be 1.98 percentage points. The effect on the placebo group
(which received calls encouraging them to buckle up their seat belts) was

17. Kevin Arceneaux, Alan S. Gerber, and Donald P. Green, 2004-04-15 “Comparing
Experimental and Matching Methods using Two Large-Scale Field Experiments on
VoterMobilization” Paper presented at the annual meeting of the The Midwest Political Science
Association, Palmer House Hilton, Chicago, Illinois Online 2009-05-26 from http://www.
allacademic.com/meta/p83115_index.html.
196 problem solving, decision making, and professional judgment

0.95 percentage points. The researchers then compared treated individuals to

untreated individuals with matching techniques. The treatment and placebo
groups were exactly matched to control groups such that the comparison groups
shared exactly the same descriptive statistics on important characteristics such
as past voting behavior and age. Using matching, the average effect of a get-
out-the vote phone call was estimated to be 7.53 percentage points—over three
times the estimate derived from a randomized experiment. Even more surpris-
ingly, the matching experiment also suggested that the buckle-up calls raised
voter turnout by 5.36 percentage points.
What accounts for the stark difference between the randomized experimental
and the matching results? The authors conclude that the variation occurred
because the treated individuals who were matched with untreated individuals
were people who actually answered calls from a commercial phone bank, and
were more likely to have elevated voting tendencies, even after controlling for
past voting habits and other demographic characteristics. While it was impossi-
ble to control for this selection bias in the matching experiment, the randomized
experiment did just that.
As evidenced by the above comparisons, randomized controlled experiments
are vastly preferable to observational studies based only on statistical controls. In
fact, many scientists and statisticians believe that one can never infer causality
from an observational study. But observational studies are often our only choice
in situations where the phenomena in question have already occurred or where
it is not feasible to conduct a controlled experiment. The analysis of legal claims,
as well as many natural sciences, including astronomy, geology, and paleontol-
ogy, depend primarily on observational studies. Even issues of public health are
not always susceptible to randomized controlled studies. An article in the British
Medical Journal entitled Parachute Use to Prevent Death and Major Trauma Related
to Gravitational Challenge: Systematic Review of Randomised Controlled Trials
concluded:18
As with many interventions intended to prevent ill health, the effectiveness of
parachutes has not been subjected to rigorous evaluation by using randomised
controlled trials. Advocates of evidence based medicine have criticised the
adoption of interventions evaluated by using only observational data. We
think that everyone might benefit if the most radical protagonists of evidence
based medicine organised and participated in a double blind, randomised,
placebo controlled, crossover trial of the parachute.
In any event, developing and testing hypotheses requires not merely quantita-
tive skills, but good empirical intuitions in order to posit possible correlates
and to identify and weed out confounding variables. Legal issues involving

18. Gordon C. S. Smith and Jill P. Pell, Parachute Use to Prevent Death and Major
Trauma Related to Gravitational Challenge: Systematic Review of Randomised Controlled
Trials, 327 British Medical Journal 1459–61 (2003).
 interpreting statistical results and evaluating policy interventions 197

figure 7.4 parachutes reduce the risk of injury after gravitational

challenges, but their effectiveness has not been proved with
randomized controlled studies.
Source: Reproduced from Gordon C. S. Smith and Jill P. Pell, Parachute Use to Prevent Death and
Major Trauma Related to Gravitational Challenge: Systematic Review of Randomised Controlled
Trials, 327 British Medical Journal 1459. © 2003. Reprinted with permission from BMJ
Publishing Group Ltd.

employment discrimination and toxic torts of the sort illustrated in the Big-Mart
and Terra Nueva examples are typical. Thus, the lawyer’s or policy maker’s
task—or the statistician’s task under their guidance—is to determine correlation
and often to attribute causation based on the relationship of plausible indepen-
dent variables (the insulation in a tenant’s apartment) with a dependent variable
of interest (the presence of illness).

7.4 some further examples of program evaluation

In formulating interventions in uncharted areas, policy makers may have little

choice but to rely on the intuitions of experts and practitioners in the field. But
intuitions about social change are notoriously inaccurate.
Consider the work of Dr. Joan McCord, a criminologist who evaluated
programs for troubled youth.19 Her best-known work was a longitudinal study
comparing participants in programs that provided mentoring, health care

19. Douglas Martin, Joan McCord, Who Evaluated Anticrime Efforts, Dies at 73, New York
Times, March 1, 2004.
198 problem solving, decision making, and professional judgment

services, and summer camp to high-risk boys with a control group of similar
youths. She found that boys in the program were more likely to become crimi-
nals, have employment and marital problems, and become alcoholics than those
in the control group. The evidence contradicted not only the expected outcome
but also the participants’ own belief that they had benefited from the program.
(Dr. McCord hypothesized that the boys in the treatment group may have felt
that they were given the attention because something was wrong with them,
making it a self-fulfilling prophecy—the Hawthorne Effect at work.) She found
similar paradoxical effects from a “just say no” drug education program and
from Scared Straight, a program designed to deter young people from following
a criminal path.
Thus, even (perhaps especially) when intuitions strongly support a particular
intervention, it is important to evaluate its actual impact. In this section, we look
at some cases where evaluation has demonstrated pretty conclusively that an
intervention works or that it doesn’t work and where it has been inconclusive.
The Good. The Abdul Latif Jameel Poverty Action Lab (J-PAL), a research
center at the Massachusetts Institute of Technology, specializes in conducting
randomized controlled studies in developing countries. J-PAL randomly assigned
girls in Kenya either to a group that was promised merit-based scholarships and
a cash grant for school supplies if they scored well on academic exams, or to a
control group. It turned out that just being eligible for scholarships led the girls
to think of themselves as “good students” and led to higher academic grades. In
fact, both student and teacher attendance improved in eligible schools, and even
boys showed improved test scores.20
In 2004–2005, SRI International employed a cross-sectional approach in eval-
uating the outcomes of the Bay Area KIPP (Knowledge is Power Program) char-
ter schools, whose “pillars” include high expectations, parental choice, long
school days, autonomous school leaders, and a relentless focus on results.
(Although a randomized controlled study is now under way, KIPP and its
sponsors also wanted this early, formative assessment of its impact to detect any

20. Michael Kremer, Edward Miguel, and Rebecca Thornton, “Incentives to Learn”
(NBER Working Paper No. 10971, National Bureau of Economic Research, 2004, http://
www.nber.org/papers/w10971). Another experiment by J-PAL demonstrates the failure of
an intervention to achieve its intended goals, though it certainly did some good along the
way. In a randomized controlled study to determine whether reducing the incidence of
parasitic worms among children could lead to better attendance and, consequently, higher
test scores, J-PAL administered low-cost deworming drugs to children at randomly chosen
schools in a poor, densely settled farming region near Lake Victoria. The drugs worked
well: after one year only 27 percent of children in treatment schools had moderate to heavy
worm infections, compared with 52 percent of children in control schools. The effects on
schooling, though, were minimal at best: though attendance rates in treatment schools
were 7 to 9 percent higher than in control schools, the test scores of children in interven-
tion and control schools were not significantly different.
 interpreting statistical results and evaluating policy interventions 199

obvious need for midcourse corrections.) Using two standard state tests, the SAT
and CST, experimenters tracked KIPP students’ achievement over two years and
compared their achievement with that of students in comparable California
schools. The report explains the limits of the methodology:
In an ideal world we would make two comparisons as a basis for determining
if KIPP students are performing better than if they were not attending KIPP
schools. The first would be to compare achievement growth of KIPP students
to their growth trajectory prior to KIPP. However, this requires individual
student data for several years prior to enrollment in KIPP that is not publicly
available. . . . The second would be to compare KIPP students’ achievement
with that of a comparison group of students, defined based on KIPP school
waitlists (which do not yet exist), or matching students in the district based on
demographic characteristics. Again, this comparison requires access to
student-level data for both KIPP and non-KIPP students.21
With these limitations, the report noted:
The spring 2005 CST data indicate that the overall percentage of students
performing at a proficient level or above is consistently higher for KIPP
schools than for comparable schools in the district—in some cases dramati-
cally so. Similarly, when comparing students in KIPP schools to all students
in the state, more fifth graders in two of the five KIPP schools scored at or
above proficient in ELA (English language arts) than students statewide; in
three of five KIPP schools, the percentage of fifth-grade students who scored
at or above proficient in math was higher than the state average. Likewise, in
three of the four KIPP schools with sixth-grade scores, a higher percentage of
sixth-grade students reached proficiency in math and ELA compared to the
state as whole. In the one KIPP school with seventh-grade scores, the percent
proficient in both ELA and math exceeded the state average.
Although not conclusive, the size of the differences was great enough to
strongly support the hypothesis that the KIPP approach made a real difference
in children’s academic outcomes—sufficient to justify continuing and even
expanding the program while conducting more precise studies.
The Bad. The next best thing to learning that a social intervention succeeds, is
determining conclusively that it does not succeed—so that funders will seek
better options rather than pouring money down the drain. A famous example of
a demonstrably ineffective intervention is the Drug Abuse Resistance Education
(DARE) program, which sought to prevent youth substance abuse through
classroom instruction. Randomized controlled studies of the DARE program

21. SRI International, Bay Area KIPP Schools, “A Study of Early Implementation:
First Year Report 2004–05” (Menlo Park, Calif.: SRI International, 2006), available at
http:// policyweb.sri.com/cep/publications/KIPPYear_1_Report.pdf.
200 problem solving, decision making, and professional judgment

consistently showed that students in treatment and control groups had the same
rates of both short- and long-term drug use.22
In another case, the evaluation firm Mathematica Policy Research Inc. was
commissioned to conduct a randomized controlled evaluation of federal absti-
nence education initiatives designed to prevent teen pregnancy. Mathematica
worked with four different states to randomly assign schools either to receive or
not to receive abstinence education, and then analyzed students’ self-reported
sexual activity rates. The results, released in 2007, show that students whose
schools received abstinence programs were just as likely to be sexually active as
students whose schools did not.23 James Wagoner, the president of Advocates for
Youth, said of the results: “After 10 years and $1.5 billion in public funds these
failed abstinence-only-until-marriage programs will go down as an ideological
boondoggle of historic proportions.”24
The Inconclusive. In 1997, the New York City Voucher Experiment randomly
assigned two thousand low-income families with K–4 students to either a treat-
ment group or a control group. Families in the treatment group received school
vouchers for private-school tuition, worth $1400 per child per year, for four
years,25 while families in the control group did not receive vouchers. After three
years, researchers administered math and reading tests to students in each group
and analyzed the results. It turned out that vouchers did not significantly affect
children’s test scores in the aggregate, though they had a small positive impact
on the test scores of African Americans.26
The voucher experiment did not show that vouchers are ineffective. Rather,
methodological problems, especially the small sample size, made it hard to draw
a conclusion one way or the other.27 Given the difficulty of designing and imple-
menting studies of social interventions, you should be prepared for inconclusive

22. Cheryl L. Perry et al., A Randomized Controlled Trial of the Middle and Junior High
School D.A.R.E. and D.A.R.E. Plus Programs, 157 Archives of Pediatrics and Adolescent
Medicine 178–184 (2003).
23. Christopher Trenholm et al., “Impacts of Four Title V, Section 510 Abstinence
Education Programs: Final Report,” report prepared by Mathematica Policy Research Inc.
for the U.S. Department of Health and Human Services, April 2007.
24. Advocates for Youth, “10-Year Government Evaluation of Abstinence-Only
Programs Comes Up Empty,” news release, April 13, 2007, http://www.advocatesforyouth.
org/news/press/041307.htm.
25. Social Programs That Work, “New York City Voucher Experiment,” http:// www.
evidencebasedprograms.org/Default.aspx?tabid=143 (last visited June 30, 2008).
26. See Alan Krueger and Pei Zhu, Another Look at the New York City SchoolVoucher
Experiment, 47 American Behavioral Scientist 658–99 (2003); and Institute of
Education Sciences, National Center for Education Evaluation and Regional Assistance,
“Evaluation of the DC Opportunity Scholarship Program: Impacts After One Year,”
http://ies.ed.gov/ncee/pubs/20074009 (last visited June 30, 2008).
27. William G. Howell, book review, Data Vacuum, 2 Education Next (2002), available
at http://www.hoover.org/publications/ednext/3366891.html.
 interpreting statistical results and evaluating policy interventions 201

results much of the time. This can happen, for example, because samples that
were originally of reasonable size diminish as a result of attrition—for example,
families in the control group move away and the experimenters lose track of them.
Or the control group coincidentally could receive some interventions from another
source—whether similar to or different from the “treatment” group.28
Unfortunately, many negative and inconclusive results never see daylight—-
the former, for reasons of motivation; the latter, for lack of interest. There is also
a strong bias in many scholarly fields against publishing statistically insignifi-
cant results. But even knowing about inconclusive results is useful, since a sub-
sequent study may be able to learn from the first and remedy some of the defects
that prevented a clearer outcome.

7.4.1 The Connecticut Crackdown on Speeding: A Legal Policy Case Study

A classic article on confounding variables and other methodological problems
in assessing the effects of legal policy changes is Donald Campbell’s and
H. Lawrence Ross’s study of the effect of the State of Connecticut’s decision to
impose draconian penalties—suspension of licenses—for speeding offenses in
1956.29 Governor Abraham Ribicoff announced the policy during the last week of
1955, a year that had a record number of 324 deaths caused by car accidents. As
shown in Figure 7.5, a year later, the number of deaths had declined to 284, and
the Governor stated: “With the saving of forty lives . . ., a reduction of 12.3 per-
cent from the 1955 motor vehicle death toll, we can say the program is definitely
worthwhile.”
“Not so fast,” in effect responded Campbell and Ross, in an article that details
the problems of observational, or quasi-experimental studies. Consider these
alternative explanations:
• Rival hypotheses. Was the speeding crackdown the only plausible change
between 1955 and 1956 that might have contributed to the reduction of
traffic fatalities? Consider that 1956 might have had an unusually mild
winter or been an unusually dry year. Perhaps that’s the year that Connecticut
instituted medevac services? Or perhaps the decline from one year to the
next is part of a long-term trend caused by improved automobile or highway
safety?
• Measurement changes. Though not likely in this example, someone
conducting series studies any time should be alert to possible changes in
measurement or reporting. Here are some examples from other domains:

28. See, for example, Paul T. Decker, Daniel P. Mayer, and Steven Glazerman, “The
Effects of Teach for America on Students: Findings from a National Evaluation” (Princeton,
N.J.: Mathematica Policy Research Inc., 2004), available at http://www.mathematica-mpr.
com/publications/pdfs/teach.pdf.
29. Donald Campbell and H. Lawrence Ross, The Connecticut Crackdown on Speeding:
Time-Series Data in Quasi-Experimental Analysis, 33 Law & Society Review 33–54 (1968).
202 problem solving, decision making, and professional judgment

320

310

300

290

280

BEFORE AFTER
CRACKDOWN CRACKDOWN
(1955) (1956)

figure 7.5 decline in fatal car accidents from 1956 to 1957.

Source: Reprinted by permission of Wiley-Blackwell.

Campbell and Ross refer to a dramatic increase in suicide statistics in

nineteenth-century Prussia when record keeping was transferred from the
local police to national civil service. For some more recent examples: What
some dermatologists have thought to be a drastic increase in melanoma
may just be an increase in people’s being screened for skin cancer.30 And
there is considerable uncertainty about the extent to which an apparent
epidemic of autism reflects an actual increase in the syndrome rather than
a more capacious definition of this diagnostic category.31 While the air traf-
fic controllers’ union believes that an increase in near-misses is the result

30. See Gina Kolata, Melanoma Is Epidemic. Or Is It, New York Times, Aug. 9, 2005.
“They found that since 1986, skin biopsies have risen by 250 percent, a figure nearly the
same as the rise in the incidence of early stage melanoma. But there was no change in
the melanoma death rate. And the incidence of advanced disease also did not change, the
researchers found.”
31. Tina Kelley, An Autism Anomaly, Partly Explained, New York Times, Feb. 18, 2007;
Graham Lawton, The Autism Myth, New Scientist, Aug. 13, 2005.
 interpreting statistical results and evaluating policy interventions 203

325

300

275

250

225 TREATMENT

200

'51 '52 '53 '54 '55 '56 '57 '58 '59

figure 7.6 fatal car accidents, 1951–1959.

Source: Reprinted by permission of Wiley-Blackwell.

of understaffing, FAA officials suggest that it is error detection rather than

the actual error rate that has increased.32
• Variability, sample size, and regression. Consider Figure 7.6, which shows the
history of Connecticut traffic fatalities for several years before and after the
speeding crackdown. Note that there is significant variation from year to
year. In effect, one year may constitute too small a sample on which to base
the conclusion that the crackdown had any effect. Consider also that the
governor instituted the crackdown because of the extraordinarily high fatal-
ity rate during 1955. What does the phenomenon of regression to the mean
predict about the rate in 1956?
When undertaking before/after observational studies, it is useful to simulate
a control group by examining similar situations where there was no interven-
tion. Similar trends in those situations provide a clue that an independent vari-
able other than the intervention may have been at work.
As shown in Figure 7.7, Campbell and Ross compared Connecticut to four
nearby “control states”—New York, New Jersey, Rhode Island, and Massachusetts.

32. Matthew Wald, Errors in the Air and a Battle on the Ground, New York Times,
Aug. 13, 2005.
204 problem solving, decision making, and professional judgment

17 Connecticut

16 Control States

15
14
FATALITY RATE

13

12

11
10

'51 '52 '53 '54 '55 '56 '57 '58 '59

figure 7.7 fatal car accidents in connecticut and nearby states,

1951–1959.
Source: Reprinted by permission of Wiley-Blackwell.

(Why do you suppose the combined data for those four states is much smoother
than for Connecticut?).
Although the comparison lends some support to the Governor’s causal
hypothesis, Campbell and Ross mention the possibility that there may have been
some diffusion—a tendency for the intervention in the experimental group to
affect the control group. Also, there is some doubt whether the differences are
statistically significant.33
This underscores one of the central points of this chapter: purported causal
relationships must be viewed with caution. One must be wary about making
decisions based on evidence derived from observational or quasi-experimental
studies. While randomized controlled experiments are subject to less potential
errors and do a better job of proving causality, such experiments are rarely
available. Thus, the policy maker must use common sense and basic statistical
knowledge to identify potential confounding factors and design effects that could
influence the observational study’s outcome and make decisions about causality
accordingly.

33. The analysis of significance in this context is complex. Using a one-tailed test, G.V.
Glass, found significance at somewhere between p < 0.05 and p < 0.07. Gene V. Glass,
Analysis of Data on the Connecticut Speeding Crackdown as a Time-Series Quasi-Experiment,
55 Law & Society Review 55–76 (1968).
 interpreting statistical results and evaluating policy interventions 205

7.5 the significance of statistical significance 34

7.5.1 Effect Size

Most end users of statistical analyses are ultimately interested in the strength or
magnitude of the relationship between variables—in how large an effect a change
in one variable has on another: Is there just a small or a large difference in the
number of reported illnesses of Terra Nueva residents who live in apartments
with and without foam insulation? Do the salaries of male and female assistant
managers at Big-Mart differ by a few dollars or a few hundred? Does spending
additional $100 per pupil have a big or a small effect on achievement?
The correlation coefficient (r), discussed in Section 6.2.2, is an indication of
effect size, and effect size can be inferred from other statistical tests as well. But
recall from our discussions of hypothesis testing that effect size is only one
component of statistical significance, which also depends on the size (and vari-
ance) of the samples. To find a statistically significant difference in the symp-
toms of residents with and without foam insulation means that we are probably
right to reject the null hypothesis and conclude that there is some correlation,
but it does not say how great the correlation is. Meeting a more demanding level
of significance (say, 0.001 rather than 0.5) tells you that it’s more likely that
differences revealed by the survey of a sample of the population actually reflects
the population, but it does not tell you that the effect size is greater. Knowing
with near certainty that the tenants living in apartments with foam insulation
have seventeen headaches a year when those in apartments without the insula-
tion have sixteen is unlikely to wow a jury or energize the mayor to take action.

7.5.2 Meta-Analysis
The Terra Nueva, Big-Mart, and education examples in the preceding chapters
showed statistically significant effects of sufficient magnitude to be of legitimate
interest to various stakeholders. But it is possible for a real effect to be masked
because the sample is too small. Apart from applying a less demanding signifi-
cance level—which trades off Type II for Type I errors—is there anything to be
done? In situations where a number of experiments have been done to study
essentially the same phenomenon, medical and social scientists sometimes use
the statistical technique of meta-analysis to aggregate and analyze information
from a variety of existing studies. Meta-analysis can increase the effective sample
size and thus reveal an effect that was not statistically significant in a single
study.

34. See generally Jeremy A. Blumenthal, Meta-Analysis: A Primer for Legal Scholars, 80
Temple L. Rev. 201, 209–10 (2007); John E. Hunter and Frank L. Schmidt, Cumulative
Research Knowledge and Social Policy Formulation: The Critical Role of Meta-Analysis, 2
Psychol. Pub. Pol’y & L. 324, 342–43 (1996).
206 problem solving, decision making, and professional judgment

For example, the American Public Health Association undertook a meta-

analysis of studies evaluating teen smoking prevention programs,35 which exam-
ined ninety-four studies of different programs. The studies were initially
evaluated for methodological rigor and placed into categories based on strength
of methodology. Researchers then focused analysis on the studies with stronger
methodologies in order to get more accurate results. The meta-analysis concluded
that programs that incorporate attention to social norms and social reinforce-
ment are more successful than those that try to convince teens to abstain from
smoking by giving them reasons.
Meta-analysis has strong proponents. But it also has critics, who note, among
other things, the dangers of relying on studies that are poorly designed or imple-
mented, the difficulty of aggregating studies of interventions that differ from
each other, and the susceptibility of meta-analysis to publication bias—the fact
that studies showing statistically significant results are more likely to be pub-
lished than those that do not.36 Without professing expertise, we believe that
these vulnerabilities can be controlled for sufficiently to make meta-analysis a
valuable tool.
7.5.2.a Practical Importance: The Representation of Effect Size How do you
interpret the statement that people with high cholesterol can reduce their risk of
death by 22 percent by taking statins, or that an arthritis pain-reliever increases
one’s chance of a heart attack by 30 percent?
A particular drug could halve or double your risk of a dreaded outcome, but
the difference might be between a risk of 0.001 and 0.002 or between 0.1 and 0.2.
A more useful way to report and think about risk is incremental risk, where a dif-
ference between 0.001 and 0.002 entails an incremental risk of 0.001 or one per
thousand and the difference between 0.1 and 0.2 entails an incremental risk of
0.1, or of 100 per thousand. Another way to describe risk is the number needed to
expose to observe one added death: the first risk would be described as 1000
needed to expose per additional death, and the second, much more serious risk,
is only 10 needed to expose per additional death.
Unfortunately, most media accounts of the benefits or dangers of particular
pharmaceuticals do not make clear what statistic they are using.

7.6 experimental design and external validity

We have already discussed some experiments—mostly by Amos Tversky and

Daniel Kahneman—designed to understand how people deal intuitively with
statistical problems, and the remainder of the book describes a variety of

35. http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=1694752.
36. For an excellent review of the pros and cons of meta-analysis, as well as description
of its techniques, see Blumenthal, supra.
 interpreting statistical results and evaluating policy interventions 207

experiments on judgment and decision making. For this reason alone, it is worth
spending some time considering the value and limitations of such experiments.
This will also provide an opportunity to review some of the concepts developed
in the preceding chapters.

7.6.1 Within-Subject vs. Between-Subjects Experiments

A number of the empirical studies discussed so far are essentially surveys that do
not involve comparison groups—for example, a survey to determine respon-
dents’ intuitions about the variance of the sex of babies born in hospitals with
many or fewer births. But, as in the Big-Mart and Terra Nueva examples, we are
often interested in comparing two groups.
Suppose that you are trying to learn which of two different sneakers, the
Zephyr or the Rapido, enables an athlete to run faster. You could do a within-
subject experiment with, say, ten athletes, first getting their average time in a
100-yard dash with the Zephyr and then with the Rapido. Or you could do a
between-subjects experiment with two groups of (say) ten athletes each running
the 100-yard dash.37 The within-subjects experiment has the practical advantage
that it only requires ten subjects and the statistical advantage that the average
differences between the two runs does not reflect different abilities of the two
groups. On the other hand, it may be difficult to replicate the experimental con-
dition for the within-subject experiment. For example, if the same ten people
run the 100-yard dash with the Rapido right after having run with the Zephyr, a
difference in the time may reflect their exhaustion from the first run; if they do
it the next day, the track conditions may be different.
Many if not most of the studies reported in this book are between-subjects
experiments, because exposure to the first condition would bias subjects’
response to the second. For example, in Section 10.1 we will describe the phe-
nomenon of anchoring, where subjects’ estimation of a quantity—say, the
number of cities in the United States—is influenced by an arbitrary number
given them before they answer the question. One group might be asked, “How
many cities are there in the United States—is it greater or fewer than 100?” while
a second group would be asked, “How many cities are there in the United
States—is it greater or fewer than 1 million?” It is not hard to imagine that some-
one who has already answered the first question may give a different answer to
the second than someone who starts fresh with the second.
Between-subjects experiments always raise the possibility that different
results reflect differences between the two groups of subjects. As discussed
earlier, however, random assignment of subjects to sufficiently large groups
minimizes the likelihood of this error.

37. The example is taken from David W. Martin, Doing Psychology Experiments
(5th ed. Belmont: Thomson Brooks/Cole, 2000).
208 problem solving, decision making, and professional judgment

7.6.2 External Validity: Generalizability

The preceding chapters have been concerned with internal validity—with whether
a change in a dependent variable is attributable to a change in the hypothesized
independent variable or to some other confounding variable. But even a well-
designed study with statistically significant results may present a question of
external validity—whether those results can be generalized to other situations.
The question can arise in both randomized controlled experiments and observa-
tional studies—though in somewhat different ways.
Although the subjects of the paradigmatic clinical trial are randomly assigned
to the treatment and control groups, they are nonetheless drawn from a popula-
tion which itself is seldom a random sample of the population of all possible
users of the pharmaceutical. For example, some medicines have different effects
on people of different races and ethnicities,38 but not all clinical trials include
representative samples of ethnic groups. Observational studies have these prob-
lems as well—for example, a drug prevention program that is very successful in
a wealthy suburban high school may have little impact in a poor urban high
school, or vice versa. And observational studies are particularly susceptible to the
additional problem of confounding discussed in Section 7.2.1.
By contrast to statistical hypothesis testing, establishing the external validity
of a study is more an art than a science, relying on knowledge or intuition about
whether differences between the group studied and the population to which the
study might be generalized are likely to make any difference in outcomes.
It is instructive to look at the problem of external validity with respect to psycho-
logical studies concerning people’s intuitions about statistics and other matters of
judgment and decision making. We have already discussed some findings, and the
rest of the book will report on many more. To say that these findings have internal
validity is to say that the studies are well designed and implemented and that the
results are statistically significant. But this does not entail external validity.
One psychological finding, known as prospect theory, has been particularly
criticized in this respect. Prospect theory underlies much contemporary behav-
ioral economics, and we will describe its implications in Chapters 14 and 15.39
Here we are interested not in the substance of the theory, but in the methodological
problems of generalizing experimental findings to explain people’s behavior in

38. For example, on March 2, 2005, the Food and Drug Administration required that
the cholesterol drug Crestor be relabeled to add a caution that starter doses should be
reduced in Asian Americans and some other patients. A clinical trial found that levels of
Crestor in Asian patients were double those of Caucasians taking the same dose, increas-
ing the chance of muscle damage. Asians, it seemed, did not metabolize the drug at as
rapid a rate, causing it to stay in their systems longer.
39. See also Charles R. Plott and Kathryn Zeiler, The Willingness to Pay/Willingness to
Accept Gap, the “Endowment Effect,” Subject Misconceptions and Experimental Procedures for
Eliciting Valuations, 95 American Economic Review 530–45 (2004), who argue that a
major series of laboratory experiments were infected by confounding variables.
 interpreting statistical results and evaluating policy interventions 209

the real world. 40 Law professor Chris Guthrie’s defense of the theory addresses
issues that apply to many other psychological experimental findings:41
Critics typically worry about three external validity factors: the subjects who
participate in the experiments, the lack of incentives they face, and the sim-
plicity of experimental designs versus the complexity of decision making in
the real world.
Subjects. Critics often contend that experimental work lacks external valid-
ity because the participants in these studies are typically undergraduates
(often enrolled in introductory Psychology classes). Although undergraduates
differ from other members of the population in terms of age, education level,
life experience, and so on, psychologists have found that they are in fact a
fairly good proxy for “real people.” Generally speaking, expert decision makers
exhibit the same decision-making patterns as undergraduates. Moreover,
many of the participants in the prospect theory studies identified in this arti-
cle were not undergraduates but rather adults with expertise relevant to the
domain being studied. For example, law students and federal magistrate
judges participated in some of the litigation studies; licensed physicians par-
ticipated in the medical treatment study; executive MBA students participated
in some of the tax studies; and corporate managers participated in some of
the managerial decision-making studies.
Incentives. When making decisions in the real world, people generally
have incentives to “get it right.” Some critics suggest that this could mean that
real-world decision makers are more “rational” than participants in experi-
mental studies. Although real-world incentives might induce decision makers
to take more care with their decisions, they do not appear to induce rational
decision-making patterns. In one rather impressive illustration of this, Steven
Kachelmeier and Mohamed Shehata conducted decision-making studies in
China where the researchers could afford to offer substantial monetary incen-
tives relative to local salaries. In one experimental condition, for instance, the
researchers were able to offer subjects an amount roughly comparable to
three months’ worth of income. Despite these substantial monetary incentives,
Kachelmeier and Shehata found that the participants, just like their uncom-
pensated counterparts in American universities, behaved consistent with
prospect theory. In short, incentives do not magically induce people to behave
in accord with rational choice theory.42

40. Chris Guthrie, Prospect Theory, Risk, Preference, and the Law, 97 Nw. U.L. Rev. 1115
(2003).
41. Id. at 1156–59. (The excerpt is based on the text of Guthrie’s article but does not
adhere strickly to it.) Reprinted by special permission of Northwestern University School
of Law, Northwestern University Law Review.
42. See also Colin Camerer and Robin Hogarth, The Effects of Financial Incentives in
Experiments: A Review and Capital-Labor-Production Framework, 7 Journal of Risk and
210 problem solving, decision making, and professional judgment

Experimental Environment. Critics also contend that it is inappropriate

to draw inferences about the complicated world we inhabit based on
responses to simple hypotheticals presented in a controlled setting.
Experimental work in cognitive psychology is generally designed to isolate
the effects of one particular phenomenon. The fact that psychologists (and
others who do similar work) construct problems that are often quite sim-
plistic does not mean, however, that these problems do not illuminate
behavior in the real world.
Lending credence to the external validity of this experimental work is the
fact that many of the prospect theory-based observations reported here are
confirmed by real-world empirical evidence. For example, Rachlinski found
evidence of framing in litigated cases; analyses of tax-returns support the
prospect-theory account of taxpayer compliance; several corporate law studies
of real-world decisions corroborate the prospect-theory account of managerial
decision making; and studies of settlement rates after the adoption of com-
parative fault in Arkansas support the prospect-theory view of the relative
advantages of comparative fault over contributory negligence.
These and other concerns about the external validity of prospect-theory
analyses are certainly legitimate. Prospect theory analyses typically rely on
experimental work conducted in the lab with college students. Extrapolating
from such experimental results should be done with some care. Nonetheless,
the available evidence suggests that external validity concerns about subjects,
the incentives they face, and the laboratory context are often overstated.
Prospect theory has received substantial attention from proponents and
opponents alike, and its empirical findings have withstood this scrutiny.
Granting the limitations of both laboratory experiments and observational
studies, the best of both worlds involves randomized controlled studies con-
ducted in natural settings. The Poverty Action Lab at MIT, mentioned above, is
one of a small but growing number of organizations doing this.

Uncertainty 3–42 (1999), reviewing seventy-four experiments to conclude that “incen-

tives sometimes improve performance, but often don’t. . . . [T]he data shows that higher
levels of incentives have the largest effects in judgment and decision tasks. Incentives
improve performance in easy tasks that are effort-responsive, like judgment, prediction,
problem-solving, recalling items from merely or clerical tasks. Incentives sometimes hurt
when problems are too difficult or when simple intuition or habit provides an optimal
answer and thinking harder makes things worse. In games, auctions, and risky choices
the most typical result is that incentives do not affect mean performance, but incentives
often reduce variance in responses.” The authors note that no replicated study has made
rationality violations disappear purely by raising incentives.
8. explaining and predicting
one-time events

Up to this point, we have been using statistics to look for regularities in situa-
tions in which there are many similar events—an employer’s setting the wages
of employees, many residents’ being exposed (or not) to foam insulation in their
apartments. Statistical analysis of this sort plays an important role not just in
law and public policy but also in many other aspects of people’s personal and
professional lives.
In this chapter, we are interested in the probability that an event will occur
one time. We begin by reviewing our treatment of conditional probability begun
in Chapter 5, and then turn to Bayesian (or subjectivist) statistics, which is a
quite different approach from the frequentist approach explored thus far.
Note to Students: This chapter derives, uses, and transforms a number of formulas,
which readers who haven’t done math for a while may find daunting. It’s not essential
that you remember any of them—only that you understand the basic concepts of prob-
ability they express.

8.1 calculating conditional probabilities (a review)

Recall that a conditional probability is the probability that event A will occur,
given that event B has occurred. It is written
P(A |B)
and is read: “the probability of A given B.” For example, in the Terra Nueva case,
P(rash present|foam) is “the probability of having a rash given that the tenant lives
in an apartment with foam insulation.” We repeat the contingency table for
Terra Nueva in Table 8.1. In view of the focus of this chapter, we’ll use its
synonym of probability table.
In Chapter 5, we found that:
P(rash present| foam) = P(foam & rash present)/P(foam)
which turns out to be 0.25/0.60 = 0.42.
Here is a generalized probability table, as shown in Table 8.2. (The symbol
“∼” (tilde) means “not.”)
Remember that the probabilities in the margins of the table, such as P(A), are
called marginal probabilities. The probability P(A & B) is called a joint probability.
212 problem solving, decision making, and professional judgment

table 8.1 probability table for terra nueva

Rash

Present Absent Marginal

Foam P(foam & rash) P(foam & no rash) P(foam)

Insulation Not Foam P(not foam & rash) P(not foam & no rash) P(not foam)

Marginal P(rash) P(no rash) 1.0

Rash
Probabilities:
Present Absent Marginal

Foam 0.25 0.35 0.60

Insulation Not Foam 0.15 0.25 0.40

Marginal 0.40 0.60 1.00

To generalize the calculation above, the conditional probability of event B given

A is the joint probability of A and B divided by the marginal probability of A:

P(A & B)
P (B | A) =
P(A)

While joint probabilities are symmetric—P(A & B) = P(B & A)—conditional

probabilities are not: P(A|B) ≠ P(B|A). As we see in the formula, the denominator
for determining P(B|A) is P(A), the row marginal, but the denominator for deter-
mining P(A|B) would be P(B), the column marginal.
P(A& B)
P(A |B) =
P(B)

table 8.2 generalized probability table

B
B ∼B Total
A P(A & B) P(A & ∼B) P(A)
A
∼A P(∼A & B) P(∼A & ∼B) P(∼A)
Total P(B) P(∼B) 1.00
 explaining and predicting one-time events 213

In the Terra Nueva example, we have

P(rash present| foam) = P(foam & rash present)/P(foam) = 0.25/0.60 = 0.42

P(foam|rash present) = P(foam & rash present)/P(rash present)

= 0.25/0.40 = 0.625
While the former is the proportion of foam-insulation residents who have a
rash, the latter is the proportion of rash-afflicted tenants whose premises have
foam insulation.

8.2 the probability of conjunctive events

In People v. Collins, an interracial couple was prosecuted for robbery.1 Here are
the facts, as summarized by the Supreme Court of California:
On June 18, 1964, about 11:30 a.m. Mrs. Juanita Brooks, who had been
shopping, was walking home along an alley in the San Pedro area of the City
of Los Angeles. She was pulling behind her a wicker basket carryall contain-
ing groceries and had her purse on top of the packages. She was using a cane.
As she stooped down to pick up an empty carton, she was suddenly pushed to
the ground by a person whom she neither saw nor heard approach. She was
stunned by the fall and felt some pain. She managed to look up and saw a
young woman running from the scene. According to Mrs. Brooks the latter
appeared to weigh about 145 pounds, was wearing “something dark,” and had
hair “between a dark blond and a light blond,” but lighter than the color of
defendant Janet Collins’ hair as it appeared at trial. Immediately after the
incident, Mrs. Brooks discovered that her purse, containing between $35 and
$40, was missing. About the same time as the robbery, John Bass, who lived
on the street at the end of the alley, was in front of his house watering his
lawn. His attention was attracted by “a lot of crying and screaming” coming
from the alley. As he looked in that direction, he saw a woman run out of the
alley and enter a yellow automobile parked across the street from him. He
was unable to give the make of the car. The car started off immediately and
pulled wide around another parked vehicle so that in the narrow street it
passed within six feet of Bass. The latter then saw that it was being driven by
a male Negro, wearing a mustache and beard. At the trial Bass identified the
defendant as the driver of the yellow automobile. However, an attempt was
made to impeach his identification by his admission that, at the preliminary
hearing, he testified to an uncertain identification at the police lineup shortly
after the attack on Mrs. Brooks, when defendant was beardless.

1. People v. Collins, 68 Cal.2d 319, 66 Cal Rptr. 497, 438 P.2d 33 (1968).
214 problem solving, decision making, and professional judgment

In his testimony Bass described the woman who ran from the alley as a
Caucasian, slightly over five feet tall, of ordinary build, with her hair in a dark
blond ponytail and wearing dark clothing. He further testified that her pony-
tail was “just like” one which Janet had in a police photograph taken on June
22, 1964.
At the trial an expert testified as to the probability of the occurrence of the
physical characteristics of the defendants and the car:

Partly yellow automobile 0.1

Man with mustache 0.25
Girl with ponytail 0.1
Girl with blond hair 0.33
Negro man with beard 0.1
Interracial couple in car 0.0001

The expert assumed that these six characteristics were independent, and cal-
culated that there was only one chance in 12 million that any couple possessed
all of these distinctive characteristics.

8.2.1 The Probability of Dependent Conjunctive Events

Even assuming the accuracy of the expert’s assessments of the probability of the
occurrence of each of the six events in the Los Angeles area, he was incorrect in
assuming their independence. In reversing the convictions, the Supreme Court
of California noted that there was some likely “overlap” in being a man with a
mustache (0.25) and being a Negro man with a beard (0.1).
If these were independent, you would just use the multiplication rule as
described below in 8.2.2. But it’s pretty obvious that they are not. Let’s assume
that a survey of the region shows that for every 10 men with mustaches, 9 have
beards. How do we calculate the probability of a man’s having both a mustache
and a beard, P(mustache & beard)?
If we take the conditional probability formula in Section 8.1,

P(A& B)
P (B |A) =
P(A)

and multiply both sides by P(A), we get the equivalent equation2

P(A&B)=P(A) P(B |A)

2. By the same token, multiplying the formula for P(A|B) by P(B) on both sides, would
give P(A & B) = P(B)P(A|B).
 explaining and predicting one-time events 215

So, we can calculate the joint probability of the events A and B, P(A & B),
using the marginal probability of A and the conditional probability of B given A,
or the marginal probability of B and the conditional probability of A given B.
In our example,
P(mustache & beard) = P(mustache) P(beard | mustache)
If for every 10 men with mustaches, 9 have beards, then
P(beard | mustache) = 0.9
Thus: P(mustache & beard)
= P(mustache) P(beard | mustache)
= 0.25 * 0.9
= 0.225

What American president had a beard but no mustache? Look at a $5 bill.

8.2.2 Conjunctive Independent Events

Recall that two events are independent if knowledge about one does not change
one’s belief about the probability of the other. That is, A and B are independent
if P(B|A) = P(B) and P(A|B) = P(A). If all the relevant events are independent, the
formulas in Section 8.2.1 for joint probabilities become
P(A & B) = P(A)P(B)
This is the multiplication rule for independent events. Under the Collins
expert’s mistaken view that mustache and beard are independent,
P (mustache & beard) = P(mustache) P(beard)
= 0.1 * 0.25
= 0.025
Noting that the expert’s assumption of independence made it seem less likely
that a randomly chosen man would have both a beard and a mustache, and thus
more likely that the accused couple was at the scene of the event than the facts
justified. The Supreme Court of California reversed the convictions.3

3. The court also noted that the expert’s conclusions about the probabilities of the
particular events were not supported by evidence. The court did not address the accuracy
of the eyewitness identification—e.g., P(beard|witness says beard)—which is presented in
the taxicab problem in Section 8.5.
216 problem solving, decision making, and professional judgment

8.2.3 Intuitions about Conjunctions

Suppose that Luis Trujillo has a brief due in a complex case, and has assigned its
three major sections to three associates. Based on past experience and the
amount of work to be done, he estimates that these are the probabilities that they
will complete their sections by next Friday:
• Amy: 0.9
• Bill: 0.8
• Cindy: 0.9
If events A, B, and C are independent of each other, you obtain their joint
probability by multiplying the (marginal) probabilities:
P(A & B & C) = P(A) * P(B) * P(C) = 0.65
Are you surprised at how low this result is? People tend to overestimate the
probability of conjunctive events—multipart events in which each stage or com-
ponent needs to be successful for the entire strategy to be successful. This leads
to unwarranted optimism about the likelihood of the success of multistage plans
(e.g., developing and launching a new product) or about the unlikelihood of the
failure of complex, interactive systems (e.g., a nuclear power plant).
Kahneman and Tversky explain this tendency in terms of anchoring and
adjustment, which we will discuss in more detail in Section 10.1. In brief, people
tend to focus on the probability of success of any one part or stage of the multi-
part event. They “anchor” their estimation of the overall probability on this
number, and make insufficient adjustment for the other stages.

National Intelligence Estimates

The simplest, easiest, cheapest and most powerful way to transform the quality
of intelligence would be to insist that analysts attach two little numbers to every
report they file.
The first number would state their confidence in the quality of the evidence
they’ve used for their analysis: 0.1 would be the lowest level of personal/
professional confidence; 1.0 would be—former CIA director George Tenet should
pardon the expression—a “slam dunk,” an absolute certainty.
The second number would represent the analyst’s own confidence in his or
her conclusions. Is the analyst 0.5— the “courage of a coin toss” confident—or
a bolder 0.75 confident in his or her analysis? Or is the evidence and environment
so befogged with uncertainty that the best analysts can offer the National Security
Council is a 0.3 level of confidence?
 explaining and predicting one-time events 217

These two little numbers would provoke intelligence analysts and intelligence
consumers alike to think extra hard about analytical quality, creativity and account-
ability. Policy makers could swiftly determine where their analysts had both the
greatest—and the least—confidence in their data and conclusions. Decision-
makers could quickly assess where “high confidence” interpretations were based
on “low-confidence” evidence and vice versa.
Michael Schrage, What Percent Is ‘Slam Dunk’?” WASHINGTON POST, Feb. 20,
2005, editorial.

8.3 the conjunction fallacy and the

representativeness heuristic

One consequence of the law of total probability is that the joint probability of two
events must always be less than or equal to the probability of one of those events.
P(A & B) can never be greater than just P(A) or just P(B).
One of the most famous results in the judgment and decisionmaking (JDM)
research suggests that people do not obey the law of total probability: they some-
times give more probability to the conjunction of two events than they give to the
two events on their own. This is called the conjunction fallacy.4
Perhaps the most common context in which people succumb to the conjunc-
tion fallacy is where the conjunction of two events seems intuitively more
plausible than one of the events alone. For example, read the paragraph below:
Linda is 31 years old, single, outspoken and very bright. She majored in phi-
losophy. As a student, she was deeply concerned with issues of discrimina-
tion and social justice, and also participated in anti-nuclear demonstrations.
Now, rate the following options in order of likelihood:
1. Linda is a teacher in an elementary school.
2. Linda is active in the feminist movement.

4. The example below, and any other examples in this section given without refer-
ences, are from studies described in detail in Amos Tversky and Daniel Kahneman,
Judgments Of and By Representativeness, quoted in Judgment Under Uncertainty:
Heuristics and Biases (Daniel Kahneman, Paul Slovic and Amos Tversky eds., New
York: Cambridge University Press, 1982) [hereafter, Heuristics and Biases] The con-
junction fallacy is discussed further in Amos Tversky and Daniel Kahneman, Extensional
Versus Intuitive Reasoning: The Conjunction Fallacy in Probabilistic Judgment, 90
Psychological Review 293–315 (1983).
218 problem solving, decision making, and professional judgment

3. Linda is a bank teller.

4. Linda is a bank teller and active in the feminist movement.
How did you order the options? Many people rate option 4 as more likely than
option 3. Since the event described in option 4 is the conjunction of options
2 and 3, the law of total probability says that we should never rate option 4 as
more probable than option 3. If your intuitions and logic are in conflict, you’re
in good company. The eminent biologist Stephen Jay Gould wrote: “Knowledge
of the truth does not dislodge the feeling that Linda is a feminist bank teller: I
know [the right answer], yet a little homunculus in my head continues to jump
up and down, shouting at me—‘but she can’t just be a bank teller; read the
description.’”5
Daniel Kahneman and Amos Tversky explain the conjunction fallacy in the
“Linda problem” in terms of the representativeness heuristic. The description of
Linda is very similar to what we might expect for somebody who is active in the
feminist movement, but not at all similar to our stereotypical idea of a bank
teller. But since the description of Linda also seems more similar to a feminist
bank teller than to that of an ordinary bank teller, the representativeness heuris-
tic leads us to commit the conjunction fallacy. In effect, the description leads one
to assign too low a probability that Linda is merely a bank teller, or too high a
probability that she is a feminist bank teller.
Kahneman and Tversky describe representativeness as “an assessment of the
degree of correspondence between a sample and a population, an instance and a
category, an act and an actor, or, more generally between an outcome and a
model.” Under the representativeness heuristic, “probability judgments (the
likelihood that X is a Y) are mediated by assessments of resemblance (the degree
to which X ‘looks like’ Y).”6 In Chapter 1, we mentioned that the representative-
ness heuristic could lead people to attribute causes to phenomena that resemble
the effect (smelly foam insulation causes distressing physical symptoms).
Determining correct probabilistic solutions is a very challenging task. Daniel
Kahneman and Shane Frederick write: “When confronted with a difficult ques-
tion people often answer an easier one instead, usually without being aware of
the substitution.”7 We substitute the target attribute (the likelihood that Linda is

5. Quoted in Daniel Kahneman and Shane Frederick, Representativeness Revisited:

Attribute Substitution in Intuitive Judgment (Aug. 2001), in Heuristics and Biases, supra
at 103. The “Linda” problem and the representativeness heuristic more generally have
spawned a considerable critical (and defensive) literature. See id.; Barbara Mellers, Ralph
Hertwig and Daniel Kahneman, Do Frequency Representations Eliminate Conjunction
Effects, 12 Psychology Science 269 (2001).
6. Daniel Kahneman and Shane Frederick, Representativeness Revisited: Attribute
Substitution in Intuitive Judgment, in Heuristics and Biases, supra at 49.
7. Id.
 explaining and predicting one-time events 219

a feminist bank teller) with the heuristic attribute (Linda’s resemblance to a femi-
nist bank teller).

8.3.1 Explanation Through Narrative

Jerome Bruner writes:8
There are two modes of cognitive functioning, two modes of thought, each
providing distinctive ways of ordering experience, of constructing reality. . . .
They differ radically in their procedure for verifications. A good story and a
well-formed argument are different natural kinds. Both can be used as means
for convincing another. Yet what they convince of is fundamentally different:
arguments convince one of their truth, stories of their lifelikeness. The one
verifies by eventual appeal to procedures for establishing formal and empiri-
cal proof. The other establishes not truth but verisimilitude.
Which of these scenarios seems more likely?9
Scenario A: A war involving the United States and North Korea.
Scenario B: A situation in which neither side intends to attack the other side,
but a war between the United States and North Korea occurs
against the background of U.S. efforts to isolate North Korea eco-
nomically; it is triggered by the U.S. Navy’s seizure of a North
Korean ship containing a shipment of missiles followed by mas-
sive movement of North Korean troops near the DMZ.
Many people choose the second and more detailed scenario, despite the fact
that providing more details actually decreases the probability of the scenario. As
Tversky and Kahneman observed, “as the amount of detail in a scenario
increases, its probability can only decrease steadily, but its representativeness
and hence its apparent likelihood may increase.” They conclude that people’s
reliance on representativeness “is a primary reason for the unwarranted appeal
of detailed scenarios and the illusory sense of insight that such constructions
often provide.”10 (This is a potential hazard of of scenario planning, discussed
in Section 3.4.)
This should not be surprising, given the role that stories or narratives play
in our lives. Narratives are part and parcel of our schematic ordering of
experience (see Section 1.5.2.); it is through narratives that we understand and

8. Jerome Bruner, Actual Minds Possible Worlds 11 (Boston, MA: Harvard

University Press, 1986).
9. This is an updated version of an example described in Scott Plous, The
Psychology of Judgment and Decision Making (New York: McGraw Hill, 1993).
10. Amos Tversky and Daniel Kahneman, Judgments of and by Representativeness, in
Heuristics and Biases, supra at 98.
220 problem solving, decision making, and professional judgment

explain the world. In this respect, consider Table 8.3 (which provides its own
narrative).

table 8.3 details, probability, and verisimilitude

Question Never Once or More than

twice twice

How often has this happened to you: Waking up 87 36 21

paralyzed with a sense of a strange person or
presence or something else in the room?
How often has this happened to you: Waking up 124 12 8
paralyzed?

Poll conducted by Robin Dawes and Matthew Mulford. In KENDRICK FRAZIER, BARRY KARR, AND
JOE NICKELL, THE UFO INVASION: THE ROSWELL INCIDENT, ALIEN ABDUCTIONS, AND GOVERNMENT
COVERUPS (1997)

In a study of jury decision making, Nancy Pennington and Reid Hastie suggest
that, given a number of possible competing versions of the facts, jurors will select
the “best story,” with their confidence being determined by several factors:11
• coverage—how well the story accounts for all the evidence;
• coherence—which consists of:
• consistency—the absence of internal contradictions,
• plausibility—correspondence with the decision maker’s knowledge of
what typically happens in the world, and
• completeness—the story’s inclusion of relevant causes, psychological
states, actions, goals, consequences;
• uniqueness—the absence of good alternative competing stories.
Notice that a number of these criteria of confidence depend on detailed rather
than parsimonious data. We share Pennington and Hastie’s belief that the same
criteria are germane to explanations in many domains beyond the courtroom.
Persuasion often consists less of convincing others of the logic of one’s argu-
ment than having them replace their narrative construction of reality with your
own.12 And the richer and more detailed the story, the more persuasive. In these
respects, Scenario B in the North Korean conflict is far superior to A.

11. Nancy Pennington and Reid Hastie, Inside the Juror: The Psychology of
Juror Decision Making (New York: Cambridge University Press, 1993). See The Story
Model for Juror Decision Making and A Theory of Explanation-Based Decision Making.
12. See Howard Gardner, Changing Minds ch. 4 (Boston, MA: Harvard Business
School, 2004); Jerome Bruner, Making Stories (Boston, MA: Harvard University Press,
 explaining and predicting one-time events 221

The phenomenon (illustrated by the Linda and Korean War examples) in

which the sum of the probabilities of necessary components of an outcome are
greater than the probability of the outcome or, indeed, greater than 1.0, is (some-
what confusingly) called subadditivity. We can think of at least one kind of situa-
tion where subadditivity makes sense. Imagine a jury determining whether or
not a defendant committed homicide. It has inconclusive evidence (1) of his
presence at the crime scene, and (2) about his motivation to kill the deceased.
A juror might rationally conclude that the defendant was more likely to kill the
victim with some particular motive than to think that he killed him at all, with no
motive given.13
Generally, though, subadditivity is simply a logical error. A set of experiments
by Craig Fox and Richard Birke suggests how the phenomenon may bias a law-
yer’s advice to a client.14 For example, when asked (in 1997) to predict the out-
come of Paula Jones’s suit against Bill Clinton, lawyers who were given the two
options of the case ending in verdict or no verdict assigned a probability of 0.2 to
the former and 0.75 to the latter. When given the options of a verdict, settlement,
dismissal, immunity, or withdrawal of the suit, the probability of a verdict
remained at 0.2, but the nonverdict probabilities summed to 1.29. “[U]npacking
the description of an event into more detailed description of disjoint compo-
nents will generally yield a higher judged probability” because it draws one’s
attention to each component.
Fox and Birke asked lawyers to advise the plaintiff ’s counsel whether to accept
an offer of settlement in a personal injury case. All of the subjects were given the
identical scenario detailing what the plaintiff must prove to prevail at trial, but
for one group, these were described as necessary to establish “liability,” while for
the other they were described in terms of the elements of a torts claim: “duty,
breach, and causation.” Seventy-four percent of the second group (compared to
52 percent of the first group) recommended settling—because unpacking the
elements made the difficulty of proof seem greater. As Fox and Birke note, an
attorney’s advice might flip depending on whether she was predicting whether a
client would “prevail” or “lose.”
To mitigate the bias, Fox and Birke suggest that attorneys (1) consider multi-
ple formulations of the probability of the event at various levels of specificity;
(2) make an effort to explicitly consider all possible scenarios while ensuring that
the all the probabilities sum to 100 percent; and (3) base their predictions on
good base rate data (See Section 8.5).

2002); Anthony G. Amsterdam and Jerome Bruner, Minding the Law (Boston, MA:
Harvard University Press, 2002).
13. Mark G. Kelman, The Heuristics Debate: Its Nature and Its Implications
for Law and Policy (Oxford University Press, forthcoming 2010).
14. Craig R. Fox and Richard Birke, Forecasting Trial Outcomes: Lawyers Assign Higher
Probabilities to Possibilities that Are Described in Greater Detail, 26 Law and Human
Behavior 159 (2002).
222 problem solving, decision making, and professional judgment

More broadly, since Kahneman and Tversky demonstrated the existence of

the conjunction fallacy, psychologists have explored ways of thinking about
probabilistic information that help us to avoid making representativeness errors.
In particular, Gerd Gigerenzer and his research group have argued that the best
way not to get confused by probabilities is to turn them into more concrete quan-
tities. A particularly effective method is to express the probabilities in frequen-
cies and consider one variable at a time. For example, out of 1000 people, how
many are bank tellers? Say the number of bank tellers is about ten. Then how
many of those bank tellers are feminists? Maybe two or three? By phrasing the
probabilities explicitly in terms of frequencies, and thinking about one variable
at a time, it becomes very hard to think that there could be more feminist bank
tellers than there are bank tellers in total.15

8.4 the probability of disjunctive events

In Section 8.2, we calculated a joint probability—the probability that event A and

event B happen. Here we discuss the probability that event A or event B happens.
In ordinary discourse, “A or B” can mean either “A or B but not both” or it
can mean “A or B or both.” In probability, we always take or to have the latter
meaning. Another way this can be stated is “at least one of A and B.”
For example, if you throw a pair of dice, what’s the probability of getting at
least one ace? Let A represent the event that the first die is an ace and B the event
that the second die is an ace. Then the probability of getting at least one ace is:
P(A or B) = P(A) + P(B) – P(A & B)
To understand this formula, consider the Venn diagram of Figure 8.1. The
area of the circle on the left represents P(A). The area of the circle on the right
represents P(B). If we were to add these two areas, we would be double-counting
the area of the shaded region, which represents P(A&B) (in the example, the
probability that both dice are aces). Therefore, we subtract the area of the shaded
region to obtain the total area.
In the dice example, since there are 6 sides of a die, P(A) = 1/6 and P(B)=1/6.
Because the two tosses are independent, P(A & B) = P(A)P(B) = (1/6)(1/6) = 1/36.
Therefore, the probability of getting at least one ace is
P(A or B) = P(A) + P(B) – P(A & B) = 1/6 + 1/6 – 1/36 = 11/36 = 0.31

15. Gerd Gigerenzer, Adaptive Thinking: Rationality in the Real World 250
(Oxford: Oxford University Press, 2000). A complete discussion of the consequences of
using frequencies for probabilistic reasoning is given in Gerd Gigerenzer and Ulrich
Hoffrage, How to Improve Bayesian Reasoning Without Instruction: Frequency Formats, 102
Psychological Review 684–704 (1995).
 explaining and predicting one-time events 223

A B

figure 8.1 venn diagram of events a and b.

or 31 percent.
If two events are mutually exclusive—that is, it is impossible for both to hap-
pen—the circles in the Venn diagram will not intersect. In this case, P(A & B) = 0,
so the formula simplifies to
P(A or B) = P(A) + P(B)
(This is Axiom 3 in Section 5.1.2) For example, what is the probability that the
toss of a single die will result in an ace or a deuce? Since these events are mutu-
ally exclusive, we just add their probabilities: 1/6 + 1/6 = 1/3 = 0.33 or 33
percent.
Whenever you are interested in the probability that at least one of a number of
events will happen, summing all the probabilities and subtracting all the over-
laps can be tedious. There’s a simpler way to calculate the probability of at least
one of the events occurring. Consider this example:
It is late in the afternoon in California and you must send a signed document
to someone in New York, who must have it by 9 a.m. the next morning.16 You
can send the document simultaneously through three overnight services,
each of which costs about $20, with these probabilities of getting the docu-
ment to the destination on time:

• AirFast: 0.9
• BeThere!: 0.8
• Chronos: 0.8

Alternatively, you can give it to a courier who will take it with him on the red-
eye, at a cost of $2000. Before deciding whether to opt for this high-cost

16. The example comes from Amir D. Aczel, Chance: A Guide to Gambling, Love,
the Stock Market, and Just about Everything Else (New York: Thunder’s Mouth
Press, 2004).
224 problem solving, decision making, and professional judgment

option, you would like to know the probability of the document getting to
New York on time if you send three signed copies, one with each service.
We are interested in the probability that at least one of these services will
deliver on time. Let A represent the event that AirFast is on time, B the event that
BeThere! is on time, and C the event that Chronos is on time. We want to find
P(A or B or C) Instead of adding the probabilities and adjusting for the various
overlaps, we notice that the event that at least one happens is the complement of the
event that none happen. Recall that the probability of an event A’s complement is
P(∼A) = 1 – P(A) In this case,
P(A or B or C) = 1 – P(~A & ~B & ~C)
If we know that the three events are independent, we can use the multiplica-
tion rule to find this last probability.
P(A or B or C) = 1 – P(~A & ~B & ~C)
= 1 – P(~A)P(~B)P(~C)

Using the complement rule again on each of the three probabilities, we get:

P(A or B or C) = 1 – P(~A & ~B & ~C)

= 1 – P(~A)P(~B)P(~C).
= 1 – (1-P(A))(1-P(B))(1-P(C))
= 1 – 0.1 * 0.2 * 0.2
= 1 – 0.004
= 0.996
Thus, it is extremely likely that the document will get to New York on time.17

In Bomber, Len Deighton observed that a World War II pilot had a 2 percent
chance of being shot down on each mission, and that he was “mathematically
certain” to be shot down in fifty missions. But this is incorrect.17
The pilot will not survive if he is shot down at least once in the fifty missions.
This phrase is a hint to use the complement rule. The event of being shot down
at least once is the complement of the event of being shot down none of the
times.
Let D represent the event of being shot down at least once.

17. David Freedman, Robert Pisani, and Roger Purves, Statistics (3rd ed. New
York: Norton, 1997).
 explaining and predicting one-time events 225

P(D) = 1 – P(never shot down)

= 1 – P(not shot on 1st mission)P(not shot on 2nd
mission . . . P(not shot on 50th mission)
= 1 – (0.98)50
= 1 – 0.36
= 0.64
There is a 64 percent chance (not 100 percent) of being shot down in fifty
missions.

8.4.1. Intuitions about Disjunctive Events

Just as people tend to overestimate the probability of conjunctive events, they
tend to underestimate the probability of disjunctive events—multipart events
where success of any one stage or component makes the entire event successful,
such as at least one of the three courier services getting the signature page to
New York on time. This is due to the same anchoring phenomenon, since the
outcome is greater than any of the component probabilities.

8.5 bayesian statistics

The statistical techniques discussed in the preceding three chapters are based
upon the frequentist interpretation of probability—the idea that probabilities
refer to the frequencies with which events occur in the world, over many trials.
If an event has a probability of 0.5, it will occur half the time. This means that the
probability assigned to an event expresses its frequency of occurrence, which can
be estimated directly from experience.
As we mentioned at the start, however, there is another way of interpreting
the axioms underlying probability theory. Many modern statisticians also sub-
scribe to a subjectivist interpretation of probability, under which probabilities
indicate the strength of belief in a proposition. If something has a probability of
0.5, it has a 50 percent chance of being true. This means that the probability you
assign to an event expresses something about your beliefs. Both the frequentist
and subjectivist interpretations are consistent with the axioms of probability,
although they involve different perspectives about what probability means.
Frequentist statistics is concerned with P(D|H) the probability of a set of data
D under a hypothesis H, typically the null hypothesis H0. A frequentist interpre-
tation assumes that a hypothesis is either true or false, not as a random variable
but as a fixed state of the world, albeit unknown. However, with a subjectivist
226 problem solving, decision making, and professional judgment

interpretation, it is also possible to talk about P(H|D) the strength of one’s belief
in a hypothesis H after observing data D. Often lawyers and courts are more
interested in this latter quantity—for example, the probability that a defendant is
guilty based on the evidence presented—so it is worth learning to evaluate it.
Consider the following problem:
A taxicab was involved in a hit-and-run accident at night. Two cab companies,
the Green and the Blue, operate in the city. You are given the following infor-
mation:
1. 85 percent of the cabs in the city are Green; 15 percent are Blue.
2. A witness identified the cab as a Blue cab. The court tested her ability to
identify cabs under appropriate visibility conditions. When presented with
a sample of cabs (half of which were Blue and half of which were Green),
the witness made correct identifications in 80 percent of the cases and
erred in 20 percent of the cases.

What is the probability that the cab involved in the accident was Blue rather
than Green?
Here our hypothesis H is that the cab was Blue; we will call it B. Our data D
is the eyewitness testimony that the cab was Blue; we will call the event that she
says it was Blue SB. Our goal is to find P(H|D) in this case P(B|SB) the probabil-
ity that the cab was actually Blue, given that the witness said it was Blue.
First, let’s go through the given information.
• The proportion of taxis in the city that are Blue is called the base rate; it is
background information not specific to the particular event. If you had no
information about the accident, the probability that the cab in the accident
was Blue would be the base rate. For this reason it is also called the prior
probability that the cab was Blue. It is your best guess for the probability
that the cab was Blue prior to hearing the testimony. In general, the prior
probability P(H) is the strength of your belief in H prior to obtaining event-
specific data. The prior probability can be determined in different ways,
depending upon the situation. Sometimes it is estimated from a belief
based on the investigator’s experience, but typically it is taken to be the
base rate. In the hit-and-run case, it is the base rate of Blue taxis, P(B) =
0.15 Since this prior probability is substantially less than 50 percent, your
best guess before hearing the testimony would be that the cab was Green.
• The fact that the witness says the taxi was Blue (SB), is the data, D.
• You have one other important additional piece of information: the proba-
bility that the witness’s perception was accurate. This is the probability that
the witness says the cab is Blue, given that it was actually Blue, P(SB|B)
In general, this is P(D|H) and is called the likelihood of the data under
 explaining and predicting one-time events 227

the hypothesis. It represents how likely the observed data would be if the
hypothesis was true. In the example, the likelihood is P(SB|B) = 0.8.

8.5.1 Structuring the Taxi Problem as a Tree

As we said before, our goal is to find P(B|SB), the probability that the cab was
actually Blue, given that the witness said it was Blue. This probability, generically
written P(H|D), represents the strength of our belief in H after we have obtained
the data D about the event. For this reason, it is called the posterior probability of
the hypothesis. In summary, we need a way to find the posterior probability based
on the information we have: the prior probability, the data, and the likelihood.
Let’s represent the problem in a tree structure as shown in Figure 8.2.18

says blue
0.17
Green 0.20
0.85 says green
0.80
says blue
Blue 0.80 0.12
0.15 says green
0.20

figure 8.2 structuring the taxi problem as a tree.

The leftmost node represents the base rate, with 85 percent of the taxis being
Green and 15 percent being Blue. The next nodes represent the likelihood, that
is, the probability of the data if the given hypothesis is true. In this example, the
likelihood is the probability that the witness says the taxi was a particular color,
given its true color. She is accurate 80 percent of the time, so when the taxi was
actually Green (upper node), she says it was Green 80 percent of the time and
Blue 20 percent of the time. When the taxi was actually Blue (lower node), she
says it was Blue 80 percent of the time and Green 20 percent of the time.
Now let’s solve, or “fold back,” the tree by using the methods of Section 8.2—
that is, by multiplying the probabilities along each branch. For example, the
probability that the cab was Blue and the witness says it was Blue is
P(B & SB) = P(B)P(SB|B) = 0.15 * 0.80 = 0.12
or 12 percent. Similarly, if we let G represent the hypothesis that the cab was
Green, the probability that it was Green and she says it was Blue is
P(G & SB) = P(G)P(SB|G) = 0.85 * 0.20 = 0.17

18. The concept of a decision tree is described in greater detail in Section 15.2.
228 problem solving, decision making, and professional judgment

or 17 percent. (Since she said it was Blue, we can ignore the second and fourth
branches.) The witness says the cab was Blue 17 + 12 = 29 out of 100 times and
is correct 12 of those times. Dividing 12 by 29, we obtain 0.41, or 41 percent.
Given her testimony that the cab was Blue, there is still only a 41 percent chance
that the cab actually was Blue, so our best guess is that the cab was Green. If this
answer is surprising, keep in mind the base rate: only 15 percent of cabs are
Blue. As we discuss below, people tend to ignore the base rate when estimating
probabilities.

8.5.2 Solving the Taxi Problem Using Bayes’ Theorem

The tree structure uses Bayes’ Theorem—developed by the Reverend Thomas
Bayes, an early eighteenth-century minister and mathematician. The equation
for Bayes’ Theorem is:

P(D |H ) P(H)
P(H |D ) =
P(D)

Note to Students: If you understand how to represent and analyze the taxi problem
using a decision tree, you’ve got the gist of Bayes’ Theorem. But the chapter would be
incomplete without walking through how the equation actually works.
Let’s replace the general formulation of Bayes’ Theorem with the particulars
of the taxi problem. We were given the likelihood, P(SB|B): it is 0.8. We were also
given P(B), the base proportion of Blue cabs: it is 0.15. Substituting in Bayes’
equation, we get:

P(SB | B )P(B) 0.8 * 0.15

P (B |SB ) = =
P(SB) P(SB)

Now we need the value of the denominator, P(SB), which is the overall, or mar-
ginal, probability that the witness will say the cab was Blue. Now letting SG
represent the event that the witness says the cab was Green, the probabilities of
this problem are displayed in Table 8.4.

table 8.4 probability table for the taxi problem

Probabilities Testimony

SB SG Total
B P(B & SB) P(B & SG) P(B)
Truth
G P(G & SB) P(G & SG) P(G)
Total P(SB) P(SG) 1.00
 explaining and predicting one-time events 229

There are only two ways the event SB—the witness saying it was a blue cab—
can come about: either the cab is blue and she says it’s blue, or the cab is green
and she says it’s blue. These are mutually exclusive and collectively exhaustive.
Therefore,19 P(SB)—the marginal probability of the testimony that the cab was
blue—is the sum of (a) the probability that the witness says the car was Blue and
it was in fact Blue, and (b) the probability that she says the car was Blue and it
was in fact Green:
P(SB) = P(B & SB) + P(G & SB)
As with the example of finishing the brief on time (in Section 8.2.1 and
Section 8.2.3), we calculate the two terms:
P(B & SB) = P(SB|B) * P(B)
P(G & SB) = P(SB|G) * P(G)
Since the witness is correct 80 percent of the time, the probability that she will
say the taxi was Blue if it was Blue is 0.8, and the probability that she will say it
was Blue if it was Green is 0.2. Plugging in these numbers, we get:
P(B & SB) = P(SB|B) * P(B) = 0.8 * 0.15 = 0.12
P(G & SB) = P(SB|G) * P(G) = 0.2 * 0.85 = 0.17
P(SB) = P(B & SB) + P(G & SB)
= 0.12 + 0.17
= 0.29
Above, we had the partially solved equation

0.8 * .015
P (B | SB) =
P(SB)

Now that we know that P(SB) is 0.29, we can plug that into the denominator,
getting:

0 .8 * 0 .15
P (B | SB) = = 0 .41
0 .29

8.5.3 Diagnosticity
Recall that we can discuss uncertainty in terms of odds instead of probabilities.
Here, just as we have prior and posterior probabilities of an event, we have prior

19. See Section 5.4.1.

230 problem solving, decision making, and professional judgment

and posterior odds, the odds of the event before and after we observe the data,
respectively.
In the previous section, with the Bayes tree method, we used the likelihood to
transform the prior probability—P(B) = 0.15—into the posterior probability—
P(B|SB) = 0.41. To transform the prior odds into the posterior odds, we use the
likelihood ratio.
Recall from the previous section that the likelihood is the probability of
observing data given a certain hypothesis, P(D|H) In the example, the data (D)
consisted of the testimony SB that the cab was Blue. The likelihood of this testi-
mony under the hypothesis (H) that the cab was Blue is P(SB|B) = 0.8 and the
likelihood under the hypothesis that the cab was Green is P(SB|G) = 0.2. The
likelihood ratio, therefore is

P (SB | B ) 0 .8
= =4
P (SB | G ) 0 .2

This means that the testimony that the cab was Blue is four times as likely if
it actually was Blue than if it was Green. In general, the likelihood ratio is

P (D | H)
P (D | ~H)

The numerator is the likelihood of the data given that H is true, and the
denominator is the likelihood of the data given that H is false.
To transform prior odds into posterior odds, we multiply by the likelihood
ratio:
Posterior odds = Likelihood ratio * Prior odds
If the data are more likely when H is true, the likelihood ratio is greater than
one, and observing the data increases the odds. If the data are more likely when
H is false, the likelihood ratio is less than one, so observing the data decreases
the odds.
The likelihood ratio is also called the diagnosticity because of its meaning in
medical diagnostic tests.
Many diagnostic tests have significant false positive rates. If we let D repre-
sent a positive test result and H the hypothesis that a patient has the disease in
question, then the true positive rate (also called the sensitivity) of the test is
P(D|H), the probability of a positive result when the patient has the disease. The
false positive rate is P(D|∼H), the probability of a positive result given that
the patient does not have the disease. The diagnosticity is therefore the ratio of
the true positive rate to the false positive rate.
Suppose you wake up one morning with a sore throat. You have heard that,
of all sore throats, 10 percent are bacterial infections and the other 90 percent
 explaining and predicting one-time events 231

are viral. In other words, for every one bacterial infection, there are nine viral
infections, so you figure that your odds of having a bacterial infection—let’s say
strep throat—are 1:9.
You go to the doctor, and she gives you a rapid strep test. She explains that,
among people with strep throat, 81 percent will get a positive test result, and
among those with a virus, 1 percent will get a positive test result. In other words,
the sensitivity or true positive rate is 0.81, and the false positive rate is 0.01. You
calculate that the diagnosticity of the rapid strep test is 81.
The doctor calls you with the news that you tested positive. Now what are your
odds of having strep?
Posterior odds = Diagnosticity * Prior odds = 81 * 1/9 = 9/1
Your odds have increased dramatically from 1:9 to 9:1, because of the high
diagnosticity of the test, and she writes a prescription for antibiotics.
A high diagnosticity is an indication of a good test. Whenever the diagnostic-
ity is greater than 1, we know two things:
• Because diagnosticity is the likelihood ratio P(D|H)/P(D|∼H) we know the
true positive rate P(D|H) is greater than the false positive rate P(D|~H).
• Because posterior odds = Diagnosticity * Prior odds, we know that the pos-
terior odds (after a positive test result) are greater than the prior odds. That
is, a positive test result increases the odds that the patient has the disease.

figure 8.3 way too general practitioner © The New Yorker Collection ,
feb. 14–21 2005, 120. all rights reserved. reprinted with permission.
232 problem solving, decision making, and professional judgment

On the other hand, if the diagnosticity of a test is 1, it does not change the
odds and is not worth performing.
The rarer the disease in question, the higher a test’s diagnosticity must be to
give one confidence in the result. For example, in order to have better-than-even
odds of having SARS, you would have to test positive on a test with very high
diagnosticity, in order to make up for SARS’s very low base odds. Even if your
symptoms are more consistent with SARS, you might still be more likely to have
the flu, just because the flu is a much more common ailment.
For an analysis of the taxi problem using the odds version of Bayes’ Theorem,
see the Appendix, Section 8.7.20

CT scans have become a significant industry even for customers who have no
particular risk of disease. Are they useful? Steven Woloshin and colleagues cite a
report that people whose lung cancer is found early by such scans have a five-year
survival rate of 80 percent, as opposed to 15 percent for the typical lung-cancer
patient whose condition is detected later.20 They are nonetheless skeptical about
whether the scans do more good than harm.
One problem is that CT scans detect tiny tumors, but do not have a very high
diagnosticity, and thus must often be followed up by biopsies, which can be pain-
ful and even dangerous. So for starters, the test can give rise to harm to people
who don’t have, and are not in danger of developing, lung cancer. And even when
cancerous abnormalities are discovered, they may be indolent and never develop
into anything dangerous. But because there’s no way to know, most people who
are diagnosed are treated with surgery, radiation therapy, or chemotherapy—also
painful and often dangerous.
But what of the survival statistics? Here’s how they are calculated: Suppose
that 1000 people were diagnosed with lung cancer five years ago. If 150 are alive
today, the five-year survival is 150/1000, or 15 percent; if 800 are alive today the
five-year survival is 80 percent.
But Woloshin and his colleagues argue that even if CT screening raised the
five-year survival rate from 15 percent to 80 percent, no one might get an extra day
of life. They suggest this thought experiment. Imagine a group of people with
lung cancer who will all die at age 70. If they are diagnosed when they are 67, their
five-year survival rate would be zero percent. But if they were diagnosed when
they were 64, their five-year survival rate would be 100 percent. Early diagnosis
increases the five-year survival statistic but doesn’t postpone death.

20. Steven Woloshin, Lisa Schwartz, and Gilbert Welch, Warned, but Worse Off,
New York Times, Aug. 22, 2005.
 explaining and predicting one-time events 233

The fact that CT scans identify small indolent tumors also inflates the survival
rate. They suggest another thought experiment. Imagine a population where,
after having symptoms such as a persistent cough, 1000 people are diagnosed
with progressive lung cancer. Five years later,150 are alive. Suppose the entire
population gets CT scans, and that 5000 are diagnosed with cancer, but the addi-
tional 4000 actually have indolent forms and are alive five years later. This would
raise the five-year survival rate to 83 percent—because these healthy people
would appear in both parts of the fraction: survivors/people diagnosed with
cancer = (150 + 4000)/(1000 + 4000) = 4150/5000 = 83%. But what has really
changed? Some people have been unnecessarily told they have cancer (and may
have experienced the harms of therapy), and the same number of people (850)
still died.
The phenomenon was borne out by a randomized trial of chest X-ray screen-
ing at the Mayo Clinic, where the five-year survival was higher for those who were
screened, but death rates were slightly higher for those who received the
screening.21

8.5.4 Bayes’ Theorem and Intuition: Ignoring the Base Rate21

When Daniel Kahneman and Amos Tversky gave the taxi problem to subjects
who were not knowledgeable about Bayesian statistics, their average answer for
the probability that the cab was Blue was 80 percent—they focused only on the
eyewitness’s testimony and completely ignored the base rates.22 In the example
of the patient with symptoms slightly more consistent with SARS than with the
flu, this is analogous to the doctor diagnosing him with SARS, ignoring the flu’s
much higher incidence rate.
A different group of subjects was given a slight variant, in which they were
told that, although the two companies are roughly equal in size, 85 percent of
cab accidents in the city involve Green cabs, and 15 percent involve Blue cabs.
Although the answers were highly variable, the average answer for the
probability that the cab involved in the accident was Blue was 55 percent—an

21. The New England Journal of Medicine published the study on which the Woloshin
article is based without disclosing that its principal investigator and her institution
held patents related to CT scanning or that the study was funded by a cigarette manufac-
turer. In January 2009 the journal announced a new disclosure policy. New York
Times, Jan. 9, 2009, available at http://www.nytimes.com/2009/01/09/us/09journal.
html?_r=1&ref=health.
22. Amos Tversky and Daniel Kahneman, Evidential Impact of Base Rates, (1982), in
Heuristics and Biases, supra at 153.
234 problem solving, decision making, and professional judgment

estimate that appears to take some account of the base rate. Kahneman and
Tversky’s explanation, which is supported by other experiments, is that the
base rate in the variant problem provides some causal information: it suggests
that the drivers of the Green cabs are less competent or careful drivers than the
drivers of Blue cabs, and this induces people to attend to the base rates of care-
ful and careless drivers.
Kahneman and Tversky23 gave two groups of undergraduate students some
short descriptions of individuals and asked them to judge whether the individu-
als were likely to be lawyers or engineers. Here is a typical description:
Jack is a 45-year-old man. He is married and has four children. He is gener-
ally conservative, careful, and ambitious. He shows no interest in political
and social issues and spends most of his free time on his many hobbies,
which include home carpentry, sailing, and mathematical puzzles.
One group of students was told that the descriptions came from a sample of
people of whom 70 percent were lawyers and 30 percent engineers. The other
group of students was told that the descriptions came from a sample of people of
whom 30 percent were lawyers and 70 percent engineers. Thus the two groups
were given quite different base rates.
Most of the students judged Jack to be very likely to be an engineer. Most
interestingly, their judgments were unaffected by the different base rates. In
Bayesian terms, the students entirely ignored P(H). Here, again, the represen-
tativeness heuristic is at work: People seem to be making decisions based upon
how representative the description is of a particular profession, rather than by
combining prior and likelihood information in a fashion consistent with
Bayesian inference.
These and other findings suggest that humans do not combine probabilistic
information in a Bayesian fashion. When told:
Suppose . . . that you are given no information whatsoever about an individual
chosen at random from the sample.
respondents correctly referred to the base rates in predicting the person’s
profession. But when given a diagnostically worthless description, such as:
Dick is a 30-year-old man. He is married with no children. A man of high
ability and high motivation, he promises to be quite successful in his field. He
is well liked by his colleagues.

23. Amos Tversky and Daniel Kahneman, Causal Schemas in Judgments Under
Uncertainty, (1973), in Heuristics and Biases, supra at 117.
 explaining and predicting one-time events 235

Legal Problem

We have focused on people’s tendency to neglect base rates. Can you think
of an instruction that judges give jurors as a matter of course in criminal
cases that is intended to counteract their tendency to take account of base
rates?

they neglected the base rates, judging Dick to be equally likely to be a lawyer or an
engineer regardless of the relative frequencies of these two professions in the
samples.

8.6 confusion of the inverse, or the prosecutor’s fallacy

In Section 5.6, we explained that it is easy to confuse P(A|B) with P(B|A), a
mistake sometimes called confusion of the inverse. Recall the difference between
P(pregnant|woman) and P(woman|pregnant).
In People v. Collins (introduced in Section 8.2), in addition to the error of
treating events that were not independent as if they were independent when
combining them, the prosecutor made the error of confusing these conditional
probabilities. Based on the claim that there was only one chance in 12 million
that any couple possessed all of these distinctive characteristics, the prosecutor
argued that there was only one chance in 12 million that the defendants were
innocent, and the jury convicted. The court caught what has since been named
the prosecutor’s fallacy. Assuming that the chances that any couple possessed
all of the distinctive characteristics were 1/12,000,000, “the prosecution’s
figures actually imply a likelihood of over 40 percent that the Collinses could be
‘duplicated’ by at least one other couple who might equally have committed the
San Pedro robbery.”24 The prosecutor’s fallacy was confusing P(not guilty|match)
with P(match|not guilty).
If this example of confusion of the inverse seems itself confusing, consider
the following problem. Approximately 3 percent of the U.S. population has blood
type AB+. The blood at the scene of the murder is AB+, the defendant’s blood is
AB+, and the match of blood type is the only evidence connecting him to the
murder.

24. In William Fairley and Frederick Mosteller, A Conversation About Collins, in

Selected Papers of Frederick Mosteller (New York: Springer New York 1974), William
Fairley and Frederick Mosteller show that the court’s particular result—though not its
fundamental critique of the prosecutor’s reasoning—was in error.
236 problem solving, decision making, and professional judgment

Let us define match as the presence of the accused’s blood type, AB+, at the
scene of the crime. And let us treat the presence or absence of the accused’s
actual blood at the scene of the crime as equivalent to his being guilty or not
guilty. We know that the probability of a match for a randomly selected member
of the population, P(match) = 0.03. Applying the prosecutor’s argument in
Collins to the present case, P(match|not guilty) = 0.03, and therefore the prosecu-
tor would want the jury instructed that there is a 97 percent chance that the
suspect committed the murder.
But we’re interested in P(not guilty|match). Calculating this number requires
knowing the population of people who might have committed the crime. Let’s
assume that the relevant population contained 1000 people, and that there is no
other evidence connecting the suspect to the crime—so that if one other person
in the relevant population has AB+, the suspect cannot be convicted “beyond a
reasonable doubt.” We can use Bayes’ Theorem,

P(D | H ) P(H)
P(H | D ) =
P(D)

to calculate P(guilty|match) or:

P(match | guilty) P(guilty)
P(guilty | match) =
P(match)

• Since there’s sure to be a match if the suspect is guilty, P(match|guilty) = 1.

• P(guilty), or the prior probability of the suspect’s being guilty, is 1/1000 or
0.001.
• The probability of a match for a random member of the population,
P(match), is about 0.03.
Thus, P(guilty | match) = 1 * 0.001/0.03 = 0.033
By the complement rule, P(not guilty|match) = 0.97. So instead of 97 percent
chance of guilt, it is 97 percent chance of innocence.

Forensic Evidence

In addition to the hazard of confusion of the inverse, identifications based on a

match of the defendant’s DNA, fingerprints, etc. with traces found at the crime
scene or on a victim present the simple problem of false positives. That is, the
laboratory tests may report a match when the defendant’s and the trace DNA
 explaining and predicting one-time events 237

are not the same25. By contrast to fingerprint matches, which courts typically
treated as certain, the relative novelty of DNA evidence opened it to a probabilis-
tic understanding, which has begun to affect more traditional forms of forensic
evidence.26

8.6.1 Countering Faulty Intuition by Presenting Frequencies2526

The cognitive scientist Gerd Gigerenzer has suggested that people are less likely
to fall prey to confusion of the inverse and the representativeness heuristic, and
more likely to correctly integrate base rates into decisions, if the data are pre-
sented in terms of natural frequencies rather than probabilities. He posed a
question along these lines to physicians:27
To facilitate early detection of breast cancer, starting at a particular age,
women are encouraged to participate at regular intervals in routine screen-
ing, even if they have no obvious symptoms. Imagine that you conduct such
breast cancer screening using mammography in a particular region of the
country. The following information is available about asymptomatic women
aged 40 to 50 in such region who participate in mammography screening.
One group of physicians was given information in terms of probabilities, in
essentially this form:
The probability that a given woman in this group has breast cancer is 1 per-
cent. If a woman has breast cancer, the probability is 80 percent that she will
have a positive mammogram. If a woman does not have breast cancer, the
probability is 10 percent that she will have a positive mammogram. Imagine
a woman who has a positive mammogram. What is the probability that she
actually has breast cancer?
Responses varied from 1 percent to 90 percent, with a median estimate of
70 percent. Only two of the twenty-four physicians in this group reasoned

25. Jonathan Koehler, quoted in Gigerenzer at 167, says 1 in 100 or 200. Hans Zeisel
and David H. Kaye, Prove it With Figures: Empirical Methods in Law and
Litigation (New York: Springer-Verlag, 1997).
26. See Michael J. Saks and Jonathan J. Koehler, The Coming Paradigm Shift in Forensic
Identification Science, 309 Science 892 (2005).
27. Gerd Gigerenzer, Calculated Risks: How to Know When Numbers Deceive
You (New York: Simon & Schuster, 2002). The particular questions quoted in the text
were put forth by one expert, leading to the development of a broader survey with results
along the same lines.
238 problem solving, decision making, and professional judgment

correctly to reach the correct result of 7.5 percent. Another two were close, but
their reasoning was not correct.28
The second group of physicians was given information in terms of natural
frequencies.
Ten out of every 1,000 women have breast cancer. Of these 10 women with
breast cancer, 8 will have a positive mammogram. Of the remaining 990
women who don’t have breast cancer, some 99 will still have a positive mam-
mogram. Imagine a sample of 100 women who have positive mammogram
screening. How many of these women actually have breast cancer?
Eleven of this group of twenty-four physicians gave the correct answer, and
quite a few of the others were close. Gigerenzer argues that the computation
based on natural frequencies is much simpler than using Bayes’ Theorem—
something that most readers who worked through the problems above would
agree with.
We think that a tree, of the sort used in the taxi problem, provides an easy way
of representing this problem, as shown in Figure 8.4.
positive test
has cancer 0.8
0.01 negative test
0.2
Positive test
does not have cancer 0.1
0.99 Negative test
0.9

figure 8.4 structuring a medical diagnosis problem as a tree.

Since we are only interested in cases where someone tested positive, we focus
on the two branches with positive tests:
has cancer * positive test = 0.01 * 0.8 = 0.008

does not have cancer * positive test = 0.99 * 0.1 = 0.099

Adding the two products—0.008 + 0.099 = 0.107—tells us how often a woman
in this population will test positive. The probability that a woman with a positive
mammogram actually has cancer is the proportion of positives that are true:
0.008/0.107 = 0.075, less than 8 percent.

28. Id. 43. Gigerenzer states that the correct result was 8 percent, but the actual
number, as derived below, is 7.476.
 explaining and predicting one-time events 239

8.7 appendix

8.7.1 Odds Version of Bayes’ Theorem Applied to the Taxi Problem

The hypothesis is that the taxi was Blue, and the data set is that the witness says
the taxi was Blue. The prior odds are P(B):P(G) = 0.15:0.85 = 3:17. So before
hearing the testimony, based on the base rate alone, the cab is much more likely
to be Green. The likelihood ratio is P(SB|B)/P(SB|G) = 0.8/0.2 = 4. Therefore,
the testimony increases the prior odds fourfold, giving posterior odds of
(4 * 3):17 = 12:17.
The posterior odds being less than one is equivalent to the posterior probabil-
ity being less than 0.5. Both tell us to guess that the cab was Green. Odds of
12 to 17 mean that it is likely to occur 12 out of every 29 times. Therefore, the
posterior probability P(B|SB), is
12
= 0.41
29
the same answer we obtained using each of the other two formulations of Bayes’
Theorem.
If the likelihood of the data were the same whether or not H were true, the
likelihood ratio would equal 1. When this is the case, the data tell us nothing, and
the posterior odds equal the prior odds. In our example, this would happen if the
witness were incorrect 50 percent of the time. She would have no credibility, so
her testimony would not change our guess of the taxi color at all. The posterior
odds would equal the prior odds of 3:17. (See Section 5.1.2a).
This page intentionally left blank
9. biases in perception and memory

9.1 introduction: stages of information processing

In the introduction to Part 2, we described the application of the lens model to a

probabilistic environment. The preceding chapters on statistics were largely con-
cerned with what inferences can appropriately be drawn from multiple fallible
indicators—in particular, what inferences can be appropriately made about an
entire population when the indicators are only a sample of the population. In
those chapters we generally proceed from the assumption that the empiricist
was accurate in perceiving and coding the sample data, and we introduced the
statistics that best allows one to make inferences. Chapter 9 focuses on errors
in perceiving and interpreting the events that provide the basis for such
inferences.
Cognitive psychologists have likened a person making a judgment to an
“information processor” that proceeds through a number of stages from the
availability of information to the generation of a response to that information.
The functional components of that information processor can be represented as
shown in Figure 9.1.

Information Attention Encoding Storage Retrieval Processing Response

figure 9.1 model of information processing.

Let us consider those components in turn. Every moment of our lives, we are
bombarded with vast amounts of information, of which we attend to only a small
fraction. We encode the information, structuring, evaluating, and interpreting it
and transforming into some sort of mental representation. You might think of
perception (not on the chart) as overlapping attention and encoding. We store
information in memory and, on occasion, retrieve it from memory (i.e., become
aware of it) and process it with respect to particular objectives. Our response to
processing may be a factual judgment or inference, an evaluative judgment or
opinion, a choice among alternatives or decision, or a solution to a problem.1
This and the following chapter examine biases and distortions that can affect
these stages of information processing. Although some of the psychological

1. See Verlin B. Hinsz, R. Scott Tindale, and David A Vollrath, The Emerging
Conceptualization of Groups as Information Processors, 43 Psychological Bulletin 121
(1997).
242 problem solving, decision making, and professional judgment

phenomena considered cut across various stages, this chapter, on biases in

perception and memory, centers around the first five stages, and Chapter 10, on
biases in judgment, centers around the last two.

9.2 biases in acquisition, retention, and retrieval

We met at nine That carriage ride

We met at eight. You walked me home.
I was on time You lost a glove
No, you were late. I lost a comb.
Ah yes! I remember it well. Ah yes! I remember it well.
We dined with friends That brilliant sky
We dined alone. We had some rain.
A tenor sang Those Russian songs
A baritone. From sunny Spain
Ah yes! I remember it well. Ah yes! I remember it well.
****
—I Remember It Well (Duet from “Gigi”)
Lyrics by ALAN JAY LERNER Music by FREDERICK LOEWE
© 1957, 1958 (Copyrights Renewed) CHAPPELL & CO., INC.
All Rights Reserved
Used by Permission of ALFRED PUBLISHING CO., INC.

A large amount of any lawyer’s time is focused on past events. It is the under-
standing and analysis of past events that give rise to legal liability—whether civil,
administrative, or criminal—and that provide the basis for the predictions
necessary for policy making and planning. By definition, understanding the past
calls upon memory. In many contexts, however, memories tend to be highly
fallible.
Our memory does not store complete representations of what was perceived,
but only fragments of our interpretations of the relevant facts or events.
Recollection requires reconstructing one’s interpretation, often using (more
or less) logical inferences to fill in missing details.2 In Eyewitness Testimony,
Elizabeth Loftus divides memory processes into three stages: acquisition, in
which the information is entered; retention, the period between acquisition and
the demand to “remember”; and retrieval, where one recollects the information.

2. Elizabeth F. Loftus, Memory: Surprising New Insights into How We

Remember and Why We Forget (Reading, MA: Addison-Wesley, 1980).
 biases in perception and memory 243

Though she focuses on visual memories, her observations apply to memories of

all sorts:3
Early on, in the acquisition stage, the observer must decide which aspects of
the visual stimulus he should attend to. Our visual environment typically con-
tains a vast amount of information, and the proportion of information that is
actually perceived is very small. . .
Once the information associated with an event has been encoded or stored
in memory, some of it may remain there unchanged while some may not.
Many things can happen to witnesses during this crucial retention phase. The
witness may engage in conversations about the event, or overhear conversa-
tions, or read a newspaper story—all of these can bring about powerful and
unexpected changes in the witness’s memory.
Finally, at any time after an event a witness may be asked questions
about it. At this point the witness must recreate from long-term memory
that portion of the event needed to answer a specific question. This recre-
ation may be based both on information acquired during the original
experience and on information acquired subsequently. In other words,
both the acquisition and the retention stages are crucial to what happens
during the retrieval. The answer the person gives is based on this
recreation.
. . . Events at any one or several of the stages can be the cause of . . .
retrieval failure. The information may simply not have been perceived in the
first place—a failure at the acquisition stage. The information might have
been accurately perceived, but then is forgotten or interfered with during the
retention stage. And finally, information may have been accurately perceived
in the first place but may have become inaccessible [or distorted] during
questioning—a failure at the retrieval stage.

9.3 biases in the acquisition of information

Before turning to more complex issues involving the acquisition of information,

we should note that some problems simply arise from perceptual inadequacies
that are fairly universal. For example, we tend to be quite inaccurate in judging
time, speed, and distance,4 and we systematically overestimate the duration of
events. More generally, Dolly Chugh and Max Bazerman describe what they call

3. Elizabeth F. Loftus, Eyewitness Testimony 21 (Cambridge, MA: Harvard

University Press, 1979) [hereafter Loftus].
4. Participants in one experiment were told in advance that they would be asked to
judge the speed of a car. The car was moving at 12 miles per hour, and their estimates
ranged from 10 to 50 miles per hour. Loftus 29.
244 problem solving, decision making, and professional judgment

bounded awareness—the phenomenon where “individuals fail to see, seek, use, or

share highly relevant, easily accessible, and readily perceivable information
during the decision-making process.”5 Because our cognitive power is limited,
we can focus only on a small part of everything that goes on around us—often
only on one thing at a time.

Mr. Hill was so engrossed in the call that he ran a red light and didn’t notice Linda
Doyle’s small sport utility vehicle until the last second. He hit her going 45 miles
per hour. She was pronounced dead shortly after. Later, a policeman asked
Mr. Hill what color the light had been. “I never saw it,” he answered.6

Much of our daily experience involves multitasking, but we aren’t as good at

multitasking as we would like to believe. Most of the time, we just don’t notice
things that are not in our mental field of vision. Sometimes this can be embar-
rassing—when we’ve lost the thread of a discussion by doing e-mail tasks at
meetings or browsing the Web during class; sometimes it’s dangerous—drivers
cannot give full attention to the road when talking even on a hands-free phone.
Eyewitness’s description of crimes are notoriously unreliable, especially when
their attention is riveted to the stress-inducing event. Elizabeth Loftus uses the
term “weapon focus” to describe an assault victim’s attention to the brandished
weapon to the exclusion of other details, including features of the assailant.
Chugh and Bazerman characterize the misalignment of available information of
which one is unaware with the information needed for a decision as a focusing
failure.

9.4 the biasing effect of schemas, expectations, theories,

and interests 6

Section 1.5.2 introduced the idea that all of our perceptions are mediated by sche-
mas or expectations: when we perceive a stimulus from our environment, our
first task is to fit that information into some existing knowledge structure repre-
sented in memory. This is essential and inevitable—and a hazard for the careful
empiricist. In a classic experiment, Jerome Bruner and Leo Postman briefly

5. Dolly Chugh and Max H. Bazerman, Bounded Awareness: What You Fail to See Can
Hurt You, 6 Mind & Society 1–18 (2007).
6. Matt Richtel, Drivers and Legislators Dismiss Cellphone Risks, New York Times,
July 19, 2009, http://www.nytimes.com/2009/07/19/technology/19distracted.html?hp.
 biases in perception and memory 245

showed participants five playing cards, one of which was a three of hearts colored
black.7 A vast majority of the participants engaged in “perceptual denial,” confi-
dently recalling that the card was a normal three of hearts. The phenomenon of
illusory correlation described in Section 5.7 exemplifies the same tendency.
Elizabeth Loftus reports on an accident where hunters killed a companion: “One
of the men . . . saw something moving and said to his friend, ‘That’s a deer, isn’t it?’
The friend replied that he thought so too, and the first man shot at the deer. The
deer pitched forward and cried out—a sound which seemed like the cry of a
wounded deer. The hunters fired more shots to bring it down.” Loftus explains:8
The hunters who eagerly scanned the landscape for a deer perceived the
moving object as a deer. They expected to hear the cry of a deer and they heard
their friend’s cry that way . . . Yet a policeman testified that when he later saw
a man under the same conditions he perceived the object as a man . . .[T]he
policeman knew he was supposed to be looking at a man; thus, he perceived
the object he saw as a man.
Consider these more complex examples:

9.4.1 They Saw a Game

A 1951 football game between Dartmouth and Princeton was unusually rough,
with injuries sustained by members of both teams. Psychologists Albert Hastorf
and Hadley Cantril asked students from each school—some of whom had seen
the game and some a film of it—to fill out a questionnaire noting any infractions
of the rules, and characterize them as “mild” or flagrant.”9
Nearly all Princeton students judged the game as “rough and dirty”—not one
of them thought it “clean and fair.” And almost nine-tenths of them thought
the other side started the rough play. . . . When Princeton students looked at
the movies of the game, they saw the Dartmouth team make over twice as
many infractions as their own team made. And they saw the Dartmouth team
make over twice as many infractions as were seen by Dartmouth students.
When Princeton students judged these infractions as “flagrant” or “mild,” the
ratio was about two “flagrant” to one “mild” on the Dartmouth team, and
about one “flagrant” to three “mild” on the Princeton team.
As for the Dartmouth students, while the plurality of answers fell in the
“rough and dirty” category, over one-tenth thought the game was “clean and

7. Jerome S. Bruner and Leo Postman, On the Perception of Incongruity: A Paradigm, 18

Journal of Personality 206 (1949).
8. Elizabeth F. Loftus and Katherine Ketcham, Witness from the Defense:
The Accused, the Eyewitness, and the Expert Who Puts Memory on Trial
(New York: St. Martin’s Press, 1991).
9. Albert H. Hastorf and Hadley Cantril, They Saw a Game: A Case Study, 49 Journal
of Abnormal Psychology 129 (1954).
246 problem solving, decision making, and professional judgment

fair” and over a third introduced their own category of “rough and fair” to
describe the action. Although a third of the Dartmouth students felt that
Dartmouth was to blame for starting the rough play, the majority of Dartmouth
students thought both sides were to blame . . .
When Dartmouth students looked at the movie of the game, they saw both
teams make about the same number of infractions. And they saw their own
team make only half the number of infractions the Princeton students saw
them make. The ratio of “flagrant” to “mild” infractions was about one to one
when Dartmouth students judged the Dartmouth team, and about one “fla-
grant” to two “mild” when Dartmouth students judged infractions made by
the Princeton team.
Granted that the students’ perceptions of fault were biased in favor of their
own schools’ teams, what explains the difference in perceptions? Hastorf and
Cantril answer:
It seems clear that the “game” actually was many different games and that
each version of the events that transpired was just as “real” to a particular
person as other versions were to other people . . .
. . . An “occurrence” on the football field or in any other social situation
does not become an experiential “event” unless and until some significance is
given to it. And a happening generally has significance only if it reactivates
learned significances already registered in what we have called a person’s
assumptive form-world.
At the very least, people interpret the same ambiguous information in differ-
ent ways: I see the tackle as gratuitously violent while you see it as hard-hitting
but fair. At the worst, from all the events going on in the environment, we some-
times notice those that are significant from our egocentric positions, and fail to
notice those that don’t.

9.4.2 The Hostile Media Phenomenon

Essentially the same phenomenon surfaces in a quite different context: people
who differ in their perceptions and recollections and interpretations of conten-
tious events inevitably have very different views of what constitutes “fair” news
coverage of those events. In 1985, Robert Vallone, Lee Ross, and Mark Lepper
conducted an experiment in which participants who had previously identified
themselves as pro-Arab, pro-Israeli, or neutral viewed extensive samples of
media coverage of the massacre of Palestinians in Lebanese refugee camps by
allies of Israel.10

10. Robert P. Vallone, Lee Ross, and Mark R. Lepper, The Hostile Media Phenomenon:
Biased Perception and Perceptions of Media Bias in Coverage of the Beirut Massacre, 49
Journal of Personal and Social Psychology 577 (1985).
 biases in perception and memory 247

The pro-Arab participants saw the media as biased toward Israel; the pro-
Israeli participants saw the news programs as biased against Israel. The pro-
Arab participants thought that the programs applied lower standards to Israel
than to other countries, that they did not focus enough on Israel’s role in the
massacre, and that the editors made a positive case for Israel and were biased in
its favor. Pro-Israeli participants thought that the programs applied a higher
standard to Israel, focused too much on Israel’s role, made a negative case
against Israel, and were biased against Israel. Furthermore, participants who
deemed themselves more knowledgeable or more emotionally involved had
stronger perceptions of bias.11 Neutrals were about midway between the parti-
sans. (Viewers who thought themselves knowledgeable thought the main source
of bias was what was omitted, particularly that the media did not provide the full
context.)
Just as Hastorf and Cantril believed that Dartmouth and Princeton students
had seen different football games, Vallone, Ross, and Lepper concluded that
“the pro-Arab and pro-Israeli participants ‘saw’ different news programs—that
is, they disagreed about the very nature of the stimulus they had viewed.”
For example, the groups reported that a higher percentage of the references to
Israel were, respectively, favorable or unfavorable. And each believed that the
programs would lead an undecided or ambivalent viewer to favor the opposing
position. The authors offer this analysis of the phenomenon:
Our results highlight two mechanisms—one apparently evaluative or cogni-
tive, the other apparently more perceptual in nature—that combine to pro-
duce the partisans’ conviction that they have been treated unfairly. According
to the first mechanism, in which opposing partisans believe, respectively, that
the truth is largely “black” or largely “white,” each complain about the fairness
and objectivity of mediated [i.e., media] accounts that suggest that the truth
might be at some particular hue of gray. According to the second mechanism,
opposing partisans further disagree about the color of the account itself: One
side reports it to be largely white (instead of the blackish hue that the other
side thinks it should be), the other side reports it to be largely black (instead
of the whitish hue that the first side thinks it should be), and both sides believe
the discrepancy between the mediated account and the unmediated truth to
be the intended result of hostile bias on the part of those responsible.
We note that our results do not permit us to speak authoritatively about
either the source or the depth of the perceptual bias we have claimed to
document; nor, obviously, do they shield us from the age-old difficulties of

11. While knowledgeable neutral participants did not perceive a bias one way or the
other, unknowledgeable neutrals more closely resembled the pro-Israeli than pro-Arab
groups. The authors note that American public opinion tends to be pro-Israeli and that the
neutrals may be more pro-Israeli than they realize.
248 problem solving, decision making, and professional judgment

ruling out all cognitive interpretations for an ostensibly perceptual phenom-

enon. Do partisans pay more attention when their side is being attacked? Do
they remember such information more, perhaps because they are frustrated
and annoyed by the absence of any sufficiently forceful rebuttals to such
attacks? The exact mechanism remains unclear, but we believe that it is not
simply a matter of differing standards or criteria in labeling particular facts,
arguments, or images as pro-Israeli or anti-Israeli. Perhaps our most impor-
tant, and interesting, finding in this regard is the tendency for both groups to
assert that neutral viewers will turn against their side when they view the
media coverage. This finding is further evidence that the specific content and
overall “hue” of the report is indeed perceived differently by the partisans,
even when they would wish the truth to be otherwise, for partisans surely
would prefer to believe, and perhaps even expect that non-partisans would
assimilate mixed information in a manner congruent with the partisans’
view of the truth.
The authors suggest that their analysis may apply to perceptions of other
types of mediation, including spouses’ perceptions of family counselors, or labor
and management’s perceptions of government arbitrators.

9.4.3 Why Cases Don’t Settle When They Should12

In many civil cases, the parties pay inordinately high litigation costs before set-
tling or they do not settle at all. Some scholars have explained the phenomenon
in terms of disputants’ uncertainty about the judge or jury. An alternative expla-
nation, put forward by Linda Babcock and her colleagues is that litigants’ predic-
tions of judicial decisions are systematically biased in a self-serving manner.
To test this hypothesis, Babcock et al. gave law and business-school students
detailed testimony in a case involving a claim for damages in a motorcycle-
automobile accident. The participants were informed that the same materials
had been given to a judge who had decided how much, if anything, to award to
the plaintiff. They were asked (1) what they thought was a fair settlement from
the viewpoint of a neutral third party and (2) their best guess of the amount the
judge would award. They were also instructed to negotiate a settlement: The
defendant was given $10 cash (with each $1 worth $10,000 in the real world).
If they could not reach a settlement in a half hour, the judge’s verdict would be
imposed on them.
The major experimental manipulation was that the participants in one group
were assigned roles as plaintiffs or defendants before reading the testimony,

12. Linda Babcock, George Loewenstein, Samuel Issacharoff, and Colin Camerer,
Biased Judgments of Fairness in Bargaining, 85 American Economic Review 1337 (1995);
Linda Babcock and George Loewenstein, Explaining Bargaining Impasse: The Role of Self-
Serving Biases, 11 Journal of Economic Perspectives 109 (1997). Abstract available at
SSRN: http://ssrn.com/abstract=11367.
 biases in perception and memory 249

answering the questions, and negotiating, while those in the second group first
read the testimony and answered the questions and were then assigned roles
before they negotiated.
The assessment of what would be a fair judgment by plaintiffs and defen-
dants who had been assigned roles before reading the testimony diverged by
$19,756 on average; their predictions of the judicial outcome diverged by $18,555.
Moreover, 28 percent of the parties in this group failed to settle. By contrast,
plaintiffs and defendants in the second group did not diverge significantly in
their assessments of either of fairness or the judicial outcome, and only 6 per-
cent failed to settle. Apparently, knowing one’s role in advance causes one to
encode the facts in a way that is self-serving and that leads to undue optimism
about the assessments likely to be made by third parties.
Not only students, but experienced litigators and negotiators are prone to this
self-serving bias. Whereas purely economic analyses attribute the failure to settle
to strategic behavior, the authors suggest: “Perhaps disputants are not trying to
maximize their expected outcome, but only trying to achieve a fair outcome.
However, what each side views as fair tends to be biased by self-interest, which
reduces the prospects for settlement.”
Merely informing disputants about the prevalence of self-serving bias had
no effect on their bias. Is it possible to counter the bias? Babcock et al. did a fol-
low-up experiment in which the participants, after being assigned roles and
reading the testimony, were instructed:
Disputants don’t always think carefully about the weaknesses in their own
case: they are therefore surprised when the judge’s ruling is worse than their
expectations. For plaintiffs, this means that the judge’s award is often less
than their expectations. For defendants, this means that the judge’s award is
often greater than their expectations. Therefore, please think carefully about
the weaknesses in your case. In the space below, please list the weaknesses in
your own case.
Being asked to assess the weaknesses of one’s own case had a significant
effect. When given the debiasing instruction, plaintiffs’ and defendants’ predic-
tions of the judge’s award differed by only $5000, and the litigants reached settle-
ment much more quickly. The authors suggest that a debiasing procedure could
be built into court-mandated pretrial settlement conferences.

9.5 distortions in retention and retrieval

It isn’t so astonishing, the number of things that I can remember, as the number of
things I can remember that aren’t so.
—Mark Twain, A Biography13

13. Http://www.twainquotes.com/Memory.html.
250 problem solving, decision making, and professional judgment

For any reader who doesn’t recall the phenomenon of forgetting, consider
this a reminder. Of particular interest in this section are events that distort the
content of what is remembered. Distorted recollections are often exacerbated by
people’s overconfidence in the correctness of their memories.
It is useful to treat retention and retrieval together, because intermediate
retrievals are among the major sources of distortion of ultimate retrievals, includ-
ing those that have legal consequences. A witness’s ultimate retrieval may take
place when she testifies at trial, but her recollections may be distorted by ques-
tions posed earlier by investigators or in a deposition, or by chatting with friends
about the event.
One group of biases results from the phenomenon of memory binding, in
which different components of an event, or different events, are bound together.14
For example, a Ryder employee who observed Timothy McVeigh rent a van two
days before the 1995 bombing of the Federal Building in Oklahoma City recalled
there being two men. One was tall and fair, and fit McVeigh’s description; the
other was short, stocky, dark-haired, and had a tattoo. It turned out that a day
after McVeigh rented the van (alone), two men, who matched the descriptions of
McVeigh and his supposed companion, came in to rent a van. As Daniel Schacter
explains: “People recall correctly a fact they learned earlier, or recognize accu-
rately a person or object they have seen before, but misattribute the source of
their knowledge.”15
More broadly, experiences after an event can distort one’s recollection of the
event. For example, people who read an inaccurate newspaper description of an
event they personally witnessed are prone to incorporate the inaccuracies into
their recollections.16 Leading or misleading questions can also distort recollec-
tion. Elizabeth Loftus and colleagues showed participants photos of a car-
pedestrian accident involving a red Datsun. One group saw the car at a “stop”
sign; the other saw the car at a “yield” sign. Within each group, half the partici-
pants were asked “Did another car pass the red Datsun while it was stopped at
the ‘stop’ sign?” and half were asked “Did another car pass the red Datsun while
it was stopped at the ‘yield’ sign?” Finally, after performing some unrelated
intervening tasks designed to put temporal distance between the intermediate
and ultimate retrieval, the participants were shown slides depicting the red
Datsun either at a “stop” or “yield” sign and asked which comported with the
original photo. Of participants who had originally seen a “stop” sign and had
been asked whether the car was stopped at a “stop” sign, 75 percent responded
correctly. Of participants who had been asked the misleading intermediate
question—“Did another car pass the red Datsun while it was stopped at the

14. Daniel L. Schacter, The Seven Sins of Memory: How the Mind Forgets and
Remembers 94 (New York: Houghton Mifflin, 2001) [hereafter Schacter].
15. Schacter 93.
16. Loftus 55.
 biases in perception and memory 251

‘yield’ sign.”—only 41 percent (less than chance) responded correctly.17 Based on

this and other experiments, Loftus concludes that “casually mentioning a nonex-
istent object during the course of questioning can increase the likelihood that a
person will later report having seen that nonexistent object.”
In another study, Loftus and her colleagues showed participants a film of a
traffic accident, and then asked either: “About how fast were the cars going when
they smashed into each other?” or: “About how fast were the cars going when they
hit each other?” The former question elicited a higher estimate. A week later,
without having viewed the film again, the participants were asked, “Did you see
any broken glass?” In fact, the film did not show any broken glass. However,
about twice as many participants who had previously been asked the smashed
question recalled seeing broken glass than those who had been asked the hit
question.
Leading questions (which invite a particular answer) can distort memory.
“Did you see the broken headlight?” elicits more inaccurate recollections that a
headlight was in fact broken than “Did you see a broken headlight?” About a year
after an El Al cargo plane crashed on take-off into an Amsterdam apartment,
Dutch psychologists conducted a survey, asking “Did you see the television film
of the moment the plane hit the building?” Fifty-five percent of the respondents
said that they had seen the film and recalled details about the plane’s speed and
angle and whether it was on fire. In fact, there was no television coverage of the
crash. Also, consistent with the anchoring and adjustment bias (described in
Section 10.1), witnesses have different recollections when asked how tall, or how
short, the suspect was.18 Loftus concludes:
Any time after a witness experiences a complex event, he may be exposed to
new information about that event. The new information may come in the
form of questions—a powerful way to introduce it—or in the form of a con-
versation, a newspaper story, and so on. The implication of these results for
courtroom examinations, police interrogations, and accident investigations
is fairly obvious: interrogators should do whatever is possible to avoid the
introduction of “external” information into the witness’s memory.
Indeed, it is not only external suggestions that may distort one’s recollection.
People who are asked to guess about the details of an event they witnessed are
likely to turn their guess into an erroneous answer when queried later. Indeed,
one’s own intervening thoughts and remarks can serve as powerful anchors in
subsequent recollections.19
Another source of distortion comes from retrieving a past event through
familiarity rather than by recollection. Schacter writes: “Recollection involves

17. Id. 58.

18. Id. 95 ff.; Schacter 112, 115.
19. Loftus 83 ff.
252 problem solving, decision making, and professional judgment

calling to mind specific details of past experiences . . . Familiarity entails a more

primitive sense of knowing that something has happened previously, without
dredging up particular details . . . A strong sense of general familiarity, together
with the absence of specific recollections, adds up to a lethal recipe for misattri-
bution.” Schacter reports on a study of police lineup procedures that invite such
misattribution:20
In standard lineup procedures, witnesses are shown a number of suspects;
after seeing them all, they attempt to identify the culprit . . . [U]nder such
conditions, witnesses tend to rely on relative judgments; they choose the
person who, relative to the others in the lineup, looks most like the suspect.
The problem is that even when the [culprit] is not in the lineup, witnesses will
still tend to choose the person who looks most like him. Witnesses rely on
general similarities between a face in the lineup and the actual culprit, even
when they lack specific recollections.
How can one guard against such distortions? Schacter proposes some spe-
cific techniques for interviewing witnesses.21 With respect to our own memories,
he suggests cultivating a distinctiveness heuristic, under which we demand recol-
lections of distinctive details of an experience before we are willing to say we
remember it.22 This would help prevent adding imagined facts to a recalled
experience, which often happens when we rely only on familiarity.

9.6 the availability heuristic — and a detour into

metacognition

The heuristics and biases agenda, begun in the 1970s by the psychologists Amos
Tversky and Daniel Kahneman, continues to inform much psychological research
today, especially in social psychology and judgment and decision making (JDM).
The heuristics identified by Tversky and Kahneman are based on processes that
operate unconsciously and automatically in System 1, and that tend to be insuf-
ficiently corrected by the conscious analytic processes of System 2.23 We discussed
the representativeness heuristic in detail in Section 8.3. Here we consider the
availability heuristic. While representativeness involves categorizing, availability
involves noticing.
Imagine that you have been considering buying a new car for some time. After
poring through Consumer Reports and some auto magazines, you decide that the
Volvo sedan is just what you want—comfortable, safe, a good maintenance

20. Schacter 44, 97.

21. Id. 119.
22. Id. 103.
23. See on Section 1.6.1.
 biases in perception and memory 253

record; all in all, a great buy. The evening before you plan to purchase the car, you
are at a dinner party where a guest describes his terrible experience with the
Volvo he bought last year—a real lemon, which has spent more time in the shop
than on the road.24 How much will this one anecdote influence your decision
compared to all the research you have done?
What Kahneman and Tversky call the availability heuristic leads people to
make judgments of probability and causation based on how readily events come
to mind because they are vivid, imaginable, or otherwise perceptually salient.
Thus, in attributing the cause of the problems at Terra Nueva, residents will
more likely be influenced by vivid stories of tenants’ illnesses and media cover-
age of “foam insulation syndrome” elsewhere in the country than in epidemio-
logical evidence.
In one experiment, two groups of participants were given samples of the
heights of a number of individuals and asked to estimate the proportion of people
over 6’ tall. Twenty percent of the individuals in both samples were over 6’ tall,
but one group had several very tall outliers. The participants estimated that this
sample had more individuals over 6’ tall. By the same token, participants esti-
mated that a sample of people that included some criminals contained more
criminals when some of them had committed heinous offenses than a similar
sample with the same number of criminals who had only committed minor
offenses.25 What impressions do you imagine various media audiences have of
the number of people with extreme versus moderate views on issues such as
abortion, gun control, and gay marriage?
Consider the following problem:
In four pages of a novel (about 2000 words), how many words would you
expect to find that have the form
_ _ _ _ _ n _ ? (five letters, then n, then another letter)
_ _ _ _ i n g ? (four letters, then ing)
What were your estimates? In between-subjects experiments, people usually
say that they expect more words of the second kind than of the first. But, of
course, this cannot be true, since ing is a subset of _n_. By the same token,
people estimate that there are more “dairy products” than “white foods” in a
supermarket.
These are logical errors—akin to the error made in the example of Linda the
(feminist) bank teller in Section 8.3. Most people understand that there cannot
be more instances of a subset than of the set of which it is a part. The errors arise
from people’s tendency to estimate the frequency or probability of an event by
how readily examples come to mind. (Consider how people would respond to the

24. Richard E. Nisbett and Lee Ross, Human Inference: Strategies and
Shortcomings in Social Judgement. (Englewood Cliffs, NJ: Prentice Hall, 1980).
25. Id. 74–75.
254 problem solving, decision making, and professional judgment

question: “How many white foods are there in a supermarket?—and by the way
don’t forget dairy products.”) It is easy to see why using availability as a proxy for
frequency can be a useful strategy—if events happen more often, we tend to
recall instances of such events more easily. But events can come to our attention
and/or be retrieved more readily for many reasons other than the frequency of
their occurrence.

9.6.1 How the Availability Heuristic Works; Metacognition

The availability heuristic can work through two different processes, reflecting
the different ways that you could make estimates of these sorts. In estimating
whether words that have n as the second-to-last letter are more or less frequent
than words that end in ing, you might rely on the recalled content, i.e., on the
actual number of words with each ending that come to mind. Or you might rely
on the ease or difficulty of recalling words with each of the endings.
People often use the latter process. In one experiment, participants were
asked to recall either six or twelve instances in which they had behaved assertively
and then assess their own assertiveness. Those who had been asked to respond
to six instances found the task relatively easy, and tended to rate themselves as
being assertive. Those who had been asked to recall twelve instances managed to
do so, but found the task relatively difficult, and therefore their rating of their
self-assertiveness decreased even though they had twice as many examples
before them.26
Analogously, participants who were given quite specific symptoms—e.g.,
muscle aches, severe headaches—of a made-up disease, believed that they were
more likely to contract the disease than those given less concrete symptoms—
e.g., a general sense of disorientation, a malfunctioning nervous system.27 The
specific symptoms were easier to imagine than the vague ones.28
These studies illustrate the phenomenon of metacognition—that is, cognitions
about one’s cognitive experience. While most theories of human judgment focus

26. Norbert Schwarz and Leigh Ann Vaughn, The Availability Heuristic Revisited: Ease
of Recall and Content of Recall as Distinct Sources of Information, in Thomas Gilovich,
Dale Griffin, and Daniel Kahneman, Heuristics and Biases, The Psychology of
Intuitive Judgment 103 (New York: Cambridge University Press, 2002).
27. Steven J. Sherman, Robert B. Cialdini, Donna F. Schwartzman, and Kim D.
Reynolds, Imagining Can Heighten or Lower the Perceived Likelihood of Contracting a Disease:
The Mediating Effect of Ease of Imagery, in Thomas Gilovich, Dale Griffin, and Daniel
Kahneman, Heuristics and Biases, The Psychology of Intuitive Judgment 98
(2002).
28. The authors of the study suggest that this has implications for preventive health,
such as persuading people to have immunization shots or to take measures against hyper-
tension: “The health profession might increase . . . compliance by making vivid presenta-
tions of the medical problems and by describing symptoms and consequences . . . in
easy-to-imagine terms.” Note, however, that unless this actually leads to changes in health
behavior, it may just increase people’s anxiety.
 biases in perception and memory 255

on how people evaluate the content of the information they are exposed to, theories
about metacognition hinge on the fact that people also make judgments by evalu-
ating features of the experience of thinking. In particular, experiencing the process-
ing of information as easy or difficult can significantly influence our judgments
of the content of that information. Of course people’s prior assumptions or expec-
tations about the ease or difficulty of a task are also relevant. For example, when
the people who listed twelve examples of assertive behavior were made aware of
the objective fact that listing twelve examples was not an easy task, they were less
likely to attribute their experienced difficulty to lack of assertiveness.29

9.6.2 Other Phenomena Resulting from Evaluation of Metacognitive Experiences

The ease with which new or recalled information can be processed is often
referred to as processing fluency. In addition to explaining some aspects of the
availability heuristic, processing fluency, and metacognition generally, have
been linked to judgments that extend beyond matters of probability. We take a
brief detour to explore these phenomena.
9.6.2.a Fluency as a Source of Judgments of Truth and Liking Familiar infor-
mation tends to be easier to process than novel information. This can lead people
to infer that easy-to-process information was already seen before and then to
conclude that “if it seems they heard it before, there’s probably something to
it.”30 (Along similar lines, people are more likely to endorse statements whose
objective truth is difficult to ascertain when the statements are presented in an
easy-to-read format than rather than one that is difficult to read.31)
Ease of processing also affects people’s liking for things they encounter.
Psychologist Robert Zajonc coined the term mere exposure effect, which refers to
the robust finding that people like things better simply because they have already
seen them before. For example, people liked meaningless symbols better after
they had been exposed to them briefly. (The use of product placements in movies
assumes the effectiveness of mere exposure—though it isn’t clear whether the
goals is to increase liking or merely to make the product in question the mind-
less default option.) People need not be aware of the prior exposure: even sub-
liminal exposure, too brief to be consciously noticeable, sufficed to enhance
people’s later liking of symbols and objects to which they had been exposed.32
Indeed, subliminal exposures are protected from metacognitive correction.

29. Norbert Schwarz, Metacognitive Experiences in Consumer Judgment and Decision

Making, 14 Journal of Consumer Psychology 332–348 (2004).
30. Id. at 340.
31. Rolf Reber, and Norbert Schwarz, Effects of Perceptual Fluency on Judgments of
Truth, 8 Consciousness and Cognition 338–342 (1999).
32. Robert B. Zajonc, Feeling and Thinking: Preferences Need No Inferences, 3 American
Psychologist 151–75 (1980); Sheila T. Murphy and Robert B. Zajonc, Affect, Cognition,
and Awareness: Affective Priming with Optimal and Suboptimal Stimulus Exposures, 64
Journal of Personality and Social Psychology 723–39 (1993).
256 problem solving, decision making, and professional judgment

The mere exposure effect can have serious public policy consequences. Since
we are constantly bombarded with facts, figures, and information, it is crucial for
consumers and citizens to discern between true and false assertions. While the
mere exposure effect robustly suggests that information that is frequently
repeated tends to stick, experiments conducted by Norbert Schwarz and his col-
leagues indicate that the ability to distinguish between the truth and falsehood of
that repeated information fades with time.33 Their study also indicates that
memory of context—which we use to retroactively determine whether a piece of
information is true or false—declines when one is engaged in cognitively
demanding tasks, while familiarity is hardly affected by such tasks.
Not only repeated false claims about an individual, product, or policy, but
efforts to debunk them, can contribute to their apparent truthfulness by height-
ening their availability. For a vivid example of this, consider Richard Nixon’s
statement, “I am not a crook,” in rebutting claims that he was involved in the
Watergate break-in.
9.6.2.b Using Metacognitive Monitoring to Repair Judgment Biases People
sometimes use metacognition to try to correct for what they perceive as biases.
For example, how common is the name Kennedy? When people are asked to
make such estimates, they might rely on the availability heuristic based on how
easily the name comes to mind. However, people’s awareness that the name of
a prominent political family appears so often in the media may lead to self-
correction of the heuristic. In a recent experiment, when asked to estimate the
frequency of names which happened to be the names of prominent celebrities
(e.g., politicians and musicians), participants estimated that the names were
much less frequent than they actually were. In other words, they overcorrected
for the bias caused by the availability heuristic.34
Metacognitive correction of perceived biases can have especially important
implications in the domain of justice. Research done by Duane Wegener and
Richard Petty35 suggests that people can use metacognition to identify potentially
biasing factors and attempt to remove bias from their perceptions and judg-
ments. Wegener and Petty posited that when trying to prevent bias or debias
initial perceptions (e.g., when on jury duty), people consult their own metacogni-
tive beliefs about how various biasing factors might influence their perceptions.

33. E.g., Ian Skurnik, Carolyn Yoon, Denise C. Park, and Norbert Schwarz. How
Warnings About False Claims Become Recommendations. 31, Journal of Consumer
Research, 713–724 (2005).
34. Daniel M. Oppenheimer, Spontaneous Discounting of Availability in Frequency
Judgment Tasks, 15 Psychological Science 100–05 (2004).
35. Reported in Duane T. Wegener, and Richard E. Petty, The Flexible Correction Model:
The Role of Naïve Theories of Bias in Bias Correction, in Advances in Experimental Social
Psychology 29, (M. Zanna ed., 1997).
 biases in perception and memory 257

In one experiment, Petty and his colleagues asked college students to judge
whether a person accused of the crime of rape was guilty or not. Participants in
the control group were presented with ambiguous information that led them to
be uncertain as to whether or not the defendant had committed the crime.36
A second group read the same ambiguous information but also learned that the
defendant had pled guilty to a prior crime of rape. A third group read the ambig-
uous information but also learned that the defendant had pled guilty to a prior
crime of burglary.
Participants were then asked to give two judgments: their own personal opin-
ion of the defendant’s guilt, and the verdict that they would render if they were a
juror. When giving their own opinions, participants were more likely to consider
the defendants with prior records (rape and burglary) guilty than the defendant
with no prior record. Perhaps they regarded some people as “criminal types.”
When acting as jurors and rendering verdicts, however, there was no difference
between the defendant with the burglary and the defendant with no record.
However, more participants judged the defendant with the prior rape conviction
to be guilty. The researchers concluded that when trying to be good jurors, people
may correct for biases due to factors they perceive as irrelevant to making an
impartial judgment (prior burglary conviction for a rape suspect), but not for
biases due to information they perceive as relevant (prior rape conviction for a
rape suspect) even when instructed to disregard it.
People may also correct for biases that they believe resulted from stereotypes
or from their mood at the time they made a judgment. But there is evidence that
attempts to suppress stereotypical thinking may actually increase stereotyping,
and that people sometimes overcorrect, for example, leading to a less favorable
evaluation of a candidate when they are in a good mood. 37

9.6.3 Availability, Vividness, and Inference

Richard Nisbett and Lee Ross note that “vivid information is more likely to be
stored and remembered than pallid information,” and thus more likely to be

36. Richard E. Petty, Duane T. Wegener, and Monique A. Fleming, “Flexible Correction
Processes: Perceived Legitimacy Of Bias And Bias Correction,” paper presented at the
annual meeting of the Society for Experimental Social Psychology (Sturbridge, MA:
1996).
37. Leonard Berkowitz et al., On the Correction of Feeling-Induced Judgmental Biases
131, in Feeling and Thinking: The Role of Affect in Social Judgments (New York:
Cambridge University Press, 2000); Linda M. Isbell and Robert S. Wyer, Jr., Correcting for
Mood-Induced Bias in the Evaluation of Political Candidates: The Roles of Intrinsic and
Extrinsic Motivation, 25 Personality & Soc. Psychol. Bull. 237 (1999). See also Ehud
Guttel, Overcorrection, 93 Geo. L.J. 241 (2004) (noting that the legal system sometimes
corrects for overcorrection); C. Neil Macrae et al., Out of Mind But Back in Sight: Stereotypes
on the Rebound, 67 J. Personality & Soc. Psychol. 808 (1994).
258 problem solving, decision making, and professional judgment

retrieved at some later date and to affect later inference.”38 Factors that contrib-
ute to the vividness of information include:
• Emotional interest. Events in which we or people we know are involved have
more emotional interest than those involving strangers.
• Concreteness. Even events involving strangers can be emotionally gripping
if they are described in sufficient detail to prompt imagery. Compare “Jack
was killed by a semi-trailer that rolled over his car and crushed his skull” to
“Jack sustained fatal injuries in a car accident.” Concrete descriptions have
more information, which, even if redundant or irrelevant to the attribution
of causation, is more likely to be remembered. Statistics may be the least
concrete form of information.
• Direct experience. First-hand information has more salience than the reports
of others. By the same token, a face-to-face recommendation is likely to be
more effective than the same recommendation in writing.
In addition to its availability, vivid information is more likely to recruit addi-
tional information from one’s memory and people are more likely to rehearse, or
mull over, vivid information, making it even more memorable.
The normative problem of using vividness, and availability more generally, as
criteria for inference is that they are not always strongly correlated with their
evidential value. This is illustrated by a number of experiments in which vivid-
ness caused participants to generalize from obviously biased samples. For exam-
ple, Ruth Hamill,Timothy Wilson and Richard Nisbett showed participants a
videotape interview of a prison guard, which depicted him as either humane or
brutal. Some were told that the guard was typical, some that he was highly atypi-
cal. When later asked to generalize about prison guards, both groups of partici-
pants concluded that prison guards generally tended to be like the one they had
seen in the interview.39 Vivid anecdotal evidence, can have a powerful effect on
people’s perceptions.

9.6.4 Vividness and the Media

With a summer season framed by a shark attack on a boy in Florida two days after the
Fourth of July and the death of a man on Labor Day on the Outer Banks of North
Carolina, the danger from sharks could easily be seen as rising dramatically.

This is being seen as the “Summer of the Shark” as Time magazine put it in a July 30
cover story bristling with images of razor-sharp teeth . . .

38. Nisbett and Ross, supra at 45.

39. Ruth Hammill, Timothy DeCamp Wilson, and Richard E. Nisbett, Insensitivity to
Sample Bias: Generalizing from Atypical Cases, 39 Journal of Personality and Social
Psychology 578–89 (1980).
 biases in perception and memory 259

But notwithstanding the bloody attacks . . . scientists said the new fears were overblown.
There is no rampage. If anything, they say, the recent global trend in shark attacks is
down, even as news media attention soars.40

. . . [Y]ou cannot talk sensibly about shark attacks this summer unless you talk first of
the sheer number of humans who flock to the beaches in the summertime. Shark attacks
will drop off precipitously now that Labor Day has come, because there will be less
human flesh in the water to be bitten . . .

Twenty-eight children in the United States were killed by falling television sets between
1990 and 1997 . . . [T]hat is four times as many people as were killed by great white
sharks in the 20th century. Loosely speaking, this means that “watching ‘Jaws’ on TV is
more dangerous than swimming in the Pacific.”41

In the aftermath of the terrorist attacks of September 11, 2001, on the World
Trade Center and on the Pentagon, and the SARS epidemic in Toronto in 2003,
Neal Feigenson and his colleagues studied Americans’ and Canadians’ percep-
tions of the two risks.42
Canadians estimated the percentage chance that they would become seriously
ill or die from SARS within the next year as 7.43%, and the percentage chance
that they would suffer a similar outcome from terrorism as 6.04%, a signifi-
cant difference. By contrast, Americans estimated the percentage chance that
they would become seriously ill or die from SARS within the next year as
2.18%, and the percentage chance that they would suffer a similar outcome
from terrorism as 8.27%, also a significant difference.
The authors estimate that the probability of a Canadian’s contracting SARS
was in fact less than .0008 percent and (albeit more speculatively) that the prob-
ability of an American dying from terrorism within the year was about .001 per-
cent. They explain the different perceptions in terms of “systematically different
media coverage of those risks, making those risks differently available to them”:
[A] sampling of national and local print coverage of SARS and terrorism indi-
cates that Canadian media sources devoted about 40% more articles to SARS
than American media sources did, while more than 14 times as many articles
about terrorism appeared in American as opposed to Canadian print media.
These threat by country interactions roughly parallel the national patterns of
respondents’ perceptions of SARS and terrorism risks.

40. William J. Broad, Scientists Say Frenzy Over Shark Attacks Is Unwarranted, New York
Times, Sept. 5, 2001.
41. The Statistical Shark, New York Times, Sept. 6, 2001.
42. Neal Feigenson, Daniel Bailis, and William Klein, Perceptions Of Terrorism And
Disease Risks: A Cross-National Comparison, 69 Mo. L. Rev. 991 (2004).
260 problem solving, decision making, and professional judgment

Paul Slovic and his colleagues have conducted numerous surveys of people’s
estimates of the frequency of causes of death. The social scientists found a strong
correlation between media coverage—especially the coverage of spectacular
events—and errors in estimation. “In general, rare causes of death were overes-
timated and common causes of death were underestimated . . . [A]ccidents were
judged to cause as many deaths as diseases, whereas diseases actually take about
15 times as many lives. Homicides were incorrectly judged to be more frequent
than diabetes and stomach cancer. Homicides were also judged to be about as
frequent as stroke, although the latter annually claims about 11 times as many
lives. Frequencies of death from botulism, tornadoes and pregnancy (including
childbirth and abortion) were also greatly overestimated.”43
Chip Heath and his colleagues studied media coverage of mad cow disease
outbreaks in France to see how language frames affect the public’s perception of
risk.44 Monthly beef consumption was affected more by the number of articles
that referred to the disease as “mad cow” than by articles that referred to the
disease by one of its scientific names, such as bovine spongiform encephalopa-
thy (BSE.). The “mad cow” articles tended to deal less with the scientific research
conducted on the disease and more with its sensational personal and social ram-
ifications, whereas the BSE articles were more likely to include technical content
and cite current studies. The authors conclude that “mad cow,” which was used
far more often than the scientific names, made the public more risk averse in
consuming beef. (Policy makers’ responses seemed more closely aligned with
scientific reports in the media, suggesting a degree of deliberation independent
of their constituents’ fears.)
Returning to the earlier discussion of fluency as a source of judgment,
research by Hyunjin Song and Norbert Schwarz suggests that people perceive
objects, diseases, foods, etc. with difficult to pronounce names to be more risky
or threatening than those with familiar-sounding names.45 In particular, when
people were asked to rate the dangerousness of a food additives named magnal-
roxate versus hnegripitrom, or roller-coaster rides named Chunta versus
Vaiveahtoishi the latter, more difficult-to-pronounce additives and rides were
judged to be more dangerous. The researchers conclude that easy-to-pronounce
or seemingly familiar names can be processed fluently, which gives rise to
the belief that one has “heard it before,” and therefore to the belief that it is
relatively safe.

43. Paul Slovic, Baruch Fischhoff, and Sarah Lichtenstein, Rating the Risks, in Paul
Slovic, The Perception of Risk 105 (London: Earthscan Publications, 2000).
44. Marwan Sinaceur, Chip Heath, and Steve Cole, Emotional and Deliberative Reactions
to a Public Crisis: Mad Cow Disease in France, American Psychological Society 247–54
(2005), http://faculty-gsb.stanford.edu/heath/documents/PsychSci-Mad%20Cow.pdf.
45. Hyunjin Song, and Norbert Schwarz, If It’s Difficult to Pronounce, It Must Be Risky,
20 Association for Psychological Science 135–38 (2009).
 biases in perception and memory 261

9.7 egocentrism: a particular memory bias in acquisition

and/or retrieval

The duet from Gigi quoted in Section 9.2 exemplifies how participants or observ-
ers may recall the same event differently even when there is no ostensible reason
for the difference. Beyond this, people’s recollections and judgments tend to be
clouded by an egocentric bias. When married couples estimate their proportion-
ate contribution to household tasks, the sum exceeds 100 percent.46 So too of
collaborators’ estimates of their contributions to joint projects, such as this book.
While the phenomenon may be partially explained by self-serving motives or
having inflatedly positive views of oneself, it extends to negative behaviors as
well. Spouses report causing, as well as resolving, most of the problems in their
relationship,47 and people are prone to overestimate their own errors and unat-
tractive behaviors.48
One plausible explanation for the egocentric bias is that our own actions are
more readily accessible to our memories than the actions of others. Furthermore,
our recollection of behaviors may be guided by a theory of one’s own behavior.
Rather than go through the laborious process of recalling particular activities, we
ask ourselves, “how often do I do this kind of thing?” and take the further short-
cut of looking to our disposition by asking, “am I the sort of person who does
this?” Of course, we might ask the same questions about the other person. But
either because we find it “difficult or time consuming to retrieve two separate
pieces of information and combine them to procedure a single judgment,” or
because we anchor on our own contributions and then (insufficiently) adjust for
the other’s, it is our own behavior that dominates.49
Even when an individual holds himself accountable for an outcome vis-à-vis
others, the question remains open whether the outcome was due to internal fac-
tors such as skill or its absence, or to external factors, such as bad luck. In an
analysis of newspaper accounts of sporting events, Richard Lau and Dan Russell
found that players attributed good outcomes to their skill 75 percent of the time,
but took responsibility for bad outcomes only 55 percent of the time.
Sportswriters—who one might expect to have more distance—were less likely
than players or coaches to attribute wins to internal factors, but even sportswriters

46. Michael Ross and Fiore Sicoly, Egocentric Biases in Availability and Attribution, 37
Journal of Personality and Social Psychology 322 (1979).
47. See Linda Babcock et al., Biased Judgments and Fairness in Bargaining, 85 American
Economic Review 1337, 1337–38 (1995).
48. Nisbett and Ross, supra at 76; Justin Kruger and Thomas Gilovich, Naive Cynicism,
in Everyday Theories of Responsibility Assessment: On Biased Perceptions of Bias, 76 Journal
of Personality and Social Psychology 743–53 (1999).
49. Suzanne C. Thompson and Harold H. Kelley, Judgments of Responsibility for Activities
in Close Relationships, 41 Journal of Personality and Social Psychology 469 (1981).
262 problem solving, decision making, and professional judgment

were more likely to provide internal attributions for their home teams’ wins than
for losses.50
Egocentrism is often correlated with self-servingness. Among people working
different amounts of time in a joint enterprise, those who work more believe
they should be paid more, while those who work less believe that both parties
should be paid equally. People may come to self-serving results indirectly, by
disproportionately valuing the particular task they contribute to the joint enter-
prise (e.g., washing the laundry compared to sweeping the floors). But people
often believe they are entitled to a greater share than others even in the absence
of joint activities. For example, in a simulation of a commons problem involving
fishing associations that needed to reduce their harvests to preserve the stock of
fish for long-run profits, participants believed they were entitled to a greater por-
tion of the catch than the others.51

9.7.1 Debiasing Egocentrism

If egocentrism is rooted in the accessibility of one’s own actions or interests,
then having an individual consider other people’s perspective should bring their
sense of fairness more into line. And it does—though with an important and
unfortunate kicker. For example, when debate team participants were just asked
to allocate their contributions to a joint effort, their net allocations summed to
156 percent; when they were asked to think about others’ contributions, the sum
was a mere 106 percent.52
And now the kicker: While putting yourself in the other person’s shoes often
brings your perception of fairness closer to reality, it does not reduce selfish
behavior, and in some cases may increase it—presumably because adopting the
other person’s perspective leads you to believe that the other person will act in an
egocentric and self-interested way, which you then counter by taking more your-
self. Thus in the fishing associations example, instructing participants to look at
the commons problem from other associations’ point of view reduced, although
did not totally eliminate, egocentric assessments of what they deemed to be their
fair share.53 But that instruction actually increased the amount of the harvest the

50. Richard R. Lau and Dan Russell, Attributions in the Sports Pages, 39 Journal of
Personality and Social Psychology 29 (1980).
51. Kimberly A. Wade-Benzoni, Ann E. Tenbrunsel, and Max H. Bazerman, Egocentric
Interpretations Of Fairness In Asymmetric, Environmental Social Dilemmas: Explaining
Harvesting Behavior and the Role of Communication, 67 Organizational Behavior and
Human Decision Processes 111–26 (1996).
52. Eugene M. Caruso, Nicholas Epley, and Max H. Bazerman, The Good, The Bad,
And The Ugly Of Perspective Taking In Groups, in E. A. Mannix, M. A. Neale, &
A. E. Tenbrunsel, (series eds), Research On Managing Groups And Teams, 8 Ethics in
Groups 201–224 (London: Elsevier, 2006).
53. Eugene M. Caruso, Nicholas Epley, and Max H. Bazerman. When Perspective
Taking Increases Taking: Reactive Egoism in Social Interaction, 91 Journal of Personality
and Social Psychology 972 (2006).
 biases in perception and memory 263

participants actually took in the simulation. In a sense, cynicism about the other
participants resulted in reactive egoism.54

9.7.2 The Recollection of One’s Own Past Beliefs, Feelings, and Events

Herein lies a difficulty in any autobiographical sketch which purports to deal with one’s
mental development. It is a story of oneself in the past, read in the light of one’s present
self. There is much supplementary inference—often erroneous inference—wherein “must
have been” masquerades as “was so.”
—C. Lloyd Morgan, History of Psychology in Autobiography (1930)55

Research suggests that emotions may not be stored directly in memory but
rather “are reconstructed based on memories of the emotion-eliciting
circumstances.”56 Reconstructions may go beyond emotions to involve the mem-
ories of attitudes and even actions. Michael Ross57 argues that such memories
are based on people’s implicit theories of human stability and change. All things
being equal, we think that attitudes are consistent over time. For this reason,
when we infer past attitudes from present ones, we tend to exaggerate their con-
sistency. For example, participants whose views changed about school busing to
achieve desegregation or about the health effects of vigorous physical exercise
misremembered their earlier views; university students whose attitudes toward
a dating partner had changed over several months recalled their earlier impres-
sion as being more consistent with their current attitudes than was the case.
Furthermore, because we tend to seek consistency in attitudes and behavior, a
change in attitudes can produce inaccurate recollection of behaviors. Thus, when
experimenters caused participants to believe that regularly brushing their teeth
was not as important as they had thought, they recalled less brushing than they
actually had done.
While a belief in personal consistency is the norm, we also have theories of
change. So, for example, people who have engaged in education and self-
improvement programs tend to exaggerate the difference between their past and
the present knowledge or behavior. In an experiment involving a program to
improve study skills, students recalled their original assessment of their study
skills to be worse than they had originally reported and reported greater improve-
ment than the facts warranted. They not only overpredicted the resulting

54. Kruger and Gilovich; Epley, Caruso, and Bazerman. However, selfish behavior was
reduced when the group project was structured or characterized as cooperative rather
than competitive.
55. Quoted in Michael Ross, Relation of Implicit Theories to the Construction of Personal
Histories, 96 Psychological Review 341 (1989).
56. Id.
57. Id.
264 problem solving, decision making, and professional judgment

improvement in their grades, but subsequently recalled the new grades as being
higher than they actually were.

9.8 conclusion: naïve realism

How narrow his vision, how cribbed and confined!

How prejudiced all of his views!
How hard is the shell of his bigoted mind!
How difficult he to excuse!

His face should be slapped and his head should be banged;

A person like that ought to die!
I want to be fair, but a man should be hanged
Who’s any less liberal than I.
—Broadmindedness, in Franklin P. Adams, Something Else Again (1920)

The biases in the acquisition and recollection of information outlined above

are exacerbated by three related convictions that people have about their percep-
tions and judgments and the perceptions and judgments of other people. These
are the elements of what Lee Ross has termed naïve realism:58
1. I see actions and events as they are in reality. My perceptions and reactions
are not biased: rather, they are an unmediated reflection of the “real nature”
of whatever it is I am responding to.
2. Other people, to the extent that they are willing and able to see things in a
similarly objective fashion, will share my perceptions and reactions.
3. When others perceive some event or react to it differently from me, they
(but not I) have been influenced by something other than the objective
features of the events in question. Their divergent views probably result
from an unwarranted ideological rigidity or a self-serving bias that I am not
influenced by. The more extreme the view is in divergence from my own,
the stronger their bias probably is.
Naïve realism can be especially pernicious in the context of disagreement or
conflict. It has the practical implications that:
• Partisans tend to overestimate the number of others who agree with their
views—the false consensus effect—or at least overestimate the number who

58. Lee Ross and Donna Shestowsky, Contemporary Psychology’s Challenge to Legal
Theory of Practice, 97 Nw. U. L. Rev. 1081 (2003).
 biases in perception and memory 265

would agree with them if apprised of the “real” facts; partisans therefore
assume that disinterested third parties would agree with them.
• Partisans tend to see viewpoints that differ from their own as highly reveal-
ing both of personal dispositions (for example, gullibility, aggressiveness,
pessimism, or charitableness) and of various cognitive and motivational
biases. In fact, differences in judgment often reflect differences in the way
a given issue or object of judgment is perceived and construed rather than
a difference in the perceivers’ values or personality traits.
• When partisans acknowledge the influence of particular features of their
experience or identity, they see it not as a bias but as a source of enlighten-
ment, while they see the other parties unique experiences or identities as a
source of bias.
• Partisans on both sides of an issue will typically perceive evenhanded
media to be biased against them and to favor their adversaries. They are apt
to see the same hostile bias in the efforts and decisions of evenhanded
third-party mediators or arbitrators.
• Partisans will be polarized and extreme in their view of others. Once they
believe a person to be biased, they tend to discount his views entirely, and
will underestimate areas of agreement and underestimate the prospects of
finding “common ground” through discussion or negotiation.
We have seen that being asked to articulate the weaknesses of one’s own posi-
tion may counter perceptual biases in some situations. A broader hope of this
book is that students and readers who understanding those biases and the phe-
nomena of naïve realism will be less likely to fall into their traps. But we should
be among the first to admit that the hope does not rest on much of an empirical
foundation.
This page intentionally left blank
10. biases in processing and judging
information

The preceding chapter examined biases in empirical judgment involving the

acquisition and recollection of information, while also necessarily touching on
processing and analysis of information. This chapter focuses on the processing
and analysis of information while necessarily touching on its recollection. The
availability heuristic continues to play an important role as we examine the
anchoring effect, hindsight bias, and the tendencies to seek evidence that con-
firms one’s views and to be overconfident in one’s judgment. The chapter ends
by reviewing the place of intuition and analysis in empirical judgments and the
role of experts.

10.1 anchoring, adjustment, and activation

In making a numerical estimate or choice, people tend to be influenced by exter-

nally-given reference points. This is true even when the reference point is entirely
arbitrary. For example, Amos Tversky and Daniel Kahneman asked people to
estimate the percentage of African countries in the United Nations. Before the
question was posed, a “wheel of fortune” with numbers from 1 to 100 was spun.
The median estimate for participants whose wheel landed on the number 10 was
25 percent; when the wheel happened to land on the number 45, the estimate
jumped to 65 percent.1
What Tversky and Kahneman call the anchoring and adjustment bias captures
the fact that we intuitively anchor our judgment on the given value and then use
rational processes to adjust away from it; but that our adjustments tend to be
insufficient. The phenomenon would more accurately described as one involv-
ing anchoring and insufficient adjustment.
The participants in the United Nations quiz were not experts on geography.
However, experts seem no less susceptible to the anchoring and adjustment
bias. To explore how experienced auditors’ estimation of fraud might be
influenced by an anchor, E. E. Joyce and G. C. Biddle gave them a questionnaire
with the following problem:2

1. Amos Tversky and Daniel Kahneman, Judgment Under Uncertainty: Heuristics and
Biases. 185 Science, 1124–1130 (1974).
2. Edward J. Joyce and Gary C. Biddle, Anchoring and Adjustment in Probabilistic
Inference in Auditing, 19 Journal of Accounting Research 120 (1981).
268 problem solving, decision making, and professional judgment

It is well known that many cases of management fraud go undetected even

when competent annual audits are performed . . . We are interested in obtain-
ing an estimate from practicing auditors of the prevalence of executive-level
management fraud as a first step in ascertaining the scope of the problem.

1. Based on your audit experience, is the incidence of significant executive-

level management fraud more than 10 [200] in each 1000 firms (i.e., 1%
[20%]) audited by Big Eight accounting firms?
2. What is your estimate of the number of Big Eight clients per 1000 that have
significant executive-level management fraud?
For auditors given the number 10 in the first question, the mean response to
the second question was 16.5 per thousand; for those given 200, the mean
response was 43 per thousand.

10.1.1 Anchoring in Negotiations

In the belief that the phenomenon would affect bargaining or negotiations,
Gregory Northcroft and Margaret Neale conducted an experiment involving the
purchase of real estate.3 Both amateurs and professional real estate agents were
asked to appraise residential properties. They were given all of the information
that would be available to real estate agents, including the owner’s listing price,
detailed information about the property in question, and the summary of sales
in the neighborhood and city—and were also shown the property in question.
The only information that differed with respect to a particular residence was the
price at which the owner listed it—a variable entirely in the owner’s control.
Four groups of amateurs were given different listing prices for the same house;
since there were fewer expert participants in the experiment, they received only
two listing prices. Table 10.1 shows the listings and appraisals for a particular
property.
Thus, professionals and amateurs alike adjusted their appraisal value signifi-
cantly based only on the listing price. In debriefing, amateurs were aware that
the listing price played a role in their judgments, while the professionals denied
this—whether through a lack of self-awareness or professional pride.
Russell Korobkin and Chris Guthrie conducted an experiment in which
parties engaging in settlement negotiations were offered different anchors.
Plaintiffs were more likely to accept a $12,000 final offer when it followed a
$2,000 opening offer than when it followed a $10,000 opening offer. The
researchers hypothesize that the low anchor developed low expectations, which

3. Gregory B. Northcroft and Margaret A. Neale, Experts, Amateurs, and Real Estate:
An Anchoring-And-Adjustment Perspective on Property Pricing Decisions, Organisational
Behavior and Human Decision Processes 39 (1987).
 biases in processing and judging information 269

table 10.1 real estate appraisals

Listing Price Amateurs’ Appraised Value Professionals’ Appraised Value

$65,900 $63,571 $67,818

$71,900 $67,452 NA
$77,900 $70,423 NA
$83,900 $72,196 $76,380

made the final offer seem comparatively generous.4 And Jeff Rachlinski, Chris
Guthrie, and Andrew Wistrich presented federal judges with the facts of a per-
sonal injury case, in which the plaintiff had been hospitalized for several months
and was left paraplegic and confined to a wheelchair. One group was anchored
by being informed that the defendant had moved for dismissal on the ground
that the case did not meet the $75,000 jurisdictional minimum for a diversity
case;5 the other group was given no such anchoring information. Though almost
none of the judges granted the motion to dismiss, the average award in the
anchored group was $882,000 compared to $1,249,000 in the unanchored
group.6
Anchoring in Jury Awards. In an experiment by John Malouff and Nicola
Shutte, when plaintiff’s lawyer requested $100,000 damages, mock jurors
awarded $90,000, and when plaintiff’s lawyer requested $500,000 damages, they
awarded nearly $300,000 on average.7 While jurors assessing punitive damages
often anchor on the amount of actual damages, they also anchor on the amount
including punitive damages requested in the closing argument of the plaintiff’s
attorney. In an experiment involving a train derailment that dumped herbicide
and caused serious environmental damage, plaintiff’s attorney told one group of
mock jurors: “So the range you may want to consider is between $15 million and
about a half a year’s profit, $50 million”; another group was given the range
between $50 million and $150 million. The median awards by the two juries

4. Russell Korobkin and Chris Guthrie, Opening Offers and Out-of-Court Settlements: A
Little Moderation May Not Go a Long Way, 10 Ohio State Journal on Dispute Resolution
1 (1994).
5. Federal courts can hear cases arising under state law if the amount in controversy
is $75,000 or more and the plaintiff and defendant reside in different states.
6. Chris Guthrie, Jeffrey J. Rachlinski, and Andrew J. Wistrich, Inside the Judicial
Mind, 86 Cornell L. Rev. 777 (2001).
7. John Malouff and Nicola S. Schutte, Shaping Juror Attitudes: Effects of Requesting
Different Damage Amounts in Personal Injury Trials, 129 Journal of Social Psychology
491 (1989).
270 problem solving, decision making, and professional judgment

were $15 million and $50 million, respectively.8 Experimental data also suggests
that statutory caps on damages may affect settlement rates.9

10.1.2 Extreme and Untrustworthy Anchors

Do extreme or implausible anchors trigger the same phenomenon? Fritz Strack
and Thomas Mussweiler asked participants to estimate the year that Einstein came
to the United States, giving them anchors of 1215, 1905, 1939, and 1992. (The cor-
rect date is 1933.) Even the ridiculously extreme numbers had anchoring effects.10
What about anchors from an untrustworthy source? In 1950, in an attack on
Secretary of State Dean Acheson, Senator Joseph McCarthy claimed that he had
a list of fifty-seven people in the State Department who were known to be mem-
bers of the American Communist Party. Even if you were highly skeptical of
McCarthy’s trustworthiness, how many Communists might you think he had
actually identified? What might your estimate have been if he had said he had a
list of five hundred Communists?

10.1.3 Cross-Modal Anchors

Studies by Danny Oppenheimer and his colleagues suggest that a nonnumeric
stimulus that conveys magnitude can bias numerical estimates. When asked
to guess the length of the Mississippi River, participants who had been exposed
to long lines drawn across a sheet of paper produced estimates significantly longer
than those produced by another group that had been exposed to shorter lines.11

10.1.4 Explanations for the Phenomenon

There are two fundamentally different accounts of the anchoring and adjustment
phenomenon. One focuses on the adjustment process and asks why adjustments

8. Reid Hastie, David A. Schkade, and John W. Payne, Do Plaintiffs’ Requests and
Plaintiffs’ Identities Matter?, in Cass R. Sunstein et al., Punitive Damages: How Juries
Decide (Chicago: University of Chicago Press, 2002).
9. A low cap increased settlement rates while a high cap decreased them. The authors
term the latter phenomenon, “motivated anchoring,” in which a relatively high damage
cap disproportionately anchors the plaintiff’s estimate of the likely damage award. The
result is a widened disparity in opposing litigants’ judgments, and less settlement. Greg
Pogarsky and Linda C. Babcock, “Damage Caps, Motivated Anchoring, and Bargaining
Impasse” (Aug. 1, 2000), available at SSRN: http://ssrn.com/abstract=235296
10. Fritz Strack and Thomas Mussweiler, Explaining the Enigmatic Anchoring Effect:
Mechanisms of Selective Accessibility, 73 Journal of Personality and Social Psychology
437 (1997).
11. Daniel M. Oppenheimer et al., Anchors Aweigh: A Demonstration of Cross-Modality
Anchoring and Magnitude Priming, Cognition (2007). When the anchors were placed in
appropriately scaled contexts—e.g., a shorter line tracing the length of the Mississippi
on a small scaled map, and a longer line tracing its length on a larger scaled map—the
influence of the anchoring effect diminished.
 biases in processing and judging information 271

away from the initial anchor are insufficient. Under this view, we “adjust the
anchor until shortly after it enters a range of plausible values for the target item.
Thus, when adjusting from a high anchor, decision makers stop at the high end
of plausible values, but stop at the low end when adjusting from low anchors.”
Adjustment is cognitively effortful, and a “lack of effort or lack of cognitive
resources cause adjustment to be terminated too soon, resulting in a final
response that is too close to the anchor.”12
Alternatively, the anchor may increase the availability of features that the
anchor and the target—e.g., the number to be determined—hold in common,
selectively activating information about the target that is consistent with the
anchor (while not activating other information).13 Gretchen Chapman and Eric
Johnson hypothesize:
Assume a decision maker, when requested to make a judgment, can access a
variety of features about the target item, either from memory or from the
external environment. Some subset of this information is retrieved and used
as the basis for . . . judgment. If a representative sample is retrieved, the judg-
ment will be unbiased. However, . . . the presence of an anchor increases the
activation of features that the anchor and target hold in common while reduc-
ing the availability of features of the target that differ from the anchor.
For example, suppose a decision maker asked to evaluate a bus trip was
presented with a low anchor [for the cost of the bus ticket]. The decision maker
might contemplate a bumpy, noisy, crowded ride aboard a city-owned vehicle
that spouts diesel fumes and has hard plastic seats and no air conditioning.
Conversely, if a high anchor were presented, the decision maker might
imagine a luxurious tour bus with reclining plush velour seats, sound system,
and bar . . ..
Under either of these explanations, a person might sometimes believe that
the anchor provides information relevant to the actual amount. This might be
true in the fraud, real estate, and jury award examples described above (though
certainly not in anchoring on the percentage of African countries in the United
Nations based on the spin of a wheel.) Even so, the phenomenon is likely to
result in a biased judgment.

12. Gretchen B. Chapman and Eric J. Johnson, Incorporating the Irrelevant: Anchors
in Judgments of Belief And Value, in Thomas Gilovich, Dale Griffin, and Daniel
Kahneman, Heuristics and Biases, The Psychology of Intuitive Judgment 127
(2002) [hereafter, Heuristics and Biases].
13. Gretchen B. Chapman and Eric J. Johnson, Anchoring, Activation, and the
Construction of Values, 79 Organizational Behavior and Human Decision Processes
115 (1999).
272 problem solving, decision making, and professional judgment

10.1.5 Debiasing
The insufficient adjustment and activation theories may account for anchoring
phenomena in different circumstances. In any event, the phenomenon is extraor-
dinarily robust. Simply warning people about it does not decrease anchoring;
nor does offering incentives for accurate estimates.14

10.2 hindsight bias

Baruch Fischhoff, who pioneered research in the subject, describes hindsight bias
as follows:
In hindsight, people consistently exaggerate what could have been anticipated
in foresight. They not only tend to view what has happened as having been
inevitable but also to view it as having appeared “relatively inevitable” before
it happened. People believe that others should have been able to anticipate
events much better than was actually the case.15
In a classic experiment, Fischhoff provided a description of the 1814 war
between the British and the Nepalese Gurkhas and informed participants that it
had four possible outcomes: British victory, Gurkha victory, military stalemate
with no peace settlement, or stalemate with a peace settlement. In each of four
conditions participants were told that one of the four outcomes had actually
occurred and were then asked to estimate its prior probability as well as those of
the three alternatives. As Fischhoff had predicted, participants’ knowledge of
what they believed to be the actual outcome significantly affected their estimates
of the prior probability.
In an experiment examining hindsight bias in a legal context, Kim Kamin and
Jeffrey Rachlinski gave two groups of participants the same facts, except that one
group was asked whether to take precautions to avert a potential accident and the
second was asked whether to assess liability after an accident had already
occurred:
Foresight condition. Participants were told that a city had constructed a draw-
bridge and needed to determine whether the risk of a flood warranted main-
taining a bridge operator during the winter when the bridge was not in use.
Hiring the operator would serve as a precaution. The operator would monitor
weather conditions and raise the bridge to let debris through if the river

14. Gretchen B. Chapman and Eric J. Johnson, supra.

15. Baruch Fischhoff, For Those Condemned to Study the Past: Heuristics and Biases in
Hindsight, in Judgement Under Uncertainty: Heuristics and Biases 335, 341–42
(Daniel Kahneman, Paul Slovic, and Amos Tversky, eds. (New York: Cambridge University
Press, 1982).
 biases in processing and judging information 273

threatened to flood. Testimony in an administrative hearing before the city

council indicated that hiring the operator would cost $100,000 annually, and
that the potential damage from a flood was $1 million. The question posed to
participants in the foresight condition was whether a flood was sufficiently
probable that the city should appropriate funds for the operator.
Hindsight condition. Participants were asked to assume the role of jurors in
a civil trial. They were told that the city had decided not to hire the operator
and that during the first winter of the bridge’s existence, debris lodged under
it, which resulted in a flood that could have been prevented had an operator
been hired. The flood damaged a neighboring bakery, whose owner then sued
the city. Participants in the hindsight condition were instructed to hold the
city liable if the flood was sufficiently probable that the city should have hired
the operator to prevent it.
In both conditions, the participants were told to use the Learned Hand for-
mula for tort liability, in which an actor bears the responsibility to take a precau-
tion if the cost of taking the precaution is less than the harm it would have
prevented (based on the total amount of the harm and the probability of its occur-
rence). In particular, with the operator costing $100,000 and flood damages pre-
dicted or stipulated to be $1 million, the city should/should have hired the
operator if the probability of a flood was greater than 10 percent. Testimony in
both conditions was ambiguous about whether, based on past events, the prob-
ability was greater or less than 10 percent. The results of the experiment were
clear cut:
• 24 percent of the participants in the foresight condition (administrative
hearing) believed the city should take the precaution.
• 57 percent of the participants in the hindsight condition (trial) believed that
the city should have taken the precaution.
It can be perfectly rational to revise your earlier judgment about the probabil-
ity of an event occurring based on what actually happened—if the occurrence
provided information from which you could draw a valid statistical inference.16
But it is obviously an error to believe, after an event has occurred, that before the
event you would have predicted its occurrence to be more likely than you could
reasonably have foreseen at that time. By the same token, it is an error to impose
liability on an actor based on the erroneous notion that a reasonable person
would have known in advance that the danger was as high as it now seems in
hindsight.
Courts are frequently called upon to decide what a policy maker, professional,
or individual “should have known” and should have done. As the drawbridge

16. See Mark Kelman, David Fallas and Hillary Folger, Decomposing Hindsight Bias, 16
Journal of Risk and Uncertainty 251 (1998).
274 problem solving, decision making, and professional judgment

case illustrates, this is an essential part of the negligence standard in tort liability
and plays a significant role in many other legal domains, ranging from trust and
estate law (did the trustee invest the beneficiary’s funds prudently?) to substan-
tive criminal law (did the defendant claiming self-defense reasonably believe that
he was in danger of being killed?) and criminal procedure (did the police officer
have probable cause for the arrest and search?). In all of these situations, the trier
of fact is vulnerable to hindsight bias.
Summarizing the rich psychological literature since this 1975 study, Jeffrey
Rachlinski17 writes that hindsight bias
results primarily from the natural (and useful) tendency for the brain to auto-
matically incorporate known outcomes into existing knowledge, and make
further inferences from that knowledge. For example, participants in the
British-Gurkha study who were told that the British had won, probably made
inferences about colonial warfare, generally. They may have concluded, for
example, that the advantages that the British had over the Gurkhas (better
weapons and training) were more important in such conflicts than the Gurkha
advantages (better knowledge of the terrain and higher motivation).
Participants informed of a Gurkha victory may have inferred the opposite.
When asked to estimate the [prior] probability of a British victory, the partici-
pants relied on their new belief that better weapons and training were more
important than knowledge of the terrain and higher motivation. These infer-
ences then induced them to make high estimates for the probability of
the known outcome, relative to the estimates of people who were unaware of
the outcome and hence had not developed such beliefs.
More generally, ignoring a known outcome is unnatural. Normally, people
should integrate new information into their existing store of knowledge and
use it to make future predictions. Assessing the predictability of past events
requires that the outcome and the subsequent inferences that depend upon
learning the outcome be ignored. Perhaps because such judgments are so
uncommon and unnatural, people cannot make them accurately.
Hindsight bias is obviously related to the availability heuristic. The concrete-
ness and vividness of the actual outcome of an event prompts us to explain it,
and hindsight explanations are usually easy to provide. Moreover, we do not usu-
ally develop casual scenarios for other possible, and even more probable, out-
comes that did not occur—although we could have readily explained those
outcomes had they occurred. The ease of explanation, and the relative availability
of explanations for occurring versus non-occurring events, becomes, in effect, a
heuristic for assessing their likelihood.

17. Jeffrey Rachlinski, Heuristics and Biases in the Courts: Ignorance or Adaptation?, 69
Oregon L. Rev. 61 (2000).
 biases in processing and judging information 275

Consider, for example, the many ways that another terrorist attack could be
launched against the United States—explosives in air freight shipments, suicide
bombers in shopping malls, dirty bombs, blowing up a chemical plant, bioterror-
ism, etc. Especially because we have heard warnings about all of these and many
other possibilities targeted at various locations, if and when an attack does occur
the causal chain will be readily available in hindsight. And, of course, people will
be held accountable for having failed to take adequate precautions against the
attack.18

10.2.1 Risk/Benefit Analysis in Hindsight

Based on an analysis of actual cases as well as a study involving mock jurors, Kip
Viscusi concludes that, notwithstanding Learned Hand’s seemingly reasonable
formulation, jurors may actually punish a corporate defendant who explicitly
used cost-benefit analysis:19
. . . [J]urors chose to award punitive damages even though thorough internal
risk analyses led the defendants to conclude that no additional safety improve-
ments were warranted. This result is the opposite of what would occur if the
legal system fostered better corporate risk behavior. More rational thinking
about risk and a conscientious effort to achieve risk-cost balancing in line
with society’s valuation of safety should signal corporate responsibility rather
than trigger corporate punishment.
. . . Why do jurors make the mistake of punishing corporations for risk-
cost balancing? A variety of conjectures are possible. People may be averse to
explicitly balancing money against human lives. Money and lives might be
considered incommensurable. Or, jurors might not be adequately trying to
place themselves in the shoes of the corporation at the time of the tradeoff
decision. At the time of the decision, the corporation sees only a small prob-
ability of an accident, not a certainty. In hindsight, a small corporate expendi-
ture would have prevented an identifiable death, whereas ex ante the
corporation would have had to make that expenditure thousands if not mil-
lions of times to decrease the risk of an abstract person’s death. Comparisons
involving identified victims and safety costs will overwhelm jurors’ sensibili-
ties, particularly for low-probability events with severe consequences. When
corporations systematically think about risk levels yet nevertheless pursue
product or environmental policies that fail to maximize safety, the juror may

18. For an argument that takes account of hindsight bias and nonetheless concludes
that airlines and government officials should have anticipated the 9/11 attack, see Max H.
Bazerman and Michael D. Watkins, Predictable Surprises: The Disasters You
Should Have Seen Coming and How to Prevent Them (Boston: Harvard Business
School, 2004).
19. W. Kip Viscusi, Corporate Risk Analysis: A Reckless Act?, 52 Stan. L. Rev. 547
(2000).
276 problem solving, decision making, and professional judgment

regard the corporate decision as “cold-blooded.” This difficulty arises, in part,

because of the well-documented role of hindsight bias with respect to retro-
spective risk judgments. What matters at the time of the corporate decision is
a comparison of the costs of the safety measure with its expected benefits,
which consist of the reduced probability of an accident multiplied by the value
of the likely damage from an accident. But . . . jurors tend not to compare the
expected benefits and costs. Rather, after the victim has been identified, they
simply compare the loss to the victim against the costs to save that individual,
neglecting the fact that before the accident the loss was only an abstract prob-
ability.

10.2.2 Debiasing Hindsight Bias

Because hindsight bias results from the high availability of the actual outcome,
one might suppose that it can be debiased by inducing someone to think about
the likelihood of other outcomes?20 In the drawbridge litigation, Kamin and
Rachlinski gave a group of jurors an instruction along these lines: “Making a fair
determination of probability may be difficult. As we all know, hindsight vision is
always 20/20. Therefore it is extremely important that before you determine the
probability of the outcome that did occur, you fully explore all other possible
alternative outcomes which could have occurred. Please take a moment to think
of all the ways in which the event in question may have happened differently or
not at all.”
The instruction had no effect, which shouldn’t be surprising in the light of
the earlier discussion of metacognition. As Norbert Schwarz and Leigh Anne
Vaughn note, “to the extent that attempts ‘to argue against the inevitability of the
outcome’ are experienced as difficult, they may only succeed in convincing us,
even more, that the outcome was, indeed, inevitable.” Replicating Fischhoff’s
experiment with the British-Gurkha War, they asked one group of participants to
give two reasons for an alternative outcome and asked another group to give ten
reasons. While hindsight bias wasn’t significantly attenuated in the first group,
it was exacerbated in the second because the counterfactual thoughts were so
difficult to imagine.
Other efforts to debias hindsight, such as paying people for accurate answers,
have not been more effective.

10.2.3 The Legal System’s Responses to Hindsight Bias

If hindsight bias is difficult to cure, the legal system nonetheless has ways of
countering it. Rachlinski notes that “allocating the burden of production and

20. Norbert Schwarz and Leigh Anne Vaughn, The Availability Heuristic Revisited:
Ease of Recall and Content of Recall as Distinct Sources of Information, in Heuristics and
Biases 103.
 biases in processing and judging information 277

setting the standard of proof . . . operate[s] against the party most likely to benefit
from the hindsight bias—the plaintiff.” And he goes on to describe other legal
mechanisms:
First, courts suppress evidence that would exacerbate the bias, such as subse-
quent remedial measures in negligence and products liability cases. Second,
they rely on standards of conduct developed ex ante, when reliable standards
are available. Examples include the use of custom as a defense in medical
malpractice cases and the business judgment rule in corporate law. Although
accurate assessments of the ex ante probabilities would be superior to these
strategies, courts seem to recognize such assessments will be biased in hind-
sight and are therefore unattainable.

10.3 confirmation bias and other resistance to modifying

prior expectations and theories

The human understanding when it has once adopted an opinion draws all things else to
support and agree with it. And though there be a greater number and weight of instances
to be found on the other side, yet these it either neglects and despises, or else by some
distinction sets aside and rejects, in order that by this great and pernicious
predetermination the authority of its former conclusion may remain inviolate.
—Francis Bacon, First Book of Aphorisms (1620)

Faced with the choice between changing one’s mind and proving that there is no need to
do so, almost everyone gets busy on the proof.
—John Kenneth Galbraith

[Herbert] Spencer began, like a scientist, with observation; he proceeded, like a scientist,
to make hypotheses; but then, unlike a scientist, he resorted not to experiment, nor to
impartial observation, but to the selective accumulation of favorable data. He had no
nose at all for “negative instances.” Contrast the procedure of Darwin, who, when he
came upon data unfavorable to his theory, hastily made note of them, knowing that they
had a way of slipping out of the memory a little more readily than the welcome facts.
—Will Durant, The Story of Philosophy 296 (1926)

Recall than in the examples of Bayesian statistics in Section 8.5, P(H), the
prior probability—or the strength of our belief in the hypothesis prior to obtain-
ing event-specific data—was determined by the base rate of the event in ques-
tion. (In our example, the base rate was the proportion of Blue taxis in a town.)
Bayes’ Theorem tells us how to revise an estimate after gaining new informa-
tion—say, about the eyewitness’s identification of the taxi involved in the
accident.
278 problem solving, decision making, and professional judgment

Consider, though, that at any given time we hold many beliefs that cannot
readily be described in terms of base rates—for example, the belief that, as of
2003, Iraq did or did not possess weapons of mass destruction. Those beliefs are
based on a variety of sources, including our own experience and the assertions of
others—teachers, scientists, politicians, journalists—which we may deem more
or less accurate or trustworthy. In these myriad instances, P(H) does not reflect
a calculable probability, but rather the subjective strength of our prior belief in
H. When we are called upon to revise a prior belief in terms of new evidence, the
task is not formulaic as much as intuitively combining the prior belief with the
current data.
In many if not most circumstances, it is entirely rational to give considerable
weight to one’s prior beliefs. However, people tend to interpret conflicting evi-
dence with a variety of confirmation biases, manifesting a pervasive tendency to
seek evidence supporting their prior beliefs or hypotheses, and to ignore or den-
igrate evidence opposing them.21 The belief by the CIA and other members of
the Western Intelligence Community that Iraq had weapons of mass destruction
provides a recent and dramatic example of the phenomenon. This was, in effect,
the conclusion of the Senate Select Committee on Intelligence:22
The Intelligence Community (IC) suffered from a collective presumption that
Iraq had an active and growing weapons of mass destruction (WMD) pro-
gram. This . . . led Intelligence Community analysts, collectors, and managers
to both interpret ambiguous evidence as conclusively indicative of a WMD
program as well as ignore or minimize evidence that Iraq did not have active
and expanding weapons of mass destruction programs . . ..
. . . [I]ntelligence analysts, in many cases based their analysis more on their
expectations than on an objective evaluation of the information. Analysts
expected to see evidence that Iraq had retained prohibited weapons and that
Iraq would resume prohibited WMD activities once United Nations (UN)
inspections ended . . ..
The . . . IC had a tendency to accept information which supported the pre-
sumption that Iraq had active and expanded WMD programs more readily
than information which contradicted it. . . . Information that contradicted the
IC’s presumption that Iraq had WMD programs, such as indications in the
intelligence reporting the dual-use materials were intended for conventional
or civilian programs, was often ignored . . ..
The presumption that Iraq had active WMD programs affected intelligence
collectors as well. None of the guidance given to human intelligence collectors

21. See generally Joshua Klayman, Varieties of Confirmation Bias, 32 The Psychology
of Learning and Motivation 385 (New York: Academic Press, 1995).
22. The committee report argued that the confirmation bias was caused or reinforced
by “group think.”
 biases in processing and judging information 279

suggested that collection be focused on determining whether Iraq had WMD.

Instead . . . analysts and collectors assumed that sources who denied the exis-
tence or continuation of WMD programs and stocks were either lying or not
knowledgeable about Iraq’s programs, while those sources who reported
ongoing WMD activities were seen as having provided valuable information.
. . . The Committee found no evidence that IC analysts, collectors, or man-
agers made any effort to question the fundamental assumptions that Iraq had
active and expanded programs, nor did they give serious consideration to
other possible explanations for Iraq’s failure to satisfy its WMD accounting
discrepancies, other than that it was hiding and preserving WMD. . . .
The IC’s failure to find unambiguous intelligence reporting of Iraqi WMD
activities should have encouraged analysts to question their presumption that
Iraq had WMD. Instead, analysts rationalized the lack of evidence as the result
of vigorous Iraqi denial and deception efforts to hide the WMD programs that
analysts were certain existed.

The Senate Select Committee on Intelligence (2004) identified another problem

in the Intelligence Community’s conclusion that Iraq possessed WMD: treating
multiple sources of evidence that rely on each other as independent. (Recall
People v. Collins in Section 8.2). The report asserted that “assessments were built
based on previous judgments without carrying forward the uncertainties of the
underlying judgments.” While “building an intelligence assessment primarily
using previous judgments without substantial new intelligence reporting . . . is a
legitimate and often useful analytical tool in . . . understanding complex analytical
problems, the process can lose its legitimacy when the cumulative uncertainties
of the underlying assessments are not factored into or conveyed through the new
assessments.”

Confirmation biases operate in various ways:

• You might search for evidence in a way that favors your hypothesis—for
example, by avoiding tests that might contradict your hypothesis.
• You might insufficiently revise your confidence in your prior hypothesis in
light of new data. You might be especially prone to treat conflicting data as
“exceptions that prove the rule” rather than call your hypothesis into ques-
tion. Indeed, the very process of “explaining away” the exceptions may
strengthen your belief in the rule.
• You might selectively evaluate and interpret the information you receive to
favor your hypothesis. For example, you might take hypothesis-confirming
data at face value and subject disconfirming data to critical scrutiny.
280 problem solving, decision making, and professional judgment

• You might have trouble generating viable new hypotheses, even when you
do feel like abandoning an old one.
The opportunity to seek confirming or disconfirming evidence can take place
at different phases of developing and testing hypotheses, and what is essential in
one phase may be counterproductive in another. Ryan Tweney et al. write:23
Success in a very difficult inference task requires that hypotheses be “good”
ones in the sense that they must possess at least some correspondence to real-
ity to be worthwhile candidates for the test. Thus, seeking confirmatory data
and ignoring disconfirmatory data may be the most appropriate strategy early
in the inference process, since it could then produce good hypotheses. Once
such hypotheses have been found, however, deliberate attempts ought to
be made to seek disconfirmatory data, as the strongest possible test of an
explanation.
Most of the instances of confirmation bias examined below do not involve the
development of hypotheses, but their testing.

10.3.1 Motives Underlying Confirmation Bias

Why might people be motivated to avoid disconfirming a hypothesis? Sometimes,
the hypothesis may be connected to deeply held ideological or religious beliefs.
If one has articulated a hypothesis in public, one may feel that abandoning it will
lose the esteem of others. And there are strong internal, as well as social, pres-
sures to be consistent.24 Abandoning a privately held hypothesis can be costly in
terms self-esteem; indeed one can feel that one is abandoning an important part
of one’s identity.

10.3.2 Motivated Skepticism

Even though one’s long-run interests are usually served by knowing the truth,
people are often motivated to be far less critical of facts and arguments that sup-
port their preferred result than an unpreferred one—a phenomenon that Peter
Ditto and David Lopez term “motivated skepticism.”25 Consider whether you
would be more likely to dismiss reports that caffeine consumption is harmful to
your health if you love to drink six cups of coffee a day than if you don’t drink
coffee. Motivated skepticism seems to provide a strong explanation for persistent

23. Ryan D. Tweney, Michael E. Doherty, W. J. Worner, D. B. Plieske, C. R. Mynatt,

K. A. Gross, and D. L. Arkkelin, Strategies of Rule Discovery in an Inference Task, 32
Quarterly Journal of Experimental Psychology 109 (1980).
24. See the discussion of the irrational escalation of commitment in Section 17.3.3.
25. Peter H. Ditto and David F. Lopez, Motivated Skepticism: Use of Differential Decision
Criteria for Preferred and Nonpreferred Conclusions, 63 Journal of Personality and Social
Psychology 568 (1992).
 biases in processing and judging information 281

skepticism about the effects of climate change in the face of an increasing

scientific consensus.
Ditto and Lopez conducted experiments in which two groups of participants
were informed that they would be tested for the presence of the (made-up) TAA
enzyme. One group was told that people who had the enzyme were ten times
less likely to contract a pancreatic disease; the other that they were ten times
more likely. Within each group, half the participants were asked to answer
questions before receiving the test results. One set of questions was prefaced
by the statement that certain factors, such as stress and lack of sleep, could
affect the accuracy of the test, and participants were asked to list whether they
experienced any of these factors. Other questions asked the participants to
rate the seriousness of the disease and the accuracy of the test. Participants
self-administered the test by putting saliva on a piece of paper, which, like litmus
paper, changed color depending on the outcome. Unbeknownst to the partici-
pants, of course, the outcomes were determined and their behavior was
observed.
When asked the questions in advance, there was no significant difference in
the answers to either set of questions by participants in the groups that were told
that the TAA enzyme was healthy or unhealthy. However, when asked the ques-
tion after being told the test results, the unhealthy group recalled many more
individual factors that could make the test inaccurate than the healthy group,
and they rated the test as far less accurate. Participants receiving bad news
engaged in frequent retesting—akin to weighing yourself again when the scale
reads too high—compared to those with good news, who tended to be satisfied
with a single test.
More generally, faced with preference-inconsistent information, people deny
both the facts and their implications. Ditto and Lopez suggest that people faced
with preference-consistent information are “less motivated to critically analyze
the available data . . . Preference-consistent information is accepted ‘at face
value,’ whereas preference-inconsistent information tends to trigger more exten-
sive cognitive analysis.”
Can motivated skepticism be debiased? A study by Claude Steele suggest that
a strategy of self-affirmation can make people more willing to accept negative or
unwelcome information by making them feel more secure about themselves.26
People who wrote a few paragraphs about an important personal value were more

26. The underlying assumption of the self-affirmation theory is that people are ulti-
mately motivated to see themselves as good. Claude M. Steele, The Psychology of Self-
Affirmation: Sustaining the Integrity of the Self, in L. Berkowitz (ed.,) 21 Advances in
Experimental Social Psychology 261 (New York: Academic Press, 1988). Most motivational
biases serve that purpose—a good (or intelligent) person would not drink a lot of coffee if
doing so were unhealthy, so drinking coffee must not be that unhealthy. Self-affirmation
doesn’t make people more skeptical about welcome news.
282 problem solving, decision making, and professional judgment

open to accepting negative health information (as well as opposing political

viewpoints).

10.3.3 Biased Assimilation

Charles Lord, Lee Ross, and Mark Lepper gave participants, who had identified
themselves as supporting or opposing the death penalty, two purported studies
about its deterrent effect.27 One study reported that murder rates had decreased
in eleven out of fourteen states when those states adopted the death penalty. The
other, which compared murder rates in ten pairs of states with different capital
punishment laws, reported that in eight of those pairs the murder rate was lower
in the state that did not impose the death penalty.
Most participants found the study that supported their prior view much more
convincing than the other; for example, they found serious methodological flaws
in the opposing study. While it makes sense to give one’s hypothesis the benefit
of the doubt in the face of mixed or ambiguous evidence, the participants went
beyond this. Death penalty supporters and opponents alike reported that their
prior views had been strengthened rather than weakened after reading the two
studies with opposite results. “Such results provide strong support for the
hypothesis that inconclusive or mixed data will lead to increased polarization
rather to uncertainty and moderation. Moreover, the degree of polarization . . .
was predicted by differences in participants’ willingness to be less critical of pro-
cedures yielding supportive evidence than of procedures yielding nonsupportive
evidence.” The authors make these normative observations about the results:
[T]here can be no real quarrel with a willingness to infer that studies support-
ing one’s theory-based expectations are more probative, or methodologically
superior to, studies that contradict one’s expectations. When an “objective
truth” is known or strongly assumed, then studies whose outcomes reflect
that truth may reasonably be given greater credence than studies whose out-
comes fail to reflect that truth. Hence, the physicist would be “biased,” but
appropriately so, if a new procedure for evaluating the speed of light were
accepted if it gave the “right answer” but rejected if it gave the “wrong answer.”
The same bias leads most of us to be skeptical about reports of miraculous
virgin births or herbal cures for cancer, and despite the risk that such theory-
based and experience-based skepticism may render us unable to recognize a
miraculous event when it occurs, overall we are surely well served by our
bias [28] Our participants’ willingness to impugn or defend findings as a

27. Charles G. Lord, Lee Ross, and Mark R. Lepper, Biased Assimilation and Attitude
Polarization: The Effects of Prior Theories on Subsequently Considered Evidence, 37 Journal
of Personality and Social Psychology 2098–2109 (1979).
28. Consider, in this respect, scientists’ belief in the theory of evolution. From time to
time one comes across complex functions—grist for the mill of so-called “intelligent
 biases in processing and judging information 283

function of their conformity to expectations can, in part, be similarly

defended. . . .
Our participants’ main inferential shortcoming, in other words, did not lie
in their inclination to process evidence in a biased manner. Willingness to
interpret new evidence in the light of past knowledge and experience is essen-
tial for any organism to make sense of, and respond adaptively to, its environ-
ment. Rather, their sin lay in their readiness to use evidence already processed
in a biased manner to bolster the very theory or belief that initially “justified”
the processing bias. In so doing, participants exposed themselves to the famil-
iar risk of making their hypotheses unfalsifiable—a serious risk in a domain
where it is clear that at least one party in a dispute holds a false hypothesis—
and not allowing themselves to be troubled by patterns of data that they ought
to have found troubling. Through such processes laypeople and professional
scientists alike find it all too easy to cling to impressions, beliefs, and theories
that have ceased to be compatible with the latest and best evidence available.
Because the evidence underlying most public policy reforms is seldom
unequivocal, the tendency to differentially evaluate evidence that confirms or
disconfirms one’s prior beliefs helps explain why policy issues sometimes seem
intractable. Not only laypersons but scientists as well are prone to biased assimi-
lation. A study by Jonathan Koehler29 reports on an agreement effect in which sci-
entists rated studies that agreed with their prior beliefs higher than those that
disagreed. The phenomenon was less the result of harsh criticism of disconfirm-
ing studies than leniency in evaluating studies congruent with their prior
beliefs.

10.3.4 Confirmation Bias and Positive Hypothesis Testing

Two other classic studies involve a rather different sort of confirmation bias, one
not readily explainable in the same terms as people’s attitudes toward the death
penalty. One involves a familiar game, in which the player is given a sequence of
three numbers that obeys a particular rule and is challenged to determine that
rule. For help, the player can present other sequences of three numbers and will
be told which of them obey the rule. It turns out that most people playing this
game come up with sequences that follow the rule that they guess initially. For
example, if someone is given the sequence “2 4 6” and hypothesizes that the
rule is “even numbers ascending by two’s,” he is likely to present sequences
such as “4 6 8,” which follow his hypotheses. Continuing with such sequences,

design”—that are not readily explicable by natural selection. But the theory has proved
robust in the 150 years since Darwin, and such phenomena have always ended up being
accounted for.
29. Jonathan J. Koehler, The Influence of Prior Beliefs on Scientific Judgments of Evidence
Quality, 56 Organizational Behavior and Human Decision Processes 28 (1993).
284 problem solving, decision making, and professional judgment

and continuing to hear the answer “yes,” he will never be able to prove his
hypothesis, because he will never have tested alternative hypothesis. It would be
more beneficial to try sequences like “3 5 7” or “3 6 9,” for which a “yes” would
disprove the hypothesis.30 Superstitious behaviors are often premised on positive
hypothesis testing: if you were wearing your favorite green striped socks when
you aced the exam, then you might as well wear them next time. (And if you
don’t do so well the next time, we’ve seen that there are lots of ways to explain
that away.)
The second study involves a logic problem, essentially asking people to apply
the rule, if p then q:
Imagine four cards with the symbols E, K, 4, and 7 written on them. Each card
has a number on one side and a letter on the other. I tell you: “Every card with
a vowel on one side has an even number on the other,” and ask you to test
whether this statement is true. If you were only allowed to turn over two
cards, which two cards would they be?31
When asked this question, most people choose the cards marked with E
and 4.32 In fact, this is not the best way to test whether the rule is true.
• Turning over the E card is a good idea, because finding an odd number on
the other side would disprove the rule.
• But turning over the 4 card is not informative if the rule is false: it can only
provide confirming evidence, giving us another instance in which the rule
turns out to be true. Consider the two possibilities: If there is a vowel on
the other side, it is consistent with the rule but cannot prove it. If there is a
consonant on the other side, the card is irrelevant. since the statement tells
us what to expect only for cards with a vowel on one side. Thus the 4 card
cannot give us evidence against the statement.
The correct choice is E and 7. If there is a vowel on the other side of the 7
card, or an odd number on the other side of the E card, the statement is false. If
there is a consonant with the 7 and an even number with the E, the statement
is true.
It is difficult to see in these examples the motives for confirmation bias
that were so apparent in the death penalty study. Rather, they seem to be instances
of a more cognitive, dispassionate form of bias involving the tendency to engage
in positive hypothesis testing, that is, testing hypotheses by looking for evidence
that would support rather than disconfirm them. Participants in an experiment

30. Peter C. Wason, On the Failure to Eliminate Hypotheses in a Conceptual Task, 12

Quarterly Journal of Experimental Psychology 129–40 (1960).
31. Peter C. Wason and Phillip N. Johnson-Laird, Psychology of Reasoning:
Structure and Content (Cambridge, MA: Harvard University Press, 1972).
32. Id.
 biases in processing and judging information 285

who were asked to determine whether a person was an extrovert asked questions
such as, “What would you do if you wanted to liven things up at a party?”
When asked to determine whether he was an introvert, participants asked questions
such as, “What makes you the most uneasy or anxious in social situations?”33
It turns out that people are better at hypothesis testing when solving prob-
lems that are familiar or situated in readily imaginable domains. For example,
Hubert L. Dreyfus and Stuart E. Dreyfus note that that people have less difficulty
with this problem than with the logically identical card-turning problem described
above:34
Suppose that you are a supervisor at a supermarket and have the checks
received that day stacked before you—some face up and some face down. The
supermarket has a rule: cashiers may only accept checks for $50 or more if
approved on the back by the manager. Which checks do you need to turn over
to see if the cashiers have followed the rule?35
As we discussed in the preceding chapters (see particularly the discussion of
diagnosticity in Section 8.5.3), when testing whether a particular hypothesis is
more likely than an alternative, one must consider not only the probability that
the data support the hypothesis, but also the probability that the data support the
alternative. This is negative hypothesis testing, which requires looking for data that

33. Mark Snyder and William B. Swann, Jr. Behavioral Confirmation in Social
Interaction: From Social Perception to Social Reality, 14 Journal of Experimental Social
Psychology 148 (1978). In a related experiment, participants read a story about a woman
with an equal number of behaviors that could be characterized as introverted (e.g., spend-
ing her office coffee break by herself) or extroverted (e.g., chatting with another patient in
the doctor’s office). Several days later, some were asked to assess her suitability for a ste-
reotypically introvert job (research librarian) and others for an extrovert job (real estate
salesperson). Those who evaluated her for the two positions respectively recalled her
introverted and extroverted behaviors and evaluated her accordingly. Mark Snyder and
Nancy Cantor, Testing Hypotheses about Other People: The Use of Historical Knowledge, 15
Journal of Experimental Social Psychology 330 (1979).
34. Hubert L. Dreyfus, Stuart E. Dreyfus, and Tom Anthanasiou, Mind Over
Machine: The Power of Human Intuition and Expertise in the Era of the
Computer 18 (New York: Free Press, 1986). See also Klayman, Varieties of Confirmation
Bias, supra at 403.
35. The evolutionary psychologists Leda Cosmides and John Tooby assert that people’s
ability to solve this problem, or an analogous one involving a bar’s checking IDs to assure that
drinks are not served to minors, involves a particular “cheater-detection” capacity. Jerome H.
Barkow, Leda Cosmides, and John Tooby, The Adapted Mind: Evolutionary Psychology
and the Generation of Culture (New York: Oxford University Press, 1995); Leda Cosmides
and John Tooby, Neurocognitive Adaptations Designed for Social Exchange. In Evolutionary
Psychology Handbook (David M. Buss, ed., New York: Wiley, 2005). For a critical review
of this work, see Mark G. Kelman, The Heuristics Debate: Its Nature and Its Implications
for Law and Policy (New York: Oxford University Press, forthcoming 2010).
286 problem solving, decision making, and professional judgment

would disconfirm the hypothesis. If I observe Alex eating nothing but junk food,
I could hypothesize that Alex likes to eat only junk food. Positive hypothesis test-
ing would involve giving Alex junk food to see if he likes it. Negative hypothesis
testing requires that I give Alex some healthy food to see if he might like that as
well. Note that if I try giving Alex broccoli and he doesn’t like it, that doesn’t
mean I have confirmed my hypothesis. Although broccoli is not junk food, many
people who like healthy food do not like it. The strongest test of my hypothesis
would be to give Alex a lot of different healthy foods and see if he likes any
of them.
In many situations, though, “people rely most heavily on information about
what happens when the presumed cause is present, and much less on informa-
tion about what happens in its absence.”36 They have “cognitive difficulty in
thinking about any information which is essentially negative in its conception.”37
This tendency is implicitly noted by the great psychologist Arthur Conan Doyle,
in this dialogue from The Adventure of Silver Blaze:
Inspector Gregory: “Is there any other point to which you would wish to draw
my attention?”
Holmes: “To the curious incident of the dog in the night-time.”
“The dog did nothing in the night-time.”
“That was the curious incident,” remarked Sherlock Holmes.
Positive hypothesis testing is not necessarily an irrational strategy. It tends to
uncover false positives—i.e., proffered hypotheses not supported by the data—
which often are more costly than false negatives.38 It also makes sense when you
believe that the circumstances in which rules apply arise only rarely.39

10.3.5 Confirmation Bias as an Instance of Anchoring

Confirmation bias may be explained in terms of the anchoring-as-activation
theory described above. “People tend to evaluate hypotheses by attempting to
confirm them, [and] such a search generates evidence disproportionately consis-
tent with the anchor. The absolute judgment is then biased by the evidence
recruited in this confirmatory search.”40 Also related to anchoring as activation is

36. Klayman, Varieties of Confirmation Bias, supra at 389.

37. Id.
38. See also Section 5.8.
39. Mike Oaksford and Nick Chater, Rational Explanation of the Selection Task, 103
Psychological Review 381–91 (1996).
40. Nicholas Epley and Thomas Gilovich, Putting Adjustment Back in the Anchoring and
Adjustment Heuristic, in Thomas Gilovich, Dale Griffin, and Daniel Kahneman,
Heuristics and Biases: The Psychology of Intuitive Judgment 120 (New York:
Cambridge University Press, 2002).
 biases in processing and judging information 287

the idea that a hypothesis acts as a suggestion, instilling at least a transient belief
in its validity.41 This is consistent with Spinoza’s view (contrary to Descartes’)
that understanding and believing are elided—that people at least tentatively accept
or believe propositions in order to comprehend them, and only then “unbelieve”
propositions that do not comport with their other beliefs.42

10.3.6 Belief Perseverance after Evidential Discrediting43

Richard Nisbett and Lee Ross make these general observations about the perse-
verance of beliefs, or theories (in a broad sense):44
1. When people already have a theory before encountering any genuinely pro-
bative evidence, exposure to such evidence (whether it supports the theory,
opposes the theory, or is mixed), will tend to result in more belief in the
correctness of the theory than normative dictates allow.
2. When people approach a set of evidence without a theory and then form a
theory based on initial evidence, the theory will be resistant to subsequent
evidence. . . .
3. When people formulate a theory based on some putatively probative evi-
dence and later discover that the evidence is false, the theory often survives
total discrediting of that evidence.
With respect to the last of these, Craig Anderson, Mark Lepper, and Lee Ross
gave participants a small sample of (spurious) correlations between firefighters’
indications of their risk preferences on a questionnaire and their actual perfor-
mance as firefighters, and asked the participants to write down an explanation
for the correlation.45 Some participants explained what they believed to be a pos-
itive relationship and some explained a negative relationship (both plausible and
explicable because one could imagine that firefighters either are more or are less
risk tolerant than people at large). Even after being told that the data was com-
pletely fictitious, and that other participants had provided an explanation for the
converse correlation, they held to their theories. In another experiment, Lee
Ross, Mark Lepper, and Michael Hubbard46 asked participants to distinguish
between real and fake suicide notes, and randomly informed the participants

41. See Karen E. Jacowitz and Daniel Kahneman, Measures of Anchoring in Estimation
Tasks, 21 Personality and Social Sciences Bulletin 1161 (1995).
42. See Daniel T. Gilbert, Inferential Correction, in Heuristics and Biases.
43. Richard E. Nisbett and Lee Ross, Human Inference: Strategies and
Shortcomings in Social Judgement 175 ff (Englewood Cliffs, NJ: Prentice Hall, 1980).
44. Id. 169.
45. Craig A. Anderson, Mark R. Lepper, and Lee Ross, Perseverance of Social Theories:
The Role of Explanation in the Persistence of Discredited Information, Journal of Personality
and Social Psychology 39 (1980).
46. Lee Ross, Mark R. Lepper, and Michael Hubbard, Perseverance in Self-Perception
and Social Perception: Biased Attributional Processes in the Debriefing Paradigm, 32 Journal
of Personality and Social Psychology 880 (1975).
288 problem solving, decision making, and professional judgment

that they had done well or poorly. Even after being informed that the evaluations
actually had no relationship to their performance, the participants continued to
believe that they were relatively good or bad, not only on the specific suicide note
task but on other tasks involving social sensitivity.
Nisbett and Ross suggest several reasons why people persevere in their beliefs
even after being explicitly told that they were duped: We are likely to recruit evi-
dence from our own memory consistent with our initial beliefs (“I have always
been good at guessing people’s occupations”) and continue to give it weight
because of its availability. Moreover, we are likely to have generated a causal
explanation of why the initial belief is true (“Of course I’m good at telling if sui-
cide notes are real, I’m very empathic”), and we tend to treat the ease of explana-
tion as evidence of its truth (ignoring how easy it is to develop explanations for
almost any outcome). Not only are the recruited evidence and causal theory not
discredited but, like the initial data that stimulated them, they remain available
even after the data turn out to be spurious.47

10.3.7 The Emergence of Coherence in Making Judgments

Consider this experiment by Keith Holyoak, Dan Simon, and their colleagues:
You are a judge considering a defamation action by a high-tech company against
a dissatisfied shareholder who posted a negative message about the company’s
prospects. The company was already facing financial difficulties and, shortly
after the shareholder posted the message, its stock plummeted and the company
went bankrupt. A number of factors may be relevant to your decision, including:
whether the defendant’s message was true, whether his message caused the
company’s demise, whether the defendant was motivated by vindictiveness or
only wanted to protect innocent shareholders, and whether the Internet site was
more analogous to a newspaper (which is subject to libel law) or a telephone
system (which is not). Ideally, you analyze each of these factors separately and
then combine them to reach a judgment.
In fact, your judgment not only tends to evolve in favor of one or the other
party as you consider the factors, but your assessment of each of these discrete
factors tends to fall into line with that judgment. If you’re heading in the direc-
tion of a judgment for the company, you will tend to conclude that the message
was false and vindictive, contributed to the company’s demise, etc. Holyoak et al.

47. In a follow-up study, an observer, who had no knowledge of the experimental

manipulation, watched the actor performing the task. The actor was then debriefed in a
procedure, watched by the observer, that explicitly discussed the perseverance phenome-
non. Although pointed debriefing attenuated the actor’s continued beliefs about her abili-
ties, the observers’ views of the actor’s abilities tended to persevere. It is less clear why
observers persevere in holding to their theories. Lee Ross suggests that perhaps the
observers care less and do shallower processing and hence they continue to anchor on
their initial impression of the actor’s ability. (Personal communication.)
 biases in processing and judging information 289

characterize this process as “bidirectional reasoning,”48 capturing the idea that

you not only reason from the premises to the conclusion but from the (nascent)
conclusion to the premises, valuing evidence that supports the conclusion over
evidence that does not. The elements of a schema or story can be related to each
other in a network of connections,49 and when particular nodes on the network
(e.g., vindictiveness) are activated, they tend to activate some (e.g., causation of
the harm) and deactivate others (e.g., truth of the statement). The process is akin
to fitting the elements of a story together to make the “best story” not when it is
over, but as you are hearing it.
The experiments by Holyoak et al. produced two other noteworthy results.
First, once participants had reached a particular result, they had difficultly recall-
ing contrary inferences they had made prior to the decision—a phenomenon
related to hindsight bias. Second, the basic scenario was designed not to favor
one outcome over the other and, indeed, the participants divided roughly equally
in finding for the plaintiff or the defendant. Yet participants reported very high
levels of confidence that they had reached the best possible decision.

10.3.8 Debiasing Strategies

Can one overcome the tendencies toward confirmation bias? Charles Lord, Mark
Lepper, and Elizabeth Preston50 found that the instruction to “be objective” had
virtually no effect on people’s tendency to assimilate new information in a biased
manner. But instructing people to consider the opposite—to seriously entertain
the possibility that the opposite of what they believe might actually be true—had
a significant effect. This suggests that people are not necessarily trying to see
things in a way that suits them; rather, their cognitive processes lead them to
interpret new information in the light of existing beliefs.

10.4 overconfidence

Edward Russo and Paul Schoemaker asked 1000 American and European busi-
ness managers to answer ten trivia questions calling for numerical estimates

48. Dan Simon, Lien B. Pham, Quang A. Le and Keith J. Holyoak, Bidirectional
Reasoning in Decision Making by Constraint Satisfaction, Journal of Experimental
Psychology—General 3 128 (1999); Dan Simon, Lien B. Pham, Quang A. Le, and Keith
J. Holyoak, The Emergence of Coherence over the Course of Decision Making, 27 Journal of
Experimental Psychology—Learning, Memory, and Cognition 1250 (2001).
49. See the discussion of connectionist networks in Section 11.3.2.
50. Charles G. Lord, Mark R. Lepper, and Elizabeth Preston, Considering the Opposite:
A Corrective Strategy For Social Judgment. 47 Journal of Personality and Social
Psychology 1231 (1984).
290 problem solving, decision making, and professional judgment

and to provide a “low and a high guess such that you are 90% sure the correct
answer falls between the two.” The questions included.51
• Martin Luther King’s age at death.
• Length of the Nile River in miles.
• Number of books in the Old Testament.
• Diameter of the moon in miles.
• Air distance from London to Tokyo.
Most respondents got considerably more than 10 percent of their answers
wrong. What is interesting about Russo and Schoemaker’s experiment is not the
state of the participants’ substantive knowledge, but their belief that they knew
more than they did. Indeed, not only amateurs answering trivia questions but
professional making judgments in their own domains of expertise tend to be
overconfident. Also noteworthy: although people are overconfident about par-
ticular answers, they are much more realistic in estimating their overall hit rate.52
Unfortunately, most judgments are made one at a time.
People tend to be overconfident about nontechnical items of moderate or
extreme difficulty.53 For example, participants given the nearly impossible task of
discriminating between Asian and European children’s drawings or predicting
fluctuations in stock prices did little if any better than chance (i.e., 50 percent
correct). But their mean confidence levels were between 70 percent to 80 per-
cent.54 Moreover, people become more confident as they are given more infor-
mation, even if the information is not diagnostic.55 Overconfidence diminishes
as the tasks become easier. Indeed, knowledgeable participants responding to
easy questions tend to be underconfident.
Granted that people often tend to be overconfident, is there a correlation
between confidence and accuracy? Alas, not. For example, in a study by Scott
Plous and Philip Zimbardo, citizens were given a description of actual Soviet
and American military actions, with the names of the countries removed. They
were asked to identify which of the superpowers was involved and to say how

51. J. Edward Russo and Paul J. H. Schoemaker, Winning Decisions: Getting it

Right the First Time 79–80 (New York: Doubleday, 2002).
52. Dale Griffin and Amos Tversky, The Weighing of Evidence and the Determinants of
Confidence, in Heuristics and Biases, 411.
53. Sarah Lichtenstein, Baruch Fischhoff, and Lawrence Phillips, Calibration
of Probabilities: The State of the Art to 1980, in Judgment Under Uncertainty: Heuristics
and Biases 306 (Daniel Kahneman and Amos Tversky eds., Cambridge University Press,
1982).
54. Sarah Lichtenstein and Baruch Fischhoff, Do Those Who Know More Also Know
More About How Much They Know? The Calibration of Probability Judgements, 3
Organizational Behavior and Human Performance 553 (1977).
55. Stuart Oskamp, Overconfidence in Case-Study Judgments, 29 Journal of Consulting
Psychology 261 (1965).
 biases in processing and judging information 291

confident they were in their answers. The respondents were correct slightly less
than half the time, and their confidence ratings were almost orthogonal to the
accuracy of their responses.56
The relationship between individuals’ assessment of the probability of their
beliefs being true and the actual truth is termed calibration. Meteorology is
among the few professions—along with bookmaking on horse races—in which
judgments prove to be well calibrated. By contrast, the calibration of physicians’
diagnosis of pneumonia in patients examined because of a cough, and their
diagnoses of skull fractures, has proven to be poor. Sarah Lichtenstein, Baruch
Fischhoff, and Lawrence Phillips conjecture:
Several factors favor the weather forecasters. First, they have been making
probabilistic forecasts for years. Second, the task is repetitive; the question to
be answered (Will it rain?) is always the same. In contrast, a practicing physi-
cian is hour by hour considering a wide array of possibilities (Is it a skull
fracture? Does she have strep? Does he need further hospitalization?). Finally,
and perhaps most important, the outcome feedback for weather forecasters is
well defined and promptly received. This is not always true for physicians;
patients fail to return or are referred elsewhere, or diagnoses remain
uncertain.57
In many situations, overconfidence seems to result from the availability and
anchoring and adjustment phenomena. Dale Griffin and Amos Tversky distin-
guish between the strength (extremity on the relevant predictive dimension) of
evidence and its weight (predictive value).58 In statistical terms, strength is analo-
gous to the size of the effect, and weight is analogous to its significance or
reliability (taking into account, for example, the size of the sample). They argue
that people focus on the perceived strength of the evidence based on the degree
to which available evidence is consistent with the hypothesis in question—and
then adjust (insufficiently) for its weight. For example, when judging whether
a coin is biased, people first focus on the proportion of heads and tails in the
sample and only then (insufficiently) adjust for the number of tosses. By the
same token, the fact that a suspect fails a lie detector test is treated as strong
evidence of guilt notwithstanding the poor reliability of such tests.59

56. Scott L. Plous and Philip G. Zimbardo, How Social Science Can Reduce Terrorism
(2004), Chronicle of Higher Education, Sept. 10, 2004, B9-10.
57. Sarah Lichtenstein, Baruch Fischhoff, and Lawrence D. Phillips, Calibration of
Probabilities: The State of the Art to 1980, in Daniel Kahneman, Paul Slovic, and Amos
Tversky, Judgment Under Uncertainty: Heuristics and Biases 306 (1982).
58. Dale Griffin and Amos Tversky, The Weighing of Evidence and Determinants of
Confidence, in Heuristics and Biases, 230.
59. Lyle A. Brenner, Derek J. Koehler, and Yuval Rottenstreich, Remarks on Support
Theory: Recent Advances and Future Directions, in Heuristics and Biases.
292 problem solving, decision making, and professional judgment

The tendency (discussed in Section 8.5.4) to ignore base rates and to over-
value the more immediate and available data also leads to overconfidence. For
example, after interviewing individuals, participants were asked to predict
whether, if given the choice, the interviewees would take a free subscription to
Playboy or The New York Review of Books. Although informed that 68 percent of
all the people interviewed preferred Playboy, participants tended to ignore this
base rate and premise their predictions on their own impressions from the inter-
views. Participants whose predictions went contrary to the base rate were over-
confident, with a mean accuracy of only about 50 percent in the face of a mean
confidence of 72 percent. When they just relied on the base rate they were quite
well calibrated.
Although people tend to be overconfident about their knowledge or predic-
tions of particular items or events, they tend to be better calibrated in terms of
their overall judgment about similar events. For example, entrepreneurs were
overconfident about the success of their own new ventures even when they were
realistic about the general failure rate for such ventures. Griffin and Tversky sug-
gest that the entrepreneurs ignore base rates and focus on factors pertaining to
the individual decision.60 Similarly, “people often make confident predictions
about individual cases on the basis of fallible data (e.g., personal interviews or
projective tests) even when they know that these data have low predictive
validity.”61
Are there individual characteristics that correlate with calibration? While men
and women were equally represented in the Plous and Zimbardo study described
above, two-thirds of the highly confident respondents were male; and about twice
as many of the highly confident respondents distrusted the Soviet Union and
were advocates of strong defense spending than their low-confidence peers.
However, in a different study of relationships between overconfidence and sev-
eral personality measures—authoritarianism, conservatism, dogmatism, and
intolerance of ambiguity—only authoritarianism showed a (modest) correlation
with overconfidence.62 The nature of the task seems to matter more than the
individual who is performing it.63
As for debiasing, merely alerting participants to the difficulty of the task does
not improve calibration. Nor does increasing the motivation to calibrate
accurately. Paralleling a strategy for correcting self-serving biases, however,

60. We address overconfidence about one’s future projects, or overoptimism, in

Section 13.5.
61. Dale Griffin and Amos Tversky, The Weighing of Evidence and the Determinants of
Confidence, in Heuristics and Biases.
62. Sarah Lichtenstein, Baruch Fischhoff, and Lawrence D. Phillips, Calibration of
Probabilities: The State of the Art to 1980, in Daniel Kahneman, Paul Slovic, and Amos
Tversky, Judgment Under Uncertainty: Heuristics and Biases 306 (1982).
63. Id.
 biases in processing and judging information 293

calibration did improve when participants were asked to write down all the rea-
sons that contradicted their answers.64 But the most powerful strategy for improv-
ing calibration is clear and immediate feedback.
Overcoming Intuition. People’s overconfidence increases to the extent their
judgments rely on intuition. Could overconfidence be overcome by “forcing” the
brain to think more deliberately? A 2007 finding by Adam Alter, Daniel
Oppenheimer, and their colleagues indicates that intuition (and the associated
overconfidence) can be overcome when faced with “metacognitive difficulty”—-
that is, when presented with information that makes them think harder about
their intuitive assumptions and judgments. The authors postulate that cognitive
disfluency (which occurs when information pertaining to a decision is difficult
to process) can compel individuals to switch from System 1 cognitive processes
(intuition) to System 2 cognitive processes (deliberation).65 In other words, cog-
nitive disfluency makes individuals “less confident” in their intuitive judgments
and forces them to switch into a more deliberative thinking mode.

10.5 intuition, analysis, and the role of experts redux

In Chapter 1, we contrasted intuitive and analytic (or deliberative) approaches to

problems. We considered their strengths and weaknesses, and their often com-
plementary roles. The preceding chapters on statistics studied a formal analytical
approach to making judgments based on probabilistic information. Chapters 9
and 10 have focused on the role of intuition, highlighting the systematic biases
of the intuitive statistician.
We conclude the chapter with two questions that relate to the distinction
between intuition and analysis. One concerns the relative merits of clinical and
statistical methods for predicting human behavior and other events. As we will
see, clinical intuitions do not fare very well. And this leads us to the second
question: What good are experts anyway?

10.5.1 Clinical vs. Statistical Prediction

It’s tough to make predictions, especially about the future

—Yogi Berra

64. Asher Koriat, Sarah Lichtenstein, and Baruch Fischhoff, Reasons for Confidence,
6 Journal of Experimental Psychology: Human Learning and Memory 107 (1980).
65. Adam L. Alter, Daniel M. Oppenheimer, Nicholas Epley, and Rebecca N. Eyre,
Overcoming Intuition: Metacognitive Difficulty Activates Analytic Reasoning, Journal of
Experimental Psychology, General 136.4 569–76 (2007).
294 problem solving, decision making, and professional judgment

Put yourself in the position of a Christine Lamm, who is considering law

school graduates for entry-level positions in her department. Your task is to pre-
dict whether the candidate is likely to develop the requisite professionals skills
(e.g., legal and policy analysis, writing) to succeed in the job. Criteria that you
might employ for this task include:
• law school GPA;
• types of law school courses taken;
• extracurricular activities;
• faculty recommendations;
• LSAT score.
Which of these strike you as most predictive—and why? Suppose that you
could pick and choose among the criteria and give them any weights you choose.
How well could you do? Suppose that the department has ten entry-level posi-
tions and that you are considering twenty applicants. How many of the ten that
you hire are likely to succeed? How many of the ten you reject would have suc-
ceeded if they had been hired (and why would you have particular difficulties in
answering this question)?
How could you improve your prediction? Consider these two alternatives:
1. Statistical: Decide which of these criteria matter and in which direction
they matter (e.g., higher GPA predicts greater success). Score them on
some common scale and add up the scores.
2. Clinical: In addition to or instead of scoring, give considerable weight to the
intangible factors that strike you from interviewing the candidate.
Both of these have elements of intuition. The first involves a rough multivari-
ate regression analysis; it relies on your intuitions about the diagnosticity or
predictive value of the relevant factors, all of which can easily be “coded.” The
second complements or replaces these codable criteria with noncodable
intuitions.
Where clinical judgments are usually accurate—e.g., identifying a person
who comes into your office as male or female—there is no need to use statistical
indicators. And when the phenomena to be predicted are essentially random—
e.g., daily variation in stock prices—neither clinical nor statistical methods are
helpful. In most other cases, however, the statistical approach alone usually
yields better predictions than the clinical approach:
• even (perhaps especially) if you only average the coded scores rather than
weight them according to some regression model;
• even if you are an expert in the field;
• even if, as an expert, you have the statistical prediction in hand before
making your own clinical prediction.
This is the implication of about one hundred experiments involving
phenomena ranging from medical and psychiatric diagnoses, to predicting
 biases in processing and judging information 295

parole violations, loan defaults and bankruptcies, and the academic success of
graduate students.66 For example, while statistical prediction of parole failures
based on three variables (type of offense, number of past convictions, and
number of violations of prison rules) produced a (very modest) 0.22 correlation
with actual parole failures, this far surpassed professional parole interviewers,
whose predictions had a correlation of only 0.06.67 In the Cook County Hospital
emergency room, the use of four objective factors did a far better job than clinical
diagnosis in predicting whether a patient with chest pain was in danger of suf-
fering a heart attack.68
Baseball offers an interesting example of the difference between clinical and
statistical decision making.69 On the clinical side, the scout, uses his intuition to
recruit players based on speed, quickness, arm strength, hitting ability, and
mental toughness, and qualities such as “a live, active lower body, quick feet,
agility, instinct, . . . [and] alertness.” To the extent statistics play a role in evaluat-
ing a hitter’s potential contribution to a team’s success, they traditionally have
centered around batting average, runs batted in, and home runs.
Bill James, who published the Baseball Abstract, found that these few statistics
did not account for a player’s contribution to the team’s winning. He began to
create new statistics that were the product of a sophisticated analysis of hun-
dreds of games, with the ultimate measure of a hitter’s success based on how
many runs he creates.
Billy Beane, general manager of the Oakland Athletics, began to experiment
with James’s approach. He discovered that on-base percentage (the percentage
of the time a batter reached base by any means) and slugging percentage (total
bases divided by at-bats) were the two best predictors of success, and on-base
percentage was three times as important as slugging percentage.
Beane thus selected players based upon an entirely different model than his
clinically-minded competitors. Just as clinicians generally resist statistical pre-
diction, the scouts were reluctant to yield their intuitions. In Moneyball, Michael
Lewis recounts a paradigmatic conversation between Beane and a recruiter:
“The guy’s an athlete, Billy,” the old scout says. “There’s a lot of upside
there.”

66. Robyn M. Dawes, David Faust, and Paul E. Meehl, Clinical Versus Actuarial
Judgment, in Heuristics and Biases; Robyn M. Dawes, House of Cards: Psychology
and Psychotherapy Built on Myth (New York: The Free Press, 2002).
67. See Robyn Dawes, House of Cards, supra.
68. The studies are summarized in Malcolm Gladwell, Blink 125–36 (2005).
69. See Michael Lewis, Moneyball: The Art of Winning an Unfair Game (New York:
Norton & Company, 2003); Cass Sunstein and Richard Thaler, Who’s on First, New
Republic (August 2003); Ehren Wassermann, Daniel R. Czech, Matthew, J. Wilson, and
A. Barry Joyner, An Examination of the Moneyball Theory: A Baseball Statistical Analysis
Abstract, 8 The Sports Journal (Winter 2005), http://www.thesportjournal.org/
2005Journal/Vol8-No1/daniel_czech.asp.
296 problem solving, decision making, and professional judgment

“He can’t hit,” says Billy.

“He’s not that bad a hitter,” says the old scout.
“Yeah, what happens when he doesn’t know a fastball is coming?” says Billy.
“He’s a tools guy,” says the old scout. . . .
“But can he hit?” asks Billy.
“He can hit,” says the old scout, unconvincingly.
[The scout] reads the player’s college batting statistics. They contain a
conspicuous lack of extra base hits and walks.
“My only question is,” says Billy, “if he’s that good a hitter why doesn’t he hit
better?”
Beane’s basic point was that if you want to predict future achievement, you
will do better to rely on past achievements than on unrealized potential. By the
same token, overachievers are more likely than underachievers to overachieve in
the future. In any event, Beane was often able to acquire players who were under-
valued by other teams. Consequently, the Athletics, with a much lower payroll
than other teams, consistently made the playoffs after Beane began implement-
ing his system of evaluating talent.70
In the large majority of the preceding examples, neither statistical nor clinical
prediction does very well, but statistical prediction almost always does better.
Why?
First, linear models tend to be quite robust in explaining or predicting a
wide range of phenomena. It turns out that simple linear models based on aver-
aging a number of independent variables, where increases in the independent
variables predict a monotonic increase the dependent variable, tend to be pretty
good approximations of empirically-derived models.71 By contrast, people are not
very good at combining or aggregating data. William Grove and Paul Meehl
write.72

70. Another proponent of Beane’s method of evaluating talent, Boston Red Sox
General Manager Theo Epstein, saw his team win the World Series for the first time in
eighty-eight years, albeit with a significantly higher payroll than Beane’s As. It should be
noted that the claim that statistics are more valuable than informed intuition in recruiting
and deploying a baseball team has met with some vehement resistance. David Leonhardt,
To Play Is the Thing, New York Times, Aug. 28, 2005, reviewing Buzz Bissinger, Three
Nights in August (2005), and Bill Shanks, Scout’s Honor (2005).
71. See Robyn M. Dawes and Bernhard Corrigan, Linear Models in Decision Making, 81
Psychology Bulletin 95 (1974). It is questionable whether one can improve on a simple
linear model through the process of bootstrapping—of developing and implementing a
regression equation based on observing an expert decision maker. Id.
72. William M. Grove and Paul E. Meehl, Comparative Efficiency of Informal (Subjective,
Impressionistic) and Formal (Mechanical, Algorithmic) Prediction Procedures—The Clinical
Statistical Controversy, 2 Psychology, Public Policy, and Law 293 (1996).
 biases in processing and judging information 297

The human brain is a relatively inefficient device for noticing, selecting,

categorizing, recording, retaining, retrieving, and manipulating information
for inferential purposes. . . . The dazzling achievements of Western post-
Galilean science are not attributable to our having better brains than Aristotle
or Aquinas, but to the scientific method of accumulating objective knowledge. . . .
However, we need not look to science for the basic point, as it holds . . . in
most areas of daily life. . . . When you check out at the supermarket, you don’t
eyeball the heap of purchases and say to the clerk, “Well it looks to me as if it’s
about $17.00 worth; what do you think?” The clerk adds it up.
Second, clinical judgments tend to be affected by extraneous factors, includ-
ing most of the biases described in Chapters 9 and 10. One might analogize
statistical versus clinical judgments to flying by instruments versus by flying by
sight and feel. At least before the advent of modern electronic navigation, flying
by instruments, though very accurate, was always subject to some error—the
plane had leeway within narrow corridors—but errors tend to be regressive, clus-
tering around the mean. Visual flying is highly accurate when conditions are
good and the pilot is alert. But put even an experienced pilot in the clouds with-
out any instruments, and his intuitions are subject to wild divergences from
reality: he may, for example, believe that the plane is turning or descending
when it’s straight and level. Such errors are nonregressive and can, indeed, be
fatally extreme. Instruments can fail, and sometimes one’s intuitions can warn
of their “statistical” anomalies; but more often what feels like an anomaly is actu-
ally a failure of the intuitive system.
Third, clinical judgments often manifest inconsistency. In one study, radiolo-
gists were asked to diagnose stomach cancer in one hundred patients. A week
later, when they were given the same X-rays in a different order, almost a quarter
of the diagnoses were different.73 As Lewis R. Goldberg writes:74
[T]he clinician . . . lacks a machine’s reliability. He “has his days.” Boredom,
fatigue, illness, situational and interpersonal distractions, all plague him,
with the result that his repeated judgments of the exact same stimulus con-
figuration are not identical. . . . If we could remove some of this human unre-
liability by eliminating the random error in his judgments, we should thereby
increase the validity of the resulting predictions.
These findings have been met with considerable resistance on two grounds.
First, especially when making selections among candidates—for admission to

73. Paul J. Hoffman, Paul Slovic, and Leonard G. Rorer, An Analysis-of-Variance Model
for Assessment of Configural Cue Utilization in Clinical Judgment, 69 Psychological
Bulletin 338 (1968).
74. Lewis R. Goldberg, Man Versus the Model of Man: A Rationale, Plus Some Evidence,
for a Method of Improving on Clinical Inferences, 73 Psychology Bulletin 422 (1970).
298 problem solving, decision making, and professional judgment

universities or for employment—the use of purely statistical methods has been

criticized as cold, inhumane, or unfair. But it is hardly clear that fairness is
improved by the arbitrariness of subjective impressions.

Clinical vs. Statistical Prediction in Affirmative Action

Admissions Criteria

In its 1993 decisions in Gratz v. Bollinger and Grutter v. Bollinger, the

Supreme Court struck down the University of Michigan’s affirmative
action program for undergraduate admissions because its actuarial for-
mula for assigning weight to diversity “lacked a meaningful individualized
review of applicants,” but upheld the law school admissions program
because it involved a “highly individualized, holistic review of each appli-
cant’s file,” where race was considered among other factors but not in a
mechanical way. What are the pros and cons of the Court’s approach? (See
William C. McGaghie and Clarence D. Kreiter, Holistic Versus Actuarial
Student Selection, 17 Teaching and Learning in Medicine 89 (2005)).

Second, the general superiority of statistical over clinical prediction raises the
question—of ego as well as fact—of what use are experts? We now turn to this
question.

10.5.2 The Role of Experts

No one doubts that an expert has knowledge or skills in a domain that laypersons
or even students aspiring to be experts do not possess. For most of us, our ability
to diagnose medical symptoms, to represent a client in a trial, or to play chess,
basketball, or the cello does not compare favorably to those accomplished in
those activities.
Also, because (as discussed in Section 1.5) experts often engage in schematic
processing—interpreting, judging, and acting based on pattern recognition—
they can often respond much more rapidly and confidently than someone who
must engage in a deliberative process. Consider this example from Gary Klein’s
study of firefighters:75

75. Gary Klein, Sources of Power: How People Make Decisions 32 (Cambridge,
MA: MIT Press, 1998).
 biases in processing and judging information 299

It is a simple fire in a one-story home in a residential neighborhood. The fire

is in the back, in the kitchen area. The lieutenant leads his house crew into
the building, to the back, to spray water on the fire, but the fire just roars back
at them.
“Odd,” he thinks. The water should have more of an impact. They try dous-
ing it again, and get the same results. They retreat a few steps to regroup.
Then, the lieutenant starts to feel as if something is not right. He doesn’t
have any clues; he just doesn’t feel right about being in that house, so he
orders his men out of the building—a perfectly standard building with noth-
ing out of the ordinary.
As soon as his men leave the building, the floor where they had been stand-
ing collapses. Had they still been there, they would have plunged into the fire
below.
When later interviewed by Klein, the lieutenant said that he had no clue that
the house had a basement, let alone that the fire was down there, but he won-
dered why the fire was not responding. In retrospect, the room in which the
firefighters were grouped was unusually warm and, while fires are generally
noisy, this one was very quiet. It was the fact that the fire didn’t fit any patterns
known to him that sent an unconscious alarm signal. It was the dog that did not
bark in the nighttime.
Along similar lines, recall the soldier in Section 1.5.3 who sensed that a parked
car was next to an improved explosive device:
One thing did not quite fit on the morning of Sergeant Tierney’s patrol in
Mosul. The nine soldiers left the police station around 9 a.m., but they did not
get their usual greeting. No one shot at them or fired a rocket-propelled
grenade. Minutes passed, and nothing. . . . Since then, Sergeant Tierney has
often run back the tape in his head, looking for the detail that tipped him off.
Maybe it was the angle of the car, or the location; maybe the absence of an
attack, the sleepiness in the market: perhaps the sum of all of the above. . . .
“I can’t point to one thing,” he said. “I just had that feeling you have when you
walk out of the house and know you forgot something—you got your keys, it’s
not that—and need a few moments to figure out what it is.”76
Though more dramatic than most, these examples are representative of
experts’ intuitions. Yet there is considerable evidence that experts fall prey to
many of the biases described above, and that their judgments are quite fallible on
important dimensions. Here is a list of concerns and some responses to them:

76. Benedict Carey, In Battle, Hunches Prove to be Valuable, New York Times, July 28,
2009, http://www.nytimes.com/2009/07/28/health/research/28brain.html?pagewanted=
2&_r=1&hp.
300 problem solving, decision making, and professional judgment

• Validity. Expert judgments in many fields are often just wrong. (Compared
to what, though? Expert judgments are usually more accurate than those of
laypersons.)
• Reliability. Given exactly the same fact situation an hour or a week later, an
expert will often come to a different judgment. A review of studies of reli-
ability found an overall correlation of 0.79 between the first judgment and
a subsequent one—though meteorologists scored 0.91.77
• Calibration. Experts’ certainty about their judgments tends to be poorly
calibrated with their validity, with meteorologists again being an excep-
tion.78 (On the other hand, there is some evidence that laypersons are
attracted to overconfident experts.79)
• Under-use of available relevant information. While one of the advantages of
expertise is the ability to exclude extraneous information, experts often do
not seek or use information that is relevant and diagnostic. One possible
explanation, related to schematic processing, is that an expert’s “feeling of
knowing”—akin to the “strong sense of general familiarity” that can mis-
lead eyewitness identification80—may demotivate further inquiry. When
an expert “must choose either to retrieve a previously computed solution or
to compute the solution anew, she may judge the likelihood that she pos-
sesses the correct answer in memory by assessing her feeling of knowing
it.”81 While the expert may often be correct, overconfidence in one’s initial
judgment can readily lead to short-circuiting any further learning.
• Combining indicators. As we saw in the preceding section, experts’ intu-
itions are not as reliable or consistent as simple linear models in combin-
ing or aggregating indicators to predict outcomes. Nonetheless, experts are
an important source for determining what indicators are relevant and for
discarding supposed indicators that are not. But more fundamentally,
many situations that experts deal with—situations like the kitchen fire—do
not have the similar, repetitive quality that allow for the application of a
linear model.82

77. Robert H. Ashton, A Review and Analysis of Research on the Test-Retest Reliability of
Professional Judgment, 13 Journal of Behavioral Decision Making 277 (2000).
78. Stuart Oskamp, Overconfidence in Case Study Judgments, 29 Journal of Consulting
Psychology 261–65 (1965).
79. Paul C. Price and Eric R. Stone, Intuitive Evaluation of Likelihood Judgment Producers:
Evidence of a Confidence Heuristic, 17 Journal of Behavioral Decision Making 39 (2004).
80. Daniel L. Schacter, The Seven Sins of Memory: How the Mind Forgets and
Remembers (New York: Houghton Mifflin, 2001).
81. Stacy L. Wood and John G. Lynch, Jr., Prior Knowledge and Complacency in New
Product Learning, 29 Journal of Consumer Research 416 (2002).
82. On the other hand, James Shanteau suggests that expert judgments are most reli-
able when dealing with relatively repetitive tasks with static stimuli. James Shanteau,
 biases in processing and judging information 301

• Expert intuitions as a safety valve. Even if purely statistical judgments are

generally more accurate than clinical judgments, statistical judgments are
susceptible to at least two sorts of errors. They cannot take account of
unusual individuating data, and an error in recording or transcription (e.g.,
the misplacement of a decimal point or the transposition of two integers in
a number) can lead to a grievously wrong result.83 On the other hand,
experts probably tend to overrate their ability to identify when statistical
judgments have gone off the rails.84
Although intuitive decision making often compares unfavorably with decision
making based on actuarial models, actuarial models can sometimes be developed
by observing experts making decisions to see what criteria they use and then test-
ing the criteria in terms of their predictive accuracy. This approach, known as
bootstrapping,85 often does better than the experts on whose judgments it is
based—in areas ranging from graduate school admission to predicting the life
expectancy of cancer patients to the performance of insurance salespersons.86

10.6 conclusion

Nisbett and Ross report that a colleague, after reading a draft of their manu-
script of Human Inference, asked: “If we’re so dumb, how come we made it
to the moon?”87 Their response is that advances in the physical and natures

Competence in Experts: The Role of Task Characteristics, 53 Organizational Behavior and

Human Decision Processes 252 (1992).
83. Contrasting intuitive and analytic judgment more generally, Hammond reports on
an experiment in which participants estimating height of a bar with their eye were never
quite right but never far out of range, while those using trigonometry were precisely
right—except when they made errors in calculation, which produced absurd results.
Kenneth R. Hammond, Human Judgment and Social Policy: Irreducible
Uncertainty, Inevitable Error, Unavoidable Injustice 160, 175 (New York: Oxford
University Press 1996). He makes the related point that while “rigorous, analytically
derived systems . . . work well . . . within the controlled circumstances for which they were
constructed,” intuition tends to be more robust, continuing to operate when those systems
unexpectedly break down.” For example, the lighthouse is infallible until it is not lit some
evening.
84. See Dawes et al., supra.
85. John A. Swets, Robyn M. Dawes, and John Monahan, Better Decisions Through
Science, Scientific American 82 (October 2000); Colin Camerer, General Conditions for
the Success of Bootstrapping Models, 27 Organizational Behavior and Human
Performance 411 (1981); J. Edward Russo and Paul J. H. Schoemaker, Winning
Decisions: Getting It Right the First Time 146 (2002).
86. Russo and Schoemaker, 147.
87. Nisbett and Ross, 249.
302 problem solving, decision making, and professional judgment

sciences and technology are ultimately due to formal research methodology and
principles of inference, and are the result of collective rather than individual
learning. For example, “the individual is not usually required to detect covaria-
tion anew,” but can learn from experts or through cultural transmission:
Each culture has experts, people of unusual acumen or specialized knowl-
edge, who detect covariations and report them to the culture at large. Thus,
most (though not by any means all) cultures recognize the covariation between
intercourse and pregnancy. This is a covariation detection task of enormous
complexity, given the interval between the two types of events and the rarity
of pregnancy relative to the frequency of intercourse, not to mention the occa-
sional unreliability of the data (“Honest, I never did it!”). . . . [O]nce a covaria-
tion is detected and the new hypothesis seems to be confirmed by perusing
available data, the entire culture is the beneficiary and may take action in
accordance with the knowledge. Such a cultural transmission principle
applies to almost all human affairs from farming (“Plant corn when the oak
leaf is as big as a mouse’s ear”) to tourism (“The German restaurants in
Minnesota are generally quite good”) to urban survival (“The South Side is
unsafe.”)”88
Although cultures can embrace and transmit false knowledge, adaptive
pressures tend to correct misinformation that has practical consequences.

88. Id. 111.

11. the social perceiver
Processes and Problems in Social Cognition

11.1 introduction

Many of the decisions that lawyers, judges, and policy professionals make involve
interpreting and predicting other people’s behavior. Indeed, making sense of
other people figures centrally in the performance of virtually every aspect of law-
yers’, judges’, and public officials’ work. Lawyers and other “forensic profession-
als” functioning as investigators, or judges acting in their roles as fact-finders,
for example, have to assess credibility and attribute motivation. As counselors,
lawyers must understand what makes their clients “tick,” win their clients’ trust,
and predict how their clients will feel and behave in the future. In evaluating
specific cases and designing advocacy strategies, lawyers have to predict how
legal decision makers, like judges, administrative officials, or jurors, will respond
to specific legal or factual arguments. What arguments will they find persuasive?
What narratives will they find compelling? What inferences will they draw from
ambiguous evidence susceptible of varying interpretations? In negotiating an
agreement or designing a process, lawyers have to structure incentives that will
encourage certain behaviors and discourage others, and this requires under-
standing a great deal about human behavior. Policy makers have analogous tasks
in dealing with constituents, legislatures, and other decision-making bodies and
in regulating or otherwise influencing the behavior of individuals and entities.
Errors of judgment in any of these, and many similar contexts, can lead to bad
decisions and ineffective policies. For this reason, understanding the sources of
bias and error that can distort social perception and judgment, and knowing how
to structure decision making so as to minimize their negative effects, will
improve the quality of a lawyer or policy maker’s performance in virtually every
aspect of his or her professional role.
Social judgment, like judgment more generally, is influenced by systematic
biases and other sources of distortion similar to those described in earlier chap-
ters. Biases in social judgment, however, are in some ways the most insidious of
all. Because social interaction constitutes so much of the grist of daily profes-
sional (and personal) life, we tend to think of ourselves as experts in understand-
ing other people and predicting their future behavior. Our self-confidence as
“common-sense” or “intuitive” psychologists often leads us to think that we
understand other people better than we do.
In this chapter, we consider how people go about making sense of other
people. In the process, we introduce readers to various sources of bias and distor-
tion in social perception and judgment, describe how “intuitive” psychological
304 problem solving, decision making, and professional judgment

theories can lead us astray in our attempts to understand and predict people’s
behavior, and make some suggestions as to what might be done about it.
The psychological processes we explore in this chapter, particularly those
having to do with racial, gender, or other forms of stereotyping and bias, raise a
number of highly controversial legal and policy issues. You have probably already
been exposed to at least parts of the large and at times contentious literature
engaging many of the controversies, whether you are studying law, public policy,
or some other discipline. It is not our purpose in this chapter to rehearse, to
weigh in on, or even to summarize, these controversies. Rather, we seek only to
describe a number of the major—and for the most part, consensus—findings in
empirical social and cognitive psychology about various psychological processes
that can conduce to intergroup misperception and perpetuate subtle—often
unintentional—forms of intergroup bias.

11.2 the biasing effects of social schemas: an overview

In Section 1.5.2, we introduced the role of schemas in organizing the vast

amounts of data that bombard our senses every moment of the day. Schemas
create expectancies, and these expectancies determine which perceptions we
attend to, how we interpret them, and how we store them in memory. The sche-
matic filters through which we make sense of incoming perceptions also influ-
ence the way we behave toward the people we perceive.
Racial, ethnic, and gender stereotypes represent one class of social schemas
that we absorb from the surrounding cultural environment. But stereotypes are
not unique in their ability to create subtle social expectancies. Person prototypes
(i.e., “my sister, Susan”) and role prototypes (i.e., “reference librarian”) function
in much the same way.
The content of our social schemas, and the expectancies those schemas gen-
erate, are influenced by culture and learning, by our past experiences in other
contexts, and, where applicable, by our prior experiences with the person in
question. But whatever an expectancy’s origin, once we have schematized a
person, the expectancies associated with that schema often lead us to interpret
their behavior in expectancy-reinforcing ways. Our schematic expectancies, like
other knowledge structures, often operate implicitly, outside of our attentional
focus, and this makes them resistant to correction or change. Consider how dif-
ficult it is for a schoolchild whose past behavior created expectancies among his
teachers and classmates to change how he is viewed, even when his behavior
changes. Sometimes the only solution seems to be to switch schools.
As we will discuss later in the chapter, implicit social expectancies have seri-
ous consequences for our understanding of other people. They affect how we
interpret ambiguous behavior, how we attribute causality and responsibility for
it, and how we predict how people will behave in the future. Moreover, our social
expectancies affect how we behave toward their targets. This, in turn, often
 the social perceiver: processes and problems in social cognition 305

affects how those targets behave toward us. If, for example, you apply for a job
and are interviewed by someone who, because of your race, gender, or grade
point average, expects you to be unqualified for the job, their attitude may leak
out in subtle aspects of their behavior, such as seating distance, eye contact, and
facial expressions, all of which may give you a vague sense that they don’t like
you. You may in turn behave in ways that project social discomfort, reticence, or
self-doubt, confirming the interviewer’s prior impression of you. We examine
these kinds of expectancy feedback effects later in this chapter, in sections on the
behavioral confirmation effect and a phenomenon known as “stereotype threat,”
through which a stereotyped person’s fear of performing in a way that will con-
firm a negative stereotype about his group actually does impair his performance
on a stereotype-relevant task.
For clarity’s sake, much of the following discussion of the biasing effects of
social expectancies centers on racial and gender stereotypes and stereotyping.
Readers should bear in mind, however, that social expectancies derive from
other knowledge structures as well, including individual person prototypes and
role prototypes. Where possible, we will provide examples and illustrations
incorporating these as well.

11.3 the cognitive mechanics of social schemas

In November 1994, in a speech to a meeting hosted by Operation PUSH in

Chicago, civil rights leader Jesse Jackson made what must have been a wrench-
ing self-observation: “There is nothing more painful for me at this stage in my
life,” Jackson said, “than to walk down the street and hear footsteps, and start
thinking about robbery, then look around and see someone White, and feel
relieved.” Jackson told the story to illustrate the obstacles facing African
Americans in the late twentieth century. We use it to illustrate certain basic con-
cepts and processes in social cognition.

11.3.1 Stereotypes and Stereotyping

At its most obvious level, Jackson’s anecdote reflects the operation of a stereo-
type, specifically, a stereotype of Black men that includes attributes of criminality
and violence and that evokes fear and a spontaneous behavioral impulse toward
avoidance in at least one particular situation. Many people believe that racial and
other social stereotypes are the result of “prejudice”—a moral defect involving
motivated hostility toward minority groups. Indeed, this was the received under-
standing of psychologists through at least the mid-twentieth century.1 It is diffi-
cult to fit Jessie Jackson into this particular model of racial stereotyping.

1. John Duckitt, The Social Psychology of Prejudice (Westport, CT: Praeger

Publishers, 1992); Gordon Allport, The Nature of Prejudice (New York: Perseus
Books Publishing,1954).
306 problem solving, decision making, and professional judgment

However, Jackson’s response, along with his disappointed and self-

disapproving reaction to it, fits easily into the models of stereotypes and stereo-
typing that have emerged over the past forty years from cognitive social
psychology and, more recently, social cognitive neuroscience. The last several
decades of research in cognitive psychology suggest that stereotypes are just a
particular instance of the schemas, categories, and other knowledge structures
introduced in Section 1.5.2, which structures we necessarily adopt and apply to
make sense of an otherwise impossibly complex perceptual environment.2
Under social cognition theory, stereotypes are viewed as mental structures com-
prising our knowledge, beliefs, affective orientation, and expectations concerning
a categorized social group.3 They contain a mixture of emotional associations,4
abstract “knowledge” about the stereotyped group, exemplars of group members,5
and theoretical causal theories about why members of the group “are the way
they are.”6 As is the case with other types of categories, stereotypes are often orga-
nized hierarchically,7 with higher level entities, such as “women” being divided
into subtypes, such as “mothers,” “old ladies,” “businesswoman,” and so forth.
Social stereotypes, not only of racial minorities, but of other groups, such as
librarians, Muslims, personal injury lawyers, doctors, or homeless people, are
culturally transmitted and then reinforced by our own experience, partially
through the operation of confirmation bias. By a very early age, children have
learned the stereotypes associated with the major social groups in their environ-
ment.8 By adulthood, these stereotypes have a long history of activation and, for
many people, have become “chronically accessible,”9 meaning that they are

2. Susan T. Fiske and Shelley. E. Taylor, Social Cognition: From Brains to

Culture (New York: McGraw-Hill, 2008); Eleanor Rosch, Human Categorization, in
Studies in Cross-Cultural Psychology, 1, 1–2 (Neil Warren ed., 1977); Jerome Bruner,
On Perceptual Readiness, 64 Psychological Review 123 (1957).
3. David L. Hamilton and Jeffrey W. Sherman, Stereotypes, in Handbook of Social
Cognition 168 (Robert S. Wyer and Thomas K. Srull eds., 2d ed. Hillsdale, NJ: Erlbaum,
1994).
4. Mark P. Zanna and John K. Rempel, Attitudes: A New Look at an Old Concept, in The
Social Psychology of Knowledge (Daniel Bar-tal and Arie W. Kruglanski eds., 1988).
5. David L. Hamilton and Jeffrey W. Sherman, Stereotypes, supra note 3
6. Bernd Wittenbrink, Charles M. Judd, and Bernadette Park, Evidence for Racial
Prejudice at the Implicit Level and its Relationship with Questionnaire Measures, 72
J. Personality & Soc. Psychol. 262 (1997).
7. Renee Weber and Jennifer Crocker, Cognitive Processes in the Revision of Stereotypic
Belief, 45 J. Personality & Soc. Psychol. 961–77 (1983).
8. Patricia G. Devine, Stereotypes and Prejudice: Their Automatic and Controlled
Components, 56 J. Personality & Soc. Psychol. 5–18 (1989); Phyllis A. Katz, The
Acquisition of Racial Attitudes in Children, in Towards the Elimination of Racism 125,
Phyllis A. Katz ed. (New York: Pergamom Press, 1976).
9. John A. Bargh, Walter J. Lombardi, and E. Tory Higgins, Automaticity of Person X
Situation Effects on Impression Formation: It’s Just a Matter of Time, 55 J. Personality &
Soc. Psychol. 599–605 (1988).
 the social perceiver: processes and problems in social cognition 307

regularly used to classify people and then to construe, encode, recall, and predict
their behavior, motivations, and character attributes.
Stereotypes often function implicitly, meaning that they can influence social
perception and judgment outside of conscious awareness, whether or not they
are believed. As social psychologist Patricia Devine has shown, even if a person
has consciously rejected a particular stereotyped belief (e.g., that females, as a
group, have less innate scientific aptitude than males), the stereotype may
linger on as an explicit expectancy, capable of influencing a perceiver’s subjec-
tive assessment of a particular male or female target’s scientific aptitude or
ability.10
Stereotypes are spontaneously activated when the stereotype-holder encoun-
ters a member of the stereotyped group. This activation is automatic, and occurs
without intention or awareness, with little cost to other, ongoing mental opera-
tions. In this way, stereotypes, and other social schemas, function in much the
same way as the availability, representativeness, and other heuristics described
in earlier chapters. Although they can serve as useful “rules of thumb,” they can
also lead to poor judgment, not to mention violations of law and important social
and personal ideals.
Law professor Jerry Kang illustrates the process by which stereotypes influ-
ence social perception, judgment, and behavior as shown in Figure 11.1.

Priming effects/
mapping rules

mapped by the
perceiver through into

Person
Perceived social
categories

which alter which activate

interaction with

meanings

figure 11.1 social categorization and person perception.

In Kang’s conception, society provides the individual perceiver with a set of

social categories (schemas). These social categories include “mapping rules”
used to assign the person perceived to a particular social category. So, for

10. Patricia G. Devine, Stereotypes and Prejudice: Their Automatic and Controlled
Components, supra n. 8.
308 problem solving, decision making, and professional judgment

example, a human being wearing a dress, having long, styled hair, and wearing
makeup will usually be mapped as female. However, category assignment may
be influenced not only by culturally provided mapping rules, but also by features
of the stimulus situation, which may “prime” particular categories, increasing
the likelihood that they will be used in categorizing the person perceived. So, in
a drag bar, the above-mentioned human might not be spontaneously mapped as
female. In either case, however, assignment of an individual to a particular cat-
egory activates the category. The category, which comprises a network of mental
associations, supplies the perceiver with a set of implicit and explicit meanings.
These meanings, in turn, systematically influence the perceiver’s judgments of
and interaction with the categorized individual.
Kang’s model can be thought of as a visual representation of the schematic
information processing model of “naturalistic,” or “System 1,” thinking described
in Section 1.6.1. Indeed, stereotyping is best understood not as something that
only “prejudiced” people do, but as one type of System 1 thinking, which is auto-
matic, rapid, associative, effortless, and process opaque.11 Of course, this does
not mean that the application of stereotypes in social judgment is inescapable.
As with other types of System 1 thinking, an initial stereotype-driven impression
can be revised through the application of deliberative, or System 2 thinking,
which is controlled, effortful, and self-conscious.10 We describe this correction
process in Section 11.7.

11.3.2 This is Your Brain on Stereotypes: The Mental Representation

of Social Knowledge
The brain comprises a vast network of individual neurons, interconnected in
unimaginably complex ways at synapses, across which move electrical impulses,
mediated by chemical neurotransmitters. Experience, either real (as through
interaction with other people) or virtual (as through watching television), actu-
ally changes the structure of the brain, as new synaptic connections are created
and strengthened through use, and as others are weakened or even eliminated
from aging, damage, or lack of use.12 But what we think of as the mind is larger
than the sum of these myriad tiny parts.13
As people acquire social knowledge through direct or vicarious experience,
they organize their knowledge into an interconnected web of associations,

11. Daniel Kahneman and Shane Frederick, Representativeness Revisited: Attribute

Substitution on Intuitive Judgment, in Thomas Gilovich, Dale Griffin, and Daniel
Kahneman, Heuristics and Biases: The Psychology of Intuitive Judgment 49
(New York: Cambridge University Press, 2002).
12. Michael S. Gazzaniga, Richard B. Irvy, and George R. Mangun, Cognitive
Neuroscience: The Biology of Mind (New York: W.W. Norton & Co., 2002).
13. Steven Pinker, The Blank Slate: The Modern Denial of Human Nature
(New York: Penguin Books, 2002).
 the social perceiver: processes and problems in social cognition 309

referred to as social “schemas”14 or “concepts.”15 Schemas, including social sche-

mas, contain myriad interrelated elements that represent a person’s accumu-
lated knowledge and beliefs about, experiences (both direct and vicarious) of,
and affective orientations toward the schematized construct. Confronted with
the task of construing a person or his behavior, we unconsciously and automati-
cally match the incoming data against a set of plausible social schemas. An
incoming bit of information that “fits” an existing schema is said to activate the
schema. When a schema is activated, it imposes meaning on the inherently
ambiguous information supplied by raw perception.15
To illustrate, imagine that one fall night, Jesse Jackson is walking down a
street in Brooklyn, on his way to visit a former colleague. He is alone, the street
is relatively deserted, and it is very dark. Hearing footsteps approaching from
behind him, Jackson turns and sees walking toward him a young Black man we
will call William Carter. Carter, an Amherst College sophomore and Amherst
Choral Society member, is in New York City over the weekend to sing in a ben-
efit production of Bach’s B Minor Mass. Carter is walking from the subway sta-
tion to a relative’s house in Brooklyn following a rehearsal, and is dressed for the
chilly November night in a parka, a ski cap, and the casual, oversized blue jeans
he often wears when he doesn’t have to dress up. Seeing Carter, Jackson feels
fear, starts walking faster, and crosses the street. What happened?
Because categories such as race, age, and gender are “chronically accessible”
in our society, the subcategory “young Black male” would likely have been highly
accessible in Jackson’s mind when he first perceived William approaching him
from behind, and Jackson would have spontaneously categorized William in this
way. This categorization would have “potentiated” the category’s content—the
web of associations comprising Jackson’s schema of “young, Black, male-ness.”
Figure 11.2 illustrates just a tiny number of the myriad concepts that might, for
Jackson or anyone else, including William himself, be associated with and in a
sense constitute Jackson’s “young Black male” schema.
Figure 11.2 unpacks the schema of “young Black man” into what Anthony
Greenwald terms a “connectionist” model of social information processing.16
Different nodes of the network may be activated by external stimuli and they, in
turn, may activate other nodes by virtue of internal mental associations. The
content of the network and the strength of the connections between its constitu-
ent parts are shaped by the surrounding culture, what one has learned from

14. Susan T. Fiske, and Shelley E. Taylor, Social Cognition, supra n. 2.

15. Zeva Kunda, Social Cognition: Making Sense of People (Cambridge: MIT
Press, 1999).
16. Anthony G. Greenwald, Mark A. Oaks, and Hunter G. Hoffman, Targets of
Discrimination: Effects of Race on Responses to Weapon Holders, 39 Journal of Experimental
Social Psychology 399–405 (2002).
 310 problem solving, decision making, and professional judgment
Civil Disobedience
SNCC SCLC
Satyagraha
Civil Rights
Activist
Law Abiding

Son, Self as Young Intelligent

Man, Other Relatives,
Political Activist Acquaintances
Divinity
School Middle Class
College Student

Liberal Political
Church
Ideals Student
Black Church
Exposure to Culture(s)
(Art, Music, Literature)
Choir Gospel
Young black male
Church Youth

Avoidance Impulse Fear

Jazz Classical Music
Music

Urban black
R&B
youth
Gangster Rap
Violence
Hip
Hop
Mugger
BlingBling
Crime “Sagging”

Mugging
Script
Oversized Clothing
Ski Cap
Gun
Poverty
Gang Member

Visual Exemplar

figure 11.2 schematic associations: young black male.

 the social perceiver: processes and problems in social cognition 311

parents, teachers, and friends, the media, or one’s own experience and imagina-
tion. Once activated, the nodes of the network mediate subsequent social percep-
tion and judgment. Like any schema, the components of the network contain
affective components (e.g., fright, pride, warmth),17 action “scripts” (e.g., walk
faster, say hello, smile), and causal theories explaining why stereotyped group
members “are the way they are.”15 This is one way of understanding what it
means to say that race is a “social construction.” The social environment “con-
structs” the associative network in the brain that constitutes racial categories.
Notice in Figure 11.2 the wide variety of concepts associated with the subcat-
egory, “young Black male.” Many of these concepts, like “crime,” “violence,”
“poverty,” “gun,” “gang member,” and “mugging script,” which are intercon-
nected at the bottom of Figure 11.2, carry negative connotations. Taken together,
they construct the schema “criminal urban Black male.” But notice also that other
concepts, such as “Black church,” “college student,” “church music,” “church
youth,” “liberal political ideals,” “exposure to culture,” “self as young man,” and
“jazz,” carry positive connotations. Grouped together, these concepts would
constitute a quite different subcategory, perhaps “Black male college student.”
Given all the different constructs that an encounter with William Carter
might activate, how is it that, on any particular occasion, he is subcategorized
and then construed—or misconstrued—in one way rather than another? Because
in American culture young Black men are so frequently associated in media and
other representations with urban violence, and because the dark, deserted New
York street might well have primed the concept of crime in Jackson’s mind,
Jackson would have been predisposed, upon first seeing Carter, to categorize
him as he did. Additionally, Carter’s physical attributes, dress, and posture,
although inherently ambiguous, would have reinforced this initial categoriza-
tion, because of their association with youth gangster culture. The net effect is
the activation of those elements of the Figure 11.2 associative network bolded
and shaded in Figure 11.3.
What might have happened if Carter had been walking down the street sing-
ing the tenor melody from the Gloria section of the B Minor Mass? This new
stimulus feature might have dramatically altered how Jackson classified the
approaching Carter. The schematic network activated by the encounter might
have looked more like Figure 11.4 than like Figure 11.3.
Presented with a William Carter singing the Gloria from Bach’s B Minor
Mass, concepts like “classical music,” “sacred choral music,” “church,” “God,”
and “exposure to culture” would have been activated. These in turn would have
activated other concepts, like “church youth,” “choir,” and “student,” and from
these, perhaps specific exemplars of young Black men from Jackson’s own

17. Russell H. Fazio, Joni R. Jackson, Bridget C. Dunton, and Carol J. Williams,
Variability in Automatic Activation as an Unobtrusive Measure of Racial Attitudes: A Bona
Fide Pipeline?, 69 Journal of Personality and Social Psychology 1013–27 (1995).
 312 problem solving, decision making, and professional judgment
Civil Disobedience
SNCC SCLC
Satyagraha
Civil Rights
Activist
Law Abiding

Son, Self as Young Intelligent

Man, Other Relatives,
Political Activist Acquaintances
Divinity
School Middle Class
College Student

Liberal Political
Church Ideals Student
Black Church
Exposure to Culture(s)
(Art, Music, Literature)
Choir Gospel
Young black male
Church Youth

Avoidance Impulse Fear

Jazz Classical Music
Music

Urban young
black male R&B
Gangster Rap
Violence
Hip
Hop
Mugger
BlingBling
Crime “Sagging”

Mugging
Script
Oversized Clothing
Ski Cap
Gun
Poverty
Gang Member

Visual Exemplar Skiing

figure 11.3 schematic associations with negative prime.

 the social perceiver: processes and problems in social cognition 313

family, religious community, or even of himself as a young man. The end result
would have been the instantiation of a particular subtype of “Black young man”
that incorporated an interconnected network of positive associations and emo-
tional reactions very different from those constituting the stereotype of young
Black urban male criminality depicted in Figure 11.4.
Indeed, recent research suggests that the activation of certain elements of a
schematic network can result not only in the activation of elements with which
they are connected, but also in the deactivation of elements with which they are
disassociated.18 Seen in this way, the various constructs comprising our mental
representations can stand in three types of relation to each other. They can be
associated, in which case the activation of one will increase the probability that
the other will be activated. They can be completely disassociated, in which case
the activation of one will have no impact on the probability of the others’ activa-
tion. Or, they can be negatively associated, or mutually inhibitory, such that the
activation of one actually decreases the probability that the other will be
invoked.
For example, many female executives and professionals report that, as soon
as they have their first child, people at work start responding to them differently.
They are given less responsible job assignments; their suggestions are taken less
seriously; they are accorded less deference and respect. One explanation of this
reported phenomenon is that it is very difficult for many people to associate the
social role construct “mother” with the trait constructs “aggressive” or “hard-
driving.” The activation of “mother” not only activates a set of trait constructs,
such as “nurturing,” or “soft,” it actually inhibits the activation of others, like
“intellectually sharp” or “professionally productive.”19
Just as it is hard for many people to think of another person simultaneously
as a “new mother” and a “hard-driving executive,” it is hard to think simultane-
ously about sacred choral musicianship and a propensity toward street violence.
By singing Bach’s B Minor Mass, William Carter has not only behaved in a way
that will activate certain constructs in Jesse Jackson’s mind, he has behaved in a
way that will inhibit the activation of others. The result is a profound change in
what Jackson “sees”—at least as profound as what Dartmouth and Princeton
students saw in the football game recounted in Section 9.4.1.
We have presented two alternative scenarios of the external stimuli in
Jackson’s encounter with Carter. In our hypothetical case, whistling Bach made
all the difference in terms of which schematic elements were activated. But other
factors can determine which aspects of an incoming perception most influence

18. Paul Thagard, and Zeva Kunda, Making Sense of People: Coherence Mechanisms, in
Connectionist Models of Social Reasoning and Social Behavior 3–26 (Stephen J.
Read and Lynn C. Miller eds., Mahwah, N.J.: Erlbaum,1998).
19. Amy. J. C. Cuddy, Susan T. Fiske, and Peter Glick, When Professionals Become
Mothers, Warmth Doesn’t Cut the Ice, 60 Journal of Social Issues 701 (2004).
 314 problem solving, decision making, and professional judgment
Civil Disobedience
SNCC SCLC
Satyagraha
Civil Rights
Emotional Activist
Warmth Law Abiding

Son, Self as Young Intelligent

Man, Other
Political Activist Relatives,
Divinity
School Acquaintances Middle Class
College Student

Liberal Political
Church Ideals Student
Black
Exposure to
Church
Culture(s) (Art,
Choir Gospel Music, Literature)
Young black male
Church Youth

Avoidance Impulse Fear

Jazz Classical Music
Music

Urban black
youth R&B
Gangster Rap
Violence
Hip
Hop
Mugger
BlingBling
Crime “Sagging”

Mugging
Script
Oversized Clothing
Ski Cap
Gun
Poverty
Gang Member

Visual Exemplar Skiing

figure 11.4 schematic associations with positive prime.

 the social perceiver: processes and problems in social cognition 315

schema activation. Some kinds of stimuli are more salient than others. All other
things being equal: visual stimuli,20 negative stimuli,21 and stimuli that are
unusual or unexpected (like a young person walking down a Brooklyn street
singing a portion of Bach’s B Minor Mass) are particularly salient.
Also, the more frequently or recently a particular schema has been activated,
the more available it is, and thus the more likely it is to be reactivated by a stimu-
lus that relates to it.22 Thus, given the prominence of news stories about crimes
committed by young Black men, seeing a young Black man on an empty Brooklyn
street late at night is likely to activate the negative elements of a schematic net-
work together with a “mugging” script and a fear/flight response.23
Priming also exerts a powerful impact on which schemas are activated by a
particular stimulus. Priming refers to the notion that people’s judgment, atti-
tudes, and behavior can be influenced by incidental cues in their surroundings.
Exposure to subtle situational cues, such as words, people, and even scent or
physical objects, can automatically activate knowledge structures in memory
such as trait concepts, stereotypes, and goals. Such automatic activation can
occur without people’s intention or awareness and carry over for at least a while
to exert an unintended influence on judgment and behavior.
For example, John Bargh and his colleagues primed people with words related
to the elderly (e.g., Florida, bingo) in what was said to be a sentence unscram-
bling task.24 After completing the task, when participants believed the session
had ended, the researchers measured how long it took participants to walk to the
elevator on their way out of the building. Surprisingly, participants primed with
words related to the elderly walked significantly more slowly than other partici-
pants who were primed with neutral words (e.g., normal, send). In another
study, Bargh and his colleagues found that people primed with words related to
rudeness (e.g., disturb, impolitely) were subsequently quicker to interrupt an

20. Susan T. Fiske and Martha G. Cox, Person Concepts: The Effect of Target Familiarity
and Descriptive Purpose on the Process of Describing Others, 47 Journal of Personality
136–61 (1979).
21. Felicia Pratto, and John P. Oliver, Automatic vigilance: The Attention-Grabbing Power
of Negative Social Information, Journal of Personality and Social Psychology (1991).
22. Felicia Pratto, and John A. Bargh, Stereotyping Based on Apparently Individuating
Information: Trait and Global Components of Sex Stereotypes Under Attention Overload, 27
Journal of Experimental Social Psychology 26 (1991); E. Tory Higgins, and Gilliam A.
King, Accessibility of Social Constructs: Information-Processing Consequences of Individual and
Contextual Variability (1981), in Personality, Cognition, and Social Interaction 69–121
(Nancy Cantor and John F. Kihlstrom eds., Hillsdale, NJ: Erlbaum, 1986); Robert S. Wyer and
Thomas K. Srull, Human Cognition in its Social Context, Psychological Review, 93, 322–59.
23. Jerry Kang, Trojan Horses of Race, 118 Harv. L. Rev. 1489–1593 (2005).
24. John A. Bargh, Mark Chen, and Lara Burrows, Automaticity of Social Behavior:
Direct Effects of Trait Construct and Stereotype Activation on Action, 71(2) Journal of
Personality and Social Psychology 230–44 (1996).
316 problem solving, decision making, and professional judgment

ongoing conversation between the experimenter and a third person than people
primed with neutral words. These findings imply that the mental circuits used to
perceive social stimuli, such as words, are closely linked to behavioral schemata
and other mental representations that affect people’s judgments and behavior.
Priming effects not only have consequences for simple, behavioral responses
such as speed of walking, but can also affect more complex behaviors and goals.25
One group of White participants was asked to write about a day in the life of
Tyrone Walker, who most people assumed to be an African American. Other White
participants were asked to write about a day in the life of Eric Walker, who they
assumed to be White. Next, all participants were asked to complete a math section
of the Graduate Record Examination (GRE). Participants who wrote about Tyrone
Walker performed poorly compared to those who wrote about Eric Walker.
As this example illustrates, racial stereotype activation can elicit maladaptive,
stereotype-consistent behavior among individuals who hold a stereotype. In other
words, because the stereotype of African Americans includes an association with
low academic achievement, activating the African American stereotype noncon-
sciously caused people to behave in a way consistent with that stereotype.
Priming can take the form of entirely unrelated stimuli occurring before the
event—for example, seeing a film (not necessarily about race) that leaves the
viewer with feelings of brotherly love on the one hand, or dread on the other.
One can also be primed by one’s own thoughts. For example, walking down a
deserted Brooklyn street late at night might give rise to scary images or feelings.
Such thoughts, even if subliminal and fleeting can promote, or prime, a negative
construal of inherently incoming perception.26
Finally, schema activation can be influenced by the perceiver’s interaction
goals. Much of our interaction with other people is motivated toward some end.
When you first meet a new coworker upon whom you must depend for the suc-
cess of your own efforts, you will probably be motivated to like the person and
feel confident in his or her abilities. To feel otherwise would prompt distress.
Likewise, if you need surgery, you will be motivated to perceive the surgeon who
will be operating on you as competent. People who are motivated to avoid preju-
dice tend to monitor their own reactions to members of other groups, attempt-
ing to notice and correct for the effects of stereotypes or other forms of bias.
We discuss issues relating to controlling or correcting for the biasing effects of
stereotype activation later in Section 11.7.

25. Based on S. Christian Wheeler, Jarvis W. Blair, and Richard E. Petty, Think
Unto Others: The Self-Destructive Impact of Negative Racial Stereotypes, 37 Journal of
Experimental Social Psychology 173–80 (2001).
26. Jerome Bruner, On Perceptual Readiness, 64 Psychological Review 123 (1957);
Jerome Bruner, Going Beyond the Information Given, in Contemporary Approaches to
Cognition, Jerome Bruner, ed., (Cambridge, MA: Harvard University Press, 1957).
 the social perceiver: processes and problems in social cognition 317

11.4 the effects of social schemas on social perception

and judgment

To summarize, the processes by which we schematize a person encountered

in our social experience are largely automatic, remarkably complex, and subject
to influences that are not within our consciousness. Once activated, a social
schema supplies the perceiver with a set of meanings about the person perceived.
It determines what we will “see” in an individual’s behavior and creates an expec-
tation that later-perceived information about the person will match the concepts
constituting the schema.
Stereotypes and other social schemas bias person perception and judgment
in four broad ways, each of which is described in the material that follows. These
include:
• the characterization of ambiguous behavior;
• causal attribution;
• the way experience of the schematized person is stored in and retrieved
from memory; and
• the way such information is used in making judgments about the person.
Moreover, social schemas, such as stereotypes or person prototypes, influ-
ence how a social perceiver behaves toward the person perceived. This behavior
can in turn alter how the target responds to the perceiver, often setting up a self-
fulfilling prophesy effect that serves to reinforce the perceiver’s stereotyped
expectancies.

11.4.1 The Effects of Social Schemas on the Characterization of

Ambiguous Behavior
In 1980, social psychologists H. Andrew Sagar and Janet Schofield conducted a
now classic experiment examining the effects of racial stereotypes on children’s
characterizations of ambiguous behavior.27 They presented elementary school
students with cartoon-like drawings of two students sitting in a classroom, one
behind the other, together with this verbal story: “Mark was sitting at his desk,
working on his social studies assignment, when David started poking him in the
back with the eraser end of his pencil. Mark just kept on working. David kept
poking him for a while, and then he finally stopped.” Participants were asked
whether they thought David’s behavior was playful, friendly, mean, or threaten-
ing. When the cartoon depicted David as Black, participants judged his behavior
to be more mean and threatening and less playful and friendly. The opposite
result obtained when he was White.

27. Andrew H. Sagar, and Janet W. Schofield, Racial and Behavioral Cues in Black and
White Children’s Perceptions of Ambiguously Aggressive Acts, 39 Journal of Personality
and Social Psychology 590–98 (1980).
318 problem solving, decision making, and professional judgment

Sagar and Schofield’s results replicated the finding of an earlier experiment

by Birt Duncan conducted in 1976. In Duncan’s study, White college students
watched one of four videotapes in which two males discussed alternative solu-
tions to a problem. Participants were told they were watching a live interaction
happening in another room, and that its purpose was to develop a new
system for rating interpersonal behavior. While the videotape played, a buzzer
rang at specified intervals, signaling participants to categorize the behavior they
were then viewing in one of ten categories and to indicate its intensity on an
8-point scale.
As the videotaped discussion progressed, the dialogue became increasingly
heated. Finally, one of the actors (the protagonist) shoved the other (the victim).
At that point, the buzzer rang (not for the first time), and participants were asked
to characterize and rate the intensity of the protagonist’s behavior. As in the
Sagar and Schofield study, the protagonist’s race significantly affected how par-
ticipants characterized the shove. If the protagonist was White, his behavior was
characterized as “playing around” or “dramatizes.” If he was Black, it was char-
acterized as “aggressive” or “violent.”28
Perhaps the most dramatic demonstration of the ways in which implicit
stereotypes can influence how ambiguous behaviors are construed comes from
a 1983 experiment on stereotyping and socioeconomic class by John Darley
and Paget Gross. Darley and Gross had participants view one of two videotapes,
each portraying a school-age child named Hannah at play. Her behavior was
identical on both tapes, but one depicted Hannah playing on a stark, fenced-in,
asphalt school yard in a rundown urban neighborhood. In this condition
researchers told participants that Hannah’s parents had only a high school
education, that her father worked as a meatpacker, and that her mother was
a seamstress who worked at home. The other videotape, which was shown to
a second group of participants, portrayed Hannah playing in a tree-lined park
in an upper middle-class neighborhood. Her school was depicted as a
large, attractive, modern building, with adjacent playing fields and a shaded
playground. Her father was described as an attorney and her mother as a
freelance writer.
Participants were then asked to predict whether Hanna would perform aca-
demically at, above, or below grade level. Reluctantly, and often protesting being
asked to base their prediction on stereotypes, most participants in both conditions
predicted that Hannah would perform at grade level. In other words, they refused
to express an initial prediction based on the stereotypes about academic achieve-
ment associated with high and low socioeconomic status.

28. Birt L. Duncan, Differential Social Perception and Attribution of Intergroup Violence:
Testing the Lower Limits of Stereotyping of Blacks, 34 J. Personality & Soc. Psychol. 590–98
(1976).
 the social perceiver: processes and problems in social cognition 319

Not all participants were asked to predict Hannah’s achievement level right
away, however. Some watched a second videotape before they were asked to
predict Hannah’s academic performance. This video, which was identical across
both conditions, depicted Hannah responding verbally to twenty-five achieve-
ment test problems. Researchers told participants that the test included easy,
moderate, and difficult problems. Hannah’s performance was designed to be
ambiguous. After watching the second video, participants who had seen the high
socioeconomic status (SES) Hannah rated her academic ability significantly
higher than did participants who saw her low SES double. Open-ended com-
ments revealed differences in perception as well. For example, participants
reported the low-income Hannah as having “difficulty accepting new informa-
tion,” while they frequently described high-income Hannah as demonstrating
the “ability to apply what she knows to unfamiliar problems.”
Darley and Gross suggested that exposure to ambiguous diagnostic informa-
tion actually promotes social stereotyping.29 People who do not believe it appropri-
ate to base social predictions on stereotypes based on socioeconomic information
alone, nonetheless did so in what they perceived to be a more diagnostic con-
text—even though it was thoroughly ambiguous. From this, Darley and Gross
argued that stereotypes function not so much as conscious ex ante decision rules,
but as dormant expectancies that impose a hypothesis-confirming bias on the
interpretation of incoming social information.
The effects of stereotypes and other schematic expectancies on social percep-
tion and judgment have been well documented across a variety of different
contexts.30 Perhaps most disturbing are recent studies showing that that a
target person’s race has a significant effect on whether both Black and White
participants think he is holding a weapon, like a knife or a gun, as opposed to
a harmless object, like a wallet or cell phone.31
Although most of the research on the biasing effects of social schemas has
focused on race, age, and gender stereotypes, the basic idea that schemas shape
how we interpret other people’s behavior applies in a variety of other contexts

29. John M. Darley and Paget H. Gross, A Hypothesis-Confirming Bias in Labeling Effects,
44 J. Personality & Soc. Psychol. 20–33 (1983).
30. David Dunning and David A. Sherman, Stereotypes and Tacit Inference, 73
J. Personality & Soc. Psychol. 459–71 (1997); Zeva Kunda and Bonnie Sherman-
Williams, Stereotypes and the Construal of Individuating Information, 19 J. Personality &
Soc. Psychol. 90–99 (1993).
31. Joshua Correll, Geoffrey R. Urland, and Tiffany A. Ito, Event-Related Potentials and
the Decision to Shoot: The Role of Threat Perception and Cognitive Control, 42 Journal of
Experimental Social Psychology 120–28 (2006); Joshua Correll, Bernadette Park, Charles
M. Judd, and Bernd Wittenbrink, The Police Officer’s Dilemma: Using Ethnicity to Disambiguate
Potentially Threatening Individuals, 83 J. Personality & Soc. Psych. 1314 (2002); B. Keith.
Payne, Prejudice and Perception: The Role of Automatic and Controlled Processes in Misperceiving
a Weapon, 81 J. Personality & Soc. Psychol . 181–192, 185–186 (2001).
320 problem solving, decision making, and professional judgment

relevant to lawyering. Consistent with naïve realism (Section 9.8), in conflict

situations, for example, people tend to stereotype their adversaries as fanatical,
grasping, and “out to get them” and then interpret adversaries’ behavior in ways
that make negotiated agreement more difficult.32
In any social context, once we have characterized another person in a particu-
lar way, we may misinterpret his actions or misattribute his intentions in ways
that preserve our impression about him but totally miss the mark in terms of
accuracy. These sorts of misapprehensions can lead to inaccurate predictions of
others’ future behavior and unfair attribution of blame and praise, and can also
injure our relationships with constituents, clients, coworkers, business partners,
witnesses, opposing counsel, and those who matter in our personal lives.

The Burden on “Tokens”

In any social situation, for any of a number of reasons, a particular person in a

group may stand out. This social salience can derive from a number of sources.
A person may be made salient, for example, by being the only woman, visibly
disabled person, or member of a racial or ethnic minority in a given group.
A person may become salient by behaving in an unexpected way, for example,
counter-stereotypically, or by dominating the visual field by virtue of his or her
physical appearance or location.33
Regardless of why a person becomes salient, that salience tends to affect in
systematic ways how the person is perceived. So, for example, in a now classic
study, Harvard psychologist Shelley Taylor and her colleagues explored the
consequences of a minority group member’s being a “token” or “solo” in a
small group. In the first experiment in that study, participants listened to a tape
recording of a discussion among six males. As each person spoke, the

32. Robert. J. Robinson and Dacher Keltner, Defending the Status Quo: Power and Bias
in Social Conflict, 23 Personality and Social Psychology Bulletin 1066–77 (1997);
Robert J. Robinson and Raymond A. Friedman, Mistrust and Misconstrual in Union-
Management Relationships: Causal Accounts in Adversarial Contexts, 6 International
Journal of Conflict Management 312–27 (1995); Roderick M. Kramer, The Sinister
Attribution Error: Paranoid Cognition and Collective Distrust in Organization, 18 Motivation
and Emotion 199–230 (1994); Dacher Keltner and Robert J. Robinson, Imagined Ideological
Differences in Conflict Escalation and Resolution, 4 International Journal of Conflict
Management 249–62 (1993).
33. Leslie Z. McArthur, What Grabs You?: The Role of Attention in Impression
Formation and Causal Attribution in Social Cognition (1981); Shelley E. Taylor
and Susan T. Fiske, Salience, Attention, and Attribution: Top of the Head Phenomena, in 11
Advances in Experimental Social Psychology 249, 264–65, Leonard Berkowitz ed.
(New York: Academic Press, 1978).
 the social perceiver: processes and problems in social cognition 321

experimenters projected a slide of that person’s picture. By playing the same tape
recording, while varying the race of the person supposedly speaking, Taylor and
her colleagues compared participants’ judgments of a Black person when he was
the only Black person in an otherwise all White group, and when he was in a fully
integrated group. In the “solo” condition, participants judged the Black person in
more extreme ways and perceived him as playing a more prominent role in the
discussion than Blacks were perceived in the “integrated” condition.
In a second experiment, Taylor and her colleagues found similar, even stron-
ger effects on perceptions of solos in gender-mixed groups. Token women and
minorities, in short, are rated more extremely and are viewed (for better or for
worse) as playing a more prominent role in group process, than are women or
minorities in integrated groups.

11.4.2 The Effects of Social Schemas on Memory

The need to reconstruct the past pervades legal problem solving and decision
making in virtually every imaginable context. In describing a problematic situa-
tion to a lawyer in an initial interview, clients are almost always asked to describe
“what happened?” Identifying the source of a presenting problem almost always
requires piecing together information about when the problem first occurred,
and what events preceded or accompanied it. Similarly, when we are called upon
to make judgments of other people with whom we have had contact, we almost
always search our memories for judgment-relevant events.
But social memory, like all memory, is not like looking into a filing cabinet
and extracting a set of intact, relevant files. But, as discussed in Section 9.2,
memory is a reconstructive process. Our recollections (or, perhaps more accu-
rately, re-collections) of events are shaped by how we perceived those events as
they were occurring, how we encoded them into and stored them in memory,
and how we extracted them from memory and combined them with other infor-
mation. Stereotypes, person prototypes, and other kinds of social expectancies
operate at every stage of the memory process.
As we have already seen, stereotypes and related mental representations
influence how we characterize other people’s behavior. This initial understand-
ing of what is happening in turn influences the way in which we remember what
happened. This phenomenon can be attributed to a number of interrelated
phenomena.
11.4.2.a Expectancies, Attention Allocation, and Encoding into Memory We
are constantly bombarded with far more information than we can process, and
for this reason, we selectively direct our attention to particular aspects of our
physical and social environment. And as a general rule, we attend preferentially
to incoming information that reinforces our prior expectancies, especially when
322 problem solving, decision making, and professional judgment

we are stressed or emotionally agitated.34 Increased focus on a particular event

increases its memorability, which in turn increases the likelihood that it will be
remembered at a later date.35
Expectancy-consistent events are more likely to be remembered not only
because we are more likely to attend to them, but also because of the way they are
encoded into memory. When we view another person’s behavior, we often spon-
taneously assign it a trait attribution. This inferred trait itself becomes part of the
stored memory of the event which triggered it.15
Experimental evidence supports the hypothesis that prior expectancies influ-
ence social memory at the encoding stage. One example can be found in a 1979
study by Myron Rothbart and his associates. They presented participants with
identical sets of fifty behavioral descriptions of a target male individual. The
items presented described actions that could be characterized as either “friendly,”
“unfriendly,” “intelligent,” “unintelligent,” or unrelated to any of these four con-
structs. Half of the participants were led to believe that the target was intellectual
and half that the target was friendly. In each group, half of the participants were
given this expectancy before being presented with the behavioral descriptions
and half afterward.
Rothbart found that participants recalled behaviors that confirmed their prior
expectancies more readily than those which disconfirmed or were unrelated to
them, but this effect was found only when the expectancy was induced prior to
the presentation of the behaviors. Participants who viewed the behaviors before
any expectancy was created were far more evenhanded in their memory for
confirming and disconfirming information. From these findings, Rothbart and
his colleagues inferred that selective retrieval alone could not account for the
observed memorial bias in favor of expectancy-confirming events. Whatever
was causing the bias operated, at least in part, much earlier in the attention/
encoding/storage/recall continuum.
11.4.2.b Prior Expectancies and Retrieval from Memory Even assuming that
we were to encode expectancy-confirming and expectancy-disconfirming infor-
mation about other people in an evenhanded fashion, those expectancies could—
and in fact do—bias memory at the point of recall. This effect is mediated by
what social psychologists Joshua Klayman and Young-won Ha refer to as a “pos-
itive test strategy.” (See Section 10.3.4)
To say that people use a positive test strategy means that, in evaluating the
soundness of a particular proposition, people intuitively search for evidence that
confirms it. This tendency applies both when searching through memory for

34. David W. Jamieson and Mark P. Zanna, Need for Structure in Attitude Formation
and Expression, in Attitude Structure and Function 383–406 (Anthony R. Pratkanis,
S. J. Breckler, and Anthony G. Greenwald eds., Hillsdale, NJ: Erlbaum, 1989).
35. Joshua Klayman and Young-Won Ha, Confirmation, Disconfirmation, and
Information in Hypothesis Testing, 94 Psychological Review 211–28 (1987).
 the social perceiver: processes and problems in social cognition 323

preexisting evidence relating to a proposition,36 and when searching the external

environment for new or additional information.35 Numerous studies have shown
that expectancy-disconfirming information is also salient, but its use in a par-
ticular judgment task is often lessened through attribution to external or tran-
sient factors that leave the expectancy itself intact.37
To summarize what we have covered so far, once we have formed an initial
impression of a person, either because we hold a stereotype relating to a social
group to which they belong, or because some prior experience with them has
caused us to characterize them in some way, that initial impression influences in
significant ways our subsequent perceptions of and judgments about the person.
Initial impressions influence subjective construal of the person’s ambiguous
behavior. They also influence how we will store information about the person in
memory, whether and how we will retrieve it from memory, and how we will
combine it with other information in making a subsequent judgment. As we will
now see, our expectancies can also influence the ways in which we behave toward
the target person, which in turn, may affect how they behave in response.

11.4.3 The Effects of Stereotypes on Causal Attribution and

Behavioral Prediction
Stereotypes, person prototypes, and other social expectancies exert powerful
effects on how we attribute people’s outcomes or behaviors and on our predic-
tions of how they will behave in the future. Specifically, when members of
stereotyped groups behave in stereotype-confirming ways, we tend to attribute
their behavior to stable, dispositional factors, like innate ability or character
traits. However, when a stereotyped target behaves in a stereotype-inconsistent
way, we tend either to explain away the discrepancy by attributing the behavior
to situational factors, or to “exceptionalize” the person, rather than question the
accuracy of our stereotype. Moreover, we tend to view stereotype-consistent
behavior as more predictive than stereotype-inconsistent behavior. For judges,
prosecutors, parole and probation officers, and social workers who make recom-
mendations to judicial tribunals, misattribution or biased prediction of outcomes
or transgressions can significantly influence, among other things, charging, sen-
tencing, probation, parole, and child custody decisions.
The most dramatic laboratory demonstration of stereotype-driven attribution
bias, and of its effects on behavior prediction and the selection of criminal penal-
ties, appears in two experiments conducted by Galen Bodenhausen and Robert
Wyer in 1985. In both experiments, participants read a case file describing a
behavioral transgression by another person. In the first experiment, the case file

36. Mark Snyder, and Nancy Cantor, Hypothesis Testing in Social Interaction, 36
J. Personality & Social Psychol. 1202–12 (1979).
37. Jennifer Crocker, Darlene B. Hanna, and Renee Weber, Person Memory and Causal
Attributions, 44 J. Personality & Social Psychol. 55 (1983).
324 problem solving, decision making, and professional judgment

contained only general biographic, demographic, and job-related information

about the person. In the second experiment, the case file also contained informa-
tion about the person’s background and life circumstances at the time the trans-
gression occurred.
After reading the case file, participants were asked to predict whether the
transgression would recur and to recommend a negative sanction. In one condi-
tion, the transgression was stereotypic of the target person’s ethnic group, in a
second, the transgression was stereotype-inconsistent, and in a third, no stereo-
type was activated. When the transgression was stereotypic of the target’s ethnic-
ity, participants perceived it as more likely to recur. Participants assigned more
severe punishments in that situation than when the transgression was stereo-
type-inconsistent or when no stereotype had been activated.38
Findings of this sort are not restricted to laboratory settings. In a 1998 inves-
tigation of the Washington State juvenile justice system, University of Washington
sociologists George Bridges and Sara Steen demonstrated that, controlling for a
large number of plausible explanatory factors, African American youth were
denied probation significantly more frequently than White youth, and that this
disparity could be traced to attribution bias on the part of court personnel.
Specifically, Bridges and Steen demonstrated that juvenile court officers who
made probation versus incarceration recommendations to judges, tended sys-
tematically to attribute criminal behavior by African American offenders to dis-
positional factors, while attributing equivalent criminal behaviors by White youth
to external, situational variables. These differences in the locus of attribution,
in turn, influenced the officers’ assessments of moral culpability and likelihood
of future criminal behavior, which in turn led to racially biased recommendation
relating to the probation versus incarceration decision.39
In short, stereotypes influence attributions. Attributions influence predic-
tions and judgments of personal responsibility. And all three influence judg-
ments and decisions made about the stereotyped individual.

11.5 on the self-fulfilling nature of social stereotypes

As we have seen, people tend to use social knowledge structures to fill in gaps in
the information they have about another person and to predict how that person

38. Galen Bodenhausen and Robert Wyer, Effects of Stereotypes in Decision Making and
Information-Processing Strategies, 48 Journal of Personality and Social Psychology
267–82 (1985).
39. George S. Bridges and Sara Steen, Racial Disparities in Official Assessments of
Juvenile Offenders: Attributional Stereotypes as Mediating Mechanisms, 63 Am. Soc. Rev. 554
(1998).
 the social perceiver: processes and problems in social cognition 325

will behave in the future.40 Once a perceiver has formed expectancies about a
target person, those expectancies can play a significant role in shaping the per-
ceiver’s behavior toward the target in subsequent interactions. This may, in turn,
affect the target’s behavior toward the perceiver, often in ways that confirm the
perceiver’s initial expectancy. This behavioral confirmation effect represents one
way in which stereotypes can function as self-fulfilling prophesies. Stereotype threat
represents a second process by which stereotypes become self-fulfilling. In stereo-
type threat situations, a person’s awareness that his or her group is negatively
stereotyped impairs his or her performance in a stereotype-relevant domain.

11.5.1 The Behavioral Confirmation Effect

If we think people are unimportant, or “not real bright,” we may treat them dis-
missively. If we think they are aggressive or “chippy,” we may be defensive when
we interact with them, betraying our discomfort through subtle facial or verbal
cues, by cutting verbal interactions short, or through our body posture. If we
expect people to be talented and interesting, we are apt to be solicitous and warm
in our interactions with them, drawing them out, smiling frequently, and listen-
ing attentively. Naturally, the way we behave toward other people influences how
they behave toward us in response. In many cases, their responsive behavior
serves to confirm our initial expectancy, providing us with further evidence of
“how they are.”
In one of the earliest demonstrations of this phenomenon, Robert Rosenthal
and Lenore Jacobson administered an intelligence test to students in an elemen-
tary school. They randomly selected 20 percent of these students and told their
teachers that their scores predicted that they would academically outperform
their classmates in the near future. Eight months later, Rosenthal and his col-
leagues retested students at the school. They found that those randomly desig-
nated as “high performers” had in fact overperformed their peers on the
subsequent test. The experimenters explained their results by suggesting that
teachers formed an expectancy of the students in the randomly designated “high
performer” group, and that this expectancy had influenced teachers’ interactions
with the students in ways that enhanced those students’ performance.41
While Rosenthal and Jacobson’s study was subsequently criticized on method-
ological grounds, its basic finding, that expectancies tend to create self-fulfilling
prophesies, is one of the best-replicated findings in social psychology. It has been
observed in connection with different types of expectancies across many different

40. James M. Olson, Neal J. Roese and Mark P. Zanna, Expectancies, in Social
Psychology: Handbook of Basic Principles 211–38, E. Tory Higgens and Arie
Kruglanski eds., (New York: Guilford, 1996).
41. Robert Rosenthal, and Lenore Jacobson, Pygmalion in the Classroom
(New York: Holt, Rinehart, & Winston, 1968).
326 problem solving, decision making, and professional judgment

domains, including interview and therapy settings, bargaining and “get-

acquainted” encounters, educational settings, and mother-infant interactions.42
Social psychologist Edward Jones illustrates the process components of
behavioral confirmation as shown in Figure 11.5.
The first three components in this model should be familiar by now. They
represent two processes discussed earlier in this chapter—the use of expectan-
cies in interpreting ambiguous behavior and in behavioral attribution. But what
evidence exists regarding step four, which Jones refers to as the expectancy-
behavior link, and step five, the behavior-behavior link?
With respect to step four, there now exists a large body of evidence demon-
strating that stereotypes influence subtle aspects of the stereotype-holder’s
behavior toward or in the presence of a stereotyped target. For example, in a now
classic study conducted at Princeton in 1974, Carl Word, Mark Zanna, and Joel
Cooper showed that White participants who interviewed Black and White job
applicants behaved quite differently toward Black and White interviewees. (The
“applicants” were actually confederates, trained to respond in uniform ways.) In
interviewing Black confederate/applicants, White participants made more speech
errors, maintained greater seating distances, and ended interviews sooner.43
In a 2001 extension of Carl Word, Mark Zanna, and Joel Cooper’s findings,
Allen McConnell and Jill Leibold demonstrated that participants high in implicit
racial bias44 interacted with Black as opposed to White trained experimenters
with less smiling, shorter speaking times, higher rates of fidgeting, more speech
errors and hesitations, and fewer extemporaneous social comments.45
More recently, research in social cognitive neuroscience has explored some of
the physiological mechanisms mediating these expectancy-behavior links. So,
for example, using functional magnetic resonance imaging (fMRI) technology

42. Steven L. Neuberg, Social Motives and Expectancy-Tinged Social Interactions, in

Handbook of Motivation and Cognition, Volume 3: The Interpersonal Context,
Richard M. Sorrentino and E. Tory Higgins eds., (New York: Guilford, 1996).
43. Carl O. Word, Mark P. Zanna, and Joel Cooper, The Nonverbal Mediation of Self-
Fulfilling Prophecies in Interracial Interaction, 10 Journal of Experimental Social
Psychology 109–20 (1974).
44. In this experiment, as in most recent social cognition research, implicit bias is
measured through the Implicit Association Test (IAT). Using a comparative reaction time
method, the IAT measures the strength of associations between group constructs (Black/
White; young/old; male/female; straight/gay) and traits or global attitudes (good/bad;
science/ humanities; street-smart/book-smart). Interested readers can read about and/or
take a demonstration IAT by visiting the Project Implicit Web site, at http://implicit.
harvard.edu/ implicit . For a discussion of the IAT, see Brian A. Nosek et. al., Harvesting
Implicit Group Attitudes and Beliefs from a Demonstration Web Site, 6 Group Dynamics
101, 105 (2002).
45. Allen R. McConnell and Jill M. Leibold, Relations Among the Implicit Association
Test, Discriminatory Behavior, and Explicit Measures of Racial Attitudes, 37 Journal of
Experimental Social Psychology 435 (2001).
 the social perceiver: processes and problems in social cognition 327

The Perceiver The Target

1. Tentative expectancy:
(I’m told he’s very bright.)

2. Ambiguous behavior:
(Could be seen as bright.)

3. Expectancy strengthened
by perceptual confirmation:
(He does seem bright!)

4. Expectancy-influenced
behavior (i.e., eye contact,
body posture, facial
expressions, patterns of
verbal interaction (the
expectancy-behavior link)
5. Lively, energized response
(behavior-behavior link)

6. Expectancy further
strengthened by behavioral
confirmation: (I was right.
He is bright!)

figure 11.5 jones’s model of the behavioral confirmation process.

and facial electromyography (EMG), a technique for measuring electrical activity

in facial muscles, a recent study showed that participants whose performance on
the Implicit Association Test (IAT) revealed negative implicit attitudes toward
Blacks exhibited both a heightened startle response (measured by EMG eyeblink
potentiation) and elevated amygdala46 activation when viewing photographs of
unfamiliar Black faces.47 A similar study showed that White participants who
favored White over Black applicants in a resume review task registered more
EMG activity in the muscles controlling smiling when they reviewed White
applicants’ resumes than when they reviewed Black applicants’ resumes.

46. The amygdala is a small, bilateral structure located in the brain’s temporal lobe. It
plays a significant role in emotional learning and evaluation, and particularly in the evoca-
tion of emotional responses to fear or aggression-inducing stimuli. Ralph Adolphs, Daniel
Tranel, and Antonio R. Damasio, The Human Amygdala in Social Judgment, 393 Nature
470 (1998).
47. Elizabeth A. Phelps, Kevin J. O’Connor, William A. Cunningham, E. Sumie
Funayama, J. Christopher Gatenby, John C. Gore, and Mahzarin R. Banaji, Performance
on Indirect Measures of Race Evaluation Predicts Amygdala Activation, 12 J. Cognitive
Neuroscience 729-738, 730 (2002).
328 problem solving, decision making, and professional judgment

Members of negatively stereotyped groups often claim that they can “just tell”
when someone they are dealing with is biased against them. Given that people
are exquisitely attuned to subtle facial, verbal, and other physical cues, and use
these to decode others’ attitudes toward them, it is certainly reasonable to believe
that that subtle behavioral “leakages” from implicitly biased perceivers can be
sensed by members of stereotyped groups.
Just as perceivers’ behavior toward stereotyped targets is affected by the
expectancies that stereotypes activate, stereotyped targets respond in kind, thus
creating the behavior-behavior link described as step 5 of Jones’s model. The
earliest evidence for the behavior-behavior link comes from Rosenthal
and Jacobson’s 1968 study, described earlier, in which students responded to
their teacher’s arbitrarily created expectancies by academically outperforming
their peers. Substantial additional evidence for the link has accumulated since
then.
For example, in the McConnell and Leibold study described above, White
confederates were trained to mimic the “friendly” and “unfriendly” interview
styles that had been observed in Word, Zanna, and Cooper’s 1974 experiment. In
the “friendly” condition, trained interviewers sat closer to interviewees, made
fewer speech errors, and extended the length of the interview. Blind raters (i.e.,
people who knew nothing about the nature of the experiment) coded the inter-
viewees’ behavior, and found that those interviewees who had been interviewed
using the “unfriendly” style performed more poorly in the interviews than White
interviewees who had been treated in the “friendly” style. Similar effects were
observed in experiments manipulating perceivers’ beliefs that they were speak-
ing with either an attractive or unattractive partner48 or to an obese or normal
weight partner. More recently, Serena Chen and John Bargh demonstrated that
African American targets who interacted with White participants who had been
subliminally primed with pictures of African American faces were rated by con-
dition-blind coders as behaving with more verbal hostility than African American
targets who had interacted with Whites primed with Caucasian faces. The tar-
gets themselves had not been primed, and yet their behavior conformed to the
stereotypic expectancies induced in their partners.49
In short, there is ample evidence, far more than can be reviewed here,
suggesting that when subtle expectancy-consistent behaviors “leak out” of
stereotype-holding perceivers, targets respond in kind, often contributing to
stereotype confirmation. While these effects can be moderated by such factors as

48. Allen. R. McConnell and Jill M. Leibold, Relations Among the Implicit Association
Test, Discriminatory Behavior, and Explicit Measures of Racial Attitudes, supra note 45.
49. Mark Chen, and John A. Bargh, Nonconscious Behavioral Confirmation Processes:
The Self-Fulfilling Consequences of Automatic Stereotype Activation, 33 Journal of
Experimental Social Psychology 541–60 (1997).
 the social perceiver: processes and problems in social cognition 329

perceiver self-presentation goals50 or conscious efforts by the target to behave in

expectancy nonconfirming ways,51 interrupting the cycle is difficult. We explore
issues relating to this and other mechanisms for debiasing social interactions in
Section 11.7.

11.5.2 Stereotype Threat

During the 1990s, Stanford psychologist Claude Steele and his colleagues con-
ducted a series of experiments investigating the well-documented tendency of
members of stigmatized groups to underperform their nonstigmatized counter-
parts on tests measuring academic aptitude or achievement.52 In the first of these
studies, Steele and Joshua Aronson gave a difficult verbal test to African American
and European American Stanford undergraduates.53 Half of the students (the
“diagnostic” condition) were told that the test would measure their verbal ability.
The other half (the “nondiagnostic” condition) were told that the purpose of the
experiment was to investigate problem-solving strategies, and that performance
on the test was in no way indicative of aptitude or achievement. In the diagnostic
condition, African American students underperformed their White counter-
parts. However, when the test was presented as nondiagnostic, Black and White
students performed equally well.
Steele and Aronson’s basic finding has now been replicated in a variety of
contexts. Rupert Brown and Robert Josephs varied the perceived diagnosticity of
a difficult mathematics test for males and females with equal backgrounds in
mathematics. Women underperformed when they thought the test was diagnos-
tic, but performed equally well when they believed it was not. In another experi-
ment, the gender effect was shown to be even stronger when the math test was
described as diagnostic of weakness in math, as opposed to strength.54 In other

50. Steven L. Neuberg, Nicole T. Judice, Lynn M. Virdin, and Mary A. Carrillo, Perceiver
Self-Presentation Goals as Moderators of Expectancy Influences: Ingratiation and the
Disconfirmation of Negative Expectancies, 64 Journal of Personality and Social
Psychology 409–20 (1993).
51. James L. Hilton, and John M. Darley, Constructing Other Persons: A Limit on the
Effect, 21 Journal of Experimental Social Psychology 1–18 (1985); William B. Swann,
Jr., and Robin J. Ely, A Battle of Wills: Self-Verification Versus Behavioral Confirmation, 46
Journal of Personality and Social Psychology 1287–1302 (1984).
52. Claude M. Steele, A Threat in the Air: How Stereotypes Shape Intellectual Identity and
Performance of African Americans, 52 American Psychologist 613–29 (1997).
53. Claude M. Steele and Joshua Aronson, Stereotype Threat and the Intellectual
Performance of African Americans, Journal of Personality and Social Psychology 69:
797–811 (1995).
54. Ryan P. Brown, and Robert A. Josephs, A Burden of Proof: Stereotype Relevance and
Gender Differences in Math Performance, 76 Journal of Personality And Social
Psychology 246–57 (1999).
330 problem solving, decision making, and professional judgment

studies, stereotype threat effects have been found in Latinos and low socioeco-
nomic status Whites.55
The way in which a test is framed and the conditions under which it is admin-
istered have a significant impact on whether stereotype threat occurs. For exam-
ple, Steven Spencer, Claude Steele, and Diane Quinn demonstrated that the
women-math stereotype threat effect could be eliminated when a test was
described as being “gender fair.” 56 Becca Levy found that elderly participants
performed more poorly on a memory test after being primed with negative ste-
reotypes of the elderly, but performed better when primed with positive stereo-
types.57 Stereotype threat is also affected by whether stigmatized group status is
made salient before group members take a test. In one particularly fascinating
study, priming Asian American women with female identity before taking a
math test depressed their performance, while priming them with Asian identity
enhanced it.58
The mechanisms underlying stereotype threat are not yet well understood.
Claude Steele hypothesizes that, when members of negatively stereotyped groups
know that their performance is being assessed in a stereotype-relevant domain,
fear of confirming the negative stereotype impairs their performance. The effect,
he suggests, occurs because this sense of self-threat raises anxiety levels and
consumes cognitive resources.59 Over time, Steele notes, the effects of stereotype
threat on performance may cause promising minority and/or female students to
disidentify with certain academic performance domains. Ironically, he notes, the
very members of stigmatized groups who are, at the outset, most talented at a
particular academic endeavor are also the most likely over time to suffer from
stereotype threat and withdraw from academic pursuits in response to its painful
effects.

55. Margaret Shih, Nalini Ambady, Jennifer A. Richeson, Kentaro Fujita, and Heather
M. Gray, Stereotype Performance Boosts: The Impact of Self-Relevance and the Manner of
Stereotype Activation, 83 Journal of Personality and Social Psychology 638 (2002).
56. Steven J. Spencer, Claude M. Steele, and Diane M. Quinn. “Stereotype Threat
and Women’s Math Performance.” Journal of Experimental Social Psychology 35
(1999): 4–28.
57. Becca Levy, Improving Memory in Old Age Through Implicit Self-Stereotyping,
Journal of Personality and Social Psychology 71: 1092–1107 (1996).
58. Id.; Margaret Shih, Todd L. Pittinsky, and Nalini Ambady, Stereotype Susceptibility:
Identity Salience and Shifts in Quantitative Performance, 10 Psychological Science 80–83
(1999).
59. Claude. M. Steele, S. J. Spencer, and Joshua Aronson, Contending with Group
Image: The Psychology of Stereotype and Social Identity Threat, in Advances in Experimental
Social Psychology (Vol. 34), 379–440 (Mark Zanna ed., New York: Academic Press,
2002).
 the social perceiver: processes and problems in social cognition 331

11.6 beyond stereotypes: dispositionism and situationism in

social attribution

The significance we attach to an event depends largely on the causes to which we

attribute it. Without even being aware of it, we constantly ask and answer the
question “why?” “Why did the head of the Terra Nueva Tenants Association cancel
a meeting scheduled with Luis Trujillo?” “Why was Christine Lamm so quiet in
that case strategy meeting this morning?” Human beings are prolific producers
of causal theories. As we move through the social world, we are constantly attribut-
ing causation to other people’s behavior and then using the resulting attributions
to assign responsibility and predict how they will behave in the future.60 Much of
this happens within System 1, automatically and unconsciously.61 Even when it is
deliberate, however, we make systematic errors in attributing causation.
In attributing the causes of other people’s behavior, we tend to search for
explanations that are stable and informative in predicting how they will act in the
future.62 Imagine, for example, that Luis Trujillo and Christine Lamm have a
meeting, which ends with a heated argument about whether Trujillo is obstruct-
ing Lamm’s access to tenants at Terra Nueva by not providing her with a list of
their names, addresses, and phone numbers. At one point, Trujillo explodes,
“You don’t think I care about the tenants? They’re my people.” Christine
responds, “don’t give me that crap! I’ll see you in court.” Lamm leaves the meet-
ing thinking, “So this is a guy with a chip on his shoulder, always ready to play
the race card.” Trujillo leaves with terms like “short-fused,” “paranoid,” “explo-
sive,” and “loose cannon” in his mind. Each of them thinks that the other’s
behavior reflects character traits that will make it very difficult to cooperate going
forward and starts devising strategies to deal with the other.
Like Lamm and Trujillo, most of us tend to attribute behavior to other’s dis-
positions, character traits, or attitudes (real or imagined), and to discount situa-
tional factors—the particular social environment or context in which behavior
occurs—in devising explanations for why a person behaved as he or she did on
a particular occasion.

60. James S. Uleman, Leonard S. Newman, and Gordon B. Moskowitz, People as

Flexible Interpreters: Evidence and Issues from Spontaneous Trait Inference, in Advances in
Experimental Social Psychology (Vol. 28) 211–79 (Mark P. Zanna ed., San Diego:
Academic Press, 1996).
61. Daniel T. Gilbert, Thinking Lightly About Others: Automatic Components of the Social
Inference Process, in Unintended Thought 189–211 (James S. Uleman, and John A.
Bargh eds., New York: Guilford Press, 1989); Yaakov Trope and Thomas Alfieri,
Effortfullness and Flexibility of Dispositional Judgment Processes, 73 Journal of Personality
and Social Psychology 662–74 (1997).
62. Edward E. Jones and Keith E. Davis, From Acts to Dispositions: The Attribution
Process in Social Psychology, in Advances in Experimental Social Psychology (Vol. 2)
219–66 (Leonard Berkowitz ed., New York: Academic Press, 1965).
332 problem solving, decision making, and professional judgment

This is one of the most significant findings of empirical social psychology in

the latter part of the twentieth century. At least in Anglo American culture, the
“intuitive psychologist” tends to overestimate the significance of dispositional
factors and understate the significance of situational factors in attributing the
causes of other people’s behavior. Over time, this significance of this phenome-
non has been judged so important that it has come to be called the “Fundamental
Attribution Error.”63
While we tend to attribute other people’s behavior to dispositions rather than
situations, when it comes to our own behavior, we favor situational attributions
over dispositional ones—particularly when we are explaining our own personal
failings.
This difference of perspective—the so-called “Actor-Observer Effect”64—
explains interpersonal and social pathologies that can ultimately foment con-
flicts and make them more difficult to resolve once they emerge. Consider, for
example, the dynamics leading to the breakdown of many marriages, business
partnerships, and other collaborations. As difficulties are encountered and ten-
sions mount, each party attributes her own behavior to features of the situation,
and the other’s to character flaws and failures. The result is an escalating cycle of
blame, abdication of personal responsibility, and inattention to features of the
environment (overwork, perverse incentive structures), which, if addressed,
might put the conflict in a new light and assist in its resolution.
In the 1950s, Fritz Heider suggested that causal attribution plays a central
role in assigning responsibility for particular outcomes and predicting people’s
behavior. We tend to distinguish between causes that differ on at least two
dimensions. First, we distinguish between causes that are internal and external to
the actor. For example, laziness or disorganization are internal explanations for
tardiness, while getting stuck in a traffic jam caused by an accident is external.
Second, we tend to distinguish between stable and unstable causal factors.
Intelligence, for example, is stable; effort is unstable. These distinctions matter,
Heider observed, because through their application we assign levels of responsi-
bility for outcomes, and we judge how well a person’s past outcomes will predict
their future performance.65

63. Lee Ross, The Intuitive Psychologist and His Shortcomings: Distortions in the Attribution
Process, in Advances in Experimental Social Psychology (Vol. 10) 173–220 (Leonard
Berkowitz ed., New York: Academic Press, 1977).
64. Edward E. Jones and Robert E. Nisbett, The Actor and the Observer: Divergent
Perceptions of the Causes of the Behavior, in Attribution: Perceiving the Causes of
Behavior 79–94 (Edward E. Jones, David E. Kanouse, Harold H. Kelley, Robert E. Nisbett,
Stuart Valins, and Bernard Weiner eds., Morristown, NJ: General Learning Press, 1972).
65. Fritz Heider, The Psychology of Interpersonal Relations (Hoboken, NJ: John
Wiley & Sons, Inc., 1958).
 the social perceiver: processes and problems in social cognition 333

Bernard Weiner and his associates built on Heider’s insights to construct the
Heider/Weiner matrix to show how these parameters affect one’s judgment and
prediction of individuals’ behavior. (See Table 11.1.)

table 11.1 the heider/weiner matrix

Stable Unstable

Skill, laziness Mood, task prioritization

Internal Locus
(dispositional) Higher predictive validity Lower predictive validity
Higher moral responsibility Higher moral responsibility

Family resources and Contributions of others,

responsibilities, stable peer transient peer group
External Locus group influences, enduring influences, extenuating
(situational) commitments circumstances

Higher predictive validity Lower predictive validity

Lower moral responsibility Lower moral responsibility

Behaviors attributed to internal, stable factors (i.e., skill, laziness, criminal pro-
pensity) are viewed both as highly predictive and warranting of moral judgment.
Behaviors attributed to internal, unstable factors (e.g., moods or how one priori-
tizes tasks on a particular occasion) are also viewed as relevant to moral judgment,
but as less predictive of future behavior. Attributions to external, stable factors dif-
fuse moral responsibility, but are seen as predictive, while attributions to external,
unstable factors do not provide a basis for either judgment or prediction.
Biases in causal attribution affect the exercise of professional judgment in
lawyering, judging, policy making, and management, not to mention in our per-
sonal lives. First, our tendency to attribute our own failings or negative outcomes
to situational factors (which we often overoptimistically view as transient) and to
attribute others’ failings or negative outcomes to internal, dispositional factors
makes us vulnerable to a host of ego-enhancing biases. These biases often impair
our willingness or ability to compromise with others, to be open to understand-
ing their points of view, and to predict accurately how well or poorly we or other
people will perform on future tasks.
Second, once we have formed an initial impression of another person, assign-
ing her a set of character traits, those traits operate as prior expectancies that
shape how we attribute her actions and outcomes in the future. Specifically, we
tend to attribute outcomes or behaviors that confirm our impressions to stable,
internal factors (the person’s character), while attributing events that contradict
our prior expectancies to transient or environmental causes. In this way as in
others, our initial person impressions tend to be self-perpetuating.
334 problem solving, decision making, and professional judgment

11.7 debiasing social judgment

First, the bad news. Social psychology can presently tell us far more about how
social perception falls prey to the biasing effects of stereotypes and other social
expectancies than about how those biases can reliably be reduced or eliminated.
But there is good news as well. First, we do know something about bias reduction.
And second, debiasing is presently a subject of intense investigation in cognitive
social psychology, and it is reasonable to expect that a decade from now we will
know much more than we know now about how to constrain what psychologist
John Bargh calls “the cognitive monster.”66

11.7.1 Dual Process Models of Social Cognition: High Road vs.

Low Road Thinking
Much of the existing bias reduction research draws on a dual process model of
social cognition, essentially the System 1/System 2 framework of automatic and
controlled processing described in Section 1.6.1. In making sense of other
people, we use two systems of information processing: a “low road” system,
which is automatic, rapid, unconscious, and demanding of few cognitive
resources, and a “high road” system, which is effortful, conscious, controlled,
and resource-intensive.
Social attribution processes, described earlier in this chapter, fit easily into
this dual process framework. Let’s return to Trujillo’s thoughts about his interac-
tion with Lamm. Cognitive social psychologists67 suggest his thought process
might have gone through the following steps. First, he would automatically
categorize Lamm’s behavior as “explosive.” Next, from this automatic categoriza-
tion, he would spontaneously characterize Lamm as a “loose cannon.” However,
if he thought about this characterization more systematically, he might correct
his initial, spontaneous characterization after considering the possibility that
Lamm was provoked by his own behavior. If he happened to know that she was
coping with the mayor’s demands for immediate action, he might also consider
the possibility that her behavior was attributable to tensions caused by the
immediate situation.

66. John A. Bargh, The Cognitive Monster: the Case against the Controllability of Automatic
Stereotype Effects, in Shelly Chaiken and Yaacov Trope, Dual-Process Theories in
Social Psychology 361 (New York: Guilford Press, 1999).
67. James S. Uleman, Leonard S. Newman, and Gordon B. Moskowitz, People as
Flexible Interpreters: Evidence and Issues from Spontaneous Trait Inference, in Advances in
Experimental Social Psychology (Vol. 28) 211–79 (Mark P. Zanna ed., San Diego:
Academic Press, 1996); Timothy D. Wilson and Nancy Brekke, Mental Contamination and
Mental Correction: Unwanted Influences on Judgments and Evaluations, 116 Psychological
Bulletin 117 (1994); Daniel T. Gilbert, Brett W. Pelham, and Douglas S. Krull, On
Cognitive Busyness: When Person Perceivers Meet Persons Perceived, 54 Journal of
Personality and Social Psychology 733–40 (1988).
 the social perceiver: processes and problems in social cognition 335

Notice, however, that correction of this kind would require three things:
Trujillo’s motivation to reconsider his initial characterization; the cognitive
resources (time, spare processing capacity) required to do so; and information
from which a reasoned corrective judgment might be made.
There is virtual consensus among social psychologists that stereotyping
includes both automatic and controlled components. If a person has learned, but
then consciously rejects, the stereotypes and attitudes associated with a devalued
social group, those stereotypes and attitudes do not just disappear. Indeed, tech-
niques like the Implicit Association Test68 show them to be present, even in
many participants who rate low on measures of explicit bias or who are them-
selves members of the negatively stereotyped group.69 In such situations, the
older implicit attitudes and associations continue to exist alongside the newer,
consciously held beliefs and commitments. The implicit attitudes function as a
Type 1 or “low road” system, while the conscious beliefs function as a Type 2 or
“high road” system. In fact, the neurological substrates underlying the implicit/
explicit distinction can even be observed through such technologies as functional
magnetic resonance imaging (fMRI).70
There is relative consensus within cognitive social psychology that stereotype
activation, when it occurs, is spontaneous. But there is also a growing consensus
that stereotype activation does not necessarily lead to stereotype expression, and
that stereotype activation itself can be affected by environmental factors.
To understand the state of current thinking on these issues, consider Figure
11.6, an extension of Jerry Kang’s model of social categorization portrayed ear-
lier in Figure 11.1.
As Figure 11.6 suggests, even where a social schema (like a stereotype) is
activated and generates a schema-consistent impression, that initial impression
can be overridden through the application of conscious, “high road” thinking.71

68. See note 44, supra. A great deal has been written about the Implicit Association Test.
For a relatively brief description of the instrument, see Anthony G. Greenwald & Linda
Hamilton Krieger, Implicit Bias: Scientific Foundations, 94 Cal. L. Rev. 945–967 (2006).
69. Nilanjana Dasgupta, Implicit Ingroup Favoritism, Outgroup Favoritism, and Their
Behavioral Manifestations, 17 Social Justice Research 143 (2004); Brian A. Nosek,
Mahzarin R. Banaji, and Anthony G. Greenwald, Harvesting Implicit Group Attitudes and
Beliefs from a Demonstration Web Site, 6 Group Dynamics 101 (2002).
70. William. A. Cunningham, Marcia K. Johnson, Carol L. Raye, J. Chris Gatenby,
John C. Gore and Mahzarin R. Banaji, Neural Components of Social Evaluation, 85 Journal
of Personality and Social Psychology 639–49 (2003); E. A. Phelps, et al., Performance
on Indirect Measures of Race Evaluation Predicts Amygdala Activation, supra note 47.
71. Margo J. Monteith and Corrine I. Voils, Exerting Control over Prejudiced Responses,
in Cognitive Social Psychology: The Princeton Symposium on the Legacy and
Future of Social Cognition (Gordon B. Moskowitz ed., Mahwah, NJ: Erlbaum, 2001).
336 problem solving, decision making, and professional judgment

Categorized Priming Effects

through Mapping Rules Which
automatically
Person activates/primes
Perceived Which moderates
schema’s effect on
judgment of Social
Schemas
Conscious Correction
Which alters
judgment of
Low Road

and interaction Which

with potentiate

d
Which May

oa
Trigger

R
gh
Hi
Spontaneous Schematic Content
Impression
Which
generates

figure 11.6 a dual-process model of social perception and judgment:

high road vs. low road social information processing.

However, certain conditions must be present.72 The perceiver must be aware of

the possibility that his initial impression might be biased, and he must be moti-
vated to correct for that bias. Of course, some behaviors, like a decision to lease
an apartment to a particular tenant, are under greater conscious control than
behaviors like facial expressions, speech errors, and fidgeting.73 For correction to
occur, the perceiver must also have time and attention to spare: like all System 2
processing, high road correction requires cognitive resources. In certain
situations, as when a police officer has to decide whether someone is pointing a
gun at her, there may be little time available, thus reducing the likelihood that
correction will occur. And finally, to correct an initially biased impression, the
perceiver must have available the information and analytical tools required for
meaningful deliberation.

72. John A. Bargh, The Cognitive Monster: The Case Against the Controllability of
Automatic Stereotype Effects, in Shelly Chaiken and Yaacov Trope, Dual-Process
Theories in Social Psychology 361 (1999); Timothy D. Wilson and Nancy Brekke,
Mental Contamination and Mental Correction: Unwanted Influences on Judgments and
Evaluations, 116 Psychological Bulletin 117 (1994).
73. There is some evidence indicating that, with sufficient motivation and effort on the
perceiver’s part, even these subtle behaviors are amenable to conscious control. See Steven
L. Neuberg, Expectancy-Confirmation Processes in Stereotype-Tinged Social Encounters: The
Moderating Role of Social Goals, in The Psychology Of Prejudice: The Ontario
Symposium (Vol. 7) 103–30 (Mark P. Zanna and John M. Olson eds., Hillsdale, NJ:
Erlbaum, 1994).
 the social perceiver: processes and problems in social cognition 337

Consider how the presence or absence of these factors might influence the
degree of racial bias present in hiring decisions. If those who make hiring deci-
sions are simply told to be “color-blind” in evaluating potential candidates, if
they do not believe themselves to be race-biased, and if there are no meaning-
fully endorsed and enforced system of goals relating to minority hiring, con-
scious correction of implicitly race-biased evaluations is unlikely to occur. Under
such conditions, hiring decision makers are unlikely to be aware of the possibil-
ity that implicit racial stereotypes and attitudes could be skewing their judgment,
and they will have little motivation to engage in conscious correction.
In the hiring or promotion context, at least with respect to high-level jobs,
ultimate decision making is often quite deliberate, with ample time available
for systematic evaluation of the final competing candidates. In the formation
of managers’ day-to-day impressions of employee performance, however, this
may not be the case. Managers often function under conditions of extreme time
constraint and high cognitive load. In these situations, correction of stereotype-
influenced impressions is far less likely to occur.74
As earlier mentioned, to correct biased impressions, decision makers also
need ample decision-relevant information and access to a structured decision-
making processes. Carefully spelling out the applicable evaluative criteria, pro-
viding decision makers with objective, criterion-relevant information, and
requiring them to write down how the information provided relates to the rele-
vant criteria can help reduce the biasing effect of stereotypes on evaluations.75
Additionally, providing decision makers with systemwide information revealing
broad patterns of good or bad outcomes can help them identify biases they likely
would not recognize from case-by-case data.76 Finally, as with any expectancy,
“considering the opposite” can help correct expectancy-confirmation bias.
In summary, even though the implicit expectancies created by stereotypes
and other social knowledge structures bias initial impression formation
spontaneously, bias is, at least under certain conditions, amenable to correction.
Individuals or organizations interested in maximizing the likelihood that correc-
tion will occur should implement the following strategies:
• educate decision makers that their evaluative judgments may be influenced
by subtle forms of bias, no matter how sincere their conscious commit-
ment to fairness, objectivity, and nondiscrimination;

74. John A. Bargh, The Cognitive Monster: The Case Against the Controllability of
Automatic Stereotype Effects, in Shelly Chaiken and Yaacov Trope, Dual-Process
Theories in Social Psychology 361 (New York: Guilford Press, 1999); Daniel T. Gilbert,
Brett W. Pelham, and Douglas S. Krull, On Cognitive Busyness: When Person Perceivers Meet
Persons Perceived, 54 Journal of Personality and Social Psychology 733–40 (1988).
75. Veronica F. Nieva and Barbara A. Gutek, Sex Effects on Evaluation, 5 Acad. of
Mgmt. Rev. 267, 270–71 (1980).
76. Faye J. Crosby, Susan Clayton, Olaf Alksnis, and Kathryn Hemker, Cognitive Biases
in the Perception of Discrimination: The Importance of Format, 14 Sex Roles 637–46 (1986).
338 problem solving, decision making, and professional judgment

• structure incentives and foster an organizational culture that motivates

decision makers to identify and correct implicitly biased impressions or
initial judgments;
• destigmatize implicit bias, so that self-examination and conscious correc-
tion are less threatening;
• monitor decisions systematically, so that broad patterns suggestive of
uncorrected bias can be identified and addressed;
• reduce, where possible, decision makers’ levels of cognitive busyness when
making decisions;
• generate clear, objective evaluative criteria, and supply evaluators with
information relevant to those criteria; and
• encourage decision makers to “consider the opposite” before acting on or
attempting to justify an initial impression.

11.7.2 How Malleable Are Implicit Stereotypes?

The bias-reduction strategies discussed above all involve the conscious correc-
tion of an initial judgment that was biased by an implicit stereotype or other
social expectancy. But can the implicit stereotypes we have acquired from the
surrounding culture be suppressed, or better yet, replaced, through individual or
organizational efforts? In other words, how malleable are implicit attitudes?
In short, we don’t yet know. Indeed, these two questions now represent per-
haps the hottest issues in social psychology.
In the early 1990s, few established empirical social psychologists believed that
the automatic components of the stereotyping process (stereotype activation and
the spontaneous formation of stereotype-consistent impressions) could be
defused through individual efforts.77 Debiasing, if it occurred at all, was thought
to require motivated, effortful, subsequent correction.78 But by the mid-1990s, a
new generation of graduate students was chafing at the field’s collective surren-
der to the inevitability of automatic stereotype activation, and these students
launched a new research agenda on the malleability of implicit attitudes.
Early results emerged from Mahzarin Banaji’s lab at Yale, when in 1996 grad-
uate student Sandra Blair and Banaji reported that, under conditions of low cog-
nitive load, participants who had rehearsed counter-stereotypic expectations

77. John A. Bargh, supra note 74, The Cognitive Monster: The Case Against the
Controllability of Automatic Stereotype Effects, in Shelly Chaiken and Yaacov Trope,
Dual-Process Theories in Social Psychology 361 (1999); Susan T. Fiske, Examining
the Role of Intent: Toward Understanding Its Role in Stereotyping and Prejudice, in
Unintended Thought, supra note 62, pp. 253–83.
78. Patricia G. Devine, Stereotypes and Prejudice: Their Automatic and Controlled
Components, supra note 8; Susan T. Fiske, Monica Lin, and Steven L. Neuberg, The
Continuum Model: Ten Years Later, in Dual-Process Theories in Social Psychology
231–54 (Shelly Chaiken and Yaacov Trope eds., New York: Guilford Press, 1999).
 the social perceiver: processes and problems in social cognition 339

showed reduced automatic activation of implicit gender stereotypes.79 Similar

results were reported by a second research team in 2000.80
Subsequent experiments indicated that automatic stereotype activation could
be reduced when participants were exposed to positive exemplars of a disliked
group and negative exemplars of a liked group,81 concentrated on counter-
stereotypic mental images82 or received positive feedback from a stereotyped
group member.83 Similar decreases in automatic stereotype activation resulted
where participants were placed in low-status positions relative to a stereotyped
target,84 when the experiment was run by a member of the stereotyped group,85
and when participants believed that their levels of automatic stereotyping were
“out of step” with those of their peers.86
These studies, and others like them, provide tantalizing—if preliminary—-
evidence that automatic stereotype activation can be reduced through environ-
mental modifications and the rehearsal of counter-stereotypic reactions. But the
research case should not be overstated. In the studies described above, effect
sizes were relatively small, and the observed debiasing effects relatively short-
lived. Moreover, in some of the studies, reduced stereotype activation was found
only under conditions of low cognitive busyness, an increasingly rare luxury in a
hypercharged social world.
Though debiasing research is still in its infancy, there are certainly some
steps that individuals and organizations motivated to reduce their levels of
implicit bias can take to achieve that end. We can control the images and other

79. Irene V. Blair and Mahzarin R. Banaji, Automatic and Controlled Processes in
Stereotype Priming, 70 Journal of Personality and Social Psychology 1142–63 (1996).
80. Kerry Kawakami, John F. Dovidio, J. Moll, S. Hermsen, and A. Russin, Just Say No
(to Stereotyping): Effect of Training in the Negation of Stereotypic Associations on Stereotype
Activation, 78 Journal of Personality and Social Psychology 871–88 (2000).
81. Nilanjana Dasgupta and Anthony G. Greenwald, On the Malleability of Automatic
Attitudes: Combating Automatic Prejudice with Images of Admired and Disliked Individuals,
81 Journal of Personality and Social Psychology 800–14 (2001).
82. Irene V. Blair, Jennifer E. Ma, and Alison P. Lenton, Imagining Stereotypes Away:
The Moderation of Implicit Stereotypes Through Mental Imagery, 81 Journal of Personality
and Social Psychology 828–41 (2001).
83. Lisa Sinclair and Zeva Kunda, Reactions to a Black Professional: Motivated Inhibition
and Activation of Conflicting Stereotypes, 7 Journal of Personality and Social
Psychology 885–904 (1999).
84. Jennifer A. Richeson and Nalini Ambady, When Roles Reverse: Stigma, Status, and
Self-Evaluation, 31 Journal of Applied Social Psychology 1350–78 (2001).
85. Brian S. Lowery, Curtis D. Hardin, and Stacey Sinclair, Social Influence Effects on
Automatic Racial Prejudice, 81 Journal of Personality and Social Psychology 842–55
(2001).
86. Gretchen B. Sechrist and Charles Stangor, Perceived Consensus Influences Intergroup
Behavior and Stereotype Accessibility, 80 Journal of Personality and Social Psychology
645–54 (2001).
340 problem solving, decision making, and professional judgment

messages we absorb from the media, making conscious decisions to limit expo-
sure to negatively stereotyped images of women and minorities, or making sure
that we supplement those we do take in with positive, non-stereotyped portrayals
as well. Organizations can take affirmative steps to ensure that members of ste-
reotyped groups are present in leadership positions. We can change the pictures
on the walls of our homes, workplaces, and schools, even on our screen savers.
These situational modifications can’t hurt, and in another decade, empirical
research may demonstrate that, alone or in combination with other factors, they
help a great deal.
part three
making decisions

People make myriad decisions every day. Most are of little consequence: What
clothes should I wear to work? What should I order for lunch? Other decisions
carry more weight: What job should I take? Should I enter into this deal, and on
what terms? What litigation strategy should I employ? Where should we site the
wastewater treatment plant? It is the presence of options that creates both oppor-
tunities and a decision problem.
Part 1 introduced a general framework for deliberative decision making. To
focus specifically on decision making (rather than problem solving in general),
the framework entails:
1. Identifying and prioritizing the relevant interests and objectives.
2. Generating a range of plausible decision alternatives. (Sometimes it’s a
simple “yes” or “no” decision; sometimes, as in the wastewater treatment
plant decision, there are numerous alternatives.)
3. Predicting the consequences of those alternatives in terms of the interests
or objectives.
4. Comparing the alternatives and selecting the course of action that opti-
mizes the interests or objectives.
We covered the first two of these in Part 1. The third is fundamentally an
empirical undertaking, often requiring probabilistic assessments of the kinds
explored in Part 2 of the book. A decision process is good to the extent that the
underlying empirical analysis follows the scientific method and that judgments
of uncertain consequences adhere to the rules of probability theory.
We got a head start on the fourth step in Chapter 4, with the survey of various
strategies for making decisions, and working through an example of a formal
model of decision making—the subjective linear model—in which the decision
making specifies the major interests involved and assigns weights to them. We
continue this exploration in Part 3.
Chapter 12, Choices, Consequences, and Trade-offs, asks what people seek to
achieve when they are making decisions. It introduces the concept and com-
plexities of utility and the criteria for rational decision making, focusing on a
widely accepted set of axioms for rational decision making—the axioms of subjec-
tive expected utility theory.
342 problem solving, decision making, and professional judgment

Chapter 13, Complexities of Decision Making: Decision Processes; Relationships to

our Future Selves, and Chapter 14, Complexities of Decision Making, Continued:
The Power of Frames, ask how one would know whether a person has made a
suboptimal decision in terms of his or her (own) subjective utility, and then
examine a variety of situations in which people tend to make suboptimal
decisions.
Chapter 15, Decision Making under Risk, considers issues of decision making
when risk or uncertainty about the consequences of a decision plays a major
role.
And Chapter 16, The Role of Affect in Risky Decisions, continues this discus-
sion, focusing on risk situations where emotions run high.
12. choices, consequences, and trade-offs

When you see a fork in the road, take it

—Yogi Berra

In Chapter 4, we surveyed a number of different strategies for choosing among

alternatives and worked through an example of a formal model of decision
making—the subjective linear model—in which the decision maker specifies the
major interests involved and assigns weights to them. Implicit in these strategies is
the idea that a decision should maximize someone’s utility—one’s own when
making a personal decision, the client’s when counseling him on a course of action,
or the utility of constituents and other stakeholders when making a policy decision.
The complexities of maximizing the utility of multiple people who may have
different interests are readily apparent. Different residents of Edenville—even
people living in the same neighborhood—may have different views about the
importance of the various factors implicated by the siting of the wastewater treat-
ment plant. Some may care deeply about protecting the habitat in the north part
of the city; others may be completely indifferent. Moreover, the decision of where
to site the plant will affect the residents in different ways.
How elected or appointed officials should deal with their constituents’ vari-
ous interests is a contested issue of political theory and beyond the scope of this
book.1 As we will see in the next several chapters, the concept of utility turns out
to be pretty complex, even in the simple case where only one person’s interests
are concerned. This chapter begins by unpacking this concept, and then intro-
duces a widely accepted set of axioms for rational decision making—the axioms
of subjective expected utility theory.

12.1 subjective utility

This section is something of a grab-bag set of observations about subjective

utility. It covers the following points:
• Utility is essentially a subjective matter and often includes nontangible values.
• The decision process itself affects utility, directly and through the percep-
tion or experience of outcomes

1. See, e.g., Hannah Pitkin, The Concept of Representation (Berkeley: University

of California Press, 1967), Adam Przeworski, Susan Stokes, and Bernard Malin, eds.,
Democracy, Accountability and Representation (Cambridge, UK: Cambridge
University Press, 1999).
344 problem solving, decision making, and professional judgment

• One’s utility is affected by looking backward and forward in time as well as

by present experience—and the relationship among these timeframes can
be complex.
• More money and other goods do not necessarily affect utility linearly or
even montonically. (We return to what Barry Schwartz has called the “para-
dox of choice”2 in Section 14.5.)
• People tend to adapt to both good and bad life circumstances. (In the next
chapter, we consider how good people are at predicting this adaptation.)

12.1.1 Sources of Utility

12.1.1.a The Arationality of Ends The goal of this book is to provide insights
and tools to help you achieve your ends or assist others in achieving theirs, with
the ultimate aim of maximizing your or their utility. The premise of our venture
is captured by Mark Kelman’s description of rational choice theory:
At the most basic level, rational choice theory posits the existence of princi-
pals who articulate ends that are beyond rational reproach. Rational choice
theorists view these ends as arbitrary—that is to say, the ends need no justifi-
cation. Ends are also subjective; they are not things that exist in the external
world that individuals apprehend, as they apprehend the natural world around
them, but rather come from the subject herself. Whether an end is correct or
not can be judged only by each individual subject herself, and the judgment
of whether an end is correct is extremely limited. In essence, the subject can
do no more than interrogate the authenticity and accuracy of the way in which
he has articulated and come to understand his wishes. Ends cannot be judged
from an external observer’s objective viewpoint to be wrong in some transcen-
dent and impersonal sense because they do not manifest what is most unique
to human nature or interfere with a particular conception of human flourish-
ing. Nor are ends derived by observing some external perceptible human
nature. Ends are not only private and individual but, perhaps most impor-
tantly, individuating. What most clearly makes us differentiable individuals is
our unique preference structure, even more so than our distinct capabilities
or capacities to meet our ends. In other words, who we uniquely are is, at
core, a reflection of who we want to be and what we aspire to attain, rather
than what we can actually attain. . . .
At the normative level, the existence of these differentiable tastes and ends is
foundational for all political theories that would describe themselves as essen-
tially liberal. A host of political theorists define liberal theory in terms of its
skepticism about theories that there are good lives, rather than mere tastes for
end states. Less negatively, these theorists define liberal theory in terms of

2. Barry Schwartz, The Paradox of Choice: Why More Is Less (New York: Ecco/
Harper Collins Publishers, 2004).
 choices, consequences, and trade-offs 345

theories that affirm that there are a multitude of good lives and that each individ-
ual is in a unique position to ascertain which of the many reasonable conceptions
of the “good” best fit her circumstances and express her individuated soul.3
Daniel Keys and Barry Schwartz write that a complete theory of rational deci-
sion making “must consider the very broadly construed consequences of a deci-
sion. That is, it must consider short- and long-term consequences, consequences
to the self and to others, consequences that are central to the decision at hand,
and consequences that may be more peripheral. It must also consider conse-
quences of decisions for the character of the decision maker, as the effects on
character may have a significant impact on a host of future decisions.”4 The next
several chapters suggest how difficult it is for even very thoughtful people to
predict whether particular courses of action will serve their individual ends, let
alone those of various stakeholders.

Gross National Happiness

The official policy of Bhutan is that “gross national happiness is more important
than gross national product,” because “happiness takes precedence over eco-
nomic prosperity in our national development process.”5
Bhutan’s constitution assesses government programs in terms of the economy,
culture, environment, and good governance. The government employs seventy-
two indicators in nine different domains: psychological well-being, ecology, health,
education, culture, living standards, time use, community vitality, and good gover-
nance. For example, indicators of psychological well-being include the frequencies
of prayer and meditation and of feelings of selfishness, jealousy, calm, compassion,
and generosity, and the absence of frustration and suicidal thoughts.6

People sometimes use “happiness” as a synonym for utility, but that is some-
what misleading. Many people have a conception of leading a “good” or “whole”

3. Mark G. Kelman, Law and Behavioral Science: Conceptual Overviews, 97 Nw. L. Rev.
1347, 1358 (2003).
4. Daniel J. Keys and Barry Schwartz, “Leaky” Rationality: How Research on Behavioral
Decision Making Challenges Normative Standards of Rationality, 2 Perspectives on
Psychological Science 162–80 (2007).
5. Orville Schell, “Frontline World, Bhutan”—The Last Place, May 2002, http://www.
pbs.org/frontlineworld/stories/bhutan/gnh.html.
6. Seth Mydans, Thimphu Journal: Recalculating Happiness in a Himalayan Kingdom,
New York Times, May 7, 2009, http://www.nytimes.com/2009/05/07/world/asia/07bhutan.
html
346 problem solving, decision making, and professional judgment

or “balanced” life that seems trivialized by reducing it to happiness. More broadly,

as Kelman has observed, “ascetics, saints, and those willing to sacrifice every-
thing for spite all have perfectly coherent utility seeking goals.”7
In Because It Is There: The Challenges of Mountaineering . . . for Utility Theory,8
George Loewenstein examines the diaries and memoirs of individuals on seri-
ous mountain climbing explorations, “which tend to be one unrelentless misery
from beginning to end.” Drawing on Jeremy Bentham’s categories in The
Principles of Morals and Legislation (1789), Loewenstein discerns a number of dif-
ferent motivations other than the immediate experience of thrills, natural beauty,
and the like.
• Mountaineers seek Bentham’s “pleasure of a good name”—the desire to
impress others.
• They seek the “pleasure of self recommendation” or what Loewenstein
calls self-signaling—the need for self-esteem or to define oneself as having
a particular set of character traits.
• They are motivated by “the almost obsessive human need to fulfill self-set
goals,” which may also be an aspect of self-signaling.
• They are motivated by the desire to master their environments, by the need
for control.
• And they seek experiences—often near-death experiences—that will give
meaning to their lives.
All of this might lead one to define utility, as does welfare economics, simply
in terms of what an individual wants, and to specify the goal of decision making
as satisfying those wants. However, people may be mistaken about what they
want—they may “miswant.”9 But that’s for the next chapter. Rather than linger
on the epistemology of utility,10 the remainder of this chapter explores some of
its major dimensions.

7. Kelman, supra at 1364 n.31.

8. George Loewenstein, Because It is There: The Challenges of Mountaineering . . . for
Utility Theory, 52 Kyklos 315 (1999).
9. See Daniel T. Gilbert and Timothy D. Wilson, Miswanting: Some Problems in the
Forecasting of Future Affective States, in Feeling and Thinking: The Role of Affect in
Social Cognition, (Joseph Forgas ed., New York: Cambridge University Press, 2001).
10. Mark G. Kelman, Hedonic Psychology and the Ambiguities of “Welfare,” 33
Philosophy & Public Affairs 413 (2005).
 choices, consequences, and trade-offs 347

Patrick has a serious phobia about flying. The very anticipation of flight creates an
almost physical sense of dread. As a result, he avoids business and pleasure trips
that require flying, and he drives extraordinarily long distances, all at tremendous
professional and personal costs. Patrick believes that statistics show that flying is
safe compared to driving and, indeed, to many other activities he engages in. He
acknowledges that the phobia is irrational. Is it irrational for him to take account
of his fear of flying in making travel decisions?

The following sections note some other sources of utility that do not seem
reducible to purely material ends.
12.1.1.b Existence and Option Value Considerable research has gone into
measuring the material benefits of ecosystems—for example, as sources of pure
drinking water. But some environmental legislation, such as the Endangered
Species Act, treats species, habitats, and natural landscapes as having an exis-
tence value independent of any human uses—for example, the value attributed to
the protection of an endangered species in a remote part of the world that most
people will never encounter. A cousin of existence value is the option value of a
natural resource—the value one assigns to the possibility of using it sometime
in the future. For example, you may have no plans to visit a wilderness area but
nonetheless place some value on the possibility of visiting it in the future. We’ll
say more about how economists assess these values in the discussion of cost-
benefit analysis in Section 12.3.
12.1.1.c Affection In a charming mind experiment, Christopher Hsee and
Howard Kunreuther demonstrate what they call the affection effect.11
On a visit to Europe, Helen and Laura each bought a painting for $100 and
shipped it back to the United States, each purchasing a $100 shipping insur-
ance policy. The paintings were damaged beyond repair, but claiming the
insurance will require a time-consuming trip to the shipping company. Helen
loved her painting a great deal. Laura thought hers was just OK. Who is more
likely to spend the time collecting the insurance compensation?
From a purely economic point of view, there is no difference between the
choices facing Helen and Laura: $100 is $100, whatever its source and rationale.
However, Helen, who has great affection for the painting, is more likely to take

11. Christopher K. Hsee and Howard C. Kunreuther, The Affection Effect in Insurance
Decisions, 20 Journal of Risk and Uncertainty 141 (2000). We have simplified the
scenarios.
348 problem solving, decision making, and professional judgment

the time to collect the insurance compensation than Laura. Hsee and Kunreuther
explain the difference in terms of what they call the “consolation hypothesis”:
Compensation has symbolic value. It is redemption for the lost object; it is a
token of consolation for the pain caused by the loss. The utility of a given
amount of compensation has two components: (a) the utility of that particular
amount of money, and (b) the utility of its symbolic consolation value. . . . The
consolation value, in turn, depends on one’s level of affection for the object.
The more affection one has, the more pain one experiences, and the more one
needs the consolation.
The main focus of this chapter is on decisions made in advance of knowing
just what the consequences will be—for example, whether one will suffer a loss.
But just as individuals anticipate the experience of regret, they can anticipate the
need for consolation. In another part of the experiment, Hsee and Kunreuther in
effect asked Laura and Helen how much they would pay (in advance) for insur-
ance to ship their paintings. Not surprisingly, Helen, who has strong affection
for her work of art, would pay significantly more than Laura—almost twice as
much. By the same token, people would be more willing to buy a warranty for a
beautiful used convertible than for a utilitarian used station wagon even if the
expected repair expenses and the cost of the warranty were identical.12

Suppose that your dog is dying of either Virus A or Virus B. They are equally likely,
but which one it is cannot be known. There are different medicines for Virus A
and Virus B, each equally effective for its intended target, that cost $1 and $20
respectively. They interact in a way that they cannot be given together. Which one
would you choose?
When the dog was described as friendly or faithful, 43 percent of respondents
would have opted for the $20 medicine and said that they would have felt better
paying more. The effects were significantly diminished when the dog was
described as unfriendly or unfaithful.
Christopher Hsee and Howard Kunreuther, The Affection Effect in Insurance
Decisions, 20 JOURNAL OF RISK AND UNCERTAINTY 141 (2000).

12.1.1.d The Allure of the Free Suppose you were offered the choice between
buying a Hershey’s Kiss for 1 cent and buying a delicious Swiss truffle, made
by Lindt, for 15 cents; which would you choose? Now imagine that instead of

12. Christopher K. Hsee, and Geeta Menon, Affection Effect in Consumer

Choices, unpublished data (University of Chicago, 1999).
 choices, consequences, and trade-offs 349

costing 1 cent, the Hershey’s Kiss were free and the Lindt truffle cost 14 cents.
Would your preference be different? Research by Dan Ariely and his colleagues13
suggests that it would. When they sold Hershey’s Kisses and Lindt truffles to the
public for 1 cent and 15 cents respectively, 73 percent of their “customers” went
for the Lindt truffle, which, considering their usual retail prices, is a much better
deal. However, when they dropped the price of each chocolate by a penny, so the
Kiss was free and the truffle cost 14 cents, only 31 percent opted for the truffle.
The rest—a large majority—now chose the Kiss.
The relative value of the two chocolates had not changed, and nor had the
price difference. So why did people’s choices change? Ariely and his colleagues
suggest that we enjoy a benefit when it is untainted by costs, which take an affec-
tive toll that compromises our enjoyment of the benefit. (By the same token, we
may enjoy a vacation more if we de-couple the experience from its cost by paying
in advance.)

12.1.2 The (Dis)utility of the Decision Process—And “Leaky” Decisions

Regardless of whether utility can be reduced to happiness, or is more complex
and nuanced, most discussions of utility focus on the utility of the outcome. The
decision process is thus evaluated solely on its contribution to the outcome.
However, the process of decision making is often itself a cost—and one that
must be weighed against the likelihood of its contributing to a better outcome.
(While Christine Lamm’s use of the subjective linear model was helpful in decid-
ing where to site a wastewater treatment plant, imagine her employing that
approach in selecting a restaurant, let alone in choosing from a menu. In addi-
tion to the time and money involved, some decision processes carry emotional
costs, such as the stress of making difficult trade-offs.
Beyond these obvious points, though, the decision process may itself be a
source of utility or disutility. To use Daniel Keys and Barry Schwartz’s evocative
phrase, the process may “leak” into the outcome.14
12.1.2.a Fairness People attribute value to the fairness of a decision process.
Litigants are more willing to accept and abide by an adverse judgment if they
believe that the process was fair, and this may be true for citizens who must
live with the outcomes of policy decisions, such as siting a wastewater

13. Kristina Shampanier, Nina Mazar, and Dan Ariely, Zero as a Special Price: The True
Value of Free Products, 26 Marketing Science 742–57 (2007); Dan Ariely, Predictably
Irrational: The Hidden Forces That Shape Our Decisions (New York: HarperCollins
2008).
14. Daniel J. Keys and Barry Schwartz, “Leaky Rationality: How Research on Behavioral
Decision Making Challenges Normative Standards of Rationality, 2 Perspectives on
Psychological Science 162–80 (2007). The authors acknowledge Deborah Frisch’s
article, Reasons for Framing Effects, 54 Organizational Behavior and Human Decision
Processes 399–429 (1993), for the concept of what they term “leakage.”
350 problem solving, decision making, and professional judgment

treatment plant.15 In a summary of the research, Donna Shestowsky suggests

that people define fairness in terms of how much control they retain over the
development of information that will be used to resolve an issue, and how
respectfully the decision maker treats them.16
People’s behavior in the so-called “ultimatum game” provides a dramatic
example of the role of fairness in decision making. In this game, two players are
given a sum of money to split between them. The “proposer” suggests a way to
split the sum. The “responder” can then either accept or reject this offer. If the
responder accepts the offer, the sum is distributed to both players as offered. If
the responder rejects the offer, neither player receives any of the money.
If the responder’s only utility consisted of the money gained, he would accept
a penny regardless of how small a percentage it was of the total sum; if the pro-
poser believed that the responder would act in this way or if the proposer’s only
utility was in money gained, she would offer only a penny. In fact, proposers usu-
ally offer 30 to 40 percent, and responders usually reject offers of less than 20
percent.17 The results seem to be the same when one hundred dollars are at stake
as for ten dollars.18 Cross-cultural comparisons of ultimatum game experiments
suggest that people from Ljubljana, Pittsburgh, and Tokyo play the game with
similar results, although people may play differently in traditional societies.19

The “dictator” game controls for the proposer’s anticipation of the respondent’s
response. The dictator divides a fixed amount of money between herself and the
“recipient,” who cannot respond. When the game is played anonymously with
participants knowing nothing about the other player, most dictators keep the
entire sum or allot a paltry amount to the recipient.20 However, when dictators are
provided information about recipients—pictures or surnames, for example—they

15. Jonathan D. Casper, Tom Tyler, and Bonnie Fisher, Procedural Justice in Felony
Cases, 22 Law and Society Review 483–508 (1988).
16. Donna Shestowsky, Misjudging: Implications for Dispute Resolution, 7 Nev. L.J. 487,
493 (2007).
17. Colin Camerer and Richard H. Thaler, Anomalies: Ultimatums, Dictators and
Manners, 9 The Journal of Economic Perspectives 210 (1995).
18. Elizabeth Hoffman, Kevin A. McCabe, and Vernon L. Smith, On Expectations
and the Monetary Stakes in Ultimatum Games, 25 International Journal of Game
Theory 289–301 (1996).
19. Alvin E. Roth, Vesna Prasnikar, Masahiro Okuno-Fujiwara, and Shmuel Zamir,
Bargaining and Market Behavior in Jerusalem, Ljubljana, Pittsburgh, and Tokyo: An
Experimental Study, 81 American Economic Review 1068–95 (1991).
20. Elizabeth Hoffman, Kevin McCabe, Keith Shachat, and Vernon Smith, Preferences,
Property Rights, and Anonymity in Bargaining Games, 7 Games and Economic Behavior
346–80 (1994); Elizabeth Hoffman, Kevin McCabe, and Vernon Smith, Social Distance and
Other-Regarding Behavior in Dictator Games, 86 American Economic Review 653–60 (1996).
 choices, consequences, and trade-offs 351

increase their giving.21 The dictator’s behavior is also affected if she adverts to
being observed. In an experiment where players conducted the game on comput-
ers to ensure anonymity, dictators gave more when the background of the com-
puter screen featured a robot with human eyes.22 Commenting on the ultimatum
game, Martin Nowak suggests that objecting to unfairness is a way of asserting
one’s reputation: “Rejecting low offers is costly, but the cost is offset by gaining
the reputation of being somebody who insists on a fair offer.”23 In the dictator
game, perhaps the image of the observer reminds the dictator of her reputational
interests.

12.1.2.b Transaction Utility Richard Thaler’s idea of transaction utility stems

at least partly from a sense of fairness as well. Thaler gives this example:
You are lying on the beach on a hot day. For the last hour you have been think-
ing about much you would enjoy a nice cold bottle of your favorite brand of
beer. A companion gets up to go make a phone call and offers to bring back a
beer from the only nearby place where beer is sold [a fancy resort hotel/a
small run-down grocery store]. He says that the beer might be expensive and
so asks how much you are willing to pay for it. He says he will buy the beer if
it costs as much or less than the price you state. What price do you tell him?
When the beer would be purchased from the resort, participants were, on
average, willing to pay $2.65. When it would be purchased from the grocery
store, they were only willing to pay $1.50. Thaler suggests that while the con-
sumption utility is the same in both cases, they have different “transaction” util-
ities. Just as we get special pleasure from a “good deal”—sometimes even if we
really didn’t want the item all that much and end up not using it—we get nega-
tive transaction utility from a bad one. We expect the resort to charge a premium,
but feel we’re being taken if the grocery store charges too much.
12.1.2.c The Effect of Expectations on Experienced Utility One’s expecta-
tions about an experience can affect the experience itself, and expectations can

21. Terrence C. Burnham, Engineering Altruism: A Theoretical and Experimental

Investigation of Anonymity and Gift Giving, 50 Journal of Economic Behavior and
Organization 133–144 (2003); Gary Charness and Uri Gneezy, What’s in a Name?
Reducing the Social Distance in Dictator and Ultimatum Games, 68 Journal of Economic
Behavior and Organization 29–35 (2008).
22. Terrence C. Burnham and Brian Hare, Engineering Human Cooperation: Does
Involuntary Neural Activation Increase Public Goods Contributions?, 18 Human Nature
88–108 (2007).
23. Martin A. Nowak, Karen M. Page, and Karl Sigmund, Fairness versus Reason in the
Ultimatum Game, 289 Science 1774 (2000).
352 problem solving, decision making, and professional judgment

be affected by framing the same item in different terms.24 People given a yogurt
described as having “only 5% fat” found it richer tasting than the same food
described as “95% fat-free.” They thought that 7UP in a yellow bottle tasted more
like lemon than the same drink in a green bottle.
Deborah Frisch asserts that “framing has an effect on decisions because it has
an effect on experience.”25 But does framing actually change people’s experiences
or just provide independent information about the utility of a product or experi-
ence? That is, do people think 7UP in a yellow bottle tastes more like lemon
because the color of the bottle changes the way the drink tastes to them, or does
the bottle’s color provide them with independent information about how lemony
the drink is? Leonard Lee, Shane Frederick, and Dan Ariely probed this question
with a series of clever experiments involving beer.26 They approached patrons of
a pub on MIT’s campus and offered them two small free samples of beer. One
was ordinary Budweiser or Samuel Adams and the other was called “MIT Brew,”
which was prepared by adding a few drops of balsamic vinegar to ordinary beer.
Participants all drank the two small samples of beer and then were offered a
full glass of the beer of their choice. In the blind condition, participants tasted the
two samples and choose their free beer without knowing MIT Brew’s secret
ingredient. In the before condition, participants were told the ingredients in both
samples before tasting them, and in the after condition, they were told the ingre-
dients after they had tasted the samples but before they chose their free beer.
Although beer mixed with vinegar may sound unappealing, the majority of
participants in the blind condition preferred the MIT Brew to the unadulterated
beer. The majority of participants in the before condition, turned off by the pros-
pect of vinegar-laced beer, preferred the unadulterated beer to the MIT Brew.
What did people choose in the after condition? If our expectations (about the
taste of vinegar) provide independent information that influences our ratings of
utility, then it should not matter whether this information comes before or after
tasting the samples. The information should simply be combined with the taste
information to reach a conclusion about preference. If, on the other hand, our
expectations affect the experience of tasting the beer, then they should have a
stronger effect when independent information is presented before tasting than
after, when it is too late to affect the tasting experience. As it turned out, when
participants were told about the vinegar in the MIT Brew after tasting the sam-
ples, they still liked it as much as those in the blind condition and significantly
more than those in the before condition. Thus participants’ expectations actually
influenced their experience of the beer.

24. The examples are taken from Leaky Rationality, supra.

25. Frisch (2003) quoted in Leaky Rationality, supra.
26. Leonard Lee, Shane Frederick, and Dan Ariely, Try It, You’ll Like It: The Influence of
Expectations, Consumption, and Revelation on Preferences for Beer, 17 Psychological
Science 1054–58 (2006).
 choices, consequences, and trade-offs 353

Lee and colleagues speculate that expectations may have affected people’s
experiences by biasing them to focus on the negative aspects of the complex expe-
rience of tasting beer. People may then have erroneously attributed those nega-
tive aspects of the experience to the presence of vinegar. (This is not unlike the
process involved in confirmation bias, discussed in Section 10.3.) When people
expected the beer to taste bad, they may have searched for evidence in the tasting
experience to confirm that expectation, ignoring evidence to the contrary.27
Another set of experiments, by Baba Shiv, Ziv Carmon, and Dan Ariely, indi-
cates that the price we pay for a product can affect the utility we get from it.28 In
a series of three studies, participants were given a number of difficult anagrams
to solve (for example, participants were given the letters TUPPIL and asked to
rearrange them to form a word—in this case PULPIT). Before doing the ana-
gram, participants were given a bottle of SoBe Adrenaline Rush (an iced tea drink
that claims on its package to improve mental functioning). Some of them were
asked to pay the full retail price for their bottle of iced tea ($1.89) while others
were given a heavily discounted price (89 ). The price participants paid affected
their subsequent performance on the anagram task. Those who paid less for the
iced tea consistently performed worse than those who paid full price. They had
lower expectations about how much of a boost they would get than those who
paid full price, and this led them to perform less well on the anagram task.
In another study, Shiv and colleagues also found that inflating expectations
about the effectiveness of the drink led to a significant increase in performance
on the anagram task. Some participants were told that numerous studies had
established that SoBe produced a “significant” boost in mental acuity, while
others were told that it only produces a “slight” boost. People experienced what
they were led to expect: those who were led to expect a “significant” boost experi-
enced just that, and those who were led to expect a “slight” boost had a slightly
deflated performance. Price, meanwhile, continued to play its role: those who
believed that SoBe produced a “significant” boost and paid full price did best on
the anagram task. Those who paid a discounted price did and believed that SoBe
only produced a “slight” boost to begin with did worse.
In a related experiment, participants were exposed to a series of painful elec-
tric shocks after being given a placebo pill that they were told was a new type of
painkiller.29 Some were told the painkiller’s retail price would be $2.50 per dose
while others were told it had been discounted from $2.50 to 10 cents. As you

27. See also Stephen J. Hoch and Young-Won Ha, Consumer Learning: Advertising and
the Ambiguity of Product Experience, 13 Journal of Consumer Research 221–33 (1986).
28. Baba Shiv, Ziv Carmon, and Dan Ariely, Placebo Effects of Marketing Actions:
Consumers May Get What They Pay For, 42 Journal of Marketing Research 383–93
(2005).
29. Rebecca Waber, Baba Shiv, Ziv Carmon, and Dan Ariely, “Paying More for Less
Pain,” working paper, MIT (2007).
354 problem solving, decision making, and professional judgment

might expect by now, participants who were told the pill was expensive reported
that they experienced significantly less pain than those who were told it was
inexpensive. This effect may depend on people not adverting to the inference
they are making based on price. In one of the iced tea studies, when participants
were asked how effective they thought the SoBe would be at boosting their per-
formance given the price they had paid for it, the effect of price on performance
was completely erased. (Recall the discussion of metacognition, in Section 9.6.2.)
Later in this chapter and in the next, we discuss other situations in which the
decision process may affect satisfaction with the outcome. Perhaps the most
pervasive of these involve the phenomenon of regret, closely followed by the
effects of choice overload—of being faced with a decision that involves too many
choices.
Disconnecting the decision process from the experience falls into what Daniel
Gilbert and Jane Ebert (2002) have called the “illusion of intrinsic satisfaction”—
people’s belief that “hedonic experiences [are] due entirely to the enduring intrin-
sic properties of their outcomes, as though the wonderfulness or awfulness they
are experiencing was always there ‘in the outcome’ waiting to be experienced,
and none of these properties were altered or induced by the mere act of making
the outcome their own.”30 Daniel Kahneman similarly writes that “a theory of
choice that completely ignores feelings such as the pain of losses and the regret
of mistakes is not just descriptively unrealistic. It also leads to prescriptions that
do not maximize the utility of outcomes as they are actually experienced.”31

12.1.3 The Effects of Looking Forward and Backward

12.1.3.a Utility from Current Experience, Memory, and Anticipation When
Christine Lamm eats at a restaurant, visits her parents, runs a marathon, or
visits the dentist, she experiences immediate pleasure or pain. She may also
experience pleasure or pain in the anticipation or recollection of these events.
Jon Elster and George Loewenstein make a number of interesting observa-
tions about utility from memory and anticipation.32 The authors define primary
emotions as those that arise in one’s immediate encounters with the external
world—the pleasure of riding a bicycle or the pain of falling off it. One can
have nonprimary emotions through the backward effect of contemplating past

30. Daniel T. Gilbert and Jane E. J. Ebert, Decisions and Revisions: The Effective
Forecasting of Changeable Outcomes, 82.4 Journal of Personality and Social Psychology
503–14 (2002).
31. Daniel Kahneman, A Perspective on Judgement and Choice, 58.9 American
Psychologist 697–720 (2003).
32. See Jon Elster and George Loewenstein, Utility from Memory and Anticipation, in
Choice Over Time (George Loewenstein and Jon Elster eds., New York: Russell Sage
Foundation 1992).
 choices, consequences, and trade-offs 355

experience and the forward effect of contemplating the future. Whether through
memory or anticipation, the nonprimary emotions can result from:
• the consumption effect, where present positive or negative emotions are
based on reliving positive or negative experiences from the past or antici-
pating them in the future (e.g., remembering a lovely concert or dreading
a trip to the dentist), or
• the contrast effect, where one’s present emotions are affected by the contrast
of one’s present primary experience with one’s recalled or anticipated expe-
rience in the past or future (e.g., remembering what it was like to have a
well-paying job when currently unemployed).

“While each of the Wosileskis had experienced their rare moments of happiness,
John thought that they might have been better off had they not; that the moments of
happiness served only to accentuate despair. Had John . . . never known those excru-
ciatingly beautiful seconds with Valentine Kessler, had Mrs. Wosileksi, twenty-four
years ago, not walked down the aisle as a grateful bride and hopeful mother-to-be,
had Mr. Wosileski not, those same twenty-four years ago minus a few months,
known a surge of pride and love at the sight of his newborn, had he not twice—-
twice!—bowled a perfect game, then the Wosileskis might not have understood that
they were unhappy now. With nothing to compare it to, a dreary life would’ve been a
flatline, which is something like a contented one. But they did know rapture and so
the loss of it, the fleetingness of it, rendered the sadness that much more acute.”
—BINNIE KIRSHENBAUM, AN ALMOST PERFECT MOMENT 180 (2004).

Under the consumption effect, the secondary emotion—empathy, recollec-

tion, or anticipation—has the same hedonic sign (pleasant or painful) as the
primary experience; also, the pleasure or pain of looking forward or backwards
tends not to depend critically on one’s present level of well being. Under the
contrast effect, the relationship is inverse: “The more perfect was the past, the
more deficient appears the present; the grander our neighbor’s house, the more
inadequate seems our own.”
Elster and Loewenstein discuss these and related phenomena in a highly
nuanced way. For example, they note that we may experience present (consump-
tion) pleasure from a miserable past experience (e.g., the successful culmination
of an arduous mountain ascent); or present (consumption) pain in recalling a
pleasurable past experience (e.g., a lovely evening with the lover who jilted you).
They write that “the consumption effect swamps the contrast effect when the
future is inferior to the present,” and conclude that this creates “a strong incen-
tive to build improvement over time into one’s plans.” Quoting the narrator’s
desire to defer a goodnight kiss in Marcel Proust’s Swann’s Way—“So much did
356 problem solving, decision making, and professional judgment

I love that goodnight kiss that I reached the stage of hoping it would come as late
as possible, so as to prolong the time of respite during which Mamma would not
have appeared.” As Elster and Loewenstein say: “On the one hand, the narrator
wants to be able to look forward to the goodnight kiss (a forward consumption
effect). On the other hand, he doesn’t want to be in a position in which the kiss
has already occurred (a backward contrast effect).”
Proust’s desire to postpone his mother’s kiss, and thus savor its prospect,
deviates from people’s general tendency to be impatient (captured in the typical
model of discounting discussed in Section 13.6)—the preference to enjoy a good
experience sooner rather than later. On the negative side, people sometimes
prefer to hasten experiencing a bad event rather than dread its prospect over a
period of time. In an experiment by Gregory Berns and his colleagues,33 partici-
pants chose to receive an electric shock sooner rather than later, and some pre-
ferred to receive a more severe shock now rather than delay a smaller one. As
with the instances of deferring enjoyment, the authors suggest that “waiting
enters the utility function separately from the outcome.” Hastening even an
unpleasant outcome affords relief from dread.34
12.1.3.b The Uncertain Relationship Between Momentary and Recollected
Experience (1) We experience events at the moment they occur, (2) we remember
events in retrospect, and (3) we predict how we are likely to feel about similar
events in the future. There can be significant differences between momentary
and recalled experience. For example, a study of students’ spring-break vacations
indicated that their recollections were more positive than their experience at the
time. The students’ prediction of future experiences was based largely on the
remembered experience.35 As Dave Barry notes, “The human race is far too
stupid to be deterred from tourism by a mere several million years of bad
experiences.”36
Daniel Kahneman and Jason Riis write:37
When we are asked “how good was the vacation,” it is not an experiencing self
that answers, but a remembering and evaluating self, the self that keeps score

33. Gregory S. Berns et al., Neurobiological Substrates of Dread, 312 Science 754–58 (2006).
34. Id. 754, 757.
35. Derrick Wirtz, Justin Kruger, Christie Napa Scollon and Ed Diener, What to Do on
Spring Break? The Role of Predicted, On-line, and Remembered Experience in Future Choice,
14 Psychological Science 520 (2003). Interestingly the students recollected more nega-
tive as well as positive moments than they experienced at the time, and overestimated the
intensity of the recalled experience. The authors speculate that one tends not to remember
the (many) moments of relatively neutral affect.
36. Dave Barry’s Only Travel Guide You’ll Ever Need (New York: Ballantine 1991).
37. Daniel Kahneman and Jason Riis, Living and Thinking About It: Two Perspectives on
Life, in The Science of Well Being (Felicia A. Huppert, Nick Baylis, and Barry Keverne
eds., Oxford: Oxford University Press 2005).
 choices, consequences, and trade-offs 357

and maintains records. Unlike the experiencing self, the remembering self is
relatively stable and permanent. It is a basic fact of the human condition
that memories are what we get to keep from our experience, and the only
perspective that we can adopt as we think about our lives is therefore that of
the remembering self. For an example of the biases that result from the
dominance of the remembering self, consider a music lover who listens raptly
to a long symphony on a disk that is scratched near the end, producing a
shocking sound. Such incidents are often described by the statement that the
bad ending “ruined the whole experience.” But in fact the [momentary] expe-
rience was not ruined, only the memory of it.
The recollection of painful experiences appears to follow a quite robust rule,
called the peak-end phenomenon. Figure 12.1 shows the pain of two patients, as
reported every ten minutes, while they were undergoing colonoscopies (before
the use of the most recent sedatives). Patient A’s procedure lasted less than ten
minutes, and included one period when he reported a pain level of 7, and one
with an ending pain level of 8 on a 10-point scale. Patient B’s procedure lasted
twenty-five minutes, with one period at a pain level of 8 and several with pain
levels ranging from 4 to 6,and with the last few minutes being relatively painless.
Taking the amount of the pain and its duration into account, it seems apparent
that Patient B had a more painful experience than Patient A. Yet, when asked
about the experience retrospectively, Patient B reports having had a less painful
procedure.
In this and similar experiments—e.g., immersing one’s hand in freezing
water—people’s retrospective evaluation follows the peak-end rule: they report
the painfulness of the event as the arithmetic mean of the most painful moment
and the last moments of the experience.38
It is not surprising that retrospective assessments of past affective experi-
ences differ from affect at the moment of the experience. Like other memories,
retrospective assessments are not simply “remembered”; they are constructed.
What is so striking about the peak-end phenomenon is its apparent duration
neglect, which violates the assumption of temporal monotonicity—that adding
moments of negative affect should make the experience worse.
The most obvious explanation for the peak-end phenomenon is that the peak
and end of an experience may be the two events most available to one’s memory.
Barbara Fredrickson also suggests that, like the representative heuristic, it is an
example of judgment by prototype, where a “global evaluation of a set is domi-
nated by the evaluation of its prototypical element or exemplar.” Analogous to
ignoring base rates in identifying Jack’s occupation (Section 8.5.4), people

38. See generally Barbara L. Fredrickson, Extracting Meaning from Past Affective
Experience: The Importance of Peaks, Ends, and Specific Emotions, 14 Cognition and
Emotion 577 (2000). The phenomenon manifests itself in rats as well as humans.
358 problem solving, decision making, and professional judgment

Patient A

6
Pain intensity

0
0 10 20
Time (minutes)

Patient B

6
Pain intensity

0
0 10 20
Time (minutes)

figure 12.1 the peak-end phenomenon. from daniel kahneman and

shane frederick , Representativeness Revisited: Attribute Substitution in
Intuitive Judgment , in gilovich , griffin , kahmenan, heuristics and
biases: the psychology of intuitive judgment 49 (cambridge university
press, new york, usa (2002).

ignore the duration of the event and focus on the prototypical moments of peak
and end.
Fredrickson proposes the alternative (but not inconsistent) explanation that
peaks and ends are particularly meaningful aspects of experience. Peak discomfort
 choices, consequences, and trade-offs 359

informs a person of the “personal capacity necessary for achieving, enduring, or

coping with that episode. . . . [J]ust as you need to know the maximum height of
the sailing boat you are towing before you drive under a low bridge, peak affect
is worth knowing to decide whether you can handle experiencing a particular
affective episode again.” And endings “signal that episodes are completed, safely
in the past, and therefore knowable. . . . [They] also carry residual meaning about
personal capacity. After all, by definition, if you have encountered the end of an
episode, you have survived it—you lived to tell the story.”39
Kahneman and his colleagues suggest that the momentary experience is real,
while the remembered experience is an illusion. As the earlier excerpt from
Kahneman and Riis concludes: “The experience of the symphony was almost
entirely good, and the bad end did not undo the pleasure of the preceding half
hour. The confusion of experience with memory that makes us believe a past
experience can be ruined is a compelling cognitive illusion. The remembering
self is sometimes simply wrong.”
Putting to one side these epistemological issues,40 the divergence between
momentary and remembered experience has interesting practical consequences.
Suppose that you are considering whether to repeat the event—a similar vaca-
tion, a colonoscopy, putting your hands in freezing water. Your prediction is
likely to be based on remembered experience (though in the case of the vacation
you might have a diary, or letters, or a friend who reminds you how miserable
you thought it was at the time). And this means that you may have a miserable
time again.41 Later on, however, you may have a positive recollection of the
second miserable time, and perhaps those good memories will prove to be longer
and stronger than the experience itself. We will return below to the so-called
psychological immune system, which tends to mitigate regret. Meanwhile, con-
sider what, if any, implications the peak-end phenomenon has for reducing
recidivism after a prisoner is released after spending time in prison,
12.1.3.c The Dubious Relationship Between Daily Experiences and One’s
General Sense of Well-Being Other work by Daniel Kahneman and colleagues
examines an analogue to momentary and retroactive assessment of pain—the
relationship between people’s (positive and negative) affective experiences
during the day and their overall, or global, assessment of their well-being. The
relationship often is quite weak. For example, parents whose daily interactions
with their children are fraught with negativity are, on the whole, quite satisfied
with their children. Kahneman et al. comment: “The contrasting results reflect

39. Frederickson, supra.

40. See Mark Kelman, Hedonic Psychology and the Ambiguities of “Welfare,” 13
Philosophy & Public Affairs 413 (2005).
41. Of course, this may ultimately conduce to the health of patients who undergo
colonoscopies or similarly uncomfortable medical procedures.
360 problem solving, decision making, and professional judgment

the difference between belief-based generic judgments (‘I enjoy my kids’) and
specific episodic reports (‘But they were a pain last night’).”42

12.1.4 How Much Better Is More?

12.1.4.a Absolute vs. Relative Wealth and Happiness The so-called Easterlin
paradox observes that while rich people tend to be happier than poor people
within a society, rich societies as a whole are not much happier than poor societ-
ies and, therefore, economic growth does not increase happiness. This has been
variously explained in terms of a “hedonic treadmill,” where people adapt to a
particular level of well-being, and to assessing their own well-being by compari-
son to others’. In Economic Growth and Subjective Well-Being: Reassessing the
Easterlin Paradox, Betsey Stevenson and Justin Wolfers examine subsequent
comparisons across countries and argue that the factual assumptions of the
Easterlin paradox do not hold and that economic growth is in fact associated with
rising happiness.43
12.1.4.b Declining Marginal Utility
Money brings some happiness. But after a certain point, it just brings more
money.
—Neil Simon
Although one can think of cases where too much money leads an individual
to make decisions not in his best interests, all things being equal, more is better.
But how much better?
The eighteenth-century scientist Daniel Bernoulli observed that for most
people, the marginal value of money declines as the amount increases. For
example, the subjective value of increasing one’s wealth from $100 to $200 is
greater than the subjective value of increasing one’s wealth from $100,100 to
$100,200. This is a psychological observation, rather than a logical one. But it does
appear to hold broadly across a wide variety of people in a wide variety of situa-
tions. A graph of this “utility function” would be concave downward, as shown
in Figure 12.2.
Before she took her current job, Christine Lamm was considering two gov-
ernment positions, which offered salaries of $90,000 and $100,000, respectively;
and two jobs in the private sector, respectively offering $190,000 and $200,000.
The idea of declining marginal utility suggests that the $10,000 difference

42. Daniel Kahneman, Alan B. Krueger, David Schkade, Norbert Schwarz, and Arthur
A. Stone, A Survey Method for Characterizing Daily Life Experience: The Day Reconstruction
Method (DRM), 306 Science 1776–80 (2004).
43. Betsey Stevenson and Justin Wolfers, “Economic Growth and Subjective Wellbeing:
Reassessing the Easterlin Paradox,” Institute For the Study of Labor Discussion Paper No.
3654 1–80 (August 2008).
 choices, consequences, and trade-offs 361

Utility

Wealth

figure 12.2 bernoulli’s curve for wealth and utility.

between the two government positions may be more important to her than the
$10,000 difference between the two higher paying positions.
We’ll return to these concepts below, but definitionally, the expected value
(EV) of Christine’s choice is her expected net salary. You could compute it just
using a calculator and without knowing anything about her individual prefer-
ences. Her subjective satisfaction is the expected utility (EU) of the choice. The
declining marginal utility of money (and some other goods as well) is an impor-
tant category in which EU deviates from EV.
12.1.4.c Psychophysical Numbing
One death is a tragedy; a million is a statistic.
—Joseph Stalin
I am deeply moved if I see one man suffering and would risk my life for him. Then
I talk impersonally about the possible pulverization of our big cities, with a hundred
million dead. I am unable to multiply one man’s suffering by a hundred million.
—Albert Szent-Gyorgi
Participants in a study were given a scenario in which a foundation was pre-
pared to make a $10 million grant to a medical institution to implement a new
treatment to reduce deaths from disease. They were told about three institutions
concerned with three different diseases with respective annual fatalities of
15,000, 160,000, and 290,000, and asked the minimum number of lives the insti-
tution must save to merit the grant. Two-thirds of the respondents required that
more lives be saved as the size of the at-risk population increased, requiring that
the institution dealing with the largest disease group save eleven times more
people than the institution addressing the smallest disease group.44

44. Quoted in David Fetherstonhaugh, Paul Slovic, Stephen Johnson, and James
Friedrich, Insensitivity to the Value of Human Life: A Study in Psychophysical Numbing, 14
Journal of Risk and Uncertainty 28 (1997).
362 problem solving, decision making, and professional judgment

David Featherstonehaugh and his colleagues relate this phenomenon to

Weber’s law,45 a psychophysical phenomenon that holds across a variety of stim-
uli. Suppose that you are hearing a soft noise or seeing a dim light, and the
sound or brightness diminishes just enough for you to notice the change. Now
suppose that you are hearing a loud noise or seeing a bright light, and the sound
or brightness diminishes by the same absolute amount as before. You would not
notice the change. To be noticed, the change must bear some proportional rela-
tionship to the intensity of the background stimulus. The difference is described
by a logarithmic rather than a linear function. Featherstonehaugh et al. write:
Just as a fixed decrease in brightness seems greater when the original inten-
sity is small than when it is large, an intervention saving a fixed number of
lives seems more valuable when fewer lives are at risk to begin with—when
the lives saved constitute a larger proportion of the number at risk. When
such psychophysical numbing occurs, the value of a life-saving intervention
is inversely proportional to the magnitude of the threat rather than being
determined by the absolute number of lives the intervention can save.46

45. Articulated by the nineteenth-century psychologists Ernst H. Weber and Gustav

Fechner.
46. Although proportion dominance is strong, a substantial minority of respondents
base their life-saving decisions on numbers alone. James Friedrich and his colleagues sug-
gest that there may be individual differences between members of the two groups. Those
in the former category tended to value saving lives more highly (i.e., requiring fewer lives
saved to justify intervention) than those in the latter, and to eschew cost-benefit analysis.
Friedrich and his colleague speculate that the phenomenon may result from applying
percentage discount heuristics used in making consumer choices. James Friedrich, Paul
Barnes, Kathryn Chapin, Ian Dawson, Valerie Garst, and David Kerr, Psychophysical
Numbing: When Lives are Valued Less as the Lives at Risk Increase, 8 J. Consumer Psych.
277 (1999).
The presentation of the decision can affect how one approaches the decision. In
a variation on the medical institution problem, respondents were asked to rank order
three institutions based on whether they should receive the $10 million grant. They were
told that:
• Institution X would reduce annual deaths from disease A from 15,000 to 5,000.
• Institution Y would reduce annual deaths from disease B from 160,000 to 145,000.
• Institution Z would reduce annual deaths from disease C from 290,000 to 270,000.
Notice that there is an inverse relationship between the number of lives saved and the
proportion of lives saved. Interestingly, 60 percent of the respondents preferred Institution
Z, thus maximizing the number of lives saved, with only 16 percent choosing Institution X.
Featherstonehaugh et al. explain that the information highlighted the magnitude of each
intervention’s life-saving potential and that the ease of comparability—evaluability—may
also have been a factor. Nonetheless, they conclude that “psychophysical numbing is a
robust phenomenon—ingrained in the workings of our cognitive and perceptual
systems.”
 choices, consequences, and trade-offs 363

12.1.4.d When More Is Less: Idiosyncratic Matching The phenomenon of

declining marginal utility suggests that more of a good thing is better—just not
as much better as the last increment. But there are some situations in which
more of a good thing is actually worse: You begin a meal with each bite having a
high utility, but the utility of bites decreases as the meal progresses and eventu-
ally declines. If you continue eating after you’re stuffed, the utility curve reverses
direction.
Are there situations where you would pay not to have more of something that
you are indifferent to? For example, suppose that you are a fan of first-run movies
and receive an offer of a new channel, “Best Newly Released Movies,” from your
cable TV company. How much would you pay for the subscription? Now sup-
pose that this movie channel comes a part of a package of three channels, includ-
ing two in which you have absolutely no interest—say, one has only infomercials
and the other is in Urdu. How much would you pay for the package?
Many people would pay less for the three-channel package that included the
movie channel than for the movie channel alone; indeed, they would pay to have
the unwanted channels removed from the package—not because they are
repulsed by the additional channels, or because of the burden of flipping through
them, but because the three channels do not match their tastes as closely as the
single one.47 Perhaps this apparent violation of dominance (see Section 12.2.2.a)
in particular choices reflects the application of a heuristic that is for the most
part useful—for example, that a package deal signals the inferiority of the com-
ponents of the package. Buying something that closely fits your tastes feels like
a relatively “good deal.” Perhaps it is also an application of the representative-
ness heuristic.
Varda Liberman and Lee Ross, who conducted the cable TV studies, did a
variation that inquired whether the value that viewers place on a channel they
desired would be affected by others’ tastes for the same channel. It turned out that
participants were willing to pay more for the channel when it matched their idio-
syncratic tastes than when many others shared their tastes. On the one hand,
people often seek validation of their own tastes by looking to others. On the other
hand, in addition to the sense of a bargain for a unique match, something that fits
your own idiosyncrasies may also affirm your distinct identity or special status.

47. See Varda Liberman and Lee Ross, Idiosyncratic Matching and Choice: When Less Is
More, 101 Organizational Behavior & Human Decision Processes (2006); Itamar
Simonson, Ziv Carmon, and Suzanne O’Curry, Experimental Evidence on the Negative
Effect of Product Features and Sales Promotion on Brand Choices, 13 Marketing Science 23
(1994); Ran Kivetz and Itamar Simonson, Earning the Right to Indulge: Effort as a
Determinant of Customer Preferences Toward Frequency Reward Programs, 39 Journal of
Marketing Research 155 (2002); Ran Kivetz and Itamar Simonson, The Role of Effort
Advantage in Consumer Response to Loyalty Programs, 40 Journal of Marketing Research
454 (2003).
364 problem solving, decision making, and professional judgment

12.1.5 Hedonic Adaptation

[With] never-failing certainty . . . all men, sooner or later, accommodate themselves
to whatever becomes their permanent situation . . . [I]n every permanent situation,
where there is no expectation of change, the mind of every man, in a longer or
shorter time, returns to its natural and usual state of tranquility. In prosperity, after
a certain time, it falls back to that state; in adversity, after a certain time, it rises up
to it.”48
—Adam Smith
Imagine that one morning your telephone rings and you find yourself speaking
with the King of Sweden, who informs you in surprisingly good English that you
have been selected as this year’s recipient of a Nobel prize. How would you feel, and
how long would you feel that way? . . . Now imagine that the telephone call is from
your college president, who regrets to inform you that the Board of Regents has dis-
solved your department, revoked your appointment, and stored your books in little
cardboard boxes in the hallway. How would you feel, and how long would you feel
that way?49
—Daniel Gilbert
Humans adapt to both good and bad events.50 Such “hedonic adaptation” may
occur through a number of mechanisms, including ones that parallel our adap-
tation to changes in sound, light, and other physical phenomena.51 In Chapter 1
we discussed people’s strong need to have a coherent view of the world—to see
patterns, causes, and explanations. Daniel Gilbert and Timothy Wilson describe
four aspects of this need that may play a special role in making our emotional
reactions to events more evanescent than we generally imagine:52
1. People orient to unexpected but relevant information in their environment.
2. People have more intense emotional reactions to unexpected information
than to other events.

48. Adam Smith, Theory of Moral Sentiments (1759).

49. Daniel T. Gilbert, Elizabeth C. Pinel, Timothy D. Wilson, Stephen J. Blumberg,
and Thalia P. Wheatley, Immune Neglect: A Source of Durability Bias in Affective Forecasting,
75 Journal of Personality & Social Psychology 617 (1998).
50. See Shane Frederick and George Loewenstein, Hedonic Adaptation, in Daniel
Kahneman, Ed Diener, and Norbert Schwarz, Well-Being: The Foundations of
Hedonic Psychology 302 (1999).
51. Although the phenomena are contested, some commentators have suggested that
people whose life circumstances improve may be on a “hedonic treadmill,” where they
became accustomed to the improvements, or on an “aspiration treadmill,” with increased
expectations or standards for what it takes to be satisfied.
52. Daniel T. Gilbert and Timothy D. Wilson, Affective Forecasting, in 35 Advances in
Experimental Social Psychology 345–411 (Mark P. Zanna ed., 2003).
 choices, consequences, and trade-offs 365

3. Once an unexpected event occurs and people have a relatively intense emo-
tional reaction, they tend to make sense of the event, quickly and auto-
matically.
4. When people make sense of an event, it no longer seems surprising or
unexpected, and as a result they think about it less and it produces a less
intense emotional reaction. The process of sensemaking “ordinizes” events
in a way that robs them of their emotional power.
The easier is it for people to make sense of an experience, the more quickly
they return to the baseline. With respect to negative events in particular, psy-
chologists have posited a psychological immune system that protects individuals
from an overdose of gloom53— analogous to the well-documented processes of
physical homeostasis that cause most life forms and complex systems to main-
tain their internal equilibrium. In addition to other adaptive mechanisms, ratio-
nalization can often play an important role. The easier it is to rationalize a bad
experience (breaking up with a romantic partner, not getting a job), the easier it
is to recover.
We tend to mitigate regret over an outcome by reconstruing our responsibil-
ity for its cause. As between blaming oneself for missing a plane by a few min-
utes (“if only I had left the house earlier”) and tending to cast the blame on
others (“if only they hadn’t shut the doors sooner than the departure time”),
most people opt for the latter.54 We also reconstrue our intentions and desires,
for example, responding to a job rejection by thinking “that wouldn’t have been
a good firm to work for anyway.” Indeed, the anticipation of regret may lead to
anticipatory reconstrual: “I don’t know how well my interview went from their
point of view, but I’ve got to say that I wasn’t particularly impressed with
them.”55

53. Gilbert et al., supra.

54. Daniel T. Gilbert, Carey K. Morewedge, Jane L. Risen, and Timothy D. Wilson,
Looking Forward to Looking Backward: The Misprediction of Regret, 15 Psychological
Science 346 (2004).
55. Anticipatory reconstrual can backfire, however: The applicant who downgrades the
employer may appear lackluster in an interview, thus creating a self-fulfilling prophecy;
or, if she gets the job she may not be as satisfied as she would have been if she had
not downgraded it in the first place. Timothy D. Wilson, Thalia P. Wheatley, Jamie L.
Kurtz, Elizabeth W. Dunn, and Daniel T. Gilbert, When to Fire: Anticipatory versus Postevent
Reconstrual of Uncontrollable Events, 30 Personality and Social Psychology Bulletin
340 (2004).
366 problem solving, decision making, and professional judgment

Cognitive Dissonance Theory

Cognitive dissonance57 theory proposes that we reduce inconsistencies in our

beliefs (cognitions) and behavior through one of three strategies: (1) reduce the
importance of dissonant cognitions, (2) increase the importance of consonant
cognitions so as to outweigh the dissonant ones, or (3) change cognitions or
behavior. To cite an example used by Leon Festinger, who developed the theory,
a smoker who learns that smoking is harmful to his health can (1) discount the
risks of smoking (“It adds little risk to inhaling air in the city every day”); (2) think
about the benefits (“It keeps my weight under control”); or (3) stop smoking.
Although the theory has undergone revisions and been challenged by alternatives
since it was introduced in 1957, the essential ideas remain quite robust.

Cultural differences.56 Most of the participants in research experiments on

judgment and decision making have been middle-class members of Western
cultures (broadly speaking). Research by Hazel Markus and her colleagues indi-
cates that there may be systematic cultural differences with respect to well-being
and the values of autonomous choice.57 It remains to be learned to what extent
the phenomena described in this and the following two chapters are applicable
in quite different cultural settings.

12.2 criteria for rational decision making: the focus on

consequences and trade-offs

People are in principle capable of pursuing their ends—whatever they may be—
in a rational manner. We will tentatively define a rational decision-making pro-
cess as one that tends to achieve the decision maker’s objectives—that is, the
quality of the process is ultimately measured in terms of its consequences.58 A
good process is one that works (most of the time). This seemingly tautological
premise is more complicated that it appears.
First, even the best possible decision-making processes do not always lead
to the desired consequences. For example, a decision maker may lack the

56. See generally Eddie Harmon-Jones and Judson Mills, eds., Cognitive
Dissonance: Progress on a Pivotal Theory in Social Psychology (American
Psychological Association (APA), 1999).
57. The literature is summarized on Hazel Markus’s home page, http://www-psych.
stanford.edu/∼hmarkus/
58. Deborah Frisch and Robert T. Clemen, Beyond Expected Utility: Rethinking
Behavioral Decision Research, 116 Psychological Bulletin 46 (1994).
 choices, consequences, and trade-offs 367

information necessary to accurately predict the consequences, or the decision

may be based on an accurate assessment of the probability of events conducive
to a desired consequence but—probabilities being what they are—the events
may or may not occur.
Second, as mentioned above, the decision maker may be concerned about the
process of decision making as well as the consequences. The most obvious and
often acknowledged concern about process is with the costs of making the deci-
sion, which must be weighed against the benefits of an optimal outcome. In
addition to the tangible costs of time and money, some decision processes carry
emotional costs, such as the stress of making difficult trade-offs. They may have
emotional benefits as well: We know people who positively enjoy the process of
shopping for the best deal, no matter how long it takes. A decision maker might
also take pleasure in a procedure, such as referring to a horoscope, that has a
dubious relationship to outcomes.
Third, real-life decision makers sometimes subordinate the consequences of
a decision to other values. For example, an outcome-oriented decision is neces-
sarily based on the decision maker’s current assets,59 not assets once possessed
but already spent. Yet who has not been tempted by, if not succumbed to, the
“sunk cost fallacy”? (See Section 14.3) Although we will generally assume that
deviations from outcome-oriented strategies are errors, their pervasiveness sug-
gests that decision makers sometimes may be pursuing interests other than
their apparent or stated outcomes—or that people’s interests can be more com-
plex than they appear to be. We are not referring to situations where deontologi-
cal principles—promises, duties to family, country, God, and the like—trump
utilitarian ones, but rather to mundane decisions such as the purchase of con-
sumer products, where, for example, ignoring sunk costs may make one appear
foolishly wasteful in one’s own or others’ eyes. We return to this question in the
next chapter.
Finally, especially because much of this chapter focuses on deliberative deci-
sion making, it is worth emphasizing that there is no a priori reason to believe
that deliberative decision processes inevitably (or even usually) produce better
decisions than naturalistic or intuitive processes. According to Paul Slovic and
his colleagues: 60
Images, marked by positive and negative affective feelings, guide judgment
and decision making. Specifically, . . . people use an affect heuristic to make
judgments. That is, representations of objects and events in people’s minds

59. These include money, tangible goods, physiological and psychological capacities,
social relationships. See Reid Hastie and Robyn M. Dawes, Rational Choice in an
Uncertain World 18 (Thousand Oaks, CA: Sage Publications, 2001).
60. Paul Slovic, Melissa L. Finucane, Ellen Peters, and Donald G. MacGregor, Rational
Actors or Rational Fools? Implications of the Affect Heuristic for Behavioral Economics 31
Journal of Socio-Economics, 329–42 (2002).
368 problem solving, decision making, and professional judgment

are tagged to varying degrees with affect. In the process of making a judg-
ment or decision, people consult or refer to an “affect pool” containing all the
positive and negative tags consciously or unconsciously associated with the
representations. Just as imaginability, memorability, and similarity serve as
cues for probability judgments (e.g., the availability and representativeness
heuristics), affect may serve as a cue for many important judgments. Using
an overall, readily available affective impression can be far easier—more
efficient—than weighing the pros and cons or retrieving from memory many
relevant examples, especially when the required judgment or decision is
complex or mental resources are limited. This characterization of a mental
short-cut leads to labeling the use of affect a “heuristic.” . . .
Like other heuristics, the affect heuristic can conduce to good decision making,
but it also has the potential to distort the process. Whether intuition and affect
play a constructive role in decision making depends on the nature and context of
the decision at hand.61

12.2.1 Two Concepts of Rational Decision Making

As the preceding discussion suggests, people’s objectives are multifaceted and
conflicting, and even an excellent decision process will not always produce the
desired outcomes. There are at least two definitions of rational decision making.62
12.2.1.a Consequential Rationality If the quality of a process ultimately
depends on the results it produces, then the ideal way of assessing its rationality
is in terms of its consequences. An intuitive decision-making process that is
opaque even to the decision maker herself—for example, “I just follow my gut
instinct”—is rational in this sense if it regularly produces the desired outcomes.
Evaluating a decision process in terms of its ability to produce desired out-
comes is a difficult task, however. It requires specifying the desired outcome and
then examining a large number of virtually identical decisions to assess the
strength of the correlation between the decision and outcome. This is feasible
with respect to repetitive decisions within an organization (for example, a bank’s
criteria for making loans). But a process that is effective in one environment
(e.g., making investment decisions) may not be effective in another (deciding
which job to take).
12.2.1.b Procedural Rationality: Subjective Expected Utility Theory
Consequential rationality is concerned with the decision procedure’s

61. See Kenneth R. Hammond, Human Judgment and Social Policy: Irreducible
Uncertainty, Inevitable Error, Unavoidable Injustice 60 (New York: Oxford
University Press 1996).
62. These roughly map on to the correspondence and coherence theories of truth in
philosophy. See, e.g., “The Stanford Encyclopedia of Philosophy,” http://plato.stanford.
edu/entries/truth-coherence/.
 choices, consequences, and trade-offs 369

correspondence to desired outcomes. By contrast, subjective expected utility theory

(SEU) sets out a theory of decision making that defines a rational procedure in
terms of its adherence to a set of axioms, or norms. (The word subjective empha-
sizes its concern with the decision maker’s own utility rather than some external
measure.) SEU is not indifferent to outcomes. On the contrary, it claims that
violation of its axioms will produce inferior outcomes.
To a large extent, the test of the axioms is intuitive. But the theory makes
stronger claims to validity. For example, in a game theory scenario a player of a
multiround game could bankrupt an opponent who systematically violated at
least some of the axioms.63 In the belief that SEU provides a pretty good bench-
mark for assessing analytic decision-making processes, we now describe its
essential axioms.

12.2.2 The Axioms of Subjective Expected Utility Theory

Rather than describe the axioms all at once, in this chapter we present those that
underlie decision making where risk is not an issue. In Chapter 15, we revisit
some of these in conditions of risk and set out several new axioms as well.
12.2.2.a Completeness, Dominance, and Trade-offs The completeness axiom
holds that, faced with a choice between alternatives A and B, you must prefer A
to B, or prefer B to A, or be indifferent between A and B. This is obviously true
when the alternatives have only one attribute, e.g., the choice between a choco-
late and vanilla ice-cream cone. However, is also true when the alternatives have
multiple attributes that require making trade-offs, as in the decision of siting the
wastewater treatment plant. Suppose that firm A offers you a higher salary, but
does not offer as good support and mentorship as firm B. Just as in making a
choice involving a single characteristic, the axioms of rational decision making
posit that ultimately you must either prefer firm A to firm B, or B to A, or be
indifferent between them—in which case you’d flip a coin.
A dominates B if A is preferable to B in at least one respect and is at least as
desirable as B in all other respects. If so, you should choose A and reject B.
Referring to the siting the wastewater treatment plant: if one potential site were
inferior to the others in every respect, it would be dominated by the others and
could be eliminated from the rankings.
But dominance seldom eliminates all but a single choice, and trade-offs are
often neither obvious nor easy. Even personal decisions often require making
trade-offs among more or less incommensurable wants. Many decisions involv-
ing public policy require making trade-offs that have deep and contested ethical
dimensions, for example between money on the one hand, and health, safety, or
the environment on the other.

63. See the discussion of transitivity, Section 12.2.2.d.

370 problem solving, decision making, and professional judgment

The problem goes beyond the cognitive incommensurability of the values

and involves what Philip Tetlock and his colleagues call constitutive incommensu-
rability—based on cultural values and norms:64
The guiding idea is that our commitments to other people forbid certain com-
parisons. To transgress these normative boundaries, to attach a monetary
value to one’s friendships or one’s children or one’s loyalty to one’s country is
to disqualify oneself from certain social roles, to demonstrate that one “just
doesn’t get it”—that one does not understand what it means to be a true
friend or parent or citizen. We run into constitutive incommensurability
whenever treating values as commensurable subverts one or both of the
values in the trade-off calculus. To compare is to destroy. Even to think about
certain trade-offs (less still to make them) is to corrupt and degrade one’s
standing as a moral being in the community.
For these among other reasons, the use of cost-benefit analysis to guide public
or corporate policy making has sometimes been criticized for requiring “immoral
commodification”65 by placing a dollar value on life’s unquantifiables—such as
human life, health, or the existence of a wilderness area.66 In Retaking Rationality,
Richard Revesz offers a rebuttal to this criticism:
“This criticism confuses pricing with commodification. Pricing, a mechanism
used to allocate society’s resources, is the most effective way to aggregate infor-
mation and allocate scarce resources to produce the most benefit . . .
Commodification has to do with the social significance of pricing—the fear that
assigning a price to the good things in life obscures their inherent worth.”67
To be sure, some decisions call for existential choices—for example, trade-
offs between peace and justice—that far transcend the capacity of cost-benefit
analysis.68 But as Robert Frank notes:

64. Philip E. Tetlock, Coping with Trade-offs: Psychological Constraints and Political
Implications, in Political Reasoning and Choice Elements of Reason: Cognition, Choice,
and the Bounds of Rationality (Arthur Lupia, Matthew D. McCubbins, and Samuel L.
Popkin eds., Berkeley: University of California Press; Cambridge, UK: Cambridge University
Press, 2001); Alan P. Fiske and Philip E. Tetlock, Taboo Tradeoffs: Constitutive Prerequisites for
Political and Social Life, in Political Psychology: Cultural and Cross Cultural
Foundations (Stanley A. Renshon and John Duckitt eds., New York: NYU Press 2000).
65. Richard L. Revesz And Michael A. Livermore, Retaking Rationality: How
Cost-Benefit Analysis Can Better Protect the Environment and Our Health
78–79 (New York, Oxford University Press 2008) [hereafter, Retaking Rationality].
66. For a sustained critique, arguing that cost-benefit analysis is fundamentally flawed,
see Frank Ackerman and Lisa Heinzerling, Priceless: On knowing the Price of
Everything and the Value of Nothing (New York: The New Press, 2004); See also
Steven Kelman, Cost-Benefit Analysis: An Ethical Critique, 5 J. Gov’t & Soc. Reg. 33 (1981).
67. Retaking Rationality 13.
68. On the national and international level, the issue arises with respect to whether to
subject the perpetrators of massive crimes against citizens to criminal punishment or to
 choices, consequences, and trade-offs 371

Scarcity is a simple fact of the human condition. To have more of one good
thing, we must settle for less of another. Claiming that different values are
incommensurable simply hinders clear thinking about difficult tradeoffs.
Notwithstanding their public pronouncements about incommensurability,
even the fiercest critics of cost-benefit analysis cannot escape such tradeoffs.
For example, they do not . . . get their brakes checked every morning. The
reason, presumably, is not that . . . auto safety does not matter, but that they
have more pressing uses of their time. Like the rest of us, they are forced to
make the best accommodations they can between competing values.69
Policy makers (and others) use various strategies to avoid explicit trade-offs
between constitutively incommensurable values.70 When they cannot avoid a
conscious or public trade-off, they engage in buck-passing, procrastination,
and obfuscation.71 And when responsibility for a decision cannot be avoided,
they tend to “spread the alternatives,” “playing down the strengths of the to-be-
slighted value and playing up the strengths of the to-be selected value.”72

Jurors’ Aversion to Corporate Cost-Benefit Analysis Involving Human

Safety

Recall Kip Viscusi’s discussion, in Section 10.2.1, of jurors’ imposing punitive

damages on corporate defendants who engaged in cost-benefit analysis. In
addition to explaining the phenomenon in terms of hindsight bias, he specu-
lates that “[p]eople may be averse to explicitly balancing money against human
lives. Money and lives might be considered incommensurable. . . . When corpo-
rations systematically think about risk levels yet nevertheless pursue product or
environmental policies that fail to maximize safety, the juror may regard the
corporate decision as ‘cold-blooded.’”

the processes of “truth and reconciliation” commisions, as well as in negotiations between

warring nations. The Israeli philosopher, Avram Margalit, has argued that when peace
brings an end to cruelty and humiliation, “peace can be justified by being just a peace
without being a just peace.” (Stanford University Tanner Lectures on Human Values, May
4, 2004, Stanford Report, May 10, 2004).
69. Robert Frank, Why Is Cost-Benefit Analysis so Controversial?, 29 Journal of Legal
Studies 913, 914 (2000).
70. See Jonathan Baron, and Mark Spranca Protected Values, 70 Organizational
Behavior and Human Decision Processes 1–16 (1997).
71. Philip E. Tetlock and Richard Boettger, Accountability Amplifies the Status
Quo Effect When Change Creates Victims, 7 Journal of Behavioral Decision Making
1–23 (1994).
72. Philip E. Tetlock, Coping with Trade-offs: Psychological Constraints and Political
Implications, supra.
372 problem solving, decision making, and professional judgment

12.2.2.b Invariance The order or manner in which choices are presented

should not affect the outcome of your decision. Thus, Christine Lamm’s choices
among the potential sites for the wastewater treatment plant should not depend
on which one she considers first. (Does Paul’s strategy of buying the first suit
he sees in a store that fits necessarily violate the axiom of invariance? Is it
rational?)
12.2.2.c Independence from Irrelevant Alternatives The relative ranking of a
particular set of options should not vary with the addition or deletion of other
options. If Lamm is deciding between sites A and C, the fact that she learns
about a possible new site, E, in a completely different location from any of the
ones she has considered, should not affect her ranking of A and C (unless it
somehow provides new information relevant to them).
12.2.2.d Transitivity The Transitivity axiom entails that if you prefer A to B,
and prefer B to C, then you must prefer A to C.
The rationality of transitivity is illustrated by considering the plight of a person
who is systematically intransitive. She could be used as a “money pump” in the
following way. Suppose that Lamm has chosen a model of car to buy and is con-
sidering three colors, and suppose that (intransitively) she
• prefers blue to green,
• prefers green to white,
• prefers white to blue.
Lamm buys the blue car. You then offer to trade her the white car for the blue
car if she gives you $1; she accepts the trade, because she prefers white to blue.
Now that she has the white car, you offer to trade the green car for it if she gives
you $1, which she accepts, because she prefers the green to the white. Now, you
offer to trade her the blue car for the green if she gives you $1, which she accepts,
because she prefers the blue to the green. And we start the process again—until
she runs out of money. Later in the book, we will show circumstances under
which people—perhaps yourself included—act in an intransitive manner.

Parsimoniousness, Completeness, Decomposability

Albert Einstein suggested: “Make everything as simple as possible, but not sim-
pler.” Thus, a decision maker should consider all those but only those factors that
could significantly affect her objectives or interests. In identifying the criteria rel-
evant to siting the wastewater treatment plant, Christine Lamm has tried to be
complete. Sometimes, however, she may lump a number of criteria together. For
example, environmental impact comprises a number of factors, but it may not be
worth her time and energy to consider each one separately, rather than make an
overall assessment.

***
 choices, consequences, and trade-offs 373

This concludes our first look at the axioms of subjective expected utility theory.
We consider some axioms relevant to decision making under risk in Section
15.1. Before moving on, we note two points about SEU.
First, SEU provides a generalized framework for rational choice. But evolu-
tionary psychologists such as Leda Cosmides, John Tooby, and Gerd Gigerenzer
have argued that much individual decision making is not based on general prin-
ciples of rationality, but is modularized to address specific problems in particu-
lar domains. They use the term ecological rationality to refer to the success of a
process in producing outcomes in a particular environment or context.73 Granted
that individuals often depart from the norms of SEU, however, there is consider-
able doubt whether decision making is as highly contextual as these social scien-
tists claim, let alone to what extent modularized decision making leads to
successful outcomes.74 In any event, lawyers and policy makers are often called
upon to make decisions that cut across many domains.
Second, most applications of expected utility theory focus on outcomes. But
earlier in the chapter we considered how the very process of choosing can produce
gains or losses of utility for the decision maker. This phenomenon can confound
decision makers, or at least confound understanding whether particular choices
are rational from the decision maker’s point of view. We will return to this matter
in the next several chapters.

12.3 from expected utility to expected return: cost-benefit

analysis

Under the concept of dominance in expected utility theory, if alternatives can be

converted to a common metric—typically dollars—it is rational to choose the
alternative that is worth the most. This is the idea behind cost-benefit analysis
(CBA), under which an activity or program is justified only if its benefits out-
weigh its costs. Cost-benefit analysis builds on program evaluations of the sort
described in Chapter 7, essentially translating the importance and size of effects
into dollar values. Governments frequently require that regulatory decisions be
subject to cost-benefit analysis. For example, Executive Order 12866,75 signed by

73. See Leda Cosmides and John Tooby, Origins of Domain Specificity: The Evolution of
Functional Organization, in Mapping the Mind: Domain Specificity in Cognition
and Culture 85, 89–90 (Lawrence A. Hirschfeld and Susan A. Gelman eds., Cambridge,
UK: Cambridge University Press, 1994); Gerd Gigerenzer, Peter M. Todd and ABC
Research Group, Simple Heuristics That Make Us Smart (New York: Oxford University
Press, 1999).
74. Jerry Fodor, The Mind Doesn’t Work That Way: The Scope and Limits of
Computational Psychology 58–61 (Cambridge, MA: The MIT Press, 2000); Mark G.
Kelman, The Heuristics Debate: Its Nature and Its Implications for Law and
Policy (Oxford University Press, forthcoming 2010).
75. See Office of Management and Budget (OMB), “Economic Analysis of Federal
Regulations Under Executive Order 12866,” Jan. 11, 1996, http://www.whitehouse.gov/
374 problem solving, decision making, and professional judgment

President Bill Clinton in 1993, requires federal agencies to “assess all costs and
benefits of available regulatory alternatives, including the alternative of not
regulating.”
The most difficult and contested applications of CBA involve risk, a matter
that we discuss in Chapter 15. Even when risk is not a major factor, however,
CBA can present difficulties in monetizing benefits and in making distributional
decisions (for example, where the benefits flow partly to an individual benefi-
ciary of a program and partly to taxpayers). And even when the calculation is
reasonably clear, there may be reasons not to follow the logic of CBA: For better
or worse, politics, ideology, and issues of fairness and rights may trump CBA.
But CBA almost always provides a useful norm or reference point for decision
making.
The longitudinal analysis of the Perry Preschool program in Michigan pro-
vide a classic example of CBA. This precursor of Head Start had low student-
teacher ratios; all the teachers were certified in both early childhood and special
education, and they made weekly visits to every family.76 The program produced
long-term gains in school attendance, academic achievement, graduation rates,
and earnings.
In constant discounted dollars, the economic return to society of the program
was more than $17 per dollar invested ($258,888 on an investment of $15,166
per student). Seventy-five percent of that return accrued to the general public
($12.90 per dollar invested), and 25 percent to each participant ($4.17 per
dollar invested). While net benefits derive from many sources, such as
increased earnings and taxes paid, with Perry Preschool students having sig-
nificantly fewer lifetime arrests and fewer arrests for violent crimes than stu-
dents not in the program, a significant share of the public benefits come from
the avoidance of these significant costs to society.77
Thus, even when limiting the benefits to those returned to taxpayers, the pro-
gram had a benefit-cost ratio of almost 13 to 1. Economists at the Federal Reserve
Bank in Minneapolis estimated that the Perry Preschool program generated a 16
percent rate of return on investment, of which 12 percent was a public return.78

omb/inforeg/riaguide.html; Environmental Protection Agency, “Executive Order 12866

Regulatory Planning and Review,” Sept. 30, 1993, http://www.epa.gov/fedrgstr/eo/
eo12866.html.
76. Jonathan Crane and Mallory Barg, “Do Early Childhood Intervention Programs
Really Work,” Coalition for Evidence-Based Policy, April 2003, 10.
77. Michael A. Stegman, The Power of Measuring Social Benefits, The John D. and
Catherine T. MacArthur Foundation (2009 unpublished).
78. Art Rolnick and Rob Grunewald, Early Childhood Development: Economic Development
with a High Public Return, FedGazette, March 2003. http://www.minneapolisfed.org/
pubs/fedgaz/03-03/earlychild.cfm.
 choices, consequences, and trade-offs 375

12.3.1 An Example: A CBA Analysis of a Workforce Development Program

Not only governments, but nonprofit organizations and their funders, evaluate
social programs in terms of costs and benefits. For example, the New York–based
Robin Hood Foundation examines the benefit-cost ratio of the workforce devel-
opment programs it supports so it can compare them not only to similar pro-
grams, but also to its grants in education, health, and other areas.79
Robin Hood measures impact in terms of the personal benefits that accrue to
its grantee organizations’ poor clients and their families. For example, in a job
training program for ex-offenders, the foundation estimates the impact of job
placements on trainees’ recidivism and on their future earnings. In supporting
early childhood programs, it estimates the impact of reading readiness on high
school graduation rates and on the children’s projected earnings as adults.
Although Robin Hood does not make information about particular grantees
publicly available, it demonstrates its approach through a fictional example:
Bob’s Jobs, a workforce training program for women to become construction
workers that is based on Robin Hood’s actual programs.
Calculating benefits. One hundred fifty women enrolled in Bob’s Jobs and sev-
enty-two completed the training. Of these seventy-two newly minted construc-
tion workers, forty-one held on to their jobs for only three months. The remaining
thirty-one still had jobs at the end of one year.
How much does Bob’s Jobs benefit trainees? Robin Hood compares the salary
of each participant before she entered the program and after she graduated. The
forty-one women with short-term employment enjoyed an average salary increase
of about $2900, or $120,000 in total. The average annual salary increase for the
thirty-one women who held jobs for at least a year was approximately $12,000, or
$372,000 in total. To compute the value of the program for these thirty-one
women, Robin Hood makes the following assumptions:
1. They will continue to be employed for thirty years.
2. Their annual salaries will increase by 1.5 percent above inflation.
3. They have an average of 1.8 children each; given research findings on the
effects of parents’ employment, each family will realize an intergenera-
tional income boost of $56,000.
4. The discount rate—that is, the number used to calculate how much an
amount of future money is worth today—is 3.5 percent.80
On the basis of these assumptions, Robin Hood estimates that the net pres-
ent value of the benefits to the thirty-one long-term workers is $9.1 million.
Adding the onetime salary boost for the forty-one short-term workers, the total
benefit of the program is $9.2 million. Robin Hood does not further discount the

79. See Paul Brest, Hal Harvey, & Kelvin Low, Calculated Impact, Stanford Social
Innovation Review, Winter 2009.
80. For more on discount rates, see Section 13.6.1.
376 problem solving, decision making, and professional judgment

value of the benefits based on the probability of success; it discounts its own role
in generating these benefits by 50 percent to account for the fact that other
donors, both private and public, support Bob’s Jobs. Thus, Robin Hood attri-
butes $4.6 million of the program’s benefits to its philanthropic investment.
Calculating costs. What about the costs? The grant to Bob’s Jobs was $200,000.81
Dividing $4.6 million by $200,000 results in a net benefit of $23 per $1 invested.
That is, for each dollar that Robin Hood spends on Bob’s Jobs, trainees and their
families gain $23—a pretty good return on investment.
Robin Hood has made similar calculations for its other poverty programs,
including a program to help people manage budgets, bank accounts, and loans
(net benefit = $1.90) and a clinic that deals with asthma, hepatitis, and cancer
(net benefit = $12).
Although Robin Hood considers cost-benefit analysis in its grant-making, it
does not rely on this measure exclusively. The foundation recognizes that the
metrics are imprecise, with each CBA calculation depending on complex and
uncertain empirical assumptions. Thus Robin Hood continues to test its metrics
against the informed intuitions of program officers and experts in the field. At
the same time, the foundation presses its staff to justify their intuitions against
the numbers generated by analysis.

12.3.2 Cost-Effective Analysis (CEA)

Since CBA monetizes benefits, it can be used (at least in principle) to compare
benefits across different domains, for example, workforce development, early
childhood, and health programs. When comparing specific benefits within a par-
ticular domain, one can use cost-effective analysis (CEA), which attempts to deter-
mine which policy option will maximize specific benefit at the least cost. CEA
does this by computing the ratio of cost to the nonmonetary benefit for each
option. For example, CEA allows one to compare two similar education pro-
grams in terms, say, of the cost per child graduating from high school. While
CBA’s monetization seeks to compare apples to oranges, CEA compares apples
to apples. CEA analyses of health often measure benefits using the metric of
QALYs (Quality-Adjusted Life Years), which we will discuss below.

12.3.3 Quantifying Reduced Risks in Mortality

Some of the most divisive debates surrounding CBA concern the valuation of
human life—an issue that arises in government regulatory efforts to reduce
workplace, toxic, and other risks. (We’ll defer discussing another divisive issue,
evaluating public environmental goods such as wilderness areas, to Section 14.6.)

81. Robin Hood does not include the administrative costs of making the grant, both
because these costs are not large and because they do not differ significantly from those
of the other poverty-fighting programs to which it will compare Bob’s Jobs.
 choices, consequences, and trade-offs 377

Suppose that 2.5 million people are exposed to an environmental pollutant

that causes cancer in every one of 10,000 people exposed, and thus is expected to
cause 250 cases of cancer. Regulations designed to reduce or obviate the risk are
likely to impose costs on governments, businesses, consumers, and their
intended beneficiaries (for example, employees whose compensation or even job
opportunities may be diminished by a regulation prohibiting exposure to toxic
chemicals). When conducting a cost-effectiveness analysis, how can one quan-
tify the value of each life saved?
Currently, there exist three main methods for the assessment of human life:
VSL (Value of a Statistical Life),VSLY (Value of Statistical Life Years), and QALYs
(Quality-Adjusted Life Years). QALYs are also often used to compare health
procedures in cost-effective analysis (CEA), which compares benefits and costs
that occur within the same domain without necessarily assigning monetary
valuations.
12.2.3a Value of a Statistical Life (VSL) Perhaps the most widely used
method, VSL assigns a fixed monetary value to every human life in a certain
population. It does not seek to estimate the intrinsic value of an individual human
life, but is based on people’s willingness to pay (WTP) to avoid an aggregation of
many small health and safety risks in their lives. As Richard Revesz summa-
rizes, “the value of a statistical life might be more accurately called ‘10,000 times
the value of eliminating a 1 in 10,000 risk.’”82
VSL can be determined through WTP by using:
(1) people’s stated preferences for how much they would pay to eliminate
risks, or
(2) people’s revealed preferences, analyzing decisions that people have already
made about risk trade-offs for goods and/or workplace risk reductions.
Utilizing both these techniques, the Environmental Protection Agency
reviewed twenty-six peer-reviewed value-of-life studies from academic literature
to arrive at a single VSL of $6.3 million for the United States (in year 2000
dollars). Subsequently, however, an Environmental Protection Agency work
group asserted that, since VSL values vary widely by social and economic status,
race, population, and country, there “does not appear to be a universal VSL
value that is applicable to all specific subpopulations.” Furthermore, the work
group recommended that VSL studies using stated and revealed preferences
be analyzed separately, since they sometimes lead to significantly different
estimates.
Citizens of a developing country, such as India, have a lower WTP and thus a
lower VSL than an economically developed country, such as the United States.
In fact, VSL evaluations conducted around the world range from $70,000 to

82. Retaking Rationality 45.

378 problem solving, decision making, and professional judgment

$16.3 million.83 Nonetheless, some scholars advocate, on moral grounds, the use
of a universal VSL for all peoples, regardless of their economic status or willing-
ness to pay. On the other hand, others, including Cass Sunstein, assert that
policy makers should establish specific, local VSLs for communities and coun-
tries. Sunstein argues that VSLs should be assessed within the population where
regulations force the beneficiaries to bear the costs. If an inflated VSL is used,
poorer citizens may be forced to “spend” their money to reduce a workplace risk
when they would prefer to put that money to other uses.84
12.3.3b Value of Statistical Life Years (VSLY) VSL treats the lives of the
very young and very old as equally valuable. Under the VSLY approach, benefits
to individuals are assessed based on their estimated remaining years of life.
VSLY values are derived from VSLs drawn from age-specific populations.85
Michael J. Moore and W. Kip Viscusi, proponents of this approach, note that
“in the case of fatalities, a young person loses a much greater amount of lifetime
utility than does an older person,” and thus assert that it doesn’t make sense
to use the same value for an elderly person with a remaining five-year life
expectancy as for a 25-year-old.86 Critics of VSLY term it the “senior death
discount.”
Revesz uses ozone emission standards to demonstrate the difference between
the VSL and VLSY approaches.87 Assume for the sake of argument that elderly
individuals with existing respiratory problems would be the most likely to bene-
fit from stricter ozone regulations, and that these individuals have an average life
expectancy of five years. Under a VSLY calculation, the benefit of a stricter regu-
lation, using a $180,000 life-year value, would be $900,000. By assessing benefits
at $900,000 per individual instead of at the $6.3 million VSL, the total estimated
regulatory benefit is reduced by 85 percent.
The VSLY approach assumes a decreasing relationship between willingness
to pay and age that is proportional to remaining life expectancy. This empirical
assumption remains contested despite considerable research.88 Some studies

83. Id.135.
84. Cass R. Sunstein, Laws of Fear: Beyond the Precautionary Principle 170
(New York: Cambridge University Press 2005). This distributional problem is further con-
founded when the costs for this population are subsidized or covered by a grant: in this
case, people “will be helped even if risk reduction is based on an excessive VSL.”
85. See Joseph E. Aldy and W. Kip Viscusi, Adjusting the Value of Statistical Life for Age
and Cohort Effects, April 2006, RFF DP 06–19.
86. Michael J. Moore and W. Kip Viscusi, The Quantity-Adjusted Value of Life, 26 Econ.
Inquiry 269 (1988).
87. Retaking Rationality 78–79.
88. Id. 81–82.
 choices, consequences, and trade-offs 379

suggest that WTP peaks around age fifty;89 but elderly citizens may be willing to
pay enormously for risk reductions that prolong life. Because both VSL and
VSLY “may change with age, life expectancy, anticipated future health, income,
and other factors,”90 the U.S. Office of Management and Budget has encouraged
federal agencies to conduct analysis using both approaches.91
12.2.2 c Quality-Adjusted Life Years (QALYs) QALYs incorporate a coeffi-
cient that measures “quality of life” and can be used to assess the efficiency of
health care policies and procedures. Consider an example by Dr. Fritz Allhoff:
“If treating a patient would lead to a life expectancy of 20 years with a high
quality of life (e.g., hypertension, which only requires daily pills and rarely
manifests negative symptoms), coverage for this patient’s treatment should
presumably have priority over treatment for a patient who would have a life
expectancy of 20 years with a comparatively low quality of life (e.g., diabetes,
which requires daily injections and often manifests symptoms such as neu-
ropathy, blindness, etc).”92
This “quality of life” coefficient ranges from 0 to1,valuing a healthy year of life
at 1 and a year where one suffers under a disease or disability at some fraction of
1. The QALYs assigned to a particular reduction in mortality, then, are based on
the expected number of years of life gained adjusted by their perceived quality of
life. Proponents of this method argue that this system distributes finite resources
as efficiently and fairly as possible.
The perceived quality-of-life coefficient, which has a large impact on the final
assigned value, is typically determined through contingent valuation, asking
“healthy people, ex ante, to evaluate various health states.”93 But healthy
individuals may fail to take adaptation into account and thus underestimate the
quality of life of disabled persons, leading to disproportionately low QALY
values. Additionally, as Revesz notes, the “empirical work comparing how will-
ingness to pay relates to QALYs confirms there is no simple dollar-per-QALY

89. Thomas J. Kniesner, W. Kip Viscusi, and James P. Ziliak, Like-Cycle Consumption
and the Age-Adjusted Value of Life, 5 Contributions, Economic Analysis and Policy 1524
(2006).
90. James K. Hammitt, Risk in Perspective: Valuing “Lives Saved” vs. “Life-Years Saved,
Harvard Center for Risk Analysis, March 2008, Volume 16, Issue 1.
91. Id.
92. Fritz Allhoff, The Oregon Plan and QALY’s, Virtual Mentor, Policy Forum, February
2005, Volume 7, Number 2.
93. Retaking Rationality 90.
380 problem solving, decision making, and professional judgment

conversion rate.”94 Although valuations of about $50,000 are often used,95 a study
by the Harvard Center for Risks Analysis reveals that willingness to pay per
QALY fluctuates depending on “both the severity and duration of the illness.”96
The State of Oregon was a leader in using QALYs to prioritize treatments
after deciding to make fewer Medicaid-funded services available to a larger
number of people. In 1989, a commission determined “quality-of-life” coeffi-
cients based on numerous community meetings and phone surveys.97 The com-
mission assigned QALYs to 709 procedures, providing coverage only for those
ranked higher than 587. More specifically, “treatments that prevented death with
a full chance of recovery were ranked first, maternity care was ranked second,
treatments that prevented death but did not guarantee full recovery were ranked
third, and treatments that led to minimal or no improvements in quality of life
were ranked last. For example, “Diagnosis: severe or moderate head injury,
hematoma or edema with loss of consciousness; Treatment: medical and surgi-
cal treatment’ was ranked at the top of the list and ‘Diagnosis: mental disorders
with no effective treatment’ . . . was ranked near the bottom.”98
Inevitably, the rankings produced some seemingly arbitrary and politically
vulnerable choices. For example, “acute headaches” were ranked higher than
some treatments for AIDS or cystic fibrosis.99 The plan was originally rejected
by Health and Human Services, but approved in 1993 under the Clinton
administration after some changes in methodologies and safeguards against
discrimination.100
Britain’s National Health Service, which provides 95 percent of the nation’s
care, also uses QALYs to ration health care. The National Institute for Health

94. The results of several studies are summarized in James K. Hammit,

“Methodological Review of WTP and QALY Frameworks for Valuing Environmental
Health Risks to Children” (paper presented at OECD VERHI-Children Advisory Group
Meeting, Sept. 2006).
95. Richard A. Hirth, Michael E. Chernew, Edward Miller, A. Mark Fendrick, and
William G. Weissert, Willingness to Pay for a Quality-Adjusted Life Year: In Search of a
Standard, 20 Med. Decis. Making 332–42 (2000).
96. See Kevin Haninger and James K. Hammitt, Willingness to Pay for Quality-
Adjusted Life Years: Empirical Inconsistency Between Cost-Effectiveness Analysis
and Economic Welfare Theory (2006).
97. See Gina Kolata, Ethicists Struggle to Judge the “Value” of Life, New York Times,
Nov. 24, 1992.
98. Fritz Allhoff, “THE OREGON PLAN AND QALYS—a controversial oregon plan used
quality-of-life assessments to provide fewer health care services to a larger pool of medic-
aid patients,” virtual mentor, February 2005, volume 7, number 2, http://virtualmentor.
ama-assn.org/2005/02/pfor2-0502.html.
99. Virginia Morell, Oregon Puts Bold Health Plan on Ice, 249 Science, 468, 469–71.
100. See Timothy Egan, Oregon Health Plan Stalled by Politics, New York Times, March
17, 1993,http://www.nytimes.com/1993/03/17/us/oregon-health-plan-stalled-by-politics.
html?sec=&spon=&partner=permalink&exprod=permalink.
 choices, consequences, and trade-offs 381

and Clinical Excellence (NIHCE), which approves treatments and drugs for the
Health Service, determined that, in effect, six months of good quality life is worth
about $22,750. It automatically approves drugs that provide six months of good
quality life for $15,150 or less.101 Dr. Andreas Seiter, a senior health specialist at
the World Bank, notes that “all the middle-income countries—in Eastern Europe,
Central and South America, the Middle East and all over Asia—are aware of
NIHCE and are thinking about setting up something similar.”102

Quantifying Effects on Psychological Health

In Metropolitan Edison Co. v. People Against Nuclear Energy, 460 U.S. 766
(1983), involving the operation of a nuclear power plant near a residential
area the Nuclear Regulatory Commission, the Court was asked to review
the Commission’s assessment of adverse psychological affects under the
National Environmental Policy Act. Justice William Rehnquist noted that
“any effect on psychological health is an effect on health,” but he won-
dered how one could distinguish between “genuine” adverse psychologi-
cal effects that are caused directly by a policy from claims that are rooted
in discontent.

12.3.4 Difficult-to-Measure Benefits

Many regulations and programs aim to improve quality of life in ways that are
not readily measurable by an increase in wage earnings or life years. (The Office
of Management and Budget (OMB) asserts that “a complete regulatory analysis
includes a discussion of non-quantified as well as quantified benefits and
costs.”103) For example, the benefits of the Perry Preschool program were mea-
sured by long-term gains in school attendance, academic achievement, gradua-
tion rates, and earnings, all of which were converted to monetary values.
However, as Lynn Karoly asserts, there still exists a subset of social benefits that

101. Michael D. Rawlins and Anthony J. Culyer, National Institute for Clinical Excellence
and Its Value Judgments, BMJ 2004 Jul 24; 329 (7459):227–9. This has “led many companies
to offer the British discounts unavailable almost anywhere else.” Gardiner Harris, The
Evidence Gap: British Balance Benefit vs. Cost of Latest Drugs, New York Times, Dec. 2, 2008.
102. See id.
103. OMB, Circular A-4, “Regulatory Analysis: Memorandum to the Heads of Executive
Agencies and Establishments” 3 (Sept. 17, 2003).
382 problem solving, decision making, and professional judgment

are often not assigned valuations (or “shadow prices”).104 She offers these catego-
ries with respect to social programs targeting youth:
• Outcomes rarely or never monetized: child/youth behavioral/emotional out-
comes (e.g., behavior problems, school suspensions/expulsions, mental
health outcomes), child/youth cognitive outcomes (e.g., IQ scores), K–12
grades, school attendance, school engagement, general health status of
children/youth, contraceptive use, adult parenting measures, marriage and
divorce, adult mental health, adult reproductive health, employment,
income, and poverty status.
• Outcomes without clear consensus on monetization: achievement test scores,
high school graduation, college attendance, teen pregnancy, tobacco use,
alcohol abuse, illicit drug use, crime and delinquency, and child abuse and
neglect.
• Outcomes with more-established methodology for monetization: special educa-
tion use, grade retention, transfer payments, and other means-tested
programs
Even if one can assign values to such benefits, questions of judgment remain
about which to include in CBA. These are appropriately questions not for eco-
nomic experts, but for policy makers.105

12.3.5 Ancillary Costs and Benefits

Government regulations and policies can give rise to both foreseen and unfore-
seen harms and benefits. Direct-risk trade-offs occur when diminishing one risk
of harm creates a countervailing risk: for example, reducing chlorine levels in
water may decrease the risk of cancer but increase microbial disease.106 A substi-
tution effect occurs when a regulation (for example making flying more oner-
ous) causes a shift in behavior toward taking other risks (driving accidents).
However, regulations can also produce ancillary benefits. To comply with the
Clean Air Act, “municipalities have constructed wetlands as an alternative to
conventional wastewater treatment facilities,” thereby generating “habitat cre-
ation and preservation, carbon sequestration, erosion control, and recreational
and research opportunities”107

104. Lynn A. Karoly, “Valuing Benefits in Benefit-Cost Studies of Social Programs,”

RAND: Labor and Population, pg. 78, chart 4.1, http://www.rand.org/pubs/technical_
reports/TR643/.
105. Jeffrey L. Harrison, Thomas D. Morgan, and Paul R. Verkuil, Regulation
and Deregulation: Cases and Materials, American Casebook Series 424 (2d ed.
St. Paul: West Publishing Co., 1997).
106. See John D. Graham and Jonathan Baert Wiener, Confronting Risk Tradeoffs, in
Risk Versus Risk: Tradeoffs in Protecting Health and the Environment
(Cambridge, MA: Harvard University Press, 1995).
107. See Steven Piper and Jonathan Platt, Benefits From Including Wetland Component
in Water Supply Projects, 124 J. Water Resources Planning & Mgmt. 230 (1998) and EPA,
 choices, consequences, and trade-offs 383

One study indicated that the regulation of carbon monoxide emissions from
motor vehicles to improve air quality had an enormous ancillary benefit of reduc-
ing cancer risks, thus saving an average of 25,000 lives per year—whereas other
health benefits had been estimated only at 212 to 515 lives annually.108 And
health or safety regulations may have what Revesz calls an “attentiveness effect,”
inducing people to become more conscious of safety issues beyond what is
mandated.109
Policy makers often only consider unintended costs. However, the OMB rec-
ommends that agencies account for ancillary benefits of regulations as well,110
and Revesz argues that a complete cost-benefit analysis requires as much.111 In a
2007 decision, the U.S. Court of Appeals for the Ninth Circuit invalidated the
National Highway Traffic Safety Administration (NHTSA)’s fuel economy stan-
dards for light trucks, noting that while NHTSA had accounted for the negative
consequences of the policy on employment and automobile sales, it failed to
include the ancillary beneficial effects on climate change. The court said that
“NHTSA could not put a thumb on the scale by undervaluing the benefits.”112

12.3.6 The Value and Limits of Cost-Benefit Analysis

Cost-benefit analysis depends on determining people’s WTP for certain benefits.
Yet individuals’ WTP is essentially irrelevant to many important policy deci-
sions, such as discrimination and abortion; as Cass Sunstein notes, these are
appropriately resolved by citizens’ preferences and values.113 At least with respect
to environmental decisions, Frank Ackerman, Lisa Heinzerling, and Mark Sagoff
argue that CBA is a “crude tool . . . that buries indefensible judgments of
morality and politics”114
But environmental policies often do impose costs in return for benefits, and
it seems equally indefensible for policy makers to blind themselves to these.
Following Sunstein, our own view is that if one is explicit about its inaccuracies,
biases, and limitations, CBA can provide a “concrete sense of what actually is

Office of Water, Constructed Wetlands for Wastewater Treatment and Wildlife Habitat
(91993), EPA 832-R-93-005, available at http://www.epa.gov/owow/wetlands/construc/.
108. See M. Shelef, Unanticipated Benefits of Automotive Emission Control: Reduction in
Fatalities by Motor Vehicle Exhaust Gas, 146/147 Sci. Total Envtl. 93 (1994).
109. Retaking Rationality 175.
110. OMB, Circular A-4, “Regulatory Analysis: Memorandum to the Heads of Executive
Agencies and Establishments” 3 (Sept. 17, 2003).
111. Retaking Rationality 145.
112. Center for Biological Diversity v. National Highway Traffic Safety Administration
(NHTSA), No. 06-71891, 13,871 (9th Cir. Nov. 15, 2007).
113. Cass R. Sunstein, Laws of Fear: Beyond the Precautionary Principle, supra.
114. Frank Ackerman and Lisa Heinzerling, supra. See also Mark Sagoff, The
Economy of the Earth: Philosophy, Law, and the Environment (Cambridge Studies
in Performance Practice).
384 problem solving, decision making, and professional judgment

at stake.”115 Consider James Hammit’s March 2003 letter to the editor of the
New York Times:116
The Office of Management and Budget’s interest in applying cost-benefit
analysis to homeland security measures is laudable. Even if precise quantifi-
cation of the benefits to security and harms to civil liberties is impossible,
analytic consideration of the trade-offs is wise.
As John Graham, director of regulatory affairs at O.M.B., says, “Simply
identifying some of these costs will help understand them and get people to
think about alternatives that might reduce those costs.”
Cost-benefit analysis should also have been used to evaluate the potential
war in Iraq. The benefits of ousting Saddam Hussein should be measured
against the costs of casualties, waging war, reconstruction, retaliatory terror-
ism and increased anti-Americanism. Such analysis would prompt us “to
think about alternatives that might reduce those costs.”

115. Cass R. Sunstein, Laws of Fear: Beyond the Precautionary Principle,

supra at 174.
116. James K. Hammitt, Security, at a Price: Letter to the Editor, New York Times,
March 14, 2003, http://query.nytimes.com/gst/fullpage.html?res=9F02E4DC103EF937A
25750C0A9659C8B63.
13. complexities of decision making
Decision Processes; Relationships to our Future Selves

Neoclassical economics and the axioms of expected utility theory posit a model
of the individual as homo economicus. In Gary Becker’s words, all human behav-
ior can be viewed as involving participants who maximize their utility from a
stable set of preferences and accumulate an optimal amount of information and
other inputs in a variety of markets.1
The next several chapters inquire whether and when decision makers deviate
from this model. This chapter begins with a compilation of characteristics of a
decision that may lead to suboptimal outcomes, and then discusses how the
choice of decision-making strategies can affect outcomes. The bulk of the chap-
ter focuses on people’s relationships with their own future selves—their beliefs
about how today’s decisions will affect their futures, and limits on their capacity
to plan and implement plans to improve their future well-being. Chapter 14 con-
siders how the outcomes of decisions can be affected by the ways that choices are
framed. Chapter 15 considers the effects of framing and other matters when
decisions are made in conditions of risk or uncertainty. Chapter 16 continues the
discussion of risk in situations of high affect.

13.1 how would one know if a person made a suboptimal decision?

Since introducing the concept of bounded rationality in Section 1.6.2, we have

referred to the fact that human beings have neither infinite computational power
nor infinite time to make decisions. Because gathering and comparing the infor-
mation that bears on a decision can be a time-intensive, costly, and mentally
effortful process, we usually satisfice rather than try to maximize expected utility.
We implicitly take into account the time, money, and psychological impact of
making the decision by using shortcuts or heuristics, including noncompensa-
tory strategies of the sort described in Section 4.3.
Doubtless, many particular decisions that seem suboptimal make perfect
sense—indeed become optimal—when one takes these decision costs into
account. Our goal for the chapter is not to assess particular decisions,
however, but to see whether there are systematic ways in which people make

1. Gary S. Becker, The Economic Approach to Human Behavior 14 (Chicago:

University of Chicago Press, 1976).
386 problem solving, decision making, and professional judgment

suboptimal decisions. This raises the preliminary question of how one would
know whether a decision maker had made a suboptimal decision. The following
characteristics do not necessarily lead to poor decisions, but they all reflect poten-
tial hazards for optimal decision making:
1. The decision is based on incorrect data or the incorrect analysis of data—
whether the result of biases (e.g., the availability or representativeness heu-
ristic), poor predictions of one’s own future desires or what it will take to
achieve them, or simple miscalculation.
2. The decision violates one of the axioms of expected utility, or rational
choice. (See Section 12.2.2). For example, it violates the principle of domi-
nance or transitivity.
3. The decision maker poorly integrates utilities from the decision-making
process with utilities from the outcome of the process—for example, giving
excessive weight to the utility of avoiding confronting a stressful choice
compared to the enduring consequences of making a poor choice.
4. The decision is sensitive to the way the issue is framed and is made in a con-
text where the framing is highly variable or manipulable. (See Chapter 14.)
5. The decision is sensitive to affect (i.e., emotion) and is made in a context
where affect is highly variable or manipulable, or where present affect is
likely to be very different from affect when experiencing the consequences
of the decision. (See Chapter 16.)
6. The decision maker was subjected to undue social influence. (We postpone
consideration of this subject to Chapter 17.)
We will examine some of these phenomena and other factors that can affect
the decision-making process, often for the worse.

13.2 the choice of decision-making strategies

13.2.1 Overview of the Decision-Making Process: Decision-Making Bricolage and

the Ad Hoc Construction of Preferences2
Christine Lamm approached the problem of siting the wastewater treatment
plant knowing in advance that she would employ a particular decision-making
strategy—the subjective linear model. And even though quantifying the relative

2. The French word bricolage means “doing odd jobs.” Borrowing from Claude Levi-
Strauss’s use of the term, a bricoleur is someone who improvises and uses any means or
materials that happen to be lying around in order to tackle a task. Claude Levi-Strauss,
The Savage Mind 19 (Chicago: The University of Chicago Press, 1966).
 complexities of decision making 387

values of the factors, or attributes, involved had an inevitable degree of arbitrari-

ness, she was reasonably clear about the city’s interests and objectives.
In actuality, though, people often do not decide on a particular decision-
making strategy in advance. Rather they begin with whatever process consciously
or unconsciously suggests itself as appropriate for the issue, and often switch
strategies as the process develops. Also, as we observed in Chapter 2, “human
beings have unstable, inconsistent, incompletely evoked, and imprecise goals.”3
For these reasons, preferences are often constructed during the decision-making
process.4 Much of this chapter is concerned with the malleability or lability of
preferences. But first a few notes about the processes of decision making.
Most important decisions require making trade-offs among a number of attri-
butes. The values of different attributes may be unclear; the time horizons of the
benefits and costs of particular outcomes may differ; and some trade-offs can
cause emotional conflict. Ola Svenson has described decision making as “a kind
of conflict resolution in which contradictory goals have to be negotiated and
reconciled.”5 A decision maker’s strategy may thus depend on the cognitive and
emotional difficulties of making the decision. The choice of strategy may itself
reflect a trade-off between accuracy on the one hand, and minimizing cognitive
effort and negative affect on the other.
For example, as the cognitive or emotional costs increase, people tend to
employ simplified noncompensatory strategies described in Section 4.3, such as
elimination by aspects (EBA). Some decisions demand making “taboo trade-
offs”6 between commodifiable and a noncommodifiable attributes, such as cost
versus safety. Merely considering such trade-offs elicits personal moral discom-
fort, and may subject the decision maker to moral censure. For example, partici-
pants in one study expressed outrage when they read about a hospital
administrator who considered spending $100,000 on improving the hospital
rather than on saving the life of one child—even when the administrator ulti-
mately chose to save the child. By contrast, an administrator who chose to save
the child without considering the trade-off elicited little outrage.7 No wonder,

3. James G. March, Bounded Rationality, Ambiguity, and the Engineering of Choice, 9

Bell Journal of Economics 587 (1978).
4. Sarah Lichtenstein and Paul Slovic, Construction of Preferences (New York:
Cambridge University Press, 2006)
5. Ola Svenson, Decision Making and the Search for Fundamental Psychological
Regularities: What Can Be Learned from a Process Perspective?, 65 Organizational
Behavior and Human Decision Processes 252 (1996).
6. Philip E. Tetlock, The Impact of Accountability on Judgment and Choice: Toward
a Social Contingency Model, in M. Zanna (ed.), 25 Advances in Experimental Social
Psychology 331–376 (New York: Academic Press, 1992).
7. Philip E. Tetlock, Orie V. Kristel, S. Beth Elson, Melanie C. Green, and Jennifer S.
Lerner, The Psychology of the Unthinkable: Taboo Trade-offs, Forbidden Base Rates, and Heretical
Counterfactuals, 78 Journal of Personality and Social Psychology 853 (2000).
388 problem solving, decision making, and professional judgment

then, that decision makers faced with taboo trade-offs often use choice-avoidance
strategies like putting off the decision indefinitely or delegating the responsibil-
ity of deciding to someone else.8 The emotional difficulties of making taboo
trade-offs—as well as the risk that others will condemn the person who consid-
ers them—can thus shape the strategy that decision makers select.

13.3 value-based and reason-based decision-making strategies

In Section 4.4, we examined two kinds of systematic decision-making strategies:

value-based and reason-based. Recall Christine Lamm’s use of a subjective linear
decision-making model in deciding where to site the wastewater treatment plant
and Charles Darwin’s approaches to marriage. Christine’s procedure was value-
based: She assigned a (subjective) value to the attributes she deemed important,
summed the values for each alternative, and chose the alternative that maxi-
mized the total value. Darwin followed a reason-based approach, giving reasons
for and against the alternatives under consideration. Each approach has its pros
and cons.

13.3.1 Value-Based Decision-Making Strategies: Difficulties in Evaluating Certain

Aspects of the Alternatives Being Considered
We return briefly to a cost-benefit analysis issue raised in Section 12.3: difficul-
ties in evaluating certain aspects of the alternatives being considered.
While making trade-offs among different values can be exceedingly difficult,
expected utility theory depends fundamentally on a decision maker’s ability to
evaluate the attributes of a particular decision. Christine finds the attributes rel-
evant to siting the wastewater treatment plant evaluable—for example, she has
very good criteria for determining cost, and reasonably good criteria for evaluat-
ing environmental and traffic impacts and the other factors.
But evaluating even quantifiable information can sometimes be difficult.
Evaluability refers to the ease or difficulty of evaluating an attribute of the alter-
native choices you are considering. Experiments suggest that when options or
alternatives “involve a trade-off between a hard-to-evaluate attribute and an easy-
to-evaluate attribute, the hard-to-evaluate attribute has a lesser impact in sepa-
rate evaluation than in joint evaluation, and the easy-to-evaluate attribute has a
greater impact.”9

8. Tetlock, The Impact of Accountability on Judgment and Choice, supra.

9. Christopher Hsee, The Evaluability Hypothesis: An Explanation for Preference Reversals
between Joint and Separate Evaluations of Alternatives, 67 Organizational Behavior and
Human Decision Processes 247–57 (1996).
 complexities of decision making 389

For example, participants in a study were asked to assume that they were
music majors purchasing a music dictionary in a used book store. In the joint-
evaluation condition, they were told that there were two music dictionaries
for sale:
• one dictionary was “like new” and had 10,000 entries;
• the other had a torn cover and had 20,000 entries.
On average, the participants were willing to pay $19 for the first and $27 for
the second. But when participants were told that there was only one music dic-
tionary in the store, those presented with only the first dictionary were willing to
pay $24 and those presented with only the second were willing to pay $20: a torn
cover is easy to evaluate without comparison, while the number of entries is dif-
ficult to evaluate. This phenomenon of preference reversals seems robust in a
variety of different contexts. Table 13.1 provides some examples where people’s
preferences were reversed in separate and joint evaluations.1011
table 13.1 easy and hard to evaluate attributes

Options Easy-to-evaluate Hard-to-evaluate

Computer programmer 4.9 GPA vs. 3.0 GPA Number of computer programs written
job applicants10 in past 2 years: 10 vs. 70
Eye surgeons for laser Harvard MD Degree Number of similar successful
surgery11 vs. Iowa MD Degree procedures: 80 vs. 300

Especially when the values of attributes are not readily monetizable, people
are often influenced more by proportions than absolute values. For example,
Paul Slovic and his colleagues found that, in separate evaluations, people sup-
ported an airport-safety measure expected to save 98 percent of 150 lives at risk
more strongly than a measure expected to save 150 lives (period).12 They explain:
“Saving 150 lives is diffusely good, hence only weakly evaluable, whereas saving
98% of something is clearly very good because it is so close to the upper bound
on the percentage scale, and hence is readily evaluable and highly weighted
in the support judgment.” They report that “subsequent reduction of the

10. Id.
11. Brian J. Zikmund-Fisher, Angela Fagerlin, and Peter A. Ubel, Is 28% Good or Bad?
Evaluability and Preference Reversals in Health Care Decisions, 24 Medical Decis. Making
142 (2004). Evaluation was on a scale of 0 (extremely bad choice) to 10 (extremely good
choice) rather than WTP.
12. Paul Slovic, Melissa Finucane, Ellen Peters, and Donald G. MacGregor, Rational
Actors Or Rational Fools: Implications Of The Affect Heuristic For Behavioral Economics, 31
Journal of Socio-Economics 329 (2002).
390 problem solving, decision making, and professional judgment

percentage of 150 lives that would be saved to 95%, 90%, and 85% led to reduced
support for the safety measure but each of these percentage conditions still gar-
nered a higher mean level of support than did the save 150 lives condition.”
Suppose that, in the 1990s, there were two camps for Rwandan refugees in
neighboring Zaire. One camp holds 250,000 refugees and the other 11,000.
Cholera is rampant in both camps. A nongovernmental organization can pro-
vide enough clean water to save the lives of a total of 4,500 refugees suffering
from cholera in one or the other camp. Which one would you chose? From a
moral or policy point of view, the number of refugees in the camps makes no
difference,13 and 42 percent of the participants in a study said so. However, 44
percent would have directed the resources to the smaller camp, where a larger
proportion of lives would have been saved.14 Slovic observes that number of lives
saved, standing alone, appears to be poorly evaluable, but becomes clearly evalu-
able and important in side-by-side comparisons.

13.3.2 Reason-Based Decision-Making Strategies15

We often think and talk about choices in terms of reasons—especially when faced
with conflicts among choices. The rule of law depends on lawyers, judges, and
policy makers explaining the reasons for their decisions; and it is difficult to
imagine a lawyer effectively counseling a client without the two engaging in rea-
soned discourse. Especially in making personal decisions, reasons may include
considerations, such as anticipating regret for a bad choice, that are not readily
captured by a value-based approach. A reason-based approach can also obviate
the seeming false precision required by assigning numbers to intangible values.
However, reasons often suffer from the opposite problem—of being vague,
or of the decision maker’s omitting potentially salient reasons pro or con.
Moreover, people may be unaware of what actually motivated their choice.
13.3.2.a The Effect of Giving Reasons on Decision Strategies and
Outcomes Consider the position of Adele, in charge of entry-level hiring at a law
firm or government agency. If she were going to use a subjective linear model to
decide which, among many applicants, to invite to spend a day interviewing, she
would initially determine a set of attributes—e.g., GPA, quality of the school—
and rank each candidate in terms of them.

13. Actually, one could imagine variables, such as preventing riots or contagion, that would
affect the policy choice. But such factors did not play a role in the participants’ decisions.
14. In another set of questions focusing on only one camp, one group of participants
was told that prior aid had met 5 percent of the camp’s water needs, and that the new aid
would bring the total to 7 percent; another group was told that prior aid met 95 perecnt of
the needs and the new aid would bring the total to 97 percent. Most participants would
have directed aid to the latter.
15. This section draws heavily on Eldar Shafir, Itamar Simonson, and Amos Tversky,
Reason-based Choice, 49 Cognition 11 (1993).
 complexities of decision making 391

Now suppose that Adele delegates the decision to Barbara, who, rather than
using a subjective linear model, looks at the files one by one, intuitively sorting
them in “yes” or “no” piles. Do you think she will reach the same result as Adele
did? Consider the role of the availability heuristic, discussed in Section 9.6.
Suppose that Adele gives the identical files to Clara and David and asks Clara
to produce a list of the names of candidates to invite to interview and David to
produce a list of those who should be sent a rejection letter. Who is likely to have
more candidates in the running? Experiments suggest that David will leave more
candidates in the running than Clara. Resumes and application letters tend to
favor a candidate’s strong points rather than weaknesses. Clara will only select
candidates who stand out among the rest, while David will find fewer negative
characteristics on which to reject the candidates.16
Along these lines, Eldar Shafir and his colleagues conducted an experiment
in which participants were given the following scenario: Imagine that you serve
on the jury of an only-child sole-custody case following a relatively messy divorce.
The facts of the case are complicated by ambiguous economic, social, and emo-
tion considerations, and you decide to base your decision entirely on the charac-
teristics shown in Table 13.2.

table 13.2 parents’ qualities in a child-custody dispute

Parent A Parent B

Average income Above-average income

Average health Minor health problems
Average working hours Lots of work-related travel
Reasonable rapport with the child Very close relationship with the child
Relatively stable social life Extremely active social life

Notice that Parent A is average in all respects, while Parent B has some strong
and weak points.
When asked to whom they would award sole custody, the majority of partici-
pants (64 percent) chose Parent B to receive custody. Yet when a different group
of participants was asked to whom they would deny sole custody, the majority (55
percent) chose the opposite outcome and denied custody to Parent B.
Based on this and similar experiments, the researchers conclude that
when people are basing decisions on reasons, “the positive features of options
(their pros) will loom larger when choosing, whereas the negative features

16. Vandra L. Huber, Margaret A. Neale, and Gregory B. Northcraft, Decision Bias and
Personnel Selection Strategies, 40 Organizational Behavior and Human Decision
Processes 137 (1987).
392 problem solving, decision making, and professional judgment

of options (their cons) will be weighted more heavily when rejecting.”17 But this
phenomenon violates the axiom of invariance (Section 12.2.2.b): the way in which
the question is framed should not, by itself, affect the outcome.
13.3.2.b Preference Reversals Between Prices and Choices Suppose you are
presented with two gambles:
• Gamble A: 0.9 chance to win $8
• Gamble B: 0.2 chance to win $30
Suppose that you are asked to choose between the two gambles? Now suppose
you are asked to price each gamble—e.g., suppose that you own tickets entitling
you to each, at what price would you would be willing to sell the tickets? On aver-
age, people asked to choose select gamble A, which has a much higher probabil-
ity of winning. But when asked to price each gamble, they demand more for B.18
There are two explanations for this violation of invariance: the compatibility
effect and the prominence effect.19 Under the compatibility effect, when subjects
are thinking in terms of dollars, as they do when asked to set a price, they are
more likely to evaluate their response in terms of dollars. Under the prominence
effect, choice invokes reasoning in a way that pricing doesn’t—people feel a need
to explain the choice at least to themselves—and the gamble with the higher
probability is easier to explain.
Such preference reversals have been manifested in other situations,
including people’s valuations of reducing the risks from chemical products,
improvements in consumer goods, and improvements in the environment.20
We will say a word more about this in a discussion of contingent valuation in
Section 14.6.
13.3.2.c The Effect of Giving Reasons on the Quality of Choices and
Satisfaction with Them Using reasons as a strategy can make decision making
easier and more transparent. However, excessive reliance on reasons and
deliberation has its own drawbacks. A number of experiments by Timothy
Wilson and his colleagues suggest that giving reasons for one’s preferences

17. Eldar Shafir, Choosing versus Rejecting: Why Some Options Are Both Better and Worse
than Others, 21 Memory & Cognition 546–56 (1993).
18. Amos Tversky and Richard H. Thaler, Anomalies: Preference Reversals, 4 Journal
of Economic Perspectives 193–205 (1990).
19. Paul Slovic, Dale Griffin, and Amos Tversky, Compatibility Effects in Judgment
and Choice, in Insights in Decision Making: A Tribute to Hillel J. Einhorn 5–27
(Robin M. Hogarth ed., Chicago, IL: University of Chicago Press, 1991); Amos Tversky,
Schmuel Sattath, and Paul Slovic, Contingent Weighting in Judgment and Choice, 85
Psychological Review 371–84 (1988).
20. Julie R. Irwin, Paul Slovic, Sarah Lichtenstein, and Gary H. McClelland, Preference
Reversals and the Measurement of Environmental Values, 6 Journal of Risk and
Uncertainly 5–18 (1993).
 complexities of decision making 393

before making a choice sometimes leads to suboptimal decisions and to less

satisfaction with the choice.21 For example, students were asked to evaluate
five posters—two of paintings by Monet and Van Gogh and three contemporary
humorous cartoons or photos. Members of the control group, who were
not asked to consider their reasons for liking or disliking the posters, tended to
prefer the paintings. By contrast, students who were asked to consider the rea-
sons underlying their preferences tended to prefer the humorous posters.
The students in both groups were then given the posters of their choice. At
the end of the semester, they were asked about their satisfaction with the poster,
including whether it was hanging on their wall. Members of the control group
were significantly more satisfied with their choices than those who had been
ask to introspect. “Apparently the qualities of the art posters that made them
appealing to our subject population were relatively difficult to verbalize, whereas
positive features of the humorous posters were easy to verbalize, producing a
biased sample of reasons” While “verbalizing these reasons caused participants
to adopt the attitude they implied,” over time the student returned to their initial,
unreflective evaluations.

A centipede was happy quite, until a toad in fun

Said, “Pray, which leg comes after which?”
This raised his doubts to such a pitch
He fell distracted in the ditch
Not knowing how to run
—Old rhyme

Similar experiments have been conducted with respect to the choice of jams,
college courses, and other items—with similar results. With respect to the effects
of reason giving, Timothy Wilson and Jonathan Schooler write:22
Forming preferences is akin to riding a bicycle; we can do it easily but cannot
easily explain how. Just as automatic behaviors can be disrupted when people
analyze and decompose them, so can preferences and decisions be disrupted
when people reflect about the reasons for their feelings. We suggest that this

21. See Timothy D. Wilson and Jonathan W. Schooler, Thinking Too Much: Introspection
Can Reduce the Quality of Preferences and Decisions, Journal of Personality and Social
Psychology 181 (1991); Timothy D. Wilson et al., Introspecting About Reasons Can Reduce
Post-Choice Satisfaction, 19 Personality and Social Psychology Bulletin 331 (1993).
22. Timothy Wilson and Jonathan Schooler, Thinking Too Much: Introspection Can
Reduce the Quality of Preferences and Decisions, 60 J Pers. Soc. Psychol. 181–192 (1991).
394 problem solving, decision making, and professional judgment

can occur as follows. First, people are often unaware of exactly why they feel
the way they do about an attitude or an object. When they reflect about their
reasons, they thus focus on explanations that are salient and plausible. The
problem is that what seems like a plausible cause and what actually deter-
mines people’s reactions are not always the same thing. As a result, when
asked why they feel the way they do, people focus on attributes that seem like
plausible reasons for liking or disliking the stimulus, even if these attributes
have no actual effect on their evaluations.
. . . [U]nder some circumstances . . . people will focus on reasons that
imply a different attitude than they held before and will adopt the attitude
implied by these reasons. . . . [P]eople often do not have a well-articulated,
accessible attitude and thus do not start out with the bias to find only those
reasons that are consistent with an initial reaction. They conduct a broader
search for reasons, focusing on factors that are plausible and easy to verbalize
even if they conflict with how they felt originally.
[W]e suggest that reflecting about reasons will change people’s attitudes
when their initial attitude is relatively inaccessible and the reasons that are
salient and plausible happen to have a different valence than people’s initial
attitude.
In short, because what you can readily articulate doesn’t necessarily reflect
what you deem most important, articulating reasons for your preferences can
lead to choices inconsistent with your actual preferences. Does articulating rea-
sons interfere with judgments more generally? Research by John McMakin and
Paul Slovic suggests that it depends on the kind of judgment being made. In one
study, McMakin and Slovic asked participants to estimate how their peers would
react to a series of advertisements. As with judging one’s own preference for
posters, this task requires a holistic, intuitive approach. As in Wilson and
Schooler’s study, participants who were asked to articulate reasons for their esti-
mate provided less accurate estimates than those not asked to articulate reasons.
In a second study, however, McMakin and Slovic had participants make judg-
ments that required analytic rather than intuitive thinking, such as estimating
the length of the Amazon river. For this task, generating reasons increased the
accuracy of estimates.
Why would the first judgment task have been better suited for intuitive think-
ing, while the second demand more analytic thinking? The complexity and
ambiguity of the task may provide part of the answer. Posters and advertise-
ments vary in countless ways that affect preferences; identifying the relevant
dimensions and keeping all in mind simultaneously is a complex task with
ambiguous criteria. By contrast, estimating the length of the Amazon is a prob-
lem that, while difficult, does not involve multiple, ambiguous dimensions.
Is it possible that holistic, intuitive thinking may lead to better judgments for
more complex tasks while analytic, reasons-based thinking may work better for
 complexities of decision making 395

less complex tasks? Some evidence for this proposition comes from work by Ap
Dijksterhuis and his colleagues.23 In one study, participants tried to choose the
best car from a selection of descriptions that varied on either twelve dimensions
(a relatively complex task) or on only four dimensions (a relatively simple task).
The descriptions were crafted so that, in each condition, one car was objectively
superior to the others. The researchers instructed half of the participants to spend
four minutes thinking carefully about their judgment, and the other half to spend
four minutes doing a task that distracted them from thinking about the matter.
Results showed that participants told to think carefully made more accurate judg-
ments about the cars when the task was simple than when it was complex. By
contrast, participants who were distracted from generating reasons for their judg-
ment were more accurate when the task was complex than when it was simple.
It would be a mistake to conclude that one should always eschew reasoned
analysis of complex decisions in favor of intuition. First, one might question the
generalizability of these findings to the sorts of consequential decisions facing
policy makers, who often have longer to deliberate than the participants in these
studies. More fundamentally, as we see throughout this book, “gut feelings” can
be biased by emotions unrelated to the decision at hand, or by prejudices and
stereotypes that can only be mitigated by deliberate reflection.
Still, the research reviewed in this section indicates that reason-based decision
making does not always lead to optimal judgments. As Eldar Shafir’s study about
custody awards demonstrates, the way a question is framed can affect the kinds
of reasons that one generates. Additionally, when we grapple with complex prob-
lems, the reasons we generate may indicate more about what is easy to articulate
than about what is most relevant to the judgment at hand. Without ignoring
the ability of reasoned analysis to enhance judgment, then, we should exercise
caution in over-relying on those reasons that come most easily to mind.

13.4 predicting our future well-being

There are two tragedies in life. One is not to get your heart’s desire; the other is to get it.
—George Bernard Shaw

Vronsky meanwhile, although what he had so long desired had come to pass, was not
altogether happy. He soon felt the fulfillment of his desires gave him only one grain
of the mountain of happiness he had expected. This fulfillment showed him the
eternal error men make in imagining that their happiness depends on the realiza-
tion of their desires.
—Leo Tolstoy, Anna Karenina (1887)

23. Ap Dijksterhuis, Maarten W. Bos, Loran F. Nordgren, and Rick B. van Baaren, On
Making the Right Choice: The Deliberation-Without-Attention Effect, 311 Science 1005 (2006).
396 problem solving, decision making, and professional judgment

It seems that the linear model used by Christine Lamm in the wastewater
treatment plant problem might be able to overcome some of the shortcomings
of the reason-based choice strategy we have just reviewed. However, decision
makers almost never have perfect information relating to a decision—
especially in conditions of risk and uncertainty. In Chapter 15, we will discuss
how to deal with uncertainties about objective matters—for example the weather
or, in Lamm’s case, problems involving acquiring the site near Edenville’s
airport.
Even more difficult to determine are uncertainties about subjective matters—
for example, how the residents of north Edenville will actually experience having
the wastewater treatment plant in their neighborhood.
The following sections explore ways that people may make systematic
errors—akin to the cognitive biases discussed in earlier chapters—in predicting
how they will feel about things in the future. In addition to considering their
relevance to individual decision making, it will be interesting to ask how a policy
maker should take them into account.
Your actual experience of the consequences of the decision is termed the
experienced utility of a decision. At the time you make a decision, however, you
can only predict the decision’s effect on experienced utility. Ideally, your
predicted utility is perfectly congruent with experienced utility. Unfortunately,
as we all know from personal experience, they can diverge. The giant-size box
of popcorn you bought when entering the movie theater no longer seems
so appealing when you’ve eaten half of it. The dress that seemed to make a
daring fashion statement in the store sits in your closet after being worn
only once. While “the ability to predict the intensity of one’s future likes and
dislikes appears to be an essential element of rational decision making,” the task
of predicting one’s future feelings and tastes “turns out to be surprisingly
difficult.”24
We separate these issues into two related parts. The first concerns what Daniel
Gilbert and Timothy Wilson have termed impact bias—mispredictions of how
you will feel in the future about a current occurrence or event, assuming
that your basic preferences have not changed in the interim.25 The second
involves what George Loewenstein and his colleagues have termed projection
bias—mispredictions of your future preferences.26

24. Daniel Kahneman and Jackie Snell, Predicting a Changing Taste: Do People Know
What They Will Like, 5 Journal of Behavioral Decision Making 187 (1992).
25. Daniel T. Gilbert, Elizabeth C. Pinel, Timothy D. Wilson, Stephen J. Blumberg, and
Thalia P. Wheatley, Immune Neglect: A Source of Durability Bias in Affective Forecasting, 75
Journal of Personality and Social Psychology 617–38 (1998).
26. See George Loewenstein and Erick Angner, Predicting and Indulging Changing
Preferences, in George Loewenstein, Daniel Read, and Roy Baumeister eds., Time and
 complexities of decision making 397

Legal Note

Your client, a historian, enraged by claims that a critic has made that portions
of the book were plagiarized, wants to sue the critic for defamation. Your tell
your client that he has a fairly good chance of ultimately winning a small
judgment but that, in addition to financial costs, litigants tend to incur heavy
emotional costs during the process, which often outweigh the satisfactions
of victory. However, your client is focused on vindication.
See Jeremy Blumenthal, Law and the Emotions: The Problems of Affective Forecasting,
80 IND. L.J. 155 (2005).

13.4.1 Impact Bias: Mispredicting Emotional Reactions to Future Events

A personal decision implicitly requires predicting your future feelings about
people (a spouse or partner), objects (a new car), or changed life circumstances
(aging). Most people make pretty good predictions much of the time. But here
are some examples of situations where predictions tend to be exaggerated:27
• People who had won the lottery predicted that they would be significantly
happier a year later than they turned out to be, and people who had suf-
fered an accident that left them paraplegic did not report themselves as
greatly less happy than a control group—though we have little doubt that
they would have paid a great deal to restore their mobility.
• Asked about how environmental and personal changes (e.g., air quality,
income, and body weight) had affected their well-being and would affect it
in the future, people predicted that future changes would have a much
greater effect than they reported from past changes of the same nature.
Though predicting that the difference in climates would make a big differ-
ence, students at California and Midwestern universities reported the same
level of well-being.
• People residing near a newly opened highway underestimated how much
they would adapt to the noise.

Decision: Economic and Psychological Perspectives on Intertemporal Choice 351

(New York: Russell Sage Foundation, 2003) [hereafter Time and Decision].
27. This section draws heavily on the excellent review essay, George Loewenstein and
David Schkade, Wouldn’t It Be Nice, in Daniel Kahneman, Ed Diener, and Norbert
Schwarz eds., Well-Being: The Foundations of Hedonic Psychology 95 (New York:
Russell Sage Foundation, 1999).
398 problem solving, decision making, and professional judgment

• Assistant professors predicted that they would be less happy during the
first five years after being turned down for tenure, and happier during the
first five years after gaining tenure, than they actually were.
• Dental patients, and especially those who were anxious about their appoint-
ments, overpredicted the amount of pain they would experience.
• People being tested for HIV or for (unwanted) pregnancy overpredicted
misery from a positive result and elation from a negative result.
• People overpredict their happiness if their favored teams or political candi-
dates wins and their unhappiness if they lose.
Most of these predictions involve errors in what Gilbert, Wilson, and their
colleagues have termed affective forecasting. Predictions can err in several
respects.
Although we are usually correct in predicting whether an emotion will have
positive or negative valence, we may oversimplify situations that evoke complex
emotions by focusing on only positive or negative affect—e.g., students looking
forward to graduation with joy and pride, overlooking sadness about leaving
friends and apprehension about their futures.28 We also sometimes mispredict
the specific positive or negative emotions we will experience—e.g., women who
predicted that they would respond with anger to sexually harassing questions in
a job interview tended instead to respond with fear.29
More generally, as illustrated by Figure 13.1,30 we often mispredict the inten-
sity and duration of emotional experiences. The remainder of this subsection
explores the possible causes of problems in impact bias—“the tendency to over-
estimate the enduring impact that future events will have on our emotional
reactions.”31
13.4.1.a Focalism One reason that we fall prey to the impact bias is that our
affective forecasts tend to place disproportionate emphasis on whatever our
attention is focused on at the moment. Such focalism can lead to systematic
errors in affective forecasting because features that are particularly salient when
predicting an event may be joined and even swamped by a host of other factors
when experiencing the event. This phenomenon is reinforced by the fact that
“predictions of future feelings inevitably involve some type of stylized represen-
tation of future events. Mental images of future vacations and holidays, for
example, typically do not include features such as rain, mosquitoes, and rude

28. Jeff T. Larsen, A. Peter McGraw, and John T. Cacioppo, Can People Feel Happy and
Sad at the Same Time?, 81 Journal of Personality and Social Psychology 684–96 (2001).
29. Julie A. Woodzicka and Marianne LaFrance, Real versus Imagined Reactions to
Sexual Harassment, 57 Journal of Social Issues, 15–30 (2001).
30. Timothy D. Wilson and Daniel T. Gilbert, Affective Forecasting, in Advances in
Experimental Social Psychology, Vol. 35 345–411 (Mark P. Zanna ed., San Diego, CA:
Academic Press, 2003).
31. Id. Much of the following discussion is drawn from Wilson and Gilbert’s excellent
review essay.
 complexities of decision making 399

Predicted

Experienced

Emotional
intensity

Time 1 Time 2 Time 3 Time 4

(Emotional event)
Time

figure 13.1 intensity and time course of predicted versus

experienced emotion.
Source: From Timothy Wilson and Daniel Gilbert, Affective Forecasting 35 Advances in Exterimental
Social Psychology (2003). Copyright © 2003 Elsevier Science (USA). Reprinted with permission.

service people . . .”32 People may overpredict the benefits of moving to a location
with nicer weather, not realizing that weather is just one of many factors affect-
ing their happiness—and that the weather is not always nice. When asked to
predict two months in advance how happy they would feel after a game if their
team won, football fans predicted a strong impact lasting a number of days.33 In
fact, although their happiness rose right after the Saturday game, it declined to
their normal baseline by the following Monday. In a parallel study that both
demonstrates the mechanics of focalism and suggests corrective action, partici-
pants made more realistic predictions about their happiness when they were
asked to write a “prospective diary” describing what particular activities they
would engage in on the Monday following the game.
Focalism also biases our affective forecasts when we focus our attention on
the immediate consequences of an event rather than on its effects in the longer
term. As Daniel Kahneman writes, we evaluate “an entire extended outcome by
evaluating the transition to it. For example, errors in predicting the well-being of
paraplegics may reflect people’s treating the tragic event of becoming a paraplegic
as a proxy in evaluating the long-term state of being a paraplegic.”34 And paraple-
gics spend much of their time in activities that are not compromised by their
disability.
Focalism is also encouraged by some decision-making processes, such as
elimination by aspects (See Section 4.3), that cancel out attributes that are shared
across alternatives and focus on those attributes that differ. Elizabeth Dunn and

32. Loewenstein and Schkade, supra.

33. Timothy D. Wilson, Thalia Wheatley, Jonathan M. Meyers, Daniel T. Gilbert, and
Danny Axsom, Focalism: A Source of Durability Bias in Affective Forecasting, 78 Journal of
Personality and Psychology 821 (2000).
34. Daniel Kahneman, Peter P. Wakker, and Rakesh Sarin, Back to Bentham?
Explorations of Experienced Utility, 112 Quarterly Journal of Economics 375 (1997).
400 problem solving, decision making, and professional judgment

her colleagues asked students to predict their happiness in living in various dor-
mitories, whose physical features differed considerably more than the social
aspects of dorm life.35 The students predicted their happiness or unhappiness
based solely on the former; yet, a year later, it turned out that their happiness was
not at all correlated with physical features, but was significantly correlated with
social life. When forecasting which of several options will make them happiest,
people tend to ignore the features that the options share (all dorms have active
social lives), while giving undue weight to features that make each option unique
(all dorms had different architecture).
Following Kahneman and Tversky,36 Dunn and colleagues called this
special case of focalism the isolation effect, because decision makers isolate the
unique features of each option. Focusing on the differences among competing
alternatives is not a poor decision strategy as such, as long as one takes account
of this phenomenon: “It would be quite reasonable to focus on location when
choosing between [college] houses that varied a great deal on this dimension
but not on others. When imagining a life in a specific house with a terrible
location, however, it would be unwise to expect lasting misery while neglecting
the happiness derived from living with one’s wonderful roommates.”
13.4.1.b Adaptation and “Ordinization” Neglect A second reason why the
impact bias occurs is that people tend not to consider the psychological pro-
cesses that keep our feelings and moods close to an emotional set-point. Processes
of hedonic adaptation blunt the effects of both good and bad events. (See Section
12.1.5) Affective reactions are generally most intense at the onset and diminish
over time. After experiencing setbacks, people usually get on with life, some-
times with reduced aspirations. Indeed, intensely disappointing experiences
tend to trigger proportionally stronger psychological coping mechanisms—such
as rationalizing and blaming others for one’s errors—than moderately disap-
pointing ones. Gilbert and Wilson call this the “psychological immune system.”
They term people’s tendency to underpredict these homeostatic tendencies
ordinization neglect.37
Predicting our emotional response to future events is no simple task. Yet,
given that we all accumulate years of experience forecasting and experiencing
emotional events, why do we continue to make forecasting errors without learn-
ing from our past mistakes? Gilbert and Wilson identify three reasons: (1) it is

35. Elizabeth W. Dunn, Timothy D. Wilson, and Daniel T. Gilbert, Location, Location,
Location: The Misprediction of Satisfaction in Housing Lotteries, 29 Personality and Social
Psychology Bulletin 1421 (2003).
36. Daniel Kahneman and Amos Tversky, Prospect Theory: An Analysis of Decision
Under Risk, 47 Econometrica 263–91 (1979).
37. Daniel T. Gilbert, Matthew D. Lieberman, Carey K. Morewedge, and Timothy D.
Wilson, The Peculiar Longevity of Things Not So Bad, 15 Psychological Science 14 (2004).
 complexities of decision making 401

difficult to remember emotional experiences accurately; (2) we commit hind-

sight bias,38 recalling our predictions as more accurate than they really were; and
(3) the processes of adaptation that make our experiences less emotionally potent
and enduring than we predict function outside of our awareness.

13.4.2 Projection Bias: Mispredicting Future Preferences

Just as we make systematic errors in predicting the intensity and duration of our
emotional responses, we err when predicting our future preferences for goods
and experiences.

Legal Note

A healthy client asks you to prepare an advance health directive, specifying that life-
sustaining measures be withheld or withdrawn if she has an illness that is so severe
that there is no reasonable prospect that she will recover and be able to live without
continuing life-sustaining measures. What concerns would you raise with her?
See Jeremy Blumenthal, Law and the Emotions: The Problems of Affective Forecasting,
80 IND. L.J. 155 (2005).

Projection bias refers to people’s tendency to “behave as if their future

preferences will be more like their current preferences than they actually will
be—as if they project their current preferences onto their future selves.”39 Our
predictions tend to be regressive—that is, they are anchored by our current
preferences. To a large extent this results from the difficulty of imagining
oneself having different preferences than one currently has. After all, one’s
individual identify consists in large part of preferences, and predicting a
significant change of one’s preferences is tantamount to predicting being a
different person.
• Most healthy people say they would not want heroic measures to be taken
if they were terminally ill. However, a majority of terminally ill patients
want treatment even if it would extend their lives by only a week. George
Loewenstein and David Schkade suggest that healthy people tend to under-
estimate the quality of life of sick people. (It is also possible that younger,
healthy people are making a normative judgment at time-1 about the

38. See Section 10.2.

39. Loewenstein and Angner, Predicting and Indulging Changing Preferences, in Time
and Decision.
402 problem solving, decision making, and professional judgment

allocation of resources to their older, sick future selves at time-2, which

they then resist at time-2.)
• When people are in a “cold” state, they tend to overestimate the strength of
their own willpower in a “hot” state. People shopping for food on an empty
stomach purchase more than they planned to. People who plan to refrain
from sex on a date, or who plan to use a contraceptive device, often do not
behave as they planned.
• People are systematically overconfident about the likelihood that they can
experiment with drugs or cigarettes without becoming addicted. People
purchasing credit cards tend, unrealistically, to believe that they will main-
tain a zero balance.
• With respect to the endowment effect (Section 14.1) people who do not
already possess an object or entitlement do not predict that their willing to
accept (WTA) to abandon the object is greater than their willingness to pay
(WTP) to acquire it.40

Students were given a choice among six snacks, one of which would be eaten in
each of three successive class sessions. Some were required to choose their snacks
for all three days on the first day of the study. Others could choose the snack on
the day it was to be eaten. Those making all three choices in advance chose a sub-
stantially more varied group of snacks than those making choices day by day.40

These are all examples of how people mispredict changes of their tastes, pref-
erences, or behaviors over time or under different circumstances. George
Loewenstein and Eric Angner describe a variety of processes in which prefer-
ences can change:
• Habit formation, where consuming a substance or having an experience
increases one’s preference for it.
• Satiation, where consuming a substance or having an experience decreases
its marginal utility. (Satiation is not necessarily a continuous function; you
may suddenly stop enjoying an experience or commodity one day.)
• Refinement, where people sometimes expose themselves to experiences in
art, literature, or music to improve their appreciation of them.

40. Itamar Simonson, The Effect of Purchase Quantity and Timing on Variety Seeking
Behavior, 27 Journal of Marketing Research 150–62 (May 1990).
 complexities of decision making 403

• Conditioning, where preference changes are induced by rewards or punish-

ments.
• Maturation, where one’s preferences change as one ages.
• Social influence, where one looks to others for cues about valid preferences
or just seeks their approval. (See Chapter 17.)
• Dissonance reduction, where preferences are realigned to fit circumstances
or behaviors.
Visceral states produce a special lack of imagination, which Loewenstein and
his colleagues have termed the hot/cold empathy gap. Strong positive or negative
emotions at the time of making a decision can bias one’s affective forecasting.
Consider someone with a bad cold who has just received an invitation to a party a
month hence, trying to imagine how she will feel then. In its strongest manifesta-
tion, people in a “hot” state—with respect to basic needs, such as hunger, thirst,
and sex—have difficulty imagining how they would feel in a “cold” state, and vice
versa. The effects of shopping without a list when hungry, or satiated, is a milder
example. The feelings need not be somatic: Loewenstein and Adler gave partici-
pants a trivia quiz and offered them, as compensation, a choice between a candy
bar and learning the right answers to the quiz. Most participants in a cold state—-
i.e., who made the choice before their interest had been piqued by answering the
questions chose the candy bar—predicted that they would choose a candy bar
afterwards. In fact, a majority of those who made the choice after taking the quiz—a
relatively hot state because of curiosity—chose instead to learn the answers.41
13.4.2.a Attitudes Toward Our Long-Run Preferences Do people sometimes
try to resist changes in long-run preferences? Loewenstein and Angner point to
concerns about one’s becoming corrupted—of one’s preferences changing in
ways that our present selves think undesirable.42 Daniel Boone is said to have
fled civilization to avoid its corrupting influence.43 For students reading this
book, the concern may be whether taking a high-paying job after graduation will
stamp out their idealism.

13.4.3 The Consequences of the Impact and Projection Biases and the
Possibility of Becoming More Accurate
The point of the preceding discussion is not that we usually mispredict our own
future feelings and preferences, but that we may do so in some situations that
have important consequences. These errors are not inevitably dysfunctional.

41. George Loewenstein and Daniel Adler, A Bias in the Prediction of Tastes, 105
Economic Journal 929–37 (July 1995).
42. Loewenstein and Angner, Predicting and Indulging Changing Preferences, in Time
and Decision.
43. John Mack Frargher, Daniel Boone: The Life and Legend of an American
Pioneer (Holt Paperbacks Press, 1993).
404 problem solving, decision making, and professional judgment

People may overestimate the duration of positive or negative events to motivate

themselves to action (“If I pass the exam tomorrow, my career will be vastly
enhanced; if I flunk it my life will be ruined forever”) or to protect themselves
against disappointment. But there are significant downsides as well. Short-term
fluctuations in visceral states coupled with poor affective forecasting may
lead to long-term commitments to marry, have children, and (very long-term)
to commit suicide.44 In making “advanced directives” or “living wills,” healthy
individuals may seriously mispredict their desires when they are in extremely
poor health.45
Loewenstein and Schkade are fairly skeptical about our general ability to
improve affective predictions. In addition to the reasons discussed above, the
memories of experiences from which one might learn may themselves be biased,
and learning from experience typically requires repeated observations of the
same error.46 But there may be ways to improve affective forecasting in particular
contexts—for example, improving informed consent to medical procedures by
focusing the patient’s attention on future scenarios. Loewenstein and Schkade
suggest that an “educational approach would seem to be least promising for
errors that result from hot/cold empathy gaps, whose very existence suggests a
resistance to cognitive interventions.” In such cases, governments or other third
parties may find it prudent to intervene to prevent any negative consequences of
forecasting error, such as by mandating a “cooling off” period before an indi-
vidual signs a contract.47

13.5 positive illusions: i’m above average, and my future looks rosy

We base important decisions not only on our forecasts of how various outcomes
would make us feel, but also on our judgments about how likely these outcomes
are to occur, as well our likelihood of contributing to bringing them about. For
example, an undergraduate might decide whether or not to apply to a competitive
law school based on her assessment of her academic skills, as well as her intu-
itions about her likelihood of acceptance. Yet these judgments tend to be

44. Loewenstein, O’Donoghue, and Rabin, Projection Bias in Predicting Future Utility,
Quarterly Journal of Economics 1209 (2003).
45. Id. The authors report on a study in which only 10 percent of healthy respondents
predict that they would accept a grueling course of chemotherapy to extend their lives by
three months, compared to 42 percent of current cancer patients.
46. George Loewenstein and David Schkade, Wouldn’t It Be Nice?: Predicting Future
Feelings, in Hedonic Psychology: Scientific Approaches to Enjoyment, Suffering,
and Well-Being (New York: Russell Sage Press, 1998).
47. See Jeremy A. Blumenthal, Emotional Paternalism, 35 Fla. St. U. L. Rev. 1 (2007).
 complexities of decision making 405

systematically biased in ways that create positive illusions about oneself.48 In particu-
lar, people tend to be overly optimistic about their futures, and overestimate their
skills in a variety of domains. We consider each of these observations in turn.

13.5.1 Overoptimism
In When Every Relationship Is Above Average: Perceptions and Expectations of
Divorce at the Time of Marriage,49 Lynn Baker and Robert Emery asked couples
applying for marriage licenses to estimate the percent of couples in the United
States who marry today who will get divorced at some time in their lives. The
average estimate was 50 percent—a fairly accurate prediction based on current
statistics. When asked about the likelihood that they would ever be divorced, vir-
tually no one thought so. Earlier, we mentioned a study that showed that entre-
preneurs who were realistic about the general failure rate for new ventures were
overconfident about the success of their own ventures. By the same token, “most
people expect that they will have a better-than-average chance of living long,
healthy lives; being successfully employed and happily married; and avoiding a
variety of unwanted experiences such as being robbed or assaulted, injured in an
automobile accident, or experiencing health problems.”50
In an article on what they term the planning fallacy, Roger Buehler, Dale
Griffin, and Michael Ross describe some mega-projects that have taken much
more time and money to complete than originally projected—such as the Denver
International Airport, Boston’s Central Artery/Tunnel project (the “Big Dig”)—
and many more quotidian ventures, such as term papers, honors theses, and
finishing shopping for Christmas presents. The authors wonder at the ability of
people “to hold two seemingly contradictory beliefs: Although aware that most of
their previous predictions were overly optimistic, they believe that their current
forecasts are realistic.”51 Does this describe the reader’s personal experience?
Although overoptimism often involves predictions of one’s own future, a
study of physicians’ predictions of the life expectancy of terminally ill patients
indicated a strong bias toward overoptimism, with only 20 percent being accu-
rate (within 33 percent of actual survival), 17 percent being overly pessimistic,

48. Shelley E. Taylor, Positive Illusions: Creative Self-Deception and the

Healthy Mind (New York: US Basic Books, 1989).
49. Lynn A. Baker and Reobert E. Emery, When Every Relationship Is Above Average.
Perceptions and Expectations of Divorce at the Time of Marriage, 17 Law and Human
Behavior 439 (1993).
50. David A. Armor and Shelley E. Taylor, When Predictions Fail: The Dilemma of
Unrealistic Optimism, in Thomas Gilovich, Dale Griffin, and Daniel Kahneman eds.,
Heuristics and Biases, The Psychology of Intuitive Judgment 334 (New York:
Cambridge University Press, 2002) [hereafter, Heuristics and Biases].
51. Roger Buehler, Dale Griffin, and Michael Ross, Inside the Planning Fallacy: The
Causes and Consequences of Optimistic Time Predictions, in Heuristics and Biases.
406 problem solving, decision making, and professional judgment

and 63 percent greatly overestimating survival.52 Interestingly, the better the

doctor knew the patient, the more likely his overoptimism.
People are not indiscriminately overoptimistic, however, and even when over-
optimistic, “correlations between predictions and outcomes are positive and
often substantial.”53
What conduces to overoptimism in planning? Buehler, Griffin, and Ross sug-
gest that when developing their own plans and predictions, people assume an
internal view, considering unique features of the task at hand and constructing
a scenario for how the project will unfold, in contradistinction to an external view
that places the task in a broader context. Focusing on a particular scenario for the
successful completion of the project diverts attention from the many ways in
which the future might unfold, including events—whether our own conduct or
external forces—that would compromise success. We don’t take adequate
account of our own past experiences, or personal base rates, because the sce-
nario necessarily focuses on the future, because we may perceive each new plan
as unique, and because we tend to “explain away” past failures. In addition (as
we saw in Section 8.2.3), people tend to underestimate the total probability of a
chain of independent conjunctive events.54 Overoptimism bears a strong resem-
blance to the anchoring bias—we anchor on the successful scenario and fail to
adjust adequately for alternative possibilities and (less vivid) base rates. Not sur-
prisingly, the planning fallacy diminishes or disappears when we forecast the
success of other people’s plans.
Overoptimism may also be fostered by temporal distortions: “Distant tasks (in
time) are analyzed at a higher level of abstraction than are tasks close at hand. . . .
In general, a more detailed and low-level construal of the task highlights difficul-
ties, and so predictions made at a more distant time should tend to be more
optimistic.”55
Overoptimism can be caused by the schematic and self-serving processing of
experiences. In an experiment similar to one described in Section 9.4.3, partici-
pants were assigned the roles of plaintiffs or defendants in advance of a negotia-
tion, then given the facts and asked to negotiate. When subsequently asked to
write down facts about the case that favored or disfavored their position, they
recalled significantly more facts favoring their own position than the other’s.56

52. Nicholas A. Christakis and Elizabeth B. Lamont, Extent and Determinant of Error in
Doctors’ Prognoses in Terminally Ill Patients: Prospective Cohort Study, 320 British Medical
Journal 469 (2000).
53. Armor and Taylor, supra.
54. See Section 8.2.3.
55. Armor and Taylor, supra.
56. George Loewenstein, Samuel Issacharoff, Colin Camerer, and Linda Babcock,
Self-Serving Assessments of Fairness and Pretrial Bargaining, 22 Journal of Legal Studies
135 (1993); Leigh Thompson and George Loewenstein, Egocentric Interpretations of Fairness
 complexities of decision making 407

Russell Korobkin invokes his own experience as a mediator to illustrate the

effects on negotiation when parties differentially focus their attention on their
strongest issues and underappreciate the issues on which their position is
weaker.57
The plaintiff alleged a variety of breaches of contract and tort claims. The
defendant’s attorney believed that the suit was duplicative of an earlier law-
suit brought by the same plaintiff in which the defendant had prevailed. She
believed that her client would prevail in a motion to dismiss on the ground of
res judicata58 and was even more confident in her ability to prevail at trial on
the merits even if she lost the motion. The plaintiff, for his part, was con-
vinced that the claims, while involving the same two parties, were completely
distinct, and that a judge would so conclude. Although I believed that the
defendant was likely to prevail on the substantive merits of the dispute, I
agreed with the plaintiff that his claims were not duplicative of earlier litiga-
tion and were, therefore, likely to survive the defendant’s motion to dismiss.
Because of the defendant’s confidence in and focus on the merits of his case,
the defendant, in my opinion, undervalued the expected transaction costs of
continuing with litigation. . . . Consequently, the defendant was unwilling to
offer more than a token amount to settle the dispute. On the other hand, the
plaintiff’s confidence in his ability to survive the motion . . . caused him to
focus too much attention on the motion and pay too little attention to the
proof problems he faced on the merits of the dispute. The result was an over-
all level of confidence in ultimate success that was . . . unjustified, and trans-
lated into a high reservation price for settlement.
We have various ways to avoid learning from errors caused by overoptimism.
We put a positive spin on bad outcomes, conveniently forget what we predicted,
explain away failure, or adapt to disappointment through the workings of the
psychological immune system.
Can overoptimism nonetheless be debiased? With respect to health risks,
informing people about risk factors, asking them to describe how the factors fit
their own profiles, and encouraging them to compare themselves to low-risk indi-
viduals who nonetheless succumbed to the risk had no effect—and were in some
instances counterproductive.59 Providing incentives for timely completion of
projects is not likely to work, since the incentives tend to focus people’s minds on

and Interpersonal Conflict, 51 Organizational Behavior and Human Decision Processes

176 (1992).
57. Russell Korobkin, Psychological Impediments to Mediation Success: Theory and
Practice, 21 Ohio St. J. on Disp. Resol. 281, 281 (2006).
58. For nonlawyers, this is roughly the civil analog to double jeopardy.
59. Neil D. Weinstein and William M. Klein, Resistance of Personal Risk Perceptions to
Debiasing Interventions, in Heuristics and Biases.
408 problem solving, decision making, and professional judgment

optimistic scenarios. Having people think about alternative scenarios is not suc-
cessful in countering overoptimism. The favored scenario provides too strong an
anchor. Although there is not much experimental evidence, Buehler, Griffin, and
Ross suggest that the most promising strategy with respect to personal projects
may be to draw people’s attentions to their past experiences in completing similar
projects, in effect tempering optimistic scenarios with actual personal base rates.

13.5.2 The Above-Average Effect

Closely related to overoptimism, people tend to believe that their own skills and
personality traits are above average in areas ranging from driving and athletics to
leadership and sensitivity. Most drivers believe that they are better and safer driv-
ers than average. Most college professors think they are better teachers than
average.60 David Dunning, Judith Meyerowitz, and Amy Holzberg did a series of
experiments that suggest that this tendency arises in situations in which the
skills or traits are not precisely defined. Given ambiguity in the characteristics,
people select those that place them in the most positive light.61 Rather than
having hopelessly distorted self-assessments, the participants may have pro-
vided rankings that were accurate given their own idiosyncratic definitions. Yet
there are examples of the phenomenon in relatively unambiguous circum-
stances. For example, in a survey of magistrate judges, 56 percent believed that
their rate of reversal on appeal was in the lowest quartile, and 88 percent believed
that at least half of their peers had a higher rate of reversal than they did.62
The above-average effect may have motivational causes, such as maintaining
people’s self-esteem, but it may also be a consequence of the availability heuristic,
in which ones own best traits are the ones most readily brought to mind.

13.5.3 Are Positive Illusions Always Detrimental?

“Each of us has his own way of deceiving himself. The important thing is to believe
in one’s own importance.
—Andre Gide”63
Psychologist Shelley Taylor argues that overoptimism, the above-average
effect, and the illusion of control (see Section 5.7.2) have important benefits as

60. The studies are summarized in Russell Korobkin, Psychological Impediments to

Mediation Success: Theory and Practice, 21 Ohio Journal on Dispute Resolution 281
(2006).
61. David Dunning, Judith A. Meyerowitz, and Amy D. Holzberg, Ambiguity and Self-
Evaluation: The Role of Idiosyncratic Trait Definitions in Self-Serving Assessments of Ability, in
Heuristics and Biases 324.
62. Chris Guthrie, Jeffrey J. Rachlinski, and Andrew J. Wistrich, Inside the Judicial
Mind, 86 Cornell Law Review 777 (2001).
63. Andre Gide, Journals of Andre Gide Vol. 1, 1889–1927 (J. O’Brien trans., 1987).
 complexities of decision making 409

well as downsides.64 She writes that overoptimism usually “is not a Panglossian
whitewash that paints all positive events as equally and commonly likely and all
negative events as equally and uncommonly unlikely. Rather, unrealistic opti-
mism shows a patterning that corresponds quite well to the objective likelihood
of events, to relevant personal experiences with events, and to the degree to
which one can actively contribute to bringing events about. Positive events are
simply regarded as somewhat more likely and negative events as somewhat less
likely to occur than is actually the case.” More broadly, Taylor argues that these
moderate illusions create self-confidence that motivates people to persevere in
tasks that they might not otherwise attempt:
[P]eople who hold positive illusions about themselves, the world, and the
future may be better able to develop the skills and organization necessary to
make their creative ideas and high level of motivation work effectively for
them. They seem more able to engage in constructive thinking. They can tie
their illusions concretely to the project at hand, developing a task-oriented
optimism and sense of control that enable them to accomplish more ambi-
tious goals. They are more likely to take certain risks that may enable them to
bring their ventures to fruition. And they seem better able to choose appropri-
ate tasks, gauge the effort involved, and make realistic estimates of their need
to persevere. Moreover, by developing the ability to postpone gratification,
they are able to commit themselves to a longer term task that may involve
sacrifices and delayed rewards along the way.”

Andy Grove (Former Intel CEO) on the Value of Self-Deception

“Well, part of [success] is self-discipline and part of it is deception. And the

deception becomes reality—deception in the sense that you pump yourself
up and put a better face on things than you start off feeling. After a while, if
you act confident, you become more confident. So the deception becomes
less of a deception. But I think it is very important for you to do two things:
act on your temporary conviction as if it was a real conviction; and when you
realize that you are wrong, correct course very quickly.”
Quoted in JEFFREY PFEFFER AND ROBERT I. SUTTON, HARD FACTS, DANGEROUS HALF-TRUTHS
AND TOTAL NONSENSE: PROFITING FROM EVIDENCE-BASED MANAGEMENT (Cambridge, MA:
Harvard Business Press, 2006) 201.

In sum, overoptimism is not inevitably harmful and may sometimes conduce

to improvements in performance and health, and to adaptation to life’s inevitable

64. Shelley E. Taylor, Positive Illusions: Creative Self-Deception and the Healthy
Mind (New York: Basic Books, 1989).
410 problem solving, decision making, and professional judgment

difficulties. Indeed, people sometimes counsel others to be overoptimistic rather

than pessimistic or even accurate about the future.65 All things considered,
though, you might prefer to keep your optimism within bounds; and to be real-
istic in assessing whether plans are unfolding as anticipated, so that you can
make corrections if they are not.
Notwithstanding the costs and benefits of overoptimism to the individual, the
phenomenon may be beneficial to the broader society. Given the likelihood of
success, how many people would start restaurants? Given an accurate appraisal
of the likelihood of being an acclaimed performer, how many musicians or actors
would undertake years of rigor and tedium of study and practice?

13.6 intertemporal choice

13.6.1 Discount Rates

People generally prefer to have a good thing today rather than at some future
time. Suppose that a firm offers you a $7500 signing bonus under alternative
plans. You can choose to receive the $7500 on your first day of work. Alternatively,
the firm will put the $7500 in an escrow account to be paid one year after you
start working for the firm, and this amount is absolutely guaranteed—even if
you are fired for cause before the year is up. Which plan would you prefer?
All things considered, you would prefer the $7500 today because having it
now gives you a choice that you wouldn’t otherwise have. Moreover, you could
put it in a savings account or a one-year certificate of deposit and almost surely
end up with more than $7500 a year from now.
Suppose the firm gives you a choice between a signing bonus of $7500 today
or $8000 a year from now? To decide between the alternatives, you must know
how much $8000 a year from now is worth to you today. This is the idea net pres-
ent value, or present discounted value. If you were planning to put the funds in a
one-year certificate of deposit at your bank, then your decision would depend on
how much interest the bank would pay you on your deposit. The graph in Figure
13.2 shows that you would be indifferent between $7500 today or $8000 a year
from now if the interest rate were 6.667 percent. At a lower interest rate you
would prefer to wait a year for the $8000 bonus; and the higher interest rate you
would take the $7500 today.
Net present value reflects how much a given amount of money in the future
is worth today. The formula is:
NPV=X/(1–r)n

65. David A. Armor, Cade Massey, and Aaron M. Sacket, Prescribed Optimism: Is It
Right to Be Wrong About the Future?, 19 Psychological Science 329–31 (2008).
 complexities of decision making 411

$8,300

$8,200
Future Value with 1 year return

$8,100

$8,000

$7,900

$7,800

$7,700

$7,600
Rate of return less than 6.667%: Rate of return more than 6.667%:
Choose to wait 1 year for $8000. Take $7500 now, invest for 1 year.
$7,500
0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10% 11% 12%
1 year risk-free rate of return

figure 13.2 do you prefer $7500 now or $8000 in one year?

where X is the amount of the future payment, r is the annual rate of return, and
n how many years in the future you will receive the payment. To say that a
discount rate is high means that you anticipate a high rate of return if you have
the funds today and are able to invest them.

13.6.2 Discounted Utility, Dynamically Inconsistent Preferences, and the

Pervasive Devaluation of the Future
To secure a maximum of benefit in life, all future events, all future pleasures and
pains, should act upon us with the same force as if they were present, allowance
being made for their uncertainty. The factor expressing the effect of remoteness
should, in short, always be unity, so that time should have no influence. But no
human mind is constituted in this perfect way; a future feeling is always less influ-
ential than a present one.
—W. S. Jevons, The Theory of Political Economy (1871)
People distribute their resources between the present, the near future and the remote
future on the basis of a wholly irrational preference. When they have a choice
between two satisfactions, they will not necessarily chose the larger of the two but
will often devote themselves to producing or obtaining a small one now in preference
to a much larger one some years hence.
—A. C. Pigou, The Economics of Welfare (1920)

Hard work often pays off after time, but laziness always pays off now.
—Author unknown
412 problem solving, decision making, and professional judgment

If choosing the alternative with the greater net present value usually seems
the sensible strategy, there are obvious exceptions. For example, suppose that
the interest rate is 6.7 percent, and
• the choice is between: $7400 today and $8000 a year from now. Even though
the $8000 is a great deal, taking the smaller amount now would allow you to
purchase a $7400 painting that you covet, which will not be available in a year;
or
• the choice is between $8000 today and $8000 a year from now. You are a
compulsive gambler and know you’ll blow any money you now have in Las
Vegas, but you’re about to enroll in a 12-step program.
These examples suggest that intertemporal preferences cannot be subject to
global criteria of rationality (like those of the axioms of subjective expected utility
theory), but rather depend on an individuals’ own tastes and plans. That said,
when people deviate significantly from the standard model or act as if the dis-
count rate were extraordinarily high, it raises a warning flag that they may not be
acting in their own best interests.
The standard model of discounted utility, first proposed by Paul Samuelson
in 1937, assumes that people discount at a constant rate over time, with the dif-
ference in utility between receiving $100 today and $200 a year from now being
proportional to the difference in utility between receiving $100 in ten years and
$200 eleven years from now. As shown in the solid line in Figure 13.3, this is an
exponential function.
1.2

1 Exponential
Value of Util Delayed by x Years

Hyperbolic

0.8

0.6

0.4

0.2

0
0 5 10 15 20 25 30 35 40 45 50
Delay (Years)

figure 13.3 present bias .

From George-Marios Angeletos, David Laibson, Andrea Repetto, Jeremy Tobacman, and
Stephen Weinberg. Figure 18.1 “Discount Functions.” In Time and Decision. © 2003 Russell
Sage Foundation, 112 East 64th Street, New York, NY 10021. Reprinted with permission.
 complexities of decision making 413

In fact, people often manifest a strong present bias, or excessive impatience.

They accord near-term gratification considerably greater value than deferred
gratification, and do not distinguish much among deferred gratifications with
different maturity dates—that is, between gains that are a few years away and
those that lie many years in the future. Rather than being time-consistent, our
preferences are dynamically inconsistent over time, and seem to follow a pattern
of hyperbolic rather than exponential discounting, as shown by the dashed line
in Figure 13.3.66 (fMRI studies indicate that immediate and future rewards acti-
vate different regions of the brain.67)
• In a dramatic example of the pervasive devaluation of the future, Howard
Kunreuther, Ayse Onculer, and Paul Slovic68 asked people to indicate the
maximum they would be willing to pay for measures that would protect
them from burglary (a dead-bolt lock) or earthquake damage (bracing the
foundation of one’s house) over time periods ranging between one and ten
years. Most respondents were willing to pay the same amounts for short-
term and long-term protection.
• A study of actual air conditioner purchases showed that in trading off
higher operating costs for a lower purchase price, consumers acted as
though the annual rate of return was an incredible 89 percent.69
• In another study, participants were asked to choose between receiving a
prize of a certified check for $1000 that could be cashed next week and a
$2000 certified check that could be cashed only after some delay, and were
asked to name the length of the delay at which they would be indifferent
between the two checks. On average, people would wait less than forty days
for the larger check, implying a staggering rate of return.70
• The British sociologist, Anthony Giddens, observes that changes in indi-
vidual behavior necessary to reduce obesity, high blood pressure, diabetes,

66. Some recent experimental evidence suggests that this may not be due to the differ-
ence between the present and future as such, but to the fact that people tend to have a
greater discount rate for periods of shorter duration whenever they occur. See Daniel
Read, Subadditive Intertemporal Choice, in Time and Decision 301.
67. Samuel M. McClure, David I. Laibson, George Loewenstein, and Jonathan D.
Cohen, Separate Neural Systems Value Immediate and Delayed Monetary Rewards, 306
Science 503–07 (Oct 15, 2004).
68. Howard Kunreuther, Ayse Onculer, and Paul Slovic, Time Insensitivity for Protective
Investments, 16 Journal of Risk and Uncertainty 279 (1998).
69. George Ainslee and Nick Haslam, Hyperbolic Discounting, in Choice Over Time
61 (George Loewenstein and Jon Elster eds., New York: Russel Sage Foundation, 1992).
70. George Ainslie and Vardim Haendel, The Motives of Will, in E. Gottheil, K.
Druley, T. Skolda and H. Waxman (eds.), Etiologic Aspects of Alcohol and Drug
Abuse 61 (Springfield, IL: Charles C. Thomas, 1983).
414 problem solving, decision making, and professional judgment

heart disease and cancer, excessive drinking, and drug dependence are
impeded by hyperbolic discounting.71
• Giddens also argues that “hyperbolic discounting is one of the main factors
explaining the lazy attitude most people have towards the threats posed by
global warming. Surveys show that the majority now accept both that cli-
mate change is real and dangerous and that it is created by our own behav-
iour. However, the proportion that has made any significant behaviour
change is very low. The implications are disturbing. Consciousness-raising
and green taxes, even if carefully thought out and organised, may have only
marginal effects—and they might be widely resisted even then.”72
In short, we “tend to grab immediate rewards and avoid immediate costs that
our ‘long-run selves’ do not appreciate.”73 Procrastination, in which postponing
inevitable costs often leads to increased costs (the paper poorly done at the last
minute, or the brief filed a day after the deadline) is a common example.
Addictions are another example. George Loewenstein notes that “visceral fac-
tors, such as the cravings associated with drug addiction, drive states (e.g.,
hunger, thirst, and sexual desire), moods and emotions, and physical pain . . .
[can] cause people to behave contrary to their own long-term self-interest, often
with full awareness that they are doing so.”74
There is evidence that children’s ability to defer gratification predicts
their success later in life. Walter Mischel conducted an experiment in which
four-year-olds had a marshmallow placed in front of them, and were given
a choice between eating it immediately or waiting a few minutes and getting

Personal Note

You are a very busy associate at a law firm. The firm’s pro bono coordinator
asks if you could commit to doing some work on a project that will require
about five hours a week for a number of weeks. You reply that this isn’t a
good time, but volunteer to do the work in several months—even though
you’ll be at least as busy then as you are now.

71. Anthony Giddens, This Time It’s Personal, The Guardian, Jan. 2, 2008, http://
www.guardian.co.uk/commentisfree/2008/jan/02/thistimeitspersonal.
72. Id.
73. Ted O’Donoghue and Matthew Rabin, Doing It Now or Later, 89 American
Economic Review 103 (1999).
74. George Loewenstein, Out of Control: Visceral Influences on Behavior, 65
Organizational Behavior and Human Decision Processes 272 (1996).
 complexities of decision making 415

a second marshmallow. In a follow-up study fourteen years later, those who

deferred on the whole were more highly rated in terms of coping skills, self-
esteem, and other social skills, and scored significantly higher on their SATs.75
Ainslie and Haslam note that “short-term interests are not aberrant
phenomena. . . . [T]hey cannot be ‘cured’ be any means, or controlled just by
insight into what one’s longest range interests are,”76 but rather require the
intentional adoption of behavioral strategies along the following lines.77
13.6.2.a Repression We can avoid thoughts and behavior that might lead to
impulses. In Walter Mischel’s experiments, the children who managed not to eat

“The belief that a person should weight all utility the same, regardless of its tem-
poral position, implicitly assumes that all parts of one’s future are equally parts
of oneself; that there is a single, enduring, irreducible entity to whom all
future utility can be ascribed. However, some philosophers—most notably,
Derek Parfitt—deny this assumption. They argue that a person is nothing
more than succession of overlapping selves related to varying degrees by
physical continuities, memories, and similarities of character and interests.
On this view, the separation between selves may be just as significant as
the separation between persons, and discounting one’s ‘own’ future utility
may be no more irrational than discounting the utility of someone else.”
—Shane Frederick, Time Preference and Personal Identity, in TIME AND DECISION 89

the marshmallows used various strategies of this sort. An observer of video foot-
age reports: “Some cover their eyes with their hands or turn around so that they
can’t see the tray. Others start kicking the desk, or tug on their pigtails, or stroke
the marshmallow as if it were a tiny stuffed animal.78
13.6.2.b Self-Control, Habits, and Routines We can develop personal rules,
such as “no desserts” or “twenty minutes on the treadmill every morning.” Self-
control strategies often rely on creating a feeling of fear or guilt when faced with
the temptation to be avoided. In George Loewenstein’s and Ted O’Donoghue’s
words they “immediatize the delayed costs in the form of immediate [negative]
emotions.” However, using dieting as an example, Loewenstein and O’Donoghue

75. Yuichi Shoda, Walter Mischel, and Philip K. Peake, Predicting Adolescent Cognitive
and Self-Regulatory Competencies from Preschool Delay of Gratification: Identifying Diagnostic
Conditions, 26 Developmental Psychology 978–86 (1990).
76. George Ainslee and Nick Halsam, Self Control, in Choice Over Time, supra.
77. We have given different names to some of the strategies.
78. Jonah Lehrer, Don’t!: The Secret of Self Control, The New Yorker, May 18, 2009.
416 problem solving, decision making, and professional judgment

observe that the immediate costs are only justifiable in terms of the long-term
benefits, which (realistically) are often not achieved.79
Perhaps a more promising strategy than pure self-control is to develop habits
or routines that guard against self-defeating behavior. Although habits must be
learned over time through repetition, a program of research by Peter Gollwitzer
suggests that people can often obtain the benefits of good habits immediately by
engaging in a very simple cognitive exercise.80 In a series of studies, Gollwitzer
and his collaborators asked people with specific goals (such as writing a school
paper over their holiday break or performing a breast self-examination in the
subsequent month) to come up with a specific plan for the implementation of
those goals. Such “implementation intentions” involved specifying when, where
and how participants intended to pursue their goals. By creating “instant habits”
and making goal pursuit automatic when the right circumstances present them-
selves implementation intentions increase the likelihood that people will accom-
plish their goals. For example, if your implementation intention is to “order
fruit” for dessert at a restaurant, you are more likely to automatically order fruit
when the waiter hands you the dessert menu.
The motivations to exercise self-control are complex. As Drazen Prelec and
Rodnit Bodner note, “the gap between a single act of dietary self-sacrifice and its
benefits—slimness, leading to beauty, health, and virtue—is not only temporal
but also causal. A single low-calorie salad does not cause slimness. . . . Rather, the
benefits or costs are properties of a long-run policy or lifestyle, and an integral
policy or lifestyle cannot be obtained with a single decision. . . .”81 The authors
posit that postponing gratification produces a diagnostic reward—the pleasure
of learning something about yourself, in this case, about your willpower.
These inferences about a person’s own character and prospects are “a source of
immediate pleasure and pain, which contributes to his/her ability to exercise
self-control.”82 More broadly, we tend to see our lives in terms of narratives,
where deferring gratification today leads to better outcomes over time.
13.6.2.c Extrapsychic Mechanisms We can make decisions now that substi-
tute for, or significantly aid, the exercise self-control in the future. The classic

79. George Loewenstein and Ted O’Donoghue, We Can Do This the Easy Way or the
Hard Way: Negative Emotions, Self-Regulation, and the Law, 73 Chicago Law Review 183
(2006).
80. Peter M. Gollwitzer, Implementation Intentions: Strong Effects of Simple Plans, 54
American Psychologist 493–503 (1999).
81. Drazen Prelec and Ronit Bodner, Self-Signaling and Self-Control, in Time and
Decision 277.
82. Baumeister and Kathleen D. Voss analogize the exercise of willpower to physical
endurance, which is subject to exhaustion. Emotional distress undermines self-regulation
by changing people’s priorities to focus on feeling good in the immediate present. Roy
Baumeister and Kathleen D. Voss, Willpower, Choice, and Self-Control, in Time and
Decision 201. In any event, the resources that contribute to self-control are limited.
 complexities of decision making 417

example is Ulysses having his sailors tie him to the mast so he could not respond
to the Sirens’ temptations. Keeping the refrigerator free of caloric snacks, or
leaving one’s credit card at home, are more mundane examples. Thomas Shelling
describes a drug addiction clinic in which patients subject themselves to self-
extortion: Patients write a self-incriminating letter confessing their drug addic-
tion, which will be mailed to an appropriate person (e.g., a physician’s letter
is addressed to the state board of medical examiners) if he or she is caught
taking drugs.83 The Web site, http://www.stickk.com/ invites people to make
“Commitment Contracts” for health and other personal goals, contracting to pay
charities (including causes they abhor) if they don’t meet milestones in achiev-
ing their goals.
Studies by Dan Ariely and Klaus Wertenbroch show how students use extra-
psychic mechanisms to counter the temptation to procrastinate. Students in
three sections of a marketing class at MIT were assigned three short papers to be
written during the semester. In one section, students were told that they could
hand in the papers anytime they wanted as long as all three papers were in by the
last day of class. In the second section, the professor imposed fixed, evenly-
spaced deadlines such that students had to complete one paper by the end of
each third of the course, and would be penalized for handing a paper in late. In
the third section, students were given the opportunity to set their own deadlines
for the papers. They could choose any deadlines they wanted as long as all three
papers were in by the end of the quarter, but once they had chosen their dead-
lines, they had to stick to them or they, too, would be penalized.
When given the opportunity to set their own deadlines, most students did not
choose the schedule that would have given them the most flexibility: setting
the deadlines for all three papers as late as possible. From a purely economic
perspective, this would have been the schedule with the greatest utility, since
you can always complete the papers earlier than the deadline. So why give up
flexibility, and risk being penalized for a late paper? The students were self-aware
enough to realize that if they weren’t forced to complete at least one of the papers
early in the semester, they were likely to procrastinate and be stuck trying to
write all three of them at the last minute.
It turned out that students who chose their own deadlines received
higher grades, on average, than students who were given no deadlines at all.
However, students whose professors imposed evenly-spaced deadlines did
even better.

83. T. C. Shelling, Self-Command: A New Discipline, in Choice Over Time.

418 problem solving, decision making, and professional judgment

Discounting and Future Generations

Forestalling climate change is one of a number situations where the benefits from
costs paid today may redound to the benefit of future generations. If people often
discount in planning for their own lives, should they do so for the lives of future
generations? Noting that you would almost certainly prefer $1000 now to $1000
in twenty years, Cass Sunstein asks: “Is a life twenty years hence worth a fraction
of a life today?”84
The Office of Management and Budget (OMB) suggests that policy makers
“consider generational discounting in the same way as individual discounting,
but with an explicit discussion of the intergenerational concerns.”85 Alternatively,
OMB suggests “discounting for future generations at a slightly lower rate.”86 But
some scholars argue that generational discounting is radically different from indi-
vidual discounting, and that it is morally objectionable because it does not give
equal consideration to those living in the future.87 They argue for a zero rate of
discounting for intergenerational decisions.88 While we flag this complex and con-
troversial issue, it is beyond the scope of the book.

8485868788

84. Cass R. Sunstein, Worst-Case Scenarios 11 (Cambridge, Massachusetts,

Harvard University Press, 2007).
85. OMB, Circular A-4, “Regulatory Analysis: Memorandum to the Heads of Executive
Agencies and Establishments” (Sept. 17, 2003).
86. Id.
87. Richard L. Revesz and Michael A. Livermore, Retaking Rationality: How
Cost-Benefit Analysis Can Better Protect The Environment and Our Health 111
(New York: Oxford University Press, 2008).
88. Tyler Cowen and Derek Parfit, Against the Social Discount Rate, in Justice Across
the Generations: Philosophy, Politics, and Society 144–61 (James S. Fishkin and
Peter Laslett eds., New Haven, CT: Yale University Press, 1992). Robert M. Solow, The
Economics of Resources or the Resources of Economics, 64 American Economic Review 1–14
(May 1974).
14. complexities of decision making
continued
The Power of Frames

In Chapter 2 we considered how the metaphors that we use to frame problems

can affect our solutions to them. Here we consider frames of a quite different
sort: how decisions can be affected by ways that choices are presented to us—for
example, as gains or losses, or in comparison with various alternatives.

14.1 prospect theory, the endowment effect, and status quo bias1

Neoclassical economics and its cousin, expected utility theory, posit that rational
choices maximize one’s overall wealth or, more generally, one’s overall utility.
Subject to Bernoulli’s observation (in Section 12.1.4) that people often display a
diminishing marginal utility of additional wealth, neoclassical economics, and
expected utility theory posit that a penny gained has as much positive value as a
penny lost has negative value. In fact, however, people systematically violate this
norm when evaluating changes from the status quo. Amos Tversky and Daniel
Kahneman formalized this aspect of the psychology of decision making in pros-
pect theory, which is illustrated by the S-shaped curve in Figure 14.1.
In this chapter, we discuss prospect theory in circumstances where risk or
uncertainty is not a significant factor. The value function described by prospect
theory has three important characteristics.
1. Individuals do not think of the outcomes of decisions in terms of their
overall wealth (as neoclassical economics holds). Rather, they consider out-
comes in terms of gains and losses from some reference point.
2. Individuals are loss averse. The slope of the of the curve for losses is steeper
than the slope for gains. This means that people experience losses from the
reference point as greater than equivalent gains. Loss aversion is captured
by the phrase that “losses loom larger than gains.”

1. This section draws heavily on Daniel Kahneman, Jack L. Knetsch, and Richard H.
Thaler, The Endowment Effect, Loss Aversion, and Status Quo Bias, in Choices, Values,
And Frames (Daniel Kahneman And Amos Tversky, eds., Cambridge, UK: Cambridge
University Press, 2000) 159. (2000), and Russell Korobkin, The Endowment Effect and Legal
Analysis, 97 Northwester University Law Review 1227 (2003).
420 problem solving, decision making, and professional judgment

Value

Losses Gains

figure 14.1 the prospect theory value function.

3. Both the gain and loss functions display diminishing sensitivity. They are not
straight lines. Rather, the gain function is concave downward and the loss
function is concave upward.
Two behavioral economists’ explanation of the so-called equity premium puzzle
provides a nice example of loss aversion. The puzzle is why people are overin-
vested in low-yield treasury bills compared to equities. A dollar in stocks in 1926
would have been worth over $1800 at the close of the twentieth century, while a
dollar invested in treasury bills would have been worth only about $15. Ordinary
degrees of risk aversion do not account for so large a disparity. Shlomo Benartzi
and Richard Thaler explain the puzzle in terms of what they call myopic loss aver-
sion: Many investors examine their portfolios’ performance on a daily, if not
more frequent, basis, and tend to reset their reference point each time they look.
If losses loom larger than gains, the pain of a dollar’s loss is greater than the
pleasure of a dollar’s gain. Thus the excessive risk aversion caused by frequent
evaluation of their portfolios prevents investors “from adopting a strategy that
would be preferred over an appropriately long time horizon.”2
As this example suggests, there is no objectively correct reference point. An
individual’s conscious or unconscious choice of a reference point in a particular
situation is subjective, context-sensitive, and labile. Two people—for example,
two parties negotiating a contract—may have different references points for
exactly the same situation or transaction. Whatever reference point one chooses,
though, a potential loss is more unattractive than an equivalent potential gain is
attractive.

2. Shlomo Benartzi and Richard H. Thaler, “Myopic Loss Aversion and the Equity
Premium Puzzle,” NBER Working Paper No. W4369. May 1993, available at SSRN: http://
ssrn.com/abstract=227015.
 complexities of decision making: frames 421

The example also illustrates that people often choose the status quo as a refer-
ence point from which to evaluate change. As a consequence of loss aversion, all
things considered, people like things the way they are more than they like change.
This general phenomenon is called the status quo bias. We begin by describing
an important subset of status quo bias, the endowment effect, which captures the
fact that people often demand more to give up a good or entitlement than they
would be willing to pay to acquire it. The endowment effect arises in many con-
texts and often confounds policy making.
For example, contingent valuation (discussed in Section 14.6) is a common
method, used in cost-benefit analysis and the assessment of damages, for plac-
ing a value on environmental goods or harms: affected individuals are surveyed
to determine how much they would be willing to pay (WTP) to gain an environ-
mental good or how much they would be willing to accept (WTA) to suffer the
loss of that good. Contingent valuation is premised on the assumption that an
individual’s WTP and WTA for a good are identical. But it turns out that WTP
and WTA often differ by large amounts. Russell Korobkin reports:
In one notable study, 2000 duck hunters were surveyed about the value they
would place on protecting a wetland from development. Hunters were willing
to pay $247 per person per season, on average, for the right to prevent devel-
opment to make hunting viable, while they would demand, on average, $1044
dollars each to give up an entitlement to hunt there. . . . [Another] survey
found that a sample of residents of a region of the southwest would pay $4.75
per month, on average, to maintain 75 miles of air visibility in the face of
potential pollution that would reduce visibility to 50 miles. In contrast, how-
ever, people drawn from the same pool and told that they collectively enjoyed
the right to prohibit the pollution, reported that they would demand $24.47
per month before being willing to permit the pollution.3
In a classic demonstration of the endowment effect,4 half the students in a
classroom—the sellers—were handed decorated university mugs (costing $5 at
the university bookstore) and given a questionnaire that stated: “You now own
the object in your possession. You have the option of selling it if a price, which
will be determined later, is acceptable to you. For each of the possible prices
below, indicate whether you wish to sell your object and receive this price; or
keep your object and take it home with you.” The participants indicated their
decision for prices ranging from $0.50 to $9.50 in increments of 50 cents.
Students who had not received a mug—the buyers—received a similar question-
naire, informing them that they would have the option of receiving either a mug

3. Russell Korobkin, supra.

4. Daniel Kahneman, Jack L. Knetsch, and Richard H. Thaler, Experimental Tests of the
Endowment Effect and the Coase Theorem, 98 Journal of Political Economy 1325–48
(December 1990).
422 problem solving, decision making, and professional judgment

or a sum of money to be determined later. They indicated their preferences

between a mug and sums of money ranging from $0.50 to $9.50.
The median price that buyers were willing to pay (WTP) for the mug was
$3.31, while the sellers’ median willingness to accept (WTA) was $7.06. Amos
Tversky and Daniel Kahneman explain that the buyers and the sellers face pre-
cisely the same decision problem, but from different reference states. The buyers
face the positive choice of receiving either the mug or cash. The sellers must
choose between keeping the mug (retaining the status quo) or giving it up in
exchange for money. “Thus, the mug is evaluated as a gain by the [buyers], and
as a loss by the sellers. Loss aversion entails that the rate of exchange of the mug
for money will be different in the two cases.”5
Why doesn’t the buyer feel a similar loss aversion with respect to the money
given up? Although the endowment effect is not limited to unique items, it is
attenuated for ordinary market goods that have close substitutes, and it disap-
pears for items, such as money, that have only an exchange value.6 Moreover:
Selling an item, even a minor one, may be thought to involve a choice in a
deeper or more vivid sense than buying; after all, selling is a far rarer activity
for the average individual, so that the choice to sell or not may seem rather
weighty. In contrast, the decision whether to buy or not is made by individu-
als nearly constantly, in a wide variety of highly impersonal settings. Another
way of seeing this point is that the act of selling seems more permanent than
that of not-buying, so that one is more reluctant to sell than not to buy.7
The endowment effect has been replicated in many studies. In one experi-
ment, participants were given either a lottery ticket that costs $2 or $2 cash, and
then offered the opportunity to trade the lottery ticket for cash or vice versa.

5. Amos Tversky and Daniel Kahneman, Loss Aversion in Riskless Choice: A Reference-
Dependent Model, 106 Quarterly Journal of Economics 1039 (1991). In another experi-
ment, half the participants were given pens, and half were given a token redeemable for
an unspecified gift. Then, all of the participants were given a choice between the pen and
two chocolate bars. Fifty-six percent of those who already had the pen preferred the pen,
compared to 24 percent of the others. In a separate part of the experiment, when asked to
rank the attractiveness of a number of items, including the pens and chocolate bars, the
participants rated the chocolate bars as at least as attractive. “This suggests that the main
effect of endowment is not to enhance the appeal of the good one owns, only the pain of
giving it up.” Kahneman, Knetsch, and Thaler, supra. In yet another experiment involving
mugs and chocolate, participants were randomly given either a coffee mug or a large
Swiss chocolate bar and offered the opportunity to trade. Only 11 percent and 10 percent
of the participants, respectively, were willing to trade.
6. Jennifer Arlen, Matthew Spitzer, and Eric Talley, Endowment Effects within Corporate
Agency Relationships, 31 Journal of Legal Studies 1 (2002).
7. Edward McCaffrey, Daniel Kahneman, and Matthew Spitzer, Framing the Jury:
Cognitive Perspectives on Pain and Suffering Awards, 81 Virginia Law Review 1341 (1995).
 complexities of decision making: frames 423

Very few switch.8 As a real-world analogue, consider the large number of indi-
vidual investors who will hold on to shares of stock when they would not pur-
chase the same stock at the same or even a lower market price. Consider whether
the various legal doctrines that reflect the saying that “possession is nine points
of the law,” or doctrines that give greater due process rights to someone already
holding a government job or benefit than one seeking it, may reflect the endow-
ment effect.
People may gain “psychological ownership” merely by having an item in their
physical possession. In a recent variation on the mug experiment, students were
asked to physically examine the mugs for different periods of time before bid-
ding on them in an auction. Those who held the mugs for longer period of time
bid more.9
As we mentioned, the endowment effect is a subset of the status quo bias—the
general phenomenon whereby people attach a value to the present state of the
world compared to alternative states, even when no property right or legal entitle-
ment is involved. For example, as of the 1990s, the standard auto insurance policy
in New Jersey did not provide a right to recover for pain and suffering for minor
injuries, though one could obtain that right by purchasing a higher priced policy.
In Pennsylvania, the standard policy did provide the right to recover, but one
could purchase a cheaper policy that relinquished the right. In a survey of resi-
dents of the two states, only 20 percent of New Jersey respondents and 25 percent
of Pennsylvanians indicated they would switch from their state’s default policy.10
Also, as we noted above, what one treats as a reference point is essentially a
subjective matter. Russell Korobkin conducted an experiment to explore whether
the parties negotiating a contract would treat a default contract term as the status
quo.11 Students played the role of lawyer negotiating a shipping contract on
behalf on an overnight delivery company. In one scenario, they were told that
there were two possible terms that could be used to deal with consequential dam-
ages for loss or delay of packages: (A) the delivery company would be liable only
for reasonably foreseeable damages, or (B) the delivery company would be liable
for all consequential damages, whether or not foreseeable. In another scenario
they were told that there were two possible terms for dealing with unforeseen
circumstances that make delivery commercially impracticable: (A) the delivery
company would be excused from its obligation to deliver, or (B) the delivery

8. Tversky and Kahneman, Loss Aversion in Riskless Choice, supra.

9. James R. Wolf, Hal R. Arkes, and Waleed A. Muhanna, The Power of Touch: An
Examination of the Effect of Duration of Physical Contact on the Valuation of Objects,
Judgment and Decision Making 476–82 (August 2008).
10. Eric J. Johnson, John Hershey, Jacqueline Meszaros, Hershey, and Howard
Kunreuther, Framing, Probability Distortions, and Insurance Decisions, 7 Journal of Risk
and Uncertainty 35 (1993).
11. Russell Korobkin, The Status Quo Bias and Contract Default Rules, 83 Cornell Law
Review 608 (1998).
424 problem solving, decision making, and professional judgment

company would be liable for damages. Participants were then given estimates of
the differential costs to their client (on a per-package basis) of performing its
obligations if the contract contained term B rather than A, and were then asked
to value term A relative to term B.
One-half of the participants were told that term A was the default rule and
were asked to provide their WTA price (how much their clients would need to be
paid per package) to agree to insert term B into the agreement to contract around
the default rule. The other half of the participants were told that term B was the
default rule and were thus asked to provide their WTP for inserting term A into
the contract in order to contract around the default rule. In both scenarios, WTA
significantly exceeding WTP; in other words, the strength of the participants’
preferences for the term that benefited their client was biased in favor of which-
ever term was identified as the default.
In Korobkin’s experiment, the contracting parties were induced to treat the
legal default rule as the status quo. However, Marcel Kahan and Michael Klausner
hypothesized that contracting parties might treat the terms embodied in a stan-
dard form contract as the reference point. “Although these parties have no formal
property (or other) rights in a standard term, the standard sets terms for an
expectational baseline . . .”12 Korobkin subsequently tested this hypothesis by
redesigning his experiments to inform the participants negotiating a contract
term about both the legal default rule and the term in the standard form con-
tract. They were told that the form term was the opposite of the legal default rule,
so that if they contracted around the form term, they would reestablish the
default term that the standard form circumvented. The participants’ preferences
turned out to be biased in favor of the form term rather than the default rule.
Korobkin concludes:
These results suggest that contract terms will be biased in favor of the status
quo, but that the reference point that identifies the status quo for contracting
parties is extremely dependent on context. The implication for legal policy is
that the content of contract law probably substantively affects the content of
contracts, but not all the time. The effect of contract law can be swamped by
other reference points, depending on the circumstances. The prescriptive
implications for contract lawyers seems more clear: “negotiators who are able
to define the status quo position, against which all proposed terms are judged,
are likely to enjoy an important bargaining advantage.”13

12. Marcel Kahan and Michael Klausner, Path Dependence in Corporate Contracting:
Increasing Returns, Herd Behavior and Cognitive Biases, 74 Washington University Law
Quarterly 347, 361 (1996).
13. Russell B. Korobkin, The Status Quo Bias and Contract Default Rules, 83 Cornell
Law Review (1998).
 complexities of decision making: frames 425

In a torts suit, the instructions to a jury assessing damages could be framed

from either of two perspectives. The jury might be asked
1. to imagine being the plaintiff before the injury and ask how much you
would demand to suffer the injury, or
2. to imagine being the plaintiff after the injury and ask how much is needed
to “make you whole.”
The former perspective frames the issue as a loss—from the plaintiff’s being
healthy to being injured—while the latter frames it as a gain, and prospect theory
predicts that the loss will be perceived as greater than an equivalent gain. A
series of experiments by Edward McCaffrey, Daniel Kahneman, and Matthew
Spitzer demonstrated just this result with respect to awards for pain and suffer-
ing: participants playing the role of jurors awarded twice as much in the
loss frame than in the gain frame.14 The authors suggest that, in addition to pro-
viding a “healthy” reference point, the ex ante perspective captures the value of
the plaintiff’s exercising the free choice of whether or not to incur the injury.
(The authors argue that the make-whole instruction is normatively preferable to
the ex ante instruction: because of the phenomenon of impact bias (Section
13.4.1), someone anticipating the effect of injury is likely to underestimate the
plaintiff’s resiliency and adaptability and therefore overestimate the disutility of
the injury.)
In discussing the concept of external validity in Section 7.6.2, we described
some skepticism about whether the endowment effect actually exists. The lead-
ing critique is by Charles R. Plott and Kathryn Zeiler,15 who assert that gaps
between WTP and WTA may be explained by participants’ misconceptions about
the parameters in the experimental auctions. They suggest that the WTP-WTA
gap may be explained by factors including the role of anonymity in the experi-
ment for both the buyers and the sellers, the elicitation mechanisms used, and
whether or not the auction participants have a thorough understanding of the
auction’s rules. Plott and Zeiler conducted a series of experimental auctions
designed specifically to control these factors and concluded that there was no
evidence of an endowment effect. While their criticism is far from conclusive,
their results suggest that the precise conditions under which a WTP-WTA gap
does and does not occur must be sorted out to more fully understand the nature
and scope of the endowment effect.
***

14. Edward McCaffrey, Daniel Kahneman, and Matthew Spitzer, Framing the Jury:
Cognitive Perspectives on Pain and Suffering Awards, 81 Virginia Law Review 1341 (1995).
15. Charles R. Plott and Kathryn Zeiler, The Willingness to Pay/Willingness to Accept
Gap, The “Endowment Effect,” Subject Misconceptions and Experimental Procedures for
Eliciting Valuations, 95 American Economic Review 530–45 (2004).
426 problem solving, decision making, and professional judgment

At the beginning of this section, we attributed the endowment effect and

status quo bias to loss aversion. But what causes or explains loss aversion? In
some, but not most, circumstances, the phenomenon can be attributed to the
individual’s emotional attachment to an object or the status quo. Indeed, people
exhibit more attachment to items that they have earned rather than gained by
chance.16 In some (but not all) circumstances it may also reflect people’s discom-
fort in participating in market transactions, especially with respect to goods that
are difficult to value. It is interesting that the effect may be diminished when
someone holds an asset for the purpose of making a deal to exchange it rather
than for its use value—as when the holder is a broker or agent acting on behalf
of a principal rather than the principal herself, or when the decision maker is a
corporation rather than an individual.17
In other situations, people may value the default because it reflects what most
(informed) people do, which is often a plausible basis for one’s own choice in the
absence of other information. Cass Sunstein thus writes of 401(k) savings
plans:18
Most 401(k) plans use an opt-in design. When employees first become eligi-
ble to participate in the 401(k) plan, they receive some plan information and
an enrollment form that must be completed in order to join. Under the alter-
native of automatic enrollment, employees receive the same information but
are told that unless they opt out, they will be enrolled in the plan (with default
options for savings rates and asset allocation). In companies that offer a
“match” (the employer matches the employee’s contributions according to
some formula, often a 50 percent match up to some cap), most employees
eventually do join the plan, but enrollments occur much sooner under auto-
matic enrollment. . . .
In many settings, any starting point will carry some informational content
and will thus affect choices. When a default rule affects behavior, it might
well be because it is taken to carry information about how sensible people
usually organize their affairs. . . . Some workers might think, for example,
that they should not enroll in a 401(k) plan and have a preference not to do so;

16. Arlen, Spitzer, and Talley, supra.

17. Arlen, Spitzer, and Talley, supra.
18. Cass R. Sunstein, Laws Of Fear: Beyond The Precautionary Principle 132,
138 (New York: Cambridge University Press, 2005). See James J. Choi, David Laibson,
Brigitte C. Madrian, and Andrew Metrick, Defined Contribution Pensions: Plan Rules,
Participant Choices, and the Path of Least Resistance, in Tax Policy and The Economy 67,
70 (James M. Poterba, ed., Cambridge, MA: MIT Press, 2002); Brigitte C. Madrian and
Dennis F. Shea, The Power of Suggestion: Inertia in 401(k) Participation and Savings
Behavior, 116 Quarterly Journal of Economics 1149, 1149–50 (2001); See Richard H.
Thaler and Shlomo Benartzi, Save More Tomorrow: Using Behavioral Economics to Increase
Employee Saving, 112 Journal of Political Economy, 2004, S164 (2004).
 complexities of decision making: frames 427

but the thought and the preference might shift with evidence that the employer
has made enrollment automatic. With respect to savings, the designated
default plan apparently carries a certain legitimacy for many employees, per-
haps because it seems to have resulted from some conscious thought about
what makes most sense for most people. This interpretation is supported by
the finding that the largest effects from the new default rule are shown by
women and African-Americans. We might speculate that members of such
groups tend to be less confident in their judgments in this domain and may
have less experience in assessing different savings plans.
But there are many situations where the status quo or the fact that one is
endowed with an object provides no informational content.
The endowment effect may be affected by the phenomenon of projection bias
described in Section 13.4.2, which may lead to exaggerated predictions of feel-
ings of loss.19 On the other hand, a failure of empathy may lead buyers to under-
estimate the strength of the endowment effect on owners and thus cause
distortions in negotiations “because buyers will tend to underestimate owners’
reservation prices. By the same token, owners will tend to overestimate buyers’
reservation prices.”20
Another explanation for the endowment effect and status quo bias is the
desire to avoid regret—“the painful feeling a person experiences upon determin-
ing she could have obtained a better outcome if she had decided or behaved
differently.”21 Of course, people also rejoice when the actual outcome is superior
to what might have happened. But, consistent with loss aversion, the negative
emotion of regret tends to be stronger.
Consistent with a “regret avoidance” account of the endowment effect, Ziv
Carmon and Dan Ariely have shown that, when contemplating a sale, both
buyers and sellers tend to focus on what the sale would cause them to forgo in
determining what they think the price should be. Since buyers and sellers forgo
different things in a sale (sellers forgo the item that is being sold and buyers
forgo the money they spend in the sale), buyers’ WTP and sellers’ WTA are likely
to diverge. Carmon and Ariely asked students at Duke University to contemplate
a sale of tickets to a basketball game. They found that buyers’ WTP was mainly

19. George Loewenstein, Ted O’Donoghue, and Matthew Rabin, Projection Bias in
Predicting Future Utility, Quarterly Journal of Economics 1209 (2003).
20. Id. Leaf Van Boven, David Dunning, and George Loewenstein conducted an exper-
iment in which sellers endowed with coffee mugs were asked to estimate how much
buyers would pay, and buyers were asked to estimate how much sellers would charge.
Sellers predicted that buyers would pay $3.93, compared to buyers’ WTP of $1.85; buyers
predicted that sellers would demand $4.39, compared to sellers WTA of $6.37. Egocentric
Empathy Gaps Between Owners and Buyers: Misperceptions of the Endowment Effect, 79
Journal of Personality and Social Psychology 66 (2000).
21. Chris Guthrie, Better Settle Than Sorry: The Regret Aversion Theory of Litigation
Behavior, 1999 U. Ill. L. Rev. 43.
428 problem solving, decision making, and professional judgment

influenced by monetary considerations such as the base price of the tickets and
how heavily they were discounted. Meanwhile, sellers’ WTA was more heavily
affected by factors related to the experience of the game, including the atmo-
sphere of the stadium and the importance of the game.22

14.2 regret aversion

If there is something to desire,

there will be something to regret.
If there is something to regret,
there will be something to recall.
If there is something to recall,
there was nothing to regret.
If there was nothing to regret,
there was nothing to desire.
—Vera Pavlova, Four Poems. From if there is something to desire: one
hundred poems by Vera Pavlova, translated by Steven Seymour, translation copy-
right © 2010 by Steven Seymour. Used by permission of Alfred A. Knopf, a division
of Random House, Inc.

In Section 13.1, we listed characteristics of the decision process that predicted

suboptimal outcomes. One ex-post indicator of a poor outcome is that the deci-
sion maker subsequently regrets the decision she made. Regretting a large pur-
chase or feeling bloated after eating too much are common examples. Subject to
several provisos, ex-post regret is a plausible indication that the decision was not
a good one:
• First, the decision maker’s values must be essentially the same when she
made the decision and when she regrets it; if her values have changed sig-
nificantly in the interim, regret does not indicate a poor decision.
• Second, people may regret that an advertently risky decision turned out
badly, but, again, this does not mean the decision was a poor one.23
• Third, people are quite adept at rationalizing bad outcomes, and may have
different regrets immediately after making a decision and in the long run.

22. Ziv Carmon and Dan Ariely, Focusing on the Forgone: How Value Can Appear So
Different to Buyers and Sellers, 27 Journal of Consumer Research 360–70 (2000).
23. Although a decision maker cannot undo the behavior that she now regrets, a deci-
sion maker may learn from her own experience and avoid the regret-causing behavior in
the future. Moreover, there may be categories of cases where regret is so ubiquitous that
a counselor to the decision maker would predict the consequences.
 complexities of decision making: frames 429

This makes it difficult to use regret to establish a neutral vantage point

from which to assess the quality of the decision.
Because regret is a disutility, decision makers often try to make decisions in a
way that avoids regret. Minimizing regret plays a role in decisions by consumers,
investors, professionals, and policy makers. Indeed, even in situations where
decision makers are risk-averse, regret-aversion may trump risk aversion.24

14.2.1 Imagining the Counterfactual Outcome

We only regret a decision if its outcome is worse than the outcome of a different
choice would have been. We can learn of the counterfactual outcome through
actual feedback or by imagining it. The easier it is to imagine the better
outcome—the more “mutable” the event is—the more likely we are to regret
actions that had bad consequences.
Mutability explains why regret is enhanced when the good outcome
almost happened. For example, Daniel Kahneman and Dale Miller posed this
problem:
Mr. C and Mr. D were scheduled to leave the airport on different flights at the
same time. They traveled from town in the same limousine, were caught in a
traffic jam, and arrived at the airport 30 minutes after the scheduled depar-
ture time of their flights. Mr. D is told that his flight left on time. Mr. C is told
that his flight was delayed, and only left 5 minutes ago.
Ninety-six percent of respondents thought that Mr. C would be more upset than
Mr. D.25
In many cases, the outcome that might have occurred is clearly superior or
inferior to what actually occurred. But sometimes this may be a matter of fram-
ing. Who is likely to be unhappier after an Olympic event, someone who won a
silver medal or someone who won a bronze? Victoria Medvec and her colleagues
studied contestants’ reactions immediately after the event, and concluded that
the former, though objectively better off, were actually less happy:26

24. Marcel Zeelenberg, Jane Beattie, Joop van der Pligt, and Nanne K. de Vries,
Consequences of Regret Aversion, 65 Organizational Behavior and Human Decision
Processes 148 (1996).
25. By the same token, we especially regret bad outcomes from exceptions to our rou-
tines: “Mr. Adams was involved in an accident when driving home after work on his regu-
lar route. Mr. White was involved in a similar accident when driving on a route that he
only takes when he wants a change of scenery.” Eighty-two percent thought that Mr. White
would be more upset than Mr. Adams. Daniel Kahneman and Dale T. Miller, Norm Theory:
Comparing Reality to Its Alternatives 93 Psychological Review 136–53 (1986).
26. Victoria H. Medvec, Scott F. Madley, and Thomas Gilovich, Where Less Is More:
Counterfactual Thinking and Satisfaction Among Olympic Medalists, Journal of Personality
and Social Psychology 625 (1995).
430 problem solving, decision making, and professional judgment

Consider the counterfactual thoughts of bronze and silver medalists. . . . One

would certainly expect the silver medalist to focus on almost winning the gold
because there is a qualitative difference between coming in first and any other
outcome. . . . Moreover, for the silver medalist, this exalted status was only
one step away. . . . In contrast, bronze medalists are likely to focus their coun-
terfactual thoughts downward. . . . [T]here is a categorical difference between
finishing third and finishing fourth.
The researchers did not study how long the disappointment lasted, but they
cite some anecdotal evidence of contestants in a spelling bee and an Olympic
race being plagued by the results later in their lifetimes.

14.2.2 Action vs. Inaction

At least in the short run, people tend to regret an action that turned out badly
(e.g., choosing to enter into a transaction that changes the status quo) more than
a failure to act that has unfortunate consequences.27 Thus, giving up an entitle-
ment is more likely to cause future regret than not gaining it in the first place.
Consider this question posed to participants by Kahneman and Tversky:28
Mr. Paul owns shares in Company A. During the past year, he considered
switching to stock in Company B, but decided against it. He now finds out that
he would have been better off by $1,200 if he had switched to the stock of
Company B. Mr. George owned shares in Company B. During the past year he
switched to stock in Company A. He now finds that he would have been better
off by $1,200 if he had kept his stock in Company B. Who feels greater
regret?
A striking 92 percent of the respondents thought that Mr. George would experi-
ence more regret.
Actions are often more salient and may involve a greater sense of personal
agency. Dale Miller and Brian Taylor suggest that some practices or supersti-
tions that counsel inaction are designed to anticipatorily avoid regret.29 For
example, contrary to an optimal strategy, most blackjack players will stand rather
than take another card when their cards total 16. And some people supersti-
tiously believe that switching lines in a supermarket risks that the former line

27. This may also be due to the availability heuristic: taking actions is often more vivid
and memorable than doing nothing.
28. Daniel Kahneman and Amos Tversky, The Psychology of Preferences, 246 Scientific
American 246 (1982). See also Daniel Kahneman and Dale Miller, Norm Theory: Comparing
Reality to Its Alternatives, in Heuristics and Biases: The Psychology of Intuitive
Judgment 397, 348 (Thomas Gilovich, Dale Griffin, and Daniel Kahneman eds.,
New York: Cambridge University Press, 2002) [hereafter Heuristics and Biases].
29. Dale Miller and Brian Taylor, Counterfactual Thought, Regret, and Superstition: How
to Avoid Kicking Yourself, in Heuristics and Biases, at 367.
 complexities of decision making: frames 431

will speed up and the new one will slow down. “To undertake a risky action
one easily imagines otherwise . . . is to tempt fate. . . . One is just ‘asking for it’
if one takes an action he or she knows, in retrospect, will be easy to imagine
otherwise.30 Daniel Keys and Barry Schwartz note the superstition that bullfight-
ers should not substitute for other matadors at the last minute as an example of
how regret in the decision process can “leak” into outcomes (See Section 12.1.2).
“This superstition may be rational, as the negative consequences of a matador
being gored include not just the physical damage to the matador but also the
regret, doubt, and other emotional suffering that can spread throughout the bull-
fighting community after a goring. A last-minute matador switch is a salient
event that tends to induce counterfactual thinking and increase susceptibility to
regret.”31
Actions can cause regret directly when they end in negative outcomes—as
when the decision to take a card makes a blackjack player “bust.” And actions
can also cause regret indirectly by changing the way one thinks about one’s past
behavior. Consider this example: Artie is about to take his first trip on Global
Airlines and is invited to join its frequent flier program. Ben is in the same posi-
tion, but Ben took a long trip on Global Airlines last month and didn’t take the
trouble to enroll in the frequent flier program then. Who is more likely to enroll
on this trip?
Research by Orit Tykocinski and Thane Pittman suggests that Artie is more
likely to enroll in the frequent flyer program than Ben.32 If he joined the frequent
flier program now, Ben would have to admit that his failure to join earlier had
been a mistake. His current action would make him regret his past inaction.
As the authors put it, “by quickly turning down the subsequent action opportu-
nity . . . one can avoid thinking about the inferior current opportunity that
triggers the perception of loss.” Unlike the blackjack players, who do not know
whether the next card will make them win or bust, Ben knows in advance that
joining the frequent flier program will have beneficial results. Yet they still avoid
taking these actions because doing so would make them regret their previous
procrastination.
In some circumstances, inaction is more likely to cause more regret than
action—even in the short run. Consider the dilemma of the soccer goalie as his
opponent swings his foot back for a penalty kick. Should the goalie move left,
right, or stay put? Although goalies are more than twice as likely to make a save

30. Miller and Taylor, supra.

31. Daniel J. Keys and Barry Schwartz, “Leaky” Rationality: How Research on Behavioral
Decision Making Challenges Normative Standards of Rationality, 2 Perspectives on
Psychological Science 162–80 (2007).
32. Orit E. Tykocinski and Thane S. Pittman, The Consequences of Doing Nothing:
Inaction Inertia as Avoidance of Anticipated Counterfactual Regret, 75 Journal of
Personality and Social Psychology 607 (1998).
432 problem solving, decision making, and professional judgment

if they stay in the center of the goal, it turns out that they only rarely stay put. The
anticipation of regret may help explain these findings: professional goalkeepers
indicated that they would feel more regret if they let a penalty kick go by without
moving than if they had taken an ineffective action. Why would inaction cause
more regret in this case than action? The answer may relate again to the ease of
generating counterfactuals. Because goalkeepers are expected to act, action may
be less mutable than inaction. In other words, it may be easier to generate the
counterfactual “if only I’d jumped for the ball” than it is to think, “if only I’d
stayed put.”33 Perhaps action is the status quo for many athletes.
Whether actions or inactions beget more regret also seems to depend on the
time horizon. While in the short run people most regret actions, in the long run
they most regret failures to act.34 Research by Thomas Gilovich and Victoria
Medvec found that people’s most lasting regrets involved not failures to take
discrete actions, such as failing to sign up for the discounted Tuscany tour, but
rather patterns of inaction, such as not investing sufficiently in educational
opportunities or spending enough time with friends. Still, some discrete inac-
tions, such as failure to seize a romantic or business opportunity, can have last-
ing consequences. A popular web comic humorously illustrates this point with a
bar graph entitled “Number of Google Results For I _______ Have Kissed Her
(Or Him).” (See Figure 14.2.) The bar that fills in the blank with “should” reveals
almost nine times more hits than the bar for “shouldn’t.” 35

NUMBER OF GOOGLE RESULTS FOR

“I HAVE KISSED HER”
(OR HIM)

10,230

1,213

SHOULDN’T SHOULD

figure 14.2 number of google results for “i____ have kissed her (or
him). http://xkcd.com/458/ . reprinted by permission of xkcd.

33. Patricia Cohen, The Art of the Save, for Goalie and Investor, New York Times, March
1, 2008, Business Section, online edition.
34. Thomas Gilovich and Victoria H. Medvec, The Experience of Regret: What, When,
and Why, 102 Psychology Review 379 (1995).
35. XKCD webcomic, http://xkcd.com/458/.
 complexities of decision making: frames 433

Gilovich and Medvec identify a number of factors that may diminish regret
about actions and increase regret about inactions over time:
• Elements that reduce the pain of regrettable actions so that the initial pain
of the regrettable action diminishes over time:
• Ameliorating the action. The pain of regrettable action is felt more
quickly and the action can often be ameliorated. (If you marry
Mr. Wrong, you can get divorced; if you fail to marry Mr. Right, he may
no longer be available.)
• Reducing dissonance, for example, by identifying silver linings (“At least
we had wonderful children”; “I learned a lot from making that bad
investment.”)
• Elements that bolster the pain of regrettable inactions.
• The passage of time makes one more confident that one could have
succeeded in the forgone action, thus making it easy to mentally “undo”
the inaction. (“We thought religious differences would get in the way of
our marriage, but we could have worked them out.”)
• While regrets about actions focus on bad things that actually happened,
regrets about inactions concern good things that might have happened—
an open-ended category that can grow in scope and importance with the
passage of time.
• Elements that differentially affect cognitive availability. Many regrets about
inaction involved unrealized ambitions and unfulfilled intentions that are
more available to our minds than goals that have been fulfilled.

14.2.3 The Effects of Anticipating Feedback about What Would Actually

Have Happened
Are people’s decisions affected by the knowledge that they will learn about the
consequences of the foregone option and that this might induce regret? An
experiment by two business school professors, Richard Larrick and Terry Boles,
found that negotiators were more prone to reach agreement when doing so
would shield them from knowing their opponents’ BATNA (best alternative to a
negotiated agreement) than when they knew they would learn their opponents’
BATNA and then possibly regret having made the deal.36
Against the background of this and similar studies, Chris Guthrie conducted
an experiment to shed light on litigants’ settlement behavior. Participants were
given the facts of two cases—one involving an employee’s claim for overtime,
the other a “slip and fall” injury—and told that they might be litigated in either
of two jurisdictions. In the traditional jurisdiction, if the parties settle, they

36. Richard P. Larrick and Terry L. Boles, Avoiding Regret in Decisions with Feedback: A
Negotiation Example, 63 Organizational Behavior and Human Decision Processes
87–97 (1995).
434 problem solving, decision making, and professional judgment

will have no idea what the judge would have awarded had the case gone to trial.
In the “regret” jurisdiction, the judge informs the parties what he would
have awarded. By a large margin, the participants predicted that the parties—-
defendants as well as plaintiffs—would be more prone to settle in the traditional
jurisdiction, where they would be shielded from knowledge of the foregone out-
come and thus shielded from regret.37 Whether to encourage settlement or just
to avoid the costs of unnecessary adjudication, most real-world procedures shield
parties negotiating a settlement from learning what the specific foregone out-
come might have been.

14.2.4 Regret and Suboptimal Learning Processes

Faced with the prospect of learning about the foregone outcome and experienc-
ing regret, a decision maker has three options: (1) make the decision that reduces
anticipated regret, (2) avoid feedback about the foregone outcome, or (3) just
ignore the anticipated regret. As we have seen, the problem with the first is that
people are often not good at forecasting future affective states, and especially
tend to ignore the anodyne workings of the psychological immune system.38
Although the second may be a useful self-protective device for one-time deci-
sions, feedback is the only way to improve one’s performance in repeated deci-
sions. Such “myopic regret aversion”39 is especially counterproductive for
professionals, whether in law, policy, business, or medicine. The third option is
often the best, but it is not costless. Even if regret dissipates over time, it can be
a strong negative emotion and thus a real hedonic loss while it lasts—and who
knows how long that may be?

14.3 sunk costs 40

Suppose that you are watching a movie and you find it to be boring and unlikely
to get any better. Do you nonetheless keep watching the movie? Suppose that
you paid $10 to see the movie. Does that affect your decision? In a study by

37. Guthrie, Regret Aversion Theory, supra.

38. Section 12.1.5.
39. Jochen Reb and Terry Connolly, “Myopic Regret Aversion and Feedback Avoidance
in Repeated Decisions” (working paper 2005), http://www.sie.arizona.edu/MURI/cd/
content/Reb%20Connolly%20Myopic%20reg%20av%20and%20reason%20based%20c
hoice%20Thrust%20A.doc.
40. See generally Daniel Friedman, Kai Pommerenke, Rajan Lukose, Garret Milam,
and Bernardo A. Huberman, Experimental 0407008, Economics Working Paper Archive
EconWPA (2004), http://ideas.repec.org/p/wpa/wuwpex/0407007.html.
 complexities of decision making: frames 435

Deborah Frisch, 57 percent of people who said they would stop watching a free
movie said they would keep watching it if they had paid $10.41
This phenomenon of valuing sunk costs was replicated in an experiment
involving a campus theater company at the University of Ohio. Theatergoers
who were about to buy season tickets were randomly placed into groups that
either paid full price or received a discount. During the first half of the season,
those who paid full price attended many more plays than those who had received
the discount—though the rates of attendance converged for the second half of
the season.42
Economic theory holds that the price they paid for the ticket should not affect
ticket-holders’ decision whether to attend. Whether they paid full price or
received a discount, the cost of the ticket is “sunk” and can have no effect on the
utility of attending—as compared to whether or not it is snowing that evening,
or the ticket holder has a paper due the next day, or is or just isn’t in a theater-
going mood. But for those who paid full price, the loss of the money was more
salient than for those who got the discount.
The phenomenon of honoring sunk costs manifests itself in many personal,
government, and business decisions. For example:43
• Members of a health club with semi-annual dues made greater of use of
the facility in the month after they paid dues; use declined over the next five
months, until the cycle repeated itself.44
• A soldier on the first day of the Gulf War in 1991: “Finally, the day has
finally come. You’ve got to think logically and realistically. Too much
money’s been spent, too many troops are over here, too many people had
too many hard times not to kick somebody’s ass.”
• The parents of American troops killed in Iraq:45 “To say the war right now is
worthless or that it is without cause or unjust, would be saying that my son, for
all he had in him to go over there to preserve the freedom of his country, would
have died in vain.” “Don’t take our soldiers out of Iraq. If you do, it means
our loved ones will have died for nothing.” (This is not the universal view.

41. Deborah Frisch, Reasons for Framing Effects, Organizational Behavior and Human
Decision Processes, 54, 399–429 (1993), cited in Daniel J. Keys and Barry Schwartz, “Leaky”
Rationality: How Research on Behavioral Decision Making Challenges Normative Standards
of Rationality, 2 Perspectives on Psychological Science, 162–80 (2007) [hereafter Leaky
Rationality].
42. Hal R. Arkes and Catherine Blumer, The Psychology of Sunk Cost, 35 Organizational
Behavior 124 (1985).
43. Many of the quotations come from Reid Hastie and Robyn Dawes, Rational
Choice in an Uncertain World: The Psychology of Judgment and Decision
Making (Sage Publications: Thousand Oaks, CA, 2001) 32.
44. John T. Gourville and Dilip Soman, Payment Depreciation: The Effects Of Temporally
Separating Payments from Consumption, 25 Journal of Consumer Research 160 (1998).
45. New York Times, Sept. 9, 2004.
436 problem solving, decision making, and professional judgment

The mother of a soldier who died in Iraq said: “It pains me to hear that more
people should die because those people have died. That makes no sense. We
can honor them by having an intelligent, honest policy.”46)
• Senator Jim Sasser arguing for further investment in a project that would
be worth less than the amount of money necessary to finish it: “Completing
Tennessee-Tombigbee is not a waste of taxpayers’ dollars. Terminating the
project at this late stage of development would, however, represent a seri-
ous waste of funds already invested.”
• A businessman: “I’ve already invested so much in the Concorde airliner . . .
that I cannot afford to scrap it now.”
• Members of Congress arguing for the dollars and lives already spent on
NASA’s space shuttle program as a reason for continuing it.47
• The more people pay for advice from consultants or other experts with
respect to business or personal decisions, the more likely they are to follow
the advice.48
If taking account of sunk costs doesn’t make economic sense, is it therefore
always irrational? Consider:
• Avoiding regret. If the $500 treadmill you bought several months ago is sit-
ting idly in your family room, you may mentally kick yourself each time you
look at it and regret the purchase decision. Like all feelings, the sense of
regret is arational. Whether it leads to behaviors that are in your long-term
self-interest is a matter of chance. While anticipatory regret about the idle
treadmill may lead you to exercise more often, anticipatory regret about
wasting the nonrefundable airline ticket to a country that has had an upsurge
in terrorism may lead you to take a dangerous and anxiety-laden trip.
• Reputation. Individuals and organizations may rationally value their reputa-
tions for not making poor decisions, for adhering to commitments, and for
not being wasteful. Thus, even if a policy maker recognizes that it is subop-
timal to continue a project or activity, she may decide that it is in her or the
agency’s interest to obscure a poor decision by sticking with it49—or not to
be seen as a “flip-flopper.” We imagine that circumstances of this sort often
give rise to conflicts of interest between agents and principals—for
example, between the interests of a decision maker and the clients or con-
stituents she represents (who bear the costs of continuing a bad project).

46. New York Times, Aug. 23, 2005.

47. See The Space Shuttle: Old, Unsafe, and Costly, The Economist, Aug. 28, 2003, 77.
48. Francesca Gino, “Getting Advice from the Same Source but at a Different Cost: Do
We Overweigh Information Just Because We Paid for It?” Harvard Business School
Working Paper, No. 05-017, 2004.
49. See Douglas Walton, “The Sunk Cost Fallacy or Argument from Waste,” 16
Argumentation 473 (2002).
 complexities of decision making: frames 437

• Sense of self, and self-education. “Reputation” refers to how other people think
of you. But you may also have a sense of yourself—a self-reputation—as a
person who is not wasteful and who follows through with plans and com-
mitments. Moreover, you may want to discipline yourself to be that kind of
a person. Keeping commitments that you wish you had not made may teach
you to be more cautious in the first place.50
• Precommitment. In Section 13.6.2 we discussed strategies through which
individuals could commit themselves to long-run valued behaviors in the
face of short-run temptations. Unlike Ulysses tying himself to the mast,
most of our precommitments are open to subsequent reneging. Giving
weight to sunk costs—e.g., the $500 treadmill sitting in your family room—
can be a psychic incentive for maintaining those commitments.51
• Respect for prior decisions and finality. Like judges’ respect for precedent,
sticking with earlier decisions even when they seem questionable may
reflect a combination of respect for their predecessors and the costs of con-
stantly revisiting earlier judgments. Even for personal decisions, individu-
als may believe that they do not have a better vantage point now than when
they initially made the decision.
• Meaning. The parents’ comments about loss of their children in Iraq sug-
gest that attending to sunk costs may give meaning to terrible events in our
lives—though sometimes at the cost of creating more terrible events.
Whether or not taking account of sunk costs sometimes makes sense for indi-
viduals, we are skeptical about its value to governments and private organiza-
tions. In any event, it is important to understand that a decision that takes
account of sunk costs inevitably compromises the pursuit of other goals by
diverting resources that might be used more productively or, occasionally, by
continuing to endanger people’s welfare and lives. We will return to the issue of
sunk costs when we discuss the escalation of commitment in Section 17.3.3.

14.4 mental accounting 52

Under classic economic theory, a decision maker views the money or other assets
she spends on, or gains from, a transaction in terms of their effect her overall

50. See Richard Epstein, Skepticism and Freedom: A Modern Case for Classical
Liberalism, ch. 8 (Chicago, IL: University of Chicago Press, 2003); Walton, supra.
51. Consider, though, whether taking account of the sunk costs to goad you into exer-
cising is best characterized as a “rational” decision, or whether it would be more accurate
to say that the overall strategy is (rationally) based on the prediction that you will (irratio-
nally) take account of sunk costs.
52. This section relies heavily on Richard H. Thaler, Mental Accounting Matters, in Choices,
Values, and Frames (Daniel Kahneman and Amos Tversky, eds., Cambridge, UK:
Cambridge University Press, 2000) 241 [hereafter, Mental Accounting Matters]. See also Richard
H. Thaler, Mental Accounting and Consumer Choice, 4 Marketing Science 199 (1995).
438 problem solving, decision making, and professional judgment

wealth or well-being. In fact, however, people tend not to consider decisions in

such comprehensive terms, but rather frame them in terms of separate accounts
that are not necessarily fungible. In a pathbreaking article, Richard Thaler exam-
ines a variety of such “mental accounting” phenomena. Starting from the pre-
sumption that people employ decision-making criteria that are intended to make
them as happy as possible, he derives the following principles of hedonic framing
from prospect theory.
1. Segregate gains (because the gain function is concave downward53). That is,
three separate gains of n are experienced as more pleasurable than one
gain of 3n. For example, when asked whether one would be happier win-
ning two lotteries that pay $50 and $25 respectively, or winning one lottery
that pays $75, most respondents choose the two lotteries.
2. Integrate losses (because the loss function is concave upward). That is, one
loss of 3n is experienced as less painful than three separate losses of n. For
example, an employer planning to require employees to pay a larger share
of their health insurance premiums may cause less unhappiness by moving
to the target all at once rather than spreading it out over a number of years.
Consider credit cards, or debt consolidation loans, which pool many small
losses into one large one, or buying a few optional features when purchas-
ing a car. Lee Ross has termed such integrations, which avoid resetting the
reference point with each cost incurred, “hedonic laundering.”
3. Integrate smaller losses with larger gains (to offset loss aversion). Thaler gives
the example of vacationing in Switzerland with his wife where he gave a
lecture for a fee. The Swiss franc was very high and everything seemed
expensive. They mentally “deducted” the meals from his speaking fee,
which they were mentally “allowed” to do because it was in the same
account.
4. Segregate small gains (silver linings) from larger losses. Because the gain func-
tion is steepest at its origin, the utility of a small gain can exceed the utility
of slightly reducing a large loss. Consider, for example, car dealers who
offer a “rebate” rather than the equivalent reduction in the price of the car.

“Injuries, therefore, should be inflicted all at once, that their ill savor being less
lasting may the less offend; whereas, benefits should be conferred little by little,
that so they may be more fully relished.”
—MACHIAVELLI, THE PRINCE

53. See Figure 14.1.

 complexities of decision making: frames 439

14.4.1 Decoupling Decisions and Consequences: The Value of Feedback

Would you rather pay a flat rate for phone service or pay for each phone call?
Many people opt for the former, even when the latter is less expensive.
“Prepayment separates or ‘decouples’ the purchase from the consumption and
in so doing seems to reduce the perceived cost of the activity.”54 Similarly, many
people prefer to prepay for a vacation, so as not to have the pleasurable experi-
enced compromised by the thought of having to pay for it later.55 At Club Med
resorts, the room cost includes meals, bars, and entertainment.
Using credit cards to defer the payment of purchases also has this decoupling
effect. But by mitigating the pain of paying for a purchase, a credit card encour-
ages overpurchasing. And, as Loewenstein and O’Donoghue note, “when the
credit-card bill comes, people still experience the pain of paying, and in fact,
because payments aren’t clearly linked to specific purchases, consumers seem to
find it especially painful to pay off credit-card debt.”56
The following are some examples of mental accounting.

14.4.2 Budgeting
Kahneman and Tversky posed essentially these questions to two randomly
selected groups in 1984.
• Imagine that you have decided to see a play presented by a campus theater
company and paid the admission price of $10 per ticket. As you enter the
theater, you discover that you have lost the ticket. Seats were not assigned,
and the ticket cannot be recovered. Would you pay $10 for another ticket?
• Now imagine the same scenario, except that you haven’t yet purchased the
ticket. As you enter the theater, you discover that you have lost a $10 bill.
Would you still pay $10 for a ticket to the play?
Only 46 percent of those who lost the ticket would purchase a new ticket,
while 88 percent of those who lost the $10 bill would purchase another one. Why
would this be? Kahneman and Tversky explain: “Going to theater is normally
viewed as a transaction in which the cost of the ticket is exchanged for the experi-
ence of seeing the play. Buying a second ticket increases the cost of seeing the
play to a level that many respondents apparently find unacceptable. In contrast,
the loss of the cash is not posted to the account of the play, and it affects the

54. Mental Accounting Matters at 252.

55. Drazen Prelec and George Loewenstein, Beyond Time Discounting, 8 Marketing
Letters 97 (1997).
56. George Loewenstein and Ted O’Donoghue, We Can Do This the Easy Way or the
Hard Way: Negative Emotions, Self-Regulation, and the Law, 73 Chicago Law Review 183
(2006).
440 problem solving, decision making, and professional judgment

purchase price of a ticket only by making the individual feel slightly less
affluent.”57
In another study, participants in one group were told that they had spent $50
for a ticket to a basketball game earlier in the week and asked if they would be
willing to buy a theater ticket. Participants in the second group were told that
they had paid $50 for a parking violation and asked if they would purchase the
theater ticket. Because the theater ticket was drawn from the “entertainment”
budget, people who had spent the $50 for a baseball ticket were less likely to buy
the theater ticket than those who paid $50 for the parking fine.
Dividing spending into budget categories serves two purposes. It can make
the trade-offs between competing uses of funds transparent and, to the extent
that certain accounts are “off limits,” it can assist in self-control. When budgets
are not treated as fungible, however, the existence of separate accounts can have
distorting economic effects.

14.4.3 Self-Control, Gift-Giving, and Income Accounting

From a purely economic point of view, a gift of cash is ideal because it is com-
pletely fungible; next best is the gift of an item that the recipient would have
purchased anyway. But a luxurious present may have great value for a recipient
who maintains budget limits for reasons of self-control or a sense of appropriate-
ness. Richard Thaler notes that the National Football League had difficulty in
getting superstar players to attend the Pro Bowl by offering money, but suc-
ceeded when they offered the players two first-class tickets and luxury accom-
modations in Hawaii.
People tend to match the source of funds with their use, so that money from
the windfall of winning an office football pool will be spent at a fancy restaurant,
while the same amount in a tax refund will be used to pay the bills.
Funds from the same source can be framed in different ways with different
consequences. Consider whether government should frame a tax give-back as a
“rebate” or a “bonus.” When, in 2001, the federal government gave tax “rebates”
in order to stimulate consumer spending, only 28 percent of people surveyed
said that they spent the money immediately.58 Because the “rebate” was simply
their money being returned to them, they were most likely to save it. Experiments
suggest that framing the same scheme as a bonus, implying a windfall, would
have led to more spending.59

57. Daniel Kahneman and Amos Tversky, Choices, Values, and Frames, 39 American
Psychologist 341 (1984).
58. Matthew D. Shapiro and Joel Slemrod, “Consumer Response to Tax Rebates,”
working paper, University of Michigan and NBER (2001).
59. Nicholas Epley, Rebate Psychology, New York Times, Op-Ed section, Jan. 1, 2008.
 complexities of decision making: frames 441

14.4.4 Budgeting Time

As you are about to purchase a clock radio for $50 at a consumer electronics
store, a reliable friend informs you that the identical radio is on sale for $40 at a
discount store fifteen minutes away. Do you make the trip? Now you are about to
purchase a television set for $500 at a consumer electronics store, when the reli-
able friend informs you that the identical TV set is on sale for $490 at a discount
store fifteen minutes away. Do you make the trip?
If you are like most people, you are more likely to take fifteen additional min-
utes to buy the clock radio than the television set at the discount store.60 Similarly,
when asked how much they would be willing to pay to avoid waiting in a ticket
line, people were willing to pay twice as much to avoid the wait for a $45 ticket
than for a $15 ticket.61
This does not make economic sense. “A rational person should allocate time
optimally,” and “the marginal value of an extra minute devoted to any activity
should be equal.”62 At the end of the day, if you purchased the clock radio for $40
your bank balance is $10 better off, and so too if you purchased the television set
for $490. However, our mental accounting considers the time taken in doing a
transaction in terms of the dollar value of the transaction. Recall, also, the discus-
sion of the evaluability of proportions in Section 13.3.1.

14.4.5 Conclusion: So What?

The endowment effect, the status quo bias, and other mental accounting phe-
nomena conflict with rational choice theory. Do they also result in suboptimal
individual decision making?
Looking at the indicators of suboptimal decisions mentioned in Section 13.1,
the phenomena render at least some decisions sensitive to the way the issue is
framed, and may subject the decision maker to manipulation by others. Consider,
for example, the framing of “status quo” contract provisions in terms of the
legal defaults or form contracts. On the other hand, the phenomenon of regret-
avoidance implies that decisions based on loss avoidance minimize the decision
maker’s subsequent regret. Moreover, to say that mental accounting violates the
precepts of rational economic choice is not to say that it is necessarily bad for us.
Thaler suggests:63

60. Amos Tversky and Daniel Kahneman, The Framing of Decisions and the Psychology
of Choice, 211 Science 453 (1981); Richard H. Thaler, Toward a Positive Theory of Consumer
Choice, in Choices, Values, and Frames 269 (New York: Cambridge University Press,
2000).
61. France LeClerc, Bernd H. Schmidt, and Laurette Dube, Decision Making and Waiting
Time: Is Time Like Money?, 22 Journal of Consumer Research 256 (1995).
62. Mental Accounting Matters 256.
63. Mental Accounting Matters 267.
442 problem solving, decision making, and professional judgment

Mental accounting procedures have evolved to economize on time and think-

ing costs and also to deal with self-control problems. As is to be expected, the
procedures do not work perfectly. People pay attention to sunk costs. They
buy things they don’t need because the deal is too good to pass up. . . . They
put their retirement money in a money market account.
It is not possible to say that the system is flawed without knowing how to
fix it. Given that optimization is not feasible (too costly), repairing one prob-
lem may create another. For example, if we teach people to ignore sunk costs,
do they stop abiding by the principle, “waste now, want not?” If we stop being
lured by good deals, do we stop paying attention to price altogether?
This sounds right for many ordinary personal decisions. However, lawyers
and policy makers are agents, who counsel clients and make public decisions in
which the dollars and human welfare at stake can be momentous. Thaler tells of
the congressional investigation of the disaster that ensued when Teton Dam in
southern Idaho broke in June 1976. The investigation focused, among other
things, on “the inclination on the part of the Bureau of Reclamation to continue
dam construction, once commenced, despite hazards which might emerge
during the course of construction.” A witness for the Bureau could not think of
a single example in which dam construction, once begun, had been even tempo-
rarily interrupted because of such concerns.

14.5 the downsides of choices or options

14.5.1 Choice Overload

No idea is more fundamental to Americans’ sense of themselves as individuals and as a
culture than choice. The United States has come to epitomize the “republic of choice.”
—Sheena S. Iyengar and Mark R. Lepper, Choice and Its Consequences: On the
Costs and Benefits of Self-Determination64

Look at this peanut butter! There must be three sizes of four brands of four consistencies!
Who demands this much choice? I know! I’ll quit my job and devote my life to choosing
peanut butter! Is “chunky” chunky enough or do I need extra chunky? I’ll compare ingre-
dients! I’ll compare brands! I’ll compare sizes and prices! Maybe I’ll drive around and see
what other stores have! So much selection and so little time!
—Bill Watterson, It’s A Magical World (1996) (Calvin & Hobbes)

64. Sheena S. Iyengar and Mark R. Lepper, Choice and Its Consequences: On the Costs
and Benefits of Self-Determination (2002), in Abraham Tessler, Diederik A. Stapel, and
Joanne V. Wood (eds), Self and Motivation: Emerging Psychological Perspectives)
(Washington, DC, American Psychological Association, 2002) 71, citing Lawrence M.
Friedman, The Republic of Choice (Cambridge: Harvard University Press, 1990).
 complexities of decision making: frames 443

All things being equal, we think it is good to have options—the more that
are available, the more likely we’ll find one that’s a good fit with our interests,
needs, or objectives. But choices carry costs as well. As the number of choices
increase, decision making becomes more difficult, time consuming, and even
stressful.
People faced with many options tend to employ simplifying decision strate-
gies.65 For example, in studies of how people evaluate home housing purchases,
participants used compensatory procedures when choosing among three options,
but switched to a noncompensatory procedure when choosing among a dozen.66
As the number of housing options increased, the percentage of choices that
reflected the participants’ real preferences (as elicited in a separate process)
declined from 70 percent when considering five houses to 37 percent when con-
sidering twenty-five.67 A study of professional health care purchasers for large
corporations showed that the large majority used noncompensatory strategies in
choosing among health plans that had many different attributes.68
The remainder of this section considers another set of problems caused
by options, most of which involve (1) violation of the axiom of independence from
irrelevant alternatives, which holds that the relative ranking of a particular set of
options should not vary with the addition or deletion of other options, or (2) vio-
lation of the axiom of invariance, which holds that the order in which choices are
presented (or the way they are described) should not affect the outcome of the
decision. (See Section 12.2.2)

14.5.2 Option Devaluation

Chris Guthrie presents this problem:69 Suppose that you are considering
which law school to attend. In one scenario, you have been admitted only to
Harvard Law School; in the second, you have been admitted to both Harvard
and Stanford. Will your assessment of the quality of Harvard differ in the two
scenarios?

65. See generally John W. Payne, James R. Bettman, and Eric J. Johnson, The
Adaptive Decision Maker (New York: Cambridge University Press, 1993).
66. See Chapter 4 for the distinction between these procedures.
67. Richard W. Olshavsky, Task Complexity and Contingent Processing in Decision
Making: A Replication and Extension, 24 Organizational Behavior and Human
Performance 300 (1979); Naresh K. Malhorta, Information Load and Consumer Decision
Making, 8 J. Consumer Res. 419 (1982)—both summarized in Chris Guthrie, Panacea or
Pandora’s Box? The Costs of Options in Negotiations, 88 Iowa Law Review 601, 632–33
(2003) [hereafter, Guthrie, Options].
68. Judith H. Hibbard, Jacquelyn J. Jewett, Mark W. Legnini, and Martin Tusler,
Choosing a Health Plan: Do Large Employers Use the Data?, Health Affairs 172 (Nov.–
Dec. 1997). Also summarized in Guthrie, Options.
69. Guthrie, Options.
444 problem solving, decision making, and professional judgment

Although there is no good reason why it should—after all, your admission to

Stanford has given you no new information about Harvard—the phenomenon
of comparative loss aversion suggests that the attractiveness of Harvard will be at
least somewhat diminished. Comparing the two schools brings to mind the
advantages and disadvantages of each one—in terms of climate, size, prestige, or
whatever other factors matter to the applicant. However, as prospect theory sug-
gests, losses loom larger than gains; thus, the disadvantages of each school will
make it more unattractive than the advantages make it attractive. So, wherever
you end up going, you will view the prospect less favorably than had you been
admitted to only one of the schools.
Lyle Brenner, Yuval Rottenstreich, and Sanjay Sood demonstrated compara-
tive loss aversion in several experiments. For example, three groups of partici-
pants were asked how much they would pay for subscriptions to one or all of the
magazines: People, Time, Business Week, and The New Yorker.
• members of the isolated group were asked what was the maximum they
would pay for one of these magazine without comparing it to the others;
• members of the accompanied group were asked what they would pay for
subscriptions to each of the four magazines; and
• members of the ranked group were asked to rank the magazines and then
state how much they would pay.
On average, participants in the isolated group were willing to pay $21.42 for
People; those in accompanied and ranked groups were willing to pay $15.96 and
$15.09, respectively. The phenomenon of participants’ willingness to pay more
for the isolated item than when it was accompanied by other options or ranked
was robust across the magazines and across other items, including videotapes
and airplane roundtrips to various locations.70 The mere presence of options
engendered some comparisons, and the ranking exercise forced the participants
to make explicit comparisons.
Guthrie replicated the experiment in the context of a car purchase. In the
one-car group, participants on average said they would pay a maximum of
$13,125 for a Toyota Corolla. After being asked to rank the Toyota with a Honda
Civic and Mazda Prestige, the average willingness to pay dropped 13 percent to
$11,447.
The effects of grouping. In the example of choosing an apartment in Section 4.3,
we noted that using the common decision-making procedure, elimination by
aspects (EBA), can alter the probability of achieving a particular outcome in a
way that is, at best, orthogonal to the decision maker’s interests. This is espe-
cially likely when, rather than considering the criteria (cost, size, etc.) in their
order of importance, they make the easiest comparisons first. Amos Tversky gives

70. Lyle Brenner, Yuval Rottenstreich, and Sanjay Sood, Comparison, Grouping, and
Preference, 10 Psychology of Science 225 (1999).
 complexities of decision making: frames 445

this example:71 Suppose that you are offered a choice between a recording of a
Beethoven symphony and Debussy piano works. You like the composers about
the same, so the probability of choosing one or the other is about 0.5. Now sup-
pose you are offered a choice among three recordings—the Debussy and two
versions of the Beethoven symphony. Is the probability of each choice now 0.33?
Regardless of the comparative excellence of the recordings, it’s closer to 0.25 for
each of the Beethoven’s and 0.50 for the Debussy. Most likely, you first consider
attributes that make you favor the Beethoven or Debussy recording at this time—
you’re in a German or French mood, or the sound of a symphony or of a piano
seems more appealing—and then, only if you choose Beethoven, do you com-
pare the two Beethoven discs.
Brenner et al. did some further experiments to see how grouping affected
people’s choices among options. Participants were asked to choose among four
items—for example four kinds of restaurants, four videotapes, or four gifts. In
each case, three of the items were grouped together and one was alone, and the
participants were asked to indicate their preference either for the solo option or
their choice of one of the three grouped options. For example, participants were
asked whether they would prefer to eat at (1) a seafood restaurant, or (2) their
choice of a Mexican, Italian, or Thai restaurant. Participants consistently deval-
ued options in the group of three restaurants compared to the solo item.
Apparently participants first compared the members of the group with each
other to choose a favorite, which they then compared to the option that stood
alone. “Comparative loss aversion implies that the greater degree of within-
group comparisons will sharply reduce the attractiveness of each of the grouped
options. However, because of the lesser degree of between-group comparisons,
the attractiveness of the lone option will not decrease as much.” The authors
observe that advertisers or salespersons can employ the grouping phenomenon
to manipulate consumers’ choices:
Consider choosing between three cars. One is a Japanese sedan, the second a
Japanese sports car, and the third is an American sports car. Our findings
suggest that the American car is more likely to be chosen when the cars are
grouped by country of origin (because it is the only American option) and less
likely to be chosen when the cars are grouped by body style (because it is in
the sports-car group). Similarly, the Japanese sedan is more likely to be chosen
when the cars are grouped by body style and less likely to be chosen when the
cars are grouped by country of origin. . . . [S]alespeople have been known to
emphasize the relative merits of Japanese and American cars, or the virtues
of sedans and sports cars, in an attempt to guide the consumer toward a par-
ticular purchase.

71. Amos Tversky, Elimination by Aspects: A Theory of Choice, 79 Psychological

Review 281 (1972).
446 problem solving, decision making, and professional judgment

Along similar lines, Donald A. Redelmeier and Eldar Shafir asked physicians
to decide whether, after unsuccessfully trying a number of anti-inflammatory
medicines on a patient with osteoarthritis in his hip, to refer the patient for hip-
replacement surgery or try different medication. Physicians given the option to
prescribe one of several drugs that had not previously been tried on the patient
were more likely to choose surgery than those told that there was only one untried
drug.72
14.5.2.a The Downside of Having Too Many Choices Sheena Iyengar and
Mark Lepper have studied how consumers dealt with different numbers of
choices of food products.73 For example, customers entering an upscale food
market on different days encountered a display inviting them to taste either six
(limited choice) or twenty-four (extensive choice) different Wilkin & Sons pre-
serves. Anyone who approached the display received a $1 certificate toward the
subsequent purchase of a Wilkin & Sons preserve from the store’s regular
shelves. Both groups of customers actually tasted about the same number of
samples, but those who were offered the limited choice were much more likely
to purchase the product. A similar experiment, involving Godiva chocolates,
indicated that although participants enjoyed choosing from the extensive choice
options, those given the limited choice liked their chocolate better than those
who chose from the large sample. Those in the extensive choice group felt more
responsible for their choices, more dissatisfaction, and more regret about their
choices. Iyengar and Lepper speculate that they were relatively unsure about the
choice and “burdened by the responsibility of distinguishing good from bad
decisions.”
There is reason to believe that this phenomenon extends to more important
decisions than the choice of gourmet foods. Many employers offer 401(k) retire-
ment savings plans, some of which allow the employee to invest in just a few
funds, some in many. A large-scale survey across many industries indicated that
75 percent of employees participated in retirement savings plans that offered two
funds, while only 60 percent participated in plans with fifty-nine funds.74

72. Donald A. Redelmeier and Eldar Shafir, Medical Decision Making in Situations that
Offer Multiple Alternatives, 274 Journal of the American Medical Association 302 (1995).
73. Sheena S. Iyengar and Mark R. Lepper, When Choice Is Demotivating: Can One
Desire Too Much of a Good Thing?, 76 Journal of Personality and Social Psychology
995–1006 (2000).
74. How Much Choice Is Too Much: Determinants of Individual Contributions in 401K
Retirement Plans, in Developments in Decision-Making Under Uncertainty:
Implications for Retirement Plan Design and Plan Sponsors (O. S. Mitchell and S.
P. Utkus eds., Oxford: Oxford University Press, 2004).
 complexities of decision making: frames 447

Are There Social or Cultural Differences in the Value of Choice?

Barry Schwartz, Hazel Markus, Alana Connor Snibbe, and Nicole Stephens
describe a series of experiments concerning whether people of different socio-
economic status placed different values on choice.75 Among their conclusions:

• For middle-class respondents, “choice” meant “freedom,” “action,” and

“control”; for working-class respondents, choice was connected with “fear,”
“doubt,” and “difficulty.”
• While middle-class respondents preferred items that they had chosen
themselves, working-class respondents were more satisfied with choices
that others made for them.
• When a neighbor bought the same car they owned, middle-class respon-
dents were upset because it undercut the uniqueness of their choice, while
lower-class respondents were pleased that it affirmed that they had made a
good choice.

Iyengar and Lepper refer to a body of research that suggests:

Both preferences for, and the benefits of, choice might well vary across differ-
ent cultural contexts. . . . [W]hereas individual agency is an essential element
of the self-constructs of American individualists, it may be considerably less
relevant to the self-constructs of members of more collectivistic cultures,
characteristic of Asia and elsewhere.
Westerners . . . [may] possess a model of the self as fundamentally inde-
pendent. Such individuals strive for personal independence, desire a sense of
autonomy, and seek to express their internal attributes in order to establish
their uniqueness from others within their environments. . . .
By contrast, members of non-Western cultures . . . [may] possess an inter-
dependent model of the self and to strive for the super-ordinate goal of a
sense of interconnectedness and belongingness with one’s social in-group. . . .
Indeed, in some situations, the exercise of personal choice might even pose a
threat to individuals whose personal preferences could prove at variance with
of their reference group. Interdependent selves, therefore, might actually
prefer to have choices made for them, especially if the situation enables them

75

75. See Barry Schwartz, Hazel K. Markus, and Alana Connor Snibbe, Is Freedom Just
Another Word for Many Things to Buy, New York Times, Feb. 26, 2006. The SES-related
variable in the various studies included parents’ educational attainment and whether the
participants were in white- or blue-collar jobs.
448 problem solving, decision making, and professional judgment

to be both relieved of the “burden” associated with identifying the socially

sanctioned option and, at the same time, to fulfill the super-ordinate cultural
goal of belongingness. Consequently, for members of interdependent cul-
tures, it is not the exercise of choice that is necessary for intrinsic motivation,
but the perception of themselves as having fulfilled their duties and obliga-
tions toward their reference groups.76

14.5.2.b The Downsides of Keeping Your Options Open after a Decision In

Chapter 3 we suggested that when you lack the information to make a well-
informed choice, a useful decision strategy may be to keep your options open.
However, there is evidence that, largely as a result of loss-aversion, “decision
makers overvalue their options and are willing to overinvest to keep those options
from disappearing.”76 77

Moreover, in some circumstances, keeping one’s options open after making a

decision can exact a psychological cost. An obvious example is someone who
chooses a romantic partner but keeps an eye out for someone better. The lack of
a whole-hearted commitment is likely to detract from his or her satisfaction with
the relationship. Even if the partner is not ideal in every respect—and who is?—
the phenomenon of adaptation (described in Section 12.1.5) tends to lead to
satisfaction over time.
The same phenomenon applies to consumer choices. For example, students
who were offered a gift of either of two art posters and were told that they
could exchange it any time in the next month were less satisfied—even minutes
later—than those who were told that it was not exchangeable. In similar studies,
the satisfaction of those stuck with their choice stabilized over time while the
dissatisfaction of the former group lingered. (However, most people mispredict
that they would be happier having the option to change their minds.78)
14.5.2.c The Effect of Options on Whether to Decide Now or Defer Suppose
you are considering buying a CD player and have not yet decided what model

76. Iyengar and Lepper, supra, citing Hazel Markus and Shinobu Kitayama, Culture
and the Self: Implications for Cognition, Emotion, and Motivation, 98 Psychological Review
224–53 (1991).
77. Jiwoong Shin and Dan Ariely, Keeping Doors Open: The Effect of Unavailability on
Incentives to Keep Options Viable, 50 Mgmt. Sci. 575, 575 (2004).
78. Daniel T. Gilbert and Jane E. J. Ebert, Decisions and Revisions: The Affective Forecasting
of Changeable Options, 82 Journal of Personality and Social Psychology 503 (2002).
 complexities of decision making: frames 449

to buy. You pass by a store that is having a one-day clearance sale. Consider these
scenarios and the percentage of responses:79
Low conflict: The store offers a popular Sony player for $99—well below the
list price. Do you:
Buy the Sony player ($99) 66%
Wait until you learn more about the model 34%
High conflict: The store offers a popular Sony player for $99 and a top-of-the-line
AIWA player for $169—both well below the list price. Do you:
Buy the AIWA player ($169) 27%
Buy the Sony player ($99) 27%
Wait until you learn more about the models 46%
While the first decision seems easy, the second creates a mental conflict,
which can be avoided by deferring. But choosing in one case and deferring in the
other does not make sense in terms of a value-maximizing approach to decision
making, for the decision maker has precisely the same information about the
Sony player in both cases. But the phenomenon is easy to understand in terms
of reason-based decision making. While there are compelling arguments for
choosing the Sony when it is the only attractive product one is considering,
“choosing one out of two competing alternatives can be difficult: the mere fact
that an alternative is attractive may not in itself provide a compelling reason for
its selection, because the other option may be equally attractive. The addition of
an alternative may thus make the decision harder to justify, and increase the
tendency to defer the decision.”80
The phenomenon applies to public policy as well as personal decisions. In a
study, legislators were asked to recommend whether to close a hospital. Hospital
A was a 100-bed community hospital providing 200 jobs in a small urban area
which was also served by two other hospitals, which had a higher quality of care
and lower costs. Hospital B was a 250-bed teaching hospital providing 500 jobs
in a large urban area served by three other hospitals, which also had a higher
quality of care and lower costs. In both cases, the legislators were told that they
could also make “no recommendation.” When asked only whether to close
Hospital A, only 26 percent declined to make a recommendation. When asked
whether to close either Hospital A or Hospital B were options, 64 percent
declined to make a recommendation.81
The obvious solution to avoiding these odd changes of preferences is to com-
pare all three options (including the status quo of doing nothing) using a value-
based approach like the subjective linear model.

79. Amos Tversky and Eldar Shafir, Choice under Conflict: The Dynamics of Deferred
Decision, 3 Psychological Science 358–61 (1992).
80. Ibid.
81. Donald A. Redelmeier and Eldar Shafir, Medical Decision Making in Situations that
Offer Multiple Alternatives, 274 Journal of the American Medical Association 302 (1995).
450 problem solving, decision making, and professional judgment

14.5.3 Context Dependence

14.5.3.a Trade-off Contrast, or Asymmetrical Dominance Suppose that a
store offers a popular Sony player for $99—well below list price—and an obvi-
ously inferior Panasonic player for the regular list price of $105. Do you:
Buy the Panasonic player 3%
Buy the Sony player 73%
Wait until you learn more about the various models 24%
Here, rather than detracting from the reasons for buying the premium Sony
player, the existence of the inferior alternative highlights the reasons for buying
the Sony rather than deferring the purchase to learn more. The trade-off between
the cost and quality of the Panasonic and Sony is so great that it make the Sony
look all the more attractive.
The phenomenon where the same option is evaluated more favorably in the
presence of similar inferior options than in their absence is called trade-off con-
trast or asymmetrical dominance, reflecting the fact that the chosen option domi-
nates (or in other examples, nearly dominates) the similar option that is not
chosen. Itamar Simonson and Amos Tversky liken the phenomenon to percep-
tual contrast effects: “The same circle appears large when surrounded by small
circles and small when surrounded by large ones. Similarly, the same product
may appear attractive on the background of less attractive alternatives and unat-
tractive on the background of more attractive alternatives.”82 In a classic study,
they asked participants randomly assigned to the two-option group to choose
between $6 and a Cross pen, and asked participants in the three-option group to
choose among $6, the Cross pen, and an obviously inferior pen. While only 36
percent of participants in the two-option group chose the Cross pen, this rose to
46 percent in the three-option group.83
Of course, the presence of options sometimes provides information that con-
tributes to value-maximizing decision making. Mark Kelman et al. suggest that
if you are ordering from a restaurant menu and you prefer chicken over pasta,
the presence of fish on the menu should not alter your preference, but that you
might reverse your preference upon learning that the restaurant also serves veal
parmesan, since this may suggest that the restaurant specializes in Italian food.84
But in the examples in this subsection, it is difficult to discern the informational
value of the added options. In these cases, trade-off contrast is inconsistent with
principles of rational decision making.

82. Itmar Simonson and Amos Tversky, Choice in Contrast: Tradeoff Contrast and
Extremeness Aversion, 29 Journal of Marketing Research 281 (1992).
83. Simonson and Tversky. Only 2 percent of participants chose the inferior pen, thus
confirming its inferiority.
84. Mark Kelman, Yuval Rottenstreich, and Amos Tversky, Context-Dependence in
Legal Decision Making, 25 Journal of Legal Studies 287 (1996).
 complexities of decision making: frames 451

Kelman and his colleagues conducted a number of experiments applying the

contrast effect to legal settlements.85 In one, participants were placed in the role
of a lawyer representing a woman faculty member in the Economics Department
who was not promoted to tenure, and claims sex discrimination. The client is
interested in being compensated for the injury and in having the university pub-
licly admit guilt in her case. She is also interested in pressing for affirmative
action for women in the Economics Department. The university’s lawyer has
offered the plaintiff a choice of settlement packages.
One group of participants faced an alternative between two settlement offers.
The percentage of those who accepted each is shown:
• Affirmative action plan; no damages 50%
• Public admission plus $45,000 damages 50%
Another group was given a choice between these two plus a third, which was
similar to the second choice, but obviously inferior to it:
• Affirmative action plan; no damages 20%
• Public admission plus $45,000 damages 74%
• Public admission plus $35,000 gift to charity of client’s choice 6%
The change of preferences exemplifies the contrast effect—that an option is
seen as more valuable in the presence of a clearly inferior alternative.
14.5.3.b Extremeness Aversion, or the Compromise Effect People often eval-
uate the same option more favorably when it is seen as intermediate in the set of
options under consideration than when it is extreme. For example, when
Simonson and Tversky asked one group to evaluate the attractiveness of a mid-
level Minolta camera and a low-end camera in terms of their features and price,
50 percent chose each camera. When a third, high-end camera was added to the
options, the number choosing the mid-level Minolta camera rose to 72 percent.
Introduction of the high-extreme option reduced the “market share” of the low
extreme option, but not of the intermediate option.
The phenomenon can be explained in terms of loss aversion: if the disadvan-
tages of the two extremes—high cost in one case, low quality in the other—loom
larger than their advantages, this tends to favor the intermediate option, which
has only small disadvantages compared to either one.86 It is also a feature of
reason-based decision making, where a “compromise” often seems easier to jus-
tify than either extreme.
Kelman et al. tested the compromise effect in decisions involving the grading
of homicides. In one scenario, the defendant admitted poisoning her husband,
causing a protracted and painful death. She did this after she overheard her

85. Id.
86. Simonson and Tversky.
452 problem solving, decision making, and professional judgment

seventeen-year-old daughter from her first marriage say that the deceased had,
once again, tried to sexually molest her. The evidence also showed that the defen-
dant stood to inherit a large amount of money from her husband and that she
had been involved with another man for the last half year. One group of partici-
pants was given the choice (that jurors would have had in the District of Columbia)
between convicting her of manslaughter (homicide under extreme emotional
disturbance) or murder. The results:
• Manslaughter (extreme emotional disturbance) 47%
• Murder 53%
A second group was given the choice (that jurors would have had in California)
among convicting her of manslaughter, murder, or special circumstances
murder (committed for financial gain, or in an exceptionally heinous fashion, or
by the administration of poison). The results:
• Manslaughter (extreme emotional disturbance) 19%
• Murder 39%
• Special circumstances (financial gain, heinous, poison) 42%
But if the defendant was acting under emotional distress, she didn’t commit
murder at all, with or without special circumstances. As the authors note, the
choice between manslaughter on the one hand, and the murder verdicts on the
other, should not be sensitive to whether, if the defendant were acting deliber-
ately, she killed for financial gain, etc.
Another scenario involved the intentional killing of an off-duty police officer
working as a security guard at a shopping mall by an African American man he
suspected of burglary. When the defendant refused to be patted down, the guard
grabbed him, the guard shouted a racist epithet, and the defendant shot him.
Participants were informed that there were four possible verdicts:
• special circumstances murder—for the killing of a police officer acting in
the line of duty;
• murder;
• voluntary manslaughter—if the defendant was provoked or acting under
extreme emotional disturbance; or
• involuntary manslaughter—if the defendant subjectively, but unreason-
ably, believed he was entitled to defend himself.
Participants in one group were told that the judge had ruled, as a matter of
law, that there was no evidence that the defendant believed he was acting in self-
defense, so that they could not treat the homicide as involuntary manslaughter.
The results were:
• Special circumstances murder 12%
• Murder 57%
 complexities of decision making: frames 453

• Voluntary manslaughter 31%

• Involuntary manslaughter
Participants in another group were told that the judge had ruled, as a matter
of law, that a police officer working as a security guard is not acting in the line of
duty, so that they could not treat the homicide as special circumstances murder.
The results:
• Special circumstances murder
• Murder 39%
• Voluntary manslaughter 55%
• Involuntary manslaughter 6%
In analyzing this case, the authors explain:
It seems reasonable to assume . . . that the options are naturally ordered in
terms of their “severity.” (Special circumstances murder is more severe than
murder, which is more severe than voluntary manslaughter, which is more
severe than involuntary manslaughter.) Consider any subset consisting of
three of the options. Preferences satisfy betweenness if someone who prefers
one of the extreme options in the subset over the intermediate option in the
subset is more likely to prefer the intermediate option over the other extreme
option than someone who does not. In the [shopping-mall guard] case,
betweenness is natural: someone preferring special circumstances murder to
murder would almost surely prefer murder to voluntary manslaughter.
Someone preferring murder to voluntary manslaughter would almost surely
prefer voluntary manslaughter to involuntary manslaughter.
Given betweenness, context-independence requires that the ratio of
murder verdicts to voluntary manslaughter verdicts be higher in the second
group than the first. To see why, take choices in the first group as given and
consider how these choices should be translated into the second group.
Anyone choosing special circumstances murder in the first group should
select murder. Any movement to involuntary manslaughter should come
from voluntary manslaughter. Thus, as we move from the first to the second
group, the number of murder verdicts should rise, and the number of man-
slaughter verdicts should decline.
14.5.3.c Do Context-Dependent Decisions Result in Suboptimal Decisions?
Do context-dependent decisions cause harm or result in suboptimal decisions?
In the case of grading homicides, it seems to conflict with the public policy
underlying the grading scheme. Kelman et al. write:
Once we declare, for instance, that the purpose of differentiating murder
from manslaughter is (for instance) to show a certain level of mercy to those
acting in atypically stressful circumstances, then we have failed to meet that
stated goal if we differentiate defendants on some other basis (for example,
454 problem solving, decision making, and professional judgment

the presence of an option in which we condemn murderers of policemen

more than ordinary murderers). . . .
What about individual decision making? Consider that marketers frequently
use the phenomenon of context-dependent decisions to their advantage—for
example by showing the customer an item that is obviously inferior to the one
they hope he or she will buy. The social psychologist Robert Cialdini writes:87
I accompanied a salesman [Phil] on a weekend of showing houses to prospec-
tive home buyers. . . . One thing I quickly noticed was that whenever Phil
began showing a new set of customers potential buys, he would start with a
couple of undesirable houses. I asked him about it, and he laughed. They
were what he called “setup” properties. The company maintained a run-down
house or two on its list at inflated prices. These houses were not intended to
be sold to customers but only to be shown to them, so that the genuine prop-
erties in the company’s inventory would benefit from the comparison. . . . He
said he like to watch prospects’ “eyes light up” when he showed the places he
really wanted to sell them after they had seen the rundown houses. “The
house I got them spotted for looks really great after they’ve first looked at a
couple of dumps.”88

Williams-Sonoma used to offer one bread-baking appliance for $275. Subsequently,

the store added a larger appliance, similar to the first, but costing $429. There
were few sales of the more expensive item, but sales of the original appliance
doubled.88

With respect to personal decisions, Kelman et al. write:

Because consumers do not have explicit policies or goals, it is less clear in
what sense differential evaluation of options on the basis of context hinders
consumers’ interests. The notion that the consumer harms herself by breaches
of context-independence is grounded in two observations. First, we suspect
that the consumer herself would be prone to reevaluate (if not alter) her deci-
sion if she became aware of the fact that she made it on the basis of nonproba-
tive facts. . . . Second, to the extent that people do not have a stable,
context-independent preference order, their choices can be manipulated by
the composition of the set of options under consideration. As we have noted,

87. Robert B. Cialdini, Influence: Science and Practice 14–16 (Needham

Heights, MA: Allyn and Bacon, 4th ed. 2001).
88. Steven Pearlstein, The Compromise Effect, Washington Post, Jan. 27, 2002. Quoted
in Guthrie, Options.
 complexities of decision making: frames 455

such manipulations are common in the marketplace. They suggest that harm
befalls context-dependent consumers since someone manipulated into choos-
ing the option favored by another party, with her own set of interests, is less
likely to maximize his own well-being.
Although there is reason for skepticism whether people can be educated to
avoid context-dependent decisions, it seems plausible that eliminating explicit
irrelevant options will reduce context-dependent decisions. How and when that
can be done depends on, well, the context.

14.6 cost-benefit analysis (continued): valuing environmental

and other public goods

In Section 12.3, we introduced cost-benefit analysis as an application of expected

utility theory. We deferred the problem of how to value the existence and option
values of environmental goods until we could cover some additional material,
which we have now done.
One valuation technique involves revealed preference technique, in which con-
sumers’ preferences and valuations of certain goods can be ascertained by an
analysis of consumption behavior in the face of multiple purchasing choices. To
use revealed preferences to determine how much people value a residence in the
San Francisco Bay Area with a view of the bay, one would look the real estate
prices properties of this sort (controlling for other important variables, such as
lot size).
Policy makers have sometimes used this technique to determine how much
consumers and employees are willing to pay for product and workplace safety.
Could you use a similar technique to assessing the public’s valuation of, say,
Yosemite National Park? Perhaps you could use visitors’ traveling costs as their
revealed willingness to pay (WTP) for the National Park (controlling for variables
such as the distance traveled). But the assumption that a decision not to “con-
sume” (i.e., visit the park) implies zero value is hardly obvious. What other weak-
nesses do you see in this approach?
Perhaps because of the difficulty of measuring revealed preferences for public
goods, policy makers often use contingent valuation, which relies on people’s
expressed preferences for how much they would be willing to pay for, say, the pres-
ervation of a natural resource. Contingent valuation often produces very differ-
ent results than revealed preference methods, and it is subject to a number of
criticisms. Because respondents are not “playing with real money”89 they may
overestimate their WTP for a particular outcome. Respondents also may enter

89. Richard Revesz and Michael Livermore, Retaking Rationality: How Cost
Benefit Analysis Can Better Protect the Environment and Our Health, 127
(New York, Oxford University Press, 2008).
456 problem solving, decision making, and professional judgment

very large or small “protest values” to indicate their opposition to a certain

policy.90 Some other weaknesses of contingent valuation include:
• When asked about their WTP to protect against injuries, healthy respon-
dents may underestimate the extent that people adapt to disabling events.
• As prospect theory suggestions, people’s valuation of things—whether
tangible items, or health or environmental goods such as clean air—may
be quite different depending on whether they already possess the benefit or
just hope to acquire it.
• Contingent valuation surveys ask respondents to price rather than choose
things. But we have seen that people make different assessments when
values are elicited by one means or the other. (See Section 13.3.2.b) In an
experiment by Julie Irwin, Paul Slovic, and their colleagues, an improve-
ment in air quality was valued more highly in 86 percent of direct choices
but only in 40 percent of choices inferred from respondents’ WTP,91 rais-
ing the question of which method elicited the “correct” value.
• Related to the phenomenon of psychophysical numbing described in
Section 12.1.4.c, WTP values often do not account for the scope of the
issue: for example, in one study, respondents were not willing to pay sub-
stantially more “to save 200,000 birds rather than 2,000 birds” from an oil
spill.92 WTP may be strongly influenced by their affective response to the
image of a bird dying, without regard to the numbers involved.93
• People are not good at distinguishing between, say, a 1/100,000 probability
and a 1/1 million chance of a bad event occurrig.

90. Id.
91. Irwin et al., supra.
92. Daniel Kahneman and Shane Frederick, A Model of Heuristic Judgment, in Keith
Holyoak and Robert Morrison eds., Cambridge Handbook of Heuristic Reasoning
(New York: Cambridge University Press, 2005).
93. See Daniel Kahneman, Ilana Ritov, and David Schkade, Economic Preferences or
Attitude Expressions? An Analysis of Dollar Responses to Public Issues, in Choices, Values,
and Frames 642–672 (D. Kahneman and A. Tversky eds., New York: Cambridge
University Press, 2000); Christopher K. Hsee and Yuval Rottenstreich, Music, Pandas, and
Muggers: On the Affective Psychology of Value, 133 J. Experimental Psychol.: Gen. 23,
23–24 (2004).
 complexities of decision making: frames 457

In 1995, a panel of economists, headed by Kenneth Arrow and Robert Solow

under the auspices of the National Oceanic and Atmospheric Administration
(NOAA), recommended methods for improving the accuracy of this inherently
imprecise survey technique. Among other things, the panel recommended that
respondents be asked to vote for or against a specific tax to protect a particular
resource, and that they be given detailed information about its ecological impor-
tance, the threats to which it was vulnerable, and possible outcomes of measures
to protect it.94
Even with the procedural reforms recommend by the Arrow-Solow panel,
contingent valuation is very inaccurate and vulnerable to biases. Nonetheless, it
is widely used for valuing natural resources and other benefits that are poorly
reflected in markets.95

94. http://en.wikipedia.org/wiki/Contingent_valuation.
95. For a discussion of alternative procedures for assessing people’s values for non-
market goods, see Jonathan Baron, Thinking and Deciding (4th ed. Cambridge, UK:
Cambridge University Press, 2000).
This page intentionally left blank
15. decision making under risk

In this chapter, we examine decision making where risk or uncertainty is a major

factor. The goal here, as earlier, is to maximize the decision maker’s own goals
or values, or those of her clients or constituents, recognizing that individuals
may have very different goals or values and that they also may have different
attitudes about taking risks. Paralleling Chapters 4 and 12, we begin by laying
out a model of rational decision making under risk; then (as in the immediately
preceding chapters), we examine factors that conduce to departures from
rationality.
For most of the chapter we will use the terms risk to refer to decisions where
the outcome is less than certain and where the probability of its occurring can be
specified or at least approximated as a value greater than 0 and less than 1. Later
we will discuss situations where, for all practical purposes, the decision maker
has no idea what the probability is—situations that economists term uncertainty
and some psychologists term ambiguity.

15.1 expected utility in decision making under risk

Section 12.2.2 introduced the concept of expected utility theory in situations

where risk was not a major issue. Where risk is an issue, expected utility is
described by the equation:

EU(E)= U(E)×P(E)
EU(E) is the expected utility of event E. U(E) is the subjective utility of the event.
P(E) is the probability of the event’s occurrence.
We begin with the simplifying assumption that the decision maker is risk
neutral. This means that the decision maker is indifferent between a payoff with
certainty and a gamble with the same expected payoff, for example, between
• a 100 percent chance to get $100, or
• a 50 percent chance to get $200 and a 50 percent chance to get nothing.
If a decision maker is risk neutral, then the expected utility (U(E)) of an event is
exactly the same as the expected value (V(E)) of the event—that is, its monetary (or
monetizable) value. Later, we address the fact that decision makers are often not
risk neutral, but may have different preferences or tolerances for risks in differ-
ent circumstances. Under these circumstances, expected utility and expected
value will diverge.
460 problem solving, decision making, and professional judgment

The axioms of rational decision making set out in Section 12.2.2 apply here as
well. We shall discuss their application to decision making in conditions of risk and
introduce some new axioms that bear on particular issues presented in this chapter.
Expected utility under the probability axiom. Underlying all the problems in this
chapter is the assumption—actually, another axiom of rational decision mak-
ing—that one can assign a probability to every possible outcome. In this context,
a probability is simply a degree of belief that the outcome will occur. (See Section
5.1.2b) Sometimes the decision maker may arrive at the probability through the
statistical analysis of a large sample. More often, however, she must use a sub-
jectivist, Bayesian approach and, particularly in the case of one-time decisions,
she can only hazard a guess about the probability. Be that as it may, every out-
come, in principle, has a probability somewhere between 0.0 and 1.0.
With the probability axiom as background, the essential assumptions about
completeness, dominance, and trade-offs that we discussed in Section 12.2.2 apply
to decisions under risk. To evaluate alternatives, you multiply the utility of each
possible outcome by the probability of its occurring to determine the expected
utility of that alternative. If the expected utility of alternative number 1 is greater
than that of alternative number 2, then the former dominates the other. Let’s
look at some decision tools that help guide decisions under risk.

15.2 dependency diagrams and decision trees

Many decisions under conditions of risk involve causal chains that are not read-
ily modeled by a matrix of the sort that Christine Lamm used for the wastewater
treatment plant decision in Section 4.5. Two conventional means of modeling
them are dependency diagrams and decision trees. These tools actually serve two
purposes. First, quite apart from any quantitative analysis, they can assist in
structuring a decision—in laying out the sequence of the decision, identifying
uncertainties, specifying relationships between issues, and describing the out-
comes associated with different scenarios. Second, if one attaches probabilities
and values to the various nodes, they can compute expected value.
We begin this section with very simple examples showing how a dependency
diagram (also called an influence diagram) can help structure a decision and how a
decision tree can be used to compute expected value. (Actually, the dependency
diagram is an even more powerful computational tool than the decision tree, but the
computations are less transparent to the user.) A dependency diagram shows causal
links where an event or decision affects another event, decision, or outcome.

15.2.1 Dependency Diagrams

Figure 15.1 shows every variable relevant to your decision whether or not to carry
an umbrella on a day when it might rain. For purposes of this example, we treat
(1) rain and (2) not-rain (or sun, or shine) like two sides of a coin, as mutually
exclusive and collectively exhaustive (MECE).
 decision making under risk 461

NWS Forecast Rain or sun

if rain is likely
carry umbrella

Carry umbrella? Satisfied

figure 15.1 dependency diagram for carrying an umbrella based on

NWS forecast.

The differently shaped objects are called nodes. Your satisfaction or utility
(diamond-shaped node) depends on the relationship of two variables: your deci-
sion to carry (or not carry) an umbrella (rectangular node) and whether it rains
(oval node). You will be
• very dissatisfied if it rains and you do not have an umbrella, and
• mildly dissatisfied if you carry the umbrella around unnecessarily.
Your decision whether to carry an umbrella is based on the National Weather
Service (NWS) forecast.
The arcs, or lines, of the dependency diagram show the relations among these
elements. The presence of an arc indicates that there is some possibility that the
node pointed to is dependent on the node from which the arc originates. The
absence of an arc is, if anything, more informative than its presence: its absence
asserts that the nodes that are not connected are, in fact, independent. There is
no arc between the forecast and your satisfaction, for the only role of the forecast
is to inform your decision whether to carry an umbrella (which will affect your
satisfaction).
Were you to make the decision based on a combination of the NWS forecast
and your own forecast, the dependency diagram might look like Figure 15.2.

past weather influences your and NWS forecasts

NWS Forecast past weather

Your Forecast
if rain is likely carry
umbrella Rain or sun

Carry umbrella? Satisfied

figure 15.2 dependency diagram for carrying an umbrella based on

NWS and your own forecast.
462 problem solving, decision making, and professional judgment

This shows that both the NWS weather forecast and your own forecast rely on
what the weather has been in the past. It shows that you take account of the NWS
forecast, but that NWS doesn’t take account of yours. Since today’s actual weather
(rain or sun) also depends on past weather, we draw those arcs as well.
Even without using a dependency diagram quantitatively, it has value as a
checklist of all the factors that should affect your decision—and equally important
it identifies factors present on the landscape that do not affect your decision.

15.2.2 Decision Trees

A decision tree represents the decision and its consequences, and makes
the choice and outcome alternatives explicit. The square node in the tree in
Figure 15.3 indicates your choices. The circular nodes indicate the uncertainties
that attend each choice.
rain
carry umbrella
sun

rain
don’t carry umbrella
sun

figure 15.3 the structure of a decision tree.

Both the decision nodes and the outcome nodes of a decision tree are mutu-
ally exclusive and collectively exhaustive (MECE). For example, we have posited
that there are only two possible outcomes—rain or sun—and if one occurs the
other cannot. But in fact, there are many other possible states of the weather,
ranging from snow, sleet, and hail to partly cloudy with occasional showers, and
we could represent all of them in the decision tree. A decision tree can have as
many outcomes as you wish—as many as may be relevant to the decision at
hand—but in aggregate they are MECE.
In a simple case like this one, the bare tree does not advance your understand-
ing of the decision. The tree’s power becomes more evident, however, when we
take a further step and add two sorts of numbers to the tree.
One set of numbers is the probability that it will (or will not) rain. Let us sup-
pose that the weather forecaster says there’s a 40 percent chance of rain, which
(following the conventions described in Chapter 5) we’ll write as 0.4. Because
rain and sun are MECE, the probability of sun is 0.6.
The second set of numbers indicates how satisfied or dissatisfied you will be with
the outcome. Let’s use a 100 point scale, and (since we’re talking mainly about dis-
satisfaction) we’ll treat 0 as your being perfectly happy if the sun shines and you
don’t have an umbrella, and –100 as being miserable if it rains and you don’t have
 decision making under risk 463

an umbrella. You don’t like either rain or carrying umbrellas. Each independently
gives you a (dis)utility of -30. So if you are carrying the umbrella and it’s a sunny day,
your utility is down -30, and if it rains it’s down -60. (See Figure 15.4.)
rain
−60
carry umbrella 0.4

sun
−30
0.6
rain
−100
don’t carry umbrella 0.4
sun
0
0.6

figure 15.4 the decision tree with utilities and probabilities.

For each branch of the tree, you calculate the expected utility by multiplying
the probability of rain (or shine) by the satisfaction of the outcome, and sum the
products as shown in Table 15.1.
table 15.1 solving the decision tree

Probability Outcome EU

Umbrella Rain 0.4 −60 −24

Sun 0.6 −30 −18

−42

No umbrella Rain 0.4 −100 −40

Sun 0.6 0 0

−40

The total expected utility of carrying an umbrella is -42, while the utility of not
carrying the umbrella is -40. What does the imply if you are risk neutral? Would
you in fact follow its implications?

15.2.3 A Problem: Newport Records

Let’s use a dependency diagram and decision tree to analyze a more realistic
problem than whether to carry an umbrella. Recall the case in Chapter 1, in
which Clyde Evers has sued Newport Records, a small recording company that
refuses to pay $600,000 due for accounting software that Evers customized and
installed. Newport Records says the software does not do what Evers said it would
do. After deciding not to seek summary judgment, Evers’ lawyers, Luis Trujillo
and his associate Anna Wilkins, are simultaneously considering settlement and
464 problem solving, decision making, and professional judgment

working through the scenario of a trial. After all, while Evers needs the money,
the amount for which he would settle—his “reservation price”—will be influ-
enced by what he could gain (or lose) by litigating the case.
In predicting the outcome of the trial, these factors seem relevant:
• If the case goes to trial, Evers’ costs of litigation will be $100,000.
• Evers made extravagant oral representations about how the software would
perform, and he acknowledges to the lawyers that its performance is not
entirely what he said it would be.
• The contract disclaims all warranties of performance. It also has a so-called
“integration clause,” which says that the written contract is the sole agree-
ment between the parties, notwithstanding any oral representations.
• The law requires that warranty disclaimers be in a particular typeface, some-
what different from the disclaimer in the contract with Newport Records.
The judge might or might not regard the difference as immaterial.
• If the judge finds that the disclaimer’s typeface deviates materially from the
statutory requirements, he will instruct the jury to find the disclaimer effective
only if the president of Newport Records actually read and understood it.
• If the disclaimer is ineffective, and notwithstanding the integration clause,
the judge may permit the jury—as part of its determination of whether the
software performed “as promised”— to hear about Evers’ oral representa-
tions, in order to resolve an alleged ambiguity in the contractual descrip-
tion of the software’s function.
Trujillo suggests that it would be helpful to model these factors in a depen-
dency diagram, to be sure to understand how they combine to affect the outcome
of the trial, as shown in Figure 15.5.
With the dependencies thus clear, the lawyers construct the skeleton of a deci-
sion tree. Although they may refine this analysis later, the applicable contract law
indicates that Evers will either obtain a judgment for the $600,000 due under the
contract, or get nothing. This is shown by the outcome values on the right side
(see Figure 15.6). (Once we solve for the case on its merits, we will take into
account the $100,000 costs of litigating the case.)
Trujillo and Wilkins then discuss how to determine the likelihood of Evers’
prevailing (or losing). Rather than leap immediately to guessing probabilities, they
ask what reasons would lead the judge or jury to decide for or against Evers on
each issue. For example, whether the written disclaimer meets the statutory
requirements is a question of law to be decided by the judge, and they consider
the reasons why the judge might decide one way or the other, as shown in
Table 15.2.
If the judge determines that the disclaimer did not meet the statutory require-
ments because it was in the wrong typeface, the jury would determine whether
the president of Newport Records nonetheless understood that any warranty was
being disclaimed, as shown in Table 15.3.
 outcome if
we litigate?

if yes, $600K how much will

will we win?
jury award?
if yes, yes
if yes, yes if no also, no

in light of the use of

an incorrect typeface, if no will jury find our
is our disclaimer of software performed
any warranty of as promised?
performance effective?
if yes, yes if yes, yes if allowed, hurts

decision making under risk 465

if no also, no if excluded, helps

will judge find the in light of the constract’s

will jury find defendant “integration clause,”
use of a different if no nonetheless read and
typeface to be will judge allow or exclude
understood that any Evers’ oral representations
inconsequential warranty of performance
“as a matter regarding performance?
was being disclaimed?
of law”?

figure 15.5 dependency diagram for newport records.

 466 problem solving, decision making, and professional judgment
judge finds our
warranty disclaimer
to be effective,
as a matter of law
{1} $600,000
jury finds defendant
read and understood
that any warranty of
performance was
IF LITIGATE being disclaimed
{2} $600,000
judge finds this in light of the contract’s
is an issue for the “integration clause,” jury finds software
jury, in light of judge excludes Evers’ performed as promised
incorrect typeface oral representations {3} $600,000
regarding performance
jury finds defendant did jury finds software did
not read or understand not perform as promised
the warranty disclaimer, {4} $0
so disclaimer not effective despite the contract’s
“integration clause.” jury finds software
judge allows jury to performed as promised
hear about Ever’s {5} $600,000
oral representations
jury finds software did
not perform as promised
{6} $0

figure 15.6 decision tree for newport records.

 decision making under risk 467

table 15.2 is the written disclaimer binding?

Reasons why the judge would find Reasons why the judge would find the
the written disclaimer does not meet written disclaimer meets the statutory
the statutory requirements requirements

• The statute is unequivocal about • The typeface used stands out as

the required typeface. much as that required by the stat-
• In the only litigated cases in the ute.
state, courts have held the seller • The litigated cases all involved con-
to strict standards. sumers rather than commercial
• Although not legally relevant, the purchasers. The buyer is a business
judge may be moved by knowl- and does not need same degree of
edge of Evers’ extravagant oral protection as an ordinary con-
representations about the soft- sumer.
ware. • The whole point of a written dis-
claimer and an integration clause is
to avoid litigating over alleged rep-
resentations.

After listing the pros and cons on the remaining issues, Trujillo and Wilkins
now begin to think hard about probabilities.1 They begin with qualitative judg-
ments, such as “likely” or “very likely,” but eventually realize they need to trans-
late these into numbers.
15.2.3.a Estimating probabilities. It is easy to estimate the probability of an
event’s occurring when there are sufficient instances that you can apply frequen-
tist statistics along the lines introduced in Chapters 5 and 6. But many decisions
faced by lawyers and policy makers are one-time events, susceptible only to a
subjective estimate. Trujillo’s and Wilkins’s strategy of articulating the reasons
why a particular outcome might or might not occur is often helpful background.
Here are some techniques—by no means mutually exclusive—for use in various
circumstances:2

1. See Marc B. Victor et al., Evaluating Legal Risks and Costs with Decision Tree Analysis
in Successful Partnering Between Inside and Outside Counsel ch. 12 (West Group
& ACCA 2000–2007).
2. See Douglas A. Wiegmann, “Developing a Methodology for Eliciting Subjective
Probability Estimates During Expert Evaluations of Safety Interventions: Application for
Bayesian Belief Networks,” final technical report, AHFD-05-13/NASA-05-4 (October 2005) 9
(Aviation Human Factors Division Institute of Aviation, University of Illinois at Urbana-
Champaign), http://www.humanfactors.uiuc.edu/Reports&PapersPDFs/TechReport/05-13.
pdf#search=%22probability%20wheel%22.
468 problem solving, decision making, and professional judgment

table 15.3 did the plaintiff understand the written disclaimer?

Reasons why the jury would find Reasons why the jury would find that
that Newport did not understand Newport understood the disclaimer
the disclaimer

• The disclaimer is written in • The language of the disclaimer is stan-

legal jargon. dard, and found in many contracts.
• Newport’s president negoti- • The typeface used stands out as much
ated the agreement himself, as that required by the statute.
without a lawyer. • Newport’s president is highly educated,
• The disclaimer appears near with a graduate degree, and will come
the end of a 30-page contract, across to the jury as such.
which most people would not • Newport’s president initialed every
have the patience to wade page.
through.

• Ask how many times out of one hundred you would expect an event to
occur.
• Choose between a certain payoff (e.g., a settlement) or a gamble (litigate)
where the payoff depends on the probability in question, and adjust the
amount of the certain payoff until you are indifferent between the two
choices.
Of course, these elicitation methods depend on your actually having a (rea-
sonably good) estimate of the probability in your head. And they are subject to
the cognitive biases discussed in earlier chapters.
Trujillo and Wilkins are aware of the hazards of overoptimism and confirma-
tion bias (discussed earlier in the book). While they believe that their profes-
sional distance creates a degree of immunity from these biases, Wilkins suggests
that they err on the side of conservatism. But Trujillo responds that they would
do best just to be as realistic as possible—to avoid undervaluing Evers’ lawsuit.
(If Evers wants to be conservative, he can adjust his settlement position accord-
ingly.) The lawyers end up filling in the probabilities, as shown in Figure 15.7.
Finally, they roll back, or solve, the tree. The process involves multiplying the
outcome of each branch of the tree by the probabilities of achieving it, and adding
up the results. Go from top to bottom and right to left in Table 15.4.
Rather than do this manually, or even using Microsoft Excel, they use dedi-
cated software—in this case, a program called TreeAge, which happily performs
just as warranted. (See Figure 15.8.)
Based on their best estimates of the probability of winning on each issue, the
expected value of the case is $420,000. But it will cost $100,000 to litigate, so the
net value is reduced to $320,000.
 judge finds our
warranty disclaimer
to be effective,
as a matter of law
{1} $600,000
.20
jury finds defendant
read and understood
that any warranty of
performance was
IF LITIGATE being disclaimed
{2} $600,000
.10
judge finds this in light of the contract’s
is an issue for the “integration clause,” jury finds software
jury, in light of judge excludes Evers’ performed as promised
incorrect typeface oral representations {3} $600,000
2/3
regarding performance
.80
jury finds defendant did jury finds software did
.75
not read or understand not perform as promised
the warranty disclaimer, {4} $0
1/3

decision making under risk 469

so disclaimer not effective despite the contract’s
“integration clause.” jury finds software
.90
judge allows jury to performed as promised
hear about Ever’s {5} $600,000
1/3
oral representations
jury finds software did
.25
not perform as promised
{6} $0
2/3

figure 15.7 newport records decision tree with probabilities.

470 problem solving, decision making, and professional judgment

table 15.4 solving the newport records decision tree by hand

(1) 0.20 = 0.20 x $600,000 = $120,000

(2) 0.10 x 0.80 = 0.08 x $600,000 = $48,000
(3) 0.67 x 0.75 x 0.90 x 0.80 = 0.36 x $600,000 = $216,000
(4) 0.33 x 0.75 x 0.90 x 0.80 = 0.18 x $0 = $0
(5) 0.33 x 0.25 x 0.90 x 0.80 = 0.06 x $600,000 = $36,000
(6) 0.67 x 0.25 x 0.90 x 0.80 = 0.12 x $0 = $0
$420,000

What size settlement should Evers be willing to accept? Assuming for the
moment that he is risk neutral (more about that later), he might take into account
these considerations:
• If the case goes to trial, it is not likely to be resolved for a couple of years.
But Evers is eager to have funds to start a new business.
• Also, any sum received today is worth more than the same amount received
a couple of years in the future.3 (See Section 13.6.1).
• Evers’ lawyers estimate that their fee for their engaging in settlement nego-
tiations will be $5,000.
• And, of course, it’s possible that Newport Records will not agree to settle
for anything that Evers deems acceptable.
All things considered, Evers and his lawyers conclude that a settlement of
about $300,000 would be acceptable.
15.2.3.b Sensitivity analysis. If the disclaimer fails to meet the statutory
requirements, the evidence bearing on whether the president of Newport Records
read and understood the warranty disclaimer will be based on his deposition and
testimony at trial. Other than being shrewd cross-examiners, there is not much
that Evers’ lawyers can do to affect that.
But they may be able to affect the jury’s judgment about whether the software
performed as promised. Perhaps, if understood in the context of the reasonable
needs of a small business and prevailing standards for accounting systems, Evers
may not have egregiously overstated the capabilities of his product. But estab-
lishing the prevailing norms will require expert testimony. Based on preliminary
conversations, the expert they have in mind (a business school professor) would
likely conclude that under the prevailing standards the software materially met
the specifications. But she will charge $20,000 to do the analysis to establish this
and then testify.
Is it worth hiring the expert witness? That depends on how her testimony
will affect the value of the case. And that, in turn, depends on how much her

3. We put aside the possibilities and complications of prejudgment interest.

 judge finds our
warranty disclaimer
to be effective,
as a matter of law
{1} $600,000; P = 0.20
0.20
jury finds defendant
read and understood
that any warranty of
performance was
IF LITIGATE being disclaimed
$420,000
{2} $600,000; P = 0.08
0.10
judge finds this in light of the contract’s
is an issue for the “integration clause,” jury finds software
jury, in light of judge excludes Evers’ performed as promised
$375,000 {3} $600,000; P = 0.36
incorrect typeface oral representations
$400,000 0.67
regarding performance
0.80
jury finds defendant did jury finds software did
0.75
not read or understand not perform as promised
the warranty disclaimer, $350,000 {4} $0; P = 0.18
0.33

decision making under risk 471

so disclaimer not effective despite the contract’s
“integration clause.” jury finds software
0.90
judge allows jury to performed as promised
hear about Ever’s {5} $600,000; P = 0.06
$200,000 0.33
oral representations
jury finds software did
0.25
not perform as promised
{6} $0; P = 0.12
0.67

figure 15.8 solving the decision tree in newport records with a tree program.
472 problem solving, decision making, and professional judgment

testimony will increase the likelihood of persuading the jury on that issue. Right
now, the lawyers estimate this likelihood as being 0.67 if the judge excludes
Evers’ oral representations, and 0.33 if he allows them into evidence. The
lawyers believe that the expert’s testimony would increase the likelihood in the
former case to 0.80, and in the latter case to 0.20.
The decision tree makes it easy to determine whether it is worth the cost of
the expert witness. Solving the tree with these increased probabilities produces
an expected value of $470,400. Since this is an increase in case value of over
$50,000 for a cost of only $20,000, the investment in the expert witness would be
well worth her cost.
15.2.3.c The probability of different jury awards. We simplified the problem
by assuming that the jury would either award $600,000 or nothing. But it is easy
to imagine reasons that the jury would award Evers some amount in between.
For example, although he claimed that the software would do more than it did, it
performed reasonably well, and Newport Records was not justified in refusing to
pay anything—especially if it actually continued using the software. At this point,
you have the concepts and tools for factoring this in as well.

15.2.4 Of Garbage, Anchoring, and Decision Costs

A decision tree can be helpful in structuring and analyzing a problem, and may
help inform your intuitions about the likelihood of achieving objectives, even if
you don’t plug any numbers into it. However, the true usefulness of a decision
tree is its ability to calculate expected value—and with this come several
cautions.
First, the expected value calculation is only as good as the data and judgments
on which it relies. In some cases, such as weather forecasts, one can assign prob-
abilities with reasonable confidence. This is even true for some aspects of litiga-
tion; for example, there are databases that collect the size of jury verdicts in
certain types of tort cases in particular jurisdictions. However, one-time guesses,
like those in the Newport Records case, are more problematic. “Garbage in, gar-
bage out” applies not only to estimates of probabilities but to estimates of the
values of outcomes and the costs of pursuing them: simply put, you cannot
create a trustworthy decision tree without accurate estimates of probabilities and
of expected costs and benefits.
Second, the danger of treating garbage like a gourmet dish is enhanced by the
anchoring bias—our tendency to anchor on a number and insufficiently adjust
away from it. (See Section 10.1) We have seen that just being aware of the phe-
nomenon does not avoid being caught in its snare. This does not mean that you
should not use decision trees even when you do not have great confidence in the
numbers. But it does counsel caution.
Finally, we should note that constructing a decision tree requires great cogni-
tive effort. A decision tree is just a procedure for determining expected utility.
 decision making under risk 473

And as Gerd Gigerenzer notes, referring to Darwin’s decision process involving

whether to marry (Section 4.4.1):4
Imagine that Darwin had attempted to resolve his Question by maximizing
his subjective expected utility. To compare his personal expected utility for
marrying, he would have had to determine all of the possible consequences
that marriage could bring (e.g., children, constant companion, and an endless
stream of further possibilities not included in his short list), attach quantita-
tive probabilities to each of these consequences, estimate the subjective utility
of each consequence, multiply each utility by its associated probability, and
finally add all these numbers up. The same procedure would have to have
been repeated for the alterative “not marry.” Finally, he would have had to
chose the alternative with the higher total expected utility. To acquire reliable
information about the consequences and their probabilities, Darwin might
have had to invest years of research—time he could have spent studying bar-
nacles or writing Origin of Species.

figure 15.9 the last word on expected value analysis. dilbert © scott
adams/dist. by united feature syndicate, inc. reprinted by permission.

15.3 risk tolerance and utility functions

So far, we have assumed that Evers is risk neutral in approaching his decision
whether to litigate or settle the case against Newport Records. But this needn’t be
the case. Evers may prefer to accept a settlement lower than the expected value of
the litigation, either because he wants funds now to start his new venture or just
as a matter of personal preference. Or he might wish to take a gamble on the
litigation even if the expected value happens to be less than that of a settlement.
There are essentially three conditions of risk tolerance (or risk attitude, or risk
preference). We’ll illustrate these with a simple example.

4. Gerd Gigerenzer, Peter M. Todd, and the ABC Research Group, Simple
Heuristics That Make Us Smart 9 (New York: Oxford University Press, 1999).
474 problem solving, decision making, and professional judgment

Suppose you are offered the choice between a certain $100 and a 0.5 chance
of winning $200 and a 0.5 chance of winning nothing. (This kind of choice,
referred to as a “gamble” or “lottery,” is used to determine a variety of individual
preferences involving risk.)
• If you are indifferent between the choices, you are risk neutral.
• If you would demand more than a $200 payoff, or a greater than 0.5 chance
of winning, before you accepted the gamble, you are risk averse. The addi-
tional amount of payoff you would demand is the risk premium.
• If you would take the gamble for less than a $200 payoff, or a smaller than
a 0.5 chance of winning, you are risk seeking.
If you are risk neutral in this situation, your (subjective) expected utility of the
certain $100 or the .05 chance of winning $200 is identical to the (objective)
expected value. But if you are risk averse or risk seeking, your expected utility
will diverge from expected value. Let’s consider your utility function if you are
risk averse.
Recall Bernoulli’s observation about the declining (subjective) marginal value
of money (Section 12.1.4.b). The subjective value of increasing your wealth from
$100 to $200 is greater than the subjective value of increasing your wealth from
$1100 to $1200. A graph of this “utility function” would be downwardly concave.
Bernoulli’s curve describes risk aversion as well as the declining marginal
utility of money: If you prefer a sure gain of $100 (lower on the curve) over a 0.5
chance of $200 (higher on the curve), then you are willing to pay a bit more (and
sacrifice some expected value) for the certainty. The extra amount you are willing
to pay is the risk premium. Risk aversion and the declining marginal utility of
money are two sides of the same coin. When we describe the declining marginal
utility of money, we say that increments to an already large amount bring less
utility than the equivalent increments to a small amount. When we describe risk
aversion, we say that a bird in the hand is worth two in the bush.
The same person may have a different risk preference in different situations
(e.g., finance, health and safety, social situations, and recreation).5 “Individuals
who exhibit high levels of risk-taking behavior in one content area (e.g., bungee
jumpers taking recreational risks) can be risk averse in other domains (e.g.,
financial).”6
And different people may have different risk preferences in the same situation.
If Alex and Andrea exhibit different risk preferences for flying on airlines with
poor safety records, it could be because they have different perceptions of the risk

5. Ann-Renee Blais and Elke U. Weber, A Domain Specific Risk-Taking (DOSPERT)

Scale for Adult Populations, 1 Judgment and Decision-Making 33 (2006).
6. Hanoch, Joseph Johnson, and Andreas Wilke, Domain Specificity in Experimental
Measures and Participant Recruitment, An Application to Risk-Taking Behavior, 17
Psychological Science 300 (2006).
 decision making under risk 475

or of the benefits from taking the risk, or differences in their willingness to take
on a (perceived) risk.7 While risk taking may be affected by the perceived benefit
of the activity,8 differences in risk preferences often result from different percep-
tions.9 (For example, entrepreneurs and people who enter crowded markets
that offer small chances of financial success tend to be highly optimistic.10) Apart
from its academic interest, the reasons for differences in risk preferences may
have practical consequences. For example, if you are negotiating in a situation
where the allocation of risks between the parties is an issue, understanding and
addressing the source of the other party’s risk preferences may help reach an
agreement.11
In any event, one cannot, a priori, describe any of these risk preferences as
more or less rational than the others. In many situations, risk preferences are as
subjective, and hence as arational, as preferences for flavors of ice cream.
Sometimes they make perfect sense in terms of your specific goals: If risk seek-
ing behavior seems irrational to you, suppose that you are in Topeka, with no
money in your pocket, and eager to take the next flight to New York to attend a
friend’s birthday party that night. A plane ticket costs $150, and a benefactor
offers you a certain $100 or a 0.5 chance of winning $150.
Even if risk preferences often are arational, we will see later in the chapter
that they can be affected simply by the way choices are framed. In these cases,
there is at least cause for concern that the frame-influenced preferences do not
always serve an individual’s best interests.

7. Ann-Renee Blais and Elke U. Weber, Domain-Specificity and Gender Differences in

Decision Making, 6 Risk Decision and Policy 47 (2001).
8. Hanoch, Joseph Johnson, and Andreas Wilke, Domain Specificity in Experimental
Measures and Participant Recruitment, An Application to Risk-Taking Behavior, 17
Psychological Science 300 (2006).
9. Elke U. Weber, Personality and Risk Taking, in International Encyclopedia of
the Social and Behavioral Sciences 11274–76 (N. J. Smelser and P. B. Baltes eds.,
Oxford, UK: Elsevier Science Limited 2001).
10. Arnold Cooper, Carolyn Woo, William Dunkelberg, Entrepreneurs’ Perceived
Chances For Success, Journal of Business Venturing 97–108 (1988); Colin Camerer and
Dan Lovallo, Overconfidence and Excess Entry: An Experimental Approach, 89,
The American Economic Review 306–318 (1999); Gideon Markman, David Balkin,
Robert Baron. Inventors and new venture formation: The effects of general self-efficacy and
regretful thinking, Entrepreneurship Theory & Practice 149–165 (Winter, 2002).
11. See Elke U. Weber and Christopher Hsee, Cross-Cultural Differences in Risk
Perception, But Cross-Cultural Similarities in Attitudes Towards Perceived Risk, 44
Management Science 1205 (1998).
476 problem solving, decision making, and professional judgment

Policy Problem

Should policy makers be risk averse, risk seeking, or risk neutral? Does the answer
depend on the circumstances and, if so, what might they be?

15.4 the framing of risky decisions: prospect theory redux

In Section 14.1 we considered the status quo bias, including the endowment
effect, in decisions where risk is not at issue. Here, we return to the same
phenomenon in conditions of risk. In addition to the three features of prospect
theory noted above—
1. the value function is defined over gains and losses with respect to some
reference point;
2. individuals are loss averse (the slope of the curve for losses is considerable
steeper than the slope for gains);
3. both the gain and loss functions display diminishing sensitivity as one moves
away from the reference point;
—we examine two other features that are relevant to risky choices, particularly:
4. while the utility curve for gains with respect to the status quo is concave
(implying risk aversion), the utility curve for losses is convex (implying risk
seeking);
5. individuals overweight very low probabilities and underweight very high
probabilities, so that (reversing the phenomenon generally described by
the utility function) they tend to be risk seeking with respect to low probability
gains and risk averse with respect to unlikely losses.
The fact that the reference point separating gains and losses for a particular
decision is indeterminate and, indeed, manipulable, poses a severe problem for
expected utility theory. In what has become a classic experiment, Amos Tversky
and Daniel Kahneman gave one group of subjects the following problem (the
percentage of responses to each question is shown in parentheses):
Imagine that the United States is preparing for the outbreak of an unusual
virus that is expected to kill 600 people. Two alternative programs to combat
the disease have been proposed. Assume that the exact scientific estimates of
the consequences of the programs are as follows:

• If Program A is adopted, 200 people will be saved (72 percent).

 decision making under risk 477

• If Program B is adopted, there is a one-third probability that 600 people

will be saved and a two-thirds probability that no people will be saved
(28 percent).
A second group of subjects was given the same scenario, but these two
choices:
• If Program C is adopted, 400 people will die (22 percent).
• If Program D is adopted, there is a one-third probability that nobody will
die and a two-thirds probability that 600 people will die (78 percent).
Notice that Program A is mathematically identical to Program C and that
Program B is identical to Program D—the only difference is that the first set of
choices is framed in terms of gains (people saved) and the second in terms of
losses (people dying). When making the choice in terms of gains, the large
majority of subjects were risk averse; when making the choice in terms of losses,
an even larger number were risk seeking.12
This violates the axiom of invariance (Section 12.2.2b)—that the preference
between choices should not depend on the manner in which they are described.
As Tversky and Kahneman note, “the failure of invariance is both pervasive and
robust”:
It is as common among sophisticated respondents as among naïve ones, and
it is not eliminated even when the same respondents answer both questions
within a few minutes. Respondents confronted with their conflicting answers
are typically puzzled. Even after rereading the problems, they still wish to be
risk averse in the “lives saved” version; they wish to be risk seeking in the
“lives lost” version; and they also wish to obey invariance and give consistent
answers in the two versions. In their stubborn appeal, framing effects often
resemble perceptual illusions more than computational errors.

15.4.1 Litigation vs. Settlement

In a comprehensive survey of the effect of prospect theory in legal settings, Chris
Guthrie notes:13 “In most lawsuits, plaintiffs choose either to accept a certain
settlement from the defendant or to proceed to trial in hopes of obtaining an
even more favorable judgment; most defendants, by contrast, must choose either

12. Amos Tversky and Daniel Kahneman, The Framing of Decisions and the Psychology
of Choice, 211 Science 453 (1981). David Mandel, Gain-Loss Framing and Choice: Separating
Outcome Formulations from Descriptor Formulations, 85 Org. Behavior and Human
Decision Processes 56 (2001), argues that the formulation of the problem may be ambig-
uous and that this and other factors may confound the framing effects.
13. Much of the following analysis is based on Chris Guthrie, Prospect Theory, Risk,
Preference, and the Law, 97 Nw. U. L. Rev. 1115 (2003).
478 problem solving, decision making, and professional judgment

to pay a certain settlement to the plaintiff or to gamble that further litigation will
reduce the amount they must pay. Thus, plaintiffs generally choose between
options that appear to represent gains, while defendants generally choose
between options that appear to represent losses.”
Economic theory predicts that their different perspectives will have no effect
on plaintiffs’ and defendants’ decisions whether to litigate or settle. Given
options with the identical expected value, a risk-neutral litigant would be indif-
ferent and a risk-averse litigant would prefer to settle. Overoptimism or the pur-
suit of nonmonetary values other than wealth (e.g., vengeance, vindication, or
stubbornness) might lead to suboptimal decisions to litigate, but theoretically
there is no reason to believe that these affect plaintiffs and defendants asym-
metrically.
In fact, however, plaintiffs tend to be risk-averse and defendants tend to be
risk-seeking. This is suggested by actual settlement patterns14 and is the robust
finding of many experiments. For example, Jeffrey Rachlinski asked students to
play the roles of parties in a copyright case.15 The plaintiffs had a choice
between:
• accepting the defendant’s $200,000 settlement offer, or
• going to trial with a 50 percent chance of winning $400,000 and a 50 per-
cent chance of winning nothing.
The defendants had a choice between:
• paying the plaintiff $200,000 in settlement, or
• going to trial with a 50 percent chance of losing $400,000 at trial and a 50
percent chance of losing nothing.
Although expected value of all of the options is identical, 77 percent of the plain-
tiffs preferred settlement, while 69 percent of the defendant chose to go to trial.
In another experiment, participants were assigned the role of the plaintiff or
defendant in litigation between a rural bed and breakfast inn (B&B) and an adja-
cent property owner.16 The B&B had unintentionally expanded to occupy a small
corner of the adjacent property, worth about $50. After unsuccessful negotia-
tions, the property owner sued for an injunction to require the B&B to remove
the encroaching structure. The judge can either order the B&B to remove the

14. Jeffrey Rachlinski, Gains, Losses and the Psychology of Litigation, 70 So. Cal. L. Rev.
113 (1996); Guthrie, Prospect Theory, supra; Russell Korobkin and Chris Guthrie, Psychology,
Economics, and Settlement: A New Look at the Role of the Lawyer, 76 Texas L. Rev. 77
(1997).
15. Rachlinski, supra.
16. Rachlinski, supra. Rachlinski did a number of versions of the experiment with dif-
ferent numbers.
 decision making under risk 479

structure at considerable cost, or order the adjacent owner to sell the corner of
his property for about $50. If the judge orders removal of the structure, then
rather than tear it down, the B&B will offer the owner $100,000—an offer that
the owner is certain to accept. In other words, depending upon the judge’s deci-
sion, the plaintiff will win, and the defendant will lose, either $100,000 or $50.
The parties are informed that the judge assigned to the case is an adamant
defender of property rights who opposes forced sales of land and is very (70 per-
cent) likely to order removal of the buildings. It is now one day before trial. The
opposing party is willing to settle for $70,000.
In running this experiment with various stakes and probabilities, 81 percent
of the plaintiffs settled, while only 45 percent of the defendants settled. While the
defendants were mildly risk seeking, the plaintiffs were highly risk averse.
The reference point from which any decision maker assesses gains and losses
is subjective. An experiment by Russell Korobkin and Chris Guthrie shows how
a plaintiff’s reference point may change in almost identical cases.17 Participants
played the role of plaintiffs, who had been slightly injured and whose cars were
totaled in a car accident, and who had a choice between accepting a settlement of
$21,000 or going to trial with a 50 percent chance of an award of full damages of
$28,000 and a 50 percent chance of receiving only $10,000 (the insurance limit)—
an expected value of $19,000. The first group of subjects was informed that
$14,000 of the damages were for medical costs that had already been reimbursed
by health insurance and the other $14,000 was the full value of their car. The
second group was told that $4,000 were for (reimbursed) medical costs and the
remaining $24,000 was the value of their car. For subjects in the first group,
the settlement reflected a net gain and they were risk averse, overwhelmingly
preferring to settle. For subjects in the second group, the settlement represented
a loss, and they tended to choose to go to trial.
Not just partisans, but neutrals may be affected by how the issue is framed.
Chris Guthrie, Jeffrey Rachlinski, and Andrew Wistrich presented a version of the
copyright infringement case described above to judges presiding over settlement
conferences, who were asked whether to recommend settlement.18 They were
informed that there was a 50 percent chance that the plaintiff would prevail at trial
and win $200,000 and a 50 percent chance that she would not win anything at
all—and that it would cost each party $50,000 in attorneys’ fees to go to trial.
Half of the judges reviewed the case from the plaintiff’s perspective: Should
the plaintiff accept defendant’s offer to pay the plaintiff $60,000 to settle the
case? Those judges thus had a choice between

17. Russell Korobkin and Chris Guthrie, Psychological Barriers to Litigation Settlement:
An Experimental Approach, 93 Mich. L. Rev. 107 (1994).
18. Chris Guthrie, Jeffrey J. Rachlinski, Andrew J. Wistrich, Inside the Judicial Mind,
86 Cornell Law Review (May 2001).
480 problem solving, decision making, and professional judgment

• a certain $60,000 gain, and

• a 50 percent chance of winning $200,000 judgment less the $50,000 attor-
neys’ fees (EV = $50,000).
The judges who reviewed the case from the defendant’s perspective were
asked: Should the defendant accede to the plaintiff’s demand that it pay $140,000
to settle the case? They saw that the defendant had a choice between
• a certain $140,000 loss or
• a 50 percent chance of a losing a $200,000 judgment minus $50,000 attor-
neys’ fees (EV = $150,000).
In both cases, then, the expected value of settlement was $10,000 greater
than the expected value of going to trial. But the plaintiff’s options appeared to
represent gains (i.e., $60,000 settlement versus $50,000 trial) while the defendant’s
options appeared to represent losses (i.e., -$140,000 settlement versus -$150,000
trial). Consistent with prospect theory, the researchers found that the framing of
decision options influenced the judges’ recommendations. Nearly 40 percent of
the judges assessing the problem from the plaintiff’s perspective indicated that
they would recommend settlement to the plaintiff, while only 25 percent of the
judges assessing the problem from the defendant’s perspective indicated that they
would recommend settlement to the defendant. Chris Guthrie suggests:
judges are likely to advocate settlement more strenuously to plaintiffs than to
defendants, even though the experimental evidence suggests that plaintiffs are
more likely than defendants to be attracted to settlement in the first place. By
urging plaintiffs to accept an amount that is less than appropriate or by failing
to urge defendants to pay an appropriate amount, judges may promote unfair
settlements that under-compensate plaintiffs and under-deter defendants.
As noted earlier, what someone views as the status quo—and hence the refer-
ence point for gains and losses—is entirely subjective. All things being equal, a
defendant tends to treat what’s in his pocketbook or bank account as an endow-
ment that litigation puts at risk, and the plaintiff views litigation as providing an
opportunity to gain something. This perspective on the plaintiff’s part depends
on her having adjusted to the harm inflicted by the defendant.19 If the feeling of
harm is still highly salient, or if for some other reason the plaintiff still feels the
need for revenge, she is more likely to adjust her utility preferences from risk
averse to risk seeking.

Appendix A (Section 15.5) discusses how people’s different risk attitudes toward
gains and losses can violate the Dominance Axiom.

19. We explored the psychological dynamics of such adjustment in Section 12.1.5.

 decision making under risk 481

15.4.2 Targets and Risk-Seeking and Unethical Behavior

Maurice Schweitzer and his colleagues asked students to solve anagrams, check
their own work, and reward themselves (in a situation where they were unaware
that the experimenters could independently check their results). Participants in
one group were just instructed to do your best to create as many words as you can;
those in a second group were given the goal of creating nine words in each round.
The goal group performed better than the do your best group (though the differ-
ences were not statistically significant). Although a majority of the goal group
reported their achievements honestly, a significant number overstated their
achievements. The closer they were to having nine words, the more they were
likely to pad their report to achieve the goal.20
How does this apply to real-world situations? Chris Guthrie writes:
An arbitrarily set “target”—for example, a business’s quarterly earnings or a
law firm associate’s minimum billable hours—may serve as a reference point,
with the individual executive or lawyer viewing achievements beyond the
target as gains, but those below as in the domain of losses. Several commenta-
tors, citing examples from the real world, have suggested that the motivation
to escape from the domain of losses, especially when one is close to the target,
encourages risk-seeking cheating and other forms of ethical misconduct.
By the same token, lawyers’ behavior may be influenced by how well they
believe the case or transaction is going compared to an assumed baseline. “When
things appear to be going well (gains), risky ethical violations will seem unat-
tractive; when things appear to be going poorly (losses), however, those same
ethical violations will hold more appeal.”21
Jeffrey Rachlinski tested this in an experiment where law students played the
role of the outside litigation counsel for a pharmaceutical company in a product
liability case brought by the parents of a child who had suffered permanent brain
damage from a drug.22 The parents are willing to settle the case for $3 million,
but they are unaware that the defendant withheld incriminating documents
during discovery. If the lawyers reveal the documents before settlement, the case
is likely to cost the defendant considerably more. If they proceed to settlement,
there is a reasonable chance that the documents will subsequently be revealed
and, if so, that the lawyers will be sanctioned.

20. Maurice Schweitzer, Lisa Ordóñez, and Bambi Douma, Goal Setting as a
Motivator of Unethical Behavior, 47 Academy of Management Journal 422 (2004);
Maurice Schweitzer, Lisa, Ordóñez, and Bambi Douma, The Dark Side of Goal Setting:
The Role of Goals in Motivating Unethical Decision Making, 47 Academy of Management
Proceedings 422 (2002).
21. Chris Guthrie, Prospect Theory, Risk Preference & the Law, 97 Northwestern
University Law Review 1115, 1140 (2003).
22. Jeffrey J. Rachlinski, Gains, Losses, and the Psychology of Litigation, 70 S. Cal. L. Rev.
113, 118 (1996).
482 problem solving, decision making, and professional judgment

Two groups were asked whether they would reveal the documents before
going forward with settlement negotiations. The problem was presented to one
group in a “gains” frame: The company had expected to have to pay the parents
$5 million to settle the case and therefore thought that the case was “going well.”
The other group was given a “loss” frame: The company had originally expected
to pay the parents only $1 million and believed that the case was “going poorly.”
Only 12.5 percent of the subjects who saw the settlement as a gain chose to settle
without disclosing the documents, while 45 percent who saw the settlement as a
loss indicated that they would settle with disclosing the documents. By contrast
to the risk aversion manifested by those in gains condition, those in the loss
condition opted for the riskier—and ethically and legally problematic—course of
action rather than face a sure loss.

15.4.3 Low-Probability Gains and Losses

Prospect theory predicts that people overweight very low probabilities and under-
weight very high probabilities. Correspondingly (and reversing the general
phenomenon described above), they tend to be risk seeking with respect to low prob-
ability gains and risk averse with respect to unlikely losses. Chris Guthrie suggests
that this explains risk attitudes toward frivolous litigation:23
In frivolous or low-probability litigation, the plaintiff typically chooses between
a relatively small settlement amount and a low likelihood of obtaining a much
larger amount at trial. Defendants, by contrast, typically must choose either to
pay some small settlement or face a low likelihood of having to pay a much
larger amount at trial. . . . Decision makers confronted with low-probability
gains . . . tend to make risk-seeking decisions, while those confronted with
low-probability losses . . . tend to make risk-averse decisions. In short, litigant
risk preferences are likely to be reversed in frivolous suits, with plaintiffs rela-
tively more attracted to trial than defendants.
Guthrie assigned law students to play the roles of plaintiffs and defendants in
a low-probability litigation. Plaintiffs could choose between a $50 settlement pay-
ment or a 1 percent chance at a $5,000 judgment at trial; defendants could either
pay a certain $50 settlement to plaintiff or face a 1 percent chance of having to
pay a $5,000 judgment at trial.
Although the expected value of the options was identical, 62 percent of the
plaintiffs were willing to risk trial, while 84 percent of the defendants chose to
settle. “Faced with low-probability gains, plaintiffs appear to make risk-seeking
choices unanticipated by the economic theory.”

23. Chris Guthrie, Framing Frivolous Litigation: A Psychological Theory, 67 University

of Chicago Law Review 163 (2000).
 decision making under risk 483

Estimating the Probabilities of Rare Events

Suppose that Desmond and Edwina are predicting whether their respective cli-
ents will prevail as plaintiff in a civil suit. Desmond has no personal experience
with this sort of litigation, but is reliably informed that plaintiffs in this jurisdic-
tion win 10 percent of the time. Edwina handles many such cases as a plaintiff’s
attorney and wins 10 percent of the time. Who is more likely to predict that her
client will win? Consistent with the prospect theory, Desmond will overweight the
10 percent probability in making his prediction. However, Edwina may actually
underweight the probability. Her estimate is likely to be affected by the availability
of the most recent events in her experience—and, by the nature of probability, the
one win out of ten is not likely to be one of the most recent occurrences.24

15.4.4 The Certainty Effect24

Tversky and Kahneman demonstrated what they call the certainty effect with the
following experiment.25
Subjects were asked to choose between:
A. a sure win of $300, and
B. 80 percent chance to win $450 and 20 percent chance to win nothing.
The large majority chose A, which has a lower expected value than B.
Other subjects were asked to choose between:
C. 25 percent chance to win $300 and 75 percent chance to win nothing, and
D. 20 percent chance to win $450 and 80 percent chance to win nothing.
The large majority of subjects chose D, which has a higher expected value than C.
Note, however, that the relationship between C and D is identical to that between
A and B — the probabilities are just reduced fourfold, as shown in Table 15.5.
The greater attraction of A lies in its certainty—in this case the increase in the
probability of winning from 0.25 to 1.0 has a greater effect than the increase
from 0.2 to 0.8.
The certainty effect was also demonstrated by this problem, given by one of
the authors to groups of in-house corporate counsel:

24. Elke U. Weber, “Origins and Functions of Risk Perception, Center for the Decision
Sciences” (presented at NCI Workshop on Conceptualizing and Measuring Risk
Perceptions, February 13–14, 2003), http://dccps.nci.nih.gov/brp/presentations/weber.p
df#search=%22Weber%2C%20Origins%20and%20Functions%20of%20Risk%20Perce
ption%20nci%22.
25. Amos Tversky and Daniel Kahneman, The Framing of Decisions and the Psychology
of Choice, supra.
484 problem solving, decision making, and professional judgment

table 15.5 the certainty effect

EV % Choosing

A Sure gain of $300 $300 78%

B 80% chance to win $450 20% chance to win nothing $360 22%

C 25% chance to win $300 75% chance to win nothing $75 42%
D 20% chance to win $450 80% chance to win nothing $90 58%

A new, damaging computer virus has been detected. A computer consultant,

in whom you have complete confidence, informs you that there is a 20%
chance that your company-wide computer system will become infected and
crash. Your accounting office estimates that the potential damage to the com-
pany is somewhere between $1 million and $10 million. How much would
you pay the computer consultant to reduce the chances of a crash to 1%? How
much additional would you pay to reduce the chances from 1% to 0%?
The large majority of respondents would pay far more—often as much as ten
times more—to reduces the chances from 1 percent to 0 percent than for any
unit percentage reduction from 20 percent to 1 percent.
Justice Stephen Breyer describes a real-world example arising out of litigation
involving cleanup of a toxic waste dump in southern New Hampshire:
The site was mostly cleaned up. All but one of the private parties had settled.
The remaining private party litigated the cost of cleaning up the last little bit,
a cost of about $9.3 million to remove a small amount of highly diluted PCBs
and “volatile organic compounds” (benzene and gasoline components) by
incinerating the dirt. How much extra safety did this $9.3 million buy? The
forty-thousand-page record of this ten-year effort indicated (and all the parties
seemed to agree) that, without the extra expenditure, the waste dump was
clean enough for children playing on the site to eat small amounts of dirt
daily for 70 days each year without significant harm. Burning the soil would
have made it clean enough for the children to eat small amounts daily for 245
days per year without significant harm. But there were no dirt-eating children
playing in the area, for it was a swamp. Nor were dirt-eating children likely to
appear there, for future building seemed unlikely. The parties also agreed
that at least half of the volatile organic chemicals would likely evaporate by the
year 2000. To spend $9.3 million to protect nonexistent dirt-eating children is
what I mean by the problem of “the last 10 percent.”26

26. Stephen Breyer, Breaking the Vicious Circle: Toward Effective Risk
Regulation (Cambridge: Harvard University Press, 1995).
 decision making under risk 485

At some point, a probability may be so low that people treat it as if it were

zero. The vast majority of participants in a study said they would not bother to
wear seat belts if the probability of being killed in a car accident was 0.0000025
per trip.27

Appendix B (Section 15.6) discusses the Certainty Effect and Violation of the
Independence Axiom.

15.4.5 The Pseudocertainty Effect and Other Violations of the Independence

Axiom
Although some manifestations of the certainty effect seem problematic as a
matter of public policy, it is hard to characterize the willingness to pay a pre-
mium for certainty as irrational if it allows you to sleep better at night. But con-
sider this experiment:
Tversky and Kahneman gave subjects the following two-stage game.
I. In the first stage there is a 75 percent chance to end the game without win-
ning anything, and a 25 percent chance to move into the second stage.
II. If you reach the second stage you have a choice between
E. a sure gain of $300, and
F. 80 percent chance to win $450 and 20 percent chance to win nothing.
There is a one-in-four chance of moving to the second stage, and Table 15.6
shows the expected value of E and F—and also the percent of subjects who made
the choices.
table 15.6 the pseudocertainty effect

EV % Choosing

E Sure win of $300 $75 74%

F 80% chance to win $450 20% chance to win nothing $90 26%

The expected values of E and F are, respectively, identical to the

expected values of C and D in the “certainty effect” example (Section 15.4.4).
Note, however, that while the majority of subjects there were risk seeking in

27. Jonathan Baron, Thinking and Deciding 255 (3rd. ed. New York: Cambridge
University Press, 2001). The respondents apparently ignored the value of seatbelts in pre-
venting serious injuries.
486 problem solving, decision making, and professional judgment

choosing D, the majority of subjects in the two-stage game made the risk averse
choice of E.
Tversky and Kahneman label this phenomenon the pseudocertainty effect
because an outcome that is actually uncertain is weighted as if it is certain. The
framing of probabilities as a two-stage game encourages respondents to apply
cancellation: the event of failing to reach the second stage is discarded prior to
evaluation because it yields the same outcomes in both options.28 They suggest
that the choice between E and F has a greater potential for inducing regret than
the choice between C and D.
A simple experiment by Paul Slovic and his colleagues demonstrates how the
pseudocertainty effect can affect public health decisions. One group was told
that a disease will afflict 20 percent of the population, but that a vaccine will
reduce their risk of getting the disease by a half. A second group was told that
there are two strains of the same disease, that each will afflict 10 percent of the
population, and that a vaccine will entirely eliminate the risk of getting one strain
but have no effect on the other. Although each will reduce an individual’s risk of
getting the disease from 20 percent to 10 percent, many more people (57 per-
cent) in the second group than in the first group (40 percent) said they were
willing to be vaccinated.29
Tversky and Kahneman replicated the two-stage game described above in
these questions given physicians attending a meeting of the California Medical
Association:30
In the treatment of tumors there is sometimes a choice between two types of
therapies: (i) a radical treatment such as extensive surgery, which involves
some risk of imminent death, (ii) a moderate treatment, such as limited sur-
gery or radiation therapy. Each of the following problems describes the pos-
sible outcome of two alternative treatments, for three different cases. In
considering each case, suppose the patient is a 40-year-old male. Assume that
without treatment death is imminent (within a month) and that only one of
the treatments can be applied. Please indicate the treatment you would prefer
in each case.

28. Amos Tversky and Daniel Kahneman, The Framing of Decisions and the Psychology
of Choice, supra.
29. See Paul Slovic Baruch Fischhoff and Sarah Lichtenstein, Response Mode, Framing,
and Information Processing Effects in Risk Assessment, in New Directions for Methodology
of Social and Behavioral Science: The Framing of Questions and the Consistency
of Responses 21 (Robin M. Hogarth ed. San Francisco: Jossey-Bass, 1982).
30. Rational Choice in the Framing of Decisions, in Daniel Kahneman and Amos
Tversky, eds. Choice, Values and Frames 209 (New York: Cambridge University Press,
2000). The order of the cases has been changed to make it parallel to the preceding
examples in the text.
 decision making under risk 487

CASE 1
Treatment A: certainty of a normal life, with an expected longevity of 18 years.
[65%]
Treatment B: 20% chance of imminent death and 80% chance of normal
life, with an expected longevity of 30 years. [35%]
CASE 2
Treatment C: 75% chance of imminent death and 25% chance of normal life,
with an expected longevity of 18 years. [32%]
Treatment D: 80% chance of imminent death and 20% chance of normal
life, with an expected longevity of 30 years. [68%]
CASE 3
Consider a new case where there is a 25% chance that the tumor is treatable
and a 75% chance that it is not. If the tumor is not treatable, death is immi-
nent. If the tumor is treatable, the outcomes of the treatment are as follows:
Treatment E: certainty of normal life, with an expected longevity of 18 years.
[68%]
Treatment F: 20% Chance of imminent death and 80% chance of normal
life, with an expected longevity of 30 years. [32%]
The three cases of this problem correspond, respectively, to the choices of
the bets A and B, C and D (Section 15.4.4), and E and F (Section 15.4.5) in the
preceding examples, and reveal the same pattern of preferences. The experi-
menters observe:
In case 1, most respondents make a risk-averse choice in favor of certain
survival with reduced longevity. In case 2, the moderate treatment no
longer ensures survival, and most respondents choose the treatment that
offers the higher expected longevity. In particular, 64% of the physicians who
chose A in case 1 selected D in case 2. This is another example of the certainty
effect.
The comparison of cases 2 and 3 provides another illustration of pseudo-
certainty. The cases are identical in terms of the relevant outcomes and
their probabilities, but the preferences differ. In particular, 56% of the phy-
sicians who chose D in case 2 selected E in case 3. The conditional framing
induces people to disregard the event of the tumor’s not being treatable
because the two treatments are equally ineffective in this case. In this frame,
treatment E enjoys the advantage of pseudocertainty: it appears to ensure
survival, but the assurance is conditional on the treatability of the tumor. In
fact, there is only a .25 chance of surviving a month if this option is
chosen.
488 problem solving, decision making, and professional judgment

15.5 appendix a

15.5.1 How People’s Different Risk Attitudes Toward Gains and Losses Can
Violate the Dominance Axiom
The axioms of completeness and dominance in conditions of uncertainty are
simply extensions of the axioms described in Section 12.2.2. Suppose a decision
maker prefers prospect (i.e., outcome) A to prospect B, and that he has a choice
between two lotteries, as shown in Figure 15.10. The first lottery, which we call
X, will result in prospect A with probability p, or prospect B with probability 1-p.
The second lottery, which we call Y, will result in prospects A or B with probabil-
ities q and 1-q, respectively. To satisfy the dominance axiom, the decision maker
will (1) prefer lottery X over lottery Y if and only if p > q; and (2) will be indifferent
between X and Y if and only if p = q.
A A
p q
X: Y:
1-p 1-q
B B

figure 15.10 dominance in lotteries.

People’s different risk attitudes toward gains and losses can violate the domi-
nance axiom.
Consider the choice between these two lotteries:31
E. 25 percent chance to win $240 and 75 percent chance to lose $760, and
F. 25 percent chance to win $250 and 75 percent chance to lose $750.
When given to a group of subjects, everyone recognizes that F dominates E
and chooses F.
Now consider these two pair of concurrent choices.
Choose between:
A. A sure gain of $240, and
B. 25 percent chance to win $1000 and 75 percent chance to win nothing.
and
Choose between:
C. Sure loss of $750, and
D. 75 percent chance to lose $1000 and 25 percent chance to lose nothing.

31. Daniel Kahneman and Amos Tversky, Choices, Values, and Frames, in Choices,
Values, and Frames, supra at 1.
 decision making under risk 489

As shown in Table 15.7, the choice of A and D has precisely the same expected
value as E, and the choice of B and C has the same value as F.
However, 73 percent of subjects choose A and D while only 3 percent chose B
and C,32 which dominates A and D: People are risk averse with respect to gains,
and risk seeking with respect to losses—and the violation of dominance was
stimulated by putting one part of the concurrent choice in the domain of losses
and the other in the domain of gains.

table 15.7 separating the domains of gains and losses

EV

A&D= E 25% chance to win $240 75% chance to lose $760 -$510
B&C= F 25% chance to win $250 75% chance to lose $750 -$500

15.6 Appendix B

15.6.1 The Certainty Effect and Violation of the Independence Axiom

Recall the independence axiom (Section 12.2.2.c). In the context of decision
making under risk, it is also called the cancellation, or sure thing, principle. It
provides: A decision maker who prefers alternative A to B should continue to
prefer A to B even though there is a specified probability that the decision maker
will receive neither A nor B, but some other alternative, C.
The French economist, Maurice Allais, noticed that the certainty effect could
cause people to violate the independence axiom. Consider this choice:
A. $1,000,000 for sure.
B. A 10 percent chance of receiving $2,5000,000, an 89 percent chance of
receiving $1,000,000, and a 1 percent chance of receiving nothing.
Most people choose A despite the fact that B has a considerable higher EV ($1.4
million compared to $1 million).
Now consider this choice
C. An 11 percent chance of receiving $1,000,000 and an 89 percent of receiv-
ing nothing.
D. A 10 percent chance of receiving $2,500,000 and a 90 percent chance of
receiving nothing.
Most people choose D, which has over twice the EV.
To see why this violates the independence axiom, let’s recast the choice
in terms of randomly drawing chips out of a bag that contains 89 red,
490 problem solving, decision making, and professional judgment

10 blue, and 1 black chip. The equivalents of A and B, and C and D are shown in
Table 15.8.
Table 15.8 Allais’ Paradox

Chips

89 Red 1 Black 10 Blue

A $1 million $1 million $1 million

B $1 million $0 $2.5 million
C $0 $1 million $1 million
D $0 $0 $2.5 million

Note that choosing the red chip has no effect in the A-B choice; thus, the inde-
pendence axiom implies that one only need to consider the black and
blue chips. By the same token, choosing the red chip has no effect in the C-D
choice, and only the black and blue chips need to be considered. And note
that, with the red chips excluded, the lottery in A-B is precisely the same as that
in C-D.
Thus does the extraordinarily high utility placed on certainty cause a violation
of the independence axiom.
16. the role of affect in risky decisions

In Chapter 1, we quoted Paul Slovic’s observation that virtually everything we

experience is accompanied by some sort of affect, ranging from a “faint
whisper of emotion” to strong feelings of fear and dread, and that images,
marked by positive and negative affective feelings, guide judgment and deci-
sion making.1 This section considers how affect affects people’s assessment of
risks.

16.1 risks as feelings

George Loewenstein and his colleagues posit a “risks-as-feelings hypothesis”—a

model of decision making that incorporates emotions experienced during the
decision-making process, as shown in Figure 16.1.2 They write:

Anticipated outcomes
(including anticipated emotions)
Cognitive
evaluation

Subjective
probabilities Behavior Outcomes
(incl. emotions)

Feelings
Other factors,
e.g., vividness, immediacy,
background mood

figure 16.1 risks as feelings.

1. “Affect” refers to “the specific quality of ‘goodness’ or ‘badness’ (1) experienced, as

a feeling state (with or without consciousness) and (2) demarcating a positive or negative
quality of a stimulus.” Paul Slovic, Melissa Finucane, Ellen Peters, and Donald G.
MacGregor, The Affect Heuristic [hereafter Affect Heuristic], in Heuristics And Biases:
The Psychology of Intuitive Judgment 397–420 (Thomas Gilovich, Dale Griffin, and
Daniel Kahneman eds., New York: Cambridge University Press, 2002) [hereafter
Heuristics and Biases].
2. George Loewenstein, Elke Weber, Christopher Hsee, and Ned Welch, Risk As
Feelings, 127 Psych. Bull. 267 (2001).
492 problem solving, decision making, and professional judgment

The risks-as-feeling hypothesis postulates that responses to risky situations

(including decision making) result in part from direct [i.e., System 1] emotional
influences, including feelings, such as worry, fear, dread, or anxiety. People . . .
evaluate risky alternatives at a cognitive level, as in traditional models, based
largely on the probability and desirability of associated consequences. Such cog-
nitive evaluations have affective consequences, and feeling states also exert a
reciprocal influence on cognitive evaluations. At the same time, however, feel-
ing states . . . respond to factors, such as the immediacy of a risk, that do not
enter into cognitive evaluations of risk and also respond to probabilities and
outcome values in a fashion that is different from the way in which these vari-
ables enter into cognitive evaluations. Because their determinants are different,
emotional reactions to risk can diverge from cognitive evaluations of the same
risks. As illustrated by Figure [16.1], behavior is then determined by the inter-
play between these two, often conflicting, responses, to a situation. . . . Thus
feelings may be more than an important input into decision making under
uncertainty; they may be necessary and, to a large degree, mediate the connec-
tion between cognitive evaluation of risk and risk-related behavior.”3
Under this model, “affect plays an informational role in decision making—
that is, it provides inputs into decision making that help people evaluate alterna-
tive courses of action, albeit not always in a normative fashion.” Note that
Loewenstein et al. use the word behavior rather than decision in the penultimate
box, noting that “many types of emotion-driven risk-related behaviors, ranging
from panic reactions (e.g., slamming on the brakes when one skids on the ice) to
the agoraphobic individual’s inability to leave the house, do not seem to reflect
“decisions” in the sense that term is usually used.
Antonio Damasio provides a hypothetical example of the role that negative
affect plays in decision making.4 Using the example of a businessperson decid-
ing whether or not to do a deal with a potentially valuable client who is the arch-
enemy of his best friend, Damasio suggests the complexities of a purely
deliberative process:
Gaining a client may bring immediate reward and also a substantial amount
of future reward. How much reward is unknown and so you must estimate its
magnitude and rate over time, so that you can pit it against the potential losses
among which you must now count the consequences of losing a friendship.
Since the latter loss will vary over time, you must also figure its “depreciation”
rate! You are, in effect, faced with a complex calculation, set at diverse imagi-
nary epochs, and burdened with the need to compare results of a different
nature which somehow much be translated into a common currency for the
comparison to make any sense at all. A substantial part of this calculation will

3. Minor textual changes throughout.

4. Antonio R. Damasio, Descartes Error: Emotion, Reason and the Human Brain
(New York: HarperCollins, 1995).
 the role of affect in risky decisions 493

depend on the continued generation of yet more imaginary scenarios, built

on visual and auditory patterns, among others, and also on the continued
generation of verbal narratives which accompany those scenarios, and which
are essential to keep the process of logical inference going.5
Consciously and systematically processing the myriad representations
involved in even this one decision is beyond our cognitive capacity. But Damasio’s
“somatic marker hypothesis” focuses on the role that these images play before
we engage in rational analysis:
Before you apply any kind of cost/benefit analysis . . . and before you reason
toward the solution of the problem, something quite important happens.
When the bad outcome connected with a given response option comes into
mind, however fleetingly, you experience an unpleasant gut feeling. Because
the feeling is about the body, I gave the phenomenon the technical term
somatic state (“soma” is Greek for body); and because it “marks” an image, I
called it a marker.
...
Somatic markers . . . are feelings that have been connected, by learning, to
predicted future outcomes of certain scenarios. When a negative somatic
marker is juxtaposed to a particular future outcome the combination func-
tions as an alarm bell. When a positive somatic marker is juxtaposed instead,
it becomes a beacon of incentive.
[The negative somatic marker] functions as an automated alarm signal which
says: “Beware of danger ahead if you choose the option which leads to this out-
come.” . . . The automated signal protects you against future losses, without
further ado, and then allows you to choose from among fewer alternatives. There
is still room for using a cost/benefit analysis and proper deductive competence,
but only after the automated step drastically reduces the number of options.6
As Damasio notes, somatic markers may also highlight some options as
favorable.7 While positive markers cannot eliminate options in advance of the
rational analytic process, they may play a facilitative role within an intuitive cost-
benefit analysis process.

16.1.1 Anxiety as the Shadow of Intelligence8

Damasio is a neurologist whose research has focused on patients with damage
to the ventromedial region of the frontal lobe of the brain—damage that curtails
their emotions or affect but does not compromise intelligence or analytic ability.9

5. Id. at 170.
6. Id. at 173–74 (paragraphs rearranged).
7. Id. at 174.
8. David H. Barlow, Anxiety and Its Disorders: The Nature and Treatment of
Anxiety and Panic (1988) (quoted in Loewenstein et al., supra note 2).
9. Damasio, supra at 193.
494 problem solving, decision making, and professional judgment

His experiments suggest that anxiety may be an important aspect of one’s capac-
ity to plan. For example, subjects were given a sum of play money and were
asked to repeatedly choose cards from among four decks, which had different
risks and payoffs. The subjects had to learn the rules that governed payoffs
through trial and error, and the rules were sufficiently complex that they were
difficult to learn them with certainty. Two of the decks provided consistent pay-
ments but unpredictably required the player to pay the experimenter a very large
sum. The other two decks provided smaller payments and unpredictably required
the player to pay a smaller amount. The expected value of choosing from the first
two decks was smaller than the latter.
Over time, members of a control group learned to avoid the two sets of cards
with higher risks and lower expected value. The feeling-impaired subjects con-
tinued to choose cards from the risk decks and went bankrupt. Members of the
control group hesitated before choosing the risky cards and they exhibited
heightened visceral reactions (e.g., change in skin conductance) before making
the choice. The feeling-impaired subjects did not hesitate and gave no physiolog-
ical indications before making the choice.
Loewenstein et al. note that “the lack of an emotional response does not nec-
essarily lead to poor decisions. . . . One could easily design an experiment where
the expected value of the high-risk deck (that contains some large losses) is actu-
ally higher than that of the low-risk deck. In this case, prefrontal damaged
patients would do better in the long run than nonpatients, because the fear in the
latter group would hinder them from choosing from the risky but higher expected
value deck.” Indeed, as we shall see below, emotions can often distort people’s
perception of or reaction to risk.

Damasio’s subjects had brain lesions that left their analytic capacities uncompro-
mised, but impaired their ability to associate emotions with the anticipated con-
sequences of their actions. People with and without similar lesions were given
this problem:

You have abandoned a sinking cruise ship and are in a crowded lifeboat that
is dangerously low in the water. If nothing is done it will sink before the rescue
boats arrive and everyone will die. However, there is an injured person who
will not survive in any case. If you throw that person overboard, the boat will
stay afloat and the remaining.

How do you suppose the different groups responded?

See Benedict Carey, Brain Injury Said to Affect Moral Choices, NEW YORK TIMES, Mar. 22, 2007.
Michael Koenigs et al., Damage to the Prefrontal Cortex Increases Utilitarian Moral Judgments,
NATURE, http://www.nature.com/news/2007/070319/full/070319-9.html.
 the role of affect in risky decisions 495

16.2 the availability heuristic and the role of vividness

Damasio writes that “the action of biological drives, body states, and emotions
may be an indispensable foundation for rationality. . . . Rationality is probably
shaped and modulated by body signals, even as it performs the most sublime
distinctions and acts accordingly.”10 Emotions and reasoning exist in a delicate
balance, however, and emotions can sometimes overwhelm reasoning to our
detriment. The following pages focus mostly on risks involving human life and
safety, where emotions tend to run fairly high.
In Section 9.6 we saw that people often estimate the frequency of events (or
their probabilities) based on how easily examples come to mind. We saw that the
availability or vividness of events like shark attacks and the 9/11 terrorist attacks
affects people’s perception of risk. As David Myers notes, such perceptions have
behavioral consequences:
Even before the horrors of September 11th and the ensuing crash at Rockaway
Beach, 44 percent of those willing to risk flying told Gallup they felt fearful. . . . After
the four crashed airliners, and with threats of more terror to come, cancellations
understandably left airlines, travel agencies, and holiday hotels flying into the red.
. . . If we now fly 20 percent less and instead drive half those unflown
miles, we will spend 2 percent more time in motor vehicles. This translates
into 800 more people dying as passengers and pedestrians. So, in just the next
year the terrorists may indirectly kill three times more people on our high-
ways than died on those four fated planes.11
A more quotidian example is that purchases of flood and earthquake insur-
ance are high in the aftermath of a natural disaster, but decline as the vividness
of the event recedes from memory.
The easier it is to imagine a particular outcome, the more likely its occurrence
seems to be. Steven Sherman and his colleagues did an experiment in which stu-
dents were told about Hyposcenia-B, a (fictious) illness that was becoming preva-
lent on their campus. In the easy-to-imagine scenario, students were told that the
disease had concrete symptoms that most of them had probably experienced: low
energy level, muscle aches, severe headaches. In the difficult-to-imagine scenario,
the symptoms were less concrete: a vague sense of disorientation, malfunctioning
nervous system, and inflamed liver. Students given the former scenario thought
they were more likely to contract the disease than those given the latter. The
experimenters conclude that the effort required to imagine contracting the more
difficult-to-imagine symptoms led to a decrease in their availability.12

10. Id. 200.

11. David G. Myers, Do We Fear the Right Things?, American Psychological Society
Observer (December 2001).
12. Steven J. Sherman et al., Imagining Can Heighten or Lower the Perceived Likelihood of
Contracting a Disease: The Mediating Effect of Ease of Imagery, in Heuristics and Biases, supra.
496 problem solving, decision making, and professional judgment

An easily imagined harm can lead people to violate the fundamental principle
that a subset cannot be greater than the larger set of which it is a part.13 A striking
example of this is people’s stated willingness to pay more for flight insurance for
losses resulting from “terrorism” than for flight insurance from all causes (includ-
ing terrorism).14 By the same token, in a between-subjects experiment, Eric
Johnson and his colleagues asked people how much they would pay for insurance
for hospitalization in four different scenarios. 15 The first two groups were respec-
tively asked their willingness to pay (WTP) for hospitalization insurance for any
reason, and for any disease or accident. The third group was asked their WTP for
insurance for any accident assuming that they already had insurance for disease,
and the fourth group was asked their WTP for insurance against any disease
assuming that they already had accident insurance. The mean WTP was:
Any reason $41.53
Any disease or accident $47.12
Any accident (followed by any disease) $69.55
Any disease (followed by any accident) $89.10
Isolating accidents and diseases makes vivid the individual imagine examples
of each, and thus leads to an irrational hierarchy of choices. The WTP is more

All the big money on an accident policy comes from [insuring] railroad accidents.
They found out pretty quick, when they began to write accident insurance, that the
apparent danger spots, the stops that people think are danger spots, aren’t
danger spots at all. I mean, people always think a railroad train is a pretty danger-
ous place to be, . . . but the figures show that not many people get killed, or even
hurt, on railroad trains. So on accident policies, they put in a feature that sounds
pretty good to the man that buys it, because he’s a little worried about train trips,
but it doesn’t cost the company much, because it knows he’s pretty sure to get
there safely. They pay double indemnity for railroad accidents.
—JAMES CAIN, DOUBLE INDEMNITY (1936).

13. Id. Recall the “Linda” problem in Section 8.2.3, where the representativeness heu-
ristic led to a similar error.
14. Id. People’s WTP for $100,000 insurance on a trans-Atlantic flight was $14.12 for
any act of terrorism, $10.31 for any nonterrorism related mechanical failure, and $12.03
for any reason.
15. Eric J. Johnson, John C. Hershey, and Howard Kunreuther Framing, Probability
Distortions, and Insurance Decisions, 7 Journal of Risk and Uncertainty, 35–52 (August
1993). Reprinted in: Choices, Values and Frames (Daniel Kahneman and Amos Tversky
eds., Cambridge, Eng.: Cambridge University Press, 2000).
 the role of affect in risky decisions 497

strongly correlated with vividness than it is with probability—the experimenters’

wording “tricked” the subjects into paying more for insurance that was less
inclusive.
Many of the same biases—particularly those related to fear—that lead indi-
viduals to make poor judgments also affect policy makers. For example, Cass
Sunstein notes that many public activities that pose relatively small risks none-
theless engender great public concern because of the biases that promote dis-
proportionate attention to certain types of risks while neglecting others.16 For
example, much of the public fear concerning nuclear power probably stems
from its association with memorable events such as Hiroshima, Chernobyl, and
Three-Mile Island. Vivid images of the likelihood of a nuclear incident may
combine with the invisibility of the risks created by coal-powered electricity to
lead policy makers to believe that coal is safer than nuclear.

16.3 other determinants of responses to risks

16.3.1 The Relative Acceptability of Risks; High Anxiety and Outrage

During the fall of 2002, snipers randomly killed nine people in the Washington,
D.C. area, causing widespread fear and the curtailment of outdoor activities by
schools and families. The actual increase in residents’ risk was very small.
(Indeed, Cass Sunstein suggests that “some of the precautionary steps, such as
driving to Virginia to purchase gasoline, almost certainly posed risks in excess of
those associated with the snipers’ attacks.”17) In commenting on reactions to the
Washington, D.C. sniper attacks, The Economist, citing Kip Viscusi, noted:
One of the victims was a child—and parents dramatically overestimate any
uncommon threat to their children’s lives (such as the risk of kidnapping by a
stranger). Similarly, the sniper represents an involuntary risk, not one you would
run willingly for a benefit (such as driving too fast to get somewhere). Perhaps
people worry less about voluntary risks. Worst of all, the risk is hard to mitigate.
You cannot easily change it by changing behavior—like wearing a seat belt. The
only way to remove yourself from the sniper’s mercy is not to go out at all.
As this suggests, people assign quite different subjective values to risks that
have the same probability of resulting in pain, serious injury, or death. Based on
the work of Paul Slovic and colleagues, Cass Sunstein and Timur Kuran created

16. Cass R. Sunstein, Laws of Fear: Beyond the Precautionary Principle

(New York: Cambridge University Press, 2005). [Hereafter Laws of Fear].
17. Id at 69.
498 problem solving, decision making, and professional judgment

a chart of factors that make a risk seem more anxiety-producing or outrageous or

more acceptable (see Table 16.1).18

table 16.1 aggravating and mitigating risk factors

Risk Traits Aggravating Mitigating

Familiarity New Old

Personal control Uncontrollable Controlled
Voluntariness Involuntary Voluntary
Media attention Heavy media coverage No media coverage
Equity Unevenly distributed Evenly distributed
Impact on children Children at special risk Children not at risk
Impact on future Future generations at risk Future generations not at risk
generations
Reversibility Irreversible Reversible
Identifiability of victims Victims known Victims unknown
Accompanying benefits Benefits clear Benefits invisible
Source Human-generated Natural origins
Trust in relevant Low trust in institutions High trust in institutions
institutions
Immediacy of adverse Adverse effects immediate Adverse effects delayed
effects
Understanding Mechanisms poorly Mechanisms well understood
understood
Precedents History of accidents No past accidents

In the public sphere, the aggravating factors can lead to high levels of con-
cern, or outrage. As Dr. Robert Scheuplein, former head of the U.S. Food and
Drug Administration’s Office of Toxicology, noted: “When risks are perceived to
be dread[ful], unfamiliar, uncontrollable by the individual, unfair, involuntary,
and potentially catastrophic, they are typically of great public concern, or high
outrage. When risks are perceived as voluntary, controllable by the individual,

18. Timur Kuran and Cass R. Sunstein, Availability Cascades and Risk Regulation, 51
Stanford Law Review 683, 709 (1999).
 the role of affect in risky decisions 499

familiar, equitable, easily reducible, decreasing, and non-catastrophic, they tend

to be minimized by the public (or low outrage).”19
A study by Peter Sandman and colleagues sought people’s reactions to sce-
narios where the risk was high but outrage was low (radiation in a single-family
house caused by naturally occurring radon) and where the risk was low but out-
rage was high (the sand used to make the concrete foundations for the house had
previously been at a storage site for spent nuclear power fuel rods and had been
illegally used to make the concrete). Participants reported much greater fear and
concern and willingness to take costly remedial action in the low-risk/high out-
rage case.20

16.3.2 Familiarity
Familiar risks, such as driving, engender less dread or outrage than unfamiliar
ones, such as the explosion of a nuclear power plant, bioterrorism, or the D.C.
snipers’ attacks—even when the familiar risks are more likely to eventuate. This
may respond to the fact that familiar risks tend to recede to the background and
do not gain one’s attention. Cass Sunstein suggests that the unacceptability of
unfamiliar risks may be a consequence of loss aversion:21
People will be closely attuned to the losses produced by any newly introduced
risk, or by any aggravation of existing risks, but far less concerned with the
benefits that are foregone as a result of regulation. . . . The opportunity costs
of regulation often register little or not at all, whereas the out-of-pocket costs
of the activity or substance in question are entirely visible. In fact this is a
form of status quo bias. The status quo marks the baseline against which
gains and losses are measured, and a loss from the status quo seems much
more bad than a gain from the status quo seems good.
. . . [Loss aversion] places a spotlight on the losses introduced by some risk,
and downplay[s] the benefits foregone as a result of controls on that risk. . . .
Consider the emphasis, in the United States, on the risks of insufficient test-
ing of medicines as compared with the risks of delaying the availability of
those medicines. . . . For many people, the possible harms of cloning register
more strongly than the potential therapeutic benefits that would be elimi-
nated by a ban on the practice.

19. Robert Scheuplein, Do Pesticides Cause Cancer?, Consumers’ Research 30–33

(December 1991).
20. Peter M. Sandman, Neil D. Weinstein, and William K. Hallman, Communications
to Reduce Risk Underestimation and Overestimation, 3 Risk Decision & Policy 93 (1998).
The “effect of outrage was practically as large as the effect of a 4000-fold difference in risk
between the high-risk and low-risk conditions.”
21. Laws of Fear 36.
500 problem solving, decision making, and professional judgment

16.3.3 Voluntariness and Controllability22

People are more willing to endure risks that they voluntarily assume or over
which they believe they have some control. For example, the risks associated
with rock-climbing or driving arouse less concern than the risks of an airplane
journey, even if the former are greater. Of course, whether or not a risk, such as
getting on an airplane, is voluntary is essentially a matter of individual, subjec-
tive framing.

16.3.4 (Un)natural Origins

People assume that nature is essentially benign, and find its risks less troubling
than those created by human technology. Cass Sunstein suggests that, aided by
loss aversion, this creates a bias against innovative technologies and
substances:23
Many people fear that any human intervention will create loss from the status
quo and that this loss should carry great weight, whereas the gains should be
regarded with some suspicion or at least be taken as less weighty. Often loss
aversion and a belief in nature’s benevolence march hand-in-hand: the status
quo forms the baseline or reference state against which to assess deviations.
Processes that interfere with nature seem, on the part of many, to be taken as
troubling “degradation”—whereas gains or improvements seem, other things
being equal, far less significant.
Thus, people believe—often incorrectly—that natural substances are inher-
ently safer than manufactured ones. For example, they overestimate the carcino-
genic risk from pesticides and underestimate the risks of natural carcinogens,24
or overestimate the dangers from genetically modified foods while underesti-
mating those of natural foods.25

16.3.5 Omission and Action Biases

In our earlier discussion of regret (Section 14.2), we noted that, in the short run,
actions tend to be more salient and more regret-inducing than failures to act. As
every first-year law student learns, tort, criminal law, and even constitutional
law26 often distinguish between actions and omissions: the actor who actively

22. These and other factors are analyzed in Paul Slovic, The Perception of Risk
(London: Earthscan Publications, 2000) [hereafter, Perception of Risk].
23. Laws of Fear 37.
24. See Perception of Risk at 291.
25. See James P. Collman, Naturally Dangerous (Sausalito, CA: 2001). Collman
writes that organic foods, favored by many people on grounds of safety and health and
creating annual revenues of $4.5 billion in the United States alone, are “actually riskier to
consume than food grown with synthetic chemicals.” (31)
26. See Washington v. Glucksberg, 521 US 702 (1997) (distinguishing between right to
have life support withdrawn and physician-assisted suicide).
 the role of affect in risky decisions 501

causes a harm is held liable, while someone who fails to prevent a harm usually
is not. The distinction is manifested in many moral intuitions as well.27 Among
other things, it reflects the generalization that it is easier to attribute causal
responsibility to action than inaction. From childhood, we are taught not to cause
direct harm, while the harms of inaction often seem indirect.
But if Hippocrates’ mandate, “First, do no harm,” captures a preference for
inaction where intervention may be counterproductive, the familiar injunction,
“Don’t just sit there. Do something!” also reflects a strong tendency.
Research suggests that people can have both an omission and an action bias,
depending on the circumstances and also on individual differences.
In a study modeled on the facts of the DPT vaccine, which causes serious
permanent neurological injury in one dose out of 310,000, Ilana Ritov and
Jonathan Baron found that many subjects preferred to subject their children to a
higher probability of dying from a disease that the vaccine would prevent, than a
smaller chance of dying from side-effects of the vaccine.28 Although a later study
by other Terry Connolly and Jochen Reb29 did not replicate the findings of omis-
sion bias with respect to vaccination, they concluded that anticipating regret was
nonetheless a significant factor in people’s decision whether to vaccinate: “inten-
tion to vaccinate was predicted by three measures: the respondent’s assessment
of the relative seriousness of the disease and the vaccine side-effects; her assess-
ment of the regret she would feel if vaccination turned out badly; and her assess-
ment of the regret she would feel if non-vaccination turned out badly.”
A later article by Baron and Ritov documents omission bias in various contexts,30
but also notes the existence of action bias. As in the case of the soccer goalie’s pen-
chant for action described in Section 14.2.2, a decision maker’s response may be
influenced by what is deemed “normal” behavior under the circumstance. By its
nature, a bias in either direction detracts from optimal decision making.
While the unfortunate outcome of an individual decision may result only in
personal regret, officials may incur public blame for bad policy decisions.
Officials in the Food and Drug Administration might well believe that criticisms
for delaying the approval of potentially life-saving drugs are mild compared to
the criticism of approving a drug that turns out to be deadly. One can think of

27. See Cass Sunstein, “Moral Heuristics,” University of Chicago Law & Economics,
Olin Working Paper No. 180 (2003).
28. Ilana Ritov and Jonathan Baron, Reluctance to Vaccinate: Omission Bias and
Ambiguity, in Behavioral Law and Economics 168 (Cass Sunstein ed., New York:
Cambridge University Press, 2000).
29. Terry Connolly and Jochen Reb, Omission Bias in Vaccination Decisions: Where’s
the “Omission”? Where’s the “Bias?, 91 Org. Behav. and Human Decision Processes
186–202 (2003).
30. Jonathan Baron and Ilana Ritov, Omission Bias, Individual Differences, and Normality,
94 Organizational Behavior and Human Decision Processes 74–85 (2004).
502 problem solving, decision making, and professional judgment

many other instances of omission bias by public officials. But there are also
apparent instances of action bias, perhaps including the Bush Administration’s
decision to invade Iraq after 9/11.

16.3.6 Betrayal Aversion

Betrayal creates a perception of social disorder beyond the particular damage it
causes, and harms caused by a violation of trust are accompanied by a particular
sense of anger or outrage. As with human agents, we may feel betrayed by objects
that cause harms of the sort they were intended to guard against, for example,
an asthma medicine that constricts one’s airway or a sunscreen that causes
skin cancer.31 One of the subjects in the vaccine study just mentioned explained
his reluctance to vaccinate in these terms: “I feel that if I vaccinated my kid and
he died I would be more responsible for his death than if I hadn’t vaccinated him
and he died . . .”
And, as in the case of omission bias, people’s aversion to the risk of betrayal
by a product may lead to seemingly suboptimal decisions. A noteworthy example
comes from an experiment in which Jonathan Koehler and Andrew Gershoff
gave participants two scenarios involving the role of air bags in fatal car acci-
dents. The betrayal scenario was:
Suppose that you are offered a choice between two equally priced cars: Car A
and Car B. Car A is equipped with Air Bag A. Scientific crash tests indicate
that there is a 2% chance that drivers of Car A who are in serious accidents
will be killed due to the impact of the crash. Car B is equipped with Air Bag B.
Scientific crash tests indicate that there is a 1% chance that drivers of Car B
who are in serious accidents will die due to the impact of the crash. However,
Car B may kill drivers who would not have died if they were driving Car A
instead. Specifically, some drivers of Car B may die due to trauma caused by
the force of the air bag deployment. Crash tests indicate that there is an addi-
tional one chance in 10,000 (0.01%) that someone who is in a serious accident
in Car B will be killed due to air bag trauma.
In the nonbetrayal situation, the risk of death due to air bag deployment was
replaced by a risk of death due to toxic fume inhalation from a damaged engine—
not a betrayal of the safety device itself.
Participants were asked which car they would prefer and how strong their
preference was. Most people in the nonbetrayal situation preferred Car B, which
presents roughly half the probability of death (1.01 percent) than Car A (2 per-
cent). However, most participants in the betrayal scenario were willing to double
the likelihood of death to avoid betrayal. A typical explanation was, “I’d hate to

31. Jonathan J. Koehler and Andrew Gershoff, Betrayal Aversion: When Agents of
Protection Become Agents of Harm, 90 Organizational Behavior and Human Decision
Processes 244 (2003).
 the role of affect in risky decisions 503

have the air bag that’s supposed to save my life be the cause of its termination.”
The authors consider and reject the possibility that the phenomenon is a result
of omission bias, since the participants were forced to actively choose between
the two safety devices.

16.3.7 Judgments of Risk and Benefit

The presence or absence of evident benefits may mitigate or aggravate people’s
fear or outrage over a risk. But there also is evidence that people do not consider
risk and benefit separately. Rather, the perception of the risk and benefit of an
activity, product, or technology (e.g., alcohol, guns, food coloring, pesticides,
nuclear power plants, and vaccines) are not independent but are in fact inversely
correlated. The relationship seems related to the strength of positive or negative
affect associated with that activity. Ali Alhakami and Paul Slovic write that
“people base their judgments of an activity or a technology not only on what they
think about it but also on what they feel about it. If they like an activity, they are
moved to judge the risks as low and the benefits as high; if they dislike it, they
tend to judge the opposite—high risk and low benefit.”32 In addition, “people
operate under a strong need for consistency among their beliefs and attitudes.
When people view an activity or technology as good, pressure toward consistency
would lead them to judge its benefits as high and its risks as low, and vice versa
for activities seen as bad.”
The phenomenon is not limited to laypersons. When members of the British
Toxicological Society were asked to give a quick intuitive affective rating of
chemicals (e.g., benzene, aspirin, second-hand smoke) on a good-bad scale and
then to judge the risk associated with a very small exposure to the chemical, nega-
tive affect was correlated with high risk and positive affect with low risk.33
Whether through this affective phenomenon or through a more cognitive
process, the anxiety or outrage caused by a hazard may dissipate when people
become aware of costs of eliminating it. For example, Howard Margolis relates
that there was overwhelming popular sentiment for closing New York city
schools that contained asbestos, but that once parents had to deal with the real
inconveniences of closed schools, the perceived magnitude of the risk dimin-
ished.34 Along the same lines, Cass Sunstein writes that Finland, whose econ-
omy depends heavily on Nokia, is not among the countries whose citizens and
policy makers are concerned about the health effects of cell phones.35 He notes

32. Ali Alhakami and Paul Slovic, A Psychological Study of the Inverse Relationship
Between Perceived Risk and Perceived Benefit, 14 Risk Analysis 1085 (1994).
33. See Affect Heuristic, supra.
34. Howard Margolis, Dealing with Risk: Why the Public and the Experts Disagree
on Environmental Issues 124–125 (Chicago: University of Chicago Press, 1998).
35. Cass Sunstein, Worst-Case Scenarios 219 (Cambridge: Harvard University
Press, 2007) [hereafter, Worst Case Scenarios].
504 problem solving, decision making, and professional judgment

that when the U.S. Environmental Protection Agency (EPA) called attention to
the toxic risks of a mining town’s major industry, the citizens did not respond by
taking precautions, but rather by demonizing the EPA.

16.3.8 Cultural Determinants of Perceptions of Risk

We noted earlier that different people may have different risk attitudes, whether
as a result of different perceptions of risk, or views of the benefits, or appetites
for risk-taking. In a review of Sunstein’s Laws of Fear: Beyond the Precautionary
Principle, Dan Kahan, Paul Slovic, Donald Braman, and John Gastil argue that
“culture is prior to facts in societal disputes over risk”:36
Normatively, culture might be prior to facts in the sense that cultural values
determine what significance individuals attach to the consequences of envi-
ronmental regulation, gun control, drug criminalization, and the like. But
more importantly, culture is cognitively prior to facts in the sense that cultural
values shape what individuals believe the consequences of such policies to be.
Individuals selectively credit and dismiss factual claims in a manner that sup-
ports their preferred vision of the good society.
The priority of culture to fact is the organizing premise of the “cultural
theory of risk.” Associated most famously with the work of anthropologist
Mary Douglas and political scientist Aaron Wildavsky, the cultural theory of
risk links disputes over environmental and technological risks to clusters of
values that form competing cultural worldviews—egalitarian, individualistic,
and hierarchical. Egalitarians, on this account, are naturally sensitive to envi-
ronmental hazards, the abatement of which justifies regulating commercial
activities that produce social inequality. Individualists, in contrast, predictably
dismiss claims of environmental risk as specious, in line with their commit-
ment to the autonomy of markets and other private orderings. Hierarchists are
similarly skeptical because they perceive warnings of imminent environmental
catastrophe as threatening the competence of social and governmental elites.
Although one can imagine alternative explanations for cultural variation in
risk perceptions, cultural cognition offers a distinctively psychometric one.
On this view, the impact of cultural worldviews is not an alternative to, but
rather a vital component of, the various psychological and social mechanisms
that determine perceptions of risk. These mechanisms, cultural cognition
asserts, are endogenous to culture. That is, the direction in which they point
risk perceptions depends on individuals’ cultural values. . . .
In sum, individuals adopt stances toward risks that express their commit-
ment to the values integral to their preferred ways of life.

36. Dan M. Kahan, Paul Slovic, Donald Braman, and John Gastil, Fear of Democracy:A
Cultural Evaluation of Sunstein on Risk, 119 Harv. L. Rev. 1071 (2006).
 the role of affect in risky decisions 505

Kahan, Slovic, et al. criticize Sunstein for, in effect, privileging experts’ per-
ceptions of risk over those of laypersons because he believes that laypersons are
more prone to the cognitive and affective biases described in this and the preced-
ing chapters. The critics doubt that policies concerning major environmental
and other issues involving risk can be adequately addressed only through cost-
benefit, or expected value, analysis.
Our own view is that issues of “cultural cognition” do play a role in people’s
perception of risk and that it is often difficult to separate cognitive and affective
biases from an individual’s values and tastes in determining his or her risk
preferences. However, as Sunstein notes in a response to his critics, most of the
risks that we face involve rather quotidian issues rather than divisive cultural and
political issues.37

16.4 risk vs. uncertainty (or ambiguity)

16.4.1 The Concept and Psychology of Uncertainty

The examples in Chapter 15 and thus far in this chapter deal with known probabil-
ities: Rachlinski’s litigants were told that they had a 50 percent chance of winning
or losing; the physicians in the medical hypothetical knew the risks of particular
treatments. But the probabilities attending most real-world decisions do not come
nearly so specified. In many cases, decision makers lack the data needed to make
estimates based on frequentist statistics, and are relegated to one-off guesses
where not even base rates are well known. Moreover, even when presented with
frequentist data, a real-world decision maker may have reason to question their
validity because of sample size, bias, collection methods, or other factors.
Drawing on an important paper by Frank Knight,38 the economic literature
distinguishes between a risk (known probability) and uncertainty (unknown prob-
ability). In a review of the literature, the psychologists Deborah Frisch and
Jonathan Baron use the word ambiguity as a synonym for uncertainty, defining
ambiguity as the subjective experience of missing information.39
Regardless of their risk preferences, people tend to be averse to uncertainty,
or ambiguity. Indeed, Ellsberg’s Paradox40 demonstrates how uncertainty aver-
sion leads to a violation of the cancellation principle of expected utility theory: a
choice between two alternatives should depend only on how those alternatives
differ, and not on any factor they have in common.

37. Cass Sunstein, Misfearing, 119 Harv. L. Rev. 1110 (2006).

38. Frank Hyneman Knight, Risk, Uncertainty, and Profit (Boston: Houghton
Mifflin, 1921).
39. Deborah Frisch and Jonathan Baron, Ambiguity and Rationality, 1 J. Behavioral
Decision Making 149 (1988).
40. The paradox was documented by the economist Daniel Ellsberg. The following text
is taken from http://en.wikipedia.org/wiki/Ellsberg_paradox.
506 problem solving, decision making, and professional judgment

Suppose you have an urn containing thirty red balls and sixty other balls that
are either black or yellow. You don’t know how many black or yellow balls there
are, but you know that the total number of black balls plus the total number of
yellow equals sixty. The balls are well mixed so that each individual ball is as
likely to be drawn as any other. You are now given a choice between two gambles,
as shown in Table 16.2a.

table 16.2A ellsberg’s paradox: gambles a and b

Gamble A Gamble B

You receive $100 if you draw a red ball You receive $100 if you draw a black ball

Also you are given the choice between these two gambles (in a different draw
from the same urn), as shown in Table 16.2b.

table 16.2B ellsberg’s paradox: gambles c and d

Gamble C Gamble D

You receive $100 if you draw a red or You receive $100 if you draw a black or
yellow ball yellow ball

Expected utility theory holds that you should prefer Gamble A to Gamble B if,
and only if, you believe that drawing a red ball is more likely than drawing a black
ball. Thus, there would be no clear preference between the choices if you thought
that a red ball was as likely as a black ball. Similarly it follows that you will prefer
Gamble C to Gamble D if, and only if, you believe that drawing a red or yellow
ball is more likely than drawing a black or yellow ball. If drawing a red ball is
more likely than drawing a black ball, then drawing a red or yellow ball is also
more likely than drawing a black or yellow ball. So, if you prefer Gamble A to
Gamble B, it follows that you will also prefer Gamble C to Gamble D. And, if
instead that you prefer Gamble D to Gamble C, it follows that you will also prefer
Gamble B to Gamble A. However, most people prefer Gamble A to Gamble B
and Gamble D to Gamble C.
Why is this paradoxical—or at least a logical error? A preference for Gamble A
to B implies that you believe that there are fewer black balls than red balls – i.e.,
fewer than 30 black balls, which entails more than thirty yellow balls. A preference
for Gamble D to C implies that you believe that there are more black balls than red
balls. So a preference for A and D implies that you believe the number of black
balls is less than the number of red balls and more than the number of red balls.
 the role of affect in risky decisions 507

Most policy and legal decisions must be made in conditions of uncertainty

rather than risk. To illustrate the difference in a legal context, consider the differ-
ence of a litigant being informed: (1) “Based on a database of jury verdicts and
his extensive experience, your lawyer believes that there’s a 50 percent chance of
a judgment in your favor,” or (2) “After much hesitation, your lawyer’s best guess
is a 50 percent chance; but given the nature of the case, he feels very uneasy
about providing you with a number.”41
In either set of choices, one’s best guess is that the probability is 0.5. But
(except at low probabilities) most people tend to be averse to the second uncer-
tain, or ambiguous, option. Indeed, people often make different choices in the
two situations.42 For example, given a fifty–fifty chance of winning where the
odds are pretty well known, most defendants follow the predictions of prospect
theory and incur the risk of litigating. Given the same odds of winning in the
uncertainty situation, a majority of defendants manifest ambiguity aversion and
choose to settle.43
Ambiguity aversion (correctly) predicts that someone making investment
decisions from scratch would prefer to buy securities whose recent performance
was known rather than unknown. Thus, when experimental subjects were given
a choice between:
• the unambiguous option of purchasing securities which during the past
ninety days had gone up thirty times, down thirty times, and remained the
same thirty times, and
• the ambiguous option of purchasing securities which had remained the
same thirty times in the past ninety days, with no information about what
happened the other sixty days,
they opted for the former. But when they were already endowed with the ambig-
uous option, they preferred to hold onto it rather than exchange it for the unam-
biguous one.44

41. Cf. Robin Hogarth and Howard Kunreuther, Decision Making Under Uncertainty: The
Effects of Role and Ambiguity, in Decision Making and Leadership 189 (Frank Heller, ed.,
New York: Cambridge University Press, 1992).
42. Noting that even when frequentist statistics cannot produce “objective” probabili-
ties, people often assign subjective probabilities to events, some commentators have ques-
tioned whether uncertainty actually exists. Worst-Case Scenarios 159 ff. But when the
range of one’s subjective estimate of the probability of an event occurring is sufficiently
large, assigning a numerical probability seems pointless.
43. Hogarth and Kunreuther, supra. It appears that, in making judgments under uncer-
tainty a decision maker anchors on the given probability, but then adjusts the estimate
based on imagining other values it might have.
44. Merce Roca, Robin M. Hogarth, and A. John Maule, Ambiguity Seeking as a Result
of the Status Quo Bias, 32 Journal of Risk and Uncertainty 175 (2006).
508 problem solving, decision making, and professional judgment

Ambiguity aversion may be based on several factors:

• in situations of ambiguity, an opponent may have information not available
to you;
• because the dispersion of a series of ambiguous gambles is greater than
that of a series of gambles with known risks, a series of ambiguous gam-
bles (with the identical missing information) is more volatile that a series
of nonambiguous gambles;
• in ambiguous situations, a decision maker might wait and obtain more
information; he may regret not having obtained the missing information,
or may be blamed by others for not obtaining it.
For these reasons, people may develop a heuristic that disfavors ambiguous
gambles.
The availability heuristic may combine with the phenomenon of ambiguity aver-
sion. Referring to the Washington, D.C. snipers, Kip Viscusi suggests that this is a
“new sort of risk—and people do not know how to evaluate something they have
never seen before. There has never been a serial killer like the sniper before. . . . The
sniper has killed randomly at a distance. This means that no one in the Washington
area can give any reason why he or she should not be the next victim.”45

16.4.2 The Principle of Continuity (or Exchange), Maximin, and

“Robust Satisficing”
The principle of continuity, an axiom of expected utility theory relevant to deci-
sion making under risk is that an individual has some certainty equivalent for
any lottery. To put it formally: Suppose A, B, and C are three prospects (possible
outcomes), and the decision maker prefers A to B, and prefers B to C. The axiom
provides that there must be some probability p such that the decision maker is
indifferent between choosing either prospect B with certainty or a lottery in
which prospect A occurs with probability p, or prospect C occurs with probability
1-p. We call this probability p a preference probability, and we call B the certainty
equivalent of the A-C lottery.

p
B ~
1-p

figure 16.2 certainty equivalent of a lottery.

45. The Economist, Oct. 19, 2002.

 the role of affect in risky decisions 509

But, citing the writing of Peter Vallentyne, Larry Temkin writes:46

Many reject continuity in cases like the following. Consider three possible
outcomes. In B one has a full rich life; in A, a full rich life, with an extra dollar
of income; in C, a life of excruciating pain and misery. Vallentyne denies that
there is any p such that he ought, rationally, to be indifferent between the
prospect of either A with a probability p or C with a probability 1-p.
Examples like Vallentyne’s are often thought to trade on our attitudes
regarding certainty and risk. But the issue is not merely one of certainty versus
risk. People do not generally mind risking a certain outcome for a slightly
better outcome if the downside risk is small; what they mind is risking a cer-
tain outcome for a slightly better one if the downside risk is high. What seems
crazy, in Vallentyne’s example, is trading the certainty of a full rich life for the
prospect of a slightly better one, when doing so means risking—however
minimally—a life of pain and misery.
While violating an axiom of expected utility theory, the decision maker in
Vallentyne’s example may be acting consistently with the principle of maximin,
which counsels: “choose the policy with the best worst-case outcome.” One
might think of Pascal’s Wager —which addresses a situation of uncertainty or
ambiguity rather than of specifiable known risk—in similar terms. Pascal argued
against atheism on the ground that even though one couldn’t know in advance
whether those who believe in God will be rewarded in the afterlife and those who
don’t will burn perpetually in hell, it was rational to believe since believing had
no present costs.
Although the maximin principle captures a strong intuition, it is difficult to
know when to apply it. Often, eliminating one worst-case outcome simply reveals
another one.47
In Section 1.6.2, we introduced the notion of satisficing—settling for a good
enough rather than optimal outcome when the informational and cognitive costs
of making a decision are too high. A recent paper by Yakov Ben Haim, Cliff
Dasco, and Barry Schwartz proposes a normative model of robust satisficing48 for
decision making in conditions of severe uncertainty. In contrast to expected util-
ity theory, robust satisficing does not seek to maximize expected utility. And in
contrast to maximin, robust satisficing does not try to guard against the “worst-
case scenario.” Rather, it seeks to give the decision maker the best chance of
enjoying a good enough outcome.

46. Larry S. Temkin, Weighing Goods: Some Questions and Comments, 23 Phil. & Public
Affairs 350 (1994).
47. See Worst-Case Scenarios, supra.
48. Yakov Ben Haim, Cliff Dasco, and Barry Schwartz, “What Makes a Good Decision?
Robust Satisficing as a Normative Standard of Rational Decision Making” (unpublished
paper 2009).
510 problem solving, decision making, and professional judgment

Suppose that you are considering a number of entry-level job offers in a large
government agency, and that your criteria include (1) salary and other benefits,
(2) the quality of city life, (3) whether the work will be interesting and fulfilling,
and (4) the quality of support and mentorship you will receive. You can ascertain
the first two within a small margin of error, but the latter two are deeply uncer-
tain: they depend on what department or division you will be assigned to and on
the particular people you work for, which cannot be known in advance. Ben
Haim et al. propose that rather than attempting to determine which decision
course will yield the highest expected value, you should identify a threshold of a
satisfactory, or “good enough,” outcome and choose the job that is most robust
to uncertainty—that is most likely to produce a satisfactory outcome. Similarly,
robust satisficing would be a useful decision-making strategy for a man pre-
sented with a choice of treatments for prostate cancer, which have different
chances of curing the cancer and different likely side effects. Even if some of
these probabilities can be specified, your prediction of how side effects will affect
your future wellbeing fall into the realm of uncertainty.
The mathematics of robust satisficing is pretty complicated—more so than
for expected utility. Ben Haim, Dasco, and Schwartz suggest these (satisficing)
rules of thumb:
• Ask yourself what you need to achieve in order to be happy or satisfied with
the outcome of the decision. . . . Remember that high aspirations are more
vulnerable to error than modest ones.
• Distinguish between two very different attributes of each option. One attri-
bute of an option, which we call its nominal outcome, is the estimate of the
outcome of that option based on your best data and understanding. . . . The
other attribute of an option, called its robustness, is the degree of immunity
of the outcome to error in your estimates and understanding. The robust-
ness is a nonprobabilistic assessment of immunity to error. Nonetheless,
robustness may assess immunity to error in assessments of probabilities.
An option may be quite attractive based on your current understanding
(good nominal outcome), but may lead to highly undesirable results due to
only small errors (low robustness). Alternatively, an option may be both
nominally very attractive and very robust to error. Or, an option may nomi-
nally be only moderately attractive, but it may be very robust to error. These
two attributes—nominal outcome and robustness—are different and not
correlated with one another.
• Evaluate the nominal outcome of each option. This is relatively easy,
since you can ignore uncertainty and assume that your data and under-
standing are correct. Rank the options from highest to lowest nominal
outcome.
• Evaluate the robustness of each option for the particular quality of outcome
that you identified in the first step. This is more difficult, since you must ask:
 the role of affect in risky decisions 511

How wrong can I be and still have the option yield a satisfactory outcome?
Listing contingencies (this could go wrong, that could go wrong, etc.) is one
approach. . . . Rank the options from highest to lowest robustness.
• Choose the most robust option. If the rankings according to nominal out-
come and robustness agree, then the most robust option will also be the
nominally most attractive option. If the rankings according to nominal out-
come and robustness disagree, then the most robust option may differ
from the nominally most attractive option.
• Step back and look carefully at your analysis. Does it make sense? . . . Ask
if your thought process fits together coherently, if it uses all relevant infor-
mation, if it avoids major assumptions or leaps of faith, and if it is free of
considerations that are not explicitly included.

16.5 complexities of evaluating risks

16.5.1 Evaluability and Framing Redux

Recall the issues of evaluability mentioned in Section 13.3.1. Howard Kunreuther
and his coauthors did a series of experiments to explore people’s sensitivities to
significant differences in low probability risks of high consequence events.49
Participants were asked to assess the relative riskiness of chemical plants that
had a 1/100,000, a 1/1 million, or a 1/10 million chance of releasing toxic gases.
The questions asked were: (1) How serious a risk would the plant pose to the
residents’ health and safety? and (2) How close they would be willing to live to
the plant?
In the within-subjects group, participants who were asked the questions about
all three plants simultaneously provided assessments of riskiness commensurate
with the probabilities. In other words, the incremental differences in their three
responses to the two questions were roughly equal to the incremental differ-
ences between the three proposed probabilities. However, in a between-subjects
experiment, three separate groups of participants were respectively shown each
of the three plants. Here there was no statistically significant difference in their
mean risk assessments; the feelings evoked by the images of the power plants
overwhelmed the probabilities, and the participants based their risk assessments
almost entirely on their emotional responses to the images.50

49. Howard Kunreuther, Nathan Novemsky, and Daniel Kahneman, Making Low
Probabilities Useful, 23 J. Risk & Uncertainty 103 (2001).
50. Hypothesizing that the chemical plant risks would be more evaluable if partici-
pants were able to compare them to more familiar risks, like a car accident, the experi-
menters told subjects that the probability of injury in a car accident was 1/6000 per year
and asked them to evaluate a 1/650, 1/6300, and 1/68,000 chance of the plant’s release of
toxics. The differences among the groups were still negligible. However, when subjects
512 problem solving, decision making, and professional judgment

One way in which we evaluate unfamiliar risks is to compare them to risks

that are more familiar. Hypothesizing that the chemical plant risks would be
more evaluable if people were able to compare them to more familiar risks, like
a car accident, the experimenters told participants that the probability of injury
in a car accident was 1/6000 per year and asked them to evaluate a 1/650, 1/6300,
and 1/68,000 chance of the plant’s release of toxics. The differences among the
groups were negligible. However, when subjects were given scenarios describ-
ing the contexts in which car accidents happened with relatively and low and
high probabilities (1/5900 chance of accident on icy mountain roads; 1/66,000
on flat desert highways), the differences in assessing the risks from the plant
were significant. The scenarios made the risks more evaluable than the statistics
alone. “There needs to be fairly rich context information available for people to
be able to judge differences between low probability events. In particular, people
need comparisons of risk that are located on the probability scale and evoke
people’s own feelings of risk.”51
People sometimes use “natural” levels of risk as a baseline for comparison.
For example, in Peter Sandman’s study of high outrage/low risk (radon) and low
outrage/high risk (nuclear waste) situations described in Section 16.3.1, partici-
pants were insensitive to changes in risk levels phrased in terms such as: “For
every 1000 people exposed to this level of radiation over a lifetime, 40 more of
them would get lung cancer than if they were not exposed.” But they were highly
sensitive to changes based on normal background radiation, e.g.: “The radiation
in your home is 20 times greater than the average outdoor background level.”
When a risk adds only a very small percentage to normal background, people
believe that it is not too serious; when it adds a large multiple of normal back-
ground, they conclude that it is quite serious. This point of comparison likely
draws on the intuition (noted earlier) that naturally caused conditions are not
harmful. The problem, as the experimenters note, is that
[f]or some hazards, normal background levels are sufficient to constitute a
meaningful health risk, and even a small increment would be unwise if it
were preventable. For other hazards, the risk due to normal background expo-
sure is negligible, and an exposure many times the background level would

were given scenarios describing the contexts in which car accidents happened with rela-
tively and low and high probabilities (1/5900 chance of accident on icy mountain roads;
1/66,000 on flat desert highways), the differences were significant. The scenarios made
the risks more evaluable than the statistics alone. “There needs to be fairly rich context
information available for people to be able to judge differences between low probability
events. In particular, people need comparisons of risk that are located on the probability
scale and evoke people’s own feelings of risk.” Kunreuther, et al., supra.
51. Michael Jones Lee and Graham Loomis, Private Values and Public Policy, in Elke
Weber, Jonathan Baron, and Graham Loomis, Conflict and Tradeoffs in Decision
Making 205 (Cambridge, UK: Cambridge University Press, 2001).
 the role of affect in risky decisions 513

still be negligible. Thus, comparisons to background can give misleading

impressions contrary to actual risk magnitudes.
A risk can seem greater or smaller based on the period during which it is
framed. For example, people who are not motivated to wear seat belts to avoid a
0.00001 chance of a serious accident per car trip are more likely to wear one
when informed of a 0.33 chance over a fifty-year lifetime of driving.52

16.5.2 Cognitive and Affective Processing of Probabilities

In Section 14.6, we mentioned that, when contemplating a harm that instills an
affective response (e.g., birds dying because of an oil spill), people may abandon
the quantitative evaluation of costs and benefits, and their willingness to pay
(WTP) to avoid the harm may become orthogonal to the scope of its harm. More
generally, Christopher Hsee and Yuval Rottenstreich note that valuation by calcu-
lation and valuation by feeling can produce quite different results.53 A person may
respond to the same event with one or the other valuation system, depending on
how it is framed: consider the difference between statistics about the number of
birds at risk and a photo of a bird covered with oil slick.
Jeremy Blumenthal summarizes an interesting line of research indicating
that people’s affective response to risk may depend on whether the same proba-
bility is described with larger or smaller numbers:
People behave as though certain low-probability events are less probable when
represented by equivalent ratios of smaller numbers (1 in 10) than of larger
numbers (10 in 100). That is, people saw the likelihood of Event X as greater
when the probability was expressed as a 10 in 100 chance than when it was
expressed as 1 in 10, and behaved accordingly. Even more striking, . . . despite
objective information that Event X had, for instance, a 7 out of 100 chance of
occurring versus a 1 in 10 chance, respondents chose and behaved as though
the former were more likely. Respondents explained that rationally and objec-
tively, they understood that the likelihood was lower; emotionally and subjec-
tively, however, they felt they had a better chance when the absolute likelihood
appeared higher (i.e., 7 chances rather than 1), and thus actually chose as
though they had a better chance of obtaining X under those circumstances.54

52. Paul Slovic, Baruch Fischhoff, and Sara Lichtenstein, Accident Probabilities and Seat
Belt Usage: A Psychological Perspective, 10 Accident Analysis and Prevention 281 (1978).
53. Christopher K. Hsee and Yuval Rottenstreich, Music, Pandas, and Muggers: On the
Affective Psychology of Value, 133 J. Experimental Psychol.: Gen. 23, 23–24 (2004).
54. Jeremy A. Blumenthal, Emotional Paternalism, 35 Fla. St. U. L. Rev. 1, 23 (2007),
discussing Veronika Denes-Raj and Seymour Epstein, Conflict Between Intuitive and Rational
Processing: When People Behave Against Their Better Judgment, 66 J. Personality & Soc.
Psychol. 819, 823 (1994); Seymour Epstein and Rosemary Pacini, Some Basic Issues
Regarding Dual-Process Theories from the Perspective of Cognitive-Experiential Self-Theory, in
514 problem solving, decision making, and professional judgment

People also have different perceptions of risks described as probabilities or

frequencies.55 Paul Slovic and his colleagues asked experienced forensic psycholo-
gists and psychiatrists to assess the likelihood that a mental patient would
commit an act of violence within six months after being discharged from
the hospital.56 One group of participants was given an expert’s prior assessment
of the risk of violence phrased in terms of probabilities, e.g., “patients similar to
Mr. Jones are estimated to have a 20 percent chance of committing an act of
violence.” Another group was given the same assessment phrased in terms of
relative frequencies, e.g., “of every one hundred patients similar to Mr. Jones,
twenty are estimated to commit an act of violence.”
Of those given the data in terms of probabilities, only 21 percent refused to
discharge the patient. Of those given the data in terms of relative frequencies, 41
percent refused to discharge the patient. Those given the probabilistic format
had relatively benign images of the patient, while those given the frequentistic
format conjured up frightening images of violent patients. “These affect-laden
images likely induced greater perceptions of risk in responsive to the relative-
frequency frames.”57 Along the same lines, a disease that kills 1,286 people out
of every 10,000 was judged more dangerous than one that kills 24.14 percent of
the population!58

16.5.3 Worry and Probability Neglect

Given the difficulties of evaluability, people may sometimes ignore probabilities
because they are just too difficult to reckon, or (given boundedly rational actors)
just not worth their attention. Thus, a study of why people don’t insure against
large, low-probability losses suggests that the costs of calculating the appropriate
premium are too high.59 And a study of risky business decisions suggests that
managers do not even ask for data on the probability of outcomes.60

Dual-Process Theories in Social Psychology 462 (Shelly Chaiken and Yaacov Trope
eds., 1999).
55. Recall the suggestion in Section 8.6.1 that describing events in terms of frequen-
cies rather than probabilities could, under some circumstances, improve individuals’
intuitive Bayesian analysis.
56. Slovic, Monahan, and MacGregor, Violence risk assessment and risk communica-
tion: The Effects of Using Actual Cases, Providing Instruction, and Employing Probability
Versus Frequency Formats, 24 Law Hum Behav. 271–96 (2000).
57. Affect Heuristic, supra.
58. Kimihiko Yamagishi, When a 12.86% Mortality Is More Dangerous than 24.14%:
Implications for Risk Communication, 11 Applied Cognitive Psychology, 495 (1997).
59. Howard Kunreuther and Mark Pauly, Neglecting Disaster: Why Don’t People Insure
Against Large Losses, 28 J. Risk and Uncertainty 5 (2004).
60. Oswald Hoberet al., Active Information Search and Complete Information Presentation,
95 Acta Psychologica 15 (1997).
 the role of affect in risky decisions 515

Information costs aside, a variety of decisions—from investments to willing-

ness to insure against losses—seem to be based more on a subjective sense of
“worry” or “concern” than on a probabilistic assessment of gains and losses.61
Indeed, in trying to avoid catastrophic risks, such as climate change, people may
be prone to what Elke Weber terms “single action bias”: being satisfied with
taking only one action that mitigates their worry rather than responding with a

Insurance as Protection Against Worry

Boone’s older brother, Pete, is orienting Boone for his first day on the job at as an
insurance claim adjuster.

“These are claims,” his older brother explained, grabbing a stack of paper-
clipped and clamped wads of papers and forms from a bin outside the cubicle
marked IN. “Your job is to deny them.”
“I see,” Boone had said. “You mean, I sort through the claims and deny all
the fraudulent ones, right?”
His brother implored heaven for patience with a roll of his eyes, then sighed
a gust of wintry disgust. “The fraudulent claims were picked out downstairs by
high school graduates and denied three months ago. Anybody can deny a
fraudulent claim. You’re a college graduate. Your job is to find a way to deny
legitimate claims. . . .
“People who file claims believe that money will make them happy and will
somehow compensate them for their losses. This idea—that money makes
misfortune easier to bear—is an illusion that can only be enjoyed by those
who have not suffered an actual loss.
“The most terrifying thing about life is knowing that, at any moment, a
freak accident, violence, mayhem, a psychotic break, an addiction, a heart
attack, a sexually transmitted disease, cancer, an earthquake, or some other
act of God, or worse, can take all of your happiness away from you in the time

61. Elke U. Weber and Christopher Hsee, Cross-Cultural Differences in Risk Perception,
But Cross-Cultural Similarities in Attitudes Towards Perceived Risk, 44 Management
Science 1205 (1998); Ch. Schade, H. Kunreuther, and K.P. Kaas (2002), “Low-Probability
Insurance Decisions: The Role of Concern,” Discussion Paper Number 23, SFB 373,
Humboldt-Universität zu Berlin/Wharton Risk Center Working Paper Number 02-10-HK,
Wharton School, University of Pennsylvania, USA; Paul Slovic, Baruch Fischhoff, and
Sarah Lichtenstein, Facts and Fears: Understanding Perceived Risk, in Societal Risk
Assessment: How Safe Is Safe Enough? 181 (R. Schwing and W. A. Albers, Jr., eds.,
1980).
516 problem solving, decision making, and professional judgment

it takes you to pick up the phone and get the news. That’s why people buy
insurance, because they think it will protect them from catastrophes.
“But we are in the insurance business,” said Pete. . . . “We know there is no
protection from catastrophes. No matter what you do, there’s always a chance
that a catastrophe will come along, tear your heart out of your chest, and rub
it in your face.
“When you’re crawling on the bathroom floor sick with grief . . ., wondering
why God failed to give you the courage to kill yourself, a big check from the
insurance company looks like a swatch of wallpaper. You’re in a place money
can’t reach.
“So, insurance only works if catastrophe does not strike. . . . We don’t sell
protection. We sell peace of mind. For a premium, we agree to give the con-
sumer the illusion that money will protect him from every possible foreseeable
catastrophe. Once the premium is paid and before catastrophe strikes, the
consumer is free to wallow in the illusion that if something terrible happens
money will take the sting out of it. When a catastrophe actually occurs, the
illusion is shattered and there’s nothing to be done but drag yourself out of
bed every morning and get on with your life.”
“But if what you say is true,” asked Boone, “then you are charging people
thousands of dollars for . . . an illusion.”
“Exactly,” said Pete. “Peace of mind. The money is irrelevant. You probably
subscribe to the notion that insurance is a way to pool risk and share liability.
You think premiums should be based upon risk. Nothing could be more
wrong. Premiums should be based upon line thirty-one of your federal tax
return, adjusted gross income. Our objective is to charge the insured just
enough to make it hurt. We are looking for the financial pain threshold,
because only when it hurts does the insured really believe that he is obtaining
something of value, and, as I’ve shown, he is indeed obtaining peace of mind
for nothing more than money.”
RICHARD DOOLING, WHITE MAN’S GRAVE 25–27 (1994) Copyright © Richard Dooling
1994. Reprinted with permission.

set of actions that would more effectively reduce the risk.62 As Joseph Conrad
wrote in Nostromo, “Action is consolatory. It is the enemy of thought and the
friend of flattering illusions.”

62. Elke Weber, Perception and Expectation of Climate Change: Precondition for Economic
and Technological Adaptation, in Environment, Ethics, and Behavior: The Psychology
of Environmental Valuation and Degradation (M. H. Bazerman, D. M. Messick, A.
E. Tenbrunsel, and K. A. Wade-Benzoni eds., 1997).
 the role of affect in risky decisions 517

The stronger the emotions, the more people tend to greatly underweight
probabilities or to ignore them altogether and focus only on the horrific, worst
case outcome. Cass Sunstein has coined the term probability neglect to describe
people’s departure “from the normative theory of rationality in giving excessive
weight to low-probability outcomes when the stakes are high” and giving low-
probability outcomes no weight at all when the risks are not vivid.
The phenomenon is illustrated by an experiment, based on an actual problem
that faced the Environmental Protection Agency. Sunstein asked law students to
indicate their willingness to pay (WTP) to reduce levels of arsenic in drinking
water to eliminate
• a cancer risk of 1/1,000,000;
• a cancer risk of 1/100,000;
• a cancer risk of 1/1,000,000 where the cancer was described in vividly grue-
some terms;
• a cancer risk of 1/100,000 also described in gruesome terms.
When given the unemotional description, people’s WTP increased signifi-
cantly as the probability of contracting cancer increased. Merely describing the
cancer in gruesome terms doubled people’s WTP to avoid the 1/1,000,000 risk,
but the tenfold increase of gruesome cancer risk did not greatly increase WTP.
Sunstein argues that probability neglect does not involve misestimating proba-
bilities based on their vividness or availability. Rather, emotion essentially
swamps considerations of probability.63
Laypersons’ intuitions about toxicology also manifest probability neglect.
Nancy Kraus, Torbjörn Malmfors, and Paul Slovic compared the basic attitudes
of professionals and laypersons toward toxic risks.64 A core assumption of toxi-
cology is that “the dose makes the poison,” meaning that there is a positive cor-
relation between the size of the dose and the likelihood of harm and that some
chemicals that are deadly in high concentrations are harmless in small amounts.
However, this view is not shared by the layperson—the “intuitive toxicologist”—
who tends to believe that “if large exposures to a chemical are harmful, then
small exposures are also harmful.” Specifically, laypersons believe:
• any exposure to a toxic chemical makes one likely to suffer adverse health
effects;
• any exposure to a carcinogen makes one likely to get cancer;
• the fact of exposure to a pesticide is the critical concern, rather than the
amount of exposure;

63. Laws of Fear 77–79, 81.

64. Nancy Kraus, Torbjörn Malmfors, and Paul Slovic, Intuitive Toxicology: Expert and
Lay Judgments of Chemical Risks, 12 Risk Analysis 215–232 (1992).
518 problem solving, decision making, and professional judgment

• reducing the concentration of a possibly harmful chemical in a city’s drink-

ing water would not reduce the danger associated with drinking that
water;
• there is no safe level of exposure to a cancer-causing agent.
Although not a majority view, a high percentage of laypersons sampled
believed that they should do everything possible to avoid contact with chemicals
and chemical products (40.0 percent) and believed that all use of prescription
drugs (17.2 percent) and chemicals (29.3 percent) must be risk-free. Thirty per-
cent of laypersons did not agree that a 1 in 10 million lifetime risk of cancer from
exposure to a chemical was too small to worry about.
The researchers speculate that the concept of contagion may account for these
attitudes. Anthropologists describe
the belief, widespread in many cultures, that things that have been in contact
with each other may influence each other through transfer of some of their
properties via an “essence.” Thus, “once in contact, always in contact,” even if
that contact (exposure) is brief. . . . [E]ven a minute amount of a toxic sub-
stance in one’s food will be seen as imparting toxicity to the food; any amount
of a carcinogenic substance will impart carcinogenicity, and so forth. The
“essence of harm” that is contagious is typically referred to as contamination.
Being contaminated clearly has an all-or-none quality to it—like being alive or
pregnant. . . . This all-or-none quality irrespective of the degree of exposure is
evident in the observation by Erikson that: “To be exposed to radiation or
other toxins . . . is to be contaminated in some deep and lasting way, to feel
dirtied, tainted, corrupted.” A contagion or contamination model is obviously
very different from the scientist’s model of how contact with a chemical
induces carcinogenesis or other adverse effects. Further examination of the
concepts of contagion and contamination may help us better understand the
origins of the public’s concerns about very small exposures to chemicals.65
The intuitive toxicologist’s tendency to view a situation as either safe or unsafe
is a form of probability neglect that extends beyond toxics and fear of contagion.
Cass Sunstein writes:
With respect to the decision whether to insure against low-probability hazards,
people show bimodal responses. When a risk probability is below a certain
threshold, people treat the risk as essentially zero and are willing to pay little
or nothing for insurance in the event of loss. But when the risk probability is
above a certain level, people are willing to pay a significant amount for insur-
ance, indeed an amount that greatly exceeds the expected value of the risk.

65. Paul Slovic, “If Hormesis Exists . . . Implications for Risk Perception and
Communication,” http://www.belleonline.com/newsletters/volume7/vol7-1/ifhormesi-
sexists.html. See also Paul Rozin and Carol Nemeroff, Sympathetic Magical Thinking: The
Contagion and Similarity “Heuristics,” in Heuristics and Biases, supra.
 the role of affect in risky decisions 519

Such bimodal responses provide further support for the intuitive suggestion
that some risks are simply “off-screen,” whereas others, statistically not much
larger, can come “on-screen” and produce behavioral changes.66
The tendency to ignore some low-probability risks is partly due to the diffi-
culty of assessing them. As Sunstein notes, “a decision to disregard low-level
risks is far from irrational, even if it is based in whole or in part on emotions; we
lack the information that would permit fine-grained risk judgments, and when
the probability really is low, it may be sensible to treat it as if it were zero.”67 We
may also ignore risks because of our tendency to be unrealistically optimistic—
especially about our own futures68—and also for peace of mind. As Kai Erickson
writes: “One of the bargains men make with one another in order to maintain
their sanity is to share an illusion that they are safe, even when the physical evi-
dence in the world around them does not seem to warrant that conclusion.”69
Preserving sanity through illusion can have its costs, though. Consider the com-
placency about airline security before September 11, 2001, and the muted recep-
tion given the January 31, 2001, report of the U.S. Commission on National
Security, co-chaired by former Senators Gary Hart and Warren Rudman, which
warned of America’s vulnerability to terrorist attacks. The absence of availability
may lull people into complacency: What is out of sight is effectively out of mind.70

16.5.4 Anchoring and Risk

Given its pervasiveness, it is not surprising that the anchoring phenomenon
discussed in Section 10.1 also affects decision making involving risks. In a study
of people’s WTP to reduce the annual risk of motor vehicle deaths, Michael
Jones-Lee and Graham Loomis presented participants with a risk of a certain
magnitude and asked if they were willing to pay a particular monetary amount to
avoid the risk. If they were willing to pay, the amount was increased in incre-
ments until they were unwilling. One group was given the initial amount of £25
and another £75. Not only was the WTP different for the two groups: the mini-
mum WTP for the £75 group was higher than the maximum WTP of the £25
group.71

66. Cass Sunstein, Probability Neglect: Emotions, Worst Cases, and Law, 112 Yale L.J. 61,
75 (2002).
67. Id.
68. See Section 13.5. Bimodalism in this context may be a close cousin of the certainty
effect, discussed above.
69. Kai Erikson, Everything in its Path: Destruction of Community in the
Buffalo Creek Flood 234 (1976). See also George Akerlof and William Dickens, The
Economic Consequences of Cognitive Dissonance, in George Akerlof, An Economic
Theorist’s Book of Tales 123, 124–28 (1984).
70. Paul Slovic, Baruch Fischhoff, and Sarah Lichtenstein, Rating the Risks, 21
Environment 14 (1979).
71. Michael Jones Lee and Graham Loomis, Private Values and Public Policy, supra.
520 problem solving, decision making, and professional judgment

16.6 cost-benefit analysis concluded

The preceding discussions have implications for the practice of cost-benefit anal-
ysis surveyed in Section 12.3.

16.6.1 Assessing Risks

The concept of willingness to pay (WTP) assumes not only that individuals are
the best judges of their own welfare, but that they are able to assess costs, bene-
fits, and risks. However, we have seen that people may have difficulty estimating
the risks of low-probability events (such as a 1/100,000 risk of cancer), and thus
that they might underestimate the potential gains of a regulation, yielding a low
WTP value.

16.6.2 Not All Risks Are Equal

Most importantly, people do not view all statistically identical risks in the same
way. Risks that are uncontrollable, dreaded, and potentially catastrophic produce
a much higher WTP to avoid them than other risks.72 Cass Sunstein notes that
“a 1/100,000 risk of dying in a workplace accident might well produce a different
WTP from a 1/100,000 risk of dying of cancer from air pollution, from a 1/100,000
risk of dying in an airplane as a result of a terrorist attack, or from a from a
1/100,000 risk of dying as a result of defective snowmobile. 73 He notes:
Many risks controlled by the EPA [e.g., cancer] are qualitatively different from
the workplace risks that EPA has used to generate its value of a statistical life
(VSL),74 and it is possible that the dread of the suffering associated with cancer
or AIDS might bring people to project higher WTPs for these illnesses. The
“cancer premium”—people seem to have a special fear of cancer and appear
willing to pay more to prevent a cancer death than a sudden unanticipated
death—might be produced by the “dread” nature of cancer. It is well estab-
lished that dreaded risks produced special social concern, holding the statisti-
cal risks constant.75
Similarly, Richard Revesz suggests that “the value of avoiding a death from an
involuntary, carcinogenic risk should be estimated as four times as large as the
value of avoiding an instantaneous workplace fatality.”76 These potential differ-
ences illustrate the weaknesses in basing WPT on revealed preference studies.

72. Perception of Risk at 291.

73. Laws of Fear 138.
74. See Section 12.2.3a.
75. Id.; Mark Kelman, “Saving Lives, Saving from Death, Saving from Dying:
Reflections on ‘Over-valuing’ Identifable Victims” (unpublished paper 2008).
76. See Richard L. Revesz, Environmental Regulation, Cost-Benefit Analysis, and the
Discounting of Human Lives, 99 Colum. L. Rev. 941, 962–74 (1999).
 the role of affect in risky decisions 521

Also, as mentioned earlier in this chapter, different people, communities, and

age groups value and evaluate different risks differently from one another. Some
people “show an intense aversion to risks that others treat with equanimity.”77

16.6.3 Discounting for Dread

People generally prefer to experience benefits as soon as possible and to defer
costs. But we sometimes prefer to hasten experiencing a bad event rather than
dread its prospect over a period of time. How should one account for the costs of
dread incurred throughout the years between someone’s first exposure to an
environmental or workplace risk (say, involving a carcinogen) and the ultimate
manifestation of the disease? Richard Revesz argues that “any discounting of the
adverse consequences of the risk during the latency period needs to be coupled
with an increase in the estimate of these consequences as a result of dread.”78

16.6.4 The Uncertain Risks of Catastrophic Harm

Catastrophic harms are ones that create tremendous and possibly irreversible
losses for communities or civilizations. The losses may be material—as in the
case of various worst-case outcomes of climate change or nuclear warfare, or of
nanoparticles or a particle accelerator turning the world into gray goo.79 Even
when not of global magnitude, a community’s or nation’s economic and social
fabric may be destroyed by tsunamis or earthquakes, or a terrorist attack. Because
most catastrophic harms have a very small n—they occur very infrequently, if at
all—their probability of occurrence is not susceptible to frequentist statistical
analysis, and even subjectivist estimates are likely to be highly speculative. In
other words, decision making concerning catastrophic harms takes place not
under conditions of risk but of uncertainty.
Whether cost-benefit analysis (CBA) can adequately or appropriately address
uncertain risks of catastrophic harms is a matter of lively and often acrimonious
dispute, often centering around environmental issues.80 Critics of CBA would
replace it with a precautionary principle (PP), which counsels avoiding any steps
that create a risk of harm. For example, the declaration following the 1998
Wingspread Conference on the Precautionary Principle stated: “When an activ-
ity raises threats of harm to human health or the environment, precautionary
measures should be taken even if some cause and effects relationships are not
established scientifically. In this context the proponent of the activity rather than

77. Id. 131.

78. Richard L. Revesz and Michael A. Livermore, Retaking Rationality: How
Cost-Benefit Analysis Can Better Protect the Environment and Our Health 104
(New York: Oxford University Press, 2008).
79. See, e.g., Richard A. Posner, Catastrophe: Risk and Response (2004). http://scholar-
ship.law.cornell.edu/lsrp_papers/50
80. See generally Douglas A. Kysar, It Might Have Been: Risk, Precaution, and Opportunity
Costs, J. Land Use & Envtl. L. 1 (2006).
522 problem solving, decision making, and professional judgment

the public, should bear the burden of proof.”81 There are many other versions of
the PP, but they all share the common element that it is worth high social oppor-
tunity costs to prevent the possibility of catastrophic harm.
In Worst-Case Scenarios, Cass Sunstein undertakes an extensive critique of the
PP, arguing that it is extremely vague, prone to biases of the sort described in
this book, and likely to be counterproductive to human welfare. Sunstein makes
these observations, among others:
• The PP makes a distinction between action and inaction that is at best
orthogonal to expected value. For example, the PP is frequently invoked to
halt the production of foods using genetically modified organizations, but
not to mobilize research and development to prevent a large meteor from
hitting the earth.
• The PP is invoked against particularly salient risks, based on the availabil-
ity heuristic, but not against risks that are not on people’s radar screens.
• By virtue of loss aversion, “people will be closely attuned to the potential
losses from any newly introduced risk . . ., but far less concerned about
future gains they may never see if a current risk is reduced.” Consider the
barriers to the introduction of new pharmaceuticals, which are on-screen,
compared to the illness and death that might be reduced by their introduc-
tion, which are off-screen.
• Prohibiting one harm under the PP may create equal or worse harms. For
example, in developing countries, bans on DDT have led to increases in
malaria, and bans on genetically modified organisms may increase deaths
through starvation.
Although Sunstein rejects most formulations of the PP, he proposes what he
calls an irreversible and catastrophic harm principle, under which “when a harm is
irreversible in the sense that restoration is very costly or impossible, special pre-
cautions may be justified; . . . it often makes sense to ‘buy’ an option to preserve
future flexibility; and . . . loss of cherished and qualitatively distinctive goods
deserve particular attention.”82
Without delving more deeply into this complex and hotly contested issue, two
things are reasonably clear to us. First, that the trade-offs between current ben-
efits and the risk of catastrophic harm cannot be left entirely to experts but, in a
democracy, must be the subject of debate and deliberation by legislators informed
by diverse civil society organizations. Second, public deliberation should be
informed by experts’ estimates of the scope and probability of harms, even if
they come with large margins of error.

81. Quoted in Worst-Case Scenarios 124.

82. Id. at 189.
conclusion
the lawyer as cognitive counselor
In concluding Part 3, we ask whether lawyers can use the insights of judgment
and decision making research to improve their clients’ decision making. (We
discuss the parallel question for policy makers in Part 4.) The difficulties of debi-
asing even one’s own decisions, coupled with the injunction “physician heal
thyself,” counsels modesty in this enterprise. However, their very distance from
their clients’ individual decisions may give lawyers a perspective that enables
them to see and mitigate biases in others to an extent that they cannot necessar-
ily do for themselves. The hazards of guiding clients’ decisions may lie less in
lawyers’ cognitive biases than in the imposition of their own values.

pervasive biases and their largely ineffectual remedies 1

Thus far, we have identified a variety of cognitive biases and heuristics that can
lead to incorrect empirical judgments and to decision making that does not max-
imize an individual’s utility. These include:
• the availability heuristic;
• the representativeness heuristic;
• social stereotyping;
• anchoring bias;
• neglect of base rates;
• confirmation bias;
• overconfidence;
• hindsight bias;
• overoptimism;
• self-serving bias
• misforecasting the effects of positive and negative experiences;
• framing effects—particularly the endowment effect and loss aversion;

1. See generally Scott O. Lilienfeld et al., Giving Debiasing Away: Can Psychological
Research on Correcting Cognitive Errors Promote Human Welfare?, 4 Perspectives on
Psychol. Sci. 390 (2009); Katherine L. Milkman et al., How Can Decision Making Be
Improved?, 4 Perspectives on Psychol. Sci. 379 (2009); Richard P. Larrick, Debiasing, in
Blackwell Handbook of Judgment and Decision Making 316 (Derek J. Koehler and
Nigel Harvey eds., Malden, MA: Blackwell Publishing, 2004).
524 problem solving, decision making, and professional judgment

• difficulties in dealing with too many choices and the framing and grouping
of choices;
• difficulties in evaluating and comparing risks.
Along the way, we have occasionally mentioned efforts to debias these
phenomena. Here we summarize what is known about debiasing techniques.
Informing people of the biases. Informing decision makers that they are
prone to particular biases has little or no effect in most circumstances.2
Incentives. The hope is that incentives will reduce error is based on the
assumption that they will motivate System 2 analysis and deliberation. But this
requires that people have the judgment and decision-making skills that the addi-
tional effort will mobilize.3 While incentives may stimulate the search for addi-
tional information, they have little or no effect in mitigating most cognitive
biases.4
Accountability. As with incentives, accountability to others can only reduce
biases if decision makers have the requisite skills. Even then, as we will discuss
in Section 19.11, the social pressures of accountability may induce biases of their
own.
Considering the alternative. One of the most robust and pervasive debiasing
techniques is to consider why one’s judgment may be wrong. Hindsight bias
can be counteracted by asking people to provide reasons why an event other than
the one that actually occurred might have occurred. Overconfidence and the
biased assimilation of information can be counteracted by asking people to pro-
vide counterarguments for their judgments. And the self-serving bias in litiga-
tion can be counteracted by asking parties to write down the weaknesses of their
own case.5 These techniques can sometime backfire, however: if people find it
difficult to think of reasons or examples on the other side, it may actually
strengthen their prior beliefs.6
Representing risks and other matters of likelihood in terms of frequencies rather
than probabilities. Under at least some circumstances, people will combine prob-
abilities more accurately when considering likelihoods as frequencies rather
than probabilities––for example, that an outcome occurs 25 out of 100 times
rather than 25 percent of the time. (See Section 8.6.1).

2. See Baruch Fischhoff, Debiasing, in Daniel Kahneman, Paul Slovic, and Amos
Tversky, Judgment Under Uncertainty: Heuristics and Biases 422 (1982).
3. See Colin Camerer and Robin Hogarth, The Effects of Financial Incentives in
Experiments: A Review and Capital-Labor-Production Framework, 19 J. Risk and
Uncertainty 7 (1999).
4. Camerer and Hogarth, supra; Larrick, supra.
5. Linda Babcock, George Loewenstein, and Samuel Issacharoff, Creating Convergence:
Debiasing Biased Litigants, 22 Law & Social Inquiry 914 (1997).
6. Larrick, supra.
 conclusion 525

Statistical rather than intuitive predictions. We saw in Section 10.5.1 that

people tend to rely on intuitive predictions in situations where using a simple
linear model would be far more accurate. Of course, the belief that one’s own
intuition beats a linear model may itself be difficult to debias.
Taking an outsider’s perspective. Trying to mentally remove oneself from a
particular situation, or viewing it as an example of a broader class of decisions,
may reduce people’s overconfidence about their knowledge and overoptimism
about their likelihood of success.7
***
In the remainder of this conclusion to Part 3, we ask whether a lawyer can
help clients avoid the errors of decision making we have just covered. In Part 4,
and particularly Chapter 18, we acquiesce in the inevitability of people’s making
such errors, and suggest ways that policy makers can “fight fire with fire” by
framing decisions so that uncorrected System 1 processes produce outcomes
that are in people’s or society’s best interests.

the lawyer as cognitive counselor

About half the practice of a decent lawyer consists in telling would-be clients that they are
damned fools and should stop.
— Elihu Root8

In both law and medicine, active client or patient participation in decision

making is widely accepted as central to good professional practice. This perspec-
tive is reflected in both the client-centered, or client autonomy,9 model of legal
counseling, and in the informed consent model of medical decision making.10

7. Milkman, supra.
8. Philip C. Jessup, Elihu Root 133 (New York: Dodd Mead, 1938).
9. The client autonomy model is similar in many respects to the client-centered
approach. For a leading defense of client autonomy, see Monroe H. Friedman,
Understanding Lawyers’ Ethics (1990).
10. For an account of the development of informed consent as a normative model of
doctor-patient decision making, and an extension of that model to the legal counseling
relationship, see Mark Spiegel, Lawyering and Client Decisionmaking: Informed Consent and
the Legal Profession, 128 U. Pa. L. Rev. 41 (1979). For a cognitive limitations critique of the
informed consent model of medical decision making, see Jon Merz and Baruch Fischhoff,
Informed Consent Does Not Mean Rational Consent, 11 J. Legal Medicine 321 (1990). See
also Donald A. Redelmeier, Paul Rozin, and Daniel Kahneman, Understanding Patient’s
Decisions: Cognitive and Emotional Perspectives, 270 J. of the American Medical
Association 72 (1993) (describing various cognitive biases and heuristics that influence
patient’s approach to making medical treatment decisions).
526 problem solving, decision making, and professional judgment

According to these models, the professional’s task is to present the client or

patient with sufficient information about the available options and their reason-
ably foreseeable outcomes to enable the client or patient to make a fully informed,
autonomous choice. It is not the professional’s role to make the decision for the
client or, on some views, even to tell the client which option she thinks is “best.”11
Clients are presumed to use the information provided by the professional to reach
a decision that best serves their interests, as they define and understand them.
There are situations in both legal and public policy decision making where
clients not only fully comprehend their own interests, but have already engaged
in whatever problem solving is required and have determined what actions they
want to take. The lawyer’s sole task in these situations may be to implement a
decision already made. In many instances, however, the lawyer can usefully act
as interlocutor, partner, or guide in identifying and clarifying objectives and
determining the best course of action. As former law dean Anthony Kronman
writes:
[O]ften the client’s objective is hazy, or in conflict with other objectives, or
clear but impetuously conceived. . . . [The lawyer’s] job in such cases is to help
clarify the client’s goal by pointing out ambiguities in its conception and by
identifying latent conflicts between it and other of the client’s goals. . . .
[Indeed, the lawyer’s] responsibilities to a client go beyond the preliminary
clarification of his goals and include helping him to make a deliberatively
wise choice among them. . . . [H]is duty [is] not merely to implement a client’s
decision . . . but also to help him assess its wisdom through a process of coop-
erative deliberation. . . .12
In recent years, client-centered models have been subjected to what might be
termed a cognitive bias critique. The critique proceeds from the observation,
based on research of the sort described in the preceding chapters, that people’s
decision-making skills are deficient in some significant respects. Their judg-
ments about causation and their ability to evaluate risk or predict outcomes are
susceptible to a variety of systematic biases. Their preferences are unstable and
may be inconsistent, context-dependent, or self-defeating. In short, the processes
by which people go about making decisions are flawed in a number of important
ways when compared with normative models of decision making derived from
statistics, economics, and decision science.

11. For a discussion of this issue, see David A. Binder, Paul B. Bergman, Susan
Price, and Paul R. Tremblay, Lawyers as Counselors: A Client-Centered Approach
(Eagan, MN: West. 2nd ed., 2004).
12. Anthony T. Kronman, The Lost Lawyer: Failing Ideals of the Legal
Profession 128–129, 131 (New York: Cambridge University Press, 1993).
 conclusion 527

Legal scholars have begun to apply insights from this research to a wide vari-
ety of subjects in law and legal practice.13 Much of this literature seeks to identify
sources of bias that, for example, systematically distort clients’ decisions to com-
mence, continue, or settle litigation and that thus lead to suboptimal outcomes
in terms of their own interests.14 The hope is that lawyers can help clients avoid
or compensate for these biases and thus make better decisions. In other words,
lawyers might add value to client decision making not only by predicting legal
outcomes, but also by structuring and improving decision-making processes.
So, for example, Jeffrey Rachlinski observes that by describing settlement
offers to clients in ways that counteract framing effects, attorneys can improve
outcomes for the clients and for society in general.15 In an intriguing set of
experiments (described below), Russell Korobkin and Chris Guthrie suggest
that lawyers are more likely than nonlawyers to apply expected value analysis

13. See, e.g., Donald C. Langevoort and Robert K. Rassmussen, Skewing the Results: The
Role of Lawyers in Transmitting Legal Rules, 5 S. Cal. Interdisc. L. J. 375 (1997) (using
insights from social and cognitive psychology to explain why lawyers might overstate legal
risk in counseling business clients); Gary L. Blasi, What Lawyers Know: Lawyering Expertise,
Cognitive Science, and the Functions of Theory, 45 J. Legal Educ. 313 (1995) (describing the
development of legal expertise in terms of cognitive theory and the debiasing of judg-
ment); Jody Armour, Stereotypes and Prejudice: Helping Legal Decisionmakers Break the
Prejudice Habit, 38 Cal. L. Rev. 733 (1995) (applying social cognition theory to the prob-
lem of debiasing jury decision making); Donald C. Langevoort, Where Were the Lawyers?
A Behavioral Inquiry into Lawyers’ Responsibility for Client’s Fraud, 46 Vand. L. Rev. 75
(1993) (applying insights from social psychology and social cognition theory to explain
how lawyers may fail to detect client fraud); and Albert J. Moore, Trial by Schema: Cognitive
Filters in the Courtroom, 37 U.C.L.A. L. Rev. 273 (1989) (application of insights from
cognitive psychology to trial advocacy).
14. See, e.g., Mark Kelman et. al., Context-Dependence in Legal Decision Making, 25 J.
Legal Stud. 287 (1996) (demonstrating empirically that choice in legal decision-making
situations may be influenced by context effects); Jeffrey J. Rachlinski, Gains, Losses, and
the Psychology of Litigation, 70 S. Cal. L. Rev. 113 (1996) (demonstrating how framing
effects may influence client decisions to settle or continue litigation); Linda Babock, et al.,
Forming Beliefs About Adjudicated Outcomes: Perceptions of Risk and Reservation Values, 15
Int’l Rev. L. and Econ. 289 (1995) (influence of framing and prior expectancy effects on
settlement decision making); Russell Korobkin and Chris Guthrie, Psychological Barriers
to Litigation Settlement: An Experimental Approach, 93 Mich. L. Rev. 107 (1994) (illustrat-
ing how framing effects and equity seeking may influence settlement decision making);
Russell Korobkin and Chris Guthrie, Opening Offers and Out of Court Settlement: A Little
Moderation May Not Go A Long Way, 10 Ohio St. J. on Dispute Resolution 1 (1994)
(anchoring effects and the evaluation of settlement offers); George Loewenstein, et al.,
Self-Serving Assessments of Fairness and Pretrial Bargaining, 22 J. Legal Stud. 135 (1993)
(demonstrating the effects of overoptimism and other self-serving evaluative biases on
settlement decision making).
15. Jeffrey J. Rachlinski, Gains, Losses, and the Psychology of Litigation, supra at 170–73.
528 problem solving, decision making, and professional judgment

in deciding whether to settle or continue with litigation, and are less likely in
such situations to be influenced by certain cognitive biases.
This line of research poses a challenge to the traditional client-centered model
of legal counseling: If, at least in certain contexts, lawyers are less likely than
clients to be influenced by cognitive biases or the indiscriminate application of
simplifying heuristics, does this support introducing a degree of paternalism
into normative models of lawyer-client decision making?
Korobkin and Guthrie respond to this question by positing what they refer to
as a cognitive error approach to legal counseling.16 Under this approach, the appro-
priateness of a lawyer’s intervention depends on the judgments that underlie the
client’s decision. As Korobkin and Guthrie state:
The cognitive error approach to counseling . . . requires the lawyer to assess
whether an observed difference between the lawyer’s and client’s analysis of
decision options is due to the client’s cognitive error or is merely the manifes-
tation of differences in utility functions. If the difference is due to cognitive
error, the lawyer should attempt to change the client’s outlook. If the differ-
ence is the result of different preference structures, the lawyer should scrupu-
lously avoid any interference.17
With this distinction in mind, and with due modesty about lawyers’ ability
to influence clients’ judgments that are based on cognitive biases or affect,18
let’s examine more closely Korobkin and Guthrie’s experiments involving
the lawyer’s role as counselor.19 They gave laypersons and lawyers these bias-
inducing scenarios and asked whether they would settle or go to trial.
• Anchoring. A car buyer’s suit against the dealer for the purchase of a
“lemon,” in which the defendant’s initial offer of settlement acted either as
a high or low anchor for the plaintiff’s expectations. (See Section 10.1)

16. Russell Korobkin and Chris Guthrie, Psychology, Economics, and Settlement:
A New Look at the Role of the Lawyer, 76 Texas L. Rev. 77, 129–130 (1977).
17. Id. at 130.
18. The modesty implied by the first section of this chapter is supported by Korobkin
and Guthrie’s efforts to increase settlement rates in the scenarios described in the text that
follows. They employed five strategies: (1) informing the client about the psychological
factors that might inform her decision, (2) asking the client to consider an opposite or
alternative point of view, (3) recommending settlement, (4) giving reasons for the recom-
mendation, and (5) recommending settlement without giving any reasons. All five tended
to increase the settlement rate, but not dramatically.
19. The experiments were motivated by the higher rate of settlement of civil cases
than predicted by prospect theory. As we have seen, plaintiffs and defendants view litiga-
tion from different frames, with defendants tending to be more risk-taking than plaintiffs.
Moreover, litigants may have goals besides money, such as vindication, revenge, and jus-
tice. The high rate of settlement could nonetheless be explained if lawyers did not share
the cognitive biases of their clients and if they influenced their clients to settle.
 conclusion 529

• Gain/loss framing. A car accident case in which the identical situation was
manipulated so that settlement appeared to the plaintiff either as a gain or
a loss. (See Section 14.1)
• Sympathy. A suit against a landlord who failed to provide the tenant with
heat during an entire winter, in which the landlord either had been will-
fully indifferent or had a sympathetic excuse (being out of the country on a
family emergency). The psychological issue here was not bias, but the par-
ticipants’ sense of justice.20
Lawyers tended to settle regardless of the frame. Clients’ settlement rates
depended on how the issue was framed.
In the lemon case, when the anchor was high (thus creating a high expecta-
tion), clients’ settlement rates were significantly lower than lawyers’. Since it is
not plausible that the client would accord any actual value to this anchor, the
lawyers’ tendency to ignore the anchor seems unequivocally utility maximizing.
In the accident case, the clients’ settlement rates were lower in the loss frame
than in the gains frame. This is predicted by prospect theory, as described in
Section 14.1. Whether the clients’ frame-dependent tendencies produce subopti-
mal decisions that affect their experienced utility in the long run depends on the
interactions of affective forecasting, adaptation, and regret.
And in the heater case, the clients’ settlement rates were lower when the land-
lord had an unsympathetic excuse for not fixing the heater, while the settlement
rates for lawyers were the same in both situations. Based on the experimental
results and follow-up interviews, Korobkin and Guthrie conclude that the law-
yers were single-mindedly focused on monetary expected value and indifferent
to the other variables, while some clients wished to maximize a value other than
wealth. (See Section 12.1) One certainly cannot characterize a tenant’s decision
to settle for less out of sympathy as irrational. Yet the lawyer may believe that the
client is overvaluing the sympathy discount in terms of his own utility func-
tion—say, because the excuse was conveyed in a personal, emotion-laden
manner, whose psychological effects will diminish in another week or so.
Korobkin and Guthrie’s distinction between a client’s errors and utilities
echoes David Luban’s suggestion in Paternalism and the Legal Profession21 that a
lawyer may permissibly compromise client autonomy where the intrusion inter-
feres only with a client’s “wants,” but not when it would interfere with the expres-
sion of a client’s “values.”22 Luban defines values as those reasons for choosing a
particular course of action “with which the agent most closely identifies—those

20. The subjects were informed that it would not affect the outcome of the small
claims court trial.
21. David Luban, Paternalism and the Legal Profession, 1981 Wisconsin L. Rev. 454
(1981).
22. Id. at 474.
530 problem solving, decision making, and professional judgment

that form the core of his personality, that make him who he is.”23 A cognitive
error approach to legal counseling is based on the assumption that biases or
errors can be distinguished from a client’s core preferences or utilities.
To put client-centered counseling in the terms used in this book, we start from
the premise that the lawyer’s role as counselor is to help a client maximize his or
her utility—taking into account the broad conception of utility considered in
Section 12.1. While the ideal, of course, is to maximize experienced utility, lawyers,
clients, and indeed all of us, can only forecast the expected utility of decisions.
In the simplest case, the lawyer provides the client with information about
substantive law and procedure, and makes predictions of the likely consequences
of particular courses of actions, based on which the client decides what course of
action to pursue. This inevitably requires that the lawyer understand the client’s
objectives, and sometimes requires working with him to clarify them. The lawyer
may also bring to the counseling process decision-making knowledge or tools
that not all clients necessarily possess—for example, how to structure a decision
process with multiple competing objectives or how to take risk into account
when calculating expected value.24
Even when only money is at stake, these can be complex matters; consider,
for example, quantifying risks, identifying uncertainties and understanding a
client’s risk tolerance in a particular context. But, as we have seen, subjective
utility can encompass a large variety of factors besides money—factors that the
client may not be able to specify, let alone predict how they will be affected by a
decision. Even relatively simple predictions of expected value are subject to cog-
nitive biases, and as one moves to a broader range of goals the possibilities of
bias grow immensely.
What role can the lawyer as counselor play in helping a client navigate these
manifold and often hidden shoals? Although not only laypersons but experts as
well are subject to biased judgment and decision making, there are several rea-
sons that a lawyer may be able to help a client address errors that the client
wouldn’t notice himself:
• the lawyer possesses expertise in decision making25—especially if he or she
has taken a course in problem solving and decision making;
• by virtue of her disinterest, the lawyer can provide the client with a differ-
ent and neutral perspective in the matter;
• an agent rather than a principal, the lawyer does not have the “endowment”
that a client may experience with respect to the status quo;

23. Id.
24. For example, the subjective linear model described in Section 4.5 and the deci-
sion tree described in Section 15.2.2.
25. See Russell Korobkin and Chris Guthrie, Psychology, Economics, and Settlement:
A New Look at the Role of the Lawyer, 76 Texas L. Rev. 77 (1997).
 conclusion 531

• as a repeat player, the lawyer has the opportunity to get better feedback on
recurrent events, and has knowledge about likely outcomes and about cli-
ents’ satisfaction or dissatisfaction in vindicating certain kinds of interests.
For all of its promise, cognitive counseling cannot escape, and indeed may con-
tribute to, some fundamental dilemmas of the decision-making process described
in the preceding chapters. Most notably, if preferences are sometimes constructed
in the very process of making a decision, there is a danger that lawyer’s prefer-
ences may unwittingly influence the client’s decision. The very act of communicat-
ing information to clients subtly shapes preferences and influences choice.26 The
client may be led, either by the lawyer directly or by the “scripted” nature of the
lawyer-client interaction, to value certain aspects of utility at the moment differ-
ently than they will be valued in life outside the lawyer’s office. Indeed, the lawyer-
client interaction may temporarily alter the client’s preferences.27
Problem: The Case of Elizabeth Fletcher
We end the discussion of the lawyer as counselor with a hypothetical case that
illustrates the dilemmas of the counseling relationship in circumstances where
deeply held personal values are at stake in situations of risk or uncertainty.
Elizabeth Fletcher is a single woman in her early 40s, who consulted a lawyer
after she had been fired from her job as a customer service representative at a
regional electrical utility company. The reason given for her termination was
excessive absenteeism. Fletcher explained to her lawyer that a year and a half ago
she was diagnosed with an autoimmune disease that causes painful and poten-
tially life-threatening inflammation of the joints, muscles, and other organs,
including the lining around the lung and the heart. Periods during which the
disease is in remission alternate with periods of severe symptoms, requiring
Fletcher to miss more work than would an otherwise similarly situated but non-
disabled employee. She has also had to miss work because of frequent medical
appointments.
Fletcher asked to be allowed to work a flexible schedule as an accommodation
under the Americans with Disabilities Act (ADA). While the human resources
director offered to help her apply for and obtain social security disability benefits
(SSDI), he declined to make this accommodation. After more missed days, the
company terminated her employment.
Fletcher has now consulted a lawyer for help in selecting between two possi-
ble courses of action. She can either file for SSDI, with the company’s coopera-
tion, or she can sue the company under the ADA for failing to accommodate her
need for a flexible work schedule. She thought about doing both simultaneously,

26. Steven Ellman, Lawyers and Clients, 34 U.C.L.A. L. Rev. 717, 733–53 (1987).
27. William H. Simon, Lawyer Advice and Client Autonomy: Mrs. Jones’ Case, 50
Maryland L. Rev. 213, 216–17 (1991).
532 problem solving, decision making, and professional judgment

but her lawyer indicates that the law requires her to choose one option or the
other.
The lawyer, an expert in disability law, knows that a study by the Commission
on Mental and Physical Disability of the American Bar Association28 indicates
that defendants prevail in over 90 percent of litigated ADA cases. Recent amend-
ments to the ADA, providing a broader definition of “disability,” may change the
estimate somewhat, but the particular facts of Fletcher’s case do not seem to
distinguish it from others filed under the ADA, so the lawyer sees little reason to
adjust significantly the 10 percent base rate of anticipated success. Furthermore,
the costs associated with prosecuting the ADA case would be substantial, includ-
ing filing fees, deposition and expert witness costs, and a contingency fee of 30
percent of any ultimate recovery.
The costs associated with applying for social security disability benefits are
much lower and the likelihood of success greater. There is no application fee,
and Fletcher could apply without an attorney. Should she have to file an appeal,
that could be done for a fraction of the cost associated with civil litigation under
the ADA in federal district court. With the employer’s cooperation, the probabil-
ity of having her SSDI application approved are quite high, perhaps 85 percent.
The lawyer tries to provide the data to explain why filing for SSDI has much
greater expected value than suing the employer under the ADA. But Fletcher
becomes increasingly agitated, and eventually says that, while she accepts that
the success rate in ADA cases is low, she is confident that she has a better than
average chance of success. Her former employer is a big company. They could
easily have put her on a more flexible schedule without having to endure an
“undue hardship,” which would provide a defense under the ADA. Besides, if
she has to rely on Social Security, she feels certain that she will never get better.
The very thought of going on “the dole,” she explains, makes her feel depressed;
it’s like acknowledging that her life is over, that she is simply waiting to get
worse and die. “If I don’t keep fighting, I’ll lose the battle with this disease.
People with HIV go on SSDI. That’s when you know they’re on their way out.”
One could easily conclude that Elizabeth Fletcher’s judgment is distorted—in
this case by unrealistic optimism about her chances of winning an ADA suit, and
by the seemingly irrational belief that going on Social Security disability insur-
ance has the power to alter the course of her disease. Yet recall Shelley Taylor’s
research on “positive illusions,” in which cancer patients, crime victims, and
others who had experienced adverse life events adapted by finding meaning in
the adverse experience, by attempting to regain mastery over the negative event

28. See Study Finds Employers Win Most ADA Title I Judicial and Administrative
Complaints, 22 Mental and Physical Disability Law Reporter 403 (May/June 1998)
(nationwide rate of defense judgments in ADA Title I cases that went to judgment
between 1992 and 1997 was 92.11 percent; rate in Fifth Circuit (highest nationwide) was
98.1 percent, in the Ninth Circuit (lowest nationwide) 83.3 percent). Id. at 405.
 conclusion 533

in particular and over their lives more generally, and by restoring self-esteem
through self-enhancing evaluations. (See Section 13.5)
One can understand Fletcher’s attraction to litigating the ADA case as a strat-
egy for gaining control and enhancing her self-esteem, even it if involves overop-
timism about her own chances of success. Even if her receipt of SSDI benefits
does not “objectively” affect her health, Fletcher’s belief in the causal connection
may affect her experienced utility of whatever option she eventually chooses to
pursue—and perhaps it will actually come to affect her health as well. On the
other hand, if she files the ADA suit, which she has a low chance of winning, she
will forego disability benefits at least for the duration of the suit, and reduce her
chances for obtaining SSDI for some time thereafter.
How would you counsel Elizabeth Fletcher under the circumstances?
Our own view is that in situations such as these, if the lawyer’s personal or
professional experience leads him to believe that the client will regret a decision
in the long run, it would be irresponsible not to raise his concerns in dialogue
with the client—but with sensitivity to the thin lines between dialogue, persua-
sion, and overbearing influence, and a realization that even raising the question
could affect Fletcher’s well-being.
This page intentionally left blank
part four
guiding and influencing
decisions

The chapters that follow are concerned mainly with changing people’s minds and
behavior—and sometimes, conversely, with preventing others from “messing
with your mind.” As a general introduction to the topic, consider this problem.
Suppose that you are concerned about the rise in obesity among American
youth, especially among disadvantaged children. You are concerned about both
the health consequences and the economic burdens that will ultimately be
imposed on society. Consider these various approaches for promoting healthy
behaviors by your target population.

external environmental factors

Environmental factors play a significant role in the obesity of poor children.

Many children do not eat healthy foods at home because their neighborhoods
lack supermarkets or because their working (and often single) parents don’t have
time to cook. Many children get inadequate exercise because they don’t have safe
and convenient places to play. Thus, in addition to considering techniques of
persuasion and influence, you should consider how to change the external envi-
ronment, for example, by providing healthier food and more opportunities for
exercise in schools, or by supporting organizations such as the Boys & Girls
Clubs that provide after-school activities, including sports.

information and persuasion

You might try to change the behavior of children by persuading them to eat more
healthily. You could do this by informing them about the consequences of obe-
sity and how to follow a healthy diet. These efforts might be aided by labeling
foods according to their calories and fat content, though labeling may require
legislation. You might try to inform parents in order to affect their behavior with
respect to their kids’ eating and exercise habits; for example, reducing the time
spent in front of the television improves behavior in both dimensions.
536 problem solving, decision making, and professional judgment

incentives

Economic or other material incentives and disincentives—ranging from cash

payments to taxes to threats of punishment—are among the fundamental tools
for changing individuals’ and organizations’ behavior. If we don’t want people to
speed on the freeway, we levy a fine for speeding. If we want people to borrow
more money, we lower interest rates. All other things being equal, people
respond to incentives and disincentives as economists would predict, and there
is a vast literature on the use of specific kinds incentives and disincentives, rang-
ing from tax policy to criminal punishment.
Several public school systems are experimenting with programs that pay
underachieving students for doing well on standardized tests.1 Might a govern-
ment agency or nonprofit organization pay kids to maintain a certain weight
level? Might this have unintended negative consequences?

psychology

You might also employ psychological techniques to influence behavior in ways

that do not appeal solely to rational considerations—for example, by inducing
fear of the consequences of obesity or by appealing, in a positive way, to feelings
about being healthy or slim. These messages might be conveyed through various
media, by school teachers, in homes and communities, or through counseling
programs. Using many of these same techniques, you might try to change the
culture of the target population by making healthy lifestyles cool and obesity
decidedly uncool. Although this seems more ambitious, it may be the most effec-
tive way to change some individual behaviors. (Again, consider whether these
strategies might have unintended consequences.)
Two different realms of psychology provide other approaches to changing
behavior:
• social psychology, which we’ll characterize as social influence;
• an amalgam of judgment and decision making (JDM) concepts, which
we’ll characterize as framing.
The social psychology approach to influencing behavior proceeds from the
observation that humans are social animals, who are extremely sensitive to sig-
nals from and interactions with other people. This has significant adaptive value.
More often than not, our personal outcomes depend on the regard in which others
hold us, and on their willingness to cooperate with us, promote our interests, and

1. http:// www.hotchalk.com/mydesk/index.php/the-buzz/418-pay-for-grades-a-
controversial-motivation http://www.nytimes.com/2007/06/19/nyregion/19schools.
html?_r=1.
 guiding and influencing decisions 537

assist us in achieving our goals. Our ability to get ahead in life, indeed to survive,
depends on social interactions including our impulse to conform to social norms
and to obey authority.
The techniques of social influence are designed to change beliefs, attitudes,
and behavior through interpersonal interaction2 through such mechanisms as
conformity, obedience, and reciprocity. By contrast, the techniques of framing
and persuasion seek to influence beliefs, attitudes, and behavior through the way
an issue or choice is presented. These techniques build on the insights of JDM
research examined above, including how the number of choices and the way
they are presented—for example, as gains or losses—affect decisions. We include
in this category what Richard Thaler and Cass Sunstein call “choice architecture”—
policy makers’ use of framing devices to assist individuals in pursuing their own
or society’s best interests.3
Part 4 considers how people’s decisions and behavior can be influenced or
guided—for better or worse—by individuals or groups.
Chapter 17, Social Influence, draws on the realm of social psychology and con-
siders how individuals’ behavior is affected by their awareness of and relation-
ships with other people.
Chapter 18, Influencing Behavior through Cognition, draws on cognitive psy-
chology, the JDM literature, and behavioral economics. It considers how argu-
ments can be framed to make them more persuasive, how decisions can be
unconsciously affected by representations or associations placed in memory
shortly before they are made, and how policy makers can use knowledge of the
biases described in earlier parts of the book to guide citizens’ and consumers’
decision making.
Chapter 19, Improving Group Decision Making, considers how one can gain
the benefits and avoid the pathologies of group decision making, and concludes
with a discussion of the effects of accountability on individual and group
decision-making performance.

2. Susan T. Fiske, Social Beings: A Core Motives Approach to Social Psychology

508 (New York: Wiley, 2004).
3. Richard Thaler and Cass Sunstein, Nudge: Improving Decisions About
Health, Wealth, and Happiness (New Haven: Yale University Press, 2008).
This page intentionally left blank
17. social influence

In the spring of 2004, photographs showing U.S. military police torturing and
humiliating prisoners at the Abu Ghraib prison in Iraq shocked the American
public. Competing explanatory theories quickly filled the news media, personal
conversations, Web sites, and blogs. Some attributed the events to a “few bad
apples” run amok. Others posited a vast military and CIA plan to use torture
systematically as an intelligence tactic. Through this frame, the guards at Abu
Ghraib were just scapegoats, taking a fall for following orders. Other people,
familiar with Philip Zimbardo’s famous “prison study,” conducted in the base-
ment of the psychology department at Stanford University in the early 1970s,1 or
with Stanley Milgram’s electrical shocks studies conducted a decade earlier,
viewed the atrocities at Abu Ghraib differently. To these observers, what hap-
pened at Abu Ghraib was a thoroughly predictable product of a particular align-
ment of social forces that caused ordinary people to do extraordinarily bad
things.2
The desire to understand the levers of social influence produced much of the
most important and interesting social psychological research in the latter half of
the twentieth century. Taken as a whole, this research shows that, just as deci-
sion making can be impaired through the application of cognitive heuristics that
work well most but not all of the time, so can it be impaired by social influence
processes that short circuit clear thinking about a particular action or choice. Of
course, social influences are not all bad. They conduce to harmony and social
order. They sometimes help us take advantage of others’ experience and exper-
tise, thereby avoiding costly mistakes. But processes of social influence have
downsides: they can induce us to make decisions we later regret.

1. In Zimbardo’s study, Stanford undergraduates were assigned the role of either pris-
oner or guard, in an experiment about role conformity that was expected to last two weeks.
Within days, the experiment spun out of control, as “guards” began sexually humiliating
and otherwise abusing “prisoners.” For Zimbardo’s reflections on the similarities between
the behavior of his subjects and the Abu Ghraib guards, see Philip G. Zimbardo, Power
Turns Good Soldiers into “Bad Apples,” Boston Globe, May 9, 2004, Op-Ed.
2. For a thorough treatment of the Stanford prison study and its findings’ power in
explaining Abu Ghraib, see Phillip G. Zimbardo, The Lucifer Effect: Understanding
How Good People Turn Evil (New York: Random House, 2008).
540 problem solving, decision making, and professional judgment

17.1 conformity to group norms (social proof)

Recall Luis Trujillo, the lawyer introduced in Chapter 1. Project his legal career
back in time fifteen years, and imagine that he is a brand new lawyer at Cooper
& Lytton, a medium-sized Los Angeles law firm specializing in the defense of
white collar criminal cases and the civil lawsuits that often accompany them. It
is Trujillo’s first month in his new job, and he feels utterly at sea. He has to
figure out how to dress, how early to arrive at the office and how late to stay, how
to behave in meetings with partners, clients, and associates. In writing the briefs
he has been assigned to draft, he hasn’t quite figured out where to draw the line
between vigorous advocacy on the one hand and candor to the tribunal on the
other. Asked to respond to discovery requests, he still cannot quite discriminate
between a valid objections on the one hand and unethical dilatory tactics on the
other. Sitting in on a deposition preparation session with one of the firm’s most
important clients, Trujillo is deeply concerned; this looks an awful lot like the
“witness coaching” he found so distasteful in a Paul Newman movie, The Verdict,
that he had watched in his law school professional responsibility class. Every day
at work, Trujillo’s ethical intuitions are challenged, but he is also aware that
there is an awful lot about being a lawyer—a good, ethical, but also tough and
effective lawyer—that he just doesn’t yet understand.
How will the young Luis Trujillo learn to be an ethical but aggressive litiga-
tor? In all likelihood, he will observe what the firm’s more senior lawyers do, and
then he will do the same. Then, as time goes on, he will become a “Cooper &
Lytton litigator.”
Trujillo would not be alone in looking to the actions of others in his social
environment to help him make sense of an uncertain new identity. Much of the
time, particularly in novel situations, we do not know how to behave. We only
hope that we can figure it out before we make a bad impression or worse. In such
situations, we often use the behavior of others like us as a heuristic, a fast and
dirty shortcut for figuring out what to do and how to do it.
Social psychologists, such as Robert Cialdini, refer to this phenomenon as
“social proof,”3 others, like Philip Zimbardo and Susan Fiske, as “conformity.”4
Cialdini defines social proof as the tendency to use the actions of similar others
to decide on proper behavior for ourselves. In other words, people tend to see
behavior as appropriate in a given situation to the degree that they see others
performing it. The behavior of others thus serves as “proof” that it is appropriate.

3. Robert B. Cialdini, Influence: The Psychology of Persuasion (rev. ed. New

York: Quill, 1993).
4. Philip G. Zimbardo, supra note 2; Philip G. Zimbardo and Mark R. Lieppe, The
Psychology of Attitude Change and Social Influence (New York: McGraw-Hill,
1991); Susan T. Fiske, Social Beings: A Core Motives Approach to Social Psychology
(New York: Wiley, 2004).
 social influence 541

Zimbardo’s definition of conformity runs along the same lines: “a change in a

person’s belief or behavior in response to real or imagined group pressure,
where there is no direct demand for obedience by an authority figure.” (By defin-
ing the concept in this way, Zimbardo distinguishes conformity, discussed here,
from compliance with authority, discussed later.) While (at least in our society)
“conformity” tends to have a negative connotation, “social proof” does not have
a positive or negative valance—modeling our behavior on others’ can be good or
bad, depending on the circumstances.
The best-known empirical illustration of the social conformity effect comes
from a now classic study by Solomon Asch. Asch recruited college students to
participate in a study that ostensibly concerned visual perception. Subjects were
placed in a room in groups of seven and shown a series of two posters. Every
time a pair of posters was displayed, the poster on the left had a single line on it
(the “standard” line), while the one on the right bore three lines of varying
lengths. The single line on the left hand poster was always quite obviously the
same size as one of the three lines on the other, as Figure 17.1 shows.
Each time a pair of posters was displayed, subjects were asked to determine
which of the lines on the right-hand poster was the same length as the standard
line. In every group of seven, six subjects were actually confederates working
with the experimenter. The subjects were seated so that five confederates
answered before the “real” subject gave his response. (See Figure 17.2.)
At first, the confederates gave the correct answer. But, as the experiment pro-
gressed, all six confederates began giving the wrong answers, always with the
same ease and assurance with which they had given their earlier, correct answers.
In the face of five incorrect answers, 80 percent of Asch’s subjects went along
with the confederates in at least one trial. In the aggregate, subjects went along

A B

figure 17.1 asch ’ s comparative lines.

542 problem solving, decision making, and professional judgment

figure 17.2 asch subject (center) and two confederates.

with the group and gave incorrect responses approximately 30 percent of the
time, despite the obviousness of the correct response.5
Asch’s initial studies generated numerous replications and extensions.6 In
133 studies, spanning 40 years, the average subject conformed 29 percent of the
time. Conformity rates were higher when the stimulus was ambiguous and
when confederates and participants were members of the same social reference
group. Conformity rates were relatively higher for female than for male subjects,
particularly where the object of conformity was gender counterstereotypic, and
were also higher in more collectivist cultures.7
Research on conformity raises interesting and as yet not fully answered ques-
tions. Foremost among these is, when subjects comply in these experiments, are
they actually misperceiving the stimulus, or are they simply going along with the
group, even though they know that the response they are giving is wrong? Or, is
something more complex and nuanced than either one of these two possibilities
going on? The best answer appears to be “all of the above.”
The earliest evidence that conformity effects can change perception itself
emerged from extensions of Muzafer Sherif’s conformity experiments in the
1930s, on which Asch’s research was based.8 In his studies, Sherif used a phe-
nomenon known as the “autokinetic effect,” the illusion of motion created when
one gazes at a stationary point of light. When looking at a point of light and

5. Solomon. E. Asch, Effects of Group Pressures upon the Modification and Distortion of
Judgment, in Groups, Leadership, and Men (Harold S. Guetzkow ed., Pittsburgh:
Carnegie Press, 1951).
6. For a thorough review, see Rod Bond and Peter B. Smith, Culture and Conformity:
A Meta-Analysis of Studies Using Asch’s Line Judgment Task, 119 Psychological Bulletin
111–37 (1996).
7. Id. For a focus on intercultural comparisons, see Heejung Kim and Hazel R.
Markus, Deviance or Uniqueness, Harmon or Conformity? A Cultural Analysis, 77 Journal
of Personality and Social Psychology 785–800 (1999).
8. Muzafer Sherif, The Psychology of Social Norms (New York: Harper & Row,
1936).
 social influence 543

reporting on its movements by themselves, subjects’ reports of how far it moved

varied significantly. Some subjects saw small movements, others quite large
ones. However, once subjects were placed in groups and listened to each others’
reports, their judgments tended to converge. Moreover, subsequent extensions
of Sherif’s experiments showed that these changes in subjects’ perceptions
endured. When retested alone, even a year later, subjects whose judgments had
been changed by group interactions continued to “see” light movements close to
the magnitude of the original group norm.9
More recently, brain imaging research suggests that social pressure can actu-
ally change people’s visual-spatial perception. In a 2005 study using function
magnetic resonance imaging (fMRI), Gregory Berns and his colleagues found
that, in a spatial judgment task modeled on Asch’s paradigm, social conformity
shows up in the right intraparietal sulcus, a region of the brain devoted to special
perception. Nonconformity, on the other hand, lit up the right amygdala and
right caudate nucleus, regions associated with emotional threat.10
While some of Asch’s subjects reported that they simply went along with the
group despite their knowledge that they were giving an incorrect response, many
reported that they thought that they must somehow be misunderstanding the
nature of the task and conformed for that reason.11 This suggests that, at least at
times, group pressure causes people to reconstrue the meaning of perceptual
phenomena, not necessarily the raw perception itself.12
At this point, the reader might object that these were just college students
being asked about something inconsequential, and wonder whether the phenom-
enon extends to adults making more consequential decisions. To address these
concerns, Stanley Milgram replicated Asch’s experiment in the early 1960s.
Milgram used adult subjects, who believed that they had been hired to test a new
signaling system for jet airliners.13 His subjects were asked to judge the pitch of
a tone in comparison with three others. Like Asch’s student subjects, Milgram’s
adults showed a high rate of conformity to the stated views of an erroneous major-
ity, even though the task on which they were working involved airline safety.
There are, of course, limits on the conformity effect. With respect to the Asch
experiments, for example, it is important to remember that almost a third of his

9. John H. Rohrer, Jonathan H. Baron, L. Richard Hoffman, and D. V. Swander, The

Stability of Autokinetic Judgments, 49 Journal of Abnormal and Social Psychology
595–97 (1954).
10. Gregory S. Berns, Jonathan Chappelow, Caroline F. Zink, Giuseppi Pagnoni,
Megan E. Martin-Skurski, and Jim Richards, Neurobiological Correlates of Social Conformity
and Independence During Mental Rotation, 58 Biological Psychiatry 245–53 (2005).
11. See generally John M. Levine, Solomon Asch’s Legacy for Group Research, 3
Personality and Social Psychology Review 358–64 (1999).
12. Susan T. Fiske, supra note 4, 512.
13. Stanley Milgram, Nationality and Conformity, Scientific American 45–51
(November 1961).
544 problem solving, decision making, and professional judgment

subjects resisted the impulse to conform. Another third resisted more often than
they succumbed. Moreover, in follow-up studies, Asch showed that when subjects
had at least one ally in dissent rates of conformity plummeted dramatically.14
Nonetheless, the conformity effect illustrated by Asch’s research has with-
stood the test of time, both in the lab and in the real world. For example, knowl-
edge about the behavior of peers influences peoples decisions about whether to
help in an emergency,15 pay their taxes,16 or properly dispose of their trash.17

17.1.1 Death by Conformity

One of the most shocking real-world illustrations of the power of social proof
unfolded in Queens, New York, in the spring of 1964. One warm March night, a
young woman named Kitty Genovese was walking home when she was attacked
outside her apartment building by a man wielding a knife. Her violent death did
not come quickly or quietly. The attacker set upon Ms. Genovese three separate
times over a period of thirty-five minutes, as she screamed for help and twice
temporarily escaped his grasp. Thirty-two of her neighbors watched from their
apartment windows, without intervening in any way—even to call the police—as
Ms. Genovese screamed, struggled, and finally died.
Grappling to understand how such a thing could happen, social psychologists
John Darley and Bibb Latané designed a series of experiments in which they
“staged” emergencies and observed how the presence of inactive bystanders
affected subjects’ helping behavior. Over and over again, their results demonstrated
the power of social proof. When subjects were alone, they responded to the appar-
ent emergency with much higher rates of intervention than when inactive bystand-
ers were present. Specifically, when people thought they were alone, they helped
85 percent of the time. But when they believed that others were present, and were
not intervening, they helped only 31 percent of the time.18 By 1980, four dozen
laboratory and field studies had been conducted testing this “bystander effect.”

14. Solomon E. Asch, Opinions and Social Pressure, Scientific American 31–35
(November 1955).
15. Bibb Latané and John M. Darley, The Unresponsive Bystander: Why Doesn’t
He Help? (Englewood Cliffs, NJ: Prentice-Hall, 1970).
16. John T. Scholz, Kathleen M. McGraw, and Marco R. Steenbergen, Will Taxpayers
Ever Like Taxes? Responses to the U.S. Tax Reform Act of 1986, 13 Journal of Economic
Psychology 625–56 (1992).
17. Robert B. Cialdini, Raymond R. Reno, and Carl A. Kallgren, A Focus Theory of
Normative Conduct: Recycling the Concept of Norms to Reduce Littering in Public Places, 58
Journal of Personality and Social Psychology 1015–26 (1990).
18. Bibb Latané and John M. Darley, The Unresponsive Bystander: Why Doesn’t
He Help?, supra note 15; John M. Darley and Bibb Latané, Bystander Intervention in
Emergencies: Diffusion of Responsibility, 8 Journal of Personality and Social Psychology
377–83 (1968).
 social influence 545

In approximately 90 percent of these, lone bystanders proved more likely to inter-

vene in an apparent emergency than did people in groups.19
Social psychologists generally explain these results in the following way:
When people witness an apparent emergency, the need for and the appropriate-
ness of intervention are often ambiguous. When people are alone, they will often
resolve this ambiguity in favor of intervention. However, when they see other
people like themselves idly standing by, they conclude that intervention is inap-
propriate and stand back.

17.1.2 Social Proof and Professional and Corporate Malfeasance

Many “ethical train wrecks” in law and business can be traced, at least in part, to
social proof effects. Consider, for example, the ruin of in-house Enron lawyer
Kristina Mordaunt, described by Kurt Eichenwald in his book, Conspiracy of
Fools.20 In 2001, Ms. Mordaunt played a minor role in working out “LJM1,” one
of the shady partnerships (known in accountant-speak as a “special purpose
entity”) that had been used to manipulate Enron’s apparent financial perfor-
mance and enrich certain Enron executives. LJM1’s status as a valid special pur-
pose entity, which accorded it favorable accounting treatment, was highly
questionable. But from Mordaunt’s point of view, the relevant accounting rules
were ambiguous, the deal had been approved by Enron financial officers and
accountants, and the executives who had put it together were rising stars in one
of the country’s best-regarded corporations.
A short time after the her work on LMJ1 was completed, Mordaunt was off-
handedly asked by Enron Treasurer, Ben Glisan, Jr., whether she wanted to
invest $10,000 in a similar special purpose entity, known by Enron insiders as
“Southampton.” She ended up investing only $5800, but a few weeks later, when
the Southampton deal wound down, she received a return of over a million dol-
lars. Weeks later, her career was ruined, and by the middle of 2002, prosecutors
had filed a forfeiture action to seize her home and car.
It is easy—perhaps too easy—to conclude that Mordaunt was a knowing par-
ticipant in a patently fraudulent scheme who got precisely what she deserved.
But Eichenwald suggests an alternative explanation. Caught up in a web of mind-
numbingly complex and ambiguous accounting rules, and watching apparently
well-respected, successful high-level executives participating in complex transac-
tions like LJM1 and Southampton, Mordaunt could easily have concluded that
the deals were legitimate and that she was doing nothing wrong. The “proof”
was all around her, in the actions of her superiors and colleagues.

19. Bibb Latané and Steve Nida, Ten Years of Research on Group Size and Helping, 89
Psychological Bulletin 308–24 (1981).
20. Kurt Eichenwald, Conspiracy of Fools: A True Story (New York: Broadway
Books, 2005).
546 problem solving, decision making, and professional judgment

The social proof phenomenon can explain lawyers’ passive acquiescence in

client misconduct in various legal contexts. For example, Donald C. Langevoort
suggests that it can explain attorney inaction in the face of client fraud:
[A]n attorney who is new to a situation, especially an inexperienced one, is
particularly prone to rely on social learning as the basis for constructing a
schema. The fact that those more senior and more familiar with the situation
are behaving as if there is no problem provides a strong cue. This, indeed, is
one explanation for the reduced tendency of people in groups to act hero-
ically, since responsibility is effectively diffused. To this, naturally, must be
added the situational pressures on junior members of groups to conform to
apparent norms.21

17.1.3 More Examples of Social Proof

An impressive volume of empirical research convincingly demonstrates that the
information conveyed to people about situational norms has a significant impact
on attitudes and behavior.22
In one study demonstrating the power of perceived norms, Robert Cialdini
and his colleagues placed door-hangers on the front doors of houses in a number
of California neighborhoods, encouraging residents to engage in various energy-
conserving behaviors. Each of the door-hangers contained one of five message
types appealing for energy conservation. The first message type was a simple
description of various energy conservation techniques. Other types included
messages highlighting the importance of protecting the environment, emphasiz-
ing the social benefits flowing from energy conservation, and highlighting the
money residents could save through conservation. The final message type
informed residents that the majority of their neighbors were already conserving.
Some time later, interviewers went door-to-door and asked residents how
much they had engaged in the energy-conserving behaviors that was promoted
on the door-hangers; they also requested copies of residents’ energy bills, and
took readings on outdoor energy meters. The researchers found that households
that received the message highlighting that their neighborhood norm was to
conserve energy saved significantly more energy than households in any of the
other conditions.
Note that the reference group in the descriptive norm message used in this
study was other people in the same neighborhood. This was a deliberate choice
based on the results of a previous study, which showed that people are most

21. Donald C. Langevoort, Where Were The Lawyers? A Behavioral Inquiry into Lawyers’
Responsibility for Clients’ Fraud, 46 Vanderbilt. L. Rev. 75, 105–106 (1993).
22. Jessica M. Nolan, Wesley Schultz, Robert B. Cialdini, Noah J. Goldstein, and
Vladas Griskevicius, “Normative Social Influence Is Underdetected” (unpublished manu-
script 2007).
 social influence 547

affected by norms of people who are similar to them and groups with which they
identify.
This category of “similar” others can actually extend to complete strangers,
who happen to be (or to have been) in situations similar to our own. For exam-
ple, Noah Goldstein, Robert Cialdini, and their colleagues found that hotel guests
were most likely to participate in a towel reuse program when they were informed
that the majority of people who had stayed in the same room had participated in
the program.23 This message was even more effective than one providing the
same information about the behavior of people of one’s gender, an identity that
generally seems more salient than people who happened to have occupied the
same hotel room. Goldstein, Cialdini, et al. suggest that when we are deciding
how to behave in a particular situation, the social information that is most rele-
vant is how other people behave in that same situation.
Descriptive norm information has also been found to affect charitable giving.
Rachel Croson and her colleagues collaborated with public radio stations to con-
duct research during the stations’ “pledge drives.” In these studies, some listen-
ers who called in to make donations were randomly selected to receive different
information regarding donation amounts. For example, before being asked how
much money they would like to donate, some listeners were told that another
listener had just donated $200. Listeners receiving this information donated, on
average, $30 more than callers who were simply asked how much they would
like to donate. (The researchers controlled for the anchoring effect.)
In many situations, it is possible to make use of these sorts of techniques
without engaging in deception. In the absence of other data, information about
one individual can be taken as an indication of the norm. And it is highly likely
that radio stations and other organizations that depend on charitable donations
can truthfully provide information about at least one generous donor. Moreover,
in some cases, such as Cialdini’s study of neighborhood energy conservation,
many people actually may engage in a desired behavior.

17.1.4 Conformity Effects and the Emergence of Cascades

In recent years, a number of scholars have used the social psychological research
on conformity to develop an account of how behavioral norms, viewpoints, and
perceptions of risk can spread rapidly through both large and small groups. This
account, often referred to as cascade theory, offers important insights into the
relationship between individual rationality and group behavior. It also helps
explain why apparently rational people, who seem to be making independent
decisions, might adopt the views of others in their social environment.

23. Noah J. Goldstein, Robert B. Cialdini, and Vladas Griskevicius, A Room with a
Viewpoint: Using Normative Appeals to Motivate Environmental Conservation in a Hotel
Setting, 35 Journal of Consumer Research (2008) (electronically published without
page numbers).
548 problem solving, decision making, and professional judgment

Timur Kuran and Cass Sunstein describe three different types of cascades:
informational cascades, reputational cascades, and availability cascades, all three of
which can derive from conformity-related effects.24 An informational cascade
occurs when people with incomplete personal information about a subject base
their own beliefs about it on the apparent beliefs of others in their social environ-
ment. People’s words and actions often convey the impression that they hold a
particular belief. Cascade theory posits that in response to such communica-
tions, other individuals, who know that they personally lack reliable information
on the subject, accept a particular belief simply by virtue of its apparent accep-
tance by others. Under proper conditions, such a belief, no matter how fanciful,
can spread rapidly and widely through a population.
Reputational cascades play a significant role in this process. Even if they doubt
the veracity of the belief spreading by means of an informational cascade, people
may act and speak as if they believe it to be true. Originally, at least, they do this
to earn social approval and avoid social disapproval. Of course, once a person
adopts and repeats a belief, even if only to avoid censure, he or she may eventu-
ally come to believe it true, if only to reduce the cognitive dissonance that would
otherwise result.25 Reputational cascades can prove very powerful, as direct social
pressures to adopt a fast-propagating belief reinforce concerns that deviance
would have negative long-term effects on social networks.26
Informational and reputational cascades look similar, often co-occur, and
drive the same rapid propagation of beliefs. As theoretical constructs, they differ
only in their motivation. People participate in an informational cascade primar-
ily because they view adoption of an apparently sound belief held by others as
rational, given their own lack of information and the costs associated with inde-
pendent information search. For this reason, information cascades are also
sometimes referred to as “rational herding.”27 In contrast, people participate in
reputational cascades to avoid social disapproval, or to enhance their standing in
a community.
Informational and reputational cascades sometimes interact with the availabil-
ity heuristic (Section 9.6) in ways that create what Kuran and Sunstein call an
availability cascade. In these situations, informational and reputational cascades

24. Timur Kuran and Cass R. Sunstein, Availability Cascades and Risk Regulation, 51
Stanford Law Review 683 (1999).
25. Daryl J. Bem, Self-Perception Theory, in Advances in Experimental Social
Psychology 6, 60–61 (Leonard Berkowitz ed., New York: Academic Press, 1972).
26. Michael Klausner, Corporations, Contracts, and Networks of Contracts, 81 Virginia
Law Review 757 (1995); Marcel Kahan and Michael Klausner, Path Dependence in Corporate
Contracting: Increasing Returns, Herd Behavior and Cognitive Bias, 74 Washington Law
Quarterly 347 (1996).
27. Eric Talley, Precedential Cascades: An Appraisal, 73 Southern California Law
Review 87 (1999).
 social influence 549

result in the frequent repetition of a particular story, trope, or claim. By virtue of

its frequent repetition, it becomes increasingly available. Through the operation of
availability bias, it then seems more plausible and likely. Availability cascades can
be accelerated through the work of availability entrepreneurs, who fix people’s atten-
tion on a particular issue, spin that issue in a particular way, and through these
processes shape people’s perceptions of its nature, importance, and meaning.
Kuran and Sunstein illustrate the cascade effects through various actors’
responses to the arguable presence of toxic wastes in a small town. Vivid infor-
mation about the dangers posed traveled through the community through infor-
mation cascades. Public officials, who wanted to maintain a good reputation
with their constituents, took up the issue and urged action, even though they had
little or no knowledge about the actual harms. And some concerned citizens and
environmental organizations acted as availability entrepreneurs to intensity and
broaden the communications with both groups.

17.1.5 The Strategic Deployment of Social Proof

People sometimes manipulate the impulse toward conformity to achieve
their own ends. For example, when a reporter is trying to get an employee to
disclose information about confidential events that occurred in her workplace,
the reporter may (falsely) tell her that some of her coworkers have already dis-
closed information about the events. The belief that others have already talked
makes the employee’s disclosure seem more acceptable and makes her more
likely to talk. Reference to the coworkers behavior provides a cue for the employ-
ee’s appropriate behavior in circumstances in which the proper response is
ambiguous.
Form contracts, such as unfavorable provisions in lease agreements or manda-
tory arbitration agreements, provide similar cues. A tenant or medical patient may
say to himself, “There are a lot of people like me signing these types of contracts.
This must be the way it’s done.” Through similar processes, as we discuss in
Section 18.6.3.c, default rules may acquire normative power through social proof.

***
In summary, people look to the behavior of others in their social environ-
ments as a kind of heuristic, applied to resolve uncertainties over what they
should think or how they should behave. Other things being equal, tendencies
toward social conformity are stronger when people are in novel environments, or
where behavioral norms are particularly complex and ambiguous, or where the
negative sanctions attached to behavioral deviance are particularly harsh. This
tendency to conform our own behavior—and our sense of what is right, norma-
tive, or accurate—makes it possible for us to get along with other people, master
new roles and environments, and achieve our individual and collective goals. But
it can also cause us to behave in ways we later recognize as abhorrent.
Unfortunately, that recognition often comes too late.
550 problem solving, decision making, and professional judgment

17.2 obedience to apparent authority

At the end of World War II, descriptions of Nazi genocide shocked American
society and triggered, in academic communities as elsewhere, attempts to explain
why so many people collaborated with Hitler’s atrocities. In 1950, a group of
social scientists led by German exile, Theodor Adorno, provided one explanation
in The Authoritarian Personality.28 They posited that certain people possessed par-
ticularly rigid personalities, characterized by social conservatism, conformity,
authoritarianism, and a strong inclination to obey authority. This personality
type, Adorno concluded, was widespread in Germany in the 1930s and ‘40s and
made National Socialism possible.
Building on Adorno’s notion that nationality was correlated with conformity
and compliance with authority social psychologist Stanley Milgram later set out
to demonstrate that the more “contentious” French were less likely to conform
in an Asch paradigm than the more “retiring” Scandinavians.29 After failing to
find significant differences between French and Scandinavian subjects, Milgram
then designed an experiment to identify the “type of person” who would obey the
instructions of an apparent authority figure—even if the instruction was to do
something harmful to another person.
In Milgram’s now classic study,30 subjects were asked to play the role of
“teacher” by asking questions of a “learner” who was hooked to an instrument
that vaguely resembled an electric chair. Upon first entering the laboratory, the
subject, or “teacher,” assisted an experimenter, dressed in a white lab coat, in
strapping the “learner” into the chair and attaching a shocking device to his arm.
(See Figure 17.3.)
The “teacher” was then led into an adjacent room and seated in front of a
panel with a row of switches on it, each marked with a particular voltage level,
ranging from 15 to 450 volts and rising in 15 volt intervals. Under the switches
were descriptions of the levels of the shock each would trigger, ranging from
“slight shock” to “danger: severe shock.” The last two switches were ominously
labeled “XXX.”
The experimenter then explained to the subject that he should ask the learner
a question and, if the learner failed to respond correctly, the subject should
administer a shock. Each time the learner failed to respond correctly, the shock
to be administered was to increase by 15 volts.

28. Theodor W. Adorno, Else Frenkel-Brunswik, Daniel J. Levinson, and R.

Nevitt Sanford, The Authoritarian Personality (New York: Harper, 1950).
29. Stanley Milgram, Nationality and Conformity, Scientific American 45–51
(November 1961).
30. Stanley Milgram, Behavioral Study of Obedience, 67 Journal of Abnormal and
Social Psychology, 371–78 (1963).
 social influence 551

figure 17.3 a subject (left) assists experimenter in strapping

shocking device onto “learner” in milgram obedience experiment
at yale university.

Milgram’s subjects were not college students. They were adult men from
diverse occupational backgrounds who responded to a newspaper advertisement
for a study on “learning” to be conducted at Yale University.
Unbeknownst to subjects, the learner was actually a confederate of the exper-
imenter. Prerecorded sounds (e.g., screams, protests) reflecting varying degrees
of pain were played, depending on the level of voltage the subject was “applying”
to the learner. When the subject indicated a desire to stop either questioning the
learner or applying the shock, he was prompted with one of four cues, depend-
ing on the nature of the subject’s hesitation: “Please continue”; “The experiment

figure 17.4 milgram ’ s “shock panel”.

552 problem solving, decision making, and professional judgment

requires that you continue”; “It is absolutely essential that you continue”; or
“You have no other choice, you must go on.”
Surprisingly, even though they were not coerced or threatened into obeying
the experimenter’s instructions, nearly 65 percent of the subjects were willing to
continue until they had applied what appeared to be life-threatening levels of
electricity to those who failed to provide correct answers. Milgram reported that
“subjects were observed to sweat, tremble, stutter, bit[e] their lips, groan, and dig
their fingernails into their own flesh.” Some subjects cried; others burst out into
hysterical laughing fits. In post-experiment debriefings, nearly all the subjects
reported that they felt that it was wrong to apply the shocks.
Neither Milgram nor any of the social psychologists, psychiatrists, or others
he consulted anticipated that people would comply at such high rates, or to so
great an extent. And yet, in subsequent replications the results were similar at
different locations with different subjects. By the end, Milgram and his associ-
ates had run literally thousands of people from a wide cross-section of American
society, including postal workers, college students, high school students and
teachers, construction workers, college professors, and salespeople. Obedience
was the norm. As Milgram concluded, “It is the extreme willingness of adults to
go to almost any lengths on the command of an authority that constitutes the
chief finding of the study.”31
Obedience effects are robust across many situations and groups of subjects.
However, as is described in greater detail below, subtle changes in the situation
can have profound effects on obedience behavior. Individual differences play
some role as well.32 Not surprisingly, authoritarianism33 sometimes predicts
higher levels of obedience. Trust, moral development, and conventionality also
show some (albeit weak or inconsistent) effects on obedience levels.34
All but one of Milgram’s trials included only male subjects, but most of the
experimental evidence shows no gender effect on obedience to apparent
authority,35 In the one trial including females, Milgram found no sex differences
in obedience rates.36 Ten of eleven replications found the same result.37 The one

31. Ibid.
32. For a review of the literature, see Thomas Blass, Understanding Behavior in the
Milgram Obedience Experiment: The Role of Personality, Situations, and Their Interactions, 60
Journal of Personality and Social Psychology 398–413 (1991).
33. Theodor W. Adorno, Else Frenkel-Brunswik, Daniel J. Levinson, and
R. Nevitt Sanford, The Authoritarian Personality (New York: Harper, 1950).
34. See Thomas Blass, The Migram Paradigm after 35 Years: Some Things We Now Know
about Obedience to Authority, 29 Journal of Applied Social Psychology 955–78 (1991).
35. Ibid. 968–69.
36. Stanley Milgram, Obedience to Authority: An Experimental View (New York,
NY: Harper & Row, 1974).
37. Thomas Blass, supra note 34, 968.
 social influence 553

exception is an Australian study in which women had significantly lower obedi-

ence rates than men.38
Other experiments conducted in the 1960s and 1970s demonstrated the power
of compliance with authority. For example, in a study of nurses’ compliance with
physician’s orders, an experimental confederate placed a call to a hospital nurs-
ing station, falsely identified himself as a hospital physician, and instructed hos-
pital nurses to administer a dangerous dose of an unauthorized drug to a specific
hospital patient.39 Even though the “doctor” was not familiar to them, and even
though most nurses would realize that the dosage level was dangerous, 95 per-
cent of the nurses attempted to comply. Experiments aside, it turns out that
nurses’ and pharmacists’ unquestioning compliance with obvious physician
errors on prescriptions is a significant cause of medical error.40

17.2.1 The Impulse Toward Obedience: Legal and Policy Implications

Research on obedience to authority has important legal and policy implications.
Consider, for example, noncustodial searches by police officers. Obedience
research suggests that powerful psychological forces are at work in police-citizen
encounters, where a pronounced discrepancy usually exists between the status
of the detained citizen and the status of the police officer, often accentuated by
his uniform. As one commentator has suggested, “Suspects faced with an offi-
cial request to search may instinctively feel that their options are limited based
on a reflexive tendency to obey authority, even when police conduct falls far short
of overt coercion.”41
Police lineup procedures present another example of the policy implications
of the psychology of obedience. An officer’s instruction to a crime victim to “pick
the one who did it,” for example, can convey a subtle but powerful message to the
victim that they must choose someone in the lineup group. Such biased instruc-
tions have been shown to increase the frequency of false identifications.42
The graphic and perverted display of prisoner abuse at Abu Ghraib prison
in Iraq, captured in photographs taken by the soldiers themselves, provides

38. Wesley Kilham and Leon Mann, “Level of Destructive Obedience as a Function of
Transmitter and Executant Roles in the Milgram Obedience Paradigm,” Journal of
Personality and Social Psychology, 29 (1974): 696–702.
39. Charles K. Hofling et al., An Experimental Study in Nurse-Physician Relationships,
143 Journal of Nervous and Mental Disease 171–80 (1966).
40. Neil M. Davis and Michael R. Cohen, Medication Errors: Causes and
Prevention (Philadelphia: G. F. Stickley, 1981).
41. Adrian J. Barrio, Rethinking Schneckloth v. Bustamonte: Incorporating Obedience
Theory into the Supreme Court’s Conception of Voluntary Consent, University of Illinois
Law Review 215 (1997).
42. C. A. Elizabeth Luus and Gary L. Wells, Eyewitness Identification Confidence, in
Adult Eyewitness Testimony: Current Trends and Developments (David F. Ross, J.
Don Read, and Michael P. Toglia eds., New York: Cambridge University Press, 1994).
554 problem solving, decision making, and professional judgment

troubling evidence of the extent to which obedience can lead us astray. One of
the military police who served in Abu Ghraib reported that they were told by
superiors, “Use your imagination. Break them. We want them broke by the time
we come back.”43 The deadly environment, the pressure for intelligence, and the
ambiguity of how they were supposed to “break” the prisoners led to the abuse,
torture, and even death of detainees.

17.2.2 Defusing the Impulses Toward Obedience

In sum, we learn from a very early age to obey authority. That is not necessarily
a bad thing: society benefits from order. We benefit individually and collectively
from the things we learn and the disasters we avoid by doing what those with
more knowledge and experience tell us to do. But, like so many rules, the injunc-
tion to “obey authority” can be overapplied, particularly in situations that supply
powerful cues for compliance. The thought should chill anyone in a subordinate
position, especially in an unfamiliar environment.
But despite the powerful pull of obedience, it is possible to defy authority to
support one’s principles. The abuse at Abu Ghraib continued until Joe Darby, a
military policeman on a different assignment at the prison, saw the photographs
of the abuse. “It didn’t sit right with me,” says Darby. He was torn between loy-
alty to his friends and the moral offensiveness of the acts, which “violated every-
thing I personally believed.” After three days of weighing the competing
demands, he reported the photographs.44
Consider Hugh Thompson, a helicopter pilot whose regiment perpetrated the
My Lai Massacre in Vietnam. With all of his comrades—including his superior
officers—shooting villagers and looting huts, Hugh Thompson got out of his
helicopter, stood in the line of fire between his fellow soldiers and a house where
wounded Vietnamese civilians were seeking refuge, and told his fellow soldiers
to cease fire.45
Darby’s and Thompson’s experiences remind us that although the forces that
compel us to obey authority are strong, people can and do stand up to authority
even when the consequences are grave. But it is also incredibly difficult to do. To
control the unthinking impulse to obey authority, we need to understand its
dynamics. Insights come from a number of sources, some of them in Milgram’s
experiments themselves.
Milgram conducted a total of nineteen variations of his basic experimental
design. Under some conditions, rates of compliance dropped from the original
63 percent baseline, and under others, they increased. Rates of compliance rose
when subjects first watched peers administering the shocks (showing the

43. Phillip C. Zimbardo, The Lucifer Effect: Understanding How Good People
Turn Evil, supra note 3, 352.
44. Id. at 330.
45. Id. at 474.
 social influence 555

interactive relationship between compliance and conformity effects). Rates of

compliance were highest when subjects were high school students and the
experiment was conducted by lab-coated experimenters on the grounds of
Princeton University. Compliance rates dropped precipitously when another
subject rebelled. Contradictory instructions by two experimenters, having
instructions given by an ordinary person rather than a white-lab-coated experi-
menter, placing the subject in close proximity to and/or visual contact with the
“learner,” and allowing subjects to choose the level of shock they would admin-
ister, all decreased compliance rates significantly.
As numerous social psychologists have observed, certain structural features
of the situation Milgram constructed made it particularly powerful in eliciting
high rates of compliance. At the beginning of the session, the procedure was
explained to subjects. By agreeing to participate, they implicitly consented to
following the instructions. The intensity of the shocks increased gradually,
15 “volts” at a time. Having given a shock of just slightly less intensity than the
one they were now being asked to administer made it more difficult for subjects
to refuse. (See Section 17.3 on consistency and commitment.) When subjects
did object, the experimenter’s icy responses “You have no choice,” The experi-
ment requires that you continue,” or “You have no other choice, you must go on,”
left open no “disobedience channel” through which subjects could exit without
“making a scene.” Most people don’t like to make a scene.
Imagine, as social psychologist Lee Ross suggested, that there had been a
large red button on the shock generator panel in front of Milgram’s subjects
labeled, “Stop the Experiment.” How would that have affected compliance rates?
So far as we know, the experiment has not been done, but our intuitions support
Ross’s suggestion that the presence of such a button would have served as a
channel factor for disobedience and would have reduced compliance.46
Milgram concluded that the impulse toward obedience is “a deeply ingrained
tendency, indeed a prepotent impulse overriding training in ethics, sympathy,
and social conduct.”47 Robert Cialdini suggests that the rule, “obey legitimate
authority figures,” functions as a heuristic, or rule of thumb, that people use to
decide how to act when conditions are ambiguous. He proposes that the heuris-
tic is automatically activated when features of the social environment provide
cues to its applicability.48

46. Lee Ross, Situationist Perspectives on the Obedience Experiments, 33 Contemporary

Psychology 101–04 (1988).
47. Stanley Milgram, Behavioral Study of Obedience, supra note 30.
48. Robert B. Cialdini, Influence: The Psychology of Persuasion, supra note 3.
Robert B. Cialdini, Compliance Principles of Compliance Professionals: Psychologists of
Necessity, in Social Influence From The Ontario Symposium 5 165–84 (Mark P.
Zanna, James M. Olson, and C. P. Herman eds., Hillsdale, NJ: Erlbaum, 1987).
556 problem solving, decision making, and professional judgment

This way of understanding Milgram’s results suggests several approaches to

inoculating oneself against automatic obedience. Psychologists Philip Zimbardo
and Michael Lieppe suggest:49
• remain engaged with alternate systems of authority, such as a those deriv-
ing from your religious, spiritual, political, or philosophical commitments;
• trust your intuition when you find yourself thinking, “something is wrong here”;
• when in doubt, seek out a knowledgeable—but independent—person to
give you an “obedience check”;
• mentally rehearse disobedience strategies and techniques for various
situations;
• be particularly vigilant when you notice people using euphemisms to
describe harmful behaviors or the people they harm;
• don’t expect not to suffer adverse consequences when refusing to obey an
authority figure—rather, consider the worst case scenario and act on that
possibility;
• chose carefully the organizations and situations in which you place your-
self, because it’s all too easy to overestimate your powers to resist.

17.3 consistency and the escalation of commitment

In Section 14.3 we discussed various factors that lead people to take “sunk costs”
into account when making decisions about the future. We continue that discus-
sion here, focusing more on social determinants.
Once we have taken a position, whether through an activity or communica-
tion, we tend to behave in a manner consistent with that position and to accede
to requests that are consistent with it. Why people might act this way is no mys-
tery: the social costs of changing one’s mind or of otherwise appearing inconsis-
tent can be steep.

17.3.1 The Foot in the Door

How might one persuade residents of Palo Alto, California, to place large, ugly
“Drive Safely” signs on their tidy front lawns? When Jonathan Freedman and
Scott Fraser asked Palo Alto homeowners to let them place an enormous, tacky,
hand-painted sign on their front lawns, only 16.7 percent agreed. But in the
experimental condition, in which homeowners had agreed two weeks earlier to
place a three-inch sign with the same message on their front lawns, 76 percent
acceded to the request.50

49. Philip. G. Zimbardo and Mark R. Lieppe, supra note 4, 75 (1991).

50. Jonathan L. Freedman and Scott C. Fraser, Compliance Without Pressure: The
Foot-in-the-Door Technique, 4 Journal of Personality and Social Psychology 195 (1966).
 social influence 557

The key to this foot-in-the-door technique is that compliance with the first
request causes the target to think of himself in a way that makes him more likely
to accede to the second through a process of self-attribution.51 That is, by freely
agreeing to perform and then performing the first action, the foot-in-the-door
target comes to view himself as “the kind of person” who would do things of that
kind or support the position expressed by the first action.
Fundraisers use the foot-in-the-door technique all the time, which is why your
graduate school will try to get you to make a donation, no matter how small,
before you have earned your first dollar as a lawyer or policy professional—
perhaps even while you are still paying tuition. Lawyers and investigators use the
technique, too, for example, when they are trying to persuade reluctant witnesses
to discuss a case with them. Later, perhaps even in the same case, a mediator
may use the technique in an attempt to help the parties reach a settlement.
Police officers use the foot-in-the-door technique when dealing with suspects.
For instance, a police officer may ask permission to enter a residence “just to
talk” (first request). Once inside, and after some friendly conversation, she may
then ask for consent to conduct a search (bigger second request). One can readily
see from this example how various social influence factors, in this case both
obedience to authority and the consistency and commitment tendency, could
work together to amplify compliance.

17.3.2 The Low-Ball Technique

The tendency toward consistency and commitment can work in other ways to
induce compliance in the unwary. After a person has made a tentative agree-
ment, say, to settle a case for a specific amount, or to accept a job offer on par-
ticular terms, she will tend to follow through with that agreement even if the
terms of the deal are subsequently made less favorable to her. Social psycholo-
gists (and car salespeople) refer to the phenomenon as the “low-ball effect.”
The best-known experimental demonstration of the low-ball effect comes
from a study by Robert Cialdini conducted in the late 1970s. In one condition,
Cialdini asked students enrolled in an introductory psychology class if they
would be willing to participate in a study “on thinking processes” to be held at
7:00 a.m. Only 24 percent agreed. A different group from the same class was
simply asked if they would participate in a study of “thinking processes,” with no
time specified; 56 percent agreed. Cialdini and his associates then told these
students that the study would start at 7:00 a.m. None reneged, despite having the
opportunity to do so.52

51. Donald R. Gorassini and James M. Olson, Does Self-Perception Change Explain the
Foot-in-the-Door Effect?, 69 Journal of Personality and Social Psychology 91 (1995).
52. Robert B. Cialdini, John T. Cacioppo, Rodney Bassett, and Geoffrey A. Miller, Low-
Ball Procedure or Producing Compliance: Commitment then Cost, 36 Journal of Personality
and Social Psychology 463–76 (1978).
558 problem solving, decision making, and professional judgment

As Richard Birke and Craig Fox have observed, foot-in-the-door, low-ball, and
other aspects of the consistency and commitment tendency often converge in
affecting the parties’ behavior in mediations.53 First, mediators often begin by
inducing the parties to commit to a general statement of principles, or an aspira-
tion. Once the mediator has this foot in the door, she is more likely to induce
party compliance with later, more substantial requests. The low-ball technique
also often comes into play, as Birke and Fox note. Once the parties have invested
significant time and energy into crafting a tentative agreement, they can often be
induced to agree to last-minute modifications, even if they make the deal less
beneficial to them.

17.3.3 The Irrational Escalation of Commitment

Many situations provide fertile ground for the irrational escalation of commitment.
These include decisions facing bank officers, who must decide whether to loan
more money to a defaulting client or write off the loan; lawyers and clients decid-
ing whether to settle a suit or continue its litigation; and managers deciding
whether to fire an employee or invest additional resources in helping him
improve.54 All these situations have a common structure. First, they involve a
continuing course of action toward a desired end with periodic opportunities for
exit. Second, at some point along the way, some loss or cost is incurred, making
achievement of the goal less likely, or more costly than previously anticipated.
Third, despite this initial cost or loss, ultimate success is still possible, although
it would require the investment of additional resources. Finally, withdrawal from
the situation would impose an immediate loss or cost, in contrast with the even-
tual success that perseverance could theoretically bring.
As business administration professors Barry Staw and Jerry Ross explain,
“escalation situations can be defined as predicaments where costs are suffered in
a course of action, where there is an opportunity to withdraw or persist, and
where the consequences of persistence and withdrawal are uncertain.”55
The escalation cycle begins when a questionable or negative outcome—the
loss of an initial motion to dismiss a lawsuit, for example—has been experi-
enced. At that point, the actors (e.g., a litigation defendant and his lawyer) must
reexamine their commitment to the original course of action (i.e., defending the
suit) and decide whether to withdraw (i.e., settle on terms less favorable than

53. R. Birke and C. R. Fox, Psychological Principles in Negotiating Civil Settlements,

1 Harvard Negotiation Law Review 61 (1999).
54. Barry M. Staw, Sigal G. Barsade, and Kenneth W. Koput, Escalation at the Credit
Window: A Longitudinal Study of Bank Executives’ Recognition and Write-off of Problem
Loans, 82 Journal of Applied Psychology 130–42 (1997).
55. Barry M. Staw and Jerry Ross, Understanding Behavior In Escalation Situations, in
Psychological Dimensions of Organizational Behavior 206–13 (Barry M. Staw ed.,
3rd ed. Upper Saddle River, NJ: Pearson Prentice Hall, 2004).
 social influence 559

previously deemed acceptable), or press forward (i.e., undertake discovery and

prepare for summary judgment and/or trial). In making this assessment, the
lawyer-client team will balance the perceived utility of continuing in the current
course of action (litigation) against the perceived utility of exit (settling) and
decide whether or not to recommit and invest additional resources.
A substantial amount of research has demonstrated that people and organiza-
tions tend to escalate their commitment to a failing course of action beyond
rational levels.56 Additional research has illuminated the particular conditions
under which such irrational escalation of commitment is most likely to occur.57

Why Cases Don’t Settle Early and Often

Although the vast majority of civil cases settle, they often settle very late, after
enormous amounts of money have been spent on pleadings, discovery, and
motion practice.58 There are, of course, many reasons for this, including the asym-
metrical valuation of losses and gains, the financial incentives motivating lawyers
who bill by the hour, the persistence of information asymmetries between the
parties, and the various cognitive distortions that comprise what was referred to
in Section 14.3 as the sunk costs fallacy. But the escalation of commitment likely
also plays a role in keeping cases from settling optimally.

Explanatory accounts of escalation tend to cluster around four sets of factors

that conduce to its occurrence. Barry M. Staw and Jerry Ross classify these as: (1)
project determinants; (2) psychological determinants; (3) social determinants;
and (4) and organizational determinants.
With respect to project determinants, some projects by their very nature entail
a long delay between investment and payoff, with setbacks and revenue short-
falls reasonably expected along the way. Research and development projects are
one example. Civil litigation is another. Decision making about projects such as
these is particularly vulnerable to escalation effects because of the expected delay
between expenditures and the resulting benefits. Because of the long time hori-
zon, initial losses or shortfalls do not raise alarms, and organizations are unlikely

56. See Max Bazerman, Judgment in Managerial Decision Making (5th ed., New
York: Wiley, 2002) for a review.
57. Barry M. Staw and Jerry Ross, Behavior in Escalation Situations: Antecedents,
Prototypes, and Solutions, in Research in Organizational Behavior (Barry M. Staw and
Larry L. Cummings eds., Greenwill, CT: JAI Press, 1987); Joel Brockner and Jerry Z.
Rubin, Entrapment in Escalating Conflicts (New York: Springer-Verlag, 1985).
58. Samuel R. Gross and Kent D. Syverud, Getting to No: A Study of Settlement
Negotiations and the Selection of Cases for Trial, 90 Michigan Law Review 319–93 (1991).
560 problem solving, decision making, and professional judgment

to alter course.59 Other projects may conduce toward escalation of commitment

because they have little or no salvage value, or because they involve very high
closing costs if abandoned mid-stream.60
We described in earlier chapters many of the cognitive biases constituting the
major psychological determinants of irrational escalation of commitment. Primary
among these is confirmation bias, through which people preferentially seek and
then overweight evidence supporting their prior beliefs and ignore or rationalize
evidence opposing them. Other psychological factors include the power of inter-
mittent reinforcement,61 ego-enhancing bias,62 belief perseverance,63 the desire
to avoid regret and negative self-attribution.64
Organizational determinants of suboptimal escalation include the absence of a
clearly defined exit strategy, past failure to hedge against the possibility that the
chosen strategy would fail, and whether the decision to escalate or withdraw is
made by the same or different people than those who made the original decision
to invest or reinvest.65 The strength of the escalation tendency may also depend
on whether the purpose of the project in question is closely tied to the values or
purposes of the organization that has undertaken it,66 or on factors as banal as
whether the administrative resources needed to formulate and effectuate an exit
strategy are readily available.
Of greatest relevance to this section, however, are the social determinants of esca-
lation, as these relate most closely to the consistency and commitment tendency

59. Barry M. Staw and Jerry Ross, Behavior in Escalation Situations: Antecedents,
Prototypes, and Solutions, supra note 57; John Platt, Social Traps, 28 American Psychologist
642–43 (1973).
60. Gregory B. Northcroft and Gerrit Wolf, Dollars, Sense, and Sunk Costs: A Life Cycle
Model of Resource Allocation Decisions, 9 Academy of Management Review 225–34
(1984).
61. Barry M. Staw and Jerry Ross, Behavior in Escalation Situations: Antecedents,
Prototypes, and Solutions, supra note 57; John Platt, Social Traps, supra note 59.
62. Barry M. Staw, Knee-Deep in the Big Muddy: A Study of Escalating Commitment to a
Chosen Course of Action, 16 Organizational Behavior and Human Performance 27–44
(1976); Max Bazerman, R.I. Beekum, and F. D. Schoorman, Performance Evaluation in a
Dynamic Context: A Laboratory Study of the Impact of Prior Commitment to the Ratee, 67
Journal of Applied Psychology 873–876 (1982).
63. Richard E. Nisbett and Lee Ross, Human Inference: Strategies and
Shortcomings of Social Judgment (Englewood Hills, NJ: Prentice-Hall, 1980).
64. Barry M. Staw and Jerry Ross, Behavior in Escalation Situations: Antecedents,
Prototypes, and Solutions, supra note 57.
65. Ibid.
66. Lynn G. Zucker, Organizations as Institutions, in Research in The Sociology
of Organizations 2, 53–111 (Samuel B. Bacharach ed., Greenwich, CT: JAI Press, 1981);
Paul S. Goodman, Max Bazerman, and Edward Conlon, Institutionalization of Planned
Organizational Change, in Research in Organizational Behavior: Vol. 2, 215–46
(Barry M. Staw and Lynn L. Cummings eds., Greenwich, CT: JAI Press, 1980).
 social influence 561

described earlier in this section. In an escalation situation, certain social determi-

nants function to increase the perceived utility of recommitment to an ongoing
course of action, while others tend to decrease the perceived utility of exit.
Consider in this regard decisions between two options: settling an ongoing law-
suit on disappointing terms, or continuing to litigate. Even after significant monetary
losses or strategic setbacks, lawyers and clients will be pushed toward recommit-
ment by social norms of perseverance, to which innumerable sayings like “keep on
keepin’ on,” “venceremos,” “against all odds,” and “dare to struggle, dare to win”
bear witness. This norm of perseverance is reinforced by media representations of
successful lawsuits and the lawyer-heroes that litigated them. Think, for example,
how bleak the prospects looked for the plaintiffs portrayed in the movies Erin
Brockovich and The Verdict, and recall how glorious were their eventual victories.
Just as perseverance norms and scripts may inflate the perceived utility of
recommitment, so other factors may diminish the perceived utility of withdrawal.
Lawyers are notoriously competitive, and the litigation process seldom turns
adversaries into friends. A litigant or his lawyer may want to persevere if only to
deny the opponent the savor of success. To the extent that the litigation has been
acrimonious (and it often is), any existing sense of rivalry be amplified.
If a party or his lawyer has taken a public position on a dispute, pressures to
escalate commitment may be intensified even further. Public statements of com-
mitment powerfully bind people to a course of action, particularly when actions
taken in connection with it are unambiguous and freely undertaken.67
In sum, the consistency and commitment tendency, combined with other
psychological forces and situational factors, conduces toward the irrational esca-
lation of commitment to a questionable or failing course of action. As with other
systematic biases and tendencies toward misjudgment, it is possible to reduce—
though probably not eliminate—the escalation syndrome. Staw and Ross, for
example, make the following suggestions:
• When an ongoing course of action has led to unexpected losses or setbacks
and a decision regarding recommitment or exit must be made, involve new
decision makers who were not responsible for the original decision to commit.
• Hedge possible losses on one project with other opportunities for gain.
• Provide organizational support for decision makers to reverse course, and
lower the personal costs of an exit decision.
• In situations where irrational escalation is a possibility, provide decision
makers with unambiguous feedback, preferably gathered and presented by
people with minimal connection to the original decision.

67. Barry M. Staw and Jerry Ross, Behavior in Escalation Situations: Antecedents,
Prototypes, and Solutions, supra note 57; Gerald R. Salancik, Commitment and the Control of
Organizational Behavior and Belief, in New Directions in Organizational Behavior
1–54 (Barry M. Staw and Gerald R. Salancik eds., 1977).
562 problem solving, decision making, and professional judgment

• Calculate the costs and make a plan for closing at the outset of the
project.

17.4 liking

We tend to like those who like us. We tend to keep track of “credits and debits”
with people we are not close to, but are less likely to do so for those we like a lot.
We tend to credit views expressed by people we like and discredit views expressed
by people we dislike. We prefer to say “yes” to those we like. For these and other
reasons, “liking” can impair our judgments and make us more susceptible to
influence.68 Several factors promote liking, including: familiarity, similarity,
cooperation, compliments, and physical attractiveness.69

17.4.1 Familiarity
All else being equal, we tend to like people and things that are more, rather
than less, familiar. As the late Stanford psychologist Robert Zajonc demon-
strated, mere exposure to an initially neutral or favorable stimulus enhances its
evaluation.70 So, for example, people express greater liking of, and express higher
levels of agreement with, people whose faces they have encountered frequently,
even in subliminal exposures.71 These sorts of exposure effects appear to be even
stronger in the real world than they are in the laboratory.72

68. A study described in Section 17.5 leaves open the question of how effective liking
is in gaining compliance. Dennis T. Regan, Effects of a Favor and Liking on Compliance, 7
Journal of Experimental Pscyhology 627–39 (1971). Liking for a male confederate was
manipulated (i.e., the subject overheard him behave in either a “pleasant” or “nasty” way
to someone with whom he was speaking on the phone), and the participants received a soft
drink (a favor) from either the confederate, or from the experimenter, or received no favor
at all. The confederate then asked the participants to purchase some raffle tickets. Regan
found that the favor strongly increased compliance with the request, but he did not find
a similar effect for liking. However, he cautioned that “[i]t would not be warranted to con-
clude, on the basis of the manipulation used in this study, that liking does not generally
affect compliance.” Id. at 635.
69. Different social psychologists proffer slightly different taxonomies. Ours repre-
sents a combination of Robert Cialdini’s and Susan Fiskes’. See Robert Cialdini,
Influence: Science and Practice 136–53 (3rd ed. New York: HarperCollins, 1993) and
Susan T. Fiske, Social Beings, supra note 4, 257–276.
70. Robert. B. Zajonc, Attitudinal Effects of Mere Exposure, 9 Journal of Personality
and Social Psychology 1–27 (1968).
71. Robert F. Bornstein, Dean R. Leone, and Donna J. Galley, The Generalizability of
Subliminal Mere Exposure Effects: Influence of Stimuli Perceived without Awareness on Social
Behavior, 53 Journal of Personality and Social Psychology 1070–79 (1987).
72. Robert F. Bornstein, Exposure and Affect: Overview and Meta-Analysis of Research
1968–1987, 106 Psychological Bulletin 265–89 (1989).
 social influence 563

17.4.2 Similarity
Not surprisingly, we tend to like those who share our cultures, values, and beliefs
more than we like people who don’t. Common notions that “opposites attract”
notwithstanding, the vast weight of social psychological evidence supports the
proposition that people are preferentially attracted to and are more satisfied with
their interactions with people whose attitudes, interests, and personalities are
similar to their own.73 With respect to the effect of attitude similarity on liking,
proportions are critical. “If a person only agrees with us on twelve out of twenty-
four topics, he or she is not liked as much as another who agrees with us on four
out of six topics.”74
The similarity principle has obvious legal policy implications, as various empir-
ical researchers have demonstrated. In one such study, mock jurors who per-
ceived a defendant as sharing many of their beliefs and attitudes were less inclined
to find him guilty and were more inclined toward leniency in sentencing.75
Results like these have obvious implications for jury selection, judicial qualifica-
tions, and legal advocacy.76

17.4.3 Reciprocity and Cooperation

As Susan Fiske observes, “Next to similarity, reciprocity is the most powerful
predictor of attraction.”77 Cooperative interdependence is a powerful engine of
good feeling.
These observations have obvious implications for the practice of law—or pol-
itics. To the extent that cooperation engenders liking, lawyers and politicians
often induce good feeling, and from good feeling, compliance, by behaving in a
cooperative way towards others. Contrary to the popular belief that a successful
negotiator is one who ruthlessly intimidates and exploits her counterparts, a
cooperative relationship can be more effective for achieving mutually beneficial
and equitable outcomes.78 Moreover, studies of lawyers as negotiators have

73. Ellen Berscheid and Harry T. Reis, Attraction and Close Relationships, in Handbook
of Social Psychology 193–281 (Daniel. T. Gilbert, Susan T. Fiske, and Gardner Lindzey
eds., 4th ed. New York: McGraw-Hill, 1988).
74. Curt Bartol and Anne Bartol, Psychology and Law: Theory, Research and
Application, 3d Edition (Florence, KY: Wadsworth Publishing, 2004)
75. William Griffitt and Thomas Jackson, Simulated Jury Decisions: The Influence of
Jury-Defendant Attitude Similarity-Dissimilarity, 1 Social Behavior and Personality
(1973).
76. Curt Bartol and Anne Bartol, Psychology and Law: Research and
Application, supra note 75.
77. Susan T. Fiske, supra note 4, 270.
78. See Richard Birke and Craig R. Fox, Psychological Principles in Negotiating Civil
Settlements, 1 Harvard Negotiation and Law Review 1–57 (1999); Max H. Bazerman
and Margaret A. Neale, The Role of Fairness Considerations and Relationships in a Judgmental
Perspective of Negotiation, in Barriers to Conflict Resolution (Robert H. Mnookin,
564 problem solving, decision making, and professional judgment

shown that those who are cooperative are perceived by others as more effective,
on average, than lawyers who are not.79

17.4.4 Compliments
In the psychological literature, the use of compliments is classified under “ingra-
tiation tactics.” Ingratiation tactics are a class of strategic behaviors that are
designed to elicit liking from the target.80 The use of compliments to engender
liking is an artful skill, for if the tactic “becomes obvious to a target (entirely
transparent) [it] is likely to fail miserably at increasing a target’s liking for the
ingratiator,” and may even be counterproductive. To avoid transparency, for
example, a low-status person who can obviously gain from successful ingratia-
tion may do best to compliment a high-status person in an indirect manner,
such as by telling a third person what they like about the high-status person in a
way that allows the high-status person to “overhear” the compliment.81

17.4.5 Physical Attractiveness

Attractive people have advantages in social interaction. Physical attractiveness
operates as a powerful “halo effect,” influencing people’s judgments of other
people’s talent, kindness, honesty, and intelligence, among other traits.82
Many studies examining the effects of defendant attractiveness on verdicts
have found attractiveness to be associated with leniency in mock criminal
cases.83 These laboratory experiments are complimented by field studies showing

Lee Ross, Kenneth J. Arrow, and Amos Tversky, eds., Stanford, CA: Stanford Center on
Conflict and Negotiation, 1995), 86–107.
79. See Gerald R. Williams, Legal Negotiation and Settlement 19 (St. Paul,
Minnesota: West Publishing, 1983).
80. Other ingratiation tactics include opinion conformity, rendering favors, self-
deprecation, and modesty. See Randall A. Gordon, Impact of Ingratiation on Judgments and
Evaluations: A Meta-Analytic Investigation, 54 Journal of Personality and Social
Psychology 71 (1996).
81. However: “Developing a relationship takes time and must progress at its own pace.
In fact, pursuing friendliness as an influence tactic usually requires that the relationship
between the agent and the target already be in place before the request is made so the
relationship can be used effectively. If the agent tries to cultivate a relationship very quickly
and to use it simply as a vehicle in which to lodge the influence request, it is likely that the
target will see the friendliness gestures as superficial and insincere, a perception that will
raise the target’s defensiveness rather than lower it.” Roy J. Lewicki, Joseph A. Litterer,
John W. Minton and David M. Saunders, Negotiation 316 (McGraw Hill, 1994).
82. Alice H. Eagly, Richard D. Ashmore, Mona G. Makhijani, and Laura C. Longo,
What Is Beautiful Is Good: A Meta-Analytic Review of Research on the Physical Attractiveness
Stereotype, 110 Psychological Bulletin 109–28 (1990).
83. See Michael G. Efran, The Effect of Physical Appearance on the Judgment of Guilt,
Interpersonal Attraction, and Severity of Recommended Punishment in a Simulated Jury
Task, 8 Journal of Research in Personality 45–53 (1974) (finding that attractive female
 social influence 565

that, when other variables are controlled for, less physically attractive defendants
generally receive more severe sentences.84 In the civil litigation context as well,
the evidence that physically attractive individuals experience better legal out-
comes is both compelling and disturbing.85
Similar effects have been found in the context of hiring, voting, and helping
behavior. In all three contexts, decision makers seem unaware of the effect of
physical attractiveness on their judgment.86
Some qualifications to the case for physical attraction effects are in order here.
First, although physical attractiveness powerfully influences people’s judgments
of others, notions of attractiveness—and the extent of their impact on person
judgment—vary cross-culturally. For example, Chinese immigrants to Canada
did not show the effect in a study by Karen K. Dion, A. Won-Ping Pak, and
Kenneth L. Dion.87 In a Taiwanese study, the effect of attractiveness was moder-
ated by Western value orientation (perhaps showing the independent effect of
familiarity).88
Second, people who score higher in measures of self-consciousness show
stronger effects of physical attractiveness on their judgments than do people

defendants were less likely than unattractive female defendants to be found guilty by,
and receive lighter sentences from, male jurors); Martin F. Kaplan and Gwen D.
Kemmerick, Juror Judgment as Information Integration: Combining Evidential and Non-
evidential Information, 30 Journal of Personality and Social Psychology 493 (1974);
Gloria Leventhal and Ronald Krate, Physical Attractiveness and Severity of Sentencing, 40
Psychol. Rep. 315, 315–17 (1977) (finding that mock jurors recommended shorter sen-
tences for physically attractive defendants than for physically unattractive defendants,
regardless of juror’s gender, defendant’s gender, or seriousness of offense); John E.
Stewart, II, Defendant’s Attractiveness as a Factor in the Outcome of Criminal Trials: An
Observational Study, 10 J. Applied Soc. Psychol. 348 (1980); Wayne Weiten, The Attraction-
Leniency Effect in Jury Research: An Examination of External Validity, 10 J. Applied Soc.
Psychol. 340 (1980).
84. See, for example, J. E. Stewart II, Defendant’s Attractiveness as a Factor in the Outcome
of Criminal Trials: An Observational Study, 10 Journal of Applied Social Psychology
348–61 (1980).
85. For a literature review, see A. Chris Downs and Phillip M. Lyons, Natural
Observations of the Links Between Attractiveness and Initial Legal Judgments, 17 Personality
and Social Psychology Bulletin 541–47 (1990).
86. The studies are collected in Robert B. Cialdini, supra note 3, 141–42.
87. Karen K. Dion, A. Won-Ping Pak, and Kenneth L. Dion, Stereotypic Physical
Attractiveness: A Sociocultural Perspective, 21 Journal of Cross-Cultural Psychology
158–79 (1990).
88. David R. Shaffer, Nicole Crepaz, and Chien-Ru Sun, Physical Attractiveness
Stereotyping in Cross-Cultural Perspective: Similarities and Differences Between Americans and
Taiwanese, 31 Journal of Cross-Cultural Psychology 557–82 (2000).
566 problem solving, decision making, and professional judgment

who score lower.89 This suggests that liking based on physical attractiveness is
mediated by the salience of motives toward self-enhancement.
These caveats aside, existing experimental evidence—in both the laboratory
and the field—suggests that physical attractiveness is a significant determinant
of liking, which in turn has a powerful effect on judgment and choice.

17.5 reciprocity

Our social inclination to reciprocate makes us prone to accede to requests from

people who have done favors for us—even when the favors were uninvited and
trivial and the subsequent requests are substantial. To inoculate themselves
against this tendency, some lawyers, for example, will refuse even the slightest
favor from an opposing counsel, even if it is as trivial as the opponent’s paying
for their cup of coffee at the downstairs Starbucks. We also tend to reciprocate
concessions. If we have refused an initial request, we are more likely than we
would be otherwise to accede to a subsequent, smaller request by the same
person. So, beware if an opponent asks you for a patently unreasonable exten-
sion of time, followed up by a more measured request.
Our tendency to reciprocate is a well-studied phenomenon. In the early 1970s,
for example, Dennis T. Regan set out to determine which of two factors—liking,
or prior receipt of a favor—would most powerfully influence people’s compli-
ance with a request. Prior receipt of a favor won hands down. Subjects were far
more willing to buy raffle tickets from a confederate they had heard behave badly
on the phone, but who had given them a free Coke, than subjects who had heard
the confederate behave kindly on the phone but from whom they had not received
a free Coke. Across all conditions, receipt of the favor had a significant effect on
compliance. The effect of liking, on the other hand, proved quite weak.90
Elaborating on the Regan study, Mark Whately and his associates sought to
determine whether the strength of the reciprocity norm would be affected by a
person’s belief that his compliance with a request would become publicly known.
Whateley placed participants, together with a confederate, in a situation where
they thought they would be evaluating art for a study on aesthetic judgment.
During a break in the evaluations, the confederate left and returned either with
a small bag of candy (the “favor”) for the subject or empty-handed. When the
evaluations were over, all participants received a written request from the con-
federate to pledge a donation to a charity.

89. Mark Snyder, Ellen Berscheid, and Peter Glick, Focusing on the Exterior and the
Interior: Two Investigations of the Initiation of Personal Relationships, 48 Journal of
Personality and Social Psychology 1427–39 (1985).
90. Dennis T. Regan, Effects of a Favor and Liking on Compliance, 7 Journal of
Experimental Social Psychology 627–39 (1971).
 social influence 567

In one condition, subjects were led to believe that the confederate would know
whether they made a pledge (the public compliance condition). In another (the
private compliance condition), subjects were led to believe that the confederate
would not. The presence of the favor increased compliance in both public and
private conditions. That is, people receiving an unsolicited favor were more likely
to comply with subsequent requests from the person doing the initial favor,
regardless of whether the compliance was public or private. But subjects in the
public compliance condition donated significantly more money than did those
in the private compliance condition.91
We reciprocate concessions as well as favors, as demonstrated by research on
what Robert Cialdini calls the door-in-the-face technique. The basic insight here is
that a person is more likely to induce compliance with a request for a small favor
if she asks for a more extreme favor first. The technique works as follows: the
persuader presents the target with a significant request, something she knows
the target will turn down (the door in the face). Then, after the target refuses, the
persuader makes the smaller request, the one she was actually interested in all
along. Viewing the second request as a concession on the persuader’s part, the
target feels compelled to accede.
In the classic study of this door-in-the-face effect, Cialdini and his associates
asked college students who were walking across campus if they would be willing
to accompany a group of juvenile offenders on a day trip to the zoo.92 Only sev-
enteen percent said “yes.” In a second condition, students were first presented
with a more extreme request: would they be willing to spend two hours a week
for two years as unpaid counselors to juvenile offenders? After refusing this
extreme request, subjects in this second condition were asked to take part in the
zoo trip. Fifty percent agreed so to do.
The door-in-the-face technique prompts action—not merely a verbal agree-
ment to act. For example, researchers investigated whether door-in-the-face vic-
tims who had agreed to work for two unpaid hours in a community mental-health
agency actually showed up to perform their duties as promised.93 The tactic of
starting with a larger request (to volunteer for two hours of work per week in the
agency for at least two years) produced more verbal agreement to the smaller
retreat request (76 percent) as compared to the tactic of asking for the smaller
request alone (29 percent). Even more remarkable was the rate of follow-through.

91. Mark A. Whately, J. Matthew Webster, Richard H. Smith, and Adele Rhodes, The
Effect of a Favor on Public and Private Compliance: How Internalized Is the Norm of
Reciprocity?, 21 Basic and Applied Social Psychology 251 (1999).
92. Robert B. Cialdini et al., Reciprocal Concessions Procedure for Inducing Compliance:
The Door-in-the-Face Technique, 31 Journal of Personality and Social Psychology 206
(1999).
93. Ibid.
568 problem solving, decision making, and professional judgment

Subjects in the door-in-the-face condition showed up at a rate of 85 percent, as

compared to 50 percent of subjects in the control condition.
A reader might wonder whether the door-in-the-face technique merely results
from a (cognitive) contrast effect. After all, a person might be more willing to
accede to a second, more modest request simply because it looks smaller in com-
parison with the earlier, more extreme request. But Cialdini’s studies suggest
that the phenomenon results from reciprocal concession giving. Merely expos-
ing subjects to an extreme request did not increase the likelihood of compliance.
Compliance with the second request must be construed as a concession on the
part of the requester.
The strategic offering of favors and its cousin, the door-in-the-face technique,
are deployed by lawyers and politicians all the time. Be wary, then, when you
are presented with a favor, no matter how trivial, or when you are presented
with serial offers or requests. By doing you a favor, the requester will have
triggered in you an inclination to reciprocate in kind, whether or not it is in your
or your client’s interest to do so. When someone asks you for a favor and you
turn it down, consider whether the requester could have reasonably foreseen
your refusal to comply. If so, and if your refusal is followed by a second, less
extreme request, you may want to give yourself some time to consider before
responding. The extra time should buffer the pressure to comply, and allow you
to think more rationally about whether it is really in your client’s interest to
accede.

17.6 conclusion

Humans are social animals—we care a great deal about what other people think
of us. There is nothing irrational in this. More often than not, our personal out-
comes depend on the regard in which others hold us, and on their willingness to
cooperate, promote our interests, and assist us in achieving our goals. It should
come as no surprise, then, that processes of social influence play a major role in
shaping behavior.
Social influence processes affect the work of lawyers and policy makers in
significant ways. For this reason, it is important to understand them. Lawyers
can affirmatively use influence mechanisms to enhance their persuasiveness
when advocating on behalf of their clients. On the defensive front, lawyers can
use their knowledge of social influence to guard against situations in which third
parties may subtly attempt to alter their attitudes, or explain to clients ways in
which others might attempt to influence them to act against their own best inter-
ests. The ability to recognize and react intelligently to various sources of social
influence makes the difference between being at their mercy and resisting
their appeal.
 social influence 569

Policy makers, too, must understand and be able to use the levers of social
influence. Policy initiatives, no matter how well intended or otherwise sound,
may founder if they fail to take such forces as social proof into account. Conversely,
the achievement of policy goals may be dramatically enhanced though the skilled
deployment of the tendency toward consistency and commitment, or liking, or
the triggering of norm cascades. Social influence is integral to persuasion, and
persuasion—both to effecting it and resisting it.
The various cognitive processes mediating message processing, persuasion,
and behavioral choice can be either systematic or heuristic in nature.94 Systematic
processing is a System II function—motivated, deliberate, conscious, and
demanding of cognitive resources. Heuristic processing is a System I phenom-
enon, which can function automatically, with little awareness, and requiring
only minimal cognitive effort. In some situations, a person might apply a quick
persuasion rule, such as “believe the authority,” or “if everyone else thinks it’s
right, it must be,” instead of carefully analyzing a message’s merit. Various fac-
tors, such as the listener’s level of distraction95 or the message’s relevance to the
listener96 play a role in determining which type of processing, systematic or heu-
ristic, occurs. Similar processes—either systematic or heuristic, mediate behav-
ioral choice.
As this discussion suggests, social influence factors function in much the
same way as other mental heuristics. In a message and choice-saturated world,
shortcuts are essential. And given the social worlds in which we live, cooperation
with and at times deference to others is both rational and effective. But, like all
heuristics, social influences can lead us into errors, so we should strive to under-
stand and seek to control their effects.

94. Shelly Chaiken, The Heuristic Model of Persuasion, in Social Influence: The
Ontario Symposium Vol. 5, 3–39 (Mark P. Zanna, James M. Olson, and C. P. Herman
eds., Hillsdale, N.J.: Erlbaum, 1987); John T. Cacioppo, Richard E. Petty, Chuan Feng Kao,
and Regina Rodriguez, Central and Peripheral Routes to Persuasion: An Individual Difference
Perspective, 51 Journal of Personality and Social Psychology 1032–43 (1986).
95. Richard. E. Petty, Gary L. Wells, and Timothy C. Brock, Distraction Can Enhance or
Reduce Yielding to Propaganda: Thought Disruption versus Effort Justification, 34 Journal of
Personality and Social Psychology 874–84 (1976).
96. Richard E. Petty, John T. Cacioppo, and Rachel Goldman, Personal Involvement as
a Determinant of Argument Based Persuasion, 41 Journal of Personality and Social
Psychology 847–55 (1981).
This page intentionally left blank
18. influencing behavior through cognition

This chapter asks how lawyers and policy makers can influence decision making
by citizens, consumers, legislators, judges, and other officials through approaches
that are mediated by people’s judgment and perception rather than social influ-
ence. We begin with argument, the paradigmatic form of persuasion, and then
turn to priming, and end by considering how policy makers can influence citi-
zens’ and consumers’ behavior by framing the context for their choices.

18.1 persuasion through argument

The chances are that readers of this book—whether students, lawyers, or policy
makers—spend much of their time making and listening to arguments based on
reason and evidence. But arguments can be more or less persuasive indepen-
dent of their merits. Paradoxically, to ensure that meritorious arguments prevail,
it often helps to use communication techniques that have nothing to do with the
merits.
The study of persuasive argument dates back at least to Aristotle’s Rhetoric.
Our focus here is on the psychology of persuasion, in which argument is some-
times modeled as a multistage process:1
1. Exposure to the message.
2. Attention to the message.
3. Comprehension of the message.
4. Acceptance of the message’s conclusion.
5. Memory of the message or its conclusion.
6. Action based on the message.
Much of the psychological research on rhetoric focuses on how communicators
can increase recipients’ comprehension, acceptance, and memory of the message.
For example, in their popular book, Made to Stick: Why Some Ideas Survive and

1. Carl I. Hovland, Arthur A. Lumsdaine, and Fred. D. Sheffield, Studies in

Social Psychology in World War II, Vol. 3: Experiments in Mass Communication
(Princeton, NJ: Princeton University Press, 1949); Carl I. Hovland, Irving L. Janis, and
Harold H. Kelly, Communication and Persuasion (New Haven, CT: Yale University
Press, 1953); Carl I. Hovland ed., Order of Presentation in Persuasion (New Haven,
CT: Yale University Press, 1957).
572 problem solving, decision making, and professional judgment

Others Die, Chip and Dan Heath provide a practical guide to making arguments
memorable and influential.2
Simplicity. Simplicity is not about dumbing down ideas or using sound bites;
rather it is about prioritizing. Because people have limited attention, it is often
best to focus on a single, core message. Adding secondary points and unneces-
sary detail can detract from the impact of a message. Heath and Heath suggest
that the ideal of simplicity is a message that gets at the core of the idea you are
trying to convey and includes nothing else. For example, Bill Clinton’s campaign
advisor, James Carville, developed the communication strategy captured by the
phrase: “It’s the economy, stupid.” This meant that Clinton’s campaign speeches
would focus almost exclusively on the economy. As Carville put it, “There has to
be message triage. If you say three things, you don’t say anything.”
Unexpectedness. Hearing things that are unexpected often leads to curiosity,
a state that increases our interest in and the attention we pay to a topic or ques-
tion.3 Heath and Heath argue that this often leads to extra thinking, which helps
make ideas stick in our memories. Describing your idea in terms that are coun-
terintuitive or surprising can be a good way to make a message stick.
Heath and Heath describe the example of Nordstrom, a department store
known for its excellent customer service. In order to maintain this reputation,
managers must impress on their employees how much they value good service.
Rather than just giving speeches about “the importance of customer service,”
managers give examples of past instances of outstanding service by Nordstrom
employees. Because they were surprising, examples such as ironing a shirt for a
customer who needed it for a meeting that afternoon and cheerfully gift-wrapping
clothing a customer bought at Macy’s, got the message across effectively.
Concreteness. There is a tendency, especially when we have a lot of knowledge
about a topic, to communicate ideas in abstract terms. But abstract ideas are dif-
ficult for nonexperts to understand and are easily misconstrued even by experts.
The Nordstrom example described above is also an example of the power of con-
creteness. If managers just told employees that they should “go above and beyond
what it expected,” that would leave a lot of room for interpretation and probably
wouldn’t leave a lasting impression. Providing concrete examples of the good
service makes clear exactly what is expected.
Credibility. The traditional way to establish credibility is to have credentials—for
example, advanced training or a degree. But people can sometimes establish cred-
ibility by alternative means. Heath and Heath refer to “testable credentials,” in
which listeners are asked to evaluate the credibility of a contention for themselves.
During a presidential debate with Jimmy Carter, Ronald Regan asked the rhetorical

2. Chip Heath and Dan Heath, Made to Stick: Why Some Ideas Survive and
Others Die (New York: Random House, 2007).
3. George Loewenstein, The Psychology of Curiosity: A Review and Reinterpretation, 116
Psychological Bulletin 75–98 (1994).
 influencing behavior through cognition 573

question: “Are you better off now than you were four years ago?” Rather than argue
directly that the economy had worsened during Carter’s tenure, he asked the audi-
ence to test the proposition for themselves.
Emotionality. Research on charitable giving has shown that people donate
more money in response to appeals that focus on a single, identifiable victim
than they do when appeals cite statistics reflecting mass-scale suffering.4
Researchers who study this phenomenon proffer the following explanation for
it: learning about a single victim evokes emotion, whereas learning about statis-
tics shifts people into a more analytical frame of mind that suppresses emotion.
Emotion is the motivational engine that drives people to respond to messages
involving appeals for charitable donations. Focusing on analytical themes under-
mines messages intended to promote charitable giving.5
In his book, The Political Brain, psychologist Drew Westen makes a similar
contention about political appeals. He argues that political messages based on
rational, emotion-free arguments fail to have much effect on voters. Because
people’s decisions are often guided primarily by emotion and rationalized after
the fact, attempts to persuade are more likely to resonate with voters if they focus
on deeply held values and principles that tend to evoke emotion.6
Stories. Stories help ideas stick. They give coherence to a message and,
because we find them entertaining and engrossing, they capture our attention
better than a simple list of facts. Also, unlike more direct persuasive messages,
stories tend not to put the audience in an evaluative frame of mind. Rather,
people tend to empathize with the story’s protagonist and to be less critical of the
underlying message.
A study by Nancy Pennington and Reid Hastie found that when participants
judged evidence in a hypothetical criminal trial, they were more likely to render
a verdict for the side that presented evidence in an order that made it easy to
construct a story of the events in question.7 Recall our discussion, in Section 8.3,
of the power of narrative. While arguments can establish formal or abstract
truths, stories are better at convincing people of the verisimilitude of a point
because stories make a point seem representative of reality. Recall that people use
the representativeness heuristic to make judgments about the likelihood that

4. Deborah A. Small, George Loewenstein, and Paul Slovic, Sympathy and Callousness:
Affect and Deliberations in Donation Decisions, 102 Organizational Behavior and
Human Decision Processes 143–53 (2007).
5. George Loewenstein and Deborah A. Small, The Scarecrow and the Tin Man: The
Vicissitudes of Human Sympathy and Caring, 11 Review of General Psychology 112
(2007).
6. Drew Westen, The Emotional Brain: The Role of Emotion in Deciding the
Fate of the Nation (Public Affairs, New York, 2007).
7. Nancy Pennington and Reid Hastie, Explanation-Based Decision Making: Effects of
Memory Structure on Judgment;14 Journal of Experimental Psychology: Learning,
Memory, & Cognition 521–33 (1988).
574 problem solving, decision making, and professional judgment

something is true. Persuasion often consists less of convincing others of the

logic of your argument than of having them replace their narrative construction
of reality with your own.8
The distinction between judgment and decision making under so-called
System I and System II, made throughout this book, applies no less to a recipi-
ent’s processing of communications. The processes can be either heuristic or
systematic in nature.9 Heuristic message processing is a System I function, while
systematic message processing is System II. The study of rhetoric since Aristotle
has shown how effective persuasion can work through both systems. Which of
Heath and Heath’s observations relate to System I processing, and which to
System II? Recall from Chapter 17 how social influence heuristics, such as
“believe an authority,” or “if everyone else thinks it’s right, it must be,” can sub-
stitute for analyzing a message on its merits.

18.2 framing arguments in terms of gains and losses

Recall the now familiar S-shaped curve of Prospect Theory of Figure 14.1, and
recall that losses from a reference point loom larger than gains. Because the
reference point is essentially subjective, even arbitrary, a strategic communicator
may try to get her audience to adopt a reference point that will influence attitudes
in the direction she wants. Consider these experiments.
George Quattrone and Amos Tversky conducted an experiment asked partici-
pants to imagine that a hypothetical country had dedicated $100 million to reduc-
ing crime by unemployed immigrant youths. They were asked to allocate the
$100 million between two groups of equal size, Alphans and Betans.
Half the participants were told that, by the age of twenty-five, 3.7 percent of
Alphans and 1.2 percent of Betans had criminal records, while the other half
were told that, by the age of twenty-five, 96.3 percent of Alphans and 98.8 percent
of Betans had no criminal record. The participants had to determine whether to
give a little bit more to the Alphans, or a lot more to the Alphans—the group with
the higher rate of criminality. Objectively, the two groups were given identical
information, but the information in the second condition was framed in terms
of a lack of criminality instead of criminality.

8. Jerome Bruner, Making Stories: Law, Literature, Life (New York: Farrar,
Straus, & Giroux, 2002).
9. Shelly Chaiken, The Heuristic Model of Persuasion, in Social Influence: The
Ontario Symposium Vol. 5, 3–39 (Mark P. Zanna, James M. Olson, and C. Peter Herman
eds., Hillsdale, NJ, 1987). Lawrence Erlbaum et al., Central And Peripheral Routes to
Persuasion: An Individual Difference Perspective, 51 Journal of Personality and Social
Psychology 1032–43 (1986).
 influencing behavior through cognition 575

In the “criminality” frame (3.7 percent Alphans versus 1.2 percent Betans)
participants allocated more of the money to the Alphans than in the noncriminality
framing (96.3 percent versus 98.8 percent). Quattrone and Tversky attributed
this to what they call the “ratio-difference principle.” In the criminality frame,
Alphans are perceived as three times more criminal as Betans; but they are
seen as only slightly less noncriminal than the Betans under the noncriminality
frame. This is a result of diminishing sensitivity, shown in the S-shaped prospect
theory curve as one moves away from the reference point. According to the ratio-
difference principle, “the impact of any fixed positive difference between two
amounts increases with their ratio.”10 In other words, the difference between 3.7
percent and 1.2 percent seems bigger than a difference between 98.8 percent and
96.3 percent because the ratio of the first pair is much bigger than that of the
second pair. Our sensitivity to differences in objective numbers like crime rates
is an approximate function of the ratio associated with that difference.
In another experiment, Quattrone and Tversky found that whether an eco-
nomic indicator was described as the unemployment rate or the employment rate
had a significant impact on how concerned people seemed about a 5 percent
worsening of it. People weighed the difference between 5 percent and 10 percent
unemployment much more heavily than the difference between 95 percent and
90 percent employment. Indeed, descriptive statistics of the sort used in these
experiments are common in discussions about public policy, and this research
suggests that subtle differences in framing can affect how much people approve
(or disapprove) of a policy, irrespective of its actual impact.

18.3 linguistic form

Heath and Heath describe how messages can be shaped at a macro level to make
them more persuasive. Recent research by other scholars suggests that subtle
manipulations of language can affect the meaning that recipients take from a
message.

18.3.1 Agentive vs. Passive

Caitlin Fausey and Lera Boroditsky studied the effects of how an accident is
described on how blame for that accident is assigned.11 Specifically, they were
interested in the difference between reactions to agentive and nonagentive descrip-
tions of the same event. Agentive language specifically mentions an agent (e.g.,
Dave broke the vase), while nonagentive language does not (e.g., the vase broke).

10. George Quattrone and Amos Tversky, Contrasting Rational and Psychological
Analyses of Political Choice, 82 APSR 719–36 (1988).
11. Caitlin M. Fausey, Neil, Snider, Lera Boroditsky (in prep). Speaking of Accidents:
“He did it” invites more punishment than “It happened.”
576 problem solving, decision making, and professional judgment

Both are common and sound natural in English, and they are generally consid-
ered to be equivalent ways to say the same thing. Consider, for example, the fol-
lowing story:
Mrs. Smith and her friends were finishing a lovely dinner at their favorite
restaurant. After they settled the bill, they decided to head to a nearby café
for coffee and dessert. Mrs. Smith followed her friends and as she stood up,
she flopped her napkin on the centerpiece candle. She had ignited the napkin!
As Mrs. Smith reached to grab the napkin, she toppled the candle and ignited
the whole tablecloth, too! As she jumped back, she overturned the table
and ignited the carpet as well. Hearing her desperate cries, the restaurant
staff hurried over and heroically managed to put the fire out before anyone
got hurt.
Contrast this story:
Mrs. Smith and her friends were finishing a lovely dinner at their favorite
restaurant. After they settled the bill, they decided to head to a nearby café
for coffee and dessert. Mrs. Smith followed her friends and as she stood up,
her napkin flopped on the centerpiece candle. The napkin had ignited! As
Mrs. Smith reached to grab the napkin, the candle toppled and the whole table-
cloth ignited, too! As she jumped back, the table overturned and the carpet ignited
as well. Hearing her desperate cries, the restaurant staff hurried over and
heroically managed to put the fire out before anyone got hurt.
Fausey and Boroditsky found that people who read the former, agentive ver-
sion of the story assigned significantly more blame for the fire to Mrs. Smith.
When asked how much of the $1500 in total damages from the fire Mrs. Smith
should have to pay, participants who read the agentive version, on average, said
she should have to pay $935, while those who read the nonagentive version said,
on average, $689 (i.e., less than half the total amount).
The implications of this research for the courtroom are apparent. Outside the
courtroom, policy makers and other actors sometimes try to blunt criticism by
using the passive, “mistakes were made.” In an interview on CBS’s 60 Minutes,
Marine Staff Sgt. Frank Wuterich, who was charged with eighteen counts of
murder after the squad he was leading killed twenty-four civilians in apparent
response to the death of one of their fellow marines, said “I am completely sorry
that that happened.” Note that the research described above deals with cases in
which the passive voice is used by a third party. A perpetrator’s use of the
nonagentive form may trigger a cynical response both in courtrooms and in the
court of public opinion.

18.3.2 Verbs and Causal Inference

Along similar lines, Dutch psychologists Gun Semin and Christianne De Poot
have found that the specific type of verb you use to ask a question can have an
 influencing behavior through cognition 577

important effect on the answers you receive.12 They distinguish between “verbs
of action” (such as to help, to push, to cheat, etc.) and “verbs of state” (such as to
respect, to dislike, to love). Semin and De Poot’s research shows that questions
about action verbs tend to yield answers that focus on the sentence subject, while
the answers to questions about state verbs tend to focus on the sentence object.
So, if I ask you “Why do you think John helped David,” you will tend to answer
by telling me something about John that explains his helping behavior. On other
hand, if I ask you “Why do you think John likes David,” your answers will likely
refer to something about David that makes him likeable.
At first this might not seem to be the kind of thing that is subject to easy
manipulation: action verbs and state verbs tend not to be interchangeable. It is
hard to imagine a defense lawyer asking her client to explain to the jury why he
“disliked” the person he assaulted instead of why he “attacked” the victim. As
Semin and De Poot point out, though, once we know the effect of these verbs,
other elements of a sentence are susceptible to customization. For example, we
sometimes have a choice about whom to use as the subject and whom to use as
the object of a sentence. Imagine, for example, a rape case in which it is undis-
puted that the victim and the defendant danced together before the alleged crime.
Knowing that to dance is an action verb and that answers to action verbs tend to
focus causal explanations on the object of a sentence, a defense attorney might
prefer to ask the victim whether she danced with the defendant rather than
whether the defendant danced with her.

18.3.3 The Power of Nouns

We sometimes have the option not to use a verb at all to describe behavior.
Gregory Walton and Mahzarin Banaji have examined the different effects of
describing behavior using a verb or a noun.13 For example, a person who “plays
the piano” can also be referred to as a “pianist,” and someone who “runs a lot”
might be called a “runner.” In one experiment, Walton and Banaji provided
behavioral descriptions of fictional people using either the verb form (e.g., Jennifer
eats a lot of chocolate) or the noun form (Jennifer is a chocolate-eater). They found
that people tended to think the behavior reflected stronger, more stable prefer-
ences when the noun form was used than when the verb form was used.

12. Gün R. Semin and Christianne J. De Poot, The Question-Answer Paradigm: You
Might Regret Not Noticing How a Question Is Worded, 54 Journal of Personality and
Social Psychology 558–68 (1997).
13. Gregory M. Walton and Mahzarin R. Banaji, Being What You Say: The Effect of
Essentialist Linguistic Labels on Preferences, 22 Social Cognition 193–213 (2004). This
paper was inspired by the following paper examining a similar difference in children:
Susan A. Gelman and Gail D. Heyman, Carrot-Eaters and Creature-Believers: The Effects of
Lexicalization on Children’s Inferences About Social Categories, 10 Psychological Science
489–93 (1999).
578 problem solving, decision making, and professional judgment

Banaji and Walton then wondered if the same thing would happen when
people were being asked to think about their own behavior. Under the guise that
they were conducting a handwriting study, they had participants write out a
behavioral tendency or preference of their own, using either the noun of the verb
form. If a person said he liked Coke, he was either asked to write “I am a Coke-
drinker” or “I drink a lot of Coke” three times in his “natural handwriting.” They
found similar effects. People who had written about their own behavioral prefer-
ences using the noun form tended to say that those preferences were stronger
than if they had used the verb form.
In subsequent research, Christopher Bryan, Gregory Walton, and Carol Dweck
investigated whether a similar manipulation might affect not only people’s atti-
tudes about products but important behavioral decisions such as whether or not
to vote in a presidential election.14 The day before the 2008 presidential election,
they asked people who were eligible to vote but not yet registered to complete a
questionnaire about their “attitudes about voting and the electoral process.” In
that questionnaire, people were either asked questions like “How important is it
to you to vote in the upcoming election” or “How important is it to you to be a
voter in the upcoming election.” After the election, the researchers used official
records to determine whether participants had turned up at the polls.
It turned out that participants who had completed the questionnaire about
their attitudes about “being a voter” were more likely to have voted than partici-
pants who had answered questions about “voting.” The fact that such a seem-
ingly minuscule manipulation can have an effect on the decision to vote in a
presidential election—behavior that one would expect to be determined by
motives such as a sense of civic duty or interest in the outcome—suggests that
this subtle difference may have significant implications.
Where do these nouns get their power? The answer is not yet completely
clear, but researchers agree that the use of a noun to refer to a preference or
behavior invokes the idea of essentialism. This is the folk notion that a person
(or object) has some fundamental essence that makes her what she is and deter-
mines her attitudes, thoughts, and behaviors.15 It may be that, when a character-
istic is described in the noun form, the implication is that this is a central aspect
of that person’s self. It is conceivable that this affects even our perceptions of our
own characteristics and behavior.

14. Christopher J. Bryan, Mark R. Lepper, and Carol S. Dweck, “Voting versus Being
a Voter: The Effect of a Subtle Linguistic Manipulation on Voting Behavior” (unpublished
data, Stanford University 2005).
15. Dale Miller and Deborah Prentice, Some Consequences of a Belief in Group Essence:
The Category Divide Hypothesis, in Deborah Prentice and Dale Miller eds., Cultural
Divides: Understanding And Overcoming Group Conflict 213–238 (New York:
Russell Sage Foundation, 1999).
 influencing behavior through cognition 579

18.3.4 The Power of Metaphors

Most people think of a metaphor as a rhetorical device by which we explain one
concept, often an unfamiliar one, in terms of another, more familiar concept.
This has the effect of highlighting the properties of the former that it shares with
the latter. Linguist George Lakoff argues that our choice of metaphors can change
the way we think about a concept or issue.16 According to Lakoff, we use meta-
phors to understand and experience many of the most important and familiar
concepts in our lives. He argues that many of these concepts, such as love, the
mind, or ideas, are too abstract and complex for us to understand directly. Instead,
we understand them through the lenses of other familiar concepts that are more
tangible and clearly defined.
Think, for example, about the concept of ideas. Lakoff argues that we talk
about ideas in terms of more concrete concepts. For example, the metaphor ideas
are commodities is evident in the following phrases: “He won’t buy that,” “it’s
important how you package your ideas,” and “your ideas don’t have a chance in
the intellectual marketplace.” Similarly, the metaphor ideas are food is evident in:
“all this paper has in it are raw facts, half-baked ideas, and warmed-over theories,”
“there are too many facts here for me to digest them all,” and “there’s a theory you
can really sink your teeth into.”
One reason that there are often numerous different metaphors available for
the same concept is that most important concepts are complex and multifaceted.
Each metaphor highlights some aspects of our experience and hides others. For
example, the metaphor, ideas are commodities, highlights the fact that we often
want to convince others of our ideas, using the similarity between that experi-
ence and the experience of buying or using a commodity. Meanwhile, the meta-
phor ideas are food highlights the fact that, once we learn about an idea, it becomes
a part of who we are and, in some sense, makes us stronger intellectually—a
phenomenon comparable to the absorption of nutrients when we consume food.
Importantly, our choice of metaphor affects how we think about ideas. For exam-
ple, using the ideas are commodities metaphor is likely to focus our attention on
how our ideas are perceived by others, while the ideas are food metaphor focuses
our thinking on the value of ideas to us.
The fact that metaphors structure how we understand and experience a concept
means that they also have the power to shape our actions. Think about the com-
monly used metaphor argument is war (exemplified in such phrases as “that posi-
tion is indefensible” and “he shot down my argument”). Lakoff asserts that this
metaphor structures the way we engage in argument. For example, if instead of
argument is war, we used an argument is exchange metaphor, we might feel less of a
need to defend our position and attack the positions of those who disagree with us.

16. George Lakoff and Mark Johnson, Metaphors We Live By (Chicago: University
of Chicago Press, 1980).
580 problem solving, decision making, and professional judgment

To recap, according to Lakoff:

• We often understand the abstract and complex concepts in our lives better
if they are presented in concrete terms.
• A number of different metaphors often are available to help us understand
a given concept.
• Each metaphor highlights particular aspects of that concept and hides
others.
• The metaphor we use to think about a concept affects how we experience
that concept and behave with respect to it.
While empirical evidence is still lacking, these points suggest that our choice
of a particular metaphor may influence people’s thinking and behavior; it could
mean the difference between a public that favors and opposes a policy. Take, for
example, the phrase tax relief. This phrase is often used by Republicans and
others seeking reductions in most taxes. The word relief invokes a metaphor of
taxes as an affliction.17 As Lakoff points out, we have a script about afflictions that
includes a notion of the reliever as a hero; those who oppose relief are seen as
villains. The use of the metaphor, taxes are an affliction, influences our thinking
about taxes and is likely to have an effect on the popularity of tax cuts. A big part
of the power of such framing lies in its subtlety; the phrase tax relief is a perfectly
reasonable way to describe reductions in taxes. It highlights an aspect of taxes
that almost everyone experiences: it is unpleasant to pay them. At the same time,
though, it hides another aspect of taxes that many people also experience: we
appreciate the programs they make possible.
Another example is the debate about the legality of abortion. Lakoff points out
that referring to an “embryo” or a “fetus” has very different implications from
referring to a “baby.” The terms “embryo” and “fetus” frame the discussion
through the medical domain, and the question becomes one about whether or not
women should be allowed to make unfettered medical decisions. The term “baby,”
on the other hand, places the debate in the moral domain, making the question
one about whether or not unborn babies should be protected by the law.
Discussing politics more broadly, Lakoff has suggested that the metaphor of
the community or nation as a “family” plays a central role in our thinking. He
argues that two different family models help to determine our political sensibili-
ties. Associated with conservative thinking is the authoritarian, rule-bound
family that rewards or punishes on the basis of perceived merit, while liberal
thinking derives from the more nurturing, forgiving, family that provides
resources as a function of need rather than merit.18 Because most people endorse

17. George Lakoff, Don’t Think of an Elephant! Know Your Values and Frame
the Debate (White River Junction, VT: Chelsea Green, 2004).
18. George Lakoff, Moral Politics: How Liberals and Conservatives Think
(Chicago: University of Chicago Press, 1996).
 influencing behavior through cognition 581

both models to some extent, it may be possible to influence people’s policy opin-
ions by invoking the relevant family model in political appeals.

18.4 from inclination to action: channel factors

Even when people understand and agree with communications that call for
action, they do not necessarily take the next step and act on them. People may
have an inclination to act, but if their intended behavior has no available outlet,
then they may be less likely to do so. For example, researchers found that the
best predictor of whether people would show up at a health clinic was not their
attitudes about health and medicine, but how far they were from the clinic.19
Channel factors are simply ways of structuring a situation to increase the likeli-
hood that the recipient of a communication will act on it.
Consider this example:20 Robert Leventhal and his colleagues were studying
the effects of fear on the likelihood that people would get a tetanus shot. They
found that when conveying the risk of tetanus, a fear-inducing appeal made
people change their attitudes in favor of getting the shot significantly more than
a neutral informational appeal. Nevertheless, even with this attitude, only 3 per-
cent of participants followed up and got inoculated against tetanus. Strikingly, the
simple act of giving participants a map and instructing them to select a particular
time when they would go to get inoculated raised this percentage to 28 percent.
By the same token, sales of war bonds improved during World War II when
the government arranged for the bonds to be sold at workplaces rather than at
banks and post offices.21 In sum, people may have every intention to act in a
certain way, but if they don’t have a “channel” for their behavior, their intention
may never be carried out.

18.5 priming

Priming involves the automatic influence of exposure to environmental cues on

people’s judgment, decision making, and behavior. Cues as subtle as the pres-
ence of particular objects may implicitly communicate meanings and norms,
thereby guiding one’s behavior in situations that are novel or ambiguous (and in

19. Berenice E. Van Dort and Rudolf H. Moos, Distance and the Utilization of a Student
Health Center, 24 Journal of American College Health Association 159–62 (1976).
20. Howard Leventhal, Robert Singer, and Susan Jones, Effects of Fear and Specificity of
Recommendation upon Attitudes and Behavior, 2 Journal of Personality and Social
Psychology 20–29 (1965).
21. Dorwin Cartwright, Some Principles of Mass Persuasion: Selected Findings of Research
on the Sale of U.S. War Bonds, 2 Human Relations 253 (1949).
582 problem solving, decision making, and professional judgment

which people are therefore more susceptible to suggestion about what the rele-
vant norms are). In Section 11.3, we described how the priming of particular
schemas can influence people’s perceptions of and behavior toward members of
stereotyped groups. Here we broaden the discussion to other forms of judgment
and decision making.
Recall the “Ultimatum Game,” described in Section 12.1.2, in which one par-
ticipant is asked to propose a division of a fixed sum of money that a second
participant can either accept or reject.22 Would priming participants with cues
that invoked a business environment lead them to behave in a more competitive,
cut-throat manner, consistent with a business stereotype? Researchers asked
some participants to play the ultimatum game seated in front of a large desk on
which a briefcase, an executive-style silver pen, and a black-leather portfolio had
been placed. For other participants, a backpack replaced the briefcase, a card-
board box replaced the executive portfolio, and the participants were given a
standard wooden pencil instead of an executive pen. It turned out that partici-
pants in the business-objects condition proposed significantly less generous
divisions of the money than participants in the second condition.
The result of the research on the ultimatum game is important because it
suggests that random and seemingly irrelevant features of our environments
can serve as primes and have a significant influence on our behavior. Perhaps
the most striking demonstration of the significance of implicit priming effects is
a study by Jonah Berger and his colleagues that suggests that the polling place in
which people vote (e.g., church, school, or firehouse) may influence how they
vote. Berger et al. analyzed data from Arizona’s 2000 general election and found
that people who cast their votes in schoolhouses were more likely than others to
vote to raise taxes to support education.23 This study suggests that subtle environ-
mental cues can influence decisions on consequential issues even in complex
real-world environments.
In Section 1.5.2 we introduced the concept of social scripts that define appro-
priate behavior for ourselves and others, including public officials. People often
have recourse to alternative scripts, and priming can affect which alternative
someone applies to a given situation. For example, Thomas Gilovich presented
participants with a hypothetical foreign conflict and asked whether the United

22. Aaron C. Kay, S. Christian Wheeler, John A. Bargh, and Lee Ross, Material Priming:
The Influence of Mundane Physical Objects on Situational Construal and Competitive
Behavioral Choice, 95 Organizational Behavior and Human Decision Processes
83–96 (2004).
23. Jonah Berger, Marc Meredith, and S. Christian Wheeler, “Can Where People Vote
Influence How They Vote? The Influence of Polling Location Type on Voting Behavior”
(2006) Stanford Graduate School of Business Research Paper, No. 1926. The researchers
controlled for the voters’ political views, demographics, and whether or not they lived near
schools.
 influencing behavior through cognition 583

States should intervene militarily.24 He found that superficial similarities between

the hypothetical conflict and either World War II or the Vietnam War—for
example, whether refugees were fleeing on trains or in small boats—affected
whether participants were likely to recommend military intervention. Those
primed with similarities to World War II applied the “Munich” script and favored
an early show of military strength, while those primed with similarities to
“Vietnam” tended to avoid a military commitment for fear that it could lead to a
quagmire.
While the Gilovich study involved priming specific historical scripts, other
research has shown that it is possible to prime broad ideological schemas that
influence people’s views on specific policy issues.25 One element of the conserva-
tive schema is the endorsement of notions of meritocracy in which success is
seen as the product of hard work, wise decision making, self-discipline, and
other aspects of personal merit. The liberal schema, by contrast, includes the
notion that success may in large measure be a matter of good fortune, social
advantage, the help one receives from others, and other factors independent of
personal merit.26
Students at Stanford University were asked to write a brief essay describing
how they got into Stanford; they were instructed either to focus on the role of
“hard work, self-discipline, and wise decisions” or of “chance, opportunity, and
help from others” in getting them where they were. Participants were then asked
to indicate their views on a number of controversial policy proposals such as the
creation of a universal guaranteed health care system and the implementation of
a flat income tax. Those who had written the essay focusing on hard work and
personal merit expressed more conservative views than those who had focused
on good fortune and help from others. Most people hold aspects of both liberal
and conservative schemas, and it seems that priming can affect which one is
active at a given time, with potentially important effects on their judgments.
One common way to prime participants in the laboratory is to have them
solve word puzzles, where the words may either be neutral or designed to influ-
ence the participants’ behavior. For example, before being asked to play a
resource management game in which they fished from a lake with limited stock,

24. Thomas Gilovich, Seeing the Past in the Present: The Effects of Associations to Familiar
Events on Judgments and Decisions, 40 Journal of Personality and Social Psychology
797–808 (1981).
25. Christopher J. Bryan, Carol S. Dweck, Lee Ross, Aaron C. Kay, and Natalia
Mislavsky, “Ideological Mindsets: A Demonstration of Political Malleability” (2006) (man-
uscript submitted for publication).
26. James R. Kluegel and Eliot R. Smith, Beliefs About Inequality: Americans’
Views of What is and What Ought to Be (New York: Aldine deGruyter, 1986); Irwin
Katz and R. Glen Hass, Racial Ambivalence and American Value Conflict: Correlation and
Priming Studies of Dual Cognitive Structures, 55 J. Pers. Soc. Psychol. 893–905 (1988);
Robert E. Lane, The Fear of Equality, 53 Am. Polit. Sci. Rev. 35–51 (1959).
584 problem solving, decision making, and professional judgment

participants worked on an unrelated puzzle that had either cooperation-related

words (dependable, helpful, support, reasonable, honest, cooperative, fair,
friendly, tolerant, and share) or neutral words (salad, umbrella, city, gasoline,
wet, purposeful, switch, lead, mountain, and zebra).27 Members of the first group
were more likely to act cooperatively, taking fewer fish, than the second—though
not as much so as the members of a third group, who were explicitly asked to
set the goal: “I will cooperate as much as possible.” Similar experiments indicate
that priming may improve people’s achievement, their persistence in the face
of obstacles, and the likelihood that they will resume a task that has been
interrupted.28
Because its effects decay quickly, priming seems most applicable to irrevers-
ible one-time decisions where the agent can manipulate the context immediately
before the decision is made. For better or (likely) worse, voting seems a prime
candidate.
How might a policy maker prime drivers to exercise caution in dangerous
places on highways?

18.6 influencing and guiding citizens’ decision making

The question we just asked implies that policy makers can use their knowledge
about the techniques of social and cognitive influence to help citizens act in their
own best interests or in the interests of society. Of course, policy makers’ use of
such techniques raises core normative questions about the obligations of liberal
democratic governments to their citizens, who have heterogeneous values and
goals and diverse views about how to achieve them. But although liberal democ-
racy implies a presumption favoring autonomy, most political theorists hold that
citizens’ autonomy may sometimes be limited for the benefit of the greater soci-
ety and many hold that their choices may be (paternalistically) limited for the
benefit of individuals themselves.

18.6.1 Counseling Citizens about Risks: When Better Information Doesn’t

Necessarily Lead to Better Choices
Respect for individuals’ autonomy implies a preference for governments provid-
ing citizens and consumers with information about choices rather than restrict-
ing their choices. And it is a tenet of rational decision making that accurate
information about the consequences of behavior, including the probability of
good or bad outcomes, will lead to better decisions.

27. John A. Bargh, Peter M. Gollwitzer, Annette Y. Lee-Chai, Kimberly. Barndollar,

and Roman Troetschel, The Automated Will: Nonconscious Activation and Pursuit of
Behavioral Goals, 81 Journal of Personality and Social Psychology 1014–27 (2001).
28. Id.
 influencing behavior through cognition 585

However, we have seen people often have considerable difficulty in under-

standing quantitative information about risks. Kip Viscusi and Wesley Magat
conducted a study that showed that consumers respond to the format of warning
labels on household items (bleach, drain openers) in ways that diverge from
expected utility.29 In the absence of vivid information about a low probability
risk, consumers may treat the risk as zero, but “when a risk information pro-
gram forces them to focus on the probability that some rare injury will occur,
they may treat the probability as if it were higher than its objective value because
of the cognitive difficulties in making risky choices when the probabilities are
extremely low.”30
If people’s judgments of risk generally move in the direction of univocal new
information about risk, what happens when they receive conflicting informa-
tion, with one source asserting that the risk is high and another that it is low? In
a different experiment, Viscusi asked participants to rate the risk of cancer from
airborne pollutants where (1) two government studies disagreed whether the
risk was high or low; (2) two industry studies disagreed with each other; and (3)
government and industry studies were in disagreement. Except in the govern-
ment versus government disagreement, participants treated the high risk infor-
mation as being more informative—whether the government study showed a
higher risk than the industry study or vice versa.31
Viscusi notes that “symmetry is not violated because a particular information
source’s credibility is more consequential. Rather, it is the divergence of judg-
ments from different sources that largely accounts for the differing information
weights . . . [People] devote excessive attention to the worst case scenarios.”
Prescriptively, he suggests that consensus risk estimates are likely to be pro-
cessed more accurately than multiple risk assessments. “There may be consider-
able advantage to focusing the risk communication effort on the mean risk not
the risk assessment range. Particular care should be taken with respect to the
worst case scenarios. In practice, the distortions in risk beliefs due to the worst
case scenarios may be even greater than found here since these experiments
presented low risk and high risk estimates symmetrically. In contrast, the media
and advocacy groups often highlight the worst case scenarios, which will tend to
intensify the kinds of biases observed here.”
Cass Sunstein cautions that simply by disclosing low probability risks—e.g.,
“this food is genetically modified” or “this water contains 5 parts per billion of
arsenic” (a very small amount)—labeling is likely to “greatly alarm people, caus-
ing various kinds of harms, without giving them any useful information at all.

29. Kip Viscusi and Welsey A. Magat, Learning About Risk: Consumer and Worker
Responses to Hazard Information (Cambridge: Harvard University Press, 1987).
30. Id. at 128.
31. W. Kip Viscusi, Alarmist Decisions with Divergent Risk Information, 107 The
Economic Journal 1657 (1997).
586 problem solving, decision making, and professional judgment

If people neglect probability, they may fix, or fixate, on the bad outcome, in a way
that will cause anxiety and distress but without altering behavior or even improv-
ing understanding. Sunstein suggests that it might be possible to tell people not
only about the risk but also about the meaning of the probability information—
for example, by comparing a risk to others encountered in ordinary life.” But, he
asks, “if the risk is low, and of the sort that usually do not trouble sensible human
beings, is it really important to force disclosure of facts that will predictably cause
high levels of alarm?”32
The problem is that the very discussion of a low-probability risk leads people
to imagine or visualize the worst case—even if the discussion consists mostly of
trustworthy assurances that the likelihood of harm really is infinitesimal.
Emphasizing the low probability that the risk will eventuate is a lame strategy
against affect-laded imagery. Of course, there’s another potential problem as
well: “People may attempt to deal with their fear by refusing to think about the
risk at all.”33

18.6.2 When Incentives Are Counterproductive

In the introduction to Part 4, we mentioned the idea of paying students to main-
tain an appropriate weight or improve their grades. Skeptics of this approach
might be concerned that external incentives will undermine people’s long-term
motivations in both spheres. Indeed, traditional economic incentives sometimes
backfire and have the opposite of the effect they are designed to have. This usu-
ally happens because people are already motivated by some nonmaterial incen-
tive (like guilt or a sense of civic duty) to engage in the behavior the incentives
are designed to encourage. Introducing an economic incentive can change the
way people think about that behavior, resulting in what psychologists refer to as
the overjustification effect.
James Heyman and Dan Ariely conducted a series of studies suggesting that
when you want someone to perform a minor task, you might be wiser to ask the
person for a “favor” than to offer to pay him or her.34 In one experiment, people
asked to perform a menial task put less effort into it when offered 50 cents than
when offered no payment at all. Heyman and Ariely explain that the introduction
of payment invokes people’s schema for monetary exchanges rather than the
schema for a “social market.” Interestingly, offering a small “gift” (like a piece of
candy) as compensation does not invoke a monetary exchange schema unless
you tell the recipient how much it is worth.

32. Cass R. Sunstein, Laws of Fear: Beyond the Precautionary Principle 61

(Cambridge: Cambridge University Press, 2005).
33. Andrew Caplin, Fear as a Policy Instrument, in Time and Decision: Economic
and Psychological Perspectives on Intertemporal Choice (George Loewenstein,
Daniel Read, and Roy Baumeister eds., New York: Russell Sage, 2003) 441, 443.
34. James Heyman and Dan Ariely, Effort for Payment: A Tale of Two Markets, 15
Psychological Science 787–93 (2004).
 influencing behavior through cognition 587

Uri Gneezy and Aldo Rustichini examined the effect of introducing a small
fine to discourage a nuisance behavior.35 They conducted a study in ten day-care
centers in Haifa, Israel. The day-care centers all operate from seven-thirty in the
morning until four in the afternoon, but teachers regularly have to stay past four
because some parents were late picking up their children. The researchers
arranged for some of the day-care centers to introduce a small fine (equivalent to
approximately one eighth the amount of a parking ticket) each time a parent was
more than ten minutes late. Surprisingly, the effect of the fine was to increase the
number of late-arriving parents in those day-care centers. After several weeks of
fines and increased rates of lateness, those centers that had introduced fines
eliminated them, but to no avail. The rate of lateness in those centers remained
at the same high level they were at under the fine system.
The researchers hypothesize that before the implementation of the fines, par-
ents had been operating with the understanding that teachers who stayed late to
watch over their children were doing so out of generosity, and therefore that the
parents should avoid imposing on them. The introduction of a fine, however,
was interpreted as a price for the service of staying late. Once this occurred, par-
ents thought of the service as a commodity that they could buy as often as was
convenient. The researchers further theorize that rates of lateness stayed high
when the fines were removed because the service continued to be seen as a com-
modity—only now it was a commodity with a price of zero.
Other research shows that even large incentives can sometimes be counter-
productive.36 Researchers went door-to-door in Switzerland asking residents if
they would be willing to accept a nuclear waste dump in their town. Although
residents were obviously apprehensive about the idea, approximately half of the
people polled said they would accept the dump. They presumably did this out of
a sense of civic responsibility; the dumps have to go somewhere after all. Other
residents were asked the same question but told that if they accepted the dump
in their neighborhood, they would be compensated with the equivalent of
approximately six weeks’ pay each year. The rate of acceptance dropped to half
the rate when no compensation was offered. Residents may have been willing to
accept a certain level of risk out of a sense of civic responsibility, but being paid
to risk their health was a losing proposition.
The common theme in these examples is that the introduction of economic
incentives into transactions that are governed by social norms changes the way
those transactions are understood by the people involved. In short, relying on
economic incentives to influence behavior that is governed by nonmaterial
motives sometimes can backfire.

35. Uri Gneezy and Aldo Rustichini, A Fine Is a Price, 24 Journal of Legal Studies
1–17 (2000).
36. Bruno S. Frey and Felix Oberholzer-Gee, The Cost of Price Incentives: An Empirical
Analysis of Motivation Crowding Out, 87 American Economic Review 746–55 (1997).
588 problem solving, decision making, and professional judgment

18.6.3 Framing Citizens’ Choices and Other Aspects of Choice Architecture37

Albert Einstein once said, “We can’t solve problems by using the same kind of
thinking we used when we created them.” However, it is possible that the uncon-
scious mental system can, in fact, do just that. In recent years, researchers have
proposed a new general strategy for improving biased decision making that
leverages our automatic cognitive processes and turns them to our advantage.
Rather than trying to change a decision maker’s thinking from System 1 to
System 2 in situations where System 1 processing is known to frequently result
in biased decisions, this strategy tries to change the environment so that System
1 thinking will lead to good results.”38
Humans are prone to many errors that can be reduced by mechanical design.
Just consider the devices in a modern car that turn on the headlights at dusk,
turn them off when you leave the car, warn you if a seatbelt is not fastened, and
keep you from losing the gas tank cap. Consider ATMs that require you to remove
the card before getting cash, or the packaging of birth control pills in twenty-
eight numbered compartments. Policy makers can be more or less intrusive in
efforts to prevent such errors. For example, there was a brief time in which you
couldn’t start a car unless the seatbelts were buckled; now, the car merely emits
loud beeps—an intentionally annoying way of providing information.
This book has focused on biases in perception and decision making rather
than on mechanical errors—though issues of attention are pervasive in both. In
Nudge: Improving Decisions About Health, Wealth, and Happiness, Richard Thaler
and Cass Sunstein describe a number of methods for counteracting biases and
other limitations of bounded rationality in individuals’ decision making. Their
overall approach, which they term choice architecture, rests on the two fundamen-
tal premises that individuals’ behavior and choices are subject to predictable
errors and that it is legitimate for policy makers (choice architects) to structure
choices and communicate information to citizens and consumers to help them
avoid these errors.
Thaler and Sunstein style the political philosophy underlying choice architec-
ture as libertarian paternalism.39 Since there is no way not to frame a decision,
they argue, decision frames should be intentionally designed to encourage
enlightened, self-interested behavior—for example, default enrollment in a
retirement plan rather than an employee’s having to take active steps to enroll.

37. Much of this section is based on Richard Thaler and Cass Sunstein, Nudge:
Improving Decisions About Health, Wealth, and Happiness (New Haven: Yale
University Press, 2008).
38. Katherine L. Milkman et al., How Can Decision Making Be Improved?, 4 Perspectives
on Psychol. Sci. 379 (2009).
39. For an earlier version of this argument, see Richard H. Thaler and Cass R. Sunstein,
Libertarian Paternalism, 93 Am. Econ. Rev. 175 (2003); Christine Jolls and Cass Sunstein,
Debiasing Through Law (2005); Christine Jolls, Cass Sunstein, and Richard Thaler, A
Behavioral Approach to Law and Economics, 50 Stanford Law Review 1471 (1998).
 influencing behavior through cognition 589

If it seems ironic that some of these interventions themselves take advantage of

biases, such as the status quo bias, at least they “nudge” people toward enlight-
ened self-interested or other prosocial behavior. That’s the paternalistic aspect of
libertarian paternalism. The libertarian aspect is that citizens and consumers are
free to reject the nudge if they so desire—that’s the difference between a nudge
and a shove.
As with other techniques of influence examined in this and the preceding
chapter, the techniques of nudging are orthogonal to the ends for which they are
used. Thaler and Sunstein give an example of Carolyn, who can determine how
food is set out in the cafeterias of a large school system. Without restricting stu-
dents’ choices, the order in which foods are placed, whether they are at eye level,
and other display variables, can have a large effect on what the students eat.
Carolyn might arrange the food:
• to encourage healthy choices that make the students better off;
• to facilitate students’ choosing the items that they would probably choose
anyway;
• to maximize profits to the cafeteria;
• to maximize bribes from food suppliers to Carolyn; or
• randomly.
All but the last are instances of choice architecture, though the third and fourth
do not fit the rubric of libertarian paternalism.
We begin with an example (not from Nudge) of how framing behavior in
terms of gains or losses may affect health, and then turn to the major categories
of Thaler and Sunstein’s nudges.
18.6.3.a Framing Health Behaviors as Gains or Losses Research by Alex
Rothman and Peter Salovey shows the effect of framing appeals to engage in
healthy behavior in terms either of gains or losses. Which frame is more effective
depends on the type of behavior we seek to encourage. Rothman and Salovey
distinguish between three types of health behavior: (1) detection behaviors, such as
getting a mammogram or a colonoscopy, where the goal is to detect a health prob-
lem early so it can be addressed, (2) prevention behaviors, like wearing sunscreen
or using condoms, where the goal is to avoid heath problems in the first place,
and (3) recuperative behavior, or getting treatment for an existing condition.
They argue that detection behaviors are likely to be seen as risky, not because
they actually pose a risk to a person’s health, but because detection behaviors
expose us to the short-term risk of getting scary news about our health. Because
people tend to be risk-seeking in the domain of losses (Section 15.4), loss-framed
appeals, which emphasize the potential cost of failing to detect a problem early,
are the best way to get people to engage in detection behavior.40 Gain-framed

40. Sara M. Banks et al., The Effects of Message Framing on Mammography Utilization,
14 Health Psychology 178–84 (1995).
590 problem solving, decision making, and professional judgment

appeals that emphasize the potential health benefits of early detection are likely
to be less effective. Because people tend to be risk-averse in the domain of gains,
such frames are likely to make people want to avoid the risk of scary news associ-
ated with detection.
In contrast, prevention behaviors such as wearing sunscreen or using con-
doms are perceived as safe. In fact, not performing such behaviors is generally
seen as the risky option. Because people tend to be risk-seeking in the domain of
losses, loss-framed appeals, like warning people of the potential costs of unpro-
tected sun exposure, might actually make people more likely to take the risk of
not wearing sunscreen. Instead, it is preferable to use gain-framed appeals, like
emphasizing the benefits of wearing sunscreen, to encourage this type of non-
risky behavior.41
Similar to prevention behaviors, recuperative behaviors (i.e., obtaining treat-
ment for an existing condition) are usually seen as nonrisky. Therefore, as with
prevention behavior, Rothman and Salovey find that loss-framed appeals—such
as those that emphasize the potential costs of failing to obtain treatment—tend to
be counterproductive because they are more likely encourage risk-seeking atti-
tudes. Gain-framed appeals that focus on the health benefits of treatment are
preferable.
18.6.3.b Providing Feedback and “Mapping” Choices to Experienced
Utility Simply making people more aware of the relationship between an action
and its consequences can shape behavior. For example, the environmental costs
of people’s actions are rarely salient. But the display in the Toyota Prius, which
shows drivers their momentary gas mileage, may affect their acceleration and
speed. Thaler and Sunstein describe a device called the Ambient Orb, which
changes color when home energy use rises above a certain level; people who
began using this device decreased their energy usage by 40 percent.
The information provided consumers as a basis for evaluating and choosing
among options is often obscure, with different options often being described
with different metrics. The problem will be familiar to readers who have shopped
for cell phone providers or cable TV packages. Thaler and Sunstein suggest that
a good choice architecture makes information about relevant dimensions of each
option salient.
Consider the advantage of displays that show supermarket shoppers the cost
per unit of different brands of the same item, or the advantage of giving purchas-
ers of home mortgages a comparison of lenders’ fee structures and interest rates.
Thaler and Sunstein suggest that consumers would make better choices if credit
card lenders were required to report a summary of fees and interest rates on a

41. Alexander J. Rothman, Peter Salovey, Carol Antone, Kelli Keough, and Christina
D. Martin, The Influence of Message Framing on Intentions to Perform Health Behaviors, 29
Journal of Experimental Social Psychology 408–33 (1993).
 influencing behavior through cognition 591

standardized form. They generalize this approach to a regulatory scheme called

RECAP—Record, Evaluate, and Compare Alternative Prices.
There are other approaches to mapping as well. Federal regulations require
car dealers to display a car’s mileage per gallon. Richard Larrick and Jack Soll
have proposed the alternative of gallons consumed per (say) 100 miles, arguing
that miles per gallon (MPG) leads consumers to underestimate the gains in fuel
savings from even a slightly more efficient car. For example, most people rank
an improvement from 34 to 50 MPG as saving more gas over 10,000 miles than
an improvement from 18 to 28 MPG, even though the latter saves twice as much
gas.42 People tend not make this error when presented with fuel efficiency
expressed in gallons used per 100 miles.43
18.6.3.c Setting Default Options In Germany, only 12 percent of people are
registered to donate their organs upon their death, while in Austria, 99 percent of
people are registered. It is less likely that cultural differences account for this dif-
ference than defaults. Austrians are automatically registered as organ donors but
may “opt out” and become unregistered at any time. Germans, by contrast, are
unregistered by default, but may “opt in” and register as donors if they wish.
The power of the default option may result from a number of factors:
• the status quo bias and loss aversion;
• inertia, laziness, and procrastination (the “yeah, whatever” heuristic);
• people’s belief that a default represents policy makers’ or peers’ sound
judgment (social proof).
Sunstein and Thaler argue that policy makers should be intentional in deter-
mining which option to designate as the default. They might select the default
that represents their best guess of individual preferences—for example, enroll-
ing Medicare recipients in a prescription drug plan that maximizes savings on
the most commonly used drugs—or that maximizes the welfare of society—for
example, consent to organ donation.
In keeping with the libertarian part of libertarian paternalism, Thaler and
Sunstein require that defaults be easy to change based on an individual’s prefer-
ence. But the better designed the choice architecture, the less necessary this
becomes. In fact, if the default is based on sound objective analysis, citizens
should be cautious about dismissing the default.
18.6.3.d Simplifying and Structuring Complex Choices We have seen that
large decision sets can demotivate choice or lead people to use simplifying strat-
egies that lead to suboptimal choices. Yet policy makers’ restriction of choices
may infringe the liberty of people who know they want something not included

42. Going from 34 to 50 MPG saves 94 gallons; but going from 18 to 28 MPG saves 198
gallons.
43. 20 Science 1593–94 (June 2008). DOI: 10.1126/science.1154983.
592 problem solving, decision making, and professional judgment

in the choice set. Thaler and Sunstein suggest the middle ground of simplifying
and structuring choice sets.
Consider the challenge of sorting through all the options for Medicare Plan D
prescription drug plans. Each plan contains information about all the different
drugs it covers at what rate and with what copayments. The challenge of sorting
through all the plans to find the right coverage would confound even a well-
informed person. But a computer program could recommend a smaller set of
options based on the prescriptions a particular individual uses most.
Large choice sets can be made easier to navigate by grouping options. For
example, some investment funds, rather than provide customers with a list of
funds in alphabetical order, group them by risk level (low, moderate, high) and
portfolio composition (primarily stocks, primarily bonds). Without limiting the
number of options available, this approach can limit the number of options that
people actually consider.
People could also be given the option of delegating their decision to an expert,
for example in choosing among 401(k) retirement plans. Delegation to an expert
could even be made the default choice.
18.6.3.e Making Incentives Salient For incentives to be effective in shaping
people’s behavior, they must be noticeable and salient. But the effects of deci-
sions often are delayed or difficult to imagine. Consider a family choosing
whether to purchase a car or rely on alternative means of transportation. The
costs of owning and operating a car may far exceed the costs of the alternatives.
But the financial shock of purchasing a car occurs only once, while one feels the
pain of losing money every time one pays a bus fare, or rents a car, or watches
the meter on a taxi. Recall from our discussion of prospect theory that losses are
more upsetting when segregated rather than integrated. (See Section 14.4) By
the same token, anticipating myriad cab fares and rental car fees may feel more
salient than the single loss of the price of purchasing a car. Perhaps making the
installment payments and operating costs of the car equally salient would help
level the playing field.
Prospect theory also suggests that several small incentives are likely to
affect behavior more than would one large incentive. Receiving a year-long
$100-per-month subsidy after buying a hybrid car might increase hybrid
purchases more than providing a single payment of $1200—and would almost
surely lead to more of an increase than discounting the car purchase by $1200.
Hefty fines for speeding that are only occasionally enforced are probably
less effective at reducing speeding than moderate fines that are frequently
enforced.
18.6.3.f Criticisms of Libertarian Paternalism The effectiveness of choice
architecture is an empirical question that is highly dependent on the contexts in
which its techniques are applied. We expect that, over time, social scientists will
develop useful generalizations about what works, what doesn’t, and why. But as
policy makers continue to explore the techniques of choice architecture, it is
 influencing behavior through cognition 593

worth noting some recurring concerns about its underlying political premise,
libertarian paternalism.
• Libertarian paternalism is an insidious use of government power because, by
not regulating conduct directly, it does not mobilize the same sort of focused
opposition as hard regulatory policies. But if there is no “neutral” way to
structure a choice, policy makers might as well choose frames and defaults
that are most likely to achieve citizens’ interests or, where there is no appar-
ent conflict, the public interest. In any event, a democratic government
should be transparent to its citizens about its use of choice architecture.
• Libertarian paternalism encourages citizens’ passivity and manipulates
their choices rather than seeking to educate them to be better decision
makers. But the literature on debiasing is not optimistic about the possi-
bilities for countering biases, and the pervasiveness of bounded rationality
means that even very thoughtful individuals will make many decisions by
default.
• Choice architects are subject to their own biases and to corruption, partial-
ity, and narrow-mindedness. Choice architecture offers the same opportu-
nities for influence by special interests as any regulation. But the deliberative
decision making that underlies good policies provide safeguards against
biases. (Though, to refer back to the first criticism, choice architecture may
embody tacit assumptions that would be more readily exposed in the con-
test over non-libertarian regulations.44)
• Policy makers’ choice of defaults, or even about what information to pro-
vide individuals, may assume an unjustified homogeneity of interests.
Maybe there are individuals who would genuinely prefer to have more cash
today than to save for retirement. When Thaler and Sunstein argue that
governments should inform parents’ choices of schools, they focus on
information about educational achievement, but some parents may be
more concerned about social or religious values. But as long as it isn’t oner-
ous to override the defaults, why shouldn’t policy makers choose defaults
that they believe to be in individuals’ or society’s best interests? And if gov-
ernments do not provide all the information people need to make a deci-
sion based on their own interests, they have access other sources of
information.
• Although choice architects endeavor to protect citizens’ liberty in pursuing
their diverse values and goals, the fact that goals are often constructed in
the very process of decision making undercuts its implicit distinction
between people’s ends and their means for achieving them. Mark Kelman
argues that efforts to assist citizens in pursuing their own values tend to

44. For example, there is an asymmetry between the defaults involving organ donation,
with some individuals believing that the practice violates religious imperatives.
594 problem solving, decision making, and professional judgment

embody policy makers’ implicit theory of the “good society.”45 But there are
plenty of instances where people’s bounded rationality and biases pretty
clearly result in bad decisions in terms of interests that were not con-
structed on the spot.
There are no simple solutions to these problems of choice architecture. But
Thaler and Sunstein offer an alternative to defaults that may obviate some of
them and improve people’s decision-making processes: mandated choice. Suppose
that you can’t obtain a driver’s license without specifying whether or not you
wish to be an organ donor if you are in a fatal accident. Suppose that an employer
cannot hire you until you have specified what, if any, retirement savings plan you
want to enroll in. Mandated choice doesn’t work in all circumstances, and it
doesn’t obviate the inevitability of framing—consider Carolyn’s cafeteria. But,
when it is feasible, it nudges citizens to deliberate about important choices rather
than nudging the choice in a particular direction—perhaps the ideal sort of
libertarian paternalism in a deliberative democracy.

45. Mark G. Kelman, The Heuristics Debate: Its Nature and Its Implications
for Law and Policy (Oxford University Press, forthcoming 2010); Mark Kelman, Hedonic
Psychology and the Ambiguities of “Welfare,” 33 Phil. & Pub. Aff. 391, 395–97 (2005).
19. group decision making and the effects
of accountability on decision quality

Many if not most important decisions by policy makers and lawyers are made in
group settings. For example, as Luis Trujillo and Christine Lamm deal with the
problem at Terra Nueva, they will involve teams of colleagues and experts and
work with other groups of stakeholders. Sometimes they will just seek advice;
sometimes the group will have the authority to make decisions. This is charac-
teristic of decision making in the private and nonprofit sectors as well as govern-
ments, and in private as well as public institutions. In situations like these, social
dynamics within the decision-making group can exert a powerful effect on deci-
sion quality.
This chapter explores a variety of questions relating to the performance of
small decision-making groups. Do groups outperform their individual members
and, if so, when? What are the most common sources of dysfunction in decision-
making groups, and how can their effects be minimized? Are there ways in
which a group can be constituted, or its work structured, to set the stage for opti-
mal group decision making? We conclude by considering how a decision mak-
er’s accountability to others—for individual as well as group decisions—can
affect decisions for better or worse.

19.1 thinking about group decision making

In a classic 1975 article entitled Suppose We Took Groups Seriously, Harvard

Business School professor Harold J. Leavitt asserted that small groups, rather
than individuals, should be the “building blocks” for organizations simply
because groups perform better than individuals working alone.1 Leavitt sug-
gested that people perform better in groups for three reasons:
1. the group setting satisfies their need for social belonging;
2. groups are more creative than individuals; and
3. (most pertinent to our present inquiry) groups can pool the decision-relevant
information possessed by some, but not all, group members and because
members can correct each other’s errors.

1. Harold J. Leavitt, Suppose We Took Groups Seriously, in Man and Work in Society
67–77 (Eugene L. Cass and Frederick G. Zimmer eds., New York: Van Nostrand Reinhold,
1975).
596 problem solving, decision making, and professional judgment

In any event, Leavitt observed, small groups will form within organizations as
a result of natural patterns of human social interaction. Given the inevitability of
groups, it is best to manage them in ways that will maximize their effectiveness.
Leavitt’s faith in the power of groups spread rapidly and became the received
wisdom in popular and academic thinking about business and government
administration.
At about the same time, the social psychologist Irving L. Janis published a
book on the perils of group decision making. In an expanded 1982 edition enti-
tled Groupthink,2 Janis sought to explain government decision-making disasters,
such as the failure to anticipate the Japanese attack on Pearl Harbor, the deci-
sions to escalate the war in Vietnam, the Bay of Pigs invasion, and the Watergate
cover-up.
The basic idea of Janis’s “groupthink” hypothesis was that highly cohesive
decision-making groups have a strong tendency to suppress their members’
independent thinking and to converge too quickly on poor-quality decisions. The
drive to avoid intragroup conflict prevents rigorous evaluation of competing
alternatives. Divergent perspectives and theory-disconfirming information are
suppressed in favor of supportive opinions and theory-confirming views.
Buckling under the overt and covert social pressure, even those group members
who harbored serious private reservations about the group’s preferred choice do
not express them and acquiesce in the decision. Janis also described a few suc-
cessful policy decisions, including President John F. Kennedy’s response to the
Cuban Missile Crisis and the development of the Marshall Plan after World War
II, in which the decision-making process did not appear to be infected with these
dynamics.
Three decades of research tell a more nuanced and complex story than either
Leavitt’s rosy or Janis’s pessimistic account. Groups can add value to a decision-
making process, but they can also run it off the rails. Small groups often make
better decisions than their average member would make alone, but they seldom
outperform their best members.3 Small groups fall prey to virtually all of the
cognitive biases that plague individual decision makers and sometimes actually
amplify the effects of those biases. Moreover, deliberation, which commentators
like Leavitt credit with improving the effectiveness of group decision making,
can actually introduce additional sources of error into the process.
Most of this chapter will focus on decision making by groups whose mem-
bers interact with each other. As a prelude, however, it is useful to contrast the

2. Irving L. Janis, Victims of Groupthink: A Psychological Study of Foreign

Policy Decisions and Fiascos (Boston: Houghton Mifflin, 1972); Irving L. Janis,
Groupthink: Psychological Studies of Policy Decisions and Fiascos (Boston:
Houghton Mifflin, 1982).
3. Gayle W. Hill, Group versus Individual Performance: Are N + 1 Heads Better than
One?, 91 Psychological Bulletin 519–39 (1983).
 group decision making and the effects of accountability 597

aggregation of judgments about an issue made by individuals who do not

interact—in effect, a statistical or nominal group.

19.2 aggregation without group process: the wisdom

of crowds

In a phenomenon that James Surowiecki has termed the “wisdom of crowds,”

combining a sufficient number of diverse and independent judgments often
yields a better outcome than the judgment of any isolated person, no matter how
smart or well informed he or she might be.4 The result requires that four condi-
tions be met:
1. The “group” must reflect a diversity of skills, opinions, information, and
perspectives.
2. Each member’s judgment must be independent from the others. That is, an
individual’s knowledge and opinions cannot be influenced by the knowl-
edge and opinions of other group members.
3. The members must make and report their “sincere” judgments rather than
skewing them for strategic purposes.
4. The decision maker needs a mechanism for aggregating the individual
judgments to turn large numbers of private judgments into a single one.
An early example of a statistical group is a 1906 experiment by Sir Francis
Galton, a proponent of eugenics who—to put it mildly—had little regard for the
common person’s potential as a decision maker. Because of his interest in breed-
ing, Galton was attending the West England Fat Stock and Poultry Exhibition. Its
organizers had devised a game in which fairgoers were asked to guess the butch-
ered and dressed weight of a particularly fat ox. Eight hundred people—farmers,
butchers, and random fairgoers, tried their luck that day.
Galton did not expect them to perform particularly well. Indeed he wrote,
“. . . like those clerks and others who have no expert knowledge of horses, but who
bet on races, guided by newspapers, friends, and their own fancies . . . [t]he aver-
age competitor was probably as well fitted for making a just estimate of the dressed
weight of the ox, as an average voter is of judging the merits of most political
issues on which he votes.”5
At the end of the fair, Galton conducted some statistical tests on the 787 con-
test entries. He correctly anticipated that most participants’ guesses would be
way off the mark, with only a small minority coming close to the beast’s actual

4. James Surowiecki, The Wisdom of Crowds: Why the Many Are Smarter than
the Few and How Collective Wisdom Shapes Business, Economies, Societies, and
Nations (New York: Doubleday, 2004).
5. Id. xii.
598 problem solving, decision making, and professional judgment

dressed weight. However, the average of the estimates, 1,197 pounds, turned out
to be only one pound less than the weight of the ox when slaughtered and
dressed.
The power of statistical averaging illustrated by Galton’s experience derives
from a law known as the Condorcet Jury Theorem.6 Here’s how the theorem
works: Suppose that a number of people are independently answering a yes or no
question, and that the probability that each individual will answer the question
correctly is at least somewhat greater than 50 percent. Then the probability that a
majority of group members will answer the question correctly increases as the
size of the group increases. If the answers are not binary, but are spread over a
continuous range, Condorcet’s theorem predicts that the average answer will
more closely approximate the correct answer as the size of the group increases.
(If the probability that each individual member will decide correctly is less than 50
percent, the probability that a majority will answer correctly approaches zero.)
Over half a century ago, Friedrich Hayek observed that aggregated knowledge
is crucial for providing information to the market.7 Yet the mechanisms for sys-
tematically accumulating and organizing information from dispersed individu-
als (other than through revealed preferences exhibited by their participation in
the market) were largely unexplored until the late 1990s, when Caltech and
Hewlett-Packard Laboratories undertook an innovative experiment to imple-
ment an Information Aggregation Mechanism within the corporation.
Kay-Yut Chen, a scientist at Hewlett-Packard Laboratories (HP) in Palo Alto,
California, and Charles R. Plott, a professor of economics and political science at
Caltech, experimented with internal “securities markets” to improve product
sales forecasting.8 They selected a small number of employees from different
parts of HP’s business operations and different geographical regions. Some
employees had information about clients, others about pricing strategies, others
about competitors’ strategies and products. Participants were given an initial
portfolio of securities (created for the experiment) with each security associated
with one of ten different levels of HP sales, and were allowed to buy and sell
securities associated with these levels. To make money, participants had to accu-
rately predict the level of sales HP would experience three months in the future.
If participants accurately did so (by holding securities within that range), they

6. Marquis de Condorcet, Essai sur l’application de l’analysé à la probabilité de decisions

rendues à la pluralité des voix (I. McLean and F. Hewitt trans., Paris: L’imprimerie Royale,
1785); McLennan A, “Consequences of the Condorcet Jury Theorem for beneficial infor-
mation aggregation by rational agents,” American Political Science Review 92, (1998):
413–18.
7. Friedrich A. Hayek, Individualism and economic order (Chicago: University
of Chicago Press, 1948).
8. Kay-Yut Chen and Charles Plott, “Information Aggregation Mechanisms: Concept,
Design, and Implementation for a Sales Forecasting Problem,” 2002, http://www.hss.
caltech.edu/SSPapers/wp1131.pdf.
 group decision making and the effects of accountability 599

would receive a payoff, but they would receive nothing for securities held at any
other level.9
Chen’s and Plott’s futures market consistently yielded better marketing fore-
casts than did HP’s official predictions, arrived at through deliberation by the
company’s sales forecasting experts.10 Why? Because it provided an effective
mechanism for aggregating the specialized knowledge of an independent and
diverse group of people, and because the participants had no reason to strategi-
cally distort their predictions.
It is this efficient aggregation of decentralized, privately held information that
makes projects like the Iowa Electronic Markets (IEM) so interesting. IEM is a
small-scale futures market where interested individuals make contracts based on
their own predictions of economic and political events, such as elections. In
2008, for example, individuals could make contracts based on whether Barack
Obama or John McCain would win the U.S. Presidential election. The market is
operated by faculty at the University of Iowa’s Henry B. Tippie College of
Business; the results routinely outperform polls and surveys in predicting elec-
toral outcomes.11 A similar idea is the Hollywood Stock Exchange (HSX), which
allows people to place wagers on movie box office returns, Oscar results, and
other film-related events.12 HSX produces surprisingly accurate predictions,
demonstrating the power of markets to aggregate the information dispersed over
a large population of independent actors.13
***
Decision makers must often forecast future events in order to assess the
advantages and disadvantages of alternative courses of action. These examples
suggest that information markets can sometimes do a better job than experts.
But it is often not feasible to use statistical groups, and the responsibility falls on

9. To ensure that participants had no knowledge of each others’ identities or of the

aggregated results until the experiment was over, thus limting the effects of influence on
their trading decisions, participants drew only on their personal, privately held informa-
tion, general information available from HP, and the patterns of trade they observed other
participants making online (based on anonymous ID numbers). Access to information
was therefore limited to a web-based trading system.
10. Kay-Yut Chen and Charles Plott, “Information Aggregation Mechanisms: Concept,
Design, and Implementation for a Sales Forecasting Problem,” 2002, http://www.hss.
caltech.edu/SSPapers/wp1131.pdf.
11. Iowa Electronic Markets Web site, operated by the University of Iowa Henry B.
Tippie College of Business http://www.biz.uiowa.edu/iem/.
12. Hollywood Stock Exchange Web site, http://www.hsx.com/about/whatishsx.htm.
13. James Surowiecki, The Wisdom of Crowds: Why the Many are Smarter
than the Few and How Collective Wisdom Shapes Business, Economies, Societies,
and Nations 20 (New York: Doubleday, 2004); Cass. R. Sunstein, Group Judgments:
Deliberation, Statistical Means, and Information Markets, 80 New York University Law
Review 962–1049 (2005).
600 problem solving, decision making, and professional judgment

actual face-to-face groups. Also, as illustrated in the previous chapters of this

book, most decision making involves far more than estimating and forecast-
ing—situations to which application of the “wisdom of crowds” is at least
untested, if not problematic.
The remainder of this chapter considers the determinants of good decision
making in actual, interactive groups: diversity, independences, candor, and an
ability to aggregate information, among other things.

19.3 groupthink

Irving Janis defined groupthink as the tendency of highly cohesive decision-

making groups to converge too quickly on a bad decision.14 The core claim of the
groupthink hypothesis is that certain antecedent conditions can lead decision-
making groups to employ defective processes, which lead to poor policy out-
comes.15
As background for understanding the groupthink hypotheses, recall the model
of fully deliberative decision making introduced in Section 4.5. As Figure 19.1
illustrates, a fully deliberative process comprises both divergent and convergent
elements. Divergent thinking predominates early in the process, as one frames
and reframes the problem, identifies the full complement of decision-relevant
goals and objectives, and generates a range of alternative courses of action. Later
in the process, these various alternatives are systematically evaluated to deter-
mine what outcomes—both intended and unintended—they might produce.
This is largely convergent thinking, characterized by an analytic evidence-based
appraisal of the likely positive and negative consequences of each possible course
of action. The process looks something like the one shown in Figure 19.1.
By contrast, groupthink is characterized by the group’s reaching premature con-
vergence without adequate analysis—all in the interest of achieving concurrence
within the group. As a result, the problem is inadequately framed, objectives are
underspecified, and an initially preferred course of action is inadequately scruti-
nized, while potential competing alternatives are too rapidly rejected or are left

14. Irving L. Janis, Groupthink, 5 Psychology Today 43–46, 74–76 (1971); Irving L. Janis,
Victims of Groupthink: A Psychological Study of Foreign Policy Decisions and
Fiascos (Boston: Houghton Mifflin, 1972); Irving L. Janis, Groupthink: Psychological
Studies of Policy Decisions and Fiascos (Boston: Houghton Mifflin, 1982); Irving L.
Janis, Crucial Decisions: Leadership in Policymaking and Crisis Management (New
York: Free Press, 1989); Irving L. Janis and Leon Mann, Decision Making: A Psychological
Analysis of Conflict, Choice and Commitment (New York: Free Press, 1977).
15. Robert S. Baron, So Right It’s Wrong: Groupthink and the Ubiquitous Nature of Polarized
Group Decision Making, in Advances in Experimental Social Psychology 37, 219 (Mark
P. Zanna, ed., San Diego, CA: Elsevier Academic Press, 2005); Mark Schafer and Scott
Crichlow, Antecedents of Groupthink, 40 Journal of Conflict Resolution 415–35 (1996).
 group decision making and the effects of accountability 601

Divergent
Thinking Frame the decision to be made
Identify objectives
Identify alternative courses of action
Evaluate alternative course of action vs.
objectives
Make a decision
Convergent Implement and monitor the decision
Thinking

figure 19.1 divergent and convergent elements in deliberative

decision making.

unidentified altogether. Rather than resembling a diamond, decision making

under conditions of groupthink looks more like the process shown in Figure 19.2.

Frame the decision

Identify objectives
Too-Rapid
Identify alternative courses of action
Convergence
Evaluate alternative course of action
vs. objectives

Make a decision
Implement and monitor the decision

figure 19.2 groupthink’s convergent processes.

From his examination of selected policy fiascos and successes, Janis identi-
fied four sets of attributes characterizing groupthink.
1. groupthink symptoms;
2. decision-making characteristics;
3. antecedents; and
4. remedial interventions.
We review the first three here and will return to the last in Section 19.8.

19.3.1 Groupthink Symptoms and Decision-Making Characteristics

The core symptoms of groupthink consist of:
• overestimation of the group’s skill, power, and morality;
• close-mindedness; and
• pressure toward uniformity
In Janis’s view, a group that has fallen prey to the groupthink syndrome tends to
have an exaggerated view of its skills, knowledge, power, and moral righteousness.
602 problem solving, decision making, and professional judgment

Misgivings and dissenting viewpoints are quickly discounted, as are cautionary

admonitions from outsiders, if they are allowed to penetrate the group’s bound-
aries at all. The competence, strength, and reasonableness of the group’s real or
imagined adversaries are underestimated; those adversaries are reduced to ste-
reotypes, and often dismissed as “evil” or as “weak and stupid.”16
But the most significant symptoms of groupthink fall into the third category:
pressures toward uniformity. In groupthink situations, a favored course of action
emerges early in the decision-making process, often because it is, or is believed
to be, the option favored by the group’s leader. Group members who harbor
doubts about the emerging consensus feel inhibited from expressing them, cre-
ating an illusion of unanimity. Group members who do voice doubts are sub-
jected to subtle or overt forms of social pressure, disparagement, or reprisal,
often through the efforts of “mindguards”—group members who police the
group’s thinking.
Under these circumstances, divergent elements of the formal decision-
making process are given short shrift, leading to underidentification of goals and
objectives and an unduly limited set of options. Additionally, in generating or
evaluating alternative courses of action, the group conducts an inadequate infor-
mation search, fails to take advantage of expert knowledge, and looks for infor-
mation that confirms the preferred course of action. The group fails to conduct
a thorough comparison of the available alternatives, fails fully to assess the risks
attending the preferred course of action, and fails to develop contingency plans.
As a result, groupthink neutralizes what would otherwise be advantages
provided by group decision making. Diversity of perspectives is squandered
and members’ specialized expertise neutralized. The potential that a group has
for generating more ideas and for bringing more information to bear on the
decision-making process goes unrealized.

19.3.2 Groupthink Antecedents

Groupthink is primarily a disorder of highly cohesive groups. However, cohesive-
ness alone is not sufficient. Janis thought that factors such as the group’s insula-
tion, the presence of partisanship, directive group leadership, and ideological
homogeneity among group members all made groupthink more likely to emerge.
Janis proposed that common contextual antecedents included high levels
of external pressure and internal stress, which could be exacerbated by the
experience of recent failure or by the problem’s complexity and “high stakes.”
He posited that, when accompanied by other situational and contextual anteced-
ents, high group cohesion was a strong predictor of the groupthink syndrome.

16. Irving L. Janis, Groupthink: Psychological Studies of Policy Decisions

and Fiascos 197 (Boston: Houghton Mifflin, 1982).
 group decision making and the effects of accountability 603

19.3.3 Empirical Perspectives on the Groupthink Framework

Janis’s own analysis of the groupthink hypothesis was based largely on case stud-
ies of policy decisions selected because of their (mostly bad) outcomes. Even if
one could control for the myriad factors involved in these complex decisions, it
would be difficult to draw conclusions about correlation, let alone causation.
Thirty years of subsequent research has not provided strong empirical evidence of
the link between the posited antecedents to the groupthink phenomenon.17 Here
is a brief summary of what’s known about the groupthink hypothesis today.
• Reanalyses of Janis’s case studies have confirmed that structural and pro-
cedural organizational defects predict the groupthink syndrome, but have
not determined that group cohesiveness and provocative situational con-
texts predict the syndrome.18
• Analyses of other policy fiascos—including the Ford Motor Company’s
decision to produce the Edsel, price fixing in the electrical industry, the
Challenger space shuttle launch decision, and the Iran-Contra fiasco—
have found present many of the groupthink antecedents, symptoms, and
decision process characteristics.19
• Some researchers argue that the groupthink hypothesis is so complex and
incorporates so many variables that, unless it is broken up into its compo-
nent parts, it cannot possibly be validated empirically.20
• It has proved difficult to recreate the groupthink antecedents (genuine in-
group cohesion, meaningful threat, group insulation, homogeneity of

17. Robert S. Baron, supra.

18. Philip E. Tetlock et al., Assessing Political Group Dynamics: A Test of the Groupthink
Model, 63 Journal of Personality and Social Psychology 403–25 (1992). McCauley
(1989) also found no support for the predictive value of group cohesion, although he did
find significant predictive value for insulation, promotional leadership, and in-group
homogeneity. The studies that do exist focus primarily on the groupthink antecedents,
attempting to determine whether groupthink symptoms emerge if, and only if, particular
antecedent variables are present.
19. Richard C. Huseman and Russell W. Driver, Groupthink: Implications for Small
Group Decision Making in Business, in Readings in Organizational Behavior 100–10
(Richard C. Huseman and Anne Carroll eds., Boston, MA: Allyn and Bacon, 1979); Thomas
R. Hensley and Glen W. Griffin, Victims of Groupthink: The Kent State University Board of
Trustees and the 1977 Gymnasium Controversy, 31 Journal of Conflict Resolution
203–26 (1986); James Esser and Joanne Lindoerfer, Groupthink and the Space Shuttle
Challenger Accident: Toward a Quantitative Case Analysis, 2 Journal of Behavioral
Decision Making 167–77 (1989); Paul t’Hart, Irving L. Janis’ Victims of Groupthink, 12
Political Psychology 247–78 (1991); Gregory Moorhead et al., Group Decision Fiascoes
Continue: Space Shuttle Challenger and a Revised Groupthink Framework, 44 Human
Relations 539–50 (1991).
20. Robert S. Baron, supra.
604 problem solving, decision making, and professional judgment

values) in laboratory settings.21 Those laboratory studies that have been

done generally do not support Janis’ notion that group cohesiveness is the
most important antecedent predictor of groupthink.22
We turn our attention to some discrete pathologies of group decision
making.

19.4 the common knowledge effect

In Chapter 1, we introduced Jack Serrano, a client of attorney Luis Trujillo’s who

owns Terra Nueva, a low-income housing development. Recall that a group of
Serrano’s Terra Nueva tenants were about to file a class action lawsuit against
him, claiming that foam insulation used in building their apartments was
making them sick.
Imagine that the lawsuit is filed and that the lawyer representing the Terra
Nueva plaintiffs convenes a meeting to discuss the litigation with representatives
from some advocacy groups. Those present include an organizer from the United
Electrical Workers and activists from East Los Angeles Tenants United and L.A.
Consumers’ Action, along with representatives of the Terra Nueva plaintiffs and
the newly formed Terra Nueva Tenant’s Union. The main purpose of the meet-
ing is to allow representatives from the various groups to discuss how the litiga-
tion could be used to focus attention on the many toxic-exposure–related
problems facing low-income residents in Southern California. More specifically,
the participants want to explore whether similar suits should be filed against
other developers and low-income housing operators in the area.
All of the individuals present at the meeting are familiar with the allegations
made in the Terra Nueva lawsuit. They have all read the media coverage, includ-
ing the articles putting the Terra Nueva situation in the larger context of toxic
residential insulation cases being filed across the country. But only the organizer
from the United Electrical Workers is aware that exposure to toxic chemicals
in the semiconductor plants, in which some of the tenants work, could cause
similar symptoms.
Imagine also that shortly after the suit is filed, Luis Trujillo, two of his firm’s
associates, and a paralegal are meeting with an epidemiologist with knowledge

21. Id.; James Esser, Alive and Well After 25 years: A Review of Groupthink Research, 73
Organizational Behavior and Human Decision Processes 116–41 (1998).
22. Matie L. Flowers, A Laboratory Test of Some Implications of Janis’s Groupthink
Hypothesis, 35 Journal of Personality and Social Psychology 888–96 (1977); Eugene
M. Fodor and Timothy Smith, The Power Motive as an Influence on Group Decision Making,
42 Journal of Personality and Social Psychology 178–85 (1982); Michael R. Callaway
and James K. Esser, Groupthink: Effects of Cohesiveness and Problem-Solving Procedures on
Group Decision Making, 12 Social Behavior and Personality 157–64 (1984).
 group decision making and the effects of accountability 605

of “sick building syndrome” and a professor of toxicology from a local university.

The purpose of the meeting is to develop a litigation strategy for the defense.
Collectively, the six people present have a great deal of relevant knowledge—-
about the Terra Nueva residents, about the potential harmful health effects of
foam insulation and other building materials, about disease cluster analysis, and
about the litigation of toxic exposure cases. But one piece of information is
known only to the paralegal who happens to have family ties to some of the Terra
Nueva residents: that some of the residents work in a shipyard where they are
exposed to solvents and other hazardous materials that could cause the symp-
toms the tenants are complaining about.
Thus, one participant in each group has a unique piece of information that
points to a possible cause of the Terra Nueva tenants’ symptoms, other than the
foam insulation in their apartments. This information could make an enormous
difference to the strategies the group formulates.
Group decision making has the potential benefit of aggregating the knowl-
edge of diverse members.23 But realizing this benefit requires that group mem-
bers who possess unshared information—information not known by other
members—recall it and actually share it with the group, and that the group com-
bine the shared information to make its decision. Will the participant share the
unique information with the rest of the group?
Unfortunately, group deliberations often focus on shared rather than unshared
information.24 The so-called “common knowledge effect”25 biases group decision
making in the direction supported by shared information. Even when group
members with unshared information recall it and share it with the group, it
tends not to be optimally considered in subsequent group deliberation.26

23. Amiram Vinokur and Eugene Burnetein, The Effects of Partially Shared Persuasive
Arguments on Group Induced Shifts: A Group Problem Solving Approach, 29 Journal of
Personality and Social Psychology 305–15 (1974).
24. James R. Larson et al., Discussion of Shared and Unshared Information in Decision
Making Groups, 67 Journal of Personality and Social Psychology 446–61 (1994);
Garold Stasser and Dennis Stewart, Discovery of Hidden Profiles by Decision Making Groups:
Solving a Problem versus Making a Judgment, 63 Journal of Personality and Social
Psychology 426–34 (1992); Garold Stasser and William Titus, Pooling of Unshared
Information in Group Decision Making: Biased Information Sampling During Discussion, 48
Journal of Personality and Social Psychology 1467 (1985).
25. Daniel Gigone and Reid Hastie, The Common Knowledge Effect: Information Sharing
and Group Judgment, 65 Journal of Personality and Social Psychology 959–74 (1993).
The phenomenon is also called the “common information sampling bias.” Gwen M.
Wittenbaum and Garold Stasser, Management of Information in Small Groups, in What’s
Social About Social Cognition? Research on Socially Shared Cognition in Small
Groups 3–28 (Judith L. Nye and Aaron M. Brower eds., Thousand Oaks, CA: Sage, 1996).
26. Gigone and Hastie, supra.
606 problem solving, decision making, and professional judgment

19.4.1 Research on the Common Knowledge Effect: The “Hidden Profile”

Paradigm
In the mid-1980s, Garold Stasser and William Titus asked randomly selected
groups of people to make a decision about which of three candidates for a par-
ticular position was the most qualified.27 All relevant information about the three
candidates was provided to all members of the control group (the shared infor-
mation condition). In the aggregate, the information strongly supported
Candidate A, and the control group tended to choose that candidate by wide
margins.
In the experimental condition, the negative information about the “best qual-
ified candidate” (the one most frequently chosen by the control group) was
shared, while positive information about that candidate was divided among
group members. If all of the information possessed by all group members had
been aggregated, Candidate A’s superiority over Candidates B and C would be
readily apparent. However, in their deliberations, the experimental group mem-
bers were far more likely to bring up and discuss shared information. As a result,
they were unable to uncover the “hidden profile” that made Candidate A’s supe-
riority so apparent to the control group.28
Stasser and Titus’s results have been replicated in a variety of experimental
contexts29 to yield these robust generalizations: small group deliberation relies

27. Garold Stasser and William Titus, Pooling of Unshared Information in Group
Decision Making: Biased Information Sampling During Discussion, 48 Journal of
Personality and Social Psychology 1467 (1985); Garold Stasser and William Titus,
Hidden Profiles: A Brief History, 14 Psychological Inquiry 304 (2003).
28. See Garold Stasser and William Titus, Hidden Profiles: A Brief History, 14
Psychological Inquiry 304 (2003).
29. Gwen M. Wittenbaum and Jonathan M. Bowman, A Social Validation Explanation
for Mutual Enhancement, 40 Journal of Experimental Social Psychology 169–84
(2004); Gwen M. Wittenbaum, Anne P. Hubbell, and Connie Zuckerman, Mutual
Enhancement: Toward an Understanding of the Collective Preference for Shared Information,
77 Journal of Personality and Social Psychology 967–78 (1999); Janice R. Kelly and
Steven J. Karau, Group Decision Making: The Effects of Initial Preferences and Time Pressure,
25 Personality and Social Psychology Bulletin 1342–54 (1999); Garold Stasser,
Sandra Vaughn, and Dennis Stewart, Pooling Unshared Information: The Benefits of Knowing
How Access to Information Is Distributed Among Group Members, 82 Organizational
Behavior and Human Decision Processes 102–16 (2000); Jennifer R. Winquist and
James R. Larson, Information Pooling: When It Impacts Group Decision Making, 74 Journal
of Personality and Social Psychology 371–77 (1998); James R. Larson, Pennie G.
Foster-Fishma, and Timothy M. Franz, Leadership Style and the Discussion of Shared and
Unshared Information in Decision Making Groups, 24 Personality and Social Psychology
Bulletin 482–95 (1998); James R. Larson et al., Diagnosing Groups: The Pooling,
Management, and Impact of Shared and Unshared Case Information in Team-based Medical
Decision Making, 75 Journal of Personality and Social Psychology 93–108 (1998);
Gwen M. Wittenbaum, Information Sampling in Decision Making Groups: The Impact of
 group decision making and the effects of accountability 607

most heavily on completely shared information, and information known by a

majority of group members has some effect. However, information known by
only one or few members tends to not be shared or, when it is, tends not to have
as much effect as common information.30
The strength of the common knowledge effect appears to be a function of
several main variables, including:
• The group’s familiarity with the decision domain. Unshared information is
accorded the least weight in situations where group members are familiar
with the subject matter to which the decision pertains.31
• Verifiability of the unshared information. Unshared information that is
objectively verifiable is used more than unshared information that is not.
• The number of group members who possess the information. The more
group members who possess a piece of unshared information, the more
weight it is given.32

Members Task-Relevant Status, 29 Small Group Research 57–84 (1998); Garold Stasser
and Gwen M. Wittenbaum, The Reevaluation of Information During Group Discussion, 1
Group Processes & Intergroup Relations 21–34 (1998); Dennis Stewart and Garold
Stasser, Expert Role Assignment and Information Sampling During Collective Recall and
Decision Making, 69 Journal of Personality and Social Psychology 619–28 (1995);
James R. Larson, Pennie G. Foster-Fishman, and Christopher B. Keys, Discussion of Shared
and Unshared Information in Decision Making Groups, 67 Journal of Personality and
Social Psychology 446–61 (1994); Daniel Gigone and Reid Hastie, The Common
Knowledge Effect: Information Sharing and Group Judgment, 65 Journal of Personality
and Social Psychology 959–74 (1993); Daniel Gigone and Reid Hastie, The Impact of
Information on Group Judgment: A Model and Computer Simulation, in Understanding
Group Behavior: Vol. 1. Consensual Action by Small Groups (Erich H. White and
James H. Davis eds., Hillsdale, NJ: Erlbaum, 1996); Garold Stasser and Dennis Stewart,
Discovery of Hidden Profiles by Decision Making Groups: Solving a Problem versus Making a
Judgment, 63 Journal of Personality and Social Psychology 426–34 (1992); Garold
Stasser, Laurie A. Taylor, and Coleen Hanna, Information Sampling in Structured and
Unstructured Discussions of Three- and Six-Person Groups, 57 Journal of Personality and
Social Psychology 67–78 (1989); Garold Stasser and William Titus, Effects of Information
Load and Percentage of Common Information on the Dissemination of Unique Information
During Group Discussion, 53 Journal of Personality and Social Psychology 81–93
(1987).
30. Craig D. Parks and Rebecca Cowlin, Acceptance of Uncommon Information into
Group Discussion When That Information Is or Is Not Demonstable, 66 Organizational
Behavior and Human Decision Processes 307–15 (1996).
31. Id.
32. Id.
608 problem solving, decision making, and professional judgment

• Contextual factors. Time pressure increases reliance on common

information,33 as does an increase in the decision’s apparent importance
and group members’ sense of accountability for it.34

19.4.2 Causal Accounts of the Common Knowledge Effect

A number of causal theories have been advanced to explain the common knowl-
edge effect. These include:
• The Probabilistic Information Sampling Account;
• The Confirmation Bias Account;
• The Social Validation Account;
• The Mutual Enhancement Account;
• The Social Identity Formation Account;
• The Conversational Norms Account.
There is no reason to believe that they are mutually exclusive.
Probabilistic sampling. The essence of the common knowledge effect is that
the probability that a particular item of information will be used in group delib-
eration is a function of the number of group members who possess the informa-
tion before discussion begins. The earliest explanation for the effect was simply
that, as a matter of basic sampling theory, the more group members who possess
a piece of decision-relevant information ex ante, the higher the probability that it
will be recalled by at least one group member and mentioned during group dis-
cussion.35
But this information sampling account was unable to account for some aspects
of the common knowledge effect. Specifically, even when group members pos-
sessing unshared information shared it with the group, it was not subsequently

33. Steven J. Karau and Janie R. Kelly, The Effects of Time Scarcity and Time Abundance
on Group Performance Quality and Interaction Process, 28 Journal of Experimental
Social Psychology 542–71 (1992).
34. James R. Larson, Pennie G. Foster-Fishman, and Christopher B. Keys, Discussion
of Shared and Unshared Information in Decision Making Groups, 67 Journal of Personality
and Social Psychology 446–61 (1994); Dennis Stewart, Robert S. Billings, and Garold
Stasser, Accountability and the Discussion of Unshared, Critical Information in Decision
Making Groups, 2 Group Dynamics 18–23 (1998).
35. Garold Stasser and William Titus, Effects of Information Load and Percentage of
Common Information on the Dissemination of Unique Information During Group Discussion,
53 Journal of Personality and Social Psychology 81–93 (1987); James R Larson,
Pennie G. Foster-Fishman, and Christopher B. Keys, Discussion of Shared and Unshared
Information in Decision Making Groups, 67 Journal of Personality and Social
Psychology 446–61 (1994); Garold Stasser, Information Salience and the Discovery of
Hidden Profiles by Decision-Making Groups: A Thought Experiment, 52 Organizational
Behavior and Human Decision Processes 5156–81 (1992).
 group decision making and the effects of accountability 609

mentioned in discussion as frequently as commonly held information, and it

received less weight than common information in making the decision.
Confirmation bias. In Section 10.3, we described the bias under which people
tend to overweight information that confirms their initial hypothesis and under-
weight information that would tend to disconfirm it. This bias has been offered
as a partial explanation for the common knowledge effect.
Group members often come to the meeting with some decision-relevant
information and an ex ante preference for a particular decision. To the extent
that more individual group members base their initial preferences on common
pieces of information, the distribution of ex ante preferences will be skewed in
the direction the information supports. Newly acquired information will be
assimilated in a confirmatory direction,36 resulting in the shared information
being mentioned more frequently and weighted more heavily than unshared
information.37
Social validation and mutual enhancement. In group decision-making situa-
tions, participants are often concerned about their task competence. This con-
cern is exacerbated when the subject matter relevant to the decision is unfamiliar,
the decision is very important, or group members feel a high level of account-
ability for the outcome. As a result, participants are constantly trying to assess
the appropriateness and accuracy of their thoughts, arguments, and opinions;
they are worried about how their competence will be assessed, both by them-
selves and by others.
Social comparison processes play a significant role in people’s monitoring of
the accuracy and appropriateness of their thoughts, as the Asch experiments,
described in Section 17.1, illustrated so vividly. The importance, relevance, even
the accuracy of shared information can be validated through a process of social
comparison. As any particular group member hears others mentioning, credit-
ing, and relying upon information she too possess, she feels more secure in
mentioning, crediting, and relying on that information as well. Because unshared
information cannot be so easily validated, mentioning, crediting, or using it in
arguing for a position is more socially risky.38 Indeed, experiments by Gwen

36. Barry M. Staw, The Escalation of Commitment: An Update and Appraisal, in

Organizational Decision Making—Cambridge Series on Judgment and Decision
Making 191–215 (Zur Shapira ed., New York: Cambridge University Press, 1997).
37. Daniel Gigone and Reid Hastie, The Common Knowledge Effect: Information Sharing
and Group Judgment, 65 Journal of Personality and Social Psychology 959–74 (1993);
Jennifer R. Winquist and James R. Larson, Information Pooling: When It Impacts Group
Decision Making, 74 Journal of Personality and Social Psychology 371–77 (1998).
38. Gwen M. Wittenbaum., Anne P. Hubbell, and Connie Zuckerman, Mutual
Enhancement: Toward an Understanding of the Collective Preference for Shared Information,
77 Journal of Personality and Social Psychology 967–78 (1999); Gwen M.
Wittenbaum and Ernest S. Park, The Collective Preference for Shared Information, 10
Current Directions in Psychological Science 172–75 (2001).
610 problem solving, decision making, and professional judgment

Wittenbaum and her collaborators indicate that the discussion of shared infor-
mation leads people both to feel more like a “group” and to feel better about that
group.39
Conversational norms. As every good lawyer and politician knows, effective
communicators tailor their messages to what they believe to be the knowledge,
background, and expectations of their audience. By creating a “cognitive common
ground,” one makes the message both more comprehensible and more persua-
sive to the communicator.
This phenomenon may play a role in the common knowledge effect. If I am a
member of a group, and want to make a point, I may begin by mentioning infor-
mation that is believed by others in the group. If group members follow the
rhetorical axiom, “begin your remarks by mentioning facts you know others
believe true,” they will mention shared information more frequently than
unshared information.40 In this way, the common knowledge effect can result
from the operation of a simple conversational norm.

19.4.3 Reducing the Bias in Favor of Shared Information: What Works, What
Doesn’t, What Makes Things Worse?
Empirical research on the conditions that exacerbate or ameliorate the common
knowledge effect has yielded two generalizations.
First, forewarning people that they may have relevant information that other
group members may not possess has no effect.41 Nor does instructing group
members not to come to judgments about the best course of action before all
information possessed by group members is surfaced.42

39. Gwen M. Wittenbaum and Jonathan M. Bowman, A Social Validation Explanation

for Mutual Enhancement, 40 Journal of Experimental Social Psychology 169–84
(2004); Gwen M. Wittenbaum, Anne P. Hubbell, and Colleen Zuckerman, Mutual
Enhancement: Toward an Understanding of the Collective Preference for Shared Information,
77 Journal of Personality and Social Psychology 967–78 (1999). See also Michael A.
Hogg, Social Identity and the Sovereignty of the Group: A Psychology of Belonging, in
Individual Self, Relational Self, Collective Self 123–43 (Constantine Sedikides and
Marilyn B. Brewer eds., Philadelphia, PA: Psychology Press/Taylor & Francis, 2001).
40. Herbert H. Clark and Susan E. Brennan, Grounding in Communication, in
Perspectives on Socially Shared Cognition 127–49 (Lauren Resnick, John Levine, and
Stephanie D. Teasley eds., Washington, DC: American Psychological Association, 1991);
Gwen M. Wittenbaum and Jonathan M. Bowman, A Social Validation Explanation for
Mutual Enhancement, 40 Journal of Experimental Social Psychology 169–84 (2004).
41. Garold Stasser, Sandra I. Vaughan, and Dennis D. Stewart, Pooling Unshared
Information: The Benefits of Knowing How Access to Information Is Distributed Among Group
Members, 82 Organizational Behavior and Human Decision Processes 102–16 (2000).
42. R. Scott Tindale et al., Shared Cognition in Small Groups, in Blackwell Handbook
of Social Psychology: Group Processes 1–30 (Michael A. Hogg and R. Scott Tindale
eds., Malden, MA: Blackwell Publishers, 2001).
 group decision making and the effects of accountability 611

Second, placing a decision-making group under time pressure, or increasing

the stakes and group members’ sense of their accountability for the ultimate
outcome, exacerbates the effect. These conditions seem to make group members
more risk averse. Using shared information is less risky than using unshared
information and bolsters the group’s confidence in its decision.43
But certain measures tend to reduce the common knowledge effect:
• Framing the task as a “problem to be solved,” implying that it has a “cor-
rect” answer.44
• Publicly assigning roles to different group members the responsibility to
supply information on a particular decision-relevant topic.45
• Extending the time for discussion, which results in the use of more
unshared information.46
• Increasing the number of options from which the group can choose, and
explicitly listing those options.47
• Providing an objective source of verification of unshared information.
(Objectively and authoritatively shared information gets treated the same
as originally shared information, suggesting that sharedness is often used
as a proxy for reliability.48)

43. Steven J. Karau and Janice R. Kelly, The Effects of Time Scarcity and Time Abundance
on Group Performance Quality and Interaction Process, 28 Journal of Experimental Social
Psychology 542–71 (1992); James R. Larson, Pennie G. Foster-Fishman, and Christopher
B. Keys, Discussion of Shared and Unshared Information in Decision Making Groups, 67
Journal of Personality and Social Psychology 446–61 (1994); Dennis D. Stewart,
Robert S. Billings, and Garold Stasser, Accountability and the Discussion of Unshared,
Critical Information in Decision Making Groups, 2 Group Dynamics 18–23 (1998).
44. Garold Stasser and Dennis Stewart, Discovery of Hidden Profiles by Decision Making
Groups: Solving a Problem versus Making a Judgment, 63 Journal of Personality and
Social Psychology 426–34 (1992).
45. Dennis Stewart and Garold Stasser, Expert Role Assignment and Information
Sampling During Collective Recall and Decision Making, 69 Journal of Personality and
Social Psychology 619–28 (1995); Garold Stasser, Sandra I. Vaughan, and Dennis D.
Stewart, Pooling Unshared Information: The Benefits of Knowing How Access to Information
Is Distributed Among Group Members, 82 Organizational Behavior and Human
Decision Processes 102–16 (2000).
46. James R. Larson, Pennie G. Foster-Fishman, and Christopher B. Keys, Discussion
of Shared and Unshared Information in Decision Making Groups, 67 Journal of Personality
and Social Psychology 446–61 (1994); Craig D. Parks and Rebecca Cowlin, Group
Discussion as Affected by Number of Alternatives and by a Time Limit, 62 Organizational
Behavior and Human Decision Processes 267–75 (1995).
47. Craig D. Parks and Rebecca Cowlin, Group Discussion as Affected by Number of
Alternatives and by a Time Limit, 62 Organizational Behavior and Human Decision
Processes 267–75 (1995).
48. Craig D. Parks and Rebecca Cowlin, Acceptance of Uncommon Information into
Group Discussion When That Information Is or Is Not Demonstrable, 66 Organizational
Behavior and Human Decision Processes 307–15 (1996).
612 problem solving, decision making, and professional judgment

• Finally, as we will discuss, using “nominal” groups, through methods like

the “stepladder” or Delphi techniques, can help reduce the bias toward
shared information.

19.5 group polarization

Consider again the Terra Nueva lawsuit, and the efforts of community and con-
sumer activists to use it to mobilize other community groups to take actions
relating to toxic exposure-related harms. Imagine that going into the meeting
described above, representatives from East Los Angeles Tenants United (ELATU),
L.A. Consumers’ Action (LACA), and the Terra Nueva Tenants’ Union are enthu-
siastic about filing similar suits elsewhere. But the organizer from the United
Electrical Workers (UEW) is more cautious. She wants to see how the Terra
Nueva suit unfolds before allocating more union resources to similar initia-
tives—in part, because she thinks that toxic chemicals, rather than the foam
insulation, might be causing the Terra Nueva tenants’ health problems. Suppose
that the plaintiffs’ lawyers have learned about this possibility as well, and that
they share her concern about taking action in other communities, which at this
point seems a risky strategy.
Assuming that the UEW representative and the lawyers make their knowl-
edge and views known to the rest of the group, what do you think will happen
during this meeting? As different strategy options are discussed, will the UEW
representative’s and lawyers’ concerns moderate the enthusiasm for expanding
the litigation strategy with which the other organizations came into the meeting?
Or will deliberation simply embolden participants to take broader actions than
they might have decided on alone?
Your intuition might be that deliberation would tend to temper the judgment
of a group’s most extreme members on either side and draw the group toward a
middle ground.49 But it turns out that rather than moderating the views of the
community organizers, the group may be pulled toward their more extreme
views.50 Group discussion tends to magnify differences within the group and
enhance the position of those whose views initially predominate.
In an early experiment involving what has come to be called group polarization,51
James Stoner presented participants with hypothetical problems involving a

49. See Rupert Brown, Group Processes (2d ed. Malden, MA: Blackwell Publishing,
2000).
50. Cass R. Sunstein, The Law of Group Polarization, 10 Journal of Political
Philosophy 175–95 (2002).
51. Serge Moscovici and Marisa Zavalloni, The Group as a Polarizer of Attitudes, 12
Journal of Personality and Social Psychology 125–35 (1969); David G. Myers and
 group decision making and the effects of accountability 613

choice between cautious and risky courses of action. After making tentative deci-
sions on their own, the participants were placed in groups and asked to reach a
unanimous decision. When participants’ median position before discussion was
risk seeking, group deliberation led to a “risky shift,”52 in which decisions were
riskier than their individual prediscussion choices. Subsequent studies have
shown that “cautious” shifts occur when the predominant view of the individual
participants is risk averse.53 (In addition to the group’s tending to shift to one
extreme or another, individual members internalize the group’s position and
tend to hold it after the discussion has ended.54)
Group polarization is not restricted to shifts in risk preference. For example,
David Myers demonstrated that groups of moderately pro-feminist women sub-
jects became more strongly pro-feminist following discussion of feminist ideas.55
Jurors are affected by group polarization in awarding damages.56 And Cass
Sunstein and his colleagues recently demonstrated that circuit court judges’
ideological tendencies are amplified when they sit on panels consisting of judges
appointed by presidents from the same political party. Republican appointees
show an increased tendency to vote in a stereotypically conservative fashion
when they are accompanied by two other Republican appointees, while

Helmut Lamm, The Group Polarization Phenomenon, 83 Psychological Bulletin 602–27

(1976); Daniel J. Isenberg, Group Polarization: A Critical Review and Meta-Analysis, 50
Journal of Personality and Social Psychology 1141–51 (1986); Cass R. Sunstein, The
Law of Group Polarization, 10 Journal of Political Philosophy 175–95 (2002).
52. James A. F. Stoner, Risky and Cautious Shifts in Group Decisions: The Influence of
Widely Held Values, 4 Journal of Experimental and Social Psychology 442–59 (1968).
James A. F. Stoner, “A Comparison of Individual and Group Decisions Including Risk”
(unpublished thesis, Massachusetts Institute of Technology, School of Management,
Cambridge, MA).
53. Serge Moscovici and Marisa Zavalloni, The Group as a Polarizer of Attitudes, 12
Journal of Personality and Social Psychology 125–35 (1969); Willem Doise,
Intergroup Relations and Polarization in Individual and Collective Judgments, 12 Journal of
Personality & Social Psychology 136–43 (1969).
54. Michael A. Wallach, Nathan Kogan, and Daryl J. Ben, Group Influence on Individual
Risk Taking, 65 Journal of Abnormal and Social Psychology 75–86 (1962). See also
Joel Cooper, Kimberly A. Kelly, and Kimberlee Weaver, Attitudes, Norms, and Social
Groups, in Blackwell Handbook of Social Psychology: Group Processes 259–82
(Michael. A. Hogg and R. Scott Tindale eds., Malden, MA: Blackwell Publishers 2001);
Johannes A. Zuber, Helmut W. Crott, and Joachim Werner, Choice Shift and Group
Polarization: An Analysis of the Status of Arguments and Social Decision Schemes, 62 Journal
of Personality and Social Psychology 50–61 (1992).
55. David G. Myers, Discussion-Induced Attitude Polarization, 28 Human Relations
699–712 (1975).
56. David Schkade, Cass Sunstein, and Daniel Kahneman, Deliberating About Dollars,
100 Columbia Law Review 1139–75 (2000).
614 problem solving, decision making, and professional judgment

Democratic appointees show the same effect in the opposite direction when
accompanied by their co-partisans.57

19.5.1 What Causes Group Polarization?

Causal accounts of group polarization divide into two broad categories, one cen-
tering on social forces, the other on cognitive or informational processes.
19.5.1.a Social Comparison and Social Identity Maintenance There are three
related theories here. One builds on the fact that people generally want to be
viewed well by others. In order to present themselves in a socially desirable light,
they constantly monitor others’ reactions to determine how to adjust their self-
presentation. As group members see how others view the issue under discus-
sion, they tend to present themselves to be congruent with the perceived
emerging trend. Indeed, to be “distinct” from their peers in the socially valued
direction they often go “one step further” in the direction of increasingly extreme
positions.58 Seen in this way, group polarization can be understood as a type of
bandwagon effect.59
Second, people generally see themselves and like to be seen by others as mod-
erate, and initially express positions that may be more moderate than their true
beliefs. But after being exposed to others having their own viewpoints, people
become disinhibited from expressing more extreme views and they become
more confident of those views. Robert Baron and his colleagues demonstrated
that discussion with like-minded people can move individuals to more extreme
positions solely by virtue of this social corroboration effect.60
A third account of polarization, social identity, concerns the relationship of the
group to others outside the group. As group members begin interacting with
each other, they collectively begin to identify the characteristics, behaviors, and
norms that differentiate their group from other groups. To differentiate their

57. Cass R. Sunstein, David Schkade, and Lisa M. Ellman, Ideological Voting on Federal
Courts of Appeals: A Preliminary Investigation, 90 Virginia Law Review 301–55 (2004).
Similar polarization effects in panel and individual judicial decision making was found
three decades earlier by Walker and Main. Thomas G. Walker and Eleanor C. Main, Choice
Shifts in Political Decisionmaking: Federal Judges and Civil Liberties Cases, 3 Journal of
Applied Social Psychology 39–48 (1973).
58. Robert S. Baron and Gard Roper, Reaffirmation of Social Comparison Views of Choice
Shifts: Averaging and Extremity Effects in an Autokinetic Situation, 33 Journal of Personality
and Social Psychology 521–30 (1976); Daniel J. Isenberg, Group Polarization: A Critical
Review and Meta-Analysis, 50 Journal of Personality and Social Psychology 1141–51
(1986); Joel Cooper, Kimberly A. Kelly, and Kimberlee Weaver, Attitudes, Norms, and Social
Groups, in Blackwell Handbook of Social psychology: Group Processes 259–82
(Michael A. Hogg and R. Scott Tindale eds., Malden, MA: Blackwell Publishers, 2001).
59. John C. Turner, Social Influence (Pacific Grove, CA: Brooks/Cole, 1991).
60. Robert S. Baron et al., Social Corroboration and Opinion Extremity, 32 Journal of
Experimental Social Psychology 537–60 (1996).
 group decision making and the effects of accountability 615

group from others, members attribute more extremity to their own group than
is objectively the case; they distinguish their group from others by accentuating
the perceived distance between their group norm and the norms they attribute to
out-groups.61 Group members then conform their own behaviors, norms, and
self-conceptions to their group’s.62 Polarization thus occurs through a process of
contrast with other groups within a specific social context.63
The social accounts of group polarization are not mutually exclusive. Indeed,
each incorporates theoretical elements of the other and experimental research
supports all of them.64
19.5.1.b Cognitive/Informational Accounts of Group Polarization A second set
of theories focuses on cognitive rather than social explanations for group polar-
ization. The persuasive argument theory builds on the observation that, before
group discussion of an issue begins, a member will usually have formulated an
initial position and formulated arguments favoring his position. The group dis-
cussion exposes the participant to new arguments that support or detract from
his initial position. The phenomenon of confirmation bias suggests that the par-
ticipant will find new information and arguments that support his initial posi-
tion more persuasive than information and arguments that oppose it. As a purely
statistical phenomenon, this predicts group polarization, in much the same way

61. Diane Mackie, Social Identification Effects in Group Polarization, 50 Journal of

Personality and Social Psychology 720–28 (1986).
62. John C. Turner, Toward a Cognitive Redefinition of the Social Group, in Social
Identity and Intergroup Relations 15–40 (Henri Tajfel ed., Cambridge, UK: Cambridge
University Press, 1982); John C. Turner, Social Categorization and the Self-Concept: A Social
Cognitive Theory of Group Behavior, in Advances in Group Processes 2, 77–122 (Edward
J. Lawler ed., Greenwich, CT: JAI, 1991); Diane Mackie, Social Identification Effects in
Group Polarization, 50 Journal of Personality and Social Psychology 720–28 (1986);
Michael A. Hogg, John C. Turner, and Barbara Davidson, Polarized Norms and Social
Frames of Reference: A Test of the Self-Categorization Theory of Group Polarization, 11 Basic
and Applied Social Psychology 77–100 (1990).
63. Michael A. Hogg, John C. Turner, and Barbara Davidson, Polarized Norms and
Social Frames of Reference: A Test of the Self-Categorization Theory of Group Polarization, 11
Basic and Applied Social Psychology 77–100 (1990); Craig McGarty et al., Group
Polarization as Conformity to the Prototypical Group Member, 31 British Journal of
Personality and Social Psychology 1–19 (1992). This social identity account is sup-
ported by experimental evidence showing that the context in which a deliberating group
is embedded influences the extent and the direction of group polarization. Specifically, if
a deliberating group is embedded with outgroups at one or another pole of an attitudinal
continuum, the group will polarize away from the direction in which those outgroups
lean. This is consistent with the operation of the in-group/outgroup contrast effect that
lies at the center of the social identity account.
64. Gary Stasser and Beth Dietz-Uhler, Collective Choice, Judgment, and Problem Solving,
in Blackwell Handbook of Social Psychology: Group Processes 31–55 (Michael A.
Hogg and R. Scott Tindale eds., Malden, MA: Blackwell Publishers, 2001).
616 problem solving, decision making, and professional judgment

as probabilistic sampling theory predicted the common information effect dis-

cussed earlier in Section 19.4.2.65
Another cognitive account of group polarization understands group interac-
tion as an exercise in rationale construction rather than information collection.
This model posits that people bring to a discussion their ex ante preferences and
that interaction calls upon them to explain these to other group members. As
people explain their positions, they gain confidence in them, whether or not they
are accurate. Indeed, as Chip Heath and Richard Gonzalez have shown, interac-
tion with others increases confidence in, but not the accuracy of, judgments.66 A
related causal theory draws on the effects of repeated expression: polarization occurs
as people repeat their own arguments and are repeatedly exposed to particular
arguments, evidence, or expressions of a conclusion by like-minded others.67

19.5.2 Reducing the Tendency Toward Group Polarization

There has been very little research directly testing the effectiveness of particular
interventions in reducing group polarization. However, some inferences can be
drawn from research on the contextual factors that tend to reduce or exacerbate
the effect:68
• Polarization is moderated by heterogeneity of initial opinions within a
deliberating group.
• As with groupthink, polarization is exacerbated where the group is an ideo-
logically cohering, high-solidarity entity.

65. Eugene Burnstein and Amiram Vinokur, What a Person Thinks upon Learning He
Has Chosen Differently from Others: Nice Evidence for the Persuasive Arguments Explanation
of Choice Shifts, 11 Journal of Experimental Social Psychology 412–26 (1975); Eugene
Burnstein and Amiram Vinokur, Persuasive Argumentation and Social Persuasion as
Determinants of Attitude Polarization, 13 Journal of Experimental Social Psychology
315–32 (1977); Markus Brauer, Charles M. Jud, and Melissa D. Gliner, The Effects of
Repeated Expressions on Attitude Polarization During Group Discussions, 68 Journal of
Personality and Social Psychology 1014–29 (1995).
66. Chip Heath and Richard Gonzalez, Interaction with Others Increases Decision
Confidence but Not Decision Quality: Evidence Against Information Collection Views of
Interactive Decisionmaking, 61 Organizational Behavior and Human Decision
Processes 305 (1995).
67. Markus Brauer, Charles M. Jud, and Melissa D. Gliner, The Effects of Repeated
Expressions on Attitude Polarization During Group Discussions, 68 Journal of Personality
and Social Psychology 1014–29 (1995).
68. Harris Bernstein, Persuasion as Argument Processing, in Group Decisionmaking
(Hermann Brandstatter, James H. Davis, and Gisela Stocker-Kreichgauer eds., London:
Academic Press, 1982); Rupert Brown, Group Processes (2d ed. Malden, MA: Blackwell
Publishing, 2000); Cass Sunstein, Deliberative Trouble? Why Groups Go to Extremes, 110
Yale Law Journal 72–119 (2000).
 group decision making and the effects of accountability 617

• Groups that work together over a sustained period of time, making numer-
ous similar decisions, may get feedback on the quality of those decisions,
which may moderate overconfidence and lead to more effective use of
information in subsequent decision making.
• Polarization is moderated where a group is subject to external shocks (e.g.,
losing an election, being sued, experiencing a severe policy failure) or
where the group must temper its preferences to preserve its prestige with
external constituencies.
• Polarization can be reduced by increasing decision-making independence
through mechanisms like the Delphi technique (Section 19.8.4), and
through norms and facilitation that encourage open-mindedness, the
inclusion of minority perspectives, and evidence-based decision making.69

19.6 strategic participation in group decision making

Group members’ concerns about their relations with others in the group play a
significant role in the pathologies of group decision making. Individual group
members may have interests that differ from those of their colleagues. So, for
example, in a meeting to help a company’s managers decide whether to settle or
litigate an employment discrimination case, the lawyer may be concerned to
maintain the impression that she is “on their side” even as she attempts to
impress upon them the nature of the exposure to risk that they face. She may
also want to convey a gender counter-stereotypical “toughness” that she hopes
will instill confidence in her competence as a vigorous advocate.
Other participants in the decision-making process may have their own sets of
competing processing goals. The manager who made the now-challenged deci-
sion may seek to save face, preventing his colleagues from concluding that he
made a mistake or otherwise acted imprudently. Yet another group member may
simply want to create a good impression and be viewed as a “team player.” And
all (or almost all) of the group members will be attempting to maintain a certain
level of social harmony, both during and after the decision making process.
In other words, the goal of making a “good” decision is only one of the items
on members’ individual and collective agendas. The susceptibility of a group to
decision-making errors will depend in substantial measure on these often hidden
competing motivations.

69. Tom Postmes, Russell Spears, and Sezgin Cihangir, Quality of Decisionmaking and
Group Norms, 80 Journal of Personality and Social Psychology 918–30 (2001).
618 problem solving, decision making, and professional judgment

19.7 the susceptibility of groups to biases and heuristics

Earlier in the book, we described a variety of biases and heuristics that system-
atically distort individuals’ judgment and decision making. Despite some hopes
to the contrary,70 groups are not less susceptible to these phenomena than indi-
viduals, and in some circumstances are more so.
Overconfidence. As we saw in Section 10.4, individuals often are overconfi-
dent in the accuracy of their own judgment. It turns out that groups don’t do
better—even when even they appoint a “devil’s advocate” or explicitly consider
reasons why their estimates might be wrong—and may actually be more over-
confident.71 Group overconfidence may result from the group polarization
effect.72 Persuasive argument theory73 suggests that if (as often happens) group
members enter the group discussion with an overconfident orientation, they will
make more confidence-inducing than caution-supporting arguments, thus skew-
ing overall group opinion in an optimistic direction. Second, social comparison
processes may come into play, as group members offer overconfident outlooks
on the subject of group discussion so as to demonstrate loyalty and commitment
to the project, and to assure that they compare favorably in this regard with other
group members.74
Confirmation bias. Small deliberating groups, no less than individuals,
fall prey to the tendency to overweight information that confirms a prior
position, theory, or attitude and to underweight information that would tend to
disconfirm it. Groups manifest a preference for information that confirms indi-
vidual members’ prediscussion preferences. The more social support for their
views participants anticipated from an upcoming group discussion, the more
preference they showed for confirming information.75 For both laypeople and

70. George P. Huber, Managerial Decision Making (Glenview, IL: Scott Foresman,
1980).
71. Scott Plous, A Comparison of Strategies for Reducing Interval Overconfidence in Group
Judgments, 80 Journal of Applied Psychology 443–54 (1995).
72. Roger Buehler, Deanna Messervey, and Dale Griffin, Collaborative Planning and
Prediction: Does Group Discussion Affect Optimistic Biases in Time Estimation?, 91
Organizational Behavior and Human Decision Processes 47–63 (2005).
73. Amiram Vinokur and Eugene Burnetein, The Effects of Partially Shared Persuasive
Arguments on Group Induced Shifts: A Group Problem Solving Approach, 29 Journal of
Personality and Social Psychology 305–15 (1974).
74. Daniel Kahneman and Dan Lovallo, Timid Choices and Bold Forecasts: A Cognitive
Perspective on Risk Taking, 39 Management Science 17–31 (1993).
75. Bozena Zdaniuk and John M. Levine, Anticipated Interaction and Thought
Generation: The Role of Faction Size, 36 British Journal of Social Psychology 201–18
(1996). On a related matter, group discussion may attenuate or eliminate the theory-per-
severance effect, in which people retain a belief in the functional relationship between two
variables even after evidence for any such relationship has been discredited. Edward F.
 group decision making and the effects of accountability 619

groups of experts making decisions within their areas of expertise (say, business
managers making economic decisions), the more homogeneous the members’
preferences, the stronger the effect.76 Only when the deliberating group is
reasonably balanced on the issue in question does confirmation bias diminish.
Various judgmental errors. The little research that exists provides no clear pic-
ture about the relative susceptibility of individuals and groups to the availability
and representativeness heuristics.77 Group discussion seems to amplify conjunc-
tion errors when the individual members’ susceptibility to conjunction bias is
high, and to attenuate the errors when it is low.78 Group interaction does not
appear to attenuate base rate neglect and may, indeed, increase it.79 The evidence
on groups’ proneness to hindsight bias is equivocal.80 Deliberating groups appear
to be even more prone to unrealistic optimism than individuals;81 this may result
from the same informational, social comparison, and social identity manage-
ment processes mentioned in connection with group polarization.82

Wright and Scott D. Christie, “The Impact of Group Discussion on the Theory-Perseverance
Effect” (unpublished manuscript, St. Francis Xavier University, Department of Psychology,
Antigonish, NS). Craig A. Anderson, Inoculation and Counterexplanation: Debiasing
Techniques in the Perseverance of Social Theories, 1 Social Cognition 126–39 (1982).
76. Stefan Schulz-Hardt et al., Biased Information Search in Group Decisionmaking, 78
Journal of Personality and Social Psychology 655–69 (2000).
77. Mark F. Stasson et al., Group Consensus Processes on Cognitive Bias Tasks: A Social
Decision Scheme Approach, 30 Japanese Psychological Research 68–77 (1988).
78. R. Scott Tindale, Susan Sheffey, and Joseph Filkins, “Conjunction Errors by
Individuals and Groups” (paper presented at the meetings of the Society for Judgment
and Decisionmaking, New Orleans, LA, 1990); R. Scott et al., “An Attempt to Reduce
Conjunction Errors in Decision-Making Groups” (paper presented at the annual meeting
of the Society for Judgment and Decisionmaking, Washington, DC, 1993). Other research-
ers obtained mixed results in studies testing susceptibility to other types of representative-
ness biases, finding bias amplification in some contexts, but not others. Mark F. Stasson
et al., Group Consensus Processes on Cognitive Bias Tasks: A Social Decision Scheme Approach,
30 Japanese Psychological Research 68–77 (1988).
79. Linda Argote, Mark A. Seabright, and Linda Dyer, Individual versus Group Use of
Base-Rate and Individuating Information, 38 Organizational Behavior and Human
Decision Processes 65–75 (1986); Linda Argote, Rukmini Devadas, and Nancy Melone,
The Base-Rate Fallacy: Contrasting Processes and Outcomes of Group and Individual Judgment,
48 Organizational Behavior and Human Decision Processes 296–311 (1990); R.
Scott Tindale, Decision Errors Made by Individuals and Groups, in N. John Castellan Jr.
(ed.), Individual and Group Decision Making: Current Issues 109–24 (Hillsdale, NJ,
England: Lawrence Erlbaum Associates, Inc., 1993).
80. Dagmar Stahlberg et al., We Knew It All Along: Hindsight Bias in Groups, 63
Organizational Behavior and Human Decision Processes 46–58 (1995).
81. Roger Buehler, Deanna Messervey, and Dale Griffin, Collaborative Planning and
Prediction: Does Group Discussion Affect Optimistic Biases in Time Estimation?, 91
Organizational Behavior and Human Decision Processes 47–63 (2005).
82. Id.
620 problem solving, decision making, and professional judgment

Gain/loss framing effects and other departures from expected utility theory.
The research on group susceptibility to framing effects shows mixed results.
One study found that group discussion amplified the asymmetrical treatment of
potential losses and gains,83 while another found that group discussion attenu-
ated framing bias.84 Yet another found bias when the outcome had been framed
by individual group members as a loss or a gain before group discussion began;
but framing bias was attenuated if the outcome was not framed until after group
discussion began, or if it was reframed in the course of the discussion.85 The
research does not suggest any robust generalizations about gain/loss framing or,
consistency with expected utility theory more generally.86
Escalation of commitment. Some research suggests that small groups are, if
anything, more prone than individuals to escalate commitment to a failing
course of action.87 This should not be surprising, since the escalation of commit-
ment is caused, at least in part, by the desire to appear consistent to others,88 and
since group polarization could readily contribute to escalation. However, other
research found no difference between individuals and groups in the amounts of
resources they were willing to invest in a failing endeavor.89
The fundamental attribution error. Curiously, group discussion appears to
attenuate susceptibility to the fundamental attribution error—the tendency to

83. Timothy W. McGuire, Sara Kiesler, and Judith Siegel, Group and Computer-
Mediated Discussion Effects in Risk Decisionmaking, 52 Journal of Personality and
Social Psychology 917–30 (1987).
84. Margaret A. Neale et al., “Choice Shift” Effects in Group Decisions: A Decision Bias
Perspective, 2 International Journal of Small Group Research 33–42 (1986).
85. Paul W. Paese, Mary Bieser, and Mark E. Tubbs, Framing Effects and Choice Shifts
in Group Decisionmaking, 56 Organizational Behavior and Human Decision
Processes 149–65 (1993).
86. Norbert L. Kerr, Robert J. MacCoun, and Geoffrey P. Kramer, Bias in Judgment:
Comparing Individuals and Groups, 103 Psychological Review 687–719 (1996); John C.
Mowen and James W. Gentry, Investigation of the Preference-Reversal Phenomenon in a New
Product Introduction Task, 65 Journal of Applied Psychology 715–22 (1980); John Bone,
John Hey, and John Suckling, Are Groups More (or Less) Consistent than Individuals?, 8
Journal of Risk and Uncertainty 63–81 (1999).
87. Glen Whyte, Escalating Commitment in Individual and Group Decisionmaking, 54
Organizational Behavior and Human Decision Processes 430 (1993).
88. Barry Staw and Jerry Ross, Behavior in Escalation Situations: Antecedents, Prototypes,
and Solutions, in Research in Organizational Behavior (Larry L. Cummings and Barry
Staw eds., Greenwich: CT: JAI Press, 1987).
89. Max H. Bazerman, Toni Giuliano and Allan Appelman, Escalation of Commitment
in Individual and Group Decisionmaking, 33 Organizational Behavior and Human
Performance 141–52 (1984).
 group decision making and the effects of accountability 621

over-attribute people’s behavior to dispositional factors and under-attribute it to

situational variables.90

19.8 improving group decision making

This section considers several approaches to reducing the pathologies that can
infect group decision making. These include measures suggested by Irving Janis
and his collaborators to prevent groupthink and other forms of premature con-
vergence, and two mechanisms—the “stepladder” and Delphi techniques—that
can be used to maximize the constructive contribution of each member to a
decision-making group.

19.8.1 Preventing Groupthink

Janis’s ultimate goal was to develop strategies for improving government deci-
sion making. To this end, he suggested a set of remedial measures that could
prevent the emergence of groupthink and facilitate what he referred to as vigilant
appraisal, or vigilant decision making.91
Vigilant appraisal looks a great deal like the optimal, or full, formal model of
decision making introduced in Section 1.4.1. It entails:
• a careful specification of all relevant goals and objectives, not only those
immediately identified by the group’s leadership;
• extensive and unbiased search for and use of relevant information;
• careful, critical consideration of a range of alternative courses of action;
• the development of contingency plans.
Janis’s remedial measures are all designed to break down the tendency of a
highly cohesive group to function as a single social and psychological entity.92
They require that the group be guided by an open, nondirective leader who takes
tangible steps to foster divergent and critical thinking, and who not only toler-
ates, but encourages, contrarian thinking and dissent.
To facilitate vigilant appraisal, the leader must adopt a neutral role, refraining
from directly endorsing or covertly signaling support for a particular choice. He
or she must encourage the expression of minority viewpoints, ensure that input

90. Edward F. Wright and Gary L. Wells, Does Group Discussion Attenuate the
Dispositional Bias?, 15 Journal of Applied Psychology 531–46 (1985).
91. Irving L. Janis, Groupthink: Psychological Studies of Policy Decisions
and Fiascos (Boston, MA: Houghton Mifflin, 1982); Irving L. Janis and Leon Mann,
Cognitive Complexity in International Decisionmaking, in Psychology and Social
Policy 33–49 (Peter Suedfeld and Philip Tetlock eds., Washington, DC: Hemisphere
Publishing Corp., 1992).
92. Alexander S. Haslam, Psychology in Organizations: The Social Identity
Approach (London, UK & Thousand Oaks, CA: Sage Publications, 2001).
622 problem solving, decision making, and professional judgment

is obtained from experts outside the group, and even appoint formal devil’s
advocates to aggressively question any emerging consensus. As Clark McCauley
has described the leader’s task, “If groupthink is a disease of insufficient
search for information, alternatives, and modes of failure, the cure is better
search procedures: impartial leadership that encourages airing of doubts and
objections in the group and encourages search for information and evaluation
from sources outside the group, procedural norms supporting systematic con-
sideration of alternatives, ‘second-chance’ meetings after preliminary consen-
sus, and special attention to the motives and communications of threatening
competitors.”93
Janis’ specific suggestions for enhancing vigilant decision making include
the following:
1. The leader should be impartial and should not state or even imply prefer-
ences or expectations about the decision at the outset.
2. The leader should assign the role of critical evaluator to each member,
encouraging all participants to vigorously air objections and doubts.
3. Periodically, the group should be broken up into smaller subgroups,
under different leaders, to work on the same question. Then the groups
should be brought back together to compare perspectives and discuss
differences.
4. Periodically, each group member should discuss the subject under delib-
eration with trusted associates who are not in the decision-making group,
and report back to the group their ideas and reactions.
5. Experts and qualified colleagues who are not members of the decision-
making group should periodically be invited to attend group meetings and
challenge the views of the group’s members.
6. At each meeting devoted to the discussion of policy alternatives, at least
one member should be assigned the role of devil’s advocate.
7. Whenever the issue being discussed involves relations with a rival organi-
zation, nation, or other collectivity, ample time should be allocated to
constructing varying conceptions of the rival’s intentions.
8. After the group reaches a preliminary consensus about the best decision,
the leader should conduct a “second chance” meeting at which group
members are instructed to express, as forcefully and persuasively as they
can, all of their doubts about the preliminary consensus, and to reconsider
the entire problem before making a final choice.

93. Clark McCauley, Group Dynamics in Janis’s Theory of Groupthink: Backward and
Forward, 73 Organizational Behavior and Human Decision Processes 142–62
(1998).
 group decision making and the effects of accountability 623

19.8.2 Preventing Premature Convergence: Devil’s Advocacy and

Dialectical Inquiry
One of the most common structures for group-based strategic decision making
is an “expert approach,” where a group of individuals, selected for their pre-
sumed expertise, present and seek consensus for a single recommended course
of action.94 This procedure invites premature convergence.
Devil’s advocacy is one way to inhibit premature convergence. After a plan is
proposed by an individual or subgroup, another individual or subgroup attempts
to identify everything wrong with the plan The larger group then uses this cri-
tique to improve or replace the original proposal.95 Problem-solving groups that
use devil’s advocacy have been shown to generate a wider range of alternative
courses of action than do conventional deliberating groups.96 Janis cites President
John F. Kennedy’s handling of the Cuban Missile Crisis as an example of suc-
cessful use of devil’s advocacy. During those critical days in October 1962,
President Kennedy assigned his brother, Attorney General Robert F. Kennedy,
the role of devil’s advocate with the task of aggressively challenging the assump-
tions and analysis underlying any emerging consensus. Devil’s advocacy is not
without its limitations: for example, if group members view the devil’s advocate
as being insincere, they may ignore his critique.
In dialectical inquiry, the entire group plays the role of devil’s advocate. After
identifying and discussing an initial plan, the entire group develops a “counter-
plan.” There follows a structured debate, in which those responsible for making
an ultimate decision hear arguments for and against the plan and the counter-
plan, with the possibility that yet another plan may emerge.
Devil’s advocacy and dialectical inquiry can be used in face-to-face decision-
making groups, or can be incorporated into computer-mediated group processes.
Because they permit anonymity, computer-mediated processes reduce the effects
of social status,97 reduce social pressures from dominant group members, and
inhibit social conformity effects more generally. However, anonymity can lead to

94. Richard Mason, A Dialectical Approach to Strategic Planning, 15 Management

Science B403–B411 (1969).
95. Joseph S. Valacich and Charles Schwenk, Devil’s Advocacy and Dialectical Inquiry
Effects on Face-to-Face and Computer-Mediated Group Decisionmaking, 63 Organizational
Behavior and Human Decision Processes 158–73 (1995).
96. Charles Schwenk, Effects of Planning Aids and Presentation Media on Performance
and Affective Responses in Strategic Decisionmaking, 30 Management Science 263–72
(1984).
97. Caryn Christenson and Ann Abbott, Team Medial Decisionmaking, in
Decisionmaking in Health Care 267 (Gretchen Chapman and Frank Sonnenberg eds.,
New York: Cambridge University Press, 2000).
624 problem solving, decision making, and professional judgment

incivility, create dissention, and injure morale.98 Computer-mediated groups can

create a sense of alienation, and lead to dissatisfaction with the process.99 And
groups that use computer-mediated devil’s advocacy tend to have difficulty reach-
ing agreement.100

19.8.3 The Stepladder Technique

Many of the dysfunctions of small decision-making groups result from social
processes that interfere with the group’s ability to fully utilize the potential con-
tribution of all of its members. The group may be dominated by particularly
aggressive individuals who stifle input from others in the group.101 Some group
members may engage in social loafing, depriving the group of their potential
contribution.102 Social pressures and social conformity effects may cause group
members to withhold valuable information, opinions, or perspectives.103
In the 1990s, Steven Rogelberg and his colleagues developed a procedure
known as the stepladder technique to minimize the impact of these negative
dynamics on group deliberation.104 The stepladder technique structures the entry
of individuals into a decision-making group so as to maximize their indepen-
dence from other group members and highlight their uniquely held informa-
tion, opinions, ideas, and perspectives.
Before entering the discussion, each member is presented with the group’s
task and sufficient time to think about it alone. After the first two members dis-
cuss the problem, new members join the group one at a time. When entering the
group, each new member must present his or her preliminary solution to the
problem before hearing what the other group members think. Then the group
discusses the problem before another new member enters. No final decision is

98. Joseph S. Valacich and Charles Schwenk, Devil’s Advocacy and Dialectical Inquiry
Effects on Face-to-Face and Computer-Mediated Group Decisionmaking, 63 Organizational
Behavior and Human Decision Processes 158–73 (1995).
99. Leonard M. Jessup, Terry Connolly, and Jolene Galegher, The Effects of Anonymity
on Group Process in an Idea-Generating Task, 14 MIS Quarterly 313–21 (1990).
100. Joseph S. Valacich and Charles Schwenk, Devil’s Advocacy and Dialectical Inquiry
Effects on Face-to-Face and Computer-Mediated Group Decisionmaking, 63 Organizational
Behavior and Human Decision Processes 158–73 (1995).
101. Dennis Falk and David W. Johnson, The Effects of Perspective-Taking and Egocentrism
on Problem Solving in Heterogeneous and Homogeneous Groups, 102 Journal of Social
Psychology 63–72 (1977).
102. Bibb Latane, Kipling Williams, and Stephen Harkins, Many Hands Make Light the
Work: The Causes and Consequences of Social Loafing, 37 Journal of Personality and
Social Psychology 822–32 (1979).
103. David Johnson and Frank Johnson, Joining Together: Group Theory and
Group Skills (Engelwood Cliffs, NJ: Prentice-Hall, 1987).
104. Steven G. Rogelberg, Janet L. Barnes-Farrell, and Charles A. Lowe, The Stepladder
Technique: An Alternative Group Structure Facilitating Effective Group Decisionmaking, 77
Journal of Applied Technology 730–37 (1992).
 group decision making and the effects of accountability 625

made until all group members have been incorporated into the group in this
manner.
The stepladder technique tends to improve a group’s decision performance.
Stepladder groups outperformed both their average and best member more
often than conventional deliberating groups.

19.8.4 Generating and Evaluating Alternatives with Nominal Groups:

The Delphi Technique
As described in Section 19.2, the members of a “nominal group” work alone to
solve a problem and their individual solutions are then aggregated. In some situ-
ations, nominal groups are more effective than either individuals or interactive
groups. For example, nominal groups tend to outperform face-to-face groups in
brainstorming tasks, where the goal is to generate lots of ideas and eventually
improve on them.105
The Delphi Technique is a well-known form of nominal group brainstorm-
ing.106 A coordinator transmits questions to group members. Working alone, the
members transmit their responses to the coordinator, who assembles the infor-
mation and generates additional information requests. Suppose, for example,
that a humanitarian relief nongovernmental organization (NGO) is deciding
what precautions to take in a conflict-ridden country in which its workers are in
physical danger, or whether to pull out altogether. The coordinator states the
problem, and asks each participant to generate as many ideas as possible for
dealing with the issue. The coordinator then asks for comments on their
strengths and weaknesses, and asks for refinements and new ideas. The process
continues, with eventual resolution either through the emergence of a consen-
sus or through voting.

19.9 avoiding “predictable surprises”

The discussion so far has focused on groups coming to conclusions about issues
recognized as requiring a decision. But organizations may have impediments to
even recognizing the need for a decision. They can become trapped in a particu-
lar mindset, and be unable or unwilling to adjust to a changing reality.

105. Norbert L. Kerr and R. Scott Tindale, Group Performance and Decisionmaking, 55
Annual Review of Psychology 623–55 (2004); Brian Mullen, Craig Johnson, and
Eduardo Salas, Productivity Loss in Brainstorming Groups: A Meta-Analytic Integration, 115
Psychological Bulletin 210–27 (1991).
106. Gene Rowe and George Wright, Expert Opinions in Forecasting: The Role of the
Delphi Technique, in Principles of Forecasting 125–44 (J. Scott Armstrong ed., Norwell,
MA: Kluwer Academic Publishers, 2001).
626 problem solving, decision making, and professional judgment

Without delving into the large field of organizational behavior, we summarize

Max Bazerman’s and Michael Watkins’ study of “predictable surprises”—
instances in which entire organizations failed to recognize impending crises and
correct their behavior in order to avert them, despite “prior awareness of all of
the information necessary to anticipate the events and their consequences.”107
Bazerman and Watkins describe six basic characteristics that lead to predict-
able surprises:
1. People know that a problem exists that would not solve itself. (For example,
Enron and Arthur Andersen managers knew that not having independent
consultants and auditors would lead to conflicts of interests.)
2. People recognize that the problem is getting worse over time, but do not
respond adequately. (For example, global warming is widely recognized to
be an impending crisis, but the nations of the world have yet to make it a
top priority.)
3. Fixing the problem is costly, while the benefits of action are delayed. (For
example, investing in alternative sources of energy requires a large upfront
investment, while the rewards may not accrue until many years into the
future.)
4. Along the same lines, fixing the problem will incur a certain cost in order to
prevent an uncertain outcome. (For example, investing in security for ports
and chemical plants has little incentive for leaders, since they are answer-
able for the high costs, but get no credit for the absence of a catastrophe.)
5. Needed changes do not occur because of the tendency to maintain the
status quo. (For example, reducing the national debt may require reduc-
tions and/or equalization of Americans’ standards of living.)
6. A “vocal minority benefits from inaction and is motivated to subvert the
actions of leaders for their own private benefit.” (For example, the U.S.
auto industry fought hard to prevent increased fuel efficiency standards
which they believe would be costly for them to meet.)
Bazerman and Watkins suggest a three-part model to help avoid predictable
surprises. Their “RPM” model consists of: Recognizing the problem, making it
a Priority, and Mobilizing a response.
First, recognizing a problem requires an effortful process that helps prevent
groups from collectively falling prey to the tendency to see what they expect or
want to see, just as individuals often do. For instance, setting clear performance
thresholds helps groups recognize when a strategy is failing, thus triggering a
reevaluation that might otherwise have been neglected.

107. Max H. Bazerman and Michael D. Watkins, Predictable Surprises: The

Disasters You Should Have Seen Coming, and How to Prevent Them: Leadership
for the Common Good (Boston, MA: Harvard Business Press, 2004).
 group decision making and the effects of accountability 627

Second, once it identifies potential problems, an organization must prioritize

them. The authors suggest assigning an advocate to focus on a particular prob-
lem and to give that person a strong voice in the organization. This helps to
ensure that as long as the problem has a committed and consistent advocate, it
cannot be easily ignored.
Finally, once a potential problem has been recognized and made a priority,
the organization must mobilize to confront it. Here the authors suggest strate-
gies, such as effective coalition-building, to ensure that various actors are work-
ing toward the same ends.

19.10 optimizing group decision making

For all of the hazards described in the preceding pages, group decision making
is inevitable. In an increasingly complex world, many problems cannot be ade-
quately addressed by a single individual with limited knowledge and cognitive
capacity. Groups can pool knowledge, information, and experience. They can
divide a complex task into multiple components. And groups have the potential
to correct members’ errors, to reduce decision variability, and moderate indi-
vidual idiosyncrasies. Moreover, participation in making decisions can increase
people’s motivation and sense of commitment, and can enhance the perceived
legitimacy of whatever course of action is chosen.
Rupert Brown noted, “[T]he question is not . . . how we can dispense with
group decision making, but how we can capitalize on its advantages while cir-
cumventing its defects.”108 The key to answering his question lies in assembling
a decision-making group whose members bring diverse knowledge sets, skills,
and perspectives to the task, and then utilizing the potential unique contribution
of each member. The independence of group members must be preserved, so
that they can, in fact, make their diverse contributions. This requires addressing
pressures toward social conformity, power imbalances, and members’ needs to
maintain their social position. The decision-making process must be designed to
ensure that the group considers a broad range of alternative options before con-
verging on a decision. And, where the group is to persist over time and make
repeated decisions, mechanisms must be designed to aggregate and re-present
information to enable the group and its members to receive feedback and learn
from experience.
Lawyers and policy makers regularly serve as the leaders of decision-making
groups. The leader’s role is key in harnessing the power of groups and in pre-
venting the pitfalls that attend group decision making. Understanding the

108. Rupert Brown, Group Processes (2d ed. Malden, MA: Blackwell Publishing,
2000).
628 problem solving, decision making, and professional judgment

dynamics of group decision making is essential for good policy making in an

increasingly complex world“.

19.11 the effects of accountability on individual and group

decision-making performance

The preceding discussion has focused on the interaction of decision makers with
other people. But knowledge that one will be required to justify one’s decision to
others may influence one’s decision, even without any actual interaction with
one’s audience. Mere anticipation of others’ reactions may cause decision makers
to change their approach to the problem.
Accountability has become a popular concept in business, politics, and else-
where, but its effects are not always clearly understood. Received wisdom tells us
that that a decision maker’s accountability to others improves the quality of the
decision. Though this sometimes is true, accountability can also lead to poorer
decision making. This section summarizes the literature on the effects of
accountability.109
All other things being equal, accountability is likely to reduce error and bias
in contexts in which, for whatever reasons, people tend to make mistakes that
they could prevent with extra attention or effort. For example, the expectation
that they will have to justify their decision leads people to think more carefully
and logically, and not to be satisfied with using unarticulated criteria or unsub-
stantiated empirical judgments to arrive at answers. Moreover, accountability
makes people more likely to identify their own sources of bias, because of the
need to justify themselves to others who do not necessarily view the decision
with the same biases.
But, like other social phenomenon, the benefits of accountability depend on a
number of factors, and holding a decision maker accountable can sometimes be
counterproductive. Like other social phenomena, the benefits of accountability
depend on a number of factors, such as:
• Whether the views of the audience are known or unknown. Decision makers
who know the views of the audience to whom they are accountable tend to
conform decisions to gain the audience’s favor. For example, officials
awarding financial aid who did not need to justify their decisions to
students tended to award aid based on recipients’ needs; accountable
officials tried to please all applicants by giving something to everyone.
When the audience’s views are unknown, decision makers tend to engage

109. This section is based largely on the excellent review essay by Jennifer S. Lerner
and Philip E. Tetlock, Accounting for the Effects of Accountability, 125 Psychological
Bulletin 225 (1999).
 group decision making and the effects of accountability 629

in preemptive self-criticism, anticipating how the decision would be viewed

from various perspectives.
• Whether one is accountable for the quality of the decision process or the decision’s
substantive outcome. For example, a foundation could hold its program
officers accountable for the success of their grants each year, or it could
focus instead on whether the program officers were making thoughtful,
informed outcome-focused decisions. Ultimately, the proof of a good pro-
cess is in its outcomes; but this may take many years. A program officer
who is held accountable only for outcomes may be afraid to take appropri-
ate risks in making grants since, even if the risks are well-considered in
terms of expected value, they could still fail and lead to a bad outcome. On
the whole, accountability for the process tends to improve the decision’s
accuracy and calibration while accountability for outcome can reduce
decision quality.
Even when accountability has the potential to improve decision making, it
does not necessarily attenuate all biases, and may actually amplify some. In a
review of the experimental literature, Jennifer Lerner and Philip Tetlock con-
cluded that “accountability attenuated biases to the extent that (a) suboptimal
performance resulted from lack of self-critical attention to the judgment process
and (b) improvement required no special training in formal decision rules, only
greater attention to the information provided.”
Accountability may actually lead to poorer decision making when a biased choice
seems the easiest to justify. For example, decision makers may anticipate less criti-
cism of a decision if they choose the middle ground between two extreme values
(see Section 14.5.3.b) or if they take into account all of the information presented,
including irrelevant information that reduces diagnosticity (see Section 8.5.3).
Finally, it should be noted that in some situations where accountability
detracts from the quality of a decision, the decision maker may actually be pursu-
ing other goals—for example, a politician may wish to satisfy members of her
political base, or members of a group may wish to manage their relationships
with each other at the expense of full deliberation of a wide range of available
policy alternatives.
This page intentionally left blank
20. conclusion
Learning from Experience

In Section 1.7, we inquired into the nature of expertise—particularly the ways

that experts integrate deliberative and intuitive processes into their judgments
and decisions. And much of the book has been designed to provide the academic
foundation for lawyers and policy makers to develop their own professional
expertise in problem solving and decision making. This concluding chapter out-
lines what we hope you have learned in the classroom, and asks how you can
develop and hone your expertise beyond the classroom.

20.1 learning in the classroom: the foundations of

problem-solving expertise

The core analytic attributes that underlie problem-solving and decision-making

expertise include:
• clarity about interests and objectives—your own, your clients’, and your
constituents’;
• creativity in considering a problem in alternative frames;
• creativity in generating plausible alternative solutions that solve the prob-
lem and best satisfy the interests at stake;
• strategic planning—the ability to chart a course from wherever you are to a
satisfactory solution;
• creativity and rigor in analyzing correlation and causation—creativity in
developing hypotheses and methods for testing them, and rigor in apply-
ing the tests.
This book introduces some basic concepts and techniques for problem solv-
ing and decision making, including:
• A deliberative process checklist—essential for novices and often useful
even for experts.
• Statistics, which provide a formal approach to hypothesis testing. Statistics
is the empiricist’s most fundamental tool, and an honest policy maker’s
best friend.
• Expected utility theory, which provides a formal approach to decision
making, and a couple of easy-to-use tools: the subjective linear model for
value-based decision making and decision trees for decision making under
risk. Whatever its limits, expected utility theory sets a baseline for rational
choice—strong enough as to require a justification for departures from it.
632 problem solving, decision making, and professional judgment

We have also devoted much attention to errors that attend intuitive decision
making—the so called “heuristics and biases” research agenda in the field of
judgment and decision making (JDM) and the insights of its offspring, behav-
ioral economics. And we have described methods for helping avoid these errors,
to the limited extent possible. Beyond the core domain of JDM, we have also
looked at the dynamics of social influence in the belief that lawyers and policy
makers are called upon both to influence others and to protect clients and citi-
zens from pernicious influence.1
How does the academic study of these matters bear on your development
of expertise as lawyers and policy makers? Expertise depends as much on intu-
ition as analysis. Yet, as the JDM and social psychology literature demonstrate,
intuition is extremely vulnerable to biases, framing, and influence. How—as the
JDM scholar Robin Hogarth asked—does one educate intuition?
The scientific method and statistics play important roles in the process. They
suggest that even when one-off intuitive judgments must be made rapidly on the
spot, they should be tested in retrospect—in aggregate, where possible—so you
can see how well you fared and learn how to improve. The scientific method and
statistics help you avoid some of the recurrent errors that we have examined.
They press us to seek disconfirming evidence: a contingency table requires you
to fill in the cells where there is no result. Statistics counter the availability and
representativeness heuristics by treating sample size as an essential determinant
of statistical significance. It counters overconfidence by requiring that statistical
hypothesis testing state the acceptable margin of error.
We believe that developing the systematic habits of thought inherent in delib-
erative decision making improves subsequent problem solving done at the intu-
itive end of the spectrum, or at least facilitates reflective monitoring of intuitive
judgments. We find a compelling parallel in Constantin Stanislavski’s descrip-
tion of an actor’s preparation:2
One cannot always create subconsciously and with inspiration. No such genius
exists in the world. Therefore our art teaches us first of all to create consciously and
rightly, because that will best prepare the way for the blossoming of the subcon-
scious, which is inspiration. The more you have of conscious creative moments in
your role, the more chance you will have of a flow of inspiration.

20.2 learning from experience

Gary Blasi, from whom we borrowed our protagonist, Luis Trujillo, says that he
“has acquired a significant body of knowledge—about opposing lawyers, about

1. Robin Hogarth, Educating Intuition (Chicago: University of Chicago Press, 2001).

2. Constantin Stanislavski, An Actor Prepares 15 (Elizabeth Reynolds Hapgood
trans., New York: Routledge, 1989).
 conclusion 633

trial judges, about the likely consequences of certain actions—from his many
previous interactions with other lawyers, other judges.” It would be more accu-
rate to say that Trujillo, like any other professional, has had the opportunity to
gain this knowledge. For example, while Trujillo’s observation (in Chapter 1)
about the judge’s reception to motions for summary judgment may well be cor-
rect, it may also be based on a few vivid personal experiences or on settled
wisdom with little empirical foundation.
While experience is inevitable, learning from experience is not. The last item
in our deliberative checklist in Section 1.4.1 included monitoring the outcome of
the decision in order to learn from it. As Gary Klein writes, experts learn from
experience in several ways:
• They engage in reflective practice, so that each opportunity for practice has
a goal and evaluation criterion.
• They compile an extensive experience bank.
• They obtain feedback that is accurate, diagnostic, and reasonably timely.
• They review experiences to derive new insights and learn from mistakes.
Personal characteristics play a role as well. Mark Kelman writes:
People who might be described as “open-minded” rather than dogmatic, especially
in the sense that they accept the possibility that their own thinking is fallible and
that they feel obliged to evaluate the quality of arguments without much regard to
their predispositions about how an issue ought to be resolved, are less prone to
make many of the errors that . . . researchers have identified as “biases.” If one
wants to get a quick, intuitive feel for the sorts of dispositions one might be trying
to measure here, think about how two subjects might answer the questions: “People
should always take into consideration evidence that goes against their own beliefs”
and “No one can talk me out of something that I know is right.”3

The ability to learn from experience depends not only on the expert’s personal
qualities and stance toward learning, but on the learning environment in which
he or she works. Learning requires reliable feedback and requires using that
feedback advertently. Robin Hogarth distinguishes kind learning structures, in
which people receive good feedback, from wicked environments, in which feed-
back can be misleading.4 Determinants of the quality of the learning environ-
ment include the ratio of relevant to irrelevant or random feedback, and how
exacting or lenient the system is with respect to incorrect judgments.5 With
respect to the speed and systematicity of feedback, consider the difference

3. Mark G. Kelman, The Heuristics Debate: Its Nature and Its Implications for
Law and Policy (Oxford University Press, forthcoming 2010), citing Keith Stanovich,
Who Is Rational? (Mahwah, NJ: Lawrence Erlbaum 1999).
4. Hogarth, supra at 98.
5. Id. 88–89.
634 problem solving, decision making, and professional judgment

between a meteorologist predicting tomorrow’s weather and one predicting

changes in climate as a result of global warming.
Recurring to Section 8.5, you can think about learning from experience in
terms of Bayesian updating. You start with a set of beliefs—the prior probabili-
ty—based on what you have learned in school, through mentorship, prior
experience, and hunches. Experience provides new data, based on which you
revise your beliefs. A kind learning environment provides highly diagnostic
feedback.
Unfortunately, the learning environments for many lawyers and policy
makers tend toward the wicked. Granted that a lawyer who specializes in han-
dling a limited number of issues—say, disability or personal injury claims—has
many opportunities to get feedback, most lawyers deal with a pretty wide variety
of situations. Because of the large number of exogenous variables affecting most
policy decisions, the frequent disconnect between a good process and a good
outcome, and the delay in learning what the outcome actually was, policy makers
often get no useful feedback at all.

20.2.1 The Focused Feedback of Mentorship

An expert’s mentorship of a novice provides a “kind” learning environment in
which the novice engages in practice with the expert’s feedback and corrections.
One of the most reflective and eloquent descriptions of the acquisition of expertise
is Mark Twain’s account of becoming a riverboat pilot under the tutelage of the
experienced pilot, Horace Bixby. In Life on the Mississippi, Mark Twain writes:6
It was plain that I had got to learn the shape of the river in all the different
ways that could be thought of—upside down, wrong end first, inside out,
fore-and-aft, and thrortships,—and then know what to do on gray nights
when it hadn’t any shape at all. So I set about it. In the course of time I began
to get the best of this knotty lesson, and my self-complacency moved to the
front once more. Mr. Bixby was all fixed, and ready to start it from the rear
again. He opened on me after this fashion—
“How much water did we have in the middle crossing at Hole-in-the-Wall,
trip before last?
I considered this an outrage. I said—
“Every trip, down and up, the leadsmen are singing through that tangled
place for three quarters of an hour on a stretch. How do you reckon I can
remember such a mess as that?”
“My boy, you’ve got to remember it. You’ve got to remember the exact spot
and the exact marks the boat lay in when we had the shoalest water, in every
one of the five hundred shoal places between St. Louis and New Orleans; and

6. Mark Twain, Life on the Mississippi 57–64 (New York: Penguin Classics, 1883:
1962).
 conclusion 635

you mustn’t get the shoal soundings and marks of one trip mixed up with the
shoal soundings and marks of another, either, for they’re not often twice
alike. You must keep them separate.” . . .
[Twain sees what he believes to be a dangerous reef, and Bixby orders him
to run over it.]
As it disappeared under our bows, I held my breath; but we slid over it like oil.
“Now don’t you see the difference? It wasn’t anything but a wind reef. The
wind does that.”
“So I see. But it is exactly like a bluff reef. How am I ever going to tell them
apart?”
“I can’t tell you. It is an instinct. By and by you will just naturally know one
from the other, but you never will be able to explain why or how you know
them apart.”
It turned out to be true. The fact of the water, in time, became a wonderful
book—a book that was a dead language to the uneducated passenger, but
which told its mind to me without reserve, delivering its most cherished
secrets as clear as if it uttered them with a voice.

20.2.2 After Mentorship: Learning from Mistakes

Experience is a great teacher but the tuition is high.

Some professions provide structured peer learning opportunities even after

the mentorship phase; the morbidity and mortality conferences at hospitals are
paradigmatic. To a large extent, however, professionals’ continuing learning
must come from their own reflections on experience. This is certainly true of
much lawyering and policy making. So how do we revise our views based on
experience? And particularly, how do we learn from mistakes?
In earlier chapters we saw that once people have committed to a course of
action, there arise psychological barriers to changing that course. Beliefs, once
formed, have a way of sticking, even in the face of evidence that they are
mistaken.
Recognizing the mistakes and biases of others is often very easy. It may be
obvious that Jim is far too optimistic about his investment portfolio, or that Anne
may have fallen prey to the availability heuristic in deciding that driving across
the country is safer than flying. Recognizing mistakes in our own reasoning,
however, can be much more difficult. Building on Lee Ross’s concept of naïve
realism (Section 9.8), Emily Pronin, Thomas Gilovich, and Lee Ross have coined
the term bias blind spot to describe people’s readiness to see others’ biases while
believing that their own thinking is relatively bias-free.7

7. Emily Pronin, Thomas Gilovich, and Lee Ross, Objectivity in the Eye of the Beholder:
Divergent Perceptions of Bias in Self Versus Others, 111 Psychological Review 781–99 (2004).
636 problem solving, decision making, and professional judgment

In concluding the book, let us review some common ways in which biases
hinder learning from experience.
Confirmation biases. As we saw in Section 10.3, people tend to seek out evi-
dence that supports their existing (or preferred) hypothesis and avoid looking for
evidence that contradicts it. Closely related is the biased assimilation of evidence—
our tendency to view evidence that favors our beliefs more favorably and less
critically than evidence that challenges them. Belief perseverance is a cousin of
these phenomena. Once we have formed an initial belief about any fact or value,
we tend to process new information in light of our belief. If we have accumulated
a network of supporting beliefs, it can become extremely difficult to reverse
course, even when new evidence suggests that it would be reasonable to do so.
What Jerome Bruner and his colleagues have termed the “thirst for confirm-
ing redundancy”8 has motivational as well as cognitive roots. We get invested—-
sometimes literally—in our own ideas and practice. The possibility of their being
wrong can be a blow to our pocketbooks as well as our egos and reputations.
Poor feedback mechanisms. Learning from mistakes depends on feedback.
Sometimes it is difficult to get feedback just because the data are not available,
or not available quickly or in sufficient quantity or in a form conducive to learn-
ing. We are also prone to distortions in acquiring data—because of stress, sche-
mas, and expectations—and to distortions in retention and retrieval—because of
intervening events or just plain forgetting. We tend not to notice omissions (the
dog that didn’t bark in the nighttime). If these are largely cognitive phenomena,
there’s a strong motivational one as well: avoiding feedback to minimize regret.
Under- and over-generalization. Data are often presented in ways that make it
difficult to generalize, to connect the dots. On the flip side, we readily overgener-
alize, seeing relationships where none exist. This results from basic statistical
errors—misunderstandings of probability, randomness, independence, sample
size, regression to the mean—exacerbated by the illusion of control and the
availability heuristic.
The flipside of the illusion of control is the self-fulfilling prophecy, where our
empirical judgment actually affects the outcome. An example is the “Pygmalion”
effect, where students who are predicted to perform better actually do so. There’s
a story about a nineteenth-century physician who didn’t wash his hands between
examining patients for a particular infectious disease and was therefore very
good in predicting that the next patient he examined would have the disease.
Hindsight bias—the erroneous belief that we “knew it all along”— is also a
barrier to learning from experience. Because many decisions are made in condi-
tions of uncertainty about the outcomes, and because improving our

8. Jerome Bruner, Jacqueline Goodnow, and George Austin, A Study of

Thinking (New York: Wiley, 1956).
 conclusion 637

decision-making prowess often depends on improving our predictions, each

time we think that we “knew it all along” we lose an opportunity to improve.
Blame-avoidance strategies provide tremendous barriers to learning from
experience. The phenomenon known as attribution bias leads us to attribute suc-
cess to our ability and our failures to bad luck, our good deeds to our superior
character and our bad deeds to compelling circumstances.”9 As the Duke of
Wellington is said to have quipped, “victory has a thousand fathers; defeat is an
orphan.”
Defensiveness. Even if you don’t kill the messenger, acting defensively will at
least signal him not to bring you bad news again. We know a few people who are
genuinely nondefensive—who will accept a criticism with dispassion or even
invite it. Here’s an alternative for those of us who don’t have this quality:
Whatever your inner feelings when someone criticizes you, act nondefensively.
Not only will you get a lot more information that way, but you may actually
become less instinctively defensive.
***

The overarching hypothesis of this book is that academic study can lay the
foundation for developing expertise in problem solving, decision making, and
professional judgment on the job. If many aspects of this hypothesis remain to be
tested, it nonetheless seems like a pretty good bet. At very least we hope that you
have learned something interesting about human behavior, including your own.

9. Lee Ross, The Intuitive Psychologist and His Shortcomings: Distortions in the Attribution
Process, in L. Berkowitz (ed.), Advances in Experimental Social Psychology (Vol. 10)
173–220. (New York: Academic Press, 1977).
This page intentionally left blank
index

Figures, notes, and tables are indicated by f, n, and t following the page numbers.

A defined, 19n26
Abdul Latif Jameel Poverty Action dread, discounting for, 521
Lab (J-PAL), 198, 210 evaluability and framing, 511–12
Above-average effect, 408 exchange principle, 508–11
Absolute vs. relative wealth and familiarity, 499
happiness, 360 feelings, 491–94, 491f
Abu Ghraib prison, 539, 553–54 high anxiety, 498–99
Accompanied groups, 444 insurance as protection against
Accountability, effects of, 628–29 worry, 515–16
Ackerman, Frank, 383 judgments of risk and benefit, 502–3
Action biases, 500–501 maximin principle, 508–11
Action vs. inaction, 430–33 mitigating factors, 497t
Actor-observer effect, 332 natural origins, 500
Adams, James, 87–88 outrage, 498–99
Adaptation and “ordinization” perceptions, cultural
neglect, 400–401 determinants of, 503–5
Ad hoc construction of precautionary principles, 521
preferences, 386–88 probability neglect, 514–19
Adjustment, 216. See also Anchoring, rational decision making, 367–68
adjustment, and activation “robust satisficing,” 508–11
Adorno, Theodor, 550 single action bias, 515
Affect, 491–522 uncertainty vs. risk, 505–11
acceptability of risks, 498–99 valuation by calculation or feeling, 513
action biases, 500–501 vividness, 495–97
affect heuristic, 367–68 voluntariness, 499
aggravating factors, 497t worry, 514–19
ambiguity vs. risk, 505–11 Affect heuristic, 367–68
anchoring and risk, 519 Affection, 347–48
anxiety, 493–94 Affective forecasting, 398
assessing risks, 520 Agentive language, 575–76
availability heuristic, 495–97 Aggravating factors, 497t
betrayal aversion, 502–3 Aggregation of views without group
catastrophic harm, 521–22 process, 597–600
certainty equivalent of lottery, 508f Agreement effect, 283
cognitive and affective processing Ainslie, George, 415
of probabilities, 513–14 Alhakami, Ali, 503
continuity principle, 508–11 Allais, Maurice, 489
controllability, 500 Allais’ paradox, 489, 490t
cost-benefit analysis (CBA), 520–22 Allhoff, Fritz, 379
cultural cognition, 504 Alter, Adam, 293
640 index

Alternative hypothesis, 163, 165f expertise and, 73–74

Alternatives, choice of, 91–114 fields, 77–80, 80f
checklists, 94 gatekeepers, 77, 79
compensatory procedures, 98–101 individuals, 77–80, 80f
consequences, 91–92 individual traits, 77
decision-making processes, spectrum intellectual skills, 76
of, 93–96. See also Decision-making intrinsic task motivation, 76–77
processes marketplace of ideas, 72
delegation, 96 propulsion theory, 78
deliberative strategies, 96–98 self-governance skills, 76
elimination by aspects, 96–98 skills required for, 76–77
guidelines, 92 deliberative processes in, 13
heuristics, 93–94, 98 guidelines, 90
intuition, 93 pickers, 61–62
noncompensatory strategies, 96–98 problem-solving frames of mind, 61–62
presumptions, 94–95 scenario planning, 80–86
quantitative decision-making availability heuristic, 85
strategies, 101 brainstorming, 85
random processes, 96 defined, 81–84
reason-based decision-making drawbacks of, 85–86
strategies, 99–101 Global Business
routines, 94 Network (GBN), 81–84
rules, 94 key principles of, 84–85
rules of thumb, 94–95 long view, 84
small steps, 95–96 Monitor Group, 81–84
standards, 95 multiple perspectives, 85
subjective linear models, 102–11. outside in thinking, 85
See also Subjective linear models STEEP trends, 85
transparency, 112 use of, 81–84
Alternatives, generation of, 61–90 solutions, frameworks for
choosers, 61–62 generating, 63–67
constraints basic principles, 63
cognitive process, 86–87 constraints, 67
personal, 89 convergent thinking and, 66–67
ping pong ball in pipe problem, 87f different kinds, 63–64
practical, 88–89 divergent thinking and, 66–67
social, 87–88 information-buying, 64–66
creativity, 67–80 lateral thinking, 63
barriers to, 86–89 nine dot problem, 67f
bisociation and, 70, 71f possibilities, 67
cognitive flexibility, 77n48 studying people, 64
conceptual blending, 73 time-buying, 64–66
confluence approach to, 76–77 Amabile, Theresa, 69, 73
convergent and divergent Ambiguity. See Risk; Uncertainty
thinking and, 74–76 Ambiguity aversion, 507
critical thinking and, 73–74 Amygdala, 22, 327, 327n46
cross-fertilization and, 70–73 Anchoring, adjustment, and
defined, 69–70 activation, 216, 267–72, 472–73, 519
domains, 77–80, 80f cross-modal anchors, 270
 index 641

debiasing, 272 scenario planning, 85

explanations, 270–71 vividness and inference, 257–58
extreme and untrustworthy vividness and media, 258–60
anchors, 270 Ayres, Ian, 181
insufficient adjustment, 267
jury awards, 269–70 B
negotiations, 268–70 Babcock, Linda, 248
Anderson, Craig, 287 Baker, Lynn, 405
Angner, Eric, 402–3 Banaji, Mahzarin, 338, 577–78
Annals of Medicine: The Checklist Bargh, John, 315, 328
(Gawande), 94 Baron, Jonathan, 501, 505
Anxiety, 493–94 Barton, Tom, 63
Apparent authority, obedience to, 550–56. Base rate, 226–28, 233–35
See also Social influence Bayes, Thomas, 228
Arceneaux, Kevin, 195 Bayesian statistics, 225–35
Arguments base rate, 226, 233–35
framing of, 574–75 Bayes’ theorem, 228–29, 233–34
persuasion through, 571–74 diagnosticity, 229–33
Ariely, Dan, 349, 352, 353, 417, 427, 586 frequentism, 225
Aronson, Joshua, 329 intuition, 233–35
Arrow, Kenneth, 457 likelihood and likelihood ratio, 226, 230
Asch, Solomon, 541–44 odds version, 239
Assimilation, biased, 282–83, 636 posterior probability, 227
Asymmetrical dominance, 450–51 prior probability, 226, 277
Attention allocation, 321–22 sensitivity, 230
Attentiveness effect, 383 structure of problems, 227–28
Attorneys. See Lawyers subjectivism, 225
Attractiveness, physical, 564–66 tree structure, 227–28, 227f, 238f
Attributes Bayes’ Theorem, 228–29, 233–34
evaluation of, 107–8, 108t Bazerman, Max, 174, 243–44, 626
major attributes, 105–6 Because It Is There: The Challenges of
quantitative description of, 109t Mountaineering . . . for Utility Theory
weighing, 110–11, 110t, 111t (Loewenstein), 346
Attribution. See also Behavior, influencing, 571–94
Fundamental attribution error argument, persuasion through, 571–74
causal, 323–24 channel factors, 581
The Authoritarian Personality citizens’ decision making, 584–94
(Adorno), 550 complex choices, simplifying and
Authority, obedience to, 550–56. See also structuring, 591–92
Social influence concreteness, 572
Automatic framing, 40–42 credibility, 572–73
Availability cascades, 548–49 default options, 591
Availability heuristic, 51–52, 162, 253–61 detection behaviors, 589
causes, 51 emotionality, 573
concreteness, 258 experienced utility, mapping
direct experience, 258 choices to, 590–91
emotional interest, 258 feedback, 590–91
metacognition and, 254–55 framing choices, 588–94
risk, role of affect in, 495–97 gains and losses, 574–75
642 index

Behavior, influencing (cont.) bounded awareness, 244

health behaviors as gains distinctiveness heuristic, 252
or losses, 589–90 egocentrism, 261–64
incentives, 586–87 expectations, 244–49
libertarian paternalism, 588, 592–94 false consensus effect, 264
linguistic form, 575–81 familiarity, 251
agentive vs. passive, 575–76 focusing failure, 244
metaphors, 579–81 hostile media phenomenon, 246–48
nouns, 577–78 information processing,
verbs and causal inference, 576–77 stages of, 241–42, 242f
mandated choice, 594 interests, 244–49
prevention behaviors, 589 interpretation, 245–46
priming, 581–84 memory and anticipation, 354–56
recuperative behaviors, 589 memory binding, 250
risk counseling, 584–86 metacognition, 252–60
simplicity, 572 naïve realism, 264–65
stories, 573–74 reactive egoism, 262–63
testable credentials, 572 retention, 242–43, 249–52
unexpectedness, 572 retrieval, 242–43, 249–52, 261–64
Behavioral confirmation schemas, 244–49
effect, 325–29, 327f settlements, 248–49
Behavior-behavior links, 326, 328 theories, 244–49
Belief perseverance, 287–88, 636 present bias, 412, 413f
Ben Haim, Yakov, 509–10 prior expectations and theories,
Benjamin, Jimmy, xxvii resistance to modification of, 277–89
Bentham, Jeremy, 346 judgments, coherence in
Berger, Jonah, 582 making, 288–89
Bernoulli, Daniel, 360, 419, 474 processing and judging
Bernoulli’s curve, 360, 361f information, 267–302. See also
Berns, Gregory, 356, 543 Anchoring, adjustment, and
Betrayal aversion, 502–3 activation
Between-subject vs. within-subject calibration, 291
experiments, 207 confirmation bias, 277–89
Bias. See also specific types agreement effect, 283
action, 500–501, 515 bidirectional reasoning, 289
assimilation bias, 282–83, 636 motives underlying, 280
blind spots, 635 negative hypothesis testing
confirmation bias, 609, 615, 618–19, 636 and, 285
counteracting, 588–94 positive hypothesis testing
debiasing social judgment, 334–40 and, 283–86
debiasing strategies, 289 self-affirmation theory, 281, 281n26
heuristics and, 252–60, 618–21 hindsight bias, 272–77
hindsight bias, 272–77, 636–37 debiasing, 276
impact, 397–401, 403–4 foresight condition, 272–73
lawyers as cognitive counselors, 523–25 legal system responses to, 276–77
perception and memory, 241–65 risk/benefit analysis, 275–76
acquisition, 242–44, 261–64 intuition, analysis, and
availability heuristic, 252–60. See also experts’ role, 293–301
Availability heuristic affirmative action example, 298
 index 643

baseball example, 295–96 Brenner, Patricia, 26–27

bootstrapping, 301 Bricolage, 386–88
calibration, 300 Bridges, George, 324
clinical vs. statistical Brown, Rupert, 329, 627
prediction, 293–98 Bruner, Jerome, 17, 219, 244, 636
combining indicators, 300 Brunswik, Egon, 117
expert intuitions as Bryan, Christopher, xxvii, 578
safety valve, 301 Budgeting, 439–41
relevant information, Buehler, Roger, 405–408
underuse of, 300 Business Rules of Thumb
reliability, 300 (Godin & Conley), 95
validity, 300 Bystander effect, 544–45
intuition, overcoming, 293
overconfidence, 289–93 C
prior expectations and Calibration, 291, 300
theories, resistance to Campbell, Donald, 201, 204
modification of, 277–89 Cancer clusters, 149–52
belief perseverance after Cantril, Hadley, 245
evidential discrediting, 287–88 Carmon, Ziv, 353, 427
biased assimilation, 282–83 Cascades, emergence of, 547–49
debiasing strategies, 289 Catastrophic harm, 521–22
motivated skepticism, 280–81 Categorical variables, 163–65
projection, 401–4 Categories and categorization. See
shared information, 610–12 Schemas; Stereotypes and stereotyping
single action bias, 515 Causal attribution, 323–24. See also
social schemas, 304–5 Fundamental attribution error
status quo bias, 419–28 Causal diagrams, 187, 187f, 195f
uncontrolled selection, 189 Causal inference, 576–77
Biased assimilation, 282–83, 636 Causes
Biddle, Gary C., 267 availability heuristic and, 51
Bidirectional reasoning, 289 identification of, 45–52
Birke, Richard, 221, 558 attributing causation, 51–52
Bisociation and creativity, 70, 71f distinctions, 50–51
Blame-avoidance, 637 hypotheses for further testing, 51
Blasi, Gary, 632 problem defining, 48–49
Blind spots, 635 representativeness heuristic and, 52
Blumenthal, Jeremy, 513 Central tendency, 153–65
Bodenhausen, Galen, 323 Certainty effect, 483–85, 484t, 489–90
Bodner, Rodnit, 416 Certainty equivalent of lottery, 508f
Boles, Terry, 433 Channel factors, 581
Bootstrapping, 301 Chapman, Gretchen, 271
Boroditsky, Lera, 575–76 Checklists
Bounded awareness, 244 alternatives, choice of, 94
Bounded rationality, 22–24 framing problems, 44–45
Brainstorming, 85 interests and objectives,
Braman, Donald, 504 identification of, 44–45
Brehmer, Berndt, 23 Chen, Kay-Yut, 598–99
Brennan, William, 72 Chen, Serena, 328
Brenner, Lyle, 444, 445 Chi-squared test, 146
644 index

Choi, Stephen, 182 Commitment, escalation of, 620

Choices, consequences, and Common knowledge effect, 604–12
trade-offs, 343–84. See also Comparative loss aversion, 444
Alternatives, choice of Compensation and subjective utility, 348
completeness, dominance, and Compensatory procedures, 99–102
trade-offs, 369–72 Complement rule, 124, 224–445
complex choices, simplifying and Completeness in rational
structuring, 591–92 decision making, 369–72
cost-benefit analysis (CBA), 373–84. Compliance with authority, 550–56
See also Cost-benefit analysis Compliments, 564
mandated choice, 594 Compromise effect, 451–53
overload, 354, 442–43, 446–48 Conceptual blending, 73
rational decision-making, Conceptual Blockbusting (Adams), 87
criteria for, 366–73 Concreteness, 258, 572
consequential rationality, 368 Conditional probabilities,
constitutive incommensurability, 370 calculation of, 211–13
corporate cost-benefit analysis Conditioning, 403
involving human safety, 371 Condorcet Jury Theorem, 598
decomposability, 372 Confirmation bias, 609, 615, 618–19, 636
ecological rationality, 373 Confluence approach to generating
invariance, 372 alternatives, 76–77
irrelevant alternatives, Conformity, 540–49. See also
independence from, 372 Social influence
parsimoniousness, 372 Confounding effects, 188–89
procedural rationality, 368–69 Confusion of the inverse, 235–38
subjective expected utility theory, Conjoint analysis, 111n22
368–73. See also Subjective utility Conjunctive events, 213–17. See also
transitivity, 372–73 One-time events
Choosers, 61–62 Conjunctive fallacy, 217–22, 220t
Chugh, Dolly, 243–44 Conley, Chip, 95
Cialdini, Robert Connecticut crackdown on
context and suboptimal decisions, 454 speeding, 201–4
descriptive norms, 547 Connolly, Terry, 501
door-in-the-face technique, 567–68 Consequences. See Choices,
low-ball technique, 557 consequences, and trade-offs
obedience, 555 Consequential rationality, 368
social proof, 540, 546 Consistency and commitment tendency.
Citizens’ decision making, 584–94. See Escalation of commitment
See also Behavior, influencing Consolation hypothesis, 348
Classroom learning, 631–32 Constitutive incommensurability, 370
Clean Air Act, 34, 382 Constructed controls, 192–93
Clinical vs. statistical prediction, 293–98 Consumption effect, 355
Coefficient of determination, 176 Context and suboptimal decisions, 454
Cognitive and affective processing of Context dependence, 450–55
probabilities, 513–14 Contingency
Cognitive dissonance theory, 366 tables, 128–30, 129t, 147t, 148t
Cognitive flexibility, 77n48 Contingent valuation, 421, 455–56
Collective learning, 301 Continuity principle, 508–11
Combining indicators, 300 Contrast effect, 355
 index 645

Control, illusion of, 137–39 gatekeepers, 77, 79

Controllability, 499 individuals, 77–80, 80f
Convergent and divergent individual traits, 77
thinking, 13–14, 15f, 66–67, 74–76, intellectual skills, 76
600–601, 601f intrinsic task motivation, 76–77
Conversational norms, 610 marketplace of ideas, 72
Cooper, Joel, 326, 328 propulsion theory, 78
Cooperation, 563–64 self-governance skills, 76
Correlation vs. causation, 185–87, 186f skills required for, 76–77
Cosmides, Leda, 373 thinking, critical, 73–74
Cost-benefit analysis thinking, defined, 77n48
(CBA), 373–84, 455–57, 520–22 Credibility, 572–73
ancillary costs and benefits, 382–83 Critical thinking, 73–74
attentiveness effect, 383 Croson, Rachel, 547
benefit calculations, 375–76 Cross-fertilization, 70–73
cost calculations, 376 Cross-modal anchors, 270
cost-effective analysis (CEA), 376 Crowds, wisdom of, 597–600
decision-making processes, 455–57 Csikszentmihalyi, Mihály, 73, 79
difficult-to-measure benefits, 381–82 Cultural differences
example, 375–76 cognition, 504
frames, power of, 455–57 subjective utility, 366
human safety and corporations, 371
mortality risks, 376–81 D
outcomes and monetization, 382 Daily experiences vs. general sense of
psychological health, well-being, 359–60
qualifying effects on, 381 Damasio, Antonio, 20, 492–95
quality-adjusted life years Darley, John, 318–19, 544
(QALY), 379–81 Darwin, Charles, 99–102, 100t, 473
rational decision making, 371 Dasco, Cliff, 509–10
risk and affect, 520–22 Deactivation and social schemas, 313
value and limits of, 383–84 deBono, Edward, 63
value of statistical life (VSLY), 377–78 Decision-making processes, 3–31.
value of statistical life years, 378–79 See also Problem-solving
Counterfactual outcomes, 429–30 above-average effect, 408
Creativity in problem solving, 61–90. accompanied groups, 444
See also Alternatives, generation of action vs. inaction, 430–33
barriers to, 86–89 adaptation and “ordinization”
bisociation and, 70, 71f neglect, 400–401
cognitive flexibility, 77n48 ad hoc construction of
conceptual blending, 73 preferences, 386–88
confluence approach to, 76–77 affective forecasting, 398
convergent and divergent thinking alternatives, choice of, 93–96. See also
and, 74–76 Alternatives, choice of
critical thinking and, 73–74 asymmetrical dominance, 450–51
cross-fertilization and, 70–73 bricolage, 386–88
defined, 69–70 budgeting, 439–40
domains, 77–80, 80f budgeting time, 441
expertise and, 73–74 characteristics, 112–13, 113t
fields, 77–80, 80f choice overload, 442–43, 446–48
646 index

Decision-making processes (cont.) crowds, wisdom of, 597–600

comparative loss aversion, 444 Delphi Technique, 625
complexities of, 385–457 devil’s advocacy, 623–24
compromise effect, 451–53 dialectical inquiry, 623–24
conditioning, 403 divergent and convergent
context dependence, 450–55 thinking, 600–601, 601f
contingent valuation, 421, 455–56 fundamental attribution error, 620–21
cost-benefit analysis. See Cost-benefit gain/loss framing effects, 620
analysis groupthink, 596, 600–604, 601f,
counterfactual outcome, 429–30 621–22
decoupling and consequences, 439 “hidden profile” paradigm, 606–8
diminishing sensitivity, 420 improvement of, 621–25
dissonance reduction, 403 judgmental errors, 619
downsides, 442–55 nominal groups, 625
emotion, predicted vs. experienced, 399f optimization, 627–28
endowment effect, 419–28 overconfidence, 618
environmental and participation, 617
public goods, 455–57 persuasive argument theory, 615
equity premium puzzle, 420 polarization, 612–17
evaluability, 388, 389t predictable surprises, 625–27
experienced utility, 396 premature convergence, 623–24
extremeness aversion, 451–53 probabilistic sampling, 608–9
feedback, 433–34, 439 rationale construction, 616
401(k) plans, 426–27 shared information, bias for, 610–12
frames, 419–57. See also Frames social comparison and identity
and framing maintenance, 614–15
future well-being, predicting, 395–404 social corroboration effect, 614
adaptation, 400–401 social validation and mutual
emotional reactions, 397–401 enhancement, 609–10
focalism, 398–400 stepladder technique, 624–25
impact bias, 397–401, 403–4 vigilant appraisal, 621
long-run preferences, 403 grouping, effects of, 444–45
“ordinization” neglect, 400–401 guidance and influence, 535–37
projection bias, 401–4 habit formation, 402
gains, segregation of, 438 hedonic framing, 438
gift-giving, 440 homo economicus model, 385
group decision making, 595–629 hot/cold empathy gap, 403
accountability, effects of, 628–29 impact bias, 397–401, 403–4
aggregation without group income accounting, 440
process, 597–600 intertemporal choice, 410–18. See also
biases and heuristics, 618–21 Intertemporal choice
cognitive/informational accounts invariance, 392
of polarization, 615–16 isolated groups, 444
commitment, escalation of, 620 isolation effect, 400
common knowledge effect, 604–12 keeping options open after
Condorcet Jury Theorem, 598 decisions, 448
confirmation long-run preferences, 403
bias, 609, 615, 618–19, 636 loss aversion, 419
conversational norms, 610 losses, integration of, 438
 index 647

meaning, 437 well-being, predicting, 395–404

mental-accounting, 437–42 Decision trees, 461–71, 463f, 463t, 466f,
mutability, 429 469f, 470t, 471f
myopic loss aversion, 420 Declining marginal utility, 360–61
open options after decisions, 448 Decomposability, 372
option devaluation, 443–49 Decoupling decisions and
options, effect on decisions, 448–49 consequences, 439
overoptimism, 405–8 Defensiveness, 637
above-average effect, 408 Definitions
detrimentality of, 408–10 affect, 19n26
overview, 341–42 affect heuristic, 367–68
planning fallacy, 405 alternative hypothesis, 142
positive illusions, 404–10 anchoring, adjustment, and
precommitment, 437 activation, 267–72
predicted utility, 396 availability heuristic, 252–53
preference reversals between prices choosers, 61
and choices, 392 convergent thinking, 66
prior decisions and finality, 437 creativity, 69–70
projection bias, 401–4 divergent thinking, 66
prospect theory, 208, 419–28, domains, 79
420f, 476–87 fields, 79
psychological immune system, 400 groupthink, 600
psychological ownership, 423 null hypothesis, 142
ranked groups, 444 pickers, 62
reference points, 419 posterior probability, 227
refinement, 403 primary emotions, 354
regret and suboptimal learning priming, 315–16, 581–84
processes, 434 prior probability, 227
regret aversion, 428–34 representativeness heuristic, 218
action vs. inaction, 430–33 scenario planning, 81–84
counterfactual outcomes, 429–30 schemas, 14, 18n20
feedback, anticipation of, 433–34 scripts, 18
suboptimal learning thinking, 77n48
processes and, 434 Type I errors, 142–43
regret avoidance, 427, 436 Type II errors, 142–43
reputation, 436 Deliberation, limits of, 22–24
revealed preference, 455 Deliberative processes, 11–13
risk, 459–90. See also Risk Deliberative strategies, 96–99
satiation, 402 Delphi Technique, 625
satisficing, 385 Dependency
self-control, 440 diagrams, 460–61, 461f, 462f, 465f
self-deception, value of, 409 Dependent variables, 147
self-education, 437 De Poot, Christianne, 576–77
social influence, 403 Descriptive norms, 547
status quo bias, 419–28 Design effects, 189–91
strategies, choice of, 386–88 Detection behaviors, 589
suboptimal decisions, 385–86, 441–42 Determinism, 123
sunk costs, 434–37 Devil’s advocacy, 623–24
trade-off contrast, 450–51 Devine, Patricia, 307
648 index

Dewey, John, 33, 36 Ecological rationality, 373

Diaconis, Persi, 141 Economic Growth and Subjective Well-Being:
Dialectical inquiry, 623–24 Reassessing the Easterlin Paradox
Dictator game, 350–51 (Stevenson & Wolfers), 360
Differential attrition, 189 Egocentrism, 261–64
Dijksterhuis, Ap, 395 Elimination by aspects, 96–99
Diminishing sensitivity, 420, 476 Ellsberg’s Paradox, 505–7, 506t
Dion, Karen K., 565 Elster, Jon, 354–56
Dion, Kenneth L., 565 Emery, Robert, 405
Direct experience, 258 Emotion, predicted vs. experienced, 399f
Discounted utility, 411–18 Emotional intelligence, 20
Discount rates, 410–11 Emotionality, 573
Disjunctive events, 222–25 Employment discrimination, 179–80
Dispersion, 154–56, 154f Endowment effect, 419–28
Dispositionism in social Ends, arationality of, 344–47
attribution, 331–33. See also Enron, 545, 626
Fundamental attribution error; Environmental and public goods, 455–57
Situationism in social attribution Environmental Protection Agency, 34
Dissonance reduction, 403 Equity premium puzzle, 420
cognitive dissonance theory, 366 Escalation of commitment, 556–62
Distinctiveness heuristic, 252 foot-in-the-door technique, 557
Ditto, Peter, 280–81 irrational escalation of
Divergent and convergent commitment, 558–59
thinking, 13–14, 15f, 66–67, 74–76, and litigation, 561
600–601, 601f low-ball technique, 557
Domains and creativity, 77–80, 80f organizational determinants of, 560
Dominance axiom, 488–89, 489f project determinants, 559
rational decision making and, 369–72 psychological determinants of, 560.
Donohue, John, 181 See also Confirmation bias
Door-in-the-face technique, 567–68 and salvage value, 560
Dread, discounting for, 521 social determinants of, 560–61
Dreyfus, Hubert, 14, 25, 26, 27, 285 ways of preventing, 561–62
Dreyfus, Stuart, 24, 27, 285 Estimates, 127–28, 159, 161, 162t, 174–75
Dual process models, 20–22, 253, 308. Ethical behavior, 481–82
See also High road vs. low road thinking Evaluability, 388, 389t, 511–12
debiasing, 335–36 Evaluation
social cognition, 334–38, 336f causal diagrams, 187, 187f, 195f
Dummy variables, 180 causation, establishment of, 188–91
Duncan, Birt, 318 confounding effects, 188–89
Dunn, Elizabeth, 399–400 constructed controls, 192–93
Dunning, David, 408 controls and unwanted effects, 191–97
Duration neglect, 357 correlation vs. causation, 185–87, 186f
Dweck, Carol, 578 design effects, 189–91
Dynamically inconsistent differential attrition, 189
preferences, 411–18 experimental design and external
validity, 206–10
E experimental environment, 210
Easterlin Paradox, 360 external validity, 208–10
Ebert, Jane, 354 generalizability, 208–10
 index 649

gross outcomes, 188 Experience, memory, and

Hawthorne Effect, 190 anticipation, 354–56
impact measurement, 185–87 Experienced utility, 351–54, 396, 590–91
incentives, 209 Experimental design and
internal validity, 208 external validity, 206–10
matching, 192, 193n12 Expert intuitions as safety valve, 301
measurement reliability, 189–90 Expertise
measurement validity, 190 creativity and, 73–74
methods, comparison of, 194–97 nature of, 25–29, 31
observational studies, 191 Expert problem frames, 40–42
placebo effect, 190 Experts, 29–31, 298–301
program evaluation, Explanation through narrative, 219–22
examples of, 197–204 External validity, 208–10
prospect theory, 208 Extrapsychic mechanisms, 417–18
Pygmalion Effect, 190–91 Extremeness aversion, 451–53
randomized controlled studies, 191–92 Eyewitness Testimony (Loftus), 242–43
reflexive controls, 193–94
selection biases, 189 F
self-selection, 189 Fairness and subjective utility, 349–51
significance, 205–6 False consensus effect, 264
speeding, Connecticut False positives, 236–37
crackdown on, 201–4 Familiarity, 251, 499, 562
statistical controls, 194 Faulty intuition, 237–38
subjects, 209 Fausey, Caitlin, 575–76
unwanted effects and controls, 191–97 Featherstonehaugh, David, 362
within-subject vs. between-subject Feedback
experiments, 207 citizens’ decision making, 590–91
Exchange principle, 508–11 decision-making processes, 433–34, 439
Existence and option value, 347 experience, learning from, 636
Expectancies, effects of, 90, 190, 317–27 frames, power of, 433–34, 439
Expectancy-behavior links, 326 regret aversion, 433–34
Expectations, 244–49 Feigenson, Neal, 259
Expected return, 57, 57f Festinger, Leon, 366
Expected utility, 459–60 Fischhoff, Baruch, 272
subjective, 368–73 Fiske, Susan, 540, 563
Expected value analysis, 473f Fluency as source of truth and liking
Experience, learning from, 631–37 judgments, 255–56
belief perseverance, 636 fMRI. See Functional magnetic
bias blind spots, 635 resonance imaging
biased assimilation of evidence, 636 Focalism, 398–400
blame-avoidance, 637 Focusing failure, 244
classroom learning, 631–32 Foot-in-the-door technique, 556–57
confirmation biases, 636 Forensic evidence, 236–37
defensiveness, 637 Foresight condition, 272–73
feedback, 636 401(k) plans and framing, 426–27
generalization, 636 Fox, Craig, 221, 558
hindsight bias, 636–37 Frames and framing
mentorship, 634–35 accompanied groups, 444
mistakes, 635–37 action vs. inaction, 430–33
650 index

Frames and framing (cont.) regret aversion, 428–34

arguments, 574–75 regret avoidance, 427, 436
asymmetrical dominance, 450–51 reputation, 436
boundary setting through, 35 revealed preference, 455
budgeting, 439–40 risk, 476–87
budgeting time, 441 schemas and, 34
comparative loss aversion, 444 self, sense of, 437
compromise effect, 451–53 self-control, 440
context dependence, 450–55 self-education, 437
contingent valuation, 421, 455–56 status quo bias, 419–28
decoupling and consequences, 439 steps in, 33, 49t, 50t
deliberative processes, 12 suboptimal decisions, 441–42
diminishing sensitivity, 420 sunk costs, 434–37
endowment effect, 419–28 time horizons, 36
equity premium puzzle, 420 trade-off contrast, 450–51
evaluability and, 511–12 Frank, Robert, 370–71
extremeness aversion, 451–53 Franklin, Benjamin, 100–01
401(k) plans, 426–27 Fraser, Scott, 556
frames, power of, 419–57 Frederick, Shane, 21, 218, 352
framing-by-objectives, 42–43 Frederickson, Barbara, 357–59
gains, segregation of, 438 Free, allure of the, 348–49
gift-giving, 440 Freedman, Jonathan, 556
grouping, effects of, 444–45 Frequencies
hedonic framing, 438 presentation of, 237–38
income accounting, 440 probabilities vs., 524–25
isolated groups, 444 Frequentism vs. subjectivism, 125
lawyers and, 40–42 Frisch, Deborah, 352, 435, 505
loss aversion, 419 Fulton, Katherine, 84
losses, integration of, 438 Functional magnetic resonance
mental accounting, 437–42 imaging (fMRI), 326, 335, 412, 543
multidimensional frames, 42–43 Fundamental attribution error, 332, 620–21
mutability, 429
myopic loss aversion, 420 G
option devaluation, 443–49 Gains and losses
pitfalls, 36–42 behavior, influencing, 574–75
automatic framing, 40–42 framing effects, 620
expert problem frames, 40–42 gains, segregation of, 438
framing by solution, 37–38 health behaviors as, 589–90
legal frames, 40n9 low-probability gains and losses, 482–83
symptoms vs. problems, 38–40 risk and, 489t
precommitment, 437 Galton, Francis, 170, 172, 597–98
prior decisions and finality, 437 Gambler’s fallacy, 139–41
prospect theory, 208, 419–28, Gardner, Howard, 69
420f, 476–87 Gastil, John, 504
psychological ownership, 423 Gatekeepers, 77, 79
ranked groups, 444 Gawande, Atul, 94
reference points, 419 Generalization, 208–10, 636
regret and suboptimal learning Genocide, 550
processes, 434 Gerber, Alan, 195
 index 651

Gershoff, Andrew, 502 Harrison, Richard, 174

Giddens, Anthony, 414 Haslam, Nick, 415
Gift-giving, 440 Hastie, Reid, 220, 573
Gigerenzer, Gerd, 98, 222, Hastorf, Albert, 245
237–38, 373, 473 Hawthorne Effect, 190
Gilbert, Daniel, 354, 364, 397–98, 400 Hayek, Friedrich, 598
Gilovich, Thomas, 140, 432, 582–83, 635 Health behaviors as gains or
Global Business Network (GBN), 81–84 losses, 589–90
Gneezy, Uri, 587 Heath, Chip, 260, 572, 575
Godin, Seth, 95 Heath, Dan, 572, 575
Goldberg, Lewis R., 297 Hedonic adaptation, 364–66
Goldstein, Noah, 547 Hedonic framing, 438
Gollwitzer, Peter, 416 Hedonic laundering, 438
Gould, Stephen Jay, 218 Heider, Fritz, 332–33, 333t
Grant-making, 59 Heider/Weiner matrix, 333t
Green, Donald, 195 Heinzerling, Lisa, 383
Greenfeld, Lawrence A., 181 Heuristics. See also Availability heuristic;
Griffin, Dale, 291–92, 405–8 Representativeness heuristic
Griffiths, Thomas, xxvii affect heuristic, 367–68
Gross, Paget, 318–19 distinctiveness heuristic, 252
Gross national happiness, 345 heuristics and biases agenda, 252
Group decision making. See Hewlett Foundation, 53–59
Decision-making processes Hewlett Packard Laboratories, 598
Group norms, 540–49 Heyman, James, 586
Group polarization, 612–17 “Hidden profile” paradigm, 606–8
Groupthink, 596, 600–604, 601f, 621–22 High anxiety, 498–99
Groupthink (Janis), 596 High road vs. low road
Grove, William, 296 thinking, 334–38, 336f
Guilford, Joy Paul, 66 Higney, Caeli, xxcii
Gulati, G. Mitu, 182 Hindsight bias, 272–77, 636–37. See also
Guthrie, Chris Processing and judging information,
anchors, 268, 269 biases in
external validity, 209 Hogarth, Robin, 28, 117, 633
lawyers as cognitive counselors, 527–30 Holmes, Oliver Wendell, 72
low-probability gains and losses, 482 Holyoak, Keith, 288
option devaluation, 443–44 Holzberg, Amy, 408
settlement behavior, 433 Homo economicus model, 385
settlement vs. litigation, 477–80 Hostile media phenomenon, 246–48
targets, 481 Hot/cold empathy gap, 403
Hoving, Thomas, 19
H Hsee, Christopher, 347–48, 513
Ha, Young-won, 322 Hubbard, Michael, 287
Habits, 402, 416–17 Human safety and corporate
Halo effect, 564 cost-benefit analysis, 371
Hamill, Ruth, 258 Hypothesis testing, 51, 141–49
Hamilton, William, 65 alternative hypothesis, 142
Hammond, Kenneth, 23, 117–18 chi-squared test, 146
Hanushek, Eric A., xxvii contingency tables, 147t, 148t
Happiness, 345, 360 dependent variables, 147
652 index

Hypothesis testing (cont.) deliberative processes, 12

expected frequencies, 144t, 145t framing-by-objectives, 42–43
intuitions about, 147–49 multidimensional frames, 42–43
multiple comparisons, 143 objectives, underspecification of, 43–45
null hypothesis, 142, 146–47 stakeholder analysis, 42–44, 43f
p-value, 146 Internal validity, 208
selection on dependent variables, 147 Intertemporal choice, 410–18
Type I errors, 142–44, 143t, 151 discounted utility, 411–18
Type II errors, 142–44, 143t discounting and future generations, 418
variables and regression, 178–79 discount rates, 410–11
dynamically inconsistent
I preferences, 411–18
IAT. See Implicit association test excessive impatience, 412
Identity maintenance, 614–15 extrapsychic mechanisms, 417–18
Idiosyncratic matching, 363–64 future, pervasive devaluation of, 411–18
IEM (Iowa Electronic Markets), 599 habits, 416–17
Illusory correlation, 136–37 net present value, 410, 411f
Impact bias, 397–401, 403–4 present bias, 412, 413f
Impact measurement, 185–87 procrastination, 414
Impatience, 412 repression, 415–16
Implicit association test routines, 416–17
(IAT), 326n44, 327, 335 self-control, 416–17
Implicit bias, 326n44, 338, 340. See also Intrinsic satisfaction, illusion of, 354
Implicit association test Intrinsic task motivation, 76–77
Incentives, 209, 524, 586–87 Intuition
Income accounting, 440 alternatives, choice of, 93
Independence, 136–41 Bayesian statistics, 233–35
Independence axiom, 485–87, 489–90 conjunctive events, probability of, 216–17
Individual traits continuous variables, 161–62
creativity, 77–80, 80f deliberation, interaction between, 21–25
Influence diagrams. See disjunctive events, 225
Dependency diagrams experts, 29–31, 298–301
Influencing behavior. See faulty, 237–38
Behavior, influencing hypotheses testing, 147–49
Informational cascades, 548 overcoming defects of, 293
Information-gathering alternative, 64–66 problem-solving processes, 15–25,
Information processing, 293–301. See also Processing and
stages of, 241–42, 242f judging information, biases in
Ingratiation tactics, 564 regression, 172–73
Insufficient adjustment, 267 sample size and variability, 161–62
Insurance as protection against schemas, 17–19
worry, 515–16 statistics and probability, 136–41
Interaction goals, 316 uncertainty, 119–20
Intercepts, 168 variables, 172–75
Interests and objectives, Intuitive predictions, 525
identification of, 42–45 Invariance, 372, 392, 476–77
causes, consequences, and Inverse, confusion of, 235–38
corrective strategies, 43–45 Ioannidis, John, 151
checklist, 44–45 Iowa Electronic Markets (IEM), 599
 index 653

Irrelevant alternatives, pseudocertainty effect, 486

independence of, 372 regression, intuitions about, 172–73
Irwin, Julie, 456 representativeness heuristic, 218
Isolation effect, 400 sample size and variability,
Iyengar, Sheena, 446–48 intuitions about, 161–62
two-systems model of
J cognition, 21
Jackson, Jesse, 305–6 Kamin, Kim, 272, 276
Jacobson, Lenore, 190, 325, 328 Kang, Jerry, 307–8, 335
James, William, 75 Karoly, Lynn, 381
Janis, Irving L., 596, 600–603, 621–23 Keeney, Ralph, 34, 38
Johnson, Eric, 271, 496 Kelman, Mark, 344, 450–51,
Joint probabilities, 211 453–54, 593, 633
Jones, Edward, 326 Kepner, Charles, 8, 46
Jones-Lee, Michael, 519 Keys, Daniel, 345, 349, 431
Josephs, Robert, 329 Klausner, Michael, 424
Joyce, Edward J., 267 Klayman, Joshua, 322
J-PAL (Abdul Latif Jameel Poverty Klein, Gary, 16–17, 28, 298–99, 633
Action Lab), 198, 210 Koehler, Jonathan, 283, 502
Judgmental errors. See Bias; specific Koestler, Arthur, 70
types of heuristics Korobkin, Russell, 268, 421,
Judgment and decision-making, xxix, xxx, 423–24, 479, 527–30
253, 536–37, 632 Kraus, Nancy, 517
Judgments Kronman, Anthony, 526
of risk and benefit, 502–3 Kunreuther, Howard, 347–48, 413, 511
social. See Social cognition Kuran, Timur, 497–98, 548–49
Jury awards, 269–70, 472
L
K Lakoff, George, 36, 579–80
Kahan, Dan, 504 Langer, Ellen, 137
Kahan, Marcel, 424 Langevoort, Donald C., 546
Kahneman, Daniel Larrick, Richard, 433, 591
action vs. inaction, 430 Latané, Bibb, 544
anchoring and adjustment, 216, 267 Lateral thinking, 63
availability heuristic, 51–52 Lau, Richard, 261
Bayes’ theorem and, 233–34 Laws of Fear: Beyond the Precautionary
budgeting, 439 Principle (Sunstein), 503–4
certainty effect, 483 Lawyers
experienced utility and as cognitive counselors, 523–33
expectations, 354 accountability, 524
explanation through narrative, 219 alternatives, 524
focalism, 399 biases and remedies, 523–25
heuristics and biases agenda, 252 cognitive error approach, 528
invariance, 476–77 frequencies vs.
loss aversion, 422 probabilities, 524–25
momentary vs. recollected incentives, 524
experience, 356, 359 perspective, 525
mutability, 429 predictions, statistical vs.
prospect theory, 419, 425 intuitive, 525
654 index

Lawyers (cont.) M
as experts, 29–31 Made to Stick: Why Some Ideas
framing problems, 40–42 Survive and Others Die
problem-solving (Heath & Heath), 571–72
processes by, 4–7, 29–31 Magat, Wesley, 585
“Leaky” decisions, 349–54 Maheswaran, Ravi, 149, 150
Leavitt, Harold J., 595–96 Malfeasance, social proof and, 545–46
Lee, Leonard, 352 Malmfors, Torbjörn, 517
Legal frames, 40n9 Malouff, John, 269
Leibold, Jill, 326, 328 Mandated choice, 594
Lepper, Mark, 246–48, 282, 287, Marginal probabilities, 129, 129t, 211
289, 446–48 Marginal utility, declining, 360–61
Lerner, Jennifer, 629 Margolis, Howard, 503
Leventhal, Robert, 581 Marketplace of ideas, 72
Levitt, Steven, 181 Markus, Hazel, 366, 447
Levy, Becca, 330 Maturation, 403
Liberman, Varda, 363 MAUT (Multi-Attribute
Libertarian paternalism, 588, 592–94 Utility Theory), 102
Lieppe, Michael, 556 Maximin principle, 508–11
Life on the Mississippi (Twain), 26, 634–35 McCaffrey, Edward, 425
Likelihood and likelihood ratio, 226, 230 McCauley, Clark, 622
Liking, 255–56, 562–66 McConnell, Allen, 326, 328
Line of best fit, 166–69 McCord, Joan, 197–98
Linguistic form, 575–81 McGarity, Thomas O., 112
Litigation Risk Analysis, xxvii McMakin, John, 394
Litigation vs. settlement, 477–80 Mean, 153, 170–72
Loewenstein, George Measurement reliability, 189
decoupling decisions and Measurement validity, 190
consequences, 439 Median, 154
memory and anticipation, Medvec, Victoria, 429, 432–33
utility from, 354–56 Meehl, Paul, 296
projection bias, 402–4 Memory. See Perception and
risks as feelings, 491–92, 494 memory, biases in
self-control, habits, and routines, 416 Menkel-Meadow, Carrie, 67–68, 73
utility, sources of, 346 Mental accounting, 437–42
visceral factors, 414 Mentorship, 634–35
Loftus, Elizabeth, 242, 244, 245, 250–51 Mere exposure effect, 255
Logic models, 58f, 59 Meta-analysis, 205–6
Long-run preferences, 403 Metacognition, 252–60
Loomis, Graham, 519 fluency as source of truth and liking
Lopez, David, 280–81 judgments, 255–56
Lopez, Gerald, 8 judgment biases, correction of, 256–57
Lord, Charles, 282, 289 mere exposure effect, 255–56
Loss aversion, 419, 422, 476 processing fluency, 255
Losses, integration of, 438 recall, 254
Lott, John R., 181 truth and liking judgments, 255–56
Lottery, certainty equivalent of, 508f Metaphors, 579–81
Low-ball technique, 557–58 Meyerowitz, Judith, 408
Luban, David, 530 Milgram, Stanley, 543, 550–56
 index 655

Miller, Dale, 429, 430 O

Mischel, Walter, 415–16 Obedience to apparent authority, 555–56
Mistakes, 635–37 legal and policy implications, 553–54
Mitigating factors, 497t methods of defusing, 554–55
Mode, 154 Objectives. See Interests and objectives,
Momentary experience, 356–59 identification of
Monetization, 382 Observational studies, 191
Monitor Group, 81–84 Odds, 125
Moore, Michael J., 378 odds version of Bayes’ Theorem, 239
More Guns, Less Crime: O’Donoghue, Ted, 416, 439
Understanding Crime and Onculer, Ayse, 413
Gun-Control Laws (Lott), 181 One-time events, 211–39
Mortality risks, 376–81 Bayesian statistics, 225–35. See also
Mosteller, Fred, 141 Bayesian statistics
Motivated skepticism, 280–81 complement rule, 224–445
Multi-Attribute Utility Theory (MAUT), 103 conditional probabilities,
Multiple regression analysis, 179–82 calculation of, 211–13
Multiplication rule, 136 conjunctive events,
Mussweiler, Thomas, 270 probability of, 213–17
Mutability, 429 adjustment, 216
Mutually exclusive and collectively anchoring, 216
exhaustive distributions, 129 dependent, 214–15
Myers, David, 495, 613 independent, 215
Myopic loss aversion, 420 intuitions about, 216–17
conjunctive fallacy, 217–22, 220t
N disjunctive events, intuitions about, 225
Naïve realism, 264–65 disjunctive events,
Nazi genocide, 550 probability of, 222–25
Neale, Margaret, 268 explanation through narrative, 219–22
Negative hypothesis testing, 285 false positives, 236–37
Negotiations, 268–70 faulty intuition, 237–38
Net present value, 410, 411f forensic evidence, 236–37
Newell, Alan, 9–10, 27 frequencies vs. probabilities,
The New Property (Reich), 72 comprehension of, 237–38
Nine dot problem, 67f inverse, confusion of, 235–38
Nisbett, Richard, 18, 257–58, 287–88, 301 joint probabilities, 211
Nodes, 461 marginal probabilities, 211
Nominal groups, 597–99, 625 probability tables, 211, 212t, 228t
Noncompensatory strategies, 96–98 prosecutor’s fallacy, 235–38
Nonlinear relationships, 167 representativeness heuristic, 217–22
Normal distributions, 156–58, 157f subadditivity, 221
Normalized values, 108–9 Venn diagrams, 222–23, 223f
Northcroft, Gregory, 268 verisimilitude, 219–20, 220t
Nouns, 577–78 On Perceptual Readiness (Bruner), 17
Nudge: Improving Decisions About Oppenheimer, Daniel, 270, 293
Health, Wealth, and Happiness Optimization, 627–28
(Thaler & Sunstein), 588 Option devaluation, 443–49
Null hypothesis, 142, 146–47, 163–64, 164f “Ordinization” neglect, 400–401
Nussbaum, David, xxvii Organizational determinants, 560
656 index

Osborn, Alex, 75–76 Petty, Richard, 256

Outliers, 153 Pickers of alternatives, 61–62
Outrage, 498–99 Pittman, Thane, 431
Outside-in thinking, 85 Placebo effect, 190
Overconfidence, 289–93, 618 Planning
Overjustification effect, 586 fallacy, 405
Overoptimism, 405–8, 478 foundations, grant-making, 53
strategic, 53–59
P advocacy, 56
Pak, Anita Wan-Ping, 565 alternatives, comparing, 56–57
The Paradox of Choice (Schwartz), 61 barriers, 54–55, 54f
Parsimoniousness, 372 benefits, 57–59
Passive language, 575–76 costs, 57
Paternalism and the Legal evaluation, 59
Profession (Luban), 530 expected return, 57, 57f
Path constraints of problem-solving goals, 53
processes, 10 grant-making, 59
Peak-end phenomenon, 357–59, 358f logic models, 58f, 59
Pennington, Nancy, 220, 573 monitoring, 59
Perception and memory, biases in, 241–65 outcomes, 55
acquisition, 242–44, 261–64 scope of work, 53, 54f
availability heuristic, 252–60. See also stakeholders, 55–56
Availability heuristic success, likelihood of, 57
bounded awareness, 244 targets, 55, 59
distinctiveness heuristic, 252 Plott, Charles R., 425, 598–99
egocentrism, 261–64 Plous, Scott, 140, 290, 292
expectations, 244–49 Polarization, 612–17
false consensus effect, 264 Policy interventions,
familiarity, 251 evaluation of, 185–210
focusing failure, 244 Policy makers, problem-solving
hostile media phenomenon, 246–48 processes by, 7–8, 29–31
information processing, The Political Brain (Westen), 573
stages of, 241–42, 242f Population means, estimating, 158–61
interests, 244–49 Positive hypothesis testing, 283–86
interpretation, 245–46 Positive illusions (Taylor), 404–10
memory and anticipation, Posner, Eric, 182
utility of, 354–56 Posner, Richard, 66, 67
memory binding, 250 Posterior probability, 227
metacognition, 252–60 Postman, Leo, 244
naïve realism, 264–65 Precautionary principles, 521
reactive egoism, 262–63 Precommitment, 437
retention, 242–43, 249–52 Predictable surprises, 625–27
retrieval, 242–43, 249–52, 261–64 Predicted utility, 396
schemas, 244–49 Predictions, statistical vs. intuitive, 525
settlements, 248–49 Predictors, explanations by, 175–77
theories, 244–49 Preferences. See specific types
Perceptions, cultural Prelec, Drazen, 416
determinants of, 503–5 Premature convergence, 623–24
Persuasive argument theory, 615 Present bias, 412, 413f
 index 657

Present value. See Net present value affect, role of, 19–20
Preston, Elizabeth, 289 recognition-primed decision
Prevention behaviors, 589 making, 16–17
Primary emotions, 354 schemas and scripts, role of, 17–19
Priming, 315–16, 581–84 by lawyers, 4–7, 29–31
The Principles of Morals and overview, 3–4
Legislation (Bentham), 346 path constraints, 10
Prior decisions and finality, 437 by policy makers, 7–8, 29–31
Prior expectancies and memory problem space, 9–10, 10f, 11f
retrieval, 322–23. See also Expectancies, schemas, defined, 14, 18n20
effects of scripts, defined, 18
Prior probability, 226, 277. See also sensemaking, 19
Base rate Procedural rationality, 368–69
Probabilistic sampling, 608–9 Processing and judging information,
Probability. See Statistics and probability biases in, 267–302. See also Bias
Probability axiom, 460 anchoring, adjustment, and
Probability neglect, 514–19 activation, 267–72
Probability tables, 211, 212t, 228t biases in, 267–302. See also Bias
Problem-solving, 3–31. See also calibration, 291
Decision-making processes confirmation, 277–89
affect, defined, 19n26 hindsight, 272–77
creativity in, 61–90. See also intuition, analysis, and
Creativity in problem solving experts’ role, 293–301
defined, 8–11 intuition, overcoming, 293
deliberative, 11–13 overconfidence, 289–93
alternatives, 13 prior expectations and theories,
causes, diagnosis of, 12–13 resistance to modification of, 277–89
course of action, choice of, 13 Procrastination, 414
framing problems, 12. See also Project determinants, 559
Frames and framing Projection bias, 401–4
interests and objectives, Pronin, Emily, 635
identification of, 12. See also Propulsion theory, 78
Interests and objectives, Prosecutor’s fallacy, 235–38. See also
identification of Confusion of the inverse
thinking, divergent and Prospect theory, 208, 419–28, 420f, 476–87
convergent, 13–14, 15f Prototypes. See Schemas; Stereotypes and
emotional intelligence, 20 stereotyping
expertise, nature of, 25–29, 31 Pseudocertainty effect, 485t, 486
experts, 29–31 Psychological determinants, 560
intuition and deliberation, Psychological health,
interaction of, 21–25 qualifying effects, 381
bounded rationality, 22–24 Psychological immune
deliberation, limits of, 22–24 system, 359, 364, 400
necessity and limits of intuition, 25 Psychological ownership, 423
quasi-rational thought, 23 Psychophysical numbing, 361–62
satisficing, 22, 385 Public policy and multiple
two-systems model of regression, 181–82
cognition, 21–22, 21t p-value, 146, 164
intuitive, 14–20 Pygmalion Effect, 190
658 index

Q Reason-based decision
Quality-adjusted life years (QALY), 379–81 strategies, 99–102, 390–95
Quantitative decision strategies, 102 Reb, Jochen, 501
Quantitative variables, 153–84. See also Reciprocity, 563–64, 566–68
Variables Recognition-primed
Quasi-rational thought, 23 decision making, 16–17, 298–99
Quattronel, George, 574–75 Recollected experience, 356–59
Quinn, Diane, 330 Recuperative behaviors, 589
Redelmeier, Donald E., 446
R Reference points, 419, 476
Rachlinski, Jeffrey Refinement, 403
anchoring in negotiations, 269 Reflexive controls, 193–94
hindsight bias, 272, 274, 276–77 Regan, Dennis T., 566
litigation vs. settlement, 478 Regression
settlement offers, 527 analysis of variance equation, 176
targets and risk-seeking, 481 chance and, 172
uncertainty, 505 coefficient of determination, 176
Racial bias, 304–30 correlation, 169–70, 171f, 177f
behavioral confirmation, 326–27 employment discrimination, 179–80
juvenile probation, 324 equations, 168
weapon, perception of possession, 319 hypothesis testing, 178–79
Randomized controlled studies, 191–92 independence, 165
Randomness, 123–24, 136–41. See also intercepts, 168
Statistics and probability intuitions about, 172–75
Random samples, 132 linear, 179
Random variables, 123–24 line of best fit, 166–69
Ranked groups, 444 to mean, 170–72
Rational decision making, 366–73 multiple, 179–81
affect heuristic, 367–68 nonlinear relationships, 167
completeness, dominance, and predictors, explanations by, 175–77
trade-offs, 369–72 public policy and multiple
consequential rationality, 368 regression, 181–82
constitutive incommensurability, 370 representativeness heuristic, 173
corporate cost-benefit analysis involving scatterplots, 166, 166f
human safety, 371 simple linear, 179
decomposability, 372 statistical significance, 178–79
ecological rationality, 373 variables, 165–82
invariance, 372 variance around line of best fit, 175
irrelevant alternatives, variance around mean, 175
independence from, 372 winner’s curse, 174, 174f
parsimoniousness, 372 Regret
procedural rationality, 368–69 aversion, 428–34
subjective expected avoidance, 427, 436
utility theory, 368–73 phenomenon, 354
transitivity, 372–73 suboptimal learning processes and, 434
Rationale construction, 616 Reich, Charles, 71–72
Rationalization, 364 Reichenbach, Hans, 119
Reactive egoism, 262–63 Relative wealth and happiness, 360
Reality checks, 109–10 Reliability, 300
 index 659

Repeated expression, effects of, 616 Risk/benefit analysis, 275–76

Representativeness heuristic, 52, 139, Ritov, Ilana, 501
173, 217–22 “Robust satisficing,” 508–11
Repression, 415–16 Rogelberg, Steven, 624
Reputation, 436 Rosenthal, Robert, 190, 325
Reputational cascades, 548 Ross, H. Lawrence, 201, 204
Retaking Rationality (Revesz), 370 Ross, Jerry, 558–62
Retention, 242–43, 249–52 Ross, Lee
Retrieval, 242–43, 249–52, 261–64 attribution bias, 637
Revealed preference, 455 belief perseverance, 287–88
Revesz, Richard, 370, 378–79, 383, 520 bias blind spots, 635
Riis, Jason, 356, 359 biased assimilation, 282
Risk. See also Uncertainty collective learning, 301
affect, role of, 491–522. See also Affect hedonic laundering, 438
Allais’ paradox, 489, 490t hostile media phenomenon, 246–48
anchoring, 472–73 idiosyncratic matching, 363
certainty effect, 483–85, 484t, 489–90 naïve realism, 264
costs, 472–73 obedience, 555
counseling, 584–86 scripts, 18
decision-making processes, 459–90 vividness, 257
decision trees, 461–63, 463f, Ross, Michael, 263, 405–8
463t, 466f, 469f, 470t, 471f Rothbart, Myron, 322
dependency diagrams, 460–61, Rothman, Alex, 589
461f, 462f, 465f Rottenstreich, Yuval, 444, 513
diminishing sensitivity, 420, 476 Routines, 416–17
dominance axiom, 488–89, 489f Russell, Daniel, 261
expected utility, 459–60 Russo, Edward, 289–90
expected value analysis, 473f Rustichini, Aldo, 587
feelings as, 491–92, 494
framing, 476–87 S
gains and losses, domains of, 489t Sagar, H. Andrew, 317–18
independence axiom, 485–87, 489–90 Sagoff, Mark, 383
influence diagrams, 460 Salovey, Peter, 589
jury awards, 472 Samples and sampling, 132–33
litigation vs. settlement, 477–80 bias, 132
loss aversion, 476 distribution, 158, 159f
low-probability gains and losses, 482–83 population mean vs. sample mean, 161
nodes, 461 probabilistic sampling, 608–9
overoptimism, 478 size and variability, 161–62
probabilities, estimating, 467–70 Samuelson, Paul, 412
probability axiom, 460 Sandman, Peter, 499, 512
prospect theory, 476–87 Satiation, 402
pseudocertainty effect, 485–87, 485t Satisficing, 22, 385, 508–11
reference points, 476 Scatterplots, 166, 166f
risk as feelings, 367–68, 491–92, 494 Scearce, Diana, 84
sensitivity analysis, 470–72 Scenario planning, 80–86. See also
targets, 481–82 Alternatives, generation of
tolerance and utility functions, 473–76 availability heuristic, 85
unethical behavior, 481–82 brainstorming, 85
660 index

Scenario planning (cont.) Self-selection, 189

defined, 81–84 Semin, Gun, 576–77
drawbacks of, 85–86 Sensemaking, 19
Global Business Network (GBN), 81–84 Sensitivity
key principles of, 84–85 analysis, 470–72
long view, 84 Bayesian statistics, 230
Monitor Group, 81–84 diminishing, 420, 476
multiple perspectives, 85 Settlement offers, 527
outside in thinking, 85 Settlements, 248–49, 559
STEEP trends, 85 Shafir, Eldar, 391, 395, 446
use of, 81–84 Shared information, bias for, 610–12
Schacter, Daniel, 250–52 Shelling, Thomas, 417
Schemas Sherif, Muzafer, 542
ambiguous information, Sherman, Steven, 495
interpretation of, 317–21 Shestowsky, Donna, xxvii, 350
biasing effects of, 303–16 Shiv, Baba, 353
causal attribution, effects on, 323–24 Shutte, Nicola, 269
defined, 14, 18n20 Simon, Dan, 288
framing problems, 34 Simon, Herbert, 9–10, 22–23, 27
intuitive problem-solving processes, 17–19 Simonson, Itamar, 450
memory, effects on, 321–23 Simplicity, 572
perception and memory, Single action bias, 515
biases in, 244–49 Situationism in social attribution, 331–33.
predicting behavior, effects of, 323–24 See also Dispositionism in social
self-fulfilling effects of, 324–30 attribution; Fundamental
social cognition, 305–24. See also attribution error
Social cognition Slovic, Paul
stereotypes and stereotyping, 305–8 choice quality, effect of giving
Scheuplein, Robert, 498 reasons on, 394
Schkade, David, 404 consequences and trade-offs, 367
Schoemaker, Paul, 289–90 cost-benefit analysis (CBA), 456
Schofield, Janet, 317–18 discounted utility, 413
Schooler, Jonathan, 393–94 high anxiety and outrage, 498
Schwartz, Barry, 61–62, 345, judgments of risk and benefit, 503
349, 431, 447, 509–10 laypersons vs. professionals, 517
Schwarz, Norbert, 255, 260, 276 value-based decision making, 389
Schweitzer, Maurice, 481 vividness and media, 260
Scores, 153–84. See also Variables Snibbe, Alana Connor, 447
Scripts, 18 Social cognition, 303–40
Seiter, Andreas, 381 actor-observer effect, 332
Sela, Aner, xxvii categorization, 307f
Selection bias, 132 debiasing social judgment, 334–40
Self, sense of, 437 dual process models, 334–38, 336f
Self-affirmation theory, 281, 281n26 high road vs. low road
Self-attribution, 557 thinking, 334–38, 336f
Self-control, 416–17, 440 implicit stereotypes,
Self-deception, value of, 409 malleability of, 338–40
Self-education, 437 dispositionism and situationism in
Self-governance skills, 76 social attribution, 331–33
 index 661

fundamental attribution error, 332 legal and policy implications, 553–54

Heider/Weiner matrix, 333t Nazi genocide, 550
overview, 303–4 conformity, 540–49
social schemas, 305–24 availability cascades, 548–49
ambiguous behavior, effects on bystander effect, 544–45
characterization of, 317–21 cascades, emergence of, 547–49
attention allocation, 321–22 death by, 544–45
biasing effects of, 304–5 informational cascades, 548
causal attribution and behavioral malfeasance, social proof and, 545–46
prediction, effects on, 323–24 reputational cascades, 548
cognitive mechanics of, 305–16 social proof, strategic
deactivation, 313 deployment of, 549
expectancies, 321–22 consistency and commitment
interaction goals, 316 escalation, 556–62
memory, effects on, 321–23 determinants, 559–62
memory encoding, 321–22 door-in-the-face technique, 567–68
priming, 315–16 foot-in-the-door technique, 556–57
prior expectancies and memory irrationality, 558–62
retrieval, 322–23 low-ball technique, 557–58
schematic associations with organizational determinants, 560
negative prime, 311–13, 312f project determinants, 559
schematic associations with psychological determinants, 560
positive prime, 311–13, 314f self-attribution, 557
schematic information settlement of cases, 559
processing, 308, 310f social determinants, 560
social knowledge, mental sunk costs fallacy, 559
representation of, 308–16 group norms, 540–49
social perception and judgment, liking, 562–66
effects on, 317–24 attractiveness, physical, 564–66
social salience or “tokens,” 320–21 compliments, 564
stereotypes and cooperation, 564
stereotyping, 305–8, 324–30 familiarity, 562
stereotypes and halo effect, 564
stereotyping, 305–8, 324–30 reciprocity, 563–64, 566–68
behavioral confirmation similarity, 563
effect, 325–29, 327f Social knowledge, mental representation
behavior-behavior links, 326, 328 of, 308–16
expectancy-behavior links, 326 Social perceivers. See Social cognition
self-fulfilling nature of, 324–30 Social proof, 540–49
social schemas, 305–8 Social salience or “tokens,” 320–21
threat of, 329–30 Soll, Jack, 591
Social cognition theory, 306, 527n13 Solow, Robert, 457
Social corroboration effect, 614 Song, Hyunjin, 260
Social determinants, 560 Sood, Sanjay, 444
Social influence, 403, 539–69 Spann, Kelly, xxiii
Abu Ghraib prison, 539, 553–54 Speeding, Connecticut
apparent authority, crackdown on, 201–4
obedience to, 550–56 Spencer, Steven, 330
impulses toward, defusing, 554–56 Spitzer, Matthew, 425
662 index

Staines, Anthony, 149, 150 frequency vs., 524–25

Stakeholder analysis, 42–44, 43f frequentism vs. subjectivism, 125
Stakeholders in strategic planning, 55–56 independence and, 133–36
Standard deviation, 154–56, 155t multiplication rule, 136
Stanislavski, Constantin, 632 odds, 125
Stasser, Garold, 606 random samples, 132
Statistical groups, 597–99, 625 random variables, 123–24
Statistics and probability, 123–52. See also sampling, 132–33
Evaluation sampling bias, 132
aim of, 127–28 selection bias, 132
Bayesian statistics, 225–35. See also simple random samples, 132
Bayesian statistics statistical inference, 132
cancer clusters, 149–52 surveys, 133
conditional probabilities, probability tables, 211, 212t, 228t
calculation of, 211–13 randomness, 136–41
conjunctive events, 213–17 control, illusion of, 137–39
contingency tables, 128–30, 129t gambler’s fallacy, 139–41
disjunctive events, illusory correlation, 136–37
probability of, 222–25 random variables, 123–24
distributions, 125–31 representativeness heuristic, 139
empirical, in relative frequencies, 131t samples, 132
frequencies, 126, 127f role in law, policy, and
known empirical, 130–31, 131t citizenship, 121–22
marginal probability, 129, 129t significance, 141–49, 205–6
mutually exclusive and effect size, 205
collectively exhaustive, 129 meta-analysis, 205–6
population vs. empirical, 125–31 null hypothesis, 146–47
probability, 125–26 “Texas sharpshooter” procedure, 149–52
unknowable population Status quo bias, 419–28
distribution, 130, 130t Staw, Barry, 558–62
evaluation. See Evaluation Steele, Claude, 281, 329, 330
frequencies vs. probabilities, 524–25 Steen, Sara, 324
hypothesis testing, 141–49. See also STEEP trends, 85
Hypothesis testing Stephens, Nicole, 447
interpretation of results, 185–210. Stepladder technique, 624–25
See also Evaluation Stereotypes and stereotyping, 303–40.
intuition, 136–41 See also Schemas
multiple testing, 149–52 attention allocation and, 321–22
policy interventions, biasing effects of, 305–6
evaluation of, 185–210 debiasing, 334–40
predictions, 525 expectancies, 322–23
probabilistic sampling, 608–9 implicit stereotypes,
probability, 123–25 malleability of, 338–40
complement rule, 124 priming, 315–16, 330
conditional, 133–36 social cognition, 324–30
deterministic variable, 123 behavioral
distributions, 125–26 confirmation effect, 325–29, 327f
estimation of, 132–33, 467–70 behavior-behavior links, 326, 328
events, 124 expectancy-behavior links, 326
foundations of, 124–25 self-fulfilling nature of, 324–30
 index 663

social schemas, 305–8 daily experiences vs. general sense of

threat of, 329–30 well-being, 359–60
social judgment errors, 331–34 decision process, 349–54
Stereotype threat, 329–30 declining marginal utility, 360–61
Sternberg, Robert, 69, 78 dictator game, 350–51
Stevenson, Betsey, 360 duration neglect, 357
Strack, Fritz, 270 Easterlin Paradox, 360
Stradonitz, Friedrich ends, arationality of, 344–47
August Kekulé von, 120 existence and option value, 347
Strategic planning. See Planning experience, memory, and
Strategies, 53–59 anticipation, 354–56
choice of, 386–88 experienced utility, 351–54
corrective strategies, 43–45 fairness, 349–51
debiasing strategies, 289 free, allure of, 348–49
decision-making processes, 386–95. gross national happiness, 345
See also Decision-making processes hedonic adaptation, 364–66
deliberative strategies, 96–99 idiosyncratic matching, 363–64
noncompensatory strategies, 96–98 intrinsic satisfaction, illusion of, 354
quantitative decision strategies, 102 “leaky” decisions, 349–54
reason-based decision momentary vs. recollected
strategies, 99–102, 390–95 experience, 356–59
value-based strategies, 388–90 peak-end phenomenon, 357–59, 358f
Subadditivity, 221 primary emotions, 354
Subjective expected utility theory, 368–73 psychological immune system, 364
Subjective linear models, 102–11 psychophysical numbing, 361–62
aggregation of evaluations, 109, 109t rationalization, 364
attributes, evaluation of, 107–8, 108t regret phenomenon, 354
attributes, quantitative sources, 344–49
description of, 109t subjective expected
conjoint analysis, 111n22 utility theory, 368–73
decision opportunities, 111 temporal monotonicity, 357
evaluations, comparison of, 109 transaction utility, 351
major attributes, 105–6 ultimatum game, 350
MAUT (Multi-Attribute Weber’s law, 362
Utility Theory), 103 Subjectivism, 225
normalized values, 108–9 Suboptimal decisions, 385–86, 441–42
options, 106–7 Suboptimal learning processes, 434
reality checks, 109–10 Sunk costs, 434–37
weighing attributes, 110–11, 110t, 111t Sunk costs fallacy, 559
Subjective utility, 343–66 Sunstein, Cass
absolute vs. relative wealth biases, counteracting, 588–94
and happiness, 360 cascades, 548–49
affection, 347–48 consequences of alternatives, 93
Bernoulli’s curve, 361f cost-benefit analysis (CBA), 383
choice overload, 354 discounting and future generations, 418
compensation, 348 familiarity, 499
consolation hypothesis, 348 group polarization, 613
consumption effect, 355 high anxiety and outrage, 498
contrast effect, 355 natural origins, 500
cultural differences, 366 precautionary principles, 522
664 index

Sunstein, Cass (cont.) certainty effect, 483

probability neglect, 516, 518–19 elimination by aspects, 97
prospect theory, 426 explanation through narrative, 219
risk counseling, 585 gains and losses, framing by, 574–75
risk inequality, 520 gamblers’ fallacy, 140
small steps in decision making, 96 grouping, effects of, 444–45
time-buying alternatives, 65–66 heuristics and biases agenda, 252
value of statistical life (VSLY), 378 invariance, 476–77
Suppose We Took Groups Seriously loss aversion, 422
(Leavitt), 595 prospect theory, 419
Surowiecki, James, 597 pseudocertainty effect, 486
Surveys, 133 regression, intuitions about, 172–73
Svenson, Ola, 387 representativeness heuristic, 218
System 1 and System 2. See Dual sample size and variability,
process models intuitions about, 161–62
trade-off contrast or asymmetrical
T dominance, 450
Targets, 55, 59, 481–82 two-systems model of cognition, 21
Taylor, Brian, 430 Twain, Mark, 26, 634–35
Taylor, Shelley, 320–21, 409 Tweney, Ryan, 280
Temkin, Larry, 508–9 Two-systems model of
Temporal monotonicity, 357 cognition, 21–22, 21t
Tetlock, Philip, 370, 629 Tykocinski, Orit, 431
“Texas sharpshooter” procedure, 149–52
Thaler, Richard, 351, 420, 438, U
440–42, 537, 588–94 Ullman-Margalit, Edna, 93, 96
Thesing, Alicia, xxvii Ultimatum game, 350
Time budgeting, 441 Uncertainty, 115–22. See also Risk
Time-buying, 64–66 causation imputation, 115–17
Time horizons, 36 correlation detection, 115–17
Titus, William, 606 fallible indicators, 118, 118n5
“Tokens” or social salience, 320–21 Hawthorne effect, 117
Tolerance and utility functions, 473–76 intuition and analysis, roles in
Tooby, John, 373 empirical judgment, 119–20
Trade-offs. See Choices, consequences, lens models, 117–19, 117f, 128
and trade-offs Pygmalion effect, 117
Transaction utility, 351 reality, 117–18, 117n4
Transitivity, 372–73 risk vs., 505
Tree structure and Bayesian scientific method, 119
statistics, 227–28, 227f, 238f statistics, role of, 121–22
Tregoe, Benjamin, 8, 46 Uncontrolled selection biases, 189
Turnbull, Brit, xxvii Unethical behavior, 481–82
Tversky, Amos Unknowable population
action vs. inaction, 430 distribution, 130, 130t
anchoring and Utility. See Declining marginal utility;
adjustment, 216, 267, 291–92 Discounted utility; Expected utility;
availability heuristic, 51–52 Experienced utility; Marginal utility,
Bayes’ theorem and, 233–34 declining; Multi-Attribute Utility
budgeting, 439 Theory (MAUT); Predicted utility;
 index 665

Subjective expected utility theory; Venn diagrams, 222–23, 223f

Subjective utility; Transaction utility Verbs and causal inference, 576–77
Verisimilitude, 219–20, 220t
V Victor, Mark B., xxvii
Validity, 300 Vigilant appraisal, 621
Vallentyne, Peter, 508–9 Visceral factors, 414
Vallone, Robert, 140, 246–48 Viscusi, W. Kip, 275, 371, 378, 497, 508, 585
Valuation by calculation or feeling, 513 Vividness, 257–60, 495–97
Value-based strategies, 388–90 Voluntariness, 500
Value of statistical life (VSLY), 377–78
Value of statistical life years, 378–79 W
Variables Walton, Gregory, 577–78
categorical, 163–65 Wasserstrom, Richard, 119–20
confounding, 182, 187 Watkins, Michael, 626
continuous, 153–65 Wealth and happiness, 360
central tendency, 153–65 Weber, Elke, 515
dispersion, 154–56, 154f Weber’s law, 362
estimates, 162t Wegener, Duane, 256
intuitions, 161–62 Weick, Karl, 19
mean, 153 Weiner, Bernard, 333, 333t
median, 154 Well-being, predicting, 395–404
mode, 154 Wertenbroch, Klaus, 417
normal distributions, 156–58, 157f Westen, Drew, 573
null hypothesis, 163–64, 164f Whately, Mark, 566
outliers, 153 Why Most Published Research Findings
population means, Are False (Ioannidis), 151
estimating, 158–61 William and Flora Hewlett
p-value, 164 Foundation, 53–59
quantitative variables, 163–65 Wilson, Timothy, 258, 364,
relationships, testing for, 163–65 392–94, 397–98, 400
sample size and variability, 161–62 Winner’s curse, 174, 174f
sample vs. population mean, 161 Wisdom of Crowds
sampling distribution, 158, 159f (Surowiecki), 597–600
standard deviation, 154–56, 155t Wistrich, Andrew, 269, 479
standard error, 158–60 Within-subject vs. between-subject
variance, 154–56 experiments, 207
dummy, 180 Wittenbaum, Gwen, 609–10
experiments, 182 Wolfers, Justin, 360
explanatory, 179 Woloshin, Steven, 232
indicator, 180 Word, Carl, 326, 328
quantitative, 123–24, 153–84 Worry, 514–19
random, 123–24 Worst-Case Scenarios (Sunstein), 522
regression, 165–82. See also Regression Wyer, Robert, 323
analysis of variance equation, 176
Variance, 154–56 Z
around line of best fit, 175 Zajonc, Robert, 255, 562
around mean, 175 Zanna, Mark, 326, 328
equation, 176 Zeiler, Kathryn, 425
Vaughn, Leigh Anne, 276 Zimbardo, Philip, 290, 292, 540–41, 556