Instant Ebook Access, One Click Away – Begin at ebookgate.

com

Learning From Data An Introduction to Statistical

Reasoning using JASP 4th Edition Arthur M.
Glenberg & Matthew E. Andrzejewski

https://ebookgate.com/product/learning-from-data-an-
introduction-to-statistical-reasoning-using-jasp-4th-
edition-arthur-m-glenberg-matthew-e-andrzejewski/

OR CLICK BUTTON

DOWLOAD EBOOK

Get Instant Ebook Downloads – Browse at https://ebookgate.com

Click here to visit ebookgate.com and download ebook now
Instant digital products (PDF, ePub, MOBI) available
Download now and explore formats that suit you...

An Introduction to Statistical Concepts 4th Edition Debbie

L. Hahs-Vaughn

https://ebookgate.com/product/an-introduction-to-statistical-
concepts-4th-edition-debbie-l-hahs-vaughn/

ebookgate.com

Introduction to bioinformatics Arthur M. Lesk

https://ebookgate.com/product/introduction-to-bioinformatics-arthur-m-
lesk/

ebookgate.com

Structural Geology An Introduction to Geometrical

Techniques 4th edition Donal M. Ragan

https://ebookgate.com/product/structural-geology-an-introduction-to-
geometrical-techniques-4th-edition-donal-m-ragan/

ebookgate.com

Plants from Test Tubes An Introduction to Micropropagation

4th Edition Lydiane Kyte

https://ebookgate.com/product/plants-from-test-tubes-an-introduction-
to-micropropagation-4th-edition-lydiane-kyte/

ebookgate.com
An Introduction to GameGuru 1st Edition Michael Matthew
Messina (Author)

https://ebookgate.com/product/an-introduction-to-gameguru-1st-edition-
michael-matthew-messina-author/

ebookgate.com

Introduction to Statistical Data Analysis for the Life

Sciences 1st Edition Sørensen

https://ebookgate.com/product/introduction-to-statistical-data-
analysis-for-the-life-sciences-1st-edition-sorensen/

ebookgate.com

An introduction to thermodynamics and statistical

mechanics 2nd Edition Stowe K.

https://ebookgate.com/product/an-introduction-to-thermodynamics-and-
statistical-mechanics-2nd-edition-stowe-k/

ebookgate.com

Inside Jokes Using Humor to Reverse Engineer the Mind 1st

Edition Matthew M. Hurley

https://ebookgate.com/product/inside-jokes-using-humor-to-reverse-
engineer-the-mind-1st-edition-matthew-m-hurley/

ebookgate.com

Recipes for Science An Introduction to Scientific Methods

and Reasoning 2nd Edition Angela Potochnik

https://ebookgate.com/product/recipes-for-science-an-introduction-to-
scientific-methods-and-reasoning-2nd-edition-angela-potochnik/

ebookgate.com
Learning From Data

This fully updated fourth edition explores the foundations of statistical reasoning, focusing
on how to interpret psychological data and statistical results. This edition includes three
important new features. First, the book is closely integrated with the free statistical analysis
program JASP. Thus, students learn how to use JASP to help with tasks such as constructing
grouped frequency distributions, making violin plots, conducting inferential statistical tests,
and creating confidence intervals. Second, reflecting the growing use of Bayesian analyses in
the professional literature, this edition includes a chapter with an introduction to Bayesian
statistics (also using JASP). Third, the revised text incorporates adjunct questions, that is,
questions that challenge the student’s understanding, after each major section. Cognitive
psychology has demonstrated how adjunct questions and related techniques such as self-
explanation can greatly improve comprehension.
Additional key features of the book include:

• A user-friendly approach, with focused attention on explaining the more difficult concepts
and the logic behind them. End of chapter tables summarize the hypothesis testing
procedures introduced, and exercises support information recall and application.
• The consistent use of a six-step procedure for all hypothesis tests that captures the logic
of statistical inference.
• Multiple examples of each of the major inferential statistical tests.
• Boxed media reports illustrate key concepts and their relevance to real-world issues.
• A focus on power, with a separate chapter, and power analysis procedures in each chapter.

With comprehensive digital resources, including large data sets integrated throughout the
textbook, and files for conducting analysis in JASP, this is an essential text for undergraduate
or beginning graduate statistics courses in psychology, education, and other applied social
and health sciences.

Arthur M. Glenberg is an emeritus professor of Psychology at Arizona State University

and at the University of Wisconsin–Madison. He is a Mercator Fellow at the University of
Tübingen in Germany, and he holds an appointment at INICO at the University of Salamanca
in Spain. He received his Ph.D. in 1974 from the University of Michigan.

Matthew E. Andrzejewski is a lecturer in the Department of Psychology at the University of

Wisconsin–Whitewater. He received his Ph.D. in 2001 from Temple University.
Learning From Data
An Introduction to Statistical Reasoning using JASP

Fourth Edition

Arthur M. Glenberg and

Matthew E. Andrzejewski
Designed cover image: Getty Images
Fourth edition published 2025
by Routledge
605 Third Avenue, New York, NY 10158
and by Routledge
4 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN
Routledge is an imprint of the Taylor & Francis Group, an informa business
© 2025 Arthur M. Glenberg and Matthew E. Andrzejewski
The right of Arthur M. Glenberg and Matthew E. Andrzejewski to be identified as authors
of this work has been asserted in accordance with sections 77 and 78 of the Copyright,
Designs and Patents Act 1988.
All rights reserved. No part of this book may be reprinted or reproduced or utilised in any
form or by any electronic, mechanical, or other means, now known or hereafter invented,
including photocopying and recording, or in any information storage or retrieval system,
without permission in writing from the publishers.
Trademark notice: Product or corporate names may be trademarks or registered
trademarks, and are used only for identification and explanation without intent to infringe.
First edition published by Harcourt Brace Jovanovich 1988
Third edition published by Psychology Press 2007
ISBN: 978-0-367-45798-3 (hbk)
ISBN: 978-0-367-45797-6 (pbk)
ISBN: 978-1-003-02540-5 (ebk)
DOI: 10.4324/9781003025405
Typeset in Galliard
by Deanta Global Publishing Services, Chennai, India
Access the Support Material: www​.routledge​.com​/9780367457976
Contents

Preface vii

1 Why statistics? 1

2 Frequency distributions and percentiles 20

3 Central tendency and variability 44

4 z scores and normal distributions 68

5 Overview of inferential statistics 91

6 Probability 105

7 Sampling distributions 120

8 Logic of hypothesis testing 144

9 Power 182

10 Logic of parameter estimation 203

11 Inferences about population proportions using the z statistic 219

12 Inferences about μ when σ is unknown: The single-sample t test 244

13 Comparing two population means: Independent samples 266

14 Random sampling, random assignment, and causality 300

15 Comparing two populations: Dependent samples 311

16 Comparing more than two population means: Independent samples 343

vi Contents

17 One-factor ANOVA for dependent samples 371

18 Introduction to factorial designs 388

19 Describing linear relationships: Regression 413

20 Measuring the strength of linear relationships: Correlation 448

21 Inferences from nominal data: The χ2 statistic 473

22 Introduction to Bayesian statistics 496

Glossary of symbols 509

Tables 512
Appendix A 523
Appendix B 525
Answers to selected exercises 527
Index 534
Preface

New to the Fourth Edition

We have made three changes to the fourth edition that make Learning from Data: An
Introduction to Statistical Reasoning a uniquely valuable textbook. First, we have integrated
the book with the free and easy-to-use statistical analysis program JASP (https://jasp​-stats​
.org/). We can now present modern graphing techniques, such as boxplots and violin plots
(see Chapter 2) to help students better understand data. Also, this integration allows us to
eliminate many computational formulas because nobody needs to do the computations by
hand anymore. The website for LFD IV (www​.routledge​.com​/9780367457976) has many
of the data files used with JASP in the worked examples. Thus, students can easily download
a file and duplicate an analysis. Perhaps even better, we encourage students to explore the
various JASP settings to produce their own analyses.
Another benefit of using JASP is that it was designed to perform both classical (null
hypothesis significance tests, NHST) and Bayesian analyses. Many researchers are becom-
ing convinced that Bayesian analyses are more useful than NHST. Consequently, we have
included a new chapter (Chapter 22), “Introduction to Bayesian Statistics.” In this chapter
we describe a number of reasons for why researchers are concerned about NHST, and some
of the reasons for optimism about Bayesian approaches. The chapter also includes several
re-analyses of data sets used in earlier chapters to illustrate how Bayesian analyses and NHST
differ.
The third change is based on an advance in the cognitive psychology of learning. A large
body of research has demonstrated the benefits of what is variously known as “adjunct ques-
tions” or “self-explanation.” Thus, each chapter is divided into multiple sections, and each
section is followed by a question designed to stimulate the student’s thinking. We hope that
students will read these questions, and when the answer isn’t immediately recalled, that the
student will pause to think it through.
These three changes do not preclude many of the other features built into earlier editions
of LFD. First, we devote extra attention to explaining difficult-to-understand concepts. For
example, some textbooks attempt to combine important concepts such as sampling distri-
butions, hypothesis testing, power, and parameter estimation in one chapter. In this book,
each concept has its own chapter. Yes, this means more reading, but it also means greater
understanding.
Second, the book uses extensive repetition to help students learn and retain concepts.
There are multiple fully explained examples of each major procedure. Many concepts (for
example, power, Type I errors) are repeated from chapter to chapter. The problem sets at the
ends of most chapters require students to apply principles introduced in earlier chapters as
well as the just-read chapter.
viii Preface

A third major learning aid is the use of a consistent schema (the six-step procedure) for
describing all statistical tests from the simplest to the most complex. The schema provides a
valuable heuristic for learning from data. Students learn (1) to consider the assumptions of
a statistical test, (2) to generate null and alternative hypotheses, (3) to formulate sampling
distributions, (4) to set a significance criterion and generate a decision rule, (5) to compute
the statistic of interest, and (6) to draw conclusions. Learning the schema at an early stage
(in Chapter 8) will ease the way through Chapters 11–22, in which the schema is applied to
many different situations. A table with a summary schema is included in the last section of
each chapter containing the hypothesis-testing procedure. Inside the front cover of the book
is a “Statistical Selection Guide” to further assist students in determining which statistical test
is most appropriate for the data analysis situation.
A unique feature of LFD is that it contains two independent treatments of power. The
major treatment begins in Chapter 9 with graphical illustrations of how power changes under
the influence of factors such as the significance level and sample size. The chapter also intro-
duces formulas for computing power and estimating sample size needed to obtain a particular
level of power. These formulas are repeated and generalized for many of the statistical proce-
dures discussed in later chapters. However, instructors may find that they do not have enough
time for an extensive treatment of power. In that case, instructors can choose to treat power
less extensively and omit Chapter 9 (and the relevant formulas in the other chapters). This less
extensive treatment of power is part of the description of each inferential procedure. It consists
of a non-mathematical discussion of how power can be enhanced for that particular procedure.
Last, but very important to us, is Chapter 14, “Random Sampling, Random Assignment,
and Causality.” A major reason for writing the first three editions of this book was to address
the issues discussed in this chapter. All of us who teach statistics courses and conduct research
have been struck by the incongruity between what we practice and what we preach. When
we teach a statistics course, we emphasize random sampling from populations. But in most
experiments, we do no such thing. Instead, we use some form of random assignment to con-
ditions. How can we perform statistical analyses of our experiments when we have ignored
the most important assumption of the statistical tests? In Chapter 14, we develop a rationale
for this practice, but the rationale extracts severe payment by placing restrictions on the
interpretation of the results when random assignment is used instead of random sampling.

Many Thanks
Many people have contributed to this book. We thank our students and colleagues at the
University of Wisconsin-Madison and Arizona State University and those instructors who
used the first three editions and provided valuable comments. We thank Roy Levy for his
review of Chapter 22, Introduction to Bayesian Statistics. Particular thanks are owed to Drs.
Timothy Baker, Janet Hyde, and Marilyn Essex for providing their data sets, which are avail-
able at the LFD website. Baker’s data examine the effectiveness of Zyban® and nicotine-
replacement gum on smoking. The Hyde and Essex data set measure the effects of having
a child on marriage. AMG thanks his instructors at the University of Michigan and Miami
University. MEA thanks his instructors at Temple University, especially Ralph Rosnow, Alan
Sockloff, and Phil Bersh. Thanks are due to the editorial and production staffs at Taylor &
Francis who tolerated delay after delay. Finally, thanks to Mina and Anna for their love and
support.
Arthur M. Glenberg
Matthew E. Andrzejewski
1 Why statistics?

1.1 Smoking kills 2
1.2 Variability 3
1.2.1 Understanding variability is the key to learning from data 3
1.2.2 Variability can come from many sources 4
1.2.3 Variables should be distinguished from constants 4
1.3 Populations and samples 5
1.3.1 A statistical population is a collection of scores that share a common
characteristic5
1.3.2 Most statistical populations are not available for us to use 6
1.3.3 Samples, especially random samples, assist scientists in learning from data 7
1.4 Descriptive and inferential statistics 7
1.4.1 Descriptive statistical procedures are used to organize and summarize data sets 7
1.4.2 Inferential statistical procedures allow us to make informed guesses
about populations 8
1.5 Measurement 9
1.5.1 Measurement is the process by which we collect data 9
1.5.2 Measurement is a human action and must be considered in a social and
political context 9
1.5.3 Measurement rules differ in important ways: validity, reliability, and
numerical properties 10
1.5.4 Numbers used as measurements can have different properties 11
1.5.5 If observations with the same number are in the same category, they
have the category property 11
1.5.6 Measurements have the ordinal property when larger numbers
correspond to more of the variable 11
1.5.7 The measurement scale has the equal intervals property when a
difference of 1 on the scale always corresponds to the same difference in
the variable 12
1.5.8 Measurements have the absolute zero property when the number 0
represents nothing of the variable being measured 13
1.5.9 The four properties of numbers result in four types of measurement
scales: nominal, ordinal, interval, and ratio 14

DOI: 10.4324/9781003025405-1
2 Why statistics?

1.5.10 The type of measurement scale has important implications 16

1.5.11 Computer programs and calculators should be used judiciously 16
1.6 Summary 18
1.7 Exercises 18
1.7.1 Terms 18
1.7.2 Questions 18
Notes19

1.1 Smoking kills
In your lifetime, you have probably heard countless times, from many sources, that “smoking
kills”: TV advertisements, teachers, doctors. Experts now say that about 480,000 people die
each year in the United States directly from the consequences of smoking and tobacco use.
They also tell us that these preventable illnesses and diseases cost our society billions of dol-
lars each year. You have probably also heard about someone’s grandfather who smoked every
day from the time he was 16 until his death at 96, or some similar story. How do we resolve
these contradictions? How do we know that “smoking kills”? On the one hand, authority
figures are telling us so, but our personal experience may be quite different. What is the best
way to learn about these matters?
Learning can result from critical thinking, asking the experts, or even from a religious
experience. However, collecting data (that is, measuring observations) is the surest way to
learn about how the world really is. And the data on smoking are compelling and clear: smok-
ers, on average, die earlier, have higher risk of cancer (over 25 times more likely), miss more
work, are more likely to have a stroke, and are more likely to have heart disease. In addition,
scientists are also beginning to know why your friend’s grandfather smoked all those years
with seemingly little effect (a likely explanation lies in genetics). We have learned these things
because of the steady, deliberative, and dedicated actions of the numerous scientists who
spent year after year systematically collecting data. Moreover, the terms “on average” and
“increase risk” are technical terms with important meanings; they require steady, deliberative,
and dedicated action on your part to understand their implications.
Data in the behavioral sciences are messy, unfortunately. For the beginning student of
statistics, initial examination of data reveals no clear facts about the world. Instead, the data
appear to be nothing but an incoherent jumble of numbers. To learn about the world from
data, you must first learn how to make sense out of data, and that is what this textbook will
teach you. Statistical procedures are tools for learning about the world by learning from data,
including the concepts of “average” and “risk.” From that learning, we can make better deci-
sions about our own lives and the lives of others we care about. For example, if someone you
care about smokes, how can we help them to quit? Scientists have carefully collected data
that tells us that many of the negative effects associated with smoking change for the better
after quitting, like the chance of stroke returning to that of a nonsmoker after about 5 years
of not smoking. Yet, quitting is very difficult. The drug in tobacco, nicotine, is highly addic-
tive, producing withdrawal symptoms and cravings that are relieved by smoking. It turns out,
psychologists have been studying the best way to quit smoking too.
To help you to understand the power and usefulness of statistical procedures, we will
explore two real (and important!) data sets throughout the course of the book. One of the
data sets is courtesy of Professor Timothy Baker at the University of Wisconsin Center for
Tobacco Research and Intervention (which we will call the “Smoking Study”). The data were
 Why statistics? 3

collected to investigate several questions about smoking, addiction, withdrawal, and how
best to quit smoking. The data set consists of a sample of 608 people who wanted to quit
smoking. These people were randomly assigned (see Chapter 14 for the benefits of random
assignment) to three groups. The participants in one group were given the drug bupropion
SR (Zyban) along with nicotine replacement gum. In a second group, the participants were
given the bupropion along with a placebo gum that did not contain any active ingredients.
The final group received both a placebo drug and a placebo gum. The major question of
interest is whether people are more successful in quitting smoking when the active gum is
added to the bupropion. These data are exciting for a couple of reasons. First, given the
tremendous social cost of cigarette smoking, we as a society need to figure out how to help
people overcome this addiction, and these data do just that. Second, the study included
measurements of about 30 other variables to help answer ancillary questions. For example,
there are data on how long people have smoked and how much they smoked; data on health
factors and drug use; and demographic data such as gender, ethnicity, age, education, and
height. These variables are described more fully within data files available for download and
in Appendix A. The statistical tools you will learn about will give you the opportunity to
explore these data to the fullest extent possible. You can ask important questions—some that
may never have been asked before—such as whether drug use affects people’s ability to quit
smoking, and you can get the answers. In addition, these data will be used to illustrate various
statistical procedures, and they will be used in the end-of-chapter exercises.
A second data set, courtesy of Professors Janet Hyde and Marilyn Essex of the University
of Wisconsin-Madison, is a subset of the data from the Wisconsin Maternity Leave and
Health Project and the Wisconsin Study of Families and Work (we will refer to it as the
“Maternity Study”). This project tried to answer questions about how having a baby affects
family dynamics such as marital satisfaction and how various factors affect child develop-
ment. There are measurements of 26 variables for 244 families. Some of these variables are
demographic, such as age, education, and family income. Marital satisfaction was measured
separately for mothers and fathers both before the child was born (during the fifth month of
pregnancy) and at three times after the birth (1, 4, and 12 months postpartum). There are
also data on how much the mother worked outside the house and how equally household
tasks were divided among the mothers and fathers. Finally, there are measures of the quality
of mother-child interactions and measures of child temperament measured when the child
was 4.5 years old. These variables are described more fully in Appendix B. As with the smok-
ing data, you are free to use these data to answer important questions, such as whether the
amount of time that a mother works affects child development.These data sets, as well as
many others that you will need for JASP examples, are available at the Learning from Data
website: www​.routledge​.com​/9780367457976.

1.2 Variability
1.2.1 Understanding variability is the key to learning from data

The first step in learning how to learn from data is to understand why data are messy. A con-
crete example is useful. Consider the CO_EOT (Carbon Monoxide content at the End of
Treatment) scores from the Smoking Study (see Appendix A). Each participant blew into a car-
bon monoxide detector (much like a breathalyzer) at the end of the study. Carbon Monoxide
(CO) content is a measure of recent cigarette smoking: people who have recently smoked have
high levels of CO in their breath as opposed to low levels in people who have not smoked
4 Why statistics?

recently. For the 428 participants for whom we have CO_EOT scores, the scores range from
0 to 58.5. Almost a quarter of the scores are below 1, but another quarter are above 10. These
data are messy in the sense that individual scores are very different from one another.

Variability is the statistical term for the degree to which scores (such as the CO_EOT scores)
differ from one another.

Chapter 3 presents statistical procedures for precisely measuring the variability in a set of
scores. For now, only an intuitive understanding of variability is needed. When the scores
differ from one another by quite a lot (such as carbon monoxide levels), variability is high.
When the scores have similar values, variability is low. When all the scores are the same, there
is no variability.

Section Check: What do we call the degree to which scores differ from one another in a data
set?

1.2.2 Variability can come from many sources

It is easy enough to see that the CO_EOT data are variable, but why are they variable? In
general, variability arises from several sources. One source of variability is individual differ-
ences: Some participants smoked more recently; some smoked unfiltered cigarettes; some
smokers’ bodies process the CO differently than others; some smokers inhale more deeply
than others. There are as many potential sources of variability due to individual differences
as there are reasons for why one person differs from another in intelligence, personality,
and physical characteristics. Another source of variability is the procedure used in collecting
the data. Perhaps some of the smokers were more rushed than others; perhaps some were
tested at the end of the day and after smoking all day. Any change in the procedures used for
collecting the data can introduce variability. Finally, some variability may be due to condi-
tions imposed on the participants, such as whether they are taking the nicotine replacement
gum. In other words, perhaps by using the gum, a smoker had fewer cravings and therefore
smoked less than others.

Section Check: What are some sources of variability?

1.2.3 Variables should be distinguished from constants

Variability does not occur only in textbook examples; it is characteristic of all data in the behav-
ioral sciences. Whenever a behavioral scientist collects data, whether on the incidence of depres-
sion, the effectiveness of a psychotherapeutic technique, or carbon monoxide level, the data
will be variable; that is, not all the scores collected will be the same. In fact, because data are
variable, collecting data is sometimes referred to as measuring a variable (or a random variable).

A variable is a measurement that changes from one observation to the next.

 Why statistics? 5

CO_EOT is a variable because it changes from one smoker (observation) to the next.
“Effectiveness of a psychotherapeutic technique” is another example of a variable, because a
given technique will be more effective for some people than for others.

Constants are measurements that stay the same from one observation to the next.

The boiling point of pure water at sea level is an example of a constant. It is always 100
degrees Centigrade. Whether you use a little water or a lot of water, whether the water is
encouraged to boil faster or not, no matter who is making the observation (as long as the
observer is careful!), the water always boils at the same temperature. Another constant is
Newton’s gravitational constant, the rate of acceleration of an object in a gravitational field
(whether the object is large or small, solid, or liquid, and so on).
Many of the observations made in the physical sciences are observations of constants.
Because of this, it is easy for the beginning student in the physical sciences to learn from
data. A single careful observation of a constant tells the whole story. You may be surprised
to learn that there is not one constant in all of the behavioral sciences. There is no such
thing as the effectiveness of a psychotherapeutic technique, or the carbon monoxide content,
because measurements of these variables change from person to person. In fact, because what
is known in the behavioral sciences is always based on measuring variables, even the begin-
ning student must have some familiarity with statistical procedures to appreciate the body of
knowledge that comprises the behavioral sciences and the limitations inherent in that body of
knowledge. In case you were wondering, this is why you are taking an introductory statistics
course, and your friends majoring in the physical sciences are not. The concept of variability
is absolutely basic to statistical reasoning, and it will motivate all discussions of learning from
data. In fact, the remainder of this chapter introduces concepts that have been developed to
help cope with variability.

Section Check: What is the difference between a constant and a variable?

1.3 Populations and samples

1.3.1 A statistical population is a collection of scores that share a common characteristic

The psychologists studying addiction might be interested in the CO_EOT scores of the
specific smokers from whom they collected data. However, it is likely that they are interested
in more than just those individuals. For example, they may be interested in the amount of
carbon dioxide in the breath among all smokers in Wisconsin, or all smokers in the United
States, or even all smokers in the world. Because CO level is a variable that changes from per-
son to person, the specific observations cannot reveal everything the researchers might want
to know about all of these CO scores.

A statistical population is a collection or set of measurements of a variable that share some

common characteristic.

One example of a population is the set of CO_EOT scores of all smokers in Wisconsin.
These scores are measurements of a variable (CO_EOT), and they have the common
6 Why statistics?

characteristic of being from a particular group of people: smokers in Wisconsin. A different

statistical population consists of all the CO_EOT scores for smokers in the United States.
And a very different population consists of the marital satisfaction scores for new mothers
who work full-time outside of the home. The point is that you should not think of sta-
tistical populations as groups of people, such as the people in the United States. There is
only one population of people for the United States, but there are an infinite number of
statistical populations depending on what variables are measured (for example, CO_EOT
or marital satisfaction), and how those scores might be grouped (for example, smokers or
working mothers).
Thinking of statistical populations as sets of numbers can cause some people to become
bored and lose interest in the enterprise. The way to counter this boredom is to remember
that the statistical procedures are operating on numbers that have meaning: The numbers are
scores that represent something interesting about the world. As you read through this book,
think about applying your new knowledge to problems that are of interest to you, and not
just as manipulation of numbers.

Section Check: Why should we consider statistical populations sets of scores rather than groups
of people?

1.3.2 Most statistical populations are not available for us to use

Some statistical populations consist of a manageable number of scores. Usually, however,

statistical populations are very large. For example, there are potentially millions of CO_
EOT scores of smokers. When dealing with large populations, it is time consuming and
expensive to actually collect all of the scores. Sometimes, for ethical reasons, all the scores
in the population cannot be obtained. For example, suppose that a medical researcher
believes that she has discovered a drug that reduces high blood pressure. One way to
determine the drug’s effectiveness is to administer it to all people suffering from high
blood pressure and then to measure their blood pressures. Clearly, this would be time
consuming and expensive. It would also be very unethical. After all, what if the medical
researcher were wrong, and the drug raised blood pressure, endangering those who took
it? Also, even with a great national effort, not all the scores could be collected, because
some of the people would die before they took the drug, others would have their blood
pressures lowered by other drugs, and others would develop high blood pressure over
the course of data collection.
We appear to have run across a problem. Usually, we are not interested in just a few scores,
but in all the scores in a population. Yet, because behavioral scientists are interested in learn-
ing about variables (not constants), it is impossible to know for sure about all the scores in
a population from measuring just a few of them. On the other hand, it is time consuming
and expensive to collect all the scores in a population, and it may be unethical or impossible.
What to do?

Section Check: Why is it difficult to work with statistical populations?

 Why statistics? 7

1.3.3 Samples, especially random samples, assist scientists in learning from data

The solution to the problem of populations is provided by statistical procedures based on

obtaining samples of scores.

A sample is a subset of measurements from a population.

That is, a sample contains some, but not all, of the scores in the population. The 428 CO_
EOT scores are a sample from the population of CO_EOT scores of all smokers. When
samples are obtained in such a way as to be “random,” special statistical procedures can be
applied and conclusions can be drawn.

A random sample is selected so that every score in the population has an equal chance of being
included in the sample, AND the selection of one score does not influence the likelihood of any
other score being included.

Whether a sample is random or not does not depend on the actual scores included in the sam-
ple, but on how the scores in the sample are selected. Only if the scores are selected in such
a way that each score in the population has an equal chance of being included in the sample
is the sample a random sample. The CO_EOT scores are not a random sample of CO_EOT
scores of all smokers. These scores are only from people living in Madison and Milwaukee,
Wisconsin, and there was no attempt to ensure that CO_EOT scores of people living else-
where were included. Procedures for producing random samples are discussed in Chapter 5.
As you will see in Chapters 5–22, random samples are used to help solve the problem of
large populations. That is, with the data in a random sample, we can learn about the popula-
tion from which the sample was obtained by using inferential statistical procedures.

Section Check: What are the two key features of random samples?

1.4 Descriptive and inferential statistics

1.4.1 Descriptive statistical procedures are used to organize and summarize data sets

Because of variability, to learn anything from data, the data must be organized.

Descriptive statistical procedures are used to organize and summarize the measurements in
samples and populations.

In other words, descriptive statistical procedures do what the name implies—they describe
the data. These procedures can be applied to samples and to populations. Most often, they
are applied to samples, because it is rare to have all the scores in a population. Descriptive sta-
tistical procedures include ways of ordering and grouping data into distributions (discussed
in Chapter 2) and ways of calculating single numbers that summarize the whole set of scores
in the sample or population (discussed in Chapters 2 and 3). Some descriptive statistical pro-
cedures are used to represent data graphically, because as everyone knows, a picture is worth
a thousand words.
8 Why statistics?

Section Check: What are the goals of descriptive statistics?

1.4.2 Inferential statistical procedures allow us to make informed guesses about populations

The most powerful tools available to the statistician are inferential statistical procedures.

Inferential statistical procedures are used to make educated guesses (inferences) about popu-
lations based on random samples from the populations.

These educated guesses are the best way to learn about a population short of collecting all
of the scores in the population. All of this may sound a bit like magic. How can you possibly
learn about a whole population that may contain millions and millions of scores by examin-
ing a small subset of scores contained in a random sample? It is not magic, however, and it
is even understandable. Part II of this book presents a detailed description of how inferential
statistical procedures work.
Inferential statistical procedures are so pervasive in our society that you have undoubt-
edly read about them and made decisions based on them. For example, think about the last
time you heard the results of an opinion poll, such as the percentages of the registered voters
who favor Candidates A, B, or C. Supposedly, your opinion is included in those percentages
(assuming that you are a registered voter so that your opinion is included in the population).
But on what grounds does the pollster presume to know your opinion? It is a safe bet that
only rarely, if ever, has a pollster actually contacted you and asked for your opinion. Instead,
the percentages reported in the poll are educated guesses based on inferential statistical pro-
cedures applied to a random sample.
In recent years, it has become fashionable for the broadcast and print media to acknowl-
edge that conclusions from opinion polls are educated guesses (rather than certainties).
This acknowledgment is in the form of a “margin of error.” The “margin of error” is how
much the reported percentages may differ from the actual percentages in the population (see
Chapter 10 for details).
Another example of the impact of inferential statistical procedures on our daily lives is in
our choices of foods and medicines. Many new food additives and medicines are tested for
safety and approved by government agencies such as the Food and Drug Administration
(FDA). But how does the FDA know that the new product is safe for you? In fact, the FDA
does not know for sure. The decision that a new drug is safe is based on inferential statistical
procedures. The FDA example raises several sobering issues about the data used by govern-
ment agencies to set standards on which our lives literally depend. It is only recently that
government agencies have insisted that data be collected from women, and without such
data, it is uncertain if a particular drug is actually safe or effective for women. The terrible
birth defects attributed to the drug Thalidomide occurred because no one had bothered to
collect the data that would verify the safety of the drug with pregnant women. Similarly, very
little data on safe levels of environmental pollutants such as PCBs and pesticides have been
collected from children. Consequently, our society may be setting the scene for a disaster by
allowing into the environment chemicals that are relatively safe for adults but disastrous for
children whose immune systems are immature and whose rapidly developing brains are sensi-
tive to disruption by chemicals.1
 Why statistics? 9

The final example of the use of inferential procedures is the behavioral sciences themselves.
Most knowledge in the behavioral sciences is derived from data. The data are analyzed using
inferential statistical procedures because interest is not confined to just the sample of scores,
but extends to the whole population of scores from which the sample was selected. If you are
to understand the data of the behavioral sciences, then you need to understand how statistical
procedures work.

Section Check: Why do we need inferential statistical procedures?

1.5 Measurement
1.5.1 Measurement is the process by which we collect data

Data are collected by measuring a variable. But what does it mean to measure a variable?

Measurement is the use of a rule to assign a number to a specific observation of a variable.

As an example, think about measuring the length of your desk. The rule for measuring length
is, “Assign a number equal to the number of lengths of a standard ruler that fit exactly from
one end of the desk to the other.” In this example, the variable being measured is “length.”
The observation is the length of a specific desk, your desk. The rule is to assign a value (for
example, 4 feet) equal to the number of lengths of a standard ruler that fit from one end of
the desk to the other. As another example, consider measuring the weight of a newborn baby.
The variable being measured is weight. The specific observation is the weight of the specific
baby. The measurement rule is something like, “Put the baby on one side of a balance scale
and assign to that baby a weight equal to the number of pound weights placed on the other
side of the scale to get the scale to balance.”
Measuring variables in the behavioral sciences also requires that we use a rule to assign
numbers to observations of a variable. For example, one way to measure marital satisfaction
is to assign a score equal to the sum of the ratings on a questionnaire. The variable is marital
satisfaction (e.g., the variable “MARSAT” in the maternity study), the specific observation is
the marital satisfaction of the person being assessed, and the rule is to assign a value equal to
the sum of the ratings. Similarly, measuring intelligence means assigning a number based on
the number of questions answered correctly on an intelligence test.

Section Check: For the purposes of statistics, what is meant by the term “measurement”?

1.5.2 Measurement is a human action and must be considered in a social and political context

The choice of what variables to measure in a study is no accident; usually those choices entail
a lot of discussion and planning and are often influenced by social or political motives of the
researcher. The measurement rules, as well, usually involve much discussion, but the details
are rarely stated in a study’s results. At the very least, there’s usually some ambiguity. Take, for
example, the LONG variable in the Smoking Study, which measures the longest time without
10 Why statistics?

smoking. Let’s say that a study participant answers “8 months,” which would result in a score of
7 (6–12 months). But, if we probe further, we may find that the participant actually answered:
“Well, I didn’t smoke for 4 months, but then one night I had one cigarette, and then didn’t
have another for 4 months. I say 8 months because it was just a minor slip-up.” Is the longest
time without smoking for this individual 8 months or 4 months? Is “smoking” defined as “one
cigarette” or “one drag” or “buying a pack”? If the researcher is interested in the effectiveness
of a particular antismoking program, she may give this participant “a break” and count it as 8
months because, clearly, to her, this participant didn’t relapse (it was only one cigarette, after
all). A different researcher, interested in showing that all addicts wind up using again (relapsing)
might say that one cigarette constitutes a relapse, and the measurement should be 4 months.
Political motives may enter a study in this way because for some people the only solution for
drug addiction may be abstinence (for example, Alcoholics Anonymous), but for others, rec-
reational drug use may be seen as OK in certain situations (for example, “harm reduction”
approaches). In addition, a researcher’s grant funding may be dependent on having and solving
a social problem, and maybe even a “growing problem,” even though the “problem” is not as
big as one might think. Note how these factors introduce variability into data, as mentioned
earlier. Nonetheless, we should remain critical of how psychologists measure and contemplate
what might have been included and what might have been left out.

Section Check: What are some of the factors that can affect both what is measured and the
rules used for measuring variables?

1.5.3 Measurement rules differ in important ways: validity, reliability, and numerical

properties

All rules for measuring variables are not equally good. They differ in three important ways.
First, they differ in validity.

Validity refers to how well the measurement rule actually measures the variable under con-
sideration as opposed to some other variable.

Some intelligence tests are better than others because they measure intelligence rather than
(accidentally) being influenced by creativity or memory for trivia. Similarly, some measures
of depression are better than others because they measure depression rather than introversion
or aggressiveness.
Measurement rules also differ in reliability.

Reliability is an index of how consistently the rule assigns the same number to the same
observation.

For example, an intelligence test is reliable if it tends to assign the same number to the person
each time the person takes the test. Books on psychological testing discuss validity and reli-
ability in detail.2
Finally, a third difference among measurement rules is that the properties of the numbers
assigned as measurements depend on the rule. At first blush, this statement may sound like
nonsense. After all, numbers are numbers; how can their properties differ?
 Why statistics? 11

Section Check: Define reliability and validity.

1.5.4 Numbers used as measurements can have different properties

We see and use numbers in a variety of ways. We call a phone number, or write down a zip
code on an envelope, or wear a jersey with our favorite player’s number. We rank our favorite
movies or count how many first places (gold medals) a country won at the Olympics. We
look up the temperature outside or check to see what size a dress is before we try it on. We
measure out a certain amount of flour for a cookie recipe or check the balance in our bank
account. But not all numbers are created equal. Some carry more information than others,
while some can be legitimately used in computations, while others cannot. The ways that
numbers differ is based on the properties of numbers. When numbers are measurements,
they can have four properties: 1) category, 2) ordinal, 3) equal intervals, and 4) absolute
zero. The point is, how we interpret the measurements depends on the properties of the
numbers, which in turn depend on the rule used in assigning the numbers.

Section Check: What are the four properties of numbers?

1.5.5 If observations with the same number are in the same category, they have the category
property

The category property is that observations assigned the same number are in the same cat-
egory, and observations assigned different numbers are in different categories.
For example, suppose that you are collecting the zip codes of people in a study. If a person
answers “13088” you can infer that they are from Liverpool, NY. Other people who provide
the answer “13088” are also part of the category “from Liverpool, NY.” People who provide
the answer “27502” are not part of the category “from Liverpool, NY” but are part of dif-
ferent category, specifically “from Apex, NC.” However, the number 27502 does not mean
MORE zip-code-stuff than 13088. The zip code 27502 is not larger than, or heavier than, or
taller than the zip code 13088; it is just a different category. Area codes also have the category
property as do the numbers on our favorite player’s jersey. In psychology, it is common to
assign a number to a category, like assigning a “1” for female and “2” for male. We refer to
this as coding the data.

Section Check: Describe another example of a measurement with the category property.

1.5.6 Measurements have the ordinal property when larger numbers correspond to more of the
variable

The ordinal property refers to the fact that the numbers can be used to order the observations
from those that have the least of the variable being measured to those that have the most.
12 Why statistics?

Consider another example. Suppose that a social psychologist investigating cooperation

has a preschool teacher rank the four pupils in the class from least cooperative (first) to
most cooperative (fourth). These cooperation scores (ranks) have two properties. First, the
scores have the category property, because children assigned different scores are in different
categories of cooperation. Second, the scores have the ordinal property because the scores
can be used to order the observations from those that have the least to those that have the
most cooperation. It is only when measurements have the ordinal property that we know that
observations with larger measurements have more of whatever is being measured.

Section Check: What does it mean to have the ordinal property?

1.5.7 The measurement scale has the equal intervals property when a difference of 1 on the
scale always corresponds to the same difference in the variable

The equal intervals property means that whenever two observations are assigned measure-
ments that differ by exactly one unit, there is always an equal interval (difference) between
the observations in the actual variable being measured.
To understand what is meant by equal intervals, consider again measuring the coopera-
tiveness of the four preschool children. The four children (call them Alana, Bob, Carol, and
Dan) have cooperation scores of 1, 2, 3, and 4. The difference between Alana’s cooperation
score (1) and Bob’s cooperation score (2) is 1. Likewise, the difference between Carol’s
cooperation score (3) and Dan’s cooperation score (4) is 1. But, is the actual difference in
cooperation (not just the score) between Alana and Bob equal to the actual difference in
cooperation between Carol and Dan?
It is very unlikely that the difference in cooperation between Alana and Bob exactly equals
the difference in cooperation between Carol and Dan. The teacher simply ranked the chil-
dren from least to most cooperative. The teacher did not take any precautions to ensure equal
intervals. Alana and Bob may both be very uncooperative, with Bob being just a bit more
cooperative than Alana (the actual difference in cooperation between Alana and Bob is a
“bit”). Carol may also be on the uncooperative side, but just a bit more cooperative than Bob
(the actual difference between Carol and Bob is a “bit”). Suppose, however, that Dan is the
teacher’s helper and is very cooperative. In this case, the difference in cooperation between
Carol and Dan may be very large, much larger than the difference in cooperation between
Alana and Bob. Because the differences in scores are equal (the difference in cooperation
scores between Alana and Bob equals the difference in cooperation scores between Carol
and Dan), but the differences in amount of cooperation (the variable) are not equal, these
cooperation scores do not have the equal interval property.
Now consider using a ruler to measure the lengths of the four lines in Figure 1.1. The lines
A, B, C, and D have lengths of 1, 2, 3, and 6 centimeters, respectively. Using a ruler to meas-
ure length generates measurements with the equal intervals property: For each pair of obser-
vations for which the measurements differ by exactly one unit, the differences in length are
exactly equal. That is, the measurements assigned lines A (1) and B (2) differ by one, as do the
measurements assigned lines B (2) and C (3); and importantly, the actual difference in lengths
between lines A and B exactly equals the actual difference in length between lines B and C. ​
A difficulty in understanding the equal intervals property is in maintaining the distinction
between the variable being measured (length or cooperation) and the number assigned as a
 Why statistics? 13

Length Length
measured measured
using ranks using ruler

1 1 A
2 2 B
3 3 C
4 6 D

Figure 1.1 Length of four lines measured using two different measurement rules.

measurement of the variable. The numbers represent or stand for certain properties of the
variable. The numbers are not the variable itself. The number 1 is no more the cooperation
of Alana (it is a measure of her cooperation) than is the number 1 the actual length of line A
(it is a measure of its length). Whether or not the measurements have properties such as equal
intervals depends on how the numbers are assigned to represent the variable being measured.
Using a ruler to measure length of a desk assigns numbers that have the equal intervals prop-
erty; using rankings to measure cooperation of preschool children assigns numbers that do
not.
The difference between the length and cooperation examples is not in what is being
measured, but in the rule used to do the measuring. A ranking rule can be used to meas-
ure the lengths of lines (this is what we do when we need a rough measure of length—
compare two lengths to see which is longer). In this case, the measured lengths of lines A,
B, C, and D would be 1, 2, 3, and 4, respectively (see Figure 1.1). These measurements
of length do not have the equal intervals property, because for each pair of observations
for which the measurements differ by exactly one unit, the real differences in length are
not exactly equal.

Section Check: Define the equal intervals property in your own words.

1.5.8 Measurements have the absolute zero property when the number 0 represents nothing of
the variable being measured

The absolute zero property means that a value of 0 is assigned as a measurement only when
there is nothing at all of the variable that is being measured.
When length is measured using a ruler (rather than ranks), the score of 0 is an absolute
zero. That is, the value of 0 is assigned only when there is no length. When measuring zip
code, 00000 may be assigned, but does not mean “no zip code.” In addition, someone
may buy a woman’s size 0 dress, but this does not mean “no dress”, it is just a really small
dress.
Another example of a measurement scale that does not have an absolute zero is the
Fahrenheit (or Centigrade) scale for measuring temperature. A temperature of 0°F does not
mean that there is no heat. In fact, there is still some heat at temperatures of –10°F, –20°F,
and so on. Because there is still some heat (the variable being measured) when 0 is assigned
as the measurement using Fahrenheit or Centigrade scales, the 0 is not an absolute zero.
14 Why statistics?

Section Check: What does a measurement of 0 mean when the scale does not have an absolute
zero?

1.5.9 The four properties of numbers result in four types of measurement scales: nominal,
ordinal, interval, and ratio

“Nominal” comes from the word name. The numbers assigned using a nominal scale name
the category to which the observation belongs but indicate nothing else. Thus, the measure-
ments of zip code form a nominal scale, because the numbers name the category (zip code
location) but have no other properties.
Several of the variables in the Smoking Study are measured using nominal scales. For
example, TYPCIG (type of cigarette smoked) is measured using a nominal scale defined as 1
= regular filter; 2 = regular no filter; 3 = light; 4 = ultralight; 5 = other. Another nominally
measured variable is SPOUSE, that is, whether the smoker’s spouse smokes (1) or does not
smoke (0). The GENDER variable in the Maternity Study (is the child male or female) is also
measured using a nominal scale.
An ordinal scale is formed when the measurement rule assigns numbers that have the
category and the ordinal properties, but no other properties.
Many of the variables in the Smoking and Maternity studies are measured using ordinal
scales. The longest time without smoking (LONG) variable is measured as 1 = less than a day;
2 = 1–7 days; 3 = 8–14 days; 4 = 15 days to a month; 5 = 1–3 months; 6 = 3–6 months; 7 =
6–12 months; 8 = more than a year. As the assigned score increases from 1 to 8, the length
of time without smoking increases, so the numbers have the ordinal property. However, the
difference between a measurement of 1 and 2 (LONG 1 – LONG 2 = about 3 days) is not
comparable to a difference between a measurement of, say, 5 and 6 (LONG 5 – LONG 6 =
about 3 months), thus the measurements do not have the equal intervals property.
Many behavioral scientists (and businesses that conduct marketing research) collect data
by having people rate observations for specific qualities. For example, a clinical psycholo-
gist may be asked to rate the severity of his patients’ psychopathologies on a scale from 1
(extremely mild) to 10 (extremely severe). As another example, a consumer may be asked
to rate the taste of a new ice cream from 1 (awful) to 100 (sublime). In both cases, the
measurements represent ordinal properties. For the clinical psychologist, the larger numbers
represent more severe psychopathology than the smaller numbers; for the ice-cream raters,
the larger numbers represent better-tasting ice cream than the smaller numbers. In neither
example, however, do the measurements have the equal intervals property. As a general rule,
ratings and rankings form ordinal scales.
An interval scale is formed when numbers have the category, nominal, and equal inter-
vals properties, but not the absolute zero property. Two examples of interval scales are the
Fahrenheit and Centigrade scales of temperature. Neither has an absolute zero because 0° (F
or C) does not mean absolutely no heat. The measurements do have the category property
(all observations assigned the same number have the same amount of heat), the ordinal prop-
erty (larger numbers indicate more heat), and the equal intervals property (on a particular
scale, a difference of 1° always corresponds to a specific amount of heat).
Many psychological variables are measured using scales that are between ordinal and interval
scales. This statement holds for many of the variables included in the Maternity Study, such
 Why statistics? 15

as marital satisfaction (for example, M1MARSAT), mother’s positive affect during free play
(MPOS), infant dysregulation during free play (IDYS), and child’s internalizing behavior during
free play (M7INT). Consider M7INT in a little more detail. To measure the variable, a mother
was asked to rate her child’s behavior in regard to nine questions such as, “Tends to be fearful or
afraid of new things or new situations.” The rating scale was 0 = does not apply; 1 = sometimes
applies; 2 = frequently applies. Thus, the rating of each question forms an ordinal scale without
the equal intervals property. But what happens when we sum the ratings from nine questions to
get the M7INT score? It is unlikely that the difference in internalizing behavior between M7INT
10 and M7INT 11 is exactly the same as the difference between, say, M7INT 20 and M7INT 21.
Nonetheless, it may well be that these two differences in internalizing behavior are fairly compa-
rable, that is, that the scale is close to having the equal intervals property.
The conservative (and always correct) approach to these “in-between” scales is to treat
them as ordinal scales. As we will see in Part II, however, ordinal scales are at a disadvantage
compared to interval scales when it comes to the range and power of statistical techniques
that can be applied to the data. Recognizing this disadvantage, many psychologists treat the
data from these in-between scales as interval data; that is, they treat the data as if the measure-
ments were collected using an interval scale. One rule of thumb is that scores from the middle
of an in-between scale are more likely to have the equal intervals property than scores from
either end. If the data include scores from the ends of an in-between scale, it is best to treat
the data conservatively as ordinal.
Scales for measuring physical qualities (length, weight, time) are often ratio scales. A ratio
scale is formed when the numbers assigned by the measurement rule have all four properties:
category, ordinal, equal interval, and absolute zero.
The reason for the name “ratio” is that statements about ratios of measurements are mean-
ingful only on a ratio scale. It makes sense to say that a line that is 2.5 centimeters long is half
(a ratio) the length of a 5-centimeter line. Similarly, it makes sense to say that 20 seconds is
twice (a ratio) the duration of 10 seconds.
On the other hand, it does not make sense to say that 68°F is twice as hot as 34°F. This
is easily demonstrated by converting to Centigrade measurements. Suppose that the tem-
perature of Object A is 34°F (corresponding to 1°C) and that the temperature of Object B
is 68°F (corresponding to 20°C). Comparing the amount of heat in the objects using the
Fahrenheit measurements seems to indicate that Object B is twice as hot as Object A, because
68 is twice 34. Comparing the measurements on the Centigrade scale (which of course does
not change the real amount of heat in the objects), it seems that Object B is 20 times as hot
as Object A. Object B cannot be 20 times as hot as Object A and at the same time be twice
as hot. The problem is that statements about ratios are not meaningful unless the measure-
ments are made using a ratio scale. Neither ratio (2:1 or 20:1) is right, because neither set of
measurements was made using a ratio scale.
This problem does not occur when using a ratio scale. A 5-centimeter (2-inch) line is twice
as long as a 2.5-centimeter (1-inch) line, and that is true whether the measurements are made
in centimeters, inches, or any other ratio measurement of length.
Several variables in the Smoking Study are measured using ratio scales. One example is the
carbon monoxide level at the end of treatment measured in parts per million (CO_EOT),
and another is the number of times the participant has tried to quit smoking (QUIT).

Section Check: Describe a new example of each measurement scale.

16 Why statistics?

1.5.10 The type of measurement scale has important implications

The question that may be uppermost in your mind is, “So what?” There are three reasons
why knowing about scale types is important. First, now that you know about scale types
you will be less likely to make unsupportable statements about data. One such statement is
the use of ratio comparisons when the data are not measured using a ratio scale. For exam-
ple, if your area code is 608, you would not say that you have twice as much area code as
someone with 304. Similarly, knowing just which medal was won (an ordinal scale), you
cannot say how much faster the gold medal winner of a race was compared to the silver
medal winner.
Second, the types of descriptive statistical procedures that can be applied to data depend in
part on the scale type. For example, the mathematical operations of finding an average only
make sense for interval or ratio data that have the equal intervals property.
Third, the types of inferential statistical procedures that can be applied to data depend in
part on the measurement scale.
Given these three reasons, it is clear that if you want to learn from data you must be able
to determine what sort of scale was used in collecting the data. The only way to know the
scale type is to determine the properties of the numbers assigned using that rule. If the only
property of the measurements is the category property, then the data are nominal; if the
measurements have both the category and ordinal properties, then the data are ordinal; if, in
addition, the data have the equal interval property, then the data are interval. Only if the data
have all four properties are they ratio.
Now that you understand the importance of scale types, it may be helpful to read this
section again. Your ability to distinguish among scale types will be used throughout this
textbook and in all of your dealings with behavioral data. Note that many calculators and
computers do not automatically distinguish between scale types. These devices treat “1” as
a quantity, with all the numerical properties described above, even though the “1” may be a
coded variable for gender and have only the category property.

Section Check: What are three reasons to differentiate between measurement scales?

1.5.11 Computer programs and calculators should be used judiciously

Statistical analysis programs, like SPSS, Excel, or JASP, have two main benefits. First, they
eliminate the drudgery of doing lots of calculations. Second, they ensure accuracy of calcula-
tion. A benefit that flows from these two is that the programs make it easy to explore data by
conducting multiple analyses.
Everything else that is important is not done by the statistical programs, however. The
essence of statistical analysis is choice (choosing the right statistical method—see the end-
papers in the front and back of the book—and interpretation of the outcome of the chosen
method). The programs cannot choose the appropriate methods for you. Similarly, the pro-
grams do not know whether a data set is a sample, a random sample, or a population. Nor can
they decide whether a scale is nominal, ordinal, interval, or ratio. Consequently, the program
 Why statistics? 17

cannot adequately interpret the output. Learning From Data teaches you how to make good
choices and how to interpret the outcome of the statistical methods; the computer eliminates
the drudgery.
Because the computer program does the calculations, you might think that you can ignore
the formulas in the text. That would be a big mistake for several reasons. First, for small sets
of data it is easier to do calculations by hand (or using a calculator) rather than using a com-
puter. But to do the calculations by hand, you need to know the formulas. Second, following
the formulas is often the best way to figure out exactly what the statistical technique is doing
and how it works. Working through the formulas can be hard intellectual labor, but that is
the only way to understand what they do.
For this edition of Learning From Data, we have made available a number of data sets
to practice techniques, explore, and literally, learn from data (see www​.routledge​.com​
/9780367457976). We also focus on the program JASP, available for download from
the web. It is user-friendly, cross platform (available for PC, Mac, and Linux), and free
to use. We have provided a number of instructions and screen captures describing its use
(Figure 1.2). ​

Section Check: What can computers help you do? What can’t they do?

Figure 1.2  JASP introductory screen.

18 Why statistics?

1.6 Summary
The behavioral sciences are built on a foundation of data. Unfortunately, because behavio-
ral data consist of measurements of variables, individual measurements will differ from one
another so that no clear picture is immediately evident. Fortunately, we can learn from vari-
able data by applying statistical procedures.
Descriptive statistical procedures organize, describe, and summarize data. Descriptive
statistical procedures can be applied to samples or to populations, but because we rarely
have all the scores in a population, descriptive procedures are generally applied to data from
samples. We use inferential statistical procedures to make educated guesses (inferences)
about a population of scores based on a random sample of scores from the population.
Although these inferences are not error-free, appropriate use of inferential statistical pro-
cedures can reduce the influence of error to acceptable levels (for example, the margin of
error in a poll).
The appropriateness of a statistical procedure depends in part on the type of measurement
scale used in collecting the data. The measurement scale is determined by the properties of
the numbers (assigned by the measurement rule). If the measurements have the category,
ordinal, equal interval, and absolute zero properties, then a ratio scale is formed; if the meas-
urements have all but the absolute zero property, an interval scale is formed. If the measure-
ments have only the category and ordinal properties, they form an ordinal scale. Finally, if the
measurements have only the category property, they form a nominal scale.

1.7 Exercises
1.7.1 Terms

Define these new terms.

variable reliability
measurement category property
constant ordinal property
sample equal intervals property
random sample absolute zero property
population nominal scale
descriptive statistical procedure ordinal scale
inferential statistical procedure interval scale
validity ratio scale

1.7.2 Questions
Q
Answer the following questions.

1. Why would there be no need for descriptive or inferential statistical procedures

if behavioral scientists could measure constants instead of variables?
2. List ten different variables and one constant in the behavioral sciences.
3. Classify each of the following as a population, a sample, or both. When the
answer is both, describe the circumstances under which the data should be con-
sidered a population and under which they should be considered a sample.
 Why statistics? 19

a. Family incomes of all families in the United States.

b. †Family incomes of all families in Wisconsin.
c. The number of words recalled from a list of 50 words by 25 first-year college
students who volunteer to take part in an experiment.
d. The number of days spent in intensive care for all people who have under-
gone heart transplant surgery.
e. The number of errors made by rats learning a maze.
4. Describe two examples of each of the four types of measurement scales. Indicate
why each is an example of its type.
5. †If you had a choice between using nominal, ordinal, interval, or ratio scales to
measure a variable, what would be the best choice? Why?
6. A set of scores can be one type of scale or another, depending on what the set
of scores represents. Consider the number of errors made by rats in learning a
maze. If the data represent simply the number of errors, then the scores form
a ratio scale. The numbers have all four properties, and it makes perfectly good
sense to say that if Rat A made 30 errors and Rat B made 15 errors, then Rat A
made twice as many errors as Rat B. Suppose, however, that the scores are used
as a measure of rat intelligence. Are these scores a ratio measure of intelligence?
Explain your answer. What are some of the implications of your answer?
7. Determine the type of measurement scale used in each of the following
situations:
a. A supervisor ranks his employees from least to most productive.
b. †Students rate their statistics teacher’s teaching ability using a scale of 1
(awful) to 10 (magnificent).
c. A sociologist classifies sexual preference as 0 (heterosexual), 1 (homosexual),
2 (bisexual), 3 (asexual), 4 (other).
d. A psychologist measures the time to complete a problem-solving task.

†Questions with this symbol have answers in the back of the book.

Notes
1 For an excellent discussion of these issues, see C. F. Moore (2003), Silent scourge. New York:
Oxford University Press.
2 A classic text is A. Anastasi (1988), Psychological testing (6th ed.). New York: Macmillan.
2 Frequency distributions and percentiles

2.1 Frequency distributions 21
2.1.1 Student loan debt 21
2.2 Frequency distributions 22
2.2.1 Frequency distributions summarize data by categorizing and counting 22
2.2.2 Benefits of frequency distributions 23
2.2.3 Relative frequency distributions display the proportions of observations
of a particular score value 24
2.2.4 Relative frequency distributions provide distinct advantages over
frequency distributions 24
2.2.5 Cumulative frequency distributions tabulate all measurements at or
smaller than a particular value 25
2.2.6 Cumulative relative frequency distributions tabulate the proportion of
scores at or smaller than a particular score value 25
2.2.7 Using JASP to construct frequency distributions 26
2.3 Grouped frequency distributions 27
2.3.1 Grouped frequency distributions are used when there are many score values 27
2.3.2 Constructing grouped distributions takes some planning 28
2.4 Graphing frequency distributions 29
2.4.1 Graphing frequency distributions is an important way to show your data 29
2.4.2 Histograms represent frequencies as heights of bars 29
2.4.3 Using JASP to create grouped frequency histograms 30
2.5 Describing distributions 31
2.5.1 Distributions can be described and characterized in three major ways 31
2.5.2 Symmetry and skew are ways in which the shape of a distribution can be
characterized32
2.5.3 Modality refers to the number of modes, or peaks, in the distributions 33
2.5.4 Measures of central tendency are near the center of the distribution 34
2.5.5 Measures of variability describe how much scores differ from one
another in a distribution 34
2.5.6 Using the three major characteristics to compare and contrast different
distributions35
2.6 Percentiles 36
2.6.1 Percentiles define a measurement’s relative standing in the distribution 36

DOI: 10.4324/9781003025405-2
 Frequency distributions and percentiles 21

2.6.2 Percentile ranks and percentiles 37

2.6.3 Three precautions when using percentile ranks 37
2.6.4 Boxplots using JASP 38
2.7 Summary 39
2.8 Exercises 41
2.8.1 Terms 41
2.8.2 Questions 41

2.1 Frequency distributions
2.1.1 Student loan debt

You have likely heard about the student loan debt “crisis” in the United States. Media outlets
and social media platforms seem to be filled with stories of huge loans, defaults, and bank-
ruptcies. Would you be surprised to hear, though, that about 75% of people with student
debt owe less than $40,000? Did you know that about one in three people owe less than
$10,000? In addition, did you know that of the people that owe over $100,000 (only about
7% of people with student debt), the vast majority owe this for graduate and professional
school (like medical and law school)? Table 2.1 shows the distribution of amount owed by
people in the United States in 2021 according to the Office of Federal Student Aid.
Table 2.1 was constructed by “categorizing and counting,” the essence of constructing a
frequency distribution. All people that have outstanding student loan debt were divided into
nine categories (for example “less than $5,000”) and the number of people that fit into that

Table 2.1 Frequency distributions of amount owed

Amount Owed Number of People (in millions) Percentage (%)

Less than $5,000 7.3 16
$5,001–$10,000 7.5 17
$10,001–$20,000 9.4 21
$20,001–$40,000 9.6 21
$40,001–$60,000 4.3 9
$60,001–$80,000 2.6 6
$80,001–$100,00 1.4 3
$100,001–$200,000 2.4 5
$200,000 + 0.9 2
Total 45.4 100

Table 2.2 A hypothetical aggressiveness scale

Meaning Score Value

potential for violence 5
very aggressive 4
somewhat aggressive 3
average 2
timid 1
very timid 0
22 Frequency distributions and percentiles

category were counted. Percentages were computed by taking the total number in a group or
category, dividing that by the total number of people with student loan debt (roughly 45.4
million people) and then multiplying by 100. With this relatively simplistic information in
hand, perhaps we might reconsider whether or not we have a crisis. The fact that more than
half of the people with student debt owe less than $20,000 may temper concerns about how
big this problem might be. In addition, the finding that most people with over $100,000 in
debt owe this for graduate school, might reduce our concern for them because medical and
law professionals generally earn high salaries. Perhaps our energies should be directed toward
those who do not wind up in high-paying professions. Nevertheless, we can learn a lot from
these data and from well-constructed frequency distributions.
Collecting data means measuring observations of a variable. And, of course, these meas-
urements will differ from one another. Given this variability, it is often difficult to make
any sense of the data until they are analyzed and described. This chapter examines a basic
technique for dealing with variability and describing data: forming a frequency distribution.
When formed correctly, frequency distributions achieve the goals of all descriptive statistical
techniques: They organize and summarize the data without distorting the information the
data provide about the world.
This chapter also introduces two related topics, graphical representation of distributions
and percentiles. Graphical representations highlight the major features of distributions to
facilitate learning from the data. Percentiles are a technique for determining the relative
standing of individual measurements within a distribution.
While reading this chapter, keep in mind that the procedures for constructing frequency
distributions can be applied to populations and to samples. Because it is so rare to actually
have all the scores in a population, however, frequency distributions are usually constructed
from samples. Reflecting this fact, most of the examples in the chapter will involve samples.

2.2 Frequency distributions
2.2.1 Frequency distributions summarize data by categorizing and counting

Suppose that you are working on a study of social development. Of particular interest is the
age at which aggressive tendencies first appear in children. You begin data collection (meas-
uring the aggressiveness variable) by asking the teacher of a preschool class to rate the aggres-
siveness of the 20 children in the class using the scale in Table 2.2:
The data are in Table 2.3. Obviously, the data are variable; that is, the measurements differ
from one child to another.
It is difficult to learn anything from these data as presented. A first step in learning from
the data should be to organize and summarize them by making a frequency distribution.

A frequency distribution is a tabulation of the number (count) of occurrences of each

score value (category).

The frequency distribution for the aggressiveness data is given in Table 2.4. The second col-
umn lists the score values. The third column in Table 2.4 lists the frequency with which each
score value appears in the data. Constructing the frequency distribution involves nothing
more than counting the number of occurrences of each score value. There is a simple way to
check whether the distribution has been properly constructed: The sum of the frequencies in
the distribution should equal the number of observations in the sample (or population). As
indicated in Table 2.4, the frequencies sum to 20, the number of observations.
 Frequency distributions and percentiles 23

Table 2.3 Aggressiveness ratings for 20 preschoolers

Child Rating Child Rating Child Rating Child Rating

a 4 f 0 k 3 p 2
b 3 g 3 l 0 q 3
c 1 h 3 m 4 r 3
d 1 i 4 n 2 s 1
e 2 j 2 o 3 t 3

Table 2.4 Frequency distributions for the aggressiveness data in Table 2.2

Meaning Score Frequency Relative Cumulative Cumulative

Value Frequency Frequency Relative Frequency
Very Timid 0 2 0.10 2 0.10
Timid 1 3 0.15 5 0.25
Average 2 4 0.20 9 0.45
Aggressive 3 8 0.40 17 0.85
Very Aggressive 4 3 0.15 20 1.00
Potential for 5 0 0.00 20 1.00
Violence
20 1.00

Section Check: What is the goal of constructing a frequency distribution?

2.2.2 Benefits of frequency distributions

It is clear that the frequency distribution has a number of advantages over the listing of the
data in Table 2.3. The frequency distribution organizes and summarizes the data, highlight-
ing the major characteristics. For example, it is easy to see that the measurements in the
sample range from a low of 0 to a high of 4. Also, most of the measurements are in the mid-
dle range of score values, and there are fewer measurements in the ends of the distribution.
Another benefit provided by the frequency distribution is that the data are now easily com-
municated. To describe the data, you need to report only five pairs of numbers (score values
and their frequencies).
Be careful not to confuse the numbers representing the score values (aggressiveness rat-
ings) and the numbers representing the frequencies of the particular score values. For exam-
ple, in Table 2.4 the number “4” appears in the column labelled “score value” and the
column labelled “frequency.” The meaning of this number is quite different in the two col-
umns, however. The score value of 4 means a particular level of aggressiveness (very aggres-
sive). The frequency of 4 means the number of times a particular score value was observed
in the data.
To help overcome any confusion, be sure that you understand the distinctions among the
following terms. “Score value” refers to a possible value on the measurement scale. Not all score
values will necessarily appear in the data, however. If a particular score value is never assigned as a
24 Frequency distributions and percentiles

measurement (for example, the score value 5, potential for violence), then that score value would
have a frequency of zero. “Frequency” refers to the number of times a particular score value
occurs in the data. Finally, the terms “measurement,” “observation,” and “score” are used inter-
changeably to refer to a particular datum—the number assigned to a particular individual. Thus,
in Table 2.4, the score value of 1 (timid) occurs with a frequency of 3. Using other words, there
are three scores (or measurements, or observations) with the score value of 1 (timid).

Section Check: What are the benefits of a frequency distribution over the raw data?

2.2.3 Relative frequency distributions display the proportions of observations of a particular

score value

An important type of frequency distribution is the relative frequency distribution.

Relative frequency of a score value is the proportion of observations in the distribution

at that score value. A relative frequency distribution is a listing of the relative frequen-
cies of each score value.

The relative frequency of a score value is obtained by dividing the score value’s frequency by
the total number of observations (measurements) in the distribution. For example, the rela-
tive frequency of aggressive children (score value of 3) is 8/20 = 0.40.
Relative frequency is closely related to percentage. Multiplying the relative frequency by
100 gives the percentage of observations at that score value. For these data, the percentage
of children rated aggressive is 0.40 × 100 = 40%.
The fourth column in Table 2.4 is the relative frequency distribution for the aggressiveness
data. Note that all of the relative frequencies are between 0.0 and 1.0, as they must be. Also, the
sum of the relative frequencies in the distribution will always equal 1.0. Thus, computing the sum
is a quick way to ensure that the relative frequency distribution has been properly constructed.

Section Check: What are some other words for relative frequencies?

2.2.4 Relative frequency distributions provide distinct advantages over frequency

distributions

Relative frequency distributions are often preferred over raw frequency distributions because
the relative frequencies combine information about frequency with information about the
number of measurements. This combination makes it easier to interpret the data. For exam-
ple, suppose that an advertisement for Nationwide Beer informs you that in a “scientifically
selected” sample, 90 people preferred Nationwide, compared to only ten who preferred Brand
X. You may conclude from these data that most people prefer Nationwide. Suppose, how-
ever, that the sample actually included 10,000 people, 90 of whom preferred Nationwide,
ten of whom preferred Brand X, and 9,900 of whom could not tell the difference. In this
 Frequency distributions and percentiles 25

case, the relative frequencies are much more informative (for the consumer). The relative
frequency of preference for Nationwide is only 0.009.
The same argument in favor of relative frequency can also be made (in a more mod-
est way) for the data on aggressiveness. It is more informative to know that the relative
frequency of timid children is 0.15 than to simply know that three children were rated as
timid.
When describing data from random samples, relative frequency has another advantage.
The relative frequency of a score value in a random sample is a good guess for the relative
frequency of that score value in the population from which the random sample was selected.
There is no corresponding relation between frequencies in a sample and frequencies in a
population.

Section Check: What are some advantages of relative frequencies over frequencies?

2.2.5 Cumulative frequency distributions tabulate all measurements at or smaller than a

particular value

Another type of distribution is the cumulative frequency distribution.

A cumulative frequency distribution is a tabulation of the frequency of all measure-

ments at or smaller than a given score value.

The fifth column in Table 2.4 is the cumulative frequency distribution for the aggressiveness
scores. The cumulative frequency of a score value is the frequency of that score value plus
the frequency of all smaller score values. The cumulative frequency of a score value of 0 (very
timid) is 2. The cumulative frequency of a score value of 1 (timid) is obtained by adding 3
(the frequency of timid) plus 2 (the frequency of very timid) to get 5. Note that the cumula-
tive frequency of the largest score value (5) equals 20, the total number of observations. This
must be the case, because cumulative frequency is the frequency of all observations at smaller
than a given score value, and all of the observations must be at or smaller than the largest
score value. Also, note that the cumulative frequencies can never decrease when going from
the lowest to the highest score value. The reason is that the cumulative frequency of the next
higher score value is always obtained by adding to the lower cumulative frequency.
The notion of “at or smaller” implies that the score values can be ordered so that we can
determine what is smaller. Thus, cumulative frequency distributions are usually not appropri-
ate for nominal data.

Section Check: Define “cumulative frequency.”

2.2.6 Cumulative relative frequency distributions tabulate the proportion of scores at or

smaller than a particular score value

A cumulative relative frequency distribution is a tabulation of the relative frequencies

of all measurements at or below a given score value.
26 Frequency distributions and percentiles

The last column in Table 2.4 lists the cumulative relative frequencies for the aggressiveness
data. These numbers are obtained by adding up the relative frequencies of all score values at
or smaller than a given score value.
Cumulative frequency distributions are most often used when computing percentiles. We
shall postpone further discussion of these distributions until that section of the chapter.

Section Check: When might we use cumulative relative frequencies?

2.2.7 Using JASP to construct frequency distributions

Programs such as JASP make constructing frequency distributions easy. Use JASP to open
the file “Table 2.3 Aggressiveness Ratings” (at the web address: www​.routledge​.com​
/9780367457976). Note that JASP correctly determines that the data are measured on an
ordinal scale (the three-line histogram next to the label “Rating”), but as we will see, JASP
doesn’t always get this right. Next, choose the “Descriptives” menu and select “Rating” as
the variable. Finally, select “Frequency tables.” Your output should look similar to Figure 2.1.
JASP automatically constructs a frequency distribution, a relative frequency distribution
(using percentages instead of proportions), and a cumulative relative frequency distribution.
Because there are no observations of “5, Potential for Violence,” JASP does not know about
that category and does not include that category in the table. If needed, you can add this
category by a) copying the table using the upside-down triangle next to “Frequency Tables”,
b) pasting the table into a word processor, and c) adding the category.

Figure 2.1 JASP frequency distributions for the data in Table 2.3.
 Frequency distributions and percentiles 27

2.3 Grouped frequency distributions

2.3.1 Grouped frequency distributions are used when there are many score values

The aggressiveness data were particularly amenable to description by frequency distributions

in part because there were only a few score values. Sometimes, however, the data are not so
accommodating, and a more sophisticated approach is called for.
Consider, for example, the first 60 measurements on the YRSMK (years smoking) variable
in the Smoking Study (Table 2.5). Because the measurements are variable, it is difficult to
learn anything from the data as presented in this table.
The frequency distribution is presented in Table 2.6. As you can see, the frequency distri-
bution for these data does not provide a very useful summary of the data. The problem is that
there are too many different score values.
The solution is to group the data into clusters called class intervals.

A class interval is a range of score values. A grouped frequency distribution is a tabu-

lation of the number of measurements in each class interval.

Table 2.5 YRSMK—Number of years smoking daily from the first 60 participants in the Smoking Study
5 13 17 20 19 35 21 28 3 22
26 13 30 30 30 32 40 27 14 4
27 33 28 45 29 25 38 35 33 39
5 4 20 24 25 27 16 25 38 9
36 20 18 11 12 23 22 27 32 49
22 30 0 32 4 23 9 29 22 23

Table 2.6 Frequency distribution of first 60 YRSMK scores

Score Value Frequency Score Value Frequency Score Value Frequency

0 1 17 1 34 0
1 0 18 1 35 2
2 0 19 1 36 1
3 1 20 3 37 0
4 3 21 1 38 2
5 2 22 4 39 1
6 0 23 3 40 1
7 0 24 1 41 0
8 0 25 3 42 0
9 2 26 1 43 0
10 1 27 4 44 0
11 1 28 2 45 1
12 0 29 2 46 0
13 2 30 4 47 0
14 1 31 0 48 0
15 0 32 3 49 1
16 1 33 2
28 Frequency distributions and percentiles

Table 2.7 Grouped frequency distributions for YRSMK scores in Table 2.5

Class Interval Frequency Relative Cumulative Cumulative

Frequency Frequency Relative Frequency
0–4 5 0.083 5 0.083
5–9 4 0.067 9 0.150
10–14 5 0.083 14 0.233
15–19 4 0.067 18 0.300
20–24 12 0.200 30 0.500
25–29 12 0.200 42 0.700
30–34 9 0.150 51 0.850
35–39 6 0.100 57 0.950
40–44 1 0.017 58 0.967
45–49 2 0.033 60 1.00
Total 60 1.000

The grouped frequency distribution is presented in Table 2.7. The class intervals are listed on
the left. The lowest interval, 0–4, contains all of the measurements between (and including)
0 and 4. The next interval, 5–9, contains the measurements between 5 and 9, and so on.
Clearly, the data in the grouped distribution are much more easily interpreted than when
the data are ungrouped. We can now see that most of the people in this sample have been
smoking for 20–30 years, although there are a few who have been smoking for more than 45
years and a few who have been smoking only a couple of years.
Relative and cumulative frequency distributions can also be formed from grouped data.
Relative frequencies are formed by dividing the frequency in each class interval by the total
number of measurements. Cumulative distributions are formed by adding up the frequencies
(or relative frequencies) of all class intervals at or below a given class interval. These distribu-
tions are also given in Table 2.7.

Section Check: What are the conditions under which we might consider grouping our data?

2.3.2 Constructing grouped distributions takes some planning

A grouped frequency distribution should summarize the data without distorting them.
Summarization is accomplished by forming class intervals; if the intervals are inappropri-
ate (for example, too big), however, the data are distorted. As an example of distortion,
Table 2.8 summarizes the YRSMK data from Table 2.5 using three large intervals. Indeed,
the data are summarized, but important information regarding how the measurements are
distributed is lost. The following steps should be used to construct grouped distributions that
summarize but do not distort.
Guidelines for grouped frequency intervals:

1 There should be between 8 and 15 intervals.

2 Use convenient class interval sizes, like 2, 3, 5, or multiples of 5.
3 Start the first interval at or below your lowest score.
 Frequency distributions and percentiles 29

Table 2.8 Grouped frequency distributions

for YRSMK scores in Table 2.5

Interval Frequency
0–19 18
20–39 39
40–59 3
Total 60

We will have more to say about grouped frequency distributions when we discuss how to use
JASP to create grouped distributions in the next section.

Section Check: Generally speaking, how many intervals should be in a grouped frequency
distribution?

2.4 Graphing frequency distributions

2.4.1 Graphing frequency distributions is an important way to show your data

Displaying a frequency distribution as a graph can highlight important features of the data.
Graphs of frequency distributions are drawn using two axes.

The abscissa or x-axis is the horizontal axis. For frequency and relative frequency distri-
butions, the abscissa is marked in units of the variable being measured, and it is labelled
with the variable’s name. The ordinate or y-axis is marked in units of frequency or rela-
tive frequency, and so labelled.

In Figure 2.2, the abscissa is labelled with values of the aggressiveness variable for the distri-
bution in Table 2.4. The ordinate is marked to represent relative frequency of the measure-
ments. Techniques for graphing frequency and relative frequency distributions are almost
exactly the same. The only difference is in how the ordinate is marked. Because relative fre-
quency is generally more useful than raw frequency, the examples that follow are for relative
frequency distributions.

Section Check: What advantages are there to graphing over data tables?

2.4.2 Histograms represent frequencies as heights of bars

Figure 2.2 is a relative frequency histogram for the aggressiveness data.

A relative frequency histogram uses the heights of bars to represent relative frequencies
of score values (or class intervals).
30 Frequency distributions and percentiles

Figure 2.2 Relative frequency histogram for the aggressiveness data in Table 2.3.

To construct the histogram, place a bar over each score value. The bar extends up to the
appropriate frequency mark on the ordinate. Thus, a bar’s height is a visual analogue of the
score value’s relative frequency: the higher the bar, the greater the relative frequency.
Relative frequency histograms can also be drawn for grouped distributions. For these dis-
tributions, a bar is placed over each class interval.
Figure 2.3 is a relative frequency histogram of the YRSMK scores in Table 2.7. Sometimes,
only the midpoints of each interval are shown on the abscissa. The midpoint of a class interval
is the average of the interval’s lower bound and the upper bound. Again, the height of each
bar corresponds to its relative frequency.
The relative frequency histogram illustrated in Figure 2.3 makes particularly clear some of
the salient characteristics of the distribution. For example, it is easy to see that most of the
scores are in the middle of the distribution and that there is a decrease in frequency from the
moderate scores to the higher scores.

Section Check: What does the height of a bar in a histogram represent?

2.4.3 Using JASP to create grouped frequency histograms

As an example, use JASP to open the file “Table 2.5 60 Observations of YRSMK,” go to the
Descriptives menu, and choose YRSMK as the variable. JASP does not have an easy way to
create grouped frequency distributions, but it is very useful for creating grouped frequency
histograms.
Under Plots, choose Distribution plots. JASP gives you four ways to decide on bin (i.e.,
interval) widths (e.g., the Sturges rule), or you can choose to set the number of bins manually.
 Frequency distributions and percentiles 31

Figure 2.3 JASP grouped frequency histogram.

If you choose “Manual” and enter 10, JASP produces the grouped frequency histogram in
Figure 2.3.
As you can see in Figure 2.3, JASP gives you several other options for plotting distribu-
tions, such as Pie charts and Dot plots. You should explore those options to see which pro-
vide the best summary of the data without distortion. We will discuss the Customizable plots
after discussion of percentiles.

2.5 Describing distributions
2.5.1 Distributions can be described and characterized in three major ways

Distributions differ in three major characteristics: shape, central tendency, and variability.
Over the next few pages, we will practice comparing distributions by using these characteris-
tics, and you will see that they play a major role in the study of statistics.

Section Check: What are the three major ways a distribution can be described?
32 Frequency distributions and percentiles

2.5.2 Symmetry and skew are ways in which the shape of a distribution can be characterized

The shape of a distribution can be broadly classified as symmetric, positively skewed, or nega-
tively skewed.

A symmetric distribution can be divided into two halves that are mirror images of each
other.

The distributions illustrated in the top row of Figure 2.4 are symmetric distributions. The
distribution of years smoking in Figure 2.3 can be characterized as “somewhat symmetric.”
In contrast, a skewed distribution cannot be divided into halves that are mirror images.
The distributions illustrated in the bottom row of Figure 2.4 are skewed.

A positively skewed distribution has score values with low frequencies that trail off
toward the positive numbers (to the right). A negatively skewed distribution has
score values with low frequencies that trail off toward the negative numbers (to the
left).

Distributions E and F in Figure 2.4 are positively skewed; Distribution D has a negative skew.
Note that a skewed distribution has one “tail” that is longer than the other. If the longer
tail is pointing toward the positive numbers, then the distribution has a positive skew; if the
longer tail is pointing toward the negative numbers, then the distribution has a negative
skew.
Salary distributions are often positively skewed. Salaries cannot be less than $0.00, so the
tail on the left cannot trail off very far. Most people have salaries in the midrange, but some
people have very large salaries, and some (although relatively few) have enormous salaries.
The people with enormous salaries produce a positive skew in the distribution.

Section Check: What is a symmetrical distribution?

Figure 2.4 Symmetric and asymmetric distributions.

 Frequency distributions and percentiles 33

2.5.3 Modality refers to the number of modes, or peaks, in the distributions

Modality is another aspect of the shape of a distribution. Modality is the number of clearly
distinguishable peaks in the distribution.

A unimodal distribution has one peak. A bimodal distribution has two peaks. A mul-
timodal distribution has more than two peaks.

In Figure 2.4, Distributions A, D, and E are unimodal, and Distributions C and F are bimodal.
Some shapes of distributions have special names. A distribution that is unimodal and sym-
metric (such as Distribution A in Figure 2.4) is called a bell-shaped distribution. (A normal
distribution is a special type of bell-shaped distribution that we will discuss in Chapter 4.)
Many psychological variables (such as intelligence) have bell-shaped distributions.
Distribution B in Figure 2.4 is called a rectangular distribution. Note that this distribution
is symmetric, but it does not have a well-defined mode. Rectangular distributions indicate
that all of the score values have the same relative frequency. Rectangular distributions often
arise in gambling situations, such as tossing a fair coin (relative frequency of heads = relative
frequency of tails = 0.5) and rolling a fair die (each number has a relative frequency of one
sixth).
Distribution E in Figure 2.4 is called a J-curve (although “backward J-curve” would be
more appropriate). This distribution is positively skewed and is frequently seen in the relative
frequencies of rare events, such as number of lotteries won. Most of us have never won a lot-
tery, so the score value of 0 has the greatest relative frequency. Some people have won one or
two lotteries, and a few have won even more, producing the long tail on the right.

Section Check: What is another term for modality?

Figure 2.5 Distributions differing in central tendency and shape.

34 Frequency distributions and percentiles

2.5.4 Measures of central tendency are near the center of the distribution

In addition to shape, distributions differ in central tendency.

The central tendency of a distribution is the score value near the center of the distribu-
tion. It is often a typical or representative score value.

You may think of the central tendency as the score value that is most representative of the
distribution—that is, the score value that you would choose to give the general flavor of the
distribution. On the other hand, you may think of the central tendency as the location of the
center of the distribution on the measurement scale (represented by the abscissa in a graph).
In Chapter 3, we will discuss the mean, the median, and the mode, which are numerical
indices of central tendency.
The distributions in the top row of Figure 2.5 have the same shape but differ in central
tendency (the central tendencies are indicated by arrows). In the bottom of the figure, the
distributions differ in both shape and central tendency. In both the top and the bottom, the
central tendency increases from left to right.

Section Check: What are measures of central tendency?

2.5.5 Measures of variability describe how much scores differ from one another in a
distribution

As you will recall, variability is the driving force behind statistics. Indeed, distributions arise
because we measure variables so that not all the measurements are the same.

Variability is the degree to which the measurements in a distribution differ from one
another.

When all the measurements in a distribution are similar, the distribution has little variability.
In fact, in the rare instance when all of the measurements in the distribution are the same,
the distribution has no variability. When the measurements in a distribution deviate greatly
from one another, the distribution has much variability. In Chapter 3, we will learn how to
compute the variance and the standard deviation, two particularly useful numerical indices
of variability.
The top half of Figure 2.6 illustrates how distributions can have the same general shape
and the same central tendency but differ in variability. Variability increases from Distribution
A to B to C.
The bottom of Figure 2.6 illustrates how distributions can have different central tenden-
cies and different variabilities. From left to right, the distributions increase in central ten-
dency. However, Distribution D has the greatest variability and Distribution E has the least
variability.

Section Check: What does variability measure?

 Frequency distributions and percentiles 35

2.5.6 Using the three major characteristics to compare and contrast different distributions

Now that you have learned how distributions can differ, you have the skills needed to quickly
compare and contrast distributions. Simply determine how the distributions compare and
contrast in terms of shape, central tendency, and variability.
Recall that in the beginning of this chapter we started with an example of a psychologist
investigating the development of aggressive behavior. Suppose that the psychologist had
segregated the data by sex of the children and had constructed separate relative frequency
distributions for the girls and the boys. Figure 2.7 illustrates two possible distributions.
The distribution for the girls is more symmetric than that for the boys; the distribution for
the boys is positively skewed. Both distributions are unimodal. The two distributions have
similar central tendencies, but they differ in variability; the distribution for the boys is more
variable than that for the girls.
Now that the data have been described, it is up to you to determine their importance.
Why is the distribution for the boys more variable than that for the girls? Does this represent

Figure 2.6 Distributions with the same general shape but differing in central tendency and variability.

Figure 2.7 Possible distributions of aggressiveness scores.

36 Frequency distributions and percentiles

a basic difference between the sexes, or is it due to socialization? Should these data be used
to argue for social change? Of course, these particular questions are not statistical questions,
and statistical analyses will not provide answers to them. The point is that statistical analysis
can transform the data into a form that helps you to ask good questions. But it is up to you
to ask the questions.

Section Check: What are the three major ways to characterize a distribution?

2.6 Percentiles
2.6.1 Percentiles define a measurement’s relative standing in the distribution

We have been working with whole distributions and paying little attention to individual
measurements within a distribution. At times, however, the individual measurements are of
great importance. Consider this example. One of your professors has just handed back your
exam. On the top, in bold red ink, is the number 32. At first you are worried. But when you
see the 25 on your neighbor’s paper, you begin to feel a little better. Still, is 32 a good score?
How good an exam score is depends in part on how you define “good.” As a first approxi-
mation, you might think that a good score on a test corresponds to achieving close to the
maximum possible. Thus, if the exam had a total of 50 points possible, your score of 32 cor-
responds to 64% correct, which is not very good.
Suppose, however, that the exam was extremely difficult and your score of 32 is one of
the best in the class. In this case, you are justified in thinking that your score of 32 (or 64%)
is really very good.
The point is this: An informative index of the goodness of a score is the standing of that
score relative to the other scores in a distribution. Scores that are near the top of the distribu-
tion are “good,” regardless of the actual score value or the percent correct. Scores near the
bottom of the distribution are “poor,” regardless of the actual score value. (If you obtained
90% correct on a test, that would be a “poor” score if everyone else in the class obtained
100% correct.)
Percentile ranks provide just such an index of goodness by giving the relative standing of
a score—the score’s location in the distribution relative to the other measurements in the
distribution. More formally:

The percentile rank of a score value is the percent of measurements in the distribution
below that score value.

Note that percentile rank and percent correct are not the same. A score value of 80% correct
might have any percentile rank between 0 and 100, depending on the number of measure-
ments in the distribution less than 80% correct.
When exam grades are given as percentiles, you have an easily interpreted index of rela-
tive standing—how well you did relative to the others in the distribution. If your score has a
percentile rank of 95%, then you did better than 95% of the others in the distribution. If your
score has a percentile rank of 30%, you did better than 30% of the others.
Percentile ranks are often used to report the results of standardized tests such as the SAT
(Scholastic Aptitude Test) and the GRE (Graduate Record Examination). Your percentile
Exploring the Variety of Random
Documents with Different Content
State."

But Andrew said again:

"Show me the statute."

And again:

"Show me the statute."

And from that he was not to be moved. Seeing that his mind was made
up, Phillips turned away abruptly, saying to me, "Come," and we departed.
As we went downstairs Phillips said:

"I will never again speak to Andrew as long as I live."

And we went back to the Temple, knowing at last we had nothing to

depend on but ourselves and our revolvers.

Again during the interval my friend came to me. He said: "You will be
allowed to hold your meeting this afternoon, though not without
interruption. But the attack I have warned you of will be made this evening,
and I once more beseech you to stay away." He knew, of course, it was
impossible. What took place after that in the councils of the rioters I know
not. I have always supposed that my friend, a man well known in Boston,
went to the Mayor and laid the case before him. I do not know. What is
known is that before the hour when the Society was to assemble in the
evening, the Mayor closed the Temple. His decision was not imparted to us.
Phillips and I drove to the Temple, and only on arriving heard what the
Mayor had done. He was a weak Mayor, disloyal, incompetent. But he had
perhaps prevented a tragedy. I think Governor Andrew, aware of the
probable course of events in the South and at Washington, desired to avoid
anything like a conflict in Massachusetts. He said as much to me afterward.
That was his excuse.
 CHAPTER XI

WENDELL PHILLIPS—GOVERNOR ANDREW—

PHILLIPS'S CONVERSION

There was one clear reason for the deadly hatred of the pro-slavery
faction in Boston to Phillips. He was the real leader of the Anti-Slavery
Party. If he could be silenced, the voices of the rest mattered little. During
twenty years Garrison's influence had been declining, and Phillips had come
steadily to the front. For the last ten years he had stood alone. It was his
voice which rang through the land. His were the counsels which governed
the Abolitionist band. His speeches were something more than eloquent;
they were full of knowledge, of hard thinking; and the rhetorical splendour
only lighted up a closely reasoned argument. What Emerson said of
speeches and writings in general was absolutely true of Phillips's oratory;
the effect of it was mathematically measurable by the depth of thought. He
spoke all over the North. The Conservatives had no match for him;
therefore he was to be put down by other means.

Passions ran, I think, higher in Boston during those winter months of

1860-1, and the early spring, than before or since. Thanks to the pro-slavery
faction on one side and the Abolitionists on the other, Massachusetts was
within measurable distance of civil war within her own borders. After Fort
Sumter and Baltimore, these passions found an outlet elsewhere. For a time,
the two Northern factions merged into one people. But during all the years
that have passed since I have known nothing quite like the state of feeling
which prevailed that winter. The solid men of Boston thought they saw the
fabric of society dissolving and their business and wealth and authority
perishing with it. The solid world was to exist no more. Naturally, they
fought for their lives and all the rest of it, and fought hard. Their hatreds
were savage. Their methods were savage. We seemed to be getting back to
the primitive days when men stood face to face, and the issue of battle
became a personal combat. The Lawrences and their friends were generally
a little stout for the business of battle, but the allies whom they brought with
them to Tremont Temple and the Music Hall and the streets were good
fighting material. During all this time the Abolitionists were, as they had
been, a minority and on the defensive.

But this was the state of things which Governor Andrew had in mind
when he challenged Phillips to show him the statute. He did not want to
make the State of Massachusetts a party to this conflict within itself. If to
keep order in the streets or to keep a platform open to Phillips he were
obliged to move, he meant to have the law with him. No refinements, no
Judge-made law, no generalizations—for the common law after an Atlantic
voyage and a hundred years' sleep is nothing—but a statute, printed, legible,
peremptory, binding alike upon Governor and citizens. There was no such
statute. If anybody had happened to think of it, no doubt there would have
been, but there was not.

Therefore the Governor sat still. He was of such a bulk that it seemed as
if, while he sat still, nothing could move. He was, in size and build, not
wholly unlike Gambetta, though he had two eyes, both blue, as against the
one black, fiery orb of the Genoese; and curling light brown hair instead of
the black lion's mane which floated to Gambetta's shoulders; and a face in
which sweetness counted for as much as strength. Like Gambetta, he was
well served by those about him. He knew accurately what was going on,
and all that was going on. He told me afterward he did not know on what
information we acted, but he was astonished we knew so much about what
the enemy intended. When I reminded him that my associations were
mostly with the other side, he reflected a moment and said: "Yes, that
explains a good deal." I did not think it necessary to add that, after Tremont
Temple, we were on good terms with the police also; since Phillips's appeal
to Andrew had been based on the alliance between the police and the
Lawrence mob; an alliance which had in truth existed, at that time.

But the winter wore on. Twice after the discourse on Mobs and
Education, Phillips spoke in the Music Hall—January 20th, 1861, on
Disunion, and February 17th, on Progress. Both times the mob supplied part
of his audience inside and part of his escort outside. No violence was
attempted. The police were too strong, and the example of Deputy Chief
Ham had proved they were in earnest. If there was any violence, it was in
Phillips's speeches and language. He was never more provocative. His
forecast of the situation was influenced by his wishes and theories. All his
life he had been preaching disunion as the one remedy for the slave.
Disunion seemed now at last within reach, and at all costs he would do what
he could to promote it. Indeed, he thought it already accomplished. Within
six weeks after Lincoln's election South Carolina had replied by an
ordinance of secession. Mississippi, Alabama, Georgia had followed, and
all over the South United States forts and arsenals had been seized by State
troops. What was Phillips's comment?

"The Lord reigneth; let the earth rejoice. The covenant with death is
annulled; the agreement with hell is broken in pieces. The chain which has
held the slave system since 1787 is parted."

He pronounced a eulogy on the Southern State which had led the way:

"South Carolina, bankrupt, alone, with a hundred thousand more slaves

than whites, four blacks to three whites within her borders, flings her
gauntlet at the feet of twenty-four millions of people—in defence of an
idea."

A month later he was in the same mood. It was a trait of Phillips—not a

good one—that he attacked most mercilessly the men who hated slavery as
much as he did, but could not go as far as he did. In this February speech
there is a long lampoon on Dana; counsel for the slave in all the fugitive
slave cases, but never denying—what lawyer ever did deny?—that there
was a constitutional obligation to return fugitives. It is human nature, but
not the best side of it.

Such a reproach came ill from a man who denounced the Constitution as
a covenant with death because of the compromises with slavery imbedded
in the great instrument of 1787. Of these compromises the rendition of
fugitive slaves was one. Phillips himself could not deny it. The difference
between him and Dana was that Dana would bow to the law and Phillips
would not. Dana would do what he could by legal means to rescue the
fugitive. He defended him in the courts. Phillips would have defended him
in the streets. Both men were needful to the time. The Abolitionists were
very far from disdaining the use of legal weapons. When Theodore Parker
had been indicted and the Court, at the instance of his counsel, quashed the
indictment on purely technical grounds, Parker exulted. "It is a triumph for
the right. We have broken their sword."

There came, however, the moment when Phillips had to cast in his lot,
for good or evil, with either North or South. He hesitated long. He thought
and thought. He talked with his friends, with the man in the street, with the
men who had lately mobbed him. One morning he came into my office. His
sunny face was clouded. He looked anxious, almost ill. He had to make the
most momentous decision of his life; and he could not yet make up his
mind. He said:

"I came to talk to you because I know you are against me. What I have
said to you before makes no impression. You still think I ought to renounce
my past, thirty years of it, belie my pledges, disown every profession of
faith, bless those whom I have cursed, start afresh with a new set of
political principles, and admit my life has been a mistake."

"Certainly not the last," I said, "and as for the others, are you not taking
a rhetorical view, a platform view? But I will go further. I don't think it
matters much what you sacrifice—consistency, principles, or anything.
They belong to the past. They have nothing to do with to-day. The war is
upon us. You must either support it or oppose it. If you oppose it, you fling
away your position and all your influence. You will never be listened to
again."

And so on. He sat silent, unmoved. Nothing I could say, nothing

anybody could say, would move him. All his life long he had thought for
himself; in a minority of one. It had to be so now. We talked on. Finally, I
said: "I will tell you what I once heard a negro say: 'When my massa and
somebody else quarrel I'm on the somebody else's side.' Don't you think the
negro knows? Do you really doubt that a war between the Slave Power and
the North, be the result what it may, must end in Freedom?" I am not sure
that I ever did hear a negro say that, but I hoped that Phillips would open
his mind to the negro if not to me. And I think he did. I trust this little
artifice of debate was not very wrong. I had to urge what I could, but I knew
Phillips would decide for himself. He left saying, "I will see you again to-
night." I went to his house. When I opened the door of the parlour, there lay
Phillips on the sofa, asleep. Ten minutes later he awoke; lay silent for
another minute, then said:

"We shall not have to discuss these things any more. I am going to speak
next Sunday at the Music Hall for the War and the Union."

And he began at once to consider how he should announce his

conversion. Having gone over, he took his whole heart with him. No
compromise, no transition, not one word to retract, not a hint of apology or
explanation. Yesterday an Abolitionist to whom the Constitution was a
covenant with death and an agreement with hell. To-day a soldier for the
Union. Presently he said:

"It will be the most important speech of my life. I don't often write, as
you know, but I shall write this and will read it to you when it is finished."

Two days later he sent for me again and these were the first sentences I
heard:

"Many times this winter, here and elsewhere, I have counselled peace—
urged as well as I knew how the expediency of acknowledging a Southern
Confederacy and the peaceful separation of these thirty-four States. One of
the journals announces to you that I come here this morning to retract those
opinions. No not one of them."

I said: "Mr. Phillips, you will never get beyond that. They will not
listen."

"Then they will be the last sentences I shall ever utter in public. But do
you listen."

And he went on, in his finest platform manner and voice:

"No, not one of them. I need them all; every word I have spoken this
winter; every act of twenty-five years of my life, to make the welcome I
give this War hearty and hot."
 He knew what he was about. When it became known he was to speak for
the Union, Charles Pollen came to me and asked whether I thought Phillips
would like the Music Hall platform hung with the American flag. "Yes,"
said Phillips, "deck the altar for the victim." And decked it was—a forest of
flags; and the flags told the story, long before Phillips opened his mouth.
There was not a note of remonstrance as he announced his refusal to retract.
And again he went on:

"Civil war is a momentous evil. It needs the soundest, most solemn

justification. I rejoice before God to-day for every word I have spoken
counselling peace, but I rejoice also, and still more deeply, that now, for the
first time in my Anti-Slavery life, I speak beneath the Stars and Stripes, and
welcome the tread of Massachusetts men marshalled for war."

I never saw such a scene. The audience sprang up and cheered and
cheered and cheered. The hall was a furnace seven times heated. The only
unmoved man was Phillips. He waited and once more went on:

"No matter what the past has been or said, to-day the slave asks God for
a sight of this banner and counts it the pledge of his redemption. Hitherto it
may have meant what you thought or what I thought: to-day it represents
sovereignty and justice. Massachusetts has been sleeping on her arms since
'83. The first cannon shot brings her to her feet with the war-cry of the
Revolution on her lips."

And so on to the end. It was a nobler speech even than in the printed
report, for that came from his manuscript and often he put his manuscript
aside and let himself go. The inspiration of the moment was more than any
written words. When it was over there was again a mob outside; a mob that
would have carried the orator shoulder-high to Essex Street. The honest,
strong face of the Deputy Chief of Police wore a broad smile. He had done
his duty. His responsibilities were ended. He, too, had fought his fight.
Phillips took it all coolly. It was such a triumph as comes to a man once in
his career, and once only—the finest hour in Phillips's life. He never
reached a greater height of oratory, nor an equal height of devotion. For his
triumph was over himself.
 CHAPTER XII

WILLIAM LLOYD GARRISON—A CRITICAL

VIEW

In explaining why Wendell Phillips was the target for every shot in the
winter of 1860-1, I said it was because he was the real leader of the anti-
slavery party during all the later and more critical years of the long struggle
for freedom. No doubt, Garrison at one time held the first place among the
Abolitionists. He was the first of them in time, or one of the first. He had
had the good fortune to be mobbed and led through the streets of Boston
with a rope about his body. He had founded a weekly paper, The Liberator.
Georgia had offered five thousand dollars reward for his arrest. He had
unflinching courage and needed it all in the 'thirties and later. But he had
very moderate abilities. His force was a moral force. He had convictions
and would go any length rather than surrender any one of them. But he had
almost no other of those gifts and capacities which make a leader. He had
no organizing power. He was not a good writer. He was not a good speaker.
He could not hold an audience. He could not keep the attention of the public
which he had won in the beginning. He did not attract to the Abolitionist
ranks the ablest of the men who were ready to make a fight against slavery.
They did not care to serve under Garrison; under a leader who could not
lead. They went into politics.

So it happened that the Abolitionists had become a dwindling force. If

Phillips had not appeared on the scene, with his wonderful oratory, his
natural authority on the platform and off, his brilliant love of battle, his
temperament, at once commanding and sympathetic, his persuasive charm
—the Abolitionists would have been wellnigh forgotten. He had all the
moral force of Garrison, and the intellectual force which Garrison had not.
 Phillips himself would never allow this to be said if he could help it. He
recognized Garrison as leader, and was perfectly loyal to him. So far as he
could, he imposed his own view on the public. It was so abroad as well as at
home. When Garrison came to London a meeting was held in St. James's
Hall in his honour. Mr. Bright spoke and others spoke, hailing the worn-out
champion as the herald of American Emancipation, which perhaps he was.
Boston, which has periods of generous penitence, gave him thirty thousand
dollars, others than Bostonians paying part of the money, and accepted a
bronze statue and put it up—I forget where. It has ever since been the
fashion to recognize Garrison as the moral educator of the North on the
slavery question; the schoolmaster of his period. Very possibly my liking
for Phillips warped my opinion at the time. But now, after all these years, I
think myself impartial. I had a knowledge of the situation. If it is a wrong
view, why was Phillips and not Garrison the shining mark at which the pro-
slavery people aimed in those critical years from 1854 to 1861? No other
theory will explain that.

When I used to express an impatient opinion of Garrison, and of

Phillips's submission to him, I was rebuked for it. Said Phillips:

"You are unjust and you do not know the facts, or you do not make
allowance for them. Like other young men, you are of to-day. Garrison's
work had been done before you were old enough to know anything about it,
and he is for all time. I don't say there would have been no Abolitionist
movement but for Garrison, since Abolition was in the air, and the anti-
slavery fight had to be fought. It would have been fought in a different way
without him, and perhaps later. You underrate the moral forces and
Garrison's capacity as a leader. He was a leader, and is. Intellectual gifts do
not make a leader. The soldier whom other soldiers follow into the breach,
and to death, need not be a great captain, nor understand the art of war.
What he understands is the art of getting himself killed, and of inducing the
men behind him to do the same. Garrison took his life in his hand. For
many years he was leader of a forlorn hope. He held extreme views. He had
to hold them. He drove men away from the Abolitionist camp. They were
better elsewhere. He was not a politician, but politics were not what we
wanted, nor what the cause wanted. What it wanted was inspiration, and
that is what it got from Garrison."
 I have put this in quotation marks, but I do not mean that Phillips said it
all at once, nor perhaps in these words. But the passage reproduces as
accurately as I can the substance of what I have heard him say in many talks
about Garrison. I do not expect anybody to accept my view against
Phillips's. But I must give my own, right or wrong. I saw something of
Garrison, publicly and privately. I had no dislike for him, but neither had I
any enthusiasm. As I recall the impressions of those days, it seems to me
that I have never known a man of so much renown as Garrison with so
slight an equipment for the business of leadership, or even of apostleship.
When I try to sum him up, I am embarrassed by the want of material. After
all, what did he say or do?

Borrowing from Isaiah a phrase of condensed passion, Garrison had

called the Constitution a covenant with death and an agreement with hell.
Without Isaiah's help, he produced the only other phrase which, out of all
his writings and speakings has kept a place in the general memory: "I will
not equivocate; I will not excuse; I will not retreat a single inch, and I will
be heard." That was his pledge in the first number of The Liberator. It was
finely said, and well he kept it; so long as it mattered what he kept. I have
often heard him speak. I cannot recall one single effort of anything that
could be thought oratory. He was a tiresome speaker. Of rhetoric, or of that
art which goes to the making of good speeches, he had no trace or tinge.
Between him and his audiences there was no give and take. He just stood up
on the platform and hammered away.

He was a fanatic, pure and simple. He had a message to deliver, and he

delivered it as a gramophone delivers its messages. He was what they call a
record. If he impressed his hearers, as he sometimes did, it was by the
passionate fervour of his beliefs, and of his animosities. He was at white
heat. More often he wearied them. They got up and went away. I suppose
people read The Liberator. Dr. Johnson said you could write anything if you
set yourself to it doggedly, and so it is of reading. But the average reader
feels himself entitled to a little help from the writer, and from Garrison he
got none.

This, however, was in the early days of journalism—it was ten years
before Horace Greeley founded The New York Tribune that The Liberator
was born. A newspaper was then a newspaper, whether it had any news or
not; and even when its editorials were written, as the elder Bennett said The
New York Herald editorials were written, for men who could not read. The
printed page had an authority because it was printed; an authority which
hardly survived Prince Bismarck's epigram on the newspaper: "Just printer's
ink on paper." The Liberator was violent, bitter, prolix, and dull. But the
Puritan preachers were all this, yet men sat contentedly for hours beneath
their intolerable outpourings, as do the Scotch to this day. Carlyle had heard
Irving preach for hours on end. I have sometimes had to sit under the
Scottish preachers, when staying at a highly ecclesiastical house. On these
occasions I used to dream that I was reading The Liberator or listening to
Garrison in the Boston Melodeon. The a priori method was common to
both, and the absence of accurate knowledge. They did not master their
subjects, nor their trade.

As to what Garrison did, I am quite willing to accept the history of his

time as it is commonly told. I take all that for granted; all his services to the
anti-slavery cause; and, with all drawbacks, they were great. Still, I do not
think they explain his immense fame. He was a Captain in the army of the
Lord, if you like, but a Captain who won no battles. There was one final
victory, based on a long series of defeats; a victory in which he had a share,
though not a great share. Perhaps a better Saint than Captain, but in Rome's
long catalogue of the canonized how many first-rate names are there? You
can become a saint quite cheaply if you know how. There are fifty or more
huge volumes of the Acta Sanctorum, mostly lies, yet extremely interesting
as examples of the use to which the human imagination can be put for
ecclesiastical purposes. A Benedictine labour, ere yet science had shaken
the foundations of clerical fairy tales by its demand for evidence. The
acutest minds accepted them. So late as the nineteenth century they were
still accepted. After his "conversion," Newman, perhaps the finest mind of
his time, swallowed whole all the fictions to which the Church of Rome had
given the imprimatur of infallibility. Garrison's exploits are less legendary,
but are they much more substantial? His fame rests on generalities.

To look at, he was neither soldier nor saint. He had not, on the one hand,
the air of command, nor, on the other, the sweetness or benignity we expect
from one of the heavenly host. His face was both angry and weak. His
attitude on the platform was half apologetic and half passionate. His speech
at times was almost shrewish. It was never authoritative though always self-
complacent. So was the expression of his face, with its smile which tried to
be amiable and succeeded in being self-conscious. There was no fire in his
pale eyes; if there had been, his spectacles would have dulled it. He
stooped, and his most vehement appeals—they were often extremely
vehement came to you sideways. It was an unlucky effect, for there was
nothing shifty or crooked in the man's nature. But he had a rôle to play—
Isaiah, if you like—and played it as well as his means would allow.

It was the indomitable honesty of the man which gave him such
authority as he had. That is not a bad eulogy in itself. Bad or good, nothing
I can say will diminish his reputation, nor do I wish it should. When a
legend has once grown up about a man it keeps on growing. It has been
decreed that Dickens shall be a great novelist, and Gladstone a great
statesman, and Browning a great poet, and Herbert Spencer a great
philosopher. Each of these men was great in other ways, but the legend is
invincible. So, no doubt, with Garrison. He will remain the Liberator of the
Slave. By the time the cold analysis of History reverses that verdict,
personal partialities will have ceased to count.

CHAPTER XIII

CHARLES SUMNER—A PRIVATE VIEW

The anti-slavery leaders who emerged about the same time from the
groups of mediocrities enveloping them were Wendell Phillips and Charles
Sumner. So essentially was Sumner an idealist that he might naturally have
cast in his lot with those who preferred ideals to party politics, but other
influences finally prevailed and he embarked on that career which, in due
time, made him the leader of the anti-slavery forces to whom freedom
seemed possible by political methods. On the whole, even among that group
of men which included Andrew, I think Sumner must be put first. His
province was larger; the range of his activities greater; and there were more
moments than one when he was the most conspicuous figure in American
public life. Of his scholarship, his legal attainments, his multifarious and
accurate knowledge, his immense powers of work, everybody has heard. I
do not enter upon that. The Sumner I shall speak of is the Sumner I knew.

In the account, first published in The New York Tribune, of my first

meeting with Bismarck, in 1866, I said that I had heard much from
Bismarck which I could not repeat. On my return, I saw Sumner. Almost
instantly he asked what it was Bismarck had told me which I could not
repeat in print. The question was embarrassing enough, and I answered
rather slowly:

"Mr. Sumner, much of what Count Bismarck said that seemed to me

confidential related to diplomatic and international matters, and you are
Chairman of the Senate Committee on Foreign Relations. It would not have
been said to you."

Sumner reflected a moment, then answered:

"I suppose you are right. I won't ask you about anything which you think
you ought not to repeat. But you must consider that, notwithstanding all that
Bismarck has accomplished, he is still an unknown force. My own belief is
that the future of Germany lies in his hands. The man who could defy the
public opinion of Europe in that business with Denmark, who could defy
the public opinion and Parliament of Prussia, who could govern four years
without a Budget or a majority, who could make war without supplies, and
without his country behind him, and his King only a convert at the last
moment to his policy—that man, though he has put Austria under his feet
and Prussia in Austria's place at the head of Germany, is, in my judgment,
only at the beginning of his career. He is the one supremely interesting
figure in Europe at this moment. I have never met him; probably may never
meet him. But it is important to me to know all I can about him. Violate no
confidence, but tell me what you can. I will make no use of it except to
inform my own mind. When I have to deal with Count Bismarck, I want to
be able to picture to myself what manner of man he is. In diplomacy, a
knowledge of men is half the battle."
 This long speech was characteristic of Sumner. He was seldom brief or
simple. His mind overflowed. In private, as in public, he was oratorical. The
sentences, as they came from his lips, seemed to have passed through a
mould. He spoke with a model before him. The most sincere of men he was
never content to be himself and nobody else. In the murmur of the flowing
periods he often uttered, you heard echoes of Cicero, of Bossuet, of Burke.
Perhaps it was true of him—as Emerson said, not of him—that his library
overloaded his wit. He moved as if in armour; a mixed but apt metaphor.
The chair in which he sat was a platform, and his one listener was an
audience. He neglected, in his private talk, none of the arts of the
rhetorician. Whoever has heard Sumner in the Senate or in Faneuil Hall
must remember the imposing presence of the man; his stature: and the
leonine head with its waving black mane which every moment he tossed
from his forehead, only to have it fall again half over his eyes. The strong
features stood out sharply, the eyes were alight, the lips moulded into
plastic form the most stubborn sentences, and the whole blended into one
expression after another at the will of the speaker; each expression the
visible image of his thought. He was so intent on bending his audience to
his will that he used without stint every weapon at his command.

In private, all this was a little overwhelming. As it comes back to me in

memory, my view of it is probably more critical than it was while I sat and
looked and listened. But it still seems to me extremely fine. In England—
the country of all others where simplicity counts for most—Sumner was
thought emphatic; and the English do not like emphasis, but they liked
Sumner. He was first here as a young man, in 1838 and 1840, when he was
still in the late 'thirties; and these mannerisms were presumably less
mannered, or less aggressive. But the men and women whom Sumner then
came to know were men and women who dwelt on the heights. I suppose
the average of serious culture at that time in that class was at least as high
as it is now. They liked a man with a full mind. Sumner had that; and he
poured it out in a flood.

Macaulay had taught his set, or the several sets to which he more or less
belonged, to endure conversation which took the form of monologue and
rivalled the laborious accuracy of a cyclopædia. People suffered under him.
Lady Holland and Hayward and Lord Melbourne and others rebelled, but
there were not many who rebelled. Sumner's path had therefore been made
plain, nor was he dogmatic in Macaulay's way. He was human and his
enthusiasms were human, and he was sympathetic.

But when Sumner, in 1869, made his indirect Claims speech in the
Senate, seeking to induce the Government to demand from England indirect
damages for the depredations of the Alabama, his popularity in this country
came to a sudden end. His best friends were those who resented this speech
most hotly; and Mr. Bright most of all. To Mr. Bright I once undertook to
defend Sumner or to explain him, for I thought he had been misunderstood.
But Mr. Bright would not have it. "The only defence is silence," he
exclaimed, and he was the more angry when I said: "That will do for an
epigram." And we never referred to it again.

So far as I could, I satisfied Sumner's interest about Bismarck, whom I

had seen at short range, and with whom, on the evening in question, I had
spent some three hours alone. Sumner asked question after question, with
one definite object; he wanted to understand the man himself. Once or twice
he put a searching interrogatory on matters of diplomacy, or on the relations
between the King and his great Minister, which had to be answered with
reserve. He showed an astonishing knowledge of purely Prussian politics
and even of Prussian politicians. He asked if it was true that Loewe and the
other Liberals had owned they were wrong in opposing Bismarck, and
when I said yes, exclaimed: "Then they showed more good sense than I
expected."

I spent some days with Mr. Sumner in his house in Lafayette Square, in
Washington, now part of a Washington hotel. A plainly furnished house,
hardly a home; chiefly remarkable for its books and for Sumner. He was a
kindly host, anxious that his guest should make the most of his visit, and
see the men he wanted to see. I wanted to ask him why he had, on a former
visit, advised me not to see Lincoln; but I did not. But Lincoln was now
dead and among the giants who survived him Sumner was the most
attractive personality.

He became more attractive still some years later, in 1872, when he came
to Europe for the rest which his long warfare, first with President Johnson
and then with President Grant, had made imperative. He came first to
London, staying—or, as the English perversely say, stopping—at Penton's
Hotel, St. James's Street; then a hostelry of repute, now extinct. He had a
large suite of rooms on the ground floor at the back; gloomy, and intensely
respectable. I dined with him the night of his arrival. "I don't know what
kind of a dinner they will give us," said Sumner, "but you shall have a bottle
of Chateau Lafitte of 1847, and the rest will matter less." He loved good
Bordeaux, as all good men do; and his talk flowed like old wine—a full,
pure stream, with both flavour and bouquet; and not much of the best claret
has both.

It is not possible to repeat much of Sumner's talk, for it was mostly

personal and intimate. But I asked him whether he still felt the effects of
those coward blows which Preston Brooks had dealt him from behind as he
sat imprisoned in his chair in the Senate. He was not sure. He doubted
whether he had ever completely recovered, though it was now some sixteen
years since that particular piece of South Carolina chivalry had been
perpetrated. He thought everything had been done for him which could be
done. What he told me may or may not have been printed. I do not know.
When the moxa was to be applied to his spine, Dr. Charcot proposed to give
him an anæsthetic. "But," said Sumner, "does not the effect you seek to
produce—the counter-irritation—depend more or less on the pain the
patient would endure without the anæsthetic?" "Yes," Charcot admitted,
reluctantly, "it probably does." "Then let us go ahead without ether," said
Sumner; and they did. I understood the treatment consisted in laying along
the spine cotton-wool soaked in oil and setting fire to it. When, after two or
three days, the burn is partly healed, the operation is renewed, and the pain,
of course, more severe. But no ether was administered. After his first attack
of angina pectoris, "the pain," said Sumner, "which I endured in a single
second from one of those spasms was more than all I ever suffered from all
the applications of the moxa."

We went together from London by way of Boulogne to Paris, staying

two nights at Boulogne at one of the beach hotels. Sumner was like a boy;
his sixty-one years sat lightly on him and his interests were as fresh as I had
ever known them. He loved the sea and the sea air; an air so much more
exhilarating on the southern coast of the Channel than the northern. He was
amused to hear that the customs authorities had passed all our luggage—his
and mine—because I had told them he was a Senator; and still more amused
later when the Dover customs on our return had shown him the same
indulgence as "The Honourable Charles Sumner"—honourable denoting in
England not political distinction, but membership of a family the head of
which is a peer. In Paris, as in London, we had rambled about the book-
shops. "I dare say," remarked Sumner, "you thought from my books at home
that I cared nothing for books as books; or for bindings. But you will see."
And he proceeded to buy a certain number of so-called fine bindings:
which, alas, were not so fine as they ought to have been.

Less than two years after his last months in Europe, he died. I have still
much to say about him, and there are many letters of his to me which I hope
to print; but they are not here and I must end. When I remember what has
been said so often of Sumner by men who did not know him or did not like
him, I may be allowed to end with a tribute of affection. I thought him, and
I shall ever think him, one of the most lovable of men; more than loyal to
his friends, delighting in kindnesses to them; of an implacable honesty,
sincerity, devotion to duty and to high ideals; an American to whom
America has paid high honour, but never yet enough.

CHAPTER XIV

EXPERIENCES AS JOURNALIST DURING THE

CIVIL WAR

My obligations to Wendell Phillips are mixed, and one of them was an

introduction to The Tribune. In the autumn of 1861 I wanted two things: a
holiday, and a chance to see something of the war and the negro question at
short range. At that time, Mr. Charles A. Dana was managing editor of The
Tribune, with Mr. Sydney Howard Gay as his first lieutenant. Phillips gave
me a letter to Mr. Gay, the result of which was that Mr. Dana asked me to
go to South Carolina for The Tribune.
 A word about Mr. Dana. He had the reputation at that time of being what
the cabman called that Mr. John Forster who was, among other things, the
friend and biographer of Dickens—"a harbitrary gent." I suppose Mr. Dana
was arbitrary; in the sense that every commanding officer must be arbitrary.
But my relations with him, or my service under him, lasted some months,
during the whole of which period I found him considerate and kindly. He
liked, I think, to assign a man to duty and judge him by the result; which
meant that the man was left free to work out his own salvation; or
damnation, as the case might be.

I was, of course, perfectly new to the business of journalism and, equally

of course, made many mistakes. But Mr. Dana was not the kind of manager
who fastened on this mistake or that as an occasion for chastising the
offender. He judged a man's work as a whole. In the office, I am told, he
sometimes thought it needful to speak plainly in order to enforce a steady
discipline. He had been known to walk into the room of one of the
departmental editors, in full view and hearing of the whole staff, and
remark: "Mr. X, you were disgracefully beaten this morning," in the tone in
which he might have said it was a fine day. But the next morning Mr. X was
not beaten; nor the next.

Very possibly, between me and Mr. Dana's wrath, if I roused it, stood Mr.
Gay; a man of soft manners and heart. I cannot remember that, directly or
indirectly, any reprimand ever came to me from Mr. Dana. From Mr.
Greeley there came more than one; all well deserved. With the business of
managing the paper Mr. Greeley did not much concern himself. With the
results he sometimes did, and when The Tribune did not contain what he
thought it ought to contain, he was apt to make remarks on the omission.
While I was at Port Royal in South Carolina there was a skirmish at
Williamston in North Carolina, a hundred miles away. Mr. Greeley thought
I ought to have been at Williamston. Very likely I ought. But Lord Curzon
had not at that time announced his memorable definition of enterprising
journalism; "an intelligent anticipation of events that never occur." That
epigram, delivered in the House of Commons, may be supplemented by an
axiom. The business of a war correspondent is to be, not where he is
ordered, but where he is wanted.
 In the early days of the Civil War—or, for that matter, in the late days—
the American Press had little of the authority it has since acquired. The
heads of great departments of Government still held themselves responsible
primarily to the President. Berths on battleships were not then at the
disposal of the first journalists who wanted one. When I asked Commodore
Steadman of the Bienville to take me to Port Royal he politely told me it
was against the naval regulations to allow a civilian on board a ship of war.
When I asked him who had a dispensing power in such matters, he said: "If
the Secretary of the Navy should order me to receive you as a guest, I
should do so with pleasure." I thanked him and with the courage of which
ignorance is the mother, telegraphed Mr. Welles. No answer. I telegraphed
again, saying it was the wish of Mr. Dana that I should go to South Carolina
on the Bienville. The effect of Mr. Dana's name was magical, and this time
an answer came; that Commodore Steadman had orders to give me a berth.
I suppose the journalists of to-day will hardly understand how there could
have been a difficulty. But there were to be many difficulties. Commodore
Steadman was as good as his word, and better; and a kind host.

Admiral Dupont had captured the Port Royal forts by the time I arrived.
A finer example of the old type of naval officer than Admiral Dupont our
naval service never had. Captain Raymond Rodgers was his flag captain;
another example not less fine. General W. T. Sherman was in command of
the land forces. The winter passed slowly away. There was not much to do
except study the negro question; which was perhaps more attractive when
studied at a distance. General Butler, bringing the mind of a lawyer to bear
on the problems of war, and desiring a legal excuse for annexing the
personal property of the enemy had announced that the negroes were
"contraband of war." For him, the maxim that laws are silent amid arms did
not hold good. He liked to make laws the servant of arms. The negroes
naturally came soon to be known as contrabands. There were some months
during which they were called hardly anything else. I called them so in my
letters. It was characteristic of Phillips that, after a time, he wrote to me to
suggest that Butler's phrase had done its work and that the negro was a
negro: a man entitled to freedom on other grounds.

But it was long before the word passed out of use. Butler had chosen the
psychological moment. The "contrabands"—with Mr. Phillips's permission
—who crowded the camps were mostly from the cotton and rice plantations
of South Carolina and Georgia. If you were not already a convinced
Abolitionist, they were not likely to convert you. But it was becoming daily
clearer that the negro had a military value; not at Port Royal, however,
where he was only a burden. It was not an eventful winter at Port Royal.
There were expeditions by land and sea, and there was the taking of Fort
Pulaski, which I saw, but I was glad to return to New York in the spring;
and then to join General Frémont in the Shenandoah Valley. The name of
that commander was still one of promise. Except the name, there was not
much else for the purposes of war, but he had a charm of manner and a
touch of romance and a staff on which one or two foreign adventurers had
places and did weird things. "General" Cluseret was one; an impostor who
afterward found a congenial home in the Paris Commune, with other
impostors. That campaign came to nought, and when General Pope, in July,
1862, was put in command of the Army of Virginia, I found my way to the
headquarters of that redoubtable warrior.

With him, in command of the Third Army Corps, was General

McDowell. I don't know why one's memory chooses trivialities as proper
objects of its activity, but it sometimes does. One of the most vivid among
the impressions of those days is the stout figure of General McDowell on
his horse, which he sat ill, his uniform awry, his sword pushed behind him
as far as it would go, his strapless trousers ending abruptly halfway between
knee and ankle; then a space of bare flesh, and then some inches of white
stocking, and then a shoe. But he had military gifts if not a military air. He
was talking with General Pope, whose unhappy proclamation about his
headquarters in the saddle had already been issued. Unlike McDowell, Pope
looked a better soldier than he was. His six weeks' generalship on the
Rappahannock ended with the Second Bull Run, which there was now no
Billy Russell to describe in words that blistered yet were honest words; and
with Chantilly. The West suited Pope better than the East, and to the West
he returned. In these six weeks he had made nothing but mistakes and
achieved only defeats.

Personally, General Pope was pleasant to deal with. It was while he

commanded the Army of Virginia that Mr. Stanton, then Secretary of War,
or perhaps General Halleck, issued orders for the expulsion of all
correspondents from the armies in the field. General Pope sent for me and
told me of the order. Impressed at that time with the sternness of War Office
rule, I answered meekly that I supposed I must go. Said General Pope,
"This is not an official interview. I imagine you needn't go till you get the
order." A battle was thought to be imminent; any respite was welcome. I
thanked him, went back to my tent, took what I most needed, and rode off
to an outpost where I had a friend. The official notification may have been
sent to my tent but never reached me. And so it happened that I saw such
fighting as there was on the Rappahannock, and at the Second Bull Run,
better called Manassas. Interesting to a student of war; not inspiriting to a
patriot; and not now to be described even in the briefest way. My only aim
is to give the reader of to-day some faint notion of what a war
correspondent's life in those days was like.

One incident I may note, as an example of what may happen to a general

who neglects the most elementary rules and precautions of war. At the end
of a day's march, at sundown but the heavens still light, General Pope
bethought himself that he should like to see what the country ahead of him
looked like. With his staff and a bodyguard of some sixty sabres he rode up
a low hill with a broad crest, open ground about it for a hundred yards, and
beyond that in front a thick, far-spreading forest line. General Pope and his
staff dismounted. The cavalry were ordered to dismount and loosen their
saddle-girths. Just as this operation had been completed there came from
the wood beyond the open ground a rifle volley. As we stood between the
sunset and the enemy we were a pretty fair target. There was no time for
orders. Everybody scrambled into his saddle as best he could and away we
went.

But the firing woke up the advance guard of our army, and they also
began firing. It soon appeared that General Pope had unwittingly passed
outside his own lines, so that, as we rode away from the fire of the Rebels
we rode into the fire of our own troops. It was hot enough but luckily did
not last long. The hill partly protected us from the sharpshooters in grey,
and our fire was silenced after a moment. But the horses were well
frightened. It was impossible to pull up. We scattered and the horses went
on for a mile or so. I never before so much respected the intelligence of that
animal. There was nothing to do but sit down in the saddle, but the horses
never made a mistake at full speed over an unknown country, stiff with
fences and brooks, and nobody came to grief; nor, which seems more
wonderful, was anybody hit by the bullets. A good many remarks were
made which hit General Pope.
 CHAPTER XV

CIVIL WAR—GENERAL McCLELLAN—

GENERAL HOOKER

The failure of Pope's campaign and his retreat upon the Capital
demoralized his army and demoralized Washington to an extent which few
remember. The degree of the demoralization may, however, be measured by
the reappointment of General McClellan to the command of the Army of
the Potomac and of Virginia. In the absence of any general whose name
inspired confidence, General McClellan was thought a synonym of safety,
or, at any rate, of caution, and he had not wholly lost the confidence of his
men. He was not expected to enter upon large operations.

An engagement near Washington was, however, thought probable. On a

hint from a friendly official I rode out one afternoon from Washington to
the army headquarters, expecting to be away at most a day or two. My
luggage consisted of a mackintosh and a tooth-brush. I was absent six
weeks. But this was not so tragic as it sounds, for Maryland was a country
in which, even with a war afoot, it was possible to buy things. In the
interval, I had seen two battles; South Mountain and Antietam, which came
as near to being real war as could be expected under General McClellan.

Correspondents were not now allowed with the army in the field any
more than in General Pope's time. We were contraband. But so long as we
yielded nominally to the inhibition of the War Office nobody seemed to
care. The War Office was then named Edwin M. Stanton. To this day I have
never been able to understand how Mr. Stanton—a man all energy,
directness of mind and purpose, scorning compromise and half measures
and scorning those who practised them—came to assent to the replacing of
General McClellan at the head of the Army of the Potomac. But he did, and
at first General McClellan seemed to justify the new hopes newly placed in
him. He might have sat still, but after providing for the defence of
Washington he moved out upon an aggressive-defensive campaign. General
Lee had entered Maryland and McClellan went in search of him. He moved
slowly, but he moved. His soldiers, so far as I could judge, believed in him
in spite of his disasters in the Peninsula. His generals, I think, did not. I saw
and talked with some of them, for I found myself making this campaign as a
volunteer aide-de-camp to General Sedgwick. I had met General Sedgwick
before, and when I had to consider how I was to get leave to go with the
troops I went to General Sedgwick and told him my difficulty. "Come along
with me," he said. That was all the appointment I had. It would not have
been possible in a European army, but in the armies of the Union many
things were possible. And it was quite sufficient to take me outside of Mr.
Stanton's order about correspondents. I was not a correspondent; I was one
of General Sedgwick's aids. His kindness to me was a service for which I
could never be too grateful.

It was a still greater service because General Sedgwick belonged in the

category of fighting generals, who were none too popular with the general
commanding, since he, mixing politics with war, believed in half-beating
the enemy. Sedgwick, so far as I know, had no politics. Certainly he had
none in the field. He was there to fight, not to build bridges over which the
Rebels might come back into the Union. It had become known that General
Lee had entered Maryland, to enable her people "to throw off a foreign
yoke." He was not, as it turned out, a welcome guest. Maryland would have
been much obliged to him if he had stayed on the other side of the Potomac.
McClellan, taking time to think things over, and perhaps not liking to be
considered a foreign yoke, advanced toward Frederick, Lee's headquarters
for the moment, at the breakneck pace of six or seven miles a day. I suppose
McClellan must have known that Lee wanted Harpers Ferry. But even after
Lee's general order had come into his possession, with specific directions
for the movement of each division, McClellan hesitated and finally took the
wrong road.

Hence the battle of South Mountain; a picturesque performance; part of

which I watched by the side of General McClellan himself. At the moment
he was quite alone; his staff away carrying orders; an officer now and then
returning only to be sent off again at once. The general presently saw that a
stranger was standing near him and asked a question or two. I offered him
my field glasses, but he said he could see very well and declined them.
 There was in his appearance something prepossessing if not
commanding: something rather scholarly than warlike; amiable, well-bred,
cold, and yet almost sympathetic. His troops were slowly forcing their way
up the steep mountain side upon which we looked. It was, in fact, from a
military point of view, a very critical moment, but this general commanding
had a singular air of detachment; almost that of a disinterested spectator: or
of a general watching manoeuvres. The business of war seemed to be to him
merely what Iago calls "the bookish theoric"; and he himself "a great
arithmetician." He had the face of a man of thought. Napoleonic, said his
idolaters, who called him the young Napoleon: not considering dates, or not
aware that when Napoleon planned and won his great Italian campaign, a
masterpiece of war, he was twenty-seven. When McClellan planned and
lost his Peninsula campaign, he was thirty-seven. But there he stood; an
interesting figure; as if stargazing. Compact, square-chested, his face well
moulded. That he was directing the assault of the forces struggling up
yonder hill no human being could have guessed. Whether his tailor had
been too stingy in the material of his uniform, or Nature too lavish in the
contents of it, he was uncomfortable; he and his clothes did not seem made
for each other. There were wrinkles. There was a missing button; nor was
he a well set-up figure. It may well enough have been because of his
military career, but I thought an air of indecision hung about him. Men had
died by hundreds and were yet to die because he could not make up his
mind, nor push an attack home. They were dying now, as he looked on; they
lay dying and dead on the opposite slope; for when he had at last made up
his mind he had made it up wrong. The battle of South Mountain was a
victory in a sense, but it need never have been fought. A position which
might have been turned had been forced, and the road to Antietam lay open.

Again it was like McClellan, on approaching Sharpsburg and the

battleground of Antietam, to halt and think it over. If he had struck at once,
he would have found Lee's army divided and the path weakly held. But
McClellan had it not in him to do anything at once, or to do it once for all.
The armies faced each other idly all that day. In the afternoon I heard that a
flank movement on the enemy's left was to be tried under General Hooker.
So I rode over and joined that general's command. It was well known that
Hooker would fight if he was allowed. He was already called "Fighting
Joe"; a well-earned sobriquet. He put his troops in motion about four
o'clock that afternoon, himself at the head as usual, doing his own
reconnoitring. I rode with the staff, not one of whom I knew. Nobody took
the trouble to ask who I was or why I was there. For aught they knew I
might have been a Rebel spy.

General Hooker had his own way of doing things. This was what might
be called a reconnaissance in force; two brigades in line pushing steadily
forward; a force of cavalry in advance, two divisions following. By the time
we came in touch with Lee's left, it was dusk. We could see the flashes of
the Rebel rifles which drove Hooker's cavalry back upon the infantry
division. Hooker played the game of war as the youngest member of a
football team plays football. He had to the full that joy of battle which
McClellan never had at all; and showed it.

Between the man by whose side I had stood two days before at South
Mountain, and the man near whom I now rode, the contrast was complete.
McClellan was not a general; he was a Council of War, and it is a military
axiom that councils of war never fight. He surveyed the field of battle
beneath him at Turner's Gap as a chess-player surveys the board. At the
naval battle of Santiago, as the Spanish ships were sinking, our bluejackets
began to cheer. Said Admiral Philip: "Don't cheer, boys. They are dying
over there." If everything else about Philip should be forgotten, that will be
remembered; and he will be loved for it; for this one touch of human feeling
for a human enemy amid the hell of war. But for the pawns and pieces the
chess-player sends to slaughter he has no regrets. I don't say McClellan had
none for the men whom his mistaken strategy drove to death. All I say is
that as I looked at him I saw no sign of it. A general, we are told, can no
more afford to have feelings amid a battle than a surgeon with the knife in
his hand can feel for his patient. It may be. But Napoleon, who is always
cited as the highest example of indifference to the lives of men, is perhaps
the best example to the contrary. He would sacrifice a brigade without
scruple for a purpose; never one single armed man without a purpose. He
had men enough to consume for victory; never one to squander. He was an
economist of human life, though for purely military reasons. It is awful to
reflect how many thousands of Americans in these early Civil War days
were sent to death uselessly by the ignorance of their commanders; or as in
McClellan's case by his irresolution, and his incapacity for the handling of
troops in the field.

General Hooker's was a face which lighted up when the battle began.
The man seemed transformed. He rode carelessly on the march, but sat
straight up in his saddle as the martial music of the bullets whistled past
him. He was a leader of men, and his men would have followed him and did
follow him wherever he led. Hesitation, delay, he hated them. "If they had
let us start earlier we might have finished to-night," he muttered. But night
was upon us, and even Hooker could not fight an unknown force on
unknown ground in the dark. It was nine o'clock when we went into camp;
Union and Rebel lines so close that the pickets got mixed and captured each
other. "Camp" is a figure of speech. We lay down on the ground as we were.
I slept with my horse's bridle round my arm. At four o'clock next morning,
with the earliest light of a coming dawn and as soon as a man could see the
sights on his rifle, the battle began.

CHAPTER XVI

CIVIL WAR—PERSONAL INCIDENTS AT

ANTIETAM

General Hooker was about the first man in the saddle. The pickets had
begun sniping long before dawn. My bivouac was within sight of his tent.
"The old man," said one of his staff, "would have liked to be with the
pickets." No doubt. He would have liked to be anywhere in the field where
the chance of a bullet coming his way was greatest. Kinglake has a passage
which might have been written for Hooker. That accomplished historian of
war remarks that the reasons against fighting a battle are always stronger
than the reasons for fighting. If it were to be decided on the balance of
arguments, no battle would ever be begun. But there are Generals who have
in them an overmastering impulse of battle; it is in the blood; temperament
Welcome to Our Bookstore - The Ultimate Destination for Book Lovers
Are you passionate about books and eager to explore new worlds of
knowledge? At our website, we offer a vast collection of books that
cater to every interest and age group. From classic literature to
specialized publications, self-help books, and children’s stories, we
have it all! Each book is a gateway to new adventures, helping you
expand your knowledge and nourish your soul
Experience Convenient and Enjoyable Book Shopping Our website is more
than just an online bookstore—it’s a bridge connecting readers to the
timeless values of culture and wisdom. With a sleek and user-friendly
interface and a smart search system, you can find your favorite books
quickly and easily. Enjoy special promotions, fast home delivery, and
a seamless shopping experience that saves you time and enhances your
love for reading.
Let us accompany you on the journey of exploring knowledge and
personal growth!

ebookgate.com