Categorical And Nonparametric Data Analysis

Choosing The Best Statistical Technique 1st

Edition E Michael Nussbaum download

https://ebookbell.com/product/categorical-and-nonparametric-data-
analysis-choosing-the-best-statistical-technique-1st-edition-e-
michael-nussbaum-34121254

Explore and download more ebooks at ebookbell.com

Here are some recommended products that we believe you will be
interested in. You can click the link to download.

Categorical And Nonparametric Data Analysis 2nd E Michael Nussbaum

https://ebookbell.com/product/categorical-and-nonparametric-data-
analysis-2nd-e-michael-nussbaum-56329242

Applied Categorical And Count Data Analysis Second Edition 2nd Edition
Wan Tang Hua He Xin M Tu

https://ebookbell.com/product/applied-categorical-and-count-data-
analysis-second-edition-2nd-edition-wan-tang-hua-he-xin-m-tu-48249068

Applied Categorical And Count Data Analysis 2nd Edition Wan Tang

https://ebookbell.com/product/applied-categorical-and-count-data-
analysis-2nd-edition-wan-tang-49420098

Statistics Categorical And Time Dependent Data Analysis Mathematics

Research Developments Suzuki

https://ebookbell.com/product/statistics-categorical-and-time-
dependent-data-analysis-mathematics-research-developments-
suzuki-32354852
Applied Categorical And Count Data Analysis 1st Edition Tang

https://ebookbell.com/product/applied-categorical-and-count-data-
analysis-1st-edition-tang-33727342

Applied Categorical And Count Data Analysis Tang Wanhe Huatu

https://ebookbell.com/product/applied-categorical-and-count-data-
analysis-tang-wanhe-huatu-232343782

Regression Models For Categorical And Count Data Peter Martin

https://ebookbell.com/product/regression-models-for-categorical-and-
count-data-peter-martin-46541038

Generalized Linear Models For Categorical And Continuous Limited

Dependent Variables Michael Smithson

https://ebookbell.com/product/generalized-linear-models-for-
categorical-and-continuous-limited-dependent-variables-michael-
smithson-52688452

Regression Models For Categorical And Count Data Martin Peter

https://ebookbell.com/product/regression-models-for-categorical-and-
count-data-martin-peter-232949664
CATEGORICAL AND NONPARAMETRIC
DATA ANALYSIS

Featuring in-depth coverage of categorical and nonparametric statistics, this book pro-
vides a conceptual framework for choosing the most appropriate type of test in various
research scenarios. Class tested at the University of Nevada, the book’s clear explanations
of the underlying assumptions, computer simulations, and Exploring the Concept boxes
help reduce reader anxiety. Problems inspired by actual studies provide meaningful illus-
trations of the techniques. The underlying assumptions of each test and the factors that
impact validity and statistical power are reviewed so readers can explain their assump-
tions and how tests work in future publications. Numerous examples from psychology,
education, and other social sciences demonstrate varied applications of the material. Basic
statistics and probability are reviewed for those who need a refresher. Mathematical deri-
vations are placed in optional appendices for those interested in this detailed coverage.
Highlights include the following:

• Unique coverage of categorical and nonparametric statistics better prepares readers to se-
lect the best technique for their particular research project; however, some chapters
can be omitted entirely if preferred.
• Step-by-step examples of each test help readers see how the material is applied in a
variety of disciplines.
• Although the book can be used with any program, examples of how to use the tests in SPSS
and Excel foster conceptual understanding.
• Exploring the Concept boxes integrated throughout prompt students to review key
material and draw links between the concepts to deepen understanding.
• Problems in each chapter help readers test their understanding of the material.
• Emphasis on selecting tests that maximize power helps readers avoid “marginally”
significant results.
• Website (www.routledge.com/9781138787827) features datasets for the book’s exam-
ples and problems, and for the instructor, PowerPoint slides, sample syllabi, answers
to the even-numbered problems, and Excel data sets for lecture purposes.

Intended for individual or combined graduate or advanced undergraduate courses

in categorical and nonparametric data analysis, cross-classified data analysis, advanced
statistics and/or quantitative techniques taught in psychology, education, human devel-
opment, sociology, political science, and other social and life sciences, the book also
appeals to researchers in these disciplines. The nonparametric chapters can be deleted if
preferred. Prerequisites include knowledge of t tests and ANOVA.

E. Michael Nussbaum is a professor of educational psychology at the University of

Nevada, Las Vegas.
 CATEGORICAL AND
NONPARAMETRIC DATA
ANALYSIS
Choosing the Best Statistical Technique

E. Michael Nussbaum
First published 2015
by Routledge
711 Third Avenue, New York, NY 10017

and by Routledge
27 Church Road, Hove, East Sussex BN3 2FA

Routledge is an imprint of the Taylor & Francis Group, an informa business

© 2015 Taylor & Francis

The right of E. Michael Nussbaum to be identified as the author of

this work has been asserted by him 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.

Library of Congress Cataloging in Publication Data

A catalog record for this book has been requested

ISBN: 978-1-84872-603-1 (hbk)

ISBN: 978-1-138-78782-7 (pbk)
ISBN: 978-0-203-12286-0 (ebk)

Typeset in Stone Serif

by Apex CoVantage, LLC
This book is dedicated to my mentors who taught me secrets of
statistics: Edward H. Haertel, Henry E. Brady, and David R. Rogosa.
It is also dedicated to my ever patient wife, Lorraine.
CATEGORICAL AND NONPARAMETRIC
DATA ANALYSIS
 BRIEF CONTENTS

Preface xv
About the Author xix

CHAPTER 1 Levels of Measurement, Probability, and

the Binomial Formula 1

CHAPTER 2 Estimation and Hypothesis Testing 25

CHAPTER 3 Random Variables and Probability Distributions 57

CHAPTER 4 Contingency Tables: The Chi-Square Test and

Associated Effect Sizes 81

CHAPTER 5 Contingency Tables: Special Situations 99

CHAPTER 6 Basic Nonparametric Tests for Ordinal Data 129

CHAPTER 7 Nonparametric Tests for Multiple or Related Samples 177

CHAPTER 8 Advanced Rank Tests (for Interactions and Robust ANOVA) 215

CHAPTER 9 Linear Regression and Generalized Linear Models 257

CHAPTER 10 Binary Logistic Regression 285

CHAPTER 11 Multinomial Logistic, Ordinal, and Poisson Regression 335

CHAPTER 12 Log-Linear Analysis 377

CHAPTER 13 General Estimating Equations 407

CHAPTER 14 Estimation Procedures 431

CHAPTER 15 Choosing the Best Statistical Technique 469

Answers to Odd-Numbered Problems 473

Index 497
CATEGORICAL AND NONPARAMETRIC
DATA ANALYSIS
 CONTENTS

Preface xv
About the Author xix

CHAPTER 1 Levels of Measurement, Probability,

and the Binomial Formula 1
Levels of Measurement 1
Probability 3
The Binomial Formula 10
Problems 21
Appendix 1.1: Logic Behind Combination Formula (Eq. 1.9) 23

CHAPTER 2 Estimation and Hypothesis Testing 25

Estimation 25
Hypothesis Testing 35
The Binomial Test 39
Summary 51
Problems 51
Technical Notes 52
Appendix 2.1: Derivation of the Mean and Standard Error of
the Sampling Distribution for P 54

CHAPTER 3 Random Variables and Probability

Distributions 57
Random Variables 57
Testing the Shape of a Distribution 68
x Contents

Problems 73
Technical Notes 78
Appendix 3.1: Linear Combination of Two Random Variables 79

CHAPTER 4 Contingency Tables: The Chi-Square

Test and Associated Effect Sizes 81
The Contingency Table 81
Effect Sizes 88
Measures of Association 92
Using SPSS 93
Summary 94
Problems 95
Technical Note 97

CHAPTER 5 Contingency Tables: Special Situations 99

The Small Sample Case: Fisher’s Exact Test 99
McNemar Test (for Related Samples) 106
Controlling for a Third Variable: The Mantel-Haenszel Test 112
Summary 118
Problems 118
Technical Notes 122
Appendix 5.1: Understanding the Hypergeometric Distribution 123
Appendix 5.2: Understanding the Mantel-Haenszel Formula 125

CHAPTER 6 Basic Nonparametric Tests for

Ordinal Data 129
Introduction to Nonparametric Statistics 130
Analysis of Association or Trend 137
Comparing Two Groups 149
Robust Methods for Detecting Outliers 160
Summary 163
Problems 163
Technical Notes 169
Appendix 6.1: Computational Formula for τ̂ b 172
Appendix 6.2: Sum of Consecutive Integers 173
 Contents xi

CHAPTER 7 Nonparametric Tests for Multiple or

Related Samples 177
Kruskal-Wallis Test for Multiple Independent Groups 177
Sign Test for Pre- and Posttest Data 187
Wilcoxon Signed-Ranks Test 190
Friedman Test 194
Page Test for Ordered Alternatives 199
Summary 202
Problems 202
Technical Notes 208
Appendix 7.1: Logic Behind Kruskal-Wallis Eq. 7.1 209
Appendix 7.2: Derivation of Eq. 7.4 for the Asymptotic Sign Test 210
Appendix 7.3: Example and Theory Behind the Wilcoxon
Signed-Ranks Test 211

CHAPTER 8 Advanced Rank Tests (for Interactions

and Robust ANOVA) 215
Aligned Rank Transform 216
Robust ANOVA-Type Statistics 230
Summary 245
Problems 246
Technical Notes 248
Appendix 8.1: Technical Details on the ATS Procedure 250
Appendix 8.2: Derivation of Eq. 8.5c (Definition of
a Two-Way Interaction Effect) 252
Appendix 8.3: Confidence Intervals for Relative Treatment Effects 253

CHAPTER 9 Linear Regression and Generalized

Linear Models 257
Review of Linear Regression 257
The General Linear Model 264
Generalized Linear Models 269
Summary 275
xii Contents

Problems 276
Technical Notes 280
Appendix 9.1: Proofs of Selected OLS Theorems 282

CHAPTER 10 Binary Logistic Regression 285

Basics of Logistic Regression 285
Model Building and Refinement 305
Summary 321
Problems 322
Technical Notes 325
Appendix 10.1: Selected Proofs 328
Appendix 10.2: Goodness of Link Test 329
Appendix 10.3: Dean’s Test for Overdispersion 331

CHAPTER 11 Multinomial Logistic, Ordinal, and

Poisson Regression 335
Multinomial Logistic Regression 335
Ordinal Regression 338
Poisson Regression 347
Summary 365
Problems 365
Technical Notes 371
Appendix 11.1: Poisson Regression Proofs 372
Appendix 11.2: The Negative Binomial Distribution 374

CHAPTER 12 Log-Linear Analysis 377

Basic Theory 377
Ordered Contingency Tables 391
Sample Considerations 396
Summary 397
Problems 397
Technical Notes 402
 from Log-Linear Dummy Parameters
Appendix 12.1: Derivation of OR 404
Appendix 12.2: Conditional Log-Linear Analysis:
Mathematical Foundations 405
 Contents xiii

CHAPTER 13 General Estimating Equations 407

Foundations of General Estimating Equations 407
Performing General Estimating Procedures in SPSS 424
Summary 425
Problems 425
Technical Notes 428

CHAPTER 14 Estimation Procedures 431

Estimating a Binomial Parameter 431
Estimation Procedures for Generalized Linear Models
and General Estimating Equations 444
Summary 453
Problems 454
Technical Notes 456
Appendix 14.1: Logic Behind the Newton-Raphson Procedure 458
Appendix 14.2: Proofs of Important Results 460

CHAPTER 15 Choosing the Best Statistical Technique 469

Summary 472

Answers to Odd-Numbered Problems 473

Index 497
CATEGORICAL AND NONPARAMETRIC
DATA ANALYSIS
 PREFACE

When scientists perform research, they often collect data that do not come in a
form that is suitable for traditional methods, such as analysis of variance (ANOVA)
or ordinary linear regression. Categorical and nonparametric data analysis are use-
ful for data that are nominal or ordinal, or when assumptions of traditional metric
tests have been violated, for example, when the sample is small. Finally, even when
all the assumptions of traditional tests have been met, alternative tests are some-
times more statistically powerful, meaning that those tests are more likely to find a
statistically significant result when there is a real effect to be found. There is noth-
ing more frustrating than to spend a year or two on a research project, only to have
your results come in as not quite significant (e.g., p = .07).
This book was developed both out of my efforts to advise students on disserta-
tions as well as from my own research in the field of argumentation. I found that
while much of what I was taught about statistics in graduate school was useful at
times, it was also somewhat limited. It was as if I was only taught half the story.
In fact, one of my statistics teachers at Stanford University was upset that cover-
age of categorical data analysis (CDA) was no longer required, because there was
too much material already crammed into the statistics core. It was only later, after
several years conducting research with actual messy data sets, that I discovered the
immense value of categorical and nonparametric data analysis.
It is important to know when to use these techniques and when to use tradi-
tional parametric analysis. One major goal of the book is to provide a conceptual
framework for choosing the most appropriate type of test in a given situation. One
has to consider the underlying assumptions of each test and the factors that impact
each test’s statistical power. One also has to be able to explain these assumptions
(and how the test works) conceptually to both oneself and others (including the
audience of a journal article or dissertation). This allows one to make statistically
based arguments that can serve as a rationale for using that test. Therefore, another
major goal of this book is to provide readers with a conceptual framework that
underlies the statistical methods examined and, to some extent, traditional para-
metric methods as well.
xvi Preface

Intended Audience

The primary intended audience is researchers and students in the social sciences
(particularly psychology, sociology, and political science) and in related profes-
sional domains (such as education). Readers should have some prior knowledge of t
tests and ANOVA. It is preferable for readers to also have some prior knowledge of
linear regression, although bright students can also pick up the basics from reading
Chapter 9.
This book is fairly unique in covering both nonparametric statistics and CDA
in one volume. With the exception of contingency tables, most textbooks address
either one or the other. In the one-semester course that I teach on this topic, I suc-
cessfully cover both and—as noted above—provide a framework for choosing the
best statistical technique. Using the framework requires knowledge of both CDA
and nonparametrics. Although it is a challenging course, given students’ financial
and time constraints, it is often not practical for graduate students who are not spe-
cializing in statistics to take separate courses in nonparametric statistics and CDA.
Instructors who can provide a two-semester course can of course cover more topics
and provide students with more extensive practice; however, a one-semester course
can still provide students with some familiarity with the topics and a framework for
choosing a statistical methodology for particular research projects. (Students can
then get more practice and mastery of that particular technique, as the best way to
learn a technique is to use it in the context of applied research.) The book could also
be used for a course in only CDA or nonparametrics by just focusing on the appli-
cable chapters, leaving the rest of the material for optional, supplementary reading.
Finally, the book is also suitable for researchers and graduate students who are not
necessarily enrolled in a course but who desire some knowledge of these alternative
techniques and approaches for enhancing statistical power.

Unique Features of the Book

The book is closely tied to those techniques currently available in SPSS and provides
direction on using SPSS. (Incorporation into SPSS was in fact a major criterion for
choosing what topics would be covered.) Although the book can be used with other
statistical packages, SPSS is widely used and there is a need for a text that addresses
that audience. Some of the examples and problems in the book also use Excel as a
pedagogical tool for building conceptual understanding. Also available is a website
for the book at www.Routledge.com/9781138787827 that contains selected data
sets for the book’s examples and problems and, for the instructor, PowerPoint slides
for each chapter, answers to even-numbered problems, and the author’s course
syllabus.
 Preface xvii

An important feature of the text is its conceptual focus. Simple computer simu-
lations and the inclusion of Exploring the Concept boxes are used to help attach
meaning to statistical formulas. Most homework problems were inspired by actual
research studies; these problems therefore provide authentic and meaningful illus-
trations of the techniques presented. Mathematical derivations have been kept to
a minimum in the main text but are available in appendices at the end of each
chapter.

Content

The book is structured as follows. Chapters 1 through 3 cover basic concepts in

probability—especially the binomial formula—that are foundational to the rest of
the book. Chapters 4 and 5 address the analysis of contingency tables (i.e., analyz-
ing the relationship between two nominal variables). Chapters 6 through 8 address
nonparametric tests involving at least one ordinal variable. (Chapter 8 presents
contemporary techniques for testing nonparametric interaction effects, a topic
omitted from many other texts.) The book then turns to situations that involve at
least one metric variable. Chapter 9 reviews some concepts from linear regression,
such as exponential growth and dummy variables, as well as the concept of general-
ized linear models. All of these concepts are foundational to CDA, which is the
focus of the remaining portion of the book. Chapters 10 and 11 cover various types
of logistic, ordinal, and Poisson regression. Chapter 12 overviews log-linear models,
and Chapter 13 presents the general estimating equations (GEE) methodology for
measuring outcomes measured at multiple time points. Chapter 14 covers estima-
tion methods, such as Newton-Raphson and Fisher scoring, for readers desiring a
deeper understanding of how the various CDA techniques work. The chapter pro-
vides preparation for reading more advanced statistical texts and articles. Finally,
Chapter 15 summarizes the various factors that need to be taken into consideration
when choosing the best statistical technique.
Overall, the book’s organization is intended to take the reader on a step-by-step
journey from basic statistical concepts into more advanced terrain.

Acknowledgments

I am truly thankful to all those who assisted me in reviewing chapter drafts. Special
thanks to Lori Griswold for reviewing the entire book for clarity and typographical
mistakes. Sections of the book were also reviewed by my graduate assistants: Ivan
Ivanovich, Jason Boggs, and Marissa Owens. I would also like to thank my editor,
Debra Riegert, for her assistance, and Mark Beasley and Hossein Mansouri, both
of whom gave valuable input on Chapter 8. I would also like to acknowledge the
xviii Preface

following reviewers who were solicited by the publisher: Randall H. Rieger, West
Chester University; Margaret Ross, Auburn University; Sara Tomek, University of
Alabama; Haiyan Wang, Kansas State University; and three anonymous reviewers.
Finally, the book would not be possible without all the encouragement and feed-
back provided by the graduate students who completed my course in categorical/
nonparametric data analysis.
 ABOUT THE AUTHOR

Dr. E. Michael Nussbaum is a professor of educational psychology at the University

of Nevada, Las Vegas. He holds a PhD from Stanford University and a MPP from
the University of California, Berkeley. He is the author of numerous research pub-
lications and serves on the editorial boards of the Journal of Educational Psychology,
Contemporary Educational Psychology, and the Educational Psychologist.
CATEGORICAL AND NONPARAMETRIC
DATA ANALYSIS
 CHAPTER 1

LEVELS OF MEASUREMENT, PROBABILITY,

AND THE BINOMIAL FORMULA

Categorical and nonparametric data analysis is designed for use with nominal or
ordinal data, or for metric data in some situations. This chapter therefore reviews
these different levels of measurement before turning to the topic of probability.

Levels of Measurement

In statistics, there are four basic types of variables: (a) nominal, (b) ordinal, (c) inter-
val, and (d) ratio.
A nominal variable relates to the presence or absence of some characteristic.
For example, an individual will be either male or female. Gender is a dichotomous
nominal variable. In contrast, with a multinomial nominal variable, cases are classi-
fied in one of several categories. For example, ethnicity is multinomial: Individuals
can be classified as Caucasian, African American, Asian/Pacific Islander, Latino, or
Other. There is not any particular ordering to these categories.
With an ordinal variable, there is an ordering. For example, an art teacher might
look at student drawings and rank them from the most to the least creative. These
rankings comprise an ordinal variable. Ordinal variables sometimes take the form
of ordered categories, for example, highly creative, somewhat creative, and uncreative.
A number of individuals may fall into these categories, so that all the drawings clas-
sified as highly creative would technically be tied with one another (the same for
the moderately creative and uncreative categories). With ranks, on the other hand,
there are typically few if any ties.
With an interval or ratio variable, a characteristic is measured on a scale with
equal intervals. A good example is height. The scale is provided by a ruler, which
may be marked off in inches. Each inch on the ruler represents the same distance;
as a result, the difference between 8 and 10 inches is the same as between 1 and
3 inches. This is not the case with an ordinal variable. If Drawing A is ranked as
2 Measurement, Probability, and Binomial Formula

more creative than Drawing B, we do not know if Drawing A is just slightly more
creative or significantly more creative; in fact, the distances are technically unde-
fined. As a result, we need to use different statistics and mathematical manipula-
tions for ordinal variables than we do for interval/ratio variables. (Much of this
book will be devoted to this topic.)
As for the difference between an interval variable and a ratio variable, the
defining difference is that in a ratio variable a score of zero indicates the complete
absence of something. Height is a ratio variable because zero height indicates that
an object has no height and is completely flat (existing in only two dimensions).
Counts of objects are also ratio variables. The number of people in a classroom can
range from zero on up, but there cannot be a negative number of people. With an
interval variable, on the other hand, there can be negative values. Temperature
is a good example of something measured by an interval scale, since 0o Celsius is
just the freezing point of water, and negative temperatures are possible. However,
for the tests discussed in this book, it will usually not be necessary to differentiate
between ratio and interval variables, so we will lump them together into one level
of measurement. The distinction will only become important when we consider the
analysis of count data with Poisson regression. For ease of exposition, in this book
I will use the term metric variable to refer to those at the interval or ratio levels of
measurement.
Metric variables are often also referred to as continuous, but this usage fails to
recognize that some metric variables are discrete. For example, a count cannot have
fractional values; for example, it would be incorrect to say that there are 30.5 people
enrolled in a class.
Figure 1.1 shows the three levels of measurement. The metric level is shown
on top because it is the most informative. Metric data can always be reduced to
ordinal data by using the numerical values to rank the data (e.g., ranking people
from the tallest to the shortest based on their heights). Likewise, ordinal data can
be reduced to nominal data by performing a median split and classifying cases as
above or below the median. Transforming data from a higher level to a lower level
is known as data reduction. Data reduction throws away information; for example,
knowing that Marie is taller than Jennifer does not tell one how much taller Marie
is. Nevertheless, data reduction is sometimes performed if the assumptions of a
statistical test designed for metric data are not met. One might then reduce the
data to ordinal and perform a statistical test that is designed for ordinal data. One
can also reduce ordinal (or metric) data to nominal. One cannot move from a
lower level to a higher level in the figure because that requires information that
is missing.
Categorical and nonparametric statistics is concerned with statistical methods
designed for ordinal- and nominal-level data. However, these methods are often
used with metric data when sample sizes are small (and therefore some of the
assumptions of t tests, ANOVA, and linear regression are not met) or when the data
 Measurement, Probability, and Binomial Formula 3

Ratio/Interval (Metric)

Ordinal

Nominal

Figure 1.1 Three levels of measurement. The figure shows that metric data can be
reduced to ordinal data, which can in turn be reduced to nominal data. Metric data are
the most informative because they carry information on how different the cases are on
a variable in quantitative terms. Nominal data are the least informative because they
contain no information regarding order or ranks.

are skewed or otherwise highly abnormal. In the latter cases, standard methods may
not be as statistically powerful as categorical and nonparametric ones. In reading
this book, it is very important to remember the definition of statistical power:

Statistical power refers to the ability to reject the null hypothesis and find a
“result.” (To be more technically precise, it is the probability that one will reject
the null hypothesis when the alternative hypothesis is true.)

Because conducting a study requires a large amount of work, one usually wants
to use the most powerful statistical methods. (Obtaining a p value of .06 or .07 is
not sufficient to reject the null hypothesis and therefore can be very disappointing
to researchers.) That is why, in planning a study, one wants to use the most valid
and statistically powerful methods one can.

Probability

All statistical methods—including categorical/nonparametric ones—require an

understanding of probability. In the remainder of this chapter, I review the basic
axioms of probability and use them to derive the binomial formula, which is the
foundation of many of the tests discussed in this book.
In this section, I use a canonical example of tossing coins up in the air and ask-
ing questions about the probability of a certain number of them coming up heads.
Note that whether a coin comes up heads or tails is a nominal outcome (it either
happens or it doesn’t)—which is why understanding probability is essential to cat-
egorical data analysis.
4 Measurement, Probability, and Binomial Formula

The Meaning of Probability

If one flips a coin, what is the probability that the coin will come up heads? You
may reply “one-half,” but what exactly does this statement mean?
One definition of probability is given by the following equation:

Number of favorable possibilities

Probability = , (Eq. 1.1)
Number of total possibilities

assuming that all possibilities are equally likely and mutually exclusive.
So a probability of ½ means that there is one favorable possibility (heads) out
of two (heads or tails). However, this example assumes that the coin is fair and not
biased, meaning that the possibilities are equally likely. A biased coin might have a
little more metal on one side so that heads results 60% of the time. Such coins have
been created to cheat at games of chance.
Eq. 1.1 is often useful but because of the restrictive equal-probability assump-
tion, a more general definition of probability is needed. The probability of some
event A occurring is the proportion of time that A will occur (as opposed to not-A)
in the limit as n approaches infinity, that is:

f ( A)
Pr ob ( A) = lim ( n  ∞) , (Eq. 1.2)
n

where f(A) is the frequency of A. Thus, the meaning of the statement “The probabil-
ity that the coin will come up heads is one-half” is that over a large number of flips,
about half the time the coin will come up heads. The amount of error decreases as
n (the number of flips) increases, so that the proportion will approach Prob(A) as n
approaches infinity. Now to assess the probability, one might flip the coin 1,000 times
and gauge the relative proportion that the coin comes up heads as opposed to tails.
This procedure will only give one an estimate of the true probability, but it will give
one a pretty good idea as to whether the coin is fair or biased. Basically, what we are
doing is taking a random sample of all the possible flips that could occur.
The definition in Eq. 1.2 reflects a frequentist view of probability. Technically,
Eq. 1.2 assigns probabilities only to general statements, not to particular facts or
events. For example, the statement that 80% of Swedes are Lutherans is a meaning-
ful probability statement because it is a generalization; but the statement that “the
probability that John is Lutheran, given that he is Swedish, is 80%” is not mean-
ingful under this definition of probability, because probability applies to relative
frequencies, not to unique events.
This position, however, is extreme, given that we apply probability to unique
events all the time in ordinary discourse (maybe not about Lutherans, but cer-
tainly about the weather, or horse racing, etc.). In my view, probability statements
can be meaningfully applied to particular cases if one makes the appropriate back-
ground assumptions. For example, if one randomly selects one individual out of the
 Measurement, Probability, and Binomial Formula 5

population (of Swedes), then one can meaningfully say that there is an 80% chance
that she or he is Lutheran. The background assumption here is that the selection
process is truly random (it may or may not be). The existence of the background
assumptions means that probability values cannot be truly objective because they
depend on whether one believes the background assumptions. Nevertheless, these
assumptions are often rational to make in many situations, so in this book I will
assign probability values to unique events. (For further discussion of subjectivist,
Bayesian notions of probability, which view probability statements as measures of
certainty in beliefs, see Nussbaum, 2011).

Probability Rules
Probability of Joint Events
What is the probability that if one flips two coins, both will come up heads? Using
Eq. 1.1, it is ¼, because there is one favorable possibility out of four, as shown below.

H H
H T
T H
T T

Another way of calculating the joint probability would be to use the following
formula:

Prob(A & B) = Prob(A) * Prob(B) [if Prob(A) and Prob(B)

are statistically independent]. (Eq. 1.3)

Here, A represents the first coin coming up heads and B represents the second coin
coming up heads; the joint probability is ½ * ½ = ¼. The formula works because
A represents one-half of all possibilities and of these, one half represent favorable
possibilities, where B also comes up heads. One-half of one-half is, mathematically,
the same as ½ * ½.
An important background assumption, and one that we shall return to repeat-
edly in this book, is that A and B are statistically independent. What this means
is that A occurring in no way influences the probability that B will occur. This
assumption is typically a reasonable one, but we could imagine a scenario where it
is violated. For example, suppose someone designs a coin with an electrical trans-
mitter, so that if the first coin comes up heads, this information will be transmitted
to the second coin. There is a device in the second coin that will tilt it so that it
will always comes up heads if the first coin does. In other words: Prob(B | A) = 1.
This statement means that the probability of B occurring, if A occurs, is certain. We
6 Measurement, Probability, and Binomial Formula

will also assume the converse: Prob(not-B | not-A) = 1. There are therefore only two
total possibilities:

H H
T T

The joint probability of two heads is therefore one-half. The more general rule for
joint probabilities (regardless of whether or not two events are statistically inde-
pendent) is:

Prob(A & B) = Prob(A) * Prob(B | A). (Eq. 1.4)

The joint probability in the previous example is ½ * 1 = ½.

EXPLORING THE CONCEPT

If two events (A and B) are statistically independent, then Prob(A) = Prob(A | B) and
Prob(B) = Prob(B | A). Can you use this fact to derive Eq. 1.3 from Eq. 1.4?

Note that in the rigged coin example, the joint probability has increased from
¼ (under statistical independence) to ½ (under complete dependence). This out-
come reflects a more general principle that the probability of more extreme events
increases if cases are not statistically independent (and positively correlated). For
example, if I throw 10 coins up in the air, the probability that they will all come
up heads is

½ * ½ * ½ * ½ * ½ * ½ * ½ * ½ * ½ * ½ = (½)10 = .001,

or 1 in 1,000. But if the coins are programmed to all come up heads if the first
coin comes up heads, then the joint probability is again just one-half. If the first
coin coming up heads only creates a general tendency for the other coins to come
up heads [say Prob(B | A) = 0.8], then the joint probability of two coins coming up
heads would be 0.5 * 0.8 = 0.4 (which is greater than ¼ th, assuming independence).
If 10 coins are flipped, the probability is 0.5 * (0.8)9 = .067. This probability is still
far greater than the one under statistical independence.
A violation of statistical independence is a serious problem, as it is a violation
of the first axiom of statistical theory. For example, my own area of research is on
how students construct and critique arguments during small-group discussions. In
small-group settings, students influence one another, so, for example, if one stu-
dent makes a counterargument, it becomes more likely that other students will do
so as well, due to modeling effects and other factors. The probability that Student
 Measurement, Probability, and Binomial Formula 7

(A) makes a counterargument is therefore not statistically independent from the

probability that Student (B) will, if they are in the same discussion group. Analyzing
whether some intervention, such as the use of a graphic organizer, increases the
number of counterarguments, without adjusting for the lack of statistical inde-
pendence, will make the occurrence of Type I errors much more likely. That is
because the probability of extreme events (i.e., lots of students making counterar-
guments) goes way up when there is statistical dependence, so p values are likely to
be seriously inflated. There are statistical techniques, such as multilevel modeling,
that address the problem, but the technique’s appropriateness in argumentation
research is still being debated (given that certain sample sizes may be required).
Further discussion of multilevel modeling is beyond the scope of this book, but the
example illustrates why an understanding of basic probability theory is important.

EXPLORING THE CONCEPT

(a) Suppose one randomly selects 500 households and sends a survey on political
attitudes toward female politicians to all the adult males in each household, and
sends the same survey to all the adult females (e.g., one to a husband and one
to a wife). Would the individual responses be statistically dependent? How might
this fact affect the results? (b) How might sampling without replacement involve a
violation of statistical independence? For example, what is the probability that if I
shuffle a complete deck of 52 cards, the first two cards will be spades? Use Eq. 1.4
to calculate this probability.

Probabilities of Alternative Events

What is the probability that if I select one card from a full deck of 52 cards the card
will be spades or clubs? This question relates to alternative events occurring (A or B)
rather than joint events (A and B). The probability of alternative events is given by

The Prob(A or B) = Prob(A) + Prob(B) [if Prob(A) and

Prob(B) are mutually exclusive]. (Eq. 1.5)

Applying Eq. 1.5 to our question, we find that the Prob(spades or clubs) =
Prob(spades) + Prob(clubs) = 14 + 14 = 21 . This result makes sense because half the
cards will be black. The reason Eq. 1.5 works is because Event A (card being a spade)
represents ¼ of the favorable possibilities, and Event B (card being a club) repre-
sents another ¼ of the favorable possibilities—so together, the probability that the
selected card will be black is the sum of all the favorable possibilities (which is why
we add). The background assumption is that the card cannot be both spades and
8 Measurement, Probability, and Binomial Formula

clubs, that is, A and B cannot both occur—the events are mutually exclusive. If the
events are not mutually exclusive, one should use Eq. 1.6:

Prob(A or B) = Prob(A) + Prob(B) - Prob(A & B). (Eq. 1.6)

For example, if one flips two coins, the probability that one or the other (or
both) will come up heads is 21 + 21 − 14 = 43 , which is the correct answer. If one were
to erroneously use Eq. 1.5 to calculate the probability, one would double count the
possibility where both coins come up heads.

Summary
In summary, remember that if joint events are involved (involving ands), one should
multiply (assuming statistical independence), and where alternative events are
involved (involving ors), one should add (assuming the events are mutually exclusive).

Conditional Probabilities
The last type of probability to be considered is conditional probability. These take an
IF . . . THEN form; for example, if a student is male, then what is the probability that
the student will complete high school? Condition probabilities will be discussed in
Chapter 4.

Some Probability Thought Experiments

We shall now progressively consider cases of increasing complexity.

Probability Distribution of the Two-Coin Example

In tossing two coins there are four possible outcomes.

Number of Heads (H)

(1) H H 2
(2) H T 1
(3) T H 1
(4) T T 0

Each of the four outcomes is called a permutation. If one counts the number of
hits (i.e., heads), outcomes (2) and (3) both reflect one head. Although these two
outcomes reflect different permutations, they reflect the same combination, that is, a
 Measurement, Probability, and Binomial Formula 9

0.6

0.5

0.4
Probablity

0.3

0.2

0.1

0
1 2 3
Hits

Figure 1.2 Probability histogram for the two-coin problem.

combination of one head and one tail. Combinations reflect the number of heads,
but whether the head comes up on the first trial or the second does not matter. The
order does matter with permutations.
Let us now make a histogram (frequency graph) of the different combinations
(see Figure 1.2).
This figure is shape like an inverted U, except that it is not smooth. Many prob-
ability distributions—like the normal curve—have this upward concave characteris-
tic. This is because there are more permutations corresponding to the combinations
in the middle than at the extremes, making the more central possibilities “more
likely.”

One Hundred–Coin Example

Let us take a more extreme case. Suppose one throws 100 coins in the air. This
resulting histogram is shown in Figure 1.3. (We have not yet covered the formula
that was used in making this figure.)
The most extreme positive case is that all the coins will come up heads. This
corresponds to only one permutation. The probability of each head is ½, and these
are independent events, so the probability of their joint occurrence, as given by the
multiplication rule, is .5100. Not surprisingly, this number is extremely small:

.000000000000000000000000000788861.

On the other hand, the probability of obtaining 50 heads out of 100 is much larger
(8%).
10 Measurement, Probability, and Binomial Formula

0.09
0.08
0.07
0.06
Probability

0.05
0.04
0.03
0.02
0.01
0
0
5
10
15
20
25
30
35
40
45
50
55
60
65
70
75
80
85
90
95
100
Hits

Figure 1.3 Probability histogram for the 100-coin problem.

This is because there are a lot of ways of getting a combination of 50 coins out of
100 (i.e., there are a lot of corresponding permutations). The first 50 coins could come
up heads and the second 50 could come up tails; the first 50 coins could come up tails
and the second 50 could come up heads; every other coin could come up heads, and
so forth. There are also a lot more “random-looking” permutations, such as

HHHTTTHTTH...,

which is a random sequence that I generated in EXCEL. In fact, the number of permu-
tations comes out to be 1.00891 * 1029. How this number is derived will be explained
later in the chapter. Notice that in comparing Figures 1.2 and 1.3, the one with 100
coins is skinnier, if one thinks of the extremes in percentage terms (100 heads is 100%
hits, tails is 0% hits). This result occurs because the more extreme outcomes become
less likely as the number of trials increases. For example, the probability of 100% hits
in the two-coin case is 25%, not .5100 (as calculated above). Later we will see that this
is the reason why larger samples provide more precise estimates.

The Binomial Formula

The binomial formula is used to measure the probability of a certain number of hits in
a series of yes/no trials. It is a complicated formula, so I will introduce it bit by bit. In
my example, I will use a biased rather than a fair coin to make the mathematics easier
to follow. The Greek letter pi (π) denotes the probability of a hit. (Notice that the
words pi and probability both start with the letter p. Statisticians often choose Greek
letter for concepts that, in English, start with the same letter.) Let us suppose that the
probability of this particular biased coin coming up heads is 70% (π = 70%), and ask:
What is the probability of tossing 10 coins (with π = .7) and obtaining 8 heads?
 Measurement, Probability, and Binomial Formula 11

This question can be broken down into two subparts:

1. What is the probability of a permutation involving 8 heads out of 10 coins?

2. How many permutations make up a combination of 8 heads out of 10 coins?

Subquestion 1: What is the Probability of One “Permutation”?

Let us consider one permutation, where the first 8 coins come up heads: H H H H H
H H H T T. Assuming the tosses are independent, the joint probability is 70% * 70% *
70% * 70% * 70% * 70% * 70% * 70% * 30% * 30% = .78 * .32 = π8 * (1 - π)2. If we
define k as the number of hits (in this case k = 8), and n as the number of trials (in
this case n = 10), then the calculation becomes

πk * (1 - π)n-k. (Eq. 1.7)

For the permutation considered previously, the probability is .78 * .32 = 0.5%.
Note that all the permutations associated with 8 coins out of 10 coming up heads
are equally likely. For example, the permutation “H H H T H H H H H T” has the
probability 70% * 70% * 70% * 30% * 70% * 70% * 70% * 70% * 30% = .78 * .32 =
0.5%. The only thing that has changed is the order in which the coins come up
heads or tails.
Remember that our overall goal is to find the probability of a combination of
8 hits out of 10. We shall see that this combination is associated with 45 different
permutations, including the two shown above. The probability of one or another
of these 45 different permutations occurring can be calculated by adding the indi-
vidual probabilities together. Because the permutations are mutually exclusive, we
can use the addition formula (Eq. 1.5). Because the permutations are equally likely,
the calculation reduces to

45 * πk *(1 - π)n-k. (Eq. 1.8)

The underlined portion is the probability of one of the permutations occurring

(from Eq. 1.7). What I have yet to address is how the 45 is calculated.

Subquestion 2: How Many Permutations Make up a Combination?

How many possible permutations are associated with a combination of 8 hits out
of 10? The applicable formula, which is derived conceptually in Appendix 1.1, is

n n!
  = . (Eq. 1.9)
 k  k !(n − k )!
12 Measurement, Probability, and Binomial Formula

Here, ( nk ) is the notation for k hits out of n trials. On the right-hand side of the equa-
tion, the exclamation mark signifies a factorial product (e.g., 4! = 4 * 3 * 2 * 1).
Substituting the values from our example yields

10 ! 3, 628, 000
= = 45.
8 !(10 − 8)! 40, 320 ∗ 2

Putting Eqs. 1.9 and 1.7 together yields the binomial formula, which gives the
probability of a certain number of k hits out of n trials:

n!
Prob (k hitsoutof n trials) = ∗ π k ∗ (1 − π )n−k . (Eq. 1.10)
k !( n − k ) !

In our example, the terms are 45 * 0.5% = 23%. The first term represents the
number of permutations associated with a combination of 8 heads out of 10 coin
tosses, and the second term reflects the probability of each permutation.
The calculation can also be performed in EXCEL with the following command:

= BINOMDIST(k, n, π, 0),

with the 0 indicating a noncumulative calculation (explained below). So

= BINOMDIST(8, 10, 0.7, 0)  23%. There is almost a one-quarter chance of obtain-
ing exactly 8 heads with this particular biased coin.

EXPLORING THE CONCEPT

Concepts can often be better understood when applied to very simple examples.
Use the binomial formula to find the probability of obtaining exactly one head in a
toss of two fair coins.

Cumulative Versus Noncumulative Probabilities

Now suppose we wanted to calculate the chance of obtaining, out a toss of 10 coins,
8 heads or less. This probability could be derived by calculating the chance of obtain-
ing exactly 8 heads or 7 heads or 6 heads or . . . 1 head or 0 heads. These are mutu-
ally exclusive outcomes, so according to Eq. 1.5 we can just add the probabilities:
23% + 27% + 20% + 10% + 4% + 1% + 0% + 0% + 0% = 85%. (The 0% values are
not perfectly equal to zero but are so small that these round to zero.) I calculated
the individual probabilities using the EXCEL command “=BINOMDIST(k, n, π, 0)”,
but an easier way is to calculate the cumulative probability by changing the zero in
 Measurement, Probability, and Binomial Formula 13

the expression to one. (A one in the last place of the EXCEL command tells EXCEL
to use the cumulative probabilities.) So we have

= BINOMDIST(8, 10, 0.7, 1)  85%.

Now I am going to change the problem a little. The question is, “What is the
probability of obtaining 9 heads or more?” (again, out of a toss of 10 coins, with
π = 70%). This is calculated by subtracting the 85% from 100% (= 15%). Note that
the probabilities of obtaining (a) 8 coins or less and (b) 9 coins or more must sum
to 100%. This is because the two possibilities are exhaustive of all possibilities and
mutually exclusive; one or the other must happen.
Now, instead of asking, “What is the probability of obtaining 9 hits or more?”
we ask, “What is the probability of obtaining 8 hits or more?” This value is 100%
minus the probability of obtaining 7 hits or less. The EXCEL syntax is

1 - BINOMDIST(7, 10, 0.7, 1)  38%.

Some students err by putting an 8 rather than a 7 in the formula, because the
question asks about the probability of getting 8 heads or more, yielding an incor-
rect answer of 15%. This approach is incorrect because it is tantamount to calculat-
ing the probability of 8 heads or more plus 8 heads or less. These are not mutually
exclusive possibilities, because if one obtains exactly 8 heads, both scenarios occur.
According to the formula for alternative probabilities, one has to subtract out the
probability of getting exactly 8 heads so as to not double count it (see Eq. 1.6). We
previously calculated the probability of obtaining eight heads or less at 85% and for
eight heads or more at 38%. The two possibilities sum to 123%, which is not a legal
probability. We have double counted the probability of obtaining exactly 8 heads
(which we calculated above at 23%).
This fact is so important that one should spend a minute now making a men-
tal rule to remember that when you are asked cumulative probability problems
that involve X or more (rather than X or less), you need to use X - 1 in the EXCEL
formula. EXCEL will only give you cumulative probabilities for X or less problems.

The Bernoulli and Binomial Distributions

A probability distribution gives the probability of each and every mutually exclu-
sive outcome of an event. The simplest probability distribution is the Bernoulli
distribution, named after the 18th-century mathematician, Daniel Bernoulli. It
is the probability distribution associated with a single, discrete event occurring.
Table 1.1 presents the Bernoulli distribution for the situation where the probability
of a biased coin coming up heads is 70%.
14 Measurement, Probability, and Binomial Formula

Table 1.1 A Bernoulli Distribution

Heads or Tails Probability

Heads (X = 1) 70%
Tails (X = 0) 30%
Sum 100%

0.8
0.7
0.6
0.5
Probability

0.4
0.3
0.2
0.1
0
0 1
Hits

Figure 1.4 Bernoulli distribution for π = 0.7.

Probability distributions always sum to 100%. Because the different outcomes

listed are exhaustive, one of them must happen, so the probability that one or
another will happen is found by just adding all the probabilities up (the possibilities
are mutually exclusive).
Charting the above Bernoulli distribution yields Figure 1.4. With a Bernoulli
distribution, we just toss one coin. With the binomial distribution, we toss more
than one coin. So my previous examples using 2, 10, or 100 coins are all associated
with binomial distributions. With the binomial distribution, the tosses of the coins
must be independent. The formal definition of the binomial distribution is “the
distribution resulting from an independent series of Bernoulli trials.”
We shall refer to this definition in later chapters, but for now it is just a fancy
way of saying that there is more than one nominal event (e.g., we toss up more
than one coin).
The shape of a binomial distribution will depend on the value of π (probability
of a hit). A binomial distribution can be generated in EXCEL using the binomial
formula, with the EXCEL syntax =BINOMDIST(k, n, π, 0).
Figure 1.5 (Panel A) shows a binomial distribution when using a fair coin
(π = 50%) for different values of k with n = 10 (in other words, when we toss 10 fair
coins up in the air). The distribution is symmetric. However, with our biased coin
(π = 70%), the distribution is skewed.
 Measurement, Probability, and Binomial Formula 15

(A)
0.3

0.25

0.2
Probability

0.15

0.1

0.05

0
0 1 2 3 4 5 6 7 8 9 10
Heads

(B)

0.3

0.25

0.2
Probability

0.15

0.1

0.05

0
0 1 2 3 4 5 6 7 8 9 10
Heads

Figure 1.5 Binomial distribution (n = 10) for different probabilities of a hit. (A) π =
50%. (B) π = 70%.

Now we are going to make something interesting and important happen. As

the number of trials increase, the distribution becomes more and more symmetric.
Figure 1.6 shows that with n = 100 (Panel B), the distribution is extremely sym-
metric. (The distribution is in fact normal except for the fact that the variable
“number of hits” is discrete rather than continuous.) Even at n = 20 (Panel A) the
distribution is somewhat symmetric. Remember that as n increases, the probabil-
ity of extreme outcomes decreases. So does the probability of somewhat extreme
values, although not at the same rate. So in Figure 1.6 (with n = 20), consider the
probability of moderately extreme values (such as k = 8, that is 40% heads), a value
which contributes to the skew in Figure 1.5 [specifically, see Panel B at Prob(k = 4)
16 Measurement, Probability, and Binomial Formula

(A)
0.25

0.2
Probability

0.15

0.1

0.05

0
0 2 4 6 8 10 12 14 16 18 20
Heads

(B)
0.1
0.09
0.08
0.07
Probability

0.06
0.05
0.04
0.03
0.02
0.01
0
0
5
10
15
20
25
30
35
40
45
50
55
60
65
70
75
80
85
90
95
100

Heads

Figure 1.6 Binomial distribution for π = 70% for different values of n. (A) n = 20.
(B) n = 100.

= 40%]. The probability of 40% hits is becoming less likely with the larger sample,
making the distribution more symmetric.

The Normal Approximation of the Binomial Distribution

Normal Distribution
A normal distribution is also known as a Gaussian distribution, and when stan-
dardized, a z distribution. The normal distribution has the property that about
two-thirds of the distribution falls within one standard deviation (SD) of the
mean, and most of the distribution falls within two standard deviations of
the mean.
 Measurement, Probability, and Binomial Formula 17

To standardize a score means to express the score in terms of the number of

standard deviations from the mean. When a variable is normally distributed, use
the following formula to compute z scores:

X −mean
z = . (Eq. 1.11)
SD

We shall use this equation repeatedly in this book.

One can compute the cumulative probability of different z scores using the
EXCEL command: =NORMSDIST(z). So, for example, if a variable is normally dis-
tributed, the cumulative probability of obtaining a value of, say, 0.23, is 59%. With
EXCEL syntax, =NORMSDIST(0.23)  59%.
It is frequently the case in statistics that we need to build a 95% confidence
interval with 2.5% of the z distribution in each of the tails of the distribution. A
95% confidence interval around the mean can be built if we find the points that
are about two standard deviations from the mean (1.96 SDs, to be precise). These
would correspond to z scores of 1.96 and -1.96. Using EXCEL, one can verify that
=NORMSDIST(-1.96)  0.025, and =NORMSDIST(+1.96)  0.975.

EXPLORING THE CONCEPT

Some students wrongly expect that there should be a 95% chance of obtaining a
z score of +1.96 because we are building a 95% confidence interval. Why is the
probability of obtaining a z score of +1.96 or less 97.5% rather than 95%?

Suppose that X is normally distributed, has a mean of 3 and a standard deviation

of 2. What is the probability of obtaining a raw score of 6? One could first compute
the z score (1.5) and then consult EXCEL for the probability: =NORMSDIST(1.5) 
93.3%. One could also use the unstandardized normal distribution and just enter the
mean and standard deviation into EXCEL using the command =NORMDIST (this
one does not have an “S” after “NORM”). Specifically, enter =NORMDIST(X, mean,
SD, 1); the 1 requests a cumulative distribution. In the example, =NORMDIST(6, 3,
2, 1)  93.3%. Which command to use is one of personal preference.

The Normal Distribution Is Continuous, not Discrete

A discrete distribution is one where X cannot take on fractional values, whereas a
continuous variable can take on any fractional value. A nominal (0, 1) variable is
always discrete, whereas metric variables can be either continuous or discrete. For
example, the probability distribution for the variable in Table 1.2 is discrete. Here,
18 Measurement, Probability, and Binomial Formula

Table 1.2 A Discrete Probability Distribution

X Probability

1 15%
2 33%
3 7%
4 12%
5 33%
Total 100%

0.35

0.3

0.25
Probability

0.2

0.15

0.1

0.05

0
1 2 3 4 5
X

Figure 1.7 Example probability distribution for a discrete variable.

X only takes on the discrete values of 1, 2, 3, 4, and 5. On the other hand, if X could
take on values such as 1.3, 2.67, or 4.3332, it would be continuous.
The probability distribution for the above discrete variable is displayed graphi-
cally in Figure 1.7. Ignoring the spaces between the bars, note that the area under
the curve sums to 1.0. The area for the value X = 2.0 is given by the width of the bar
(1 unit) times the height (which is 0.33), so the area is 0.33. The individual subareas
sum to 1 because the probabilities sum to 1.
If X were continuous, the area for a value (such as 2.67, a number I arbitrarily
chose) is given by the height (2.67) times the width, except that the width would
be 0. This produces a paradox that the probability of any specific value of a continu-
ous variable is 0. Obviously, this situation cannot occur, as the variable will take
on specific values in particular cases. To resolve this paradox, we stipulate that for
continuous variables, one can only meaningfully talk about the values of cumula-
tive probabilities, for example, the probability of obtaining a value of 2.67 or less (or
the probability of obtaining a value of 2.67 or more). The probability of obtaining
exactly 2.67 is undefined.
 Measurement, Probability, and Binomial Formula 19

The normal distribution is a continuous distribution. Z scores are therefore

also continuous and the probability of obtaining a particular z score is cumulative.
Therefore, while it is meaningful to write such expressions as Prob(z ≤ 0.23), it is
not meaningful to write expressions such as Prob(z = 0.23).

EXPLORING THE CONCEPT

In thinking about the normal curve and calculating the probability of z scores, why
do we typically compute the area of the curve that is to the right (or left) of a certain
point?

The Normal Approximation (the 100-Coin Example)

Suppose our problem is to compute the probability that if one tosses 100 fair coins,
that 60 or less will come up heads. I present in this section two different methods
for approaching this problem: (a) with the binomial formula, and (b) with the
normal curve. The latter is more complicated with this particular problem, but the
approach will prove to be useful on problems considered in other chapters.

The Binomial Formula

To use the binomial formula, simply use the EXCEL syntax:

=BINOMDIST(k, n, π, 1) (use 1 because we want cumulative probabilities).

=BINOMDIST(60, 100, 0.5, 1)  98.2%.

The Normal Approximation of the Binomial

It can be shown that as n approaches infinity, the binomial distribution approaches
the normal distribution. When n is large, the differences between the distribu-
tions are negligible, so we speak about the normal approximation of the binomial.
However, the binomial distribution is discrete, whereas the normal distribution is
continuous, and this fact causes the approximation to be off a little, requiring what
is known as a continuity correction. Let us set the continuity correction aside for a
moment, as it will be the last step in our calculations.
We will first need to calculate a z score, and to do that we need to know the
mean and standard deviation of the binomial distribution. Table 1.3 shows the
mean and variances of both the Bernoulli and binomial distributions.
In our example, the mean is nπ (100 * 0.50 = 50). If we toss 100 coins in the
air, the most likely outcome is 50 heads (see Figure 1.3). Although the likelihood of
this occurring is only 8%, it is still the most likely outcome. It is certainly possible
20 Measurement, Probability, and Binomial Formula

Table 1.3 Means and Variances of the Bernoulli and Binomial Distributions

Parameter Distribution

Bernoulli Binomial

Mean π nπ
Variance π(1 - π) nπ(1 - π)

that we might get 49 or 53 coins coming up heads rather than exactly 50, but the
number of heads is still likely to be around 50. The mean is also known as the
expected value.
In our example, the variance is nπ(1 - π). This is sometimes written as nPQ,
with P representing the probability of a hit and Q representing the probability of a
miss. In statistics, by convention, Greek letters refer to population parameters and
English letters to sample estimates of these parameters. Because we are not dealing
with samples here, I use the Greek letters in writing the formulas.
To continue with the example, the variance is nπ(1 - π) = 100 * 50% * 50% = 25.
The standard deviation is the square root of the variance, or 5. The z score is then
z = X −SDmean = 60 − 50 = 2.00,
5 Prob(z ≤ 2.00) = 97.7%. This result does not jive
with the calculation using the binomial formula (98.2%) because we have not yet
applied the continuity correction. When using z scores, it is assumed that the val-
ues are continuous. But we are here using discrete raw score values, for example,
62, 61, 60, 59, 58, and so forth. We want to find the probability of getting a value
of X ≤ 60. The area of the probability distribution corresponding to 60 is assumed
with the normal-curve method to have a width of zero (because continuous vari-
ables are assumed), when in fact the bar has a width of 1. The area of the bar to the
right of the midpoint corresponds to the probability excluded from the calculation
with the normal-curve method. This method leaves out “half a bar” associated
with the binomial formula method, which gives the exact value. The point of the
continuity correction when using the normal approximation is basically to “add
half a bar” to the probability estimate. We can do this if we use the value of X =
60.5 rather than 60 in the z-score calculation:

X − mean 60.5 − 50
z= = = 2.10, Prob( z ≤ 2.10 ) = 98.2%.
SD 5

This result is equal to the result from using the binomial method.
The normal approximation is an important procedure for the material discussed
in this book. It may seem unnecessary here to go through all the gyrations when we
can calculate the exact probability with the binomial formula, but in many other
situations we will not have recourse to the binomial formula or some other “exact”
formula and therefore will need to use the normal approximation.
 Measurement, Probability, and Binomial Formula 21

Continuity Corrections With X or More Problems

One last problem: What is the probability of obtaining 60 or more heads when toss-
ing 100 fair coins? Once again, we should apply a continuity correction. However,
there is a second adjustment we will need to make, namely, the one described a
few pages back for “X or more” problems (specifically using X - 1 in the EXCEL
formula). Therefore, instead of using X = 60 in the formula, we should use X = 59.
But for the continuity correction, we also need to add a half a bar. Therefore, the
correct X to use is 59.5:

X − mean 59.5 − 50
z= = = 1.90, Prob( z ≤ 1.90 ) = 97.1%.
SD 5

Subtracting from 100%, the probability of obtaining 60 or more heads is 2.9%.

EXPLORING THE CONCEPT

The probability of obtaining 60 or more heads is 2.9%. The probability of obtaining
60 or less heads is 98.2%. These probabilities sum to more than 100%; they sum to
101.1%. Can you explain why?

Problems

1. What is the scale of measurement for the following random variables?

a) Someone’s weight (in pounds).
b) The state in which a person resides (Nevada, California, Wisconsin, etc.).
c) Someone’s IQ score.
2. What is the scale of measurement for the following random variables?
a) A student’s percentile ranking on a test.
b) Self-report of the number of close friends an adult has.
c) Political party affiliation.
d) A child’s categorization as normal, mildly emotionally disturbed, and
severely emotionally disturbed.
3. One throws two dice. What is the probability of
a) obtaining a 7 (combining the two numbers)? (Hint: Try to think of the
different permutations that would produce a 7, and use the multiplication
rule for joint events); and
b) obtaining a 9?
4. In a single throw of two dice, what is the probability that
a) two of the same kind will appear (a “doublet”)?
b) a doublet or a 6 will appear? (Hint: These are not mutually exclusive events.)
5. In three tosses of a fair coin, what is the probability of obtaining at least one head?
22 Measurement, Probability, and Binomial Formula

6. Three cards are drawn at random (with replacement) from a card deck with
52 cards. What is the probability that all three will be spades?
7. Three cards are drawn at random (without replacement) from a card deck with
52 cards. What is the probability that all three will be clubs? [Hint: When
two events are not statistically independent, and A occurs first, Prob(A & B) =
Prob(A) * Prob(B | A).]
4
8. Evaluate the following expression ( 1 ). [Hint: Remember that ( nk ) = n ! .]
k !( n−k !)
9. Evaluate the following expression: ( 34 ) .
10. You toss four coins. Using the binomial coefficient formula, compute how many
permutations are associated with a combination of two heads and two tails.
11. You toss four coins. Using the binomial formula, compute the probability of
obtaining exactly two heads if the coin is fair. [Hint: The probability is given
n k ( n−k )
by ( k ) ∗ π (1 − π ) .]
12. In a toss of four coins, use the binomial formula to compute the probability of
obtaining exactly two heads if the coin is biased and the probability of a head
is 75%.
13. Repeat the previous problem using EXCEL.
14. Using EXCEL, find the probability of obtaining two heads or less (again, with
n = 4 and π = 75%).
15. Using EXCEL, find the probability of obtaining two heads or more (again, with
n = 4 and π = 75%).
16. Chart the frequency distribution when n = 4 and π = 75%.
17. You construct a science achievement test consisting of 100 multiple-choice
questions. Each question has four alternatives, so the probability of obtaining a
correct answer, based on guessing alone, is 25%. If a student randomly guesses
on each item, what is the probability of obtaining a score of 30 or more correct?
a) Use the binomial function in EXCEL. (Hint: Find the probability of 29 or
less, and then subtract from 1.)
b) Use the normal approximation method, without a continuity correction.
You will need to calculate a z score, which means you will need to calculate
the mean expected correct (nπ) and standard deviation of this nπ(1- π ) .
(Hint: Find the probability of 29 or less, and then subtract from 1.)
c) Use the normal approximation with a continuity correction. (Hint: Find
the probability of 29.5 or less, then subtract from 1.)
18. If Y is a binomial random variable with parameters n = 60 and p = 0.5, esti-
mate the probability that Y will equal or exceed 45, using the normal approxi-
mation with a continuity correction.
19. Let X be the number of people who respond to an online survey of attitudes
toward social media. Assume the survey is sent to 500 people, and each has
a probability of 0.40 of responding. You would like at least 250 people to
respond to the survey. Estimate Prob(X ≥ 250), using the normal approxima-
tion with a continuity correction.
 Measurement, Probability, and Binomial Formula 23

APPENDIX 1.1

LOGIC BEHIND COMBINATION FORMULA (EQ. 1.9)

The number of ways of combining n things so that there are k hits is ( nk ) , for
example, combining 100 coins so that there are 60 heads showing. The formula is
( nk ) = k !( nn−! k )! . This appendix explains the logic behind this formula.
First, consider a simpler scenario. Suppose three people (Harry, Dick, and Tom)
go to the theatre. There are a number of possible seating arrangements:

HDT
DHT
HTD
DTH
THD
TDH

There are six permutations. Any one of the three people can fill the first seat.
Once the first seat is filled, there are two people left, so either one could fill the
second seat. Once the second seat is filled, there is just one person left, and so there
is just one possibility for filling the third seat. The total number of possibilities is
3 * 2 * 1 = 6. This is the factorial of 3 (3!). More generally, there are n! distinct ways
of arranging n objects.
Suppose now that Harry and Dick are identical twins. Some of the six permu-
tations will no longer appear distinct. If H = D and substituting H for D, the six
permutations become

HDT  HHT
DHT  HHT
HTD  HTH
DTH  HTH
THD  THH
TDH  THH

The number of distinct arrangements has been cut in half. We could just modify
the formula by dividing 6 by 2, or more specifically, 2!. In general, when k of the
objects (or people) are identical, the number of distinct permutations is given by nk!! .
We need to divide by k!, and not just k, because there are k! ways of arranging
the k identical objects, and each of these correspond to a possibility that is duplica-
tive of another possibility. There are therefore only three possible outcomes: HHT,
HTH, and THH. If instead of people, we let H represent a coin toss coming up heads
and T represent tails, we can see that there are three permutations corresponding
24 Measurement, Probability, and Binomial Formula

to a combination of two heads and one tail, and that ( 32 ) = 2 !( 33−! 2 )! = 3 . More gener-
ally, ( nk ) = k !( nn−! k )! , where (n - k!) is the number of misses. Although in the example,
(n - k)! = 1, in more complex examples (n - k)! may reflect duplicative possibilities,
so we need to divide by this term as well as k!.

Reference

Nussbaum, E. M. (2011). Argumentation, dialogue theory, and probability modeling: Alter-

native frameworks for argumentation research in education. Educational Psychologist, 46,
84–106. doi:10.1080/00461520.2011.558816
 CHAPTER 2

ESTIMATION AND HYPOTHESIS TESTING

In this chapter, we examine how to estimate a value of a population parameter such

as a probability (π) or mean (m). We will also examine the binomial test, which tests
hypotheses about π.

Estimation

The Estimation Problem(s)

In teaching this topic in my course on categorical data analysis, I begin the lesson
by asking students to answer on a piece of paper, “Have you ever been married?”
This question asks about a nominal variable and therefore it is appropriate to cal-
culate or estimate proportions. The goal of the exercise is to infer what percentage
of students at the university as a whole have ever been married (N = 30,000). In
other words, we need to estimate the population proportion. Doing so requires
taking a sample and using sample statistics to estimate the population parameter.
(A population parameter is some numerical feature of a population that we are
trying to estimate.) The sample should be random, but in this exercise we pretend
that the class represents a random sample. The population is all the students at
the university.
By convention, population parameters are represented by Greek letters: in this
case, pi (π) for a proportion. (For a metric variable, we would estimate μ, the popula-
tion mean.) By convention, English letters (e.g., P or X ) represent sample statistics:
Note that π, the population proportion, is also a probability. Therefore, if 60% of
the students at the university have been married, then if one randomly draws an
individual out of the population, there will be a 60% chance that she or he has been
married. Suppose our sample consists of 20 students (n = 20). If X is the number of
students in our sample who have been married, the expected value (nπ) is the most
26 Estimation and Hypothesis Testing

likely value of X in the sample. This number also corresponds to an expected value
of P of 12
20
=60%. This example shows why it is important to use random samples:
The laws of probability make it likely that the characteristics of the sample will be
representative (i.e., similar) to the population. In this case, if the sample contained
12 students who had been married, the P will be 60%, just like the population,
where π = 60%.
However, recall that this expected value is only the most likely value of our
sample estimate P; it is possible that we could draw a sample with 11 or 14 married
students. We should expect that we will obtain a sample of about 12 students who
have been married but that our estimate will be off a little. The “off a little” reflects
sampling error, the fact that even with a random sampling the sample might not
represent the population perfectly. Because of sampling error, we always represent
statistical estimates in the form of confidence intervals, for example .58 ± .02 (with
95% confidence).

Forming a Confidence Interval for a Proportion

In the above example, we stipulated that 60% of the population had been married.
However, we typically will not know the population parameters, which is why we
must estimate them. Suppose we take a random sample of 20 individuals, and 13
of them indicate that they have been married. (This category would include people
who are presently married as well as those who have been divorced or widowed.)
Then P = 13/20 = 65%. P is our sample estimate of π. How do we form a confidence
interval?
To address this question, we need to understand the concept of a sampling dis-
tribution. This concept is, next to probability, the most foundational one in all of
inferential statistics. A sampling distribution for a sample statistic, such as P, given
a particular sample size (e.g., n = 20), is the distribution of the statistic associ-
ated with all the different possible samples of size n that could be drawn from the
population. For each such sample, there is a P. The sampling distribution of P is the
distribution of all the possible Ps (one for each sample).
In the above example, we drew a sample with a P of 65%. If we were to draw
another sample, we might obtain a P of 70%. If we were to draw a third sample,
we might obtain a P of 55%. Theoretically, we could go on indefinitely, drawing
samples and plotting the values of the sampling distribution. In practice, we typi-
cally draw only one sample, at most, two. We therefore only observe one or two
points of the sampling distribution.
Although we never observe most of the points of the sampling distribution,
we can still make some theoretical inferences about it. In order to construct a con-
fidence interval, we need to make inferences about the sampling distribution’s
(a) mean, (b) standard deviation (standard error), and (c) shape.
 Estimation and Hypothesis Testing 27

It can be shown that the mean of the sampling distribution is π, the popu-
lation proportion. (This claim will be demonstrated in the next chapter and in
Appendix 2.1.)

EXPLORING THE CONCEPT

The mean of a sampling distribution is by definition the expected value. If the popu-
lation mean π were 50%, why is the most likely value of P, from all the samples that
could be drawn, 50%?

The population distribution is a Bernoulli distribution (see Figure 1.4 in the pre-
vious chapter), and so the graph of the population distribution will have two bars:
one for X = 1 [with Prob(X = 1) = π] and one for X = 0 [with Prob(X = 0) = 1 - π].
Recall that the variance of a Bernoulli distribution is π(1 - π ). Because we are using
P to estimate π, we can use PQ to estimate the population variance, where Q is the
probability of a miss (Q = 1 - P).
However, our interest is in finding the standard deviation of the sampling dis-
tribution, not of the population distribution. (The former is known as the standard
error, or SE). The applicable formula is

PQ
Estimated SE of a Proportion = . (Eq. 2.1)
n
Appendix 2.1 presents the proof (see also Chapter 3).

EXPLORING THE CONCEPT

Conduct a thought experiment by answering the following questions: If your sam-
ple consisted of just one observation (n = 1) from a population of 30,000, (a) How
many different samples could you draw? (b) Would the shape of the sampling dis-
tribution be just like the population distribution (i.e., Bernoulli)? (c) Would the stan-
dard deviations of the two distributions be the same?

Note from Eq. 2.1 that as n increases, the standard error decreases, meaning
that the sampling distribution becomes “thinner.” Eq. 2.1 implies that the standard
error approaches zero as n approaches infinity. In fact, the standard error would
approach zero as n approaches N¸ which is the size of the population. Technically,
therefore, we should include this additional constraint in the formula by writing

PQ PQ
Estimated SE of a Proportion = − . (Eq. 2.2)
n N
28 Estimation and Hypothesis Testing

When the population size is large, the second term is negligible, so it is typi-
cally left out of the formula for the standard error. One should use Eq. 2.2 when the
population size is small.

EXPLORING THE CONCEPT

Suppose you sample the entire population where n = N = 10, and P = 50%. To what
value would Eq. 2.2 reduce? How many different samples could one draw out of
the population? Does it make sense that the standard error would be zero?

In summary, the mean of the sampling distribution of P is π and the estimated

standard error is PQ n . In regard to shape, whereas the shape of the population dis-
tribution is Bernoulli, the shape of the sampling distribution will be binomial when
n > 1. This fact is illustrated in Figure 2.1. The sampling distribution for n = 1 is
Bernoulli (same as the population distribution). For larger n (n = 2 and n = 3 are
illustrated), the shape is binomial. Here is the rationale. The shape of the sampling
distribution for each observation will be Bernoulli (identical to the population dis-
tribution), and a series of independent Bernoulli trials yields a binomial distribu-
tion. Thus in a sample of size n = 2, there are four possible outcomes, as shown
in Table 2.1. Combining the middle two permutations where the number of hits
is 1 yields the sampling distribution of P, shown in Table 2.2 and Figure 2.1. Just
as the binomial distribution characterizes the probabilities resulting from flipping
two coins, it also characterizes the distribution associated with selecting a sample
of two.

50%
45%
40%
35%
Probability

30% n =1 (& pop. dist.)

25% n=2
20% n=3
15%
10%
5%
0%
0 1 2 3
Hits

Figure 2.1 Bernoulli and binomial sampling distributions for varying sample sizes. For
n = 1, the distribution is Bernoulli (and equal to the population distribution). For n = 2
and n = 3, the distributions are binomial.
 Estimation and Hypothesis Testing 29

Table 2.1 Sampling Distribution of P with n = 2, by Permutations

Possible outcomes

X1 X2 P Prob(P) Prob(P) if π = 50%

1 1 100% π2 25%
1 0 50% π(1 - π) 25%
0 1 50% (1 - π) π 25%
0 0 0% (1 - π)2 25%

Table 2.2 Sampling Distribution of P with n = 2, by Combinations

Xtotal P Prob(P) if π = 50%

2 100% 25%
1 50% 50%
0 0% 25%

In the previous chapter, we noted that the variance of a binomial distribution

could be estimated with the expression nPQ. However, the astute reader may have
noticed that I indicated above that the variance of the sampling distribution is PQ n ,
where we are dividing by the sample size rather than multiplying by it. The differ-
ence is that in this chapter we were dealing with a proportion and not a straight
number. In the previous chapter, we considered the example of tossing 100 coins
and asking what is the probability that 60 coins will come up heads.
The mean chance expectation was nπ, which, if π = 50%, was 50 coins. Expressed
as a proportion, the mean chance expectation is 50% (50 coins out of 100 coming
up heads is 50%). The variance of the binomial distribution, when we are predict-
ing the number of hits, is estimated by nPQ (100 * 50% * 50% = 25). The variance,
when we are predicting a proportion of hits, is PQ
n (
50% ∗ 50% = 0.25%) . (The proof is
100
in Chapter 3). In the former case, the standard error is 25 = 5. In the latter case, it
is .25% = 5%, which is the same as in the former case but cast in percentage terms.

The Central Limit Theorem and the Normal Approximation

Although the shape of the sampling distribution of P is binomial, we saw in the
last chapter that the sampling distribution can be approximated by the normal
distribution when n ≥ 20 and a continuity correction is applied. This is actually a
special case of a more general, important theorem, the central limit theorem (CLT).
The CLT holds that the sampling distribution of any random variable will approach
30 Estimation and Hypothesis Testing

normality as the sample size approaches infinity (see Technical Note 2.1). So if the
sample size is large enough, the sampling distribution will be approximately nor-
mal. When we assume normality because the sampling distribution is so close to
being normal, this is known as using a normal approximation. How large the sample
needs to be for a normal approximation to be reasonable is a complex question that
will be addressed later in this chapter.
For now, I wish to build the reader’s conceptual intuitions about the CLT with
another thought experiment. Suppose I take a sample of students at my university
(N = 30,000), but my interest is now in a metric rather than a categorical variable,
specifically, estimating the average height of males at the university. If we were to
take a sample of n = 1 and drew an individual with a height of 6 feet, then for that
sample, X =6′ . If we drew someone of 5 feet, then X =5′ . The distribution of pos-
sible X ′s would be identical to the population distribution. Because it is well known
that the distribution of heights by gender in a large population is normal, it would
follow that the sampling distribution is normal. This leads to the following impor-
tant result, which holds even for larger samples (n > 1): If the population distribution
is normal, then the sampling distribution will be exactly normal.
However, many if not most population distributions are not normal. Many are
skewed—for example, the distribution of household income or the number of high
school students in different schools who become professional actors. In addition,
we usually have little information about the shape of the population distribution;
although the sample can provide some information, we will nevertheless still be
uncertain about the shape of the sampling distribution.

EXPLORING THE CONCEPT

Examine Figure 2.2 on the next page. Can you explain the difference between the
empirical distribution (i.e., distribution of X in the data) and the sampling distribution?
Which one provides us with a hint regarding the shape of the population distribution?

Fortunately, when we are uncertain about the shape of the population distribu-
tion, or when we know it is not normal, we can invoke the CLT if the sample is large
enough. Suppose we want to estimate the average annual income of students at
the university and (going to the other extreme) we sampled the entire population,
so n = 30,000. There is only one possible sample we could draw, so the sampling
distribution will have no variance, as indicated by Eq. 2.1. Graphically, it would
be a spike, as shown in Figure 2.3 (light gray line in the middle). This spike is the
central limit.
Now let us take the next step and give the sampling distribution a little variance.
Suppose we sample all individuals in the population but one, so that n = 29,999.
Because there are 30,000 people in the population, there are 30,000 different
Discovering Diverse Content Through
Random Scribd Documents
The Project Gutenberg eBook of Gardens of the
Caribbees, v. 2/2
This ebook is for the use of anyone anywhere in the United States
and most other parts of the world at no cost and with almost no
restrictions whatsoever. You may copy it, give it away or re-use it
under the terms of the Project Gutenberg License included with this
ebook or online at www.gutenberg.org. If you are not located in the
United States, you will have to check the laws of the country where
you are located before using this eBook.

Title: Gardens of the Caribbees, v. 2/2

Author: Ida May Hill Starr

Release date: September 20, 2013 [eBook #43771]

Most recently updated: October 23, 2024

Language: English

Credits: Produced by Chuck Greif and the Online Distributed

Proofreading Team at http://www.pgdp.net (This file was
produced from images available at The Internet Archive)

*** START OF THE PROJECT GUTENBERG EBOOK GARDENS OF THE

CARIBBEES, V. 2/2 ***
 Every attempt has been made to replicate
the original as printed.
Some illustrations have been moved from
mid-paragraph for ease of reading.
In certain versions of this etext, in certain
browsers, clicking on this symbol will
bring up a larger version of the image.
Contents
List of Illustrations, Volume II.
Index
(etext transcriber's note)

GARDENS OF THE CARIBBEES

Volume II.

Travel Lovers’ Library

Each in two volumes profusely illustrated
Florence
By Grant Allen
Romance and Teutonic Switzerland
By W. D. McCrackan
Old World Memories
By Edward Lowe Temple
Paris
By Grant Allen
Feudal and Modern Japan
By Arthur May Knapp
The Unchanging East
 By Robert Barr
Venice
By Grant Allen
Gardens of the Caribbees
By Ida M. H. Starr
Belgium: Its Cities
By Grant Allen
L. C. PAGE AND COMPANY
Publishers
200 Summer Street, Boston, Mass.

FROM OUR BALCONY

CARACAS, VENEZUELA.

GARDENS OF
THE CARIBBEES
Sketches of a C r u i s e to the W e s t
Indies and the Spanish Main

By
Ida M. H. Starr
IN TWO VOLUMES
VOL. II.
ILLUSTRATED

Boston
L. C. Page & Company
MDCCCCIV

Copyright, 1903
By L. C. Page & Company
(INCORPORATED)
———
All rights reserved

Published July, 1903

Colonial Press
Electrotyped and Printed by C. H. Simonds & Co
Boston Mass., U. S. A.
 CONTENTS
CHAPTER PAGE
I. Island of Trinidad. “Iere” 11
II. Island of Trinidad. La Brea 35
III. The Spanish Main 64
IV. In Venezuela. Caracas 101
V. In Venezuela. Caracas to Puerto Cabello 125
VI. Curaçao. City of Willemstad 153
VII. The Southern Cross 189
VIII. Kingston, Jamaica 198
IX. “Cuando Salide La Habana” 239
X. A Memory of Martinique 247
INDEX
 LIST OF ILLUSTRATIONS

Volume II.
PAGE
From Our Balcony, Caracas, Venezuela Frontispiece
The Barracks, through Live-oaks and
Mahogany-Trees, Trinidad 17
Governor’s Palace and Public Gardens,
Port of Spain, Trinidad 21
On the Way to the Savannah, Port of
Spain, Trinidad 31
The Beach of la Brea, Trinidad 39
Asphalt for Northern Pavements, Pitch
Lake, Trinidad 47
Loading Cars, Pitch Lake, Trinidad 53
A Native Washerwoman on the Pitch
Lake, Trinidad 57
Where the Mountains Meet the Sea, La
Guayra, Venezuela 65
Caracas and the Mountains, Venezuela 75
Equestrian Statue of Bolivar, the
Liberator, Caracas, Venezuela 85
An Interior Court, Caracas, Venezuela 91
Cathedral and Plaza, Caracas,
Venezuela 111
A House Beside the Sea, Puerto
Cabello, Venezuela 127
A South American Street, Puerto
Cabello, Venezuela 149
Across Ste. Anne Bay, Harbour of
Willemstad, Curaçao 157
Some of Our Friends at Willemstad.—
Where the Basket-Women Waited,
Willemstad, Curaçao 161
The Landing, Willemstad, Curaçao 165
A Jolly Dutch Port, Willemstad,
Curaçao 173
A Snug Harbour, Willemstad, Curaçao 185
Kingston, Jamaica, from the Bay 199
Rio Cobre, near Spanish Town, Jamaica 203
A Native Hut, Jamaica 209
The Bog Walk Road, near Spanish
Town, Jamaica 213
Where We Landed, Kingston, Jamaica 223
El Morro, Entrance to Harbour,
Santiago de Cuba 229
The Plaza, Cienfuegos, Cuba 233
The Grave of Cervera’s Fleet, West of
Santiago de Cuba 237
The Wreck of the Maine, Havana
Harbour, Cuba 241
Cabañas, La Punta, and Harbour
Entrance, Havana, Cuba 245
St. Pierre and Mt. Pelée before the
Eruption, Martinique 249
St. Pierre and Mt. Pelée after the
Eruption, Martinique 253
Rue Victor Hugo Before the Eruption,
St. Pierre, Martinique 257
Rue Victor Hugo after the Eruption,
St. Pierre, Martinique 261
Gardens of the Caribbees
 CHAPTER I.

ISLAND OF TRINIDAD, “IERE”

I.

H AD we known just a little more about Trinidad, it would have made a

great difference in that luncheon, but it all came out wrong because
some of us didn’t know. Too late to influence us in the least, we read in
the Daily Gleaner, of Jamaica, that the beef sold in Trinidad is exported
alive from Venezuela. To be sure, we were aware that Venezuela occupies a
large part of the northern coast of South America, and were conscious that
Trinidad lies enclosed in a great bay of that coast, called the Gulf of Paria,
off the delta of the Orinoco River; also, in a hazy way, we knew that the
Spanish Main is a name applied somewhat vaguely to that same South
American coast—a relic of the days of pirates, buccaneers, and freebooting
English admirals; but we no more expected to be served a roast of beef
from the Spanish Main than a dish of Boston baked beans from our castles
in Spain. The two dimly intangible names had ever borne a close
comradeship in our minds, a poetic association affiliated them in closest
bonds. The same sun kissed into rose tints the turrets of our castles in Spain
and the lofty summits of the Spanish Main. The same romance lifted them
both away from reality into that land just bordering upon the Islands of the
Blest, and much as we longed to materialise our dreams, and make the
Spanish Main a usable fact, when the opportunity came for us to do so, it
slipped away from us before we were conscious of its existence.
Unaware that the illuminated postal-card menu on the table at the
Queen’s Park Hotel, Port of Spain, could in any sense lift the veil from our
enchantments, we read the following bill of fare:
Mayonnaise of Fish, with Lettuce
Oysters en Poulet
Scrambled Eggs with Asparagus Tips
Irish Stew
 Haricot of Oxtail
Brain Fritters
Curry of Veal à l’Indien
Boiled Turkey and Rice
Ham and Spinach
Fried Sausages and Potatoes
Salad
Assorted Cold Meats
String Beans Rice Mashed Potatoes
Macaroni au Gratin
Chocolate Ice-cream Cakes
Cheese
Eight of us sat down at a table on the veranda, white-walled, white-
ceilinged, and white pillared. A white-gravelled walk led out into the white
sun, through a stiff, boxed-in, English garden, stuffed with plants in green
tubs, and redeemed only by those natural things that will grow and be
beautiful in spite of all conventions. Thirsting for cool ices and delectable
fruits, looking wistfully for our vanishing fancies of West Indian ambrosia,
we turn in a listless, disappointed way to that bill of fare, where ham and
spinach and Irish stew and fried sausages send our hopes a-scampering off
like a lot of frightened children.
What man in his sane mind would order an haricot of ox-tail in the
tropics, when he needs but lift his hand for the food of Paradise; what man,
with any sense of the fitness of things, would eat curried veal, when, for the
asking, he might sup a libation fit for the gods? Alas! The asking never
brought it, and we—that is, one, at least—settled down to scrambled eggs,
and felt and looked unutterable scorn upon the one next at table who began
at mayonnaise of fish, and took every course to cheese. Ah! friends, this
was a case where the one who didn’t know fared ill. She lost her first
opportunity of paying her respects to the Spanish Main.
Hungry and disillusioned, the one and the only thing to do is to forget
those steaming sausages and the Irish stew as quickly as possible. We shall
not stay here a moment longer. Hotels are makeshifts at the best. Let us
leave these unromantic, unscrupulous venders of ham and spinach!
There, over yonder on the other side of the savannah, there is a delicious
retreat where we can make good our escape.
 II.
We shall never again see anything which can compare in beauty, of its
kind, with the Jardin des Plantes of Martinique. No, we never shall—still,
we must be just to all. Trinidad’s Botanical Garden is beautiful in its own
way, and we were impressed with the idea that it possessed some features
which that of Martinique lacked. However, that might have been owing to
the fact that we did not view the Martinique Garden in its entirety. Had we
done so, we might have found the same species in both places.
From casual observation there seemed to me to be one distinctive
characteristic of tropical vegetation; the trees did not appear to grow so
much in great social orders as do those of temperate zones. In the North,
vast families of the same species of trees gather together and keep together
with as rigid a pertinacity as any Scotch clan; the beech, birch, oak, maple,
pine, hemlock, walnut, hickory, all have their pet homes and their own
relations, and no amount of coddling or persuasion will ever induce them to
a wide change of habitat; but in the far South, the tropical trees seem
willing to settle anywhere in this land of endless summer. Of course, one
finds that certain trees love the swamps, and others prefer the high lands;
and some will grow in greater magnificence in some places where the
conditions are absolutely congenial, than in other places where they are not
so. There is the mangrove; it loves the wet and the mire—the mosquito-
ridden, miasmatic river borders—and wherever, on these coasts, you find a
swamp, whether in the very hottest spots, or in others only moderately so,
there you’ll find the mangrove sending out ærial roots, reaching down into
the muck for new strength, forming—banyan-like—a family of new trunks,
all under one leafy canopy, quite content if only it has the water about its
roots and a certain degree of heat.
Away up there in Haïti, we find the ceiba, and down here in Trinidad it is
equally at home. These conditions make the formation of a botanical
garden, representing the world-growth of sunlit vegetation, peculiarly
favourable. Trinidad is said to possess the most superb collection of tropical
plants in existence; and though gathered from all lands, growing not as
strangers or even stepchildren, but as rightful heirs to the immeasurable
vital force which pours forth from a rich soil warmed by a blazing sun the
year around.
 The Barracks, through Live-Oaks and Mahogany-Trees
Trinidad

The garden once entered, we pass a great, squarely built mansion, the
governor’s residence, and are in the midst of a wonderful vegetation from
the first step. At the very entrance, we are greeted with, perhaps, the most
unique tree in these latitudes.
After all, there is something stupefying in the effort to describe tropical
wonders. When they are passing before one’s eyes, each has a feature
distinct to itself, which, in a way, is its own manner of description. Each has
its peculiar wonder, its own glory,—no two alike—and yet, when one sits
down to think it over, there is the same old alphabet from which to draw
new pictures, new miracles; and how to make each different with the same
letters is a question indeed.
If I could only tell you the name of this particular tree which stands at
the entrance to the garden, you might some day hunt it up yourself, but as I
know neither its family nor home, we will let that all go, and just tell you
how it is dressed.
It is a heavily, glossily leafed, symmetrical, low tree, just about the size
of those dear old cherry-trees we used to climb, oh, so long,—so long ago!
From the tip of every branch there drops a cord-like fibre about a foot and a
half long, and at the end of this little brown string there hangs a cluster of
delicate pink flowers. These are suspended in almost exact length in rows
from the lowest to the highest branch, and it really seems as if Nature were
experimenting to see what wonderful living garlands she could create for a
canopy above our heads.

III.
The character of the garden is defined at once upon entrance. It is a
botanical garden, pure and simple, a place for strange plants from far away,
a sort of orphan asylum for everybody’s vegetable baby. It is not, like
Martinique, an enchanted forest with cascades and glens fit for nymphs and
dryads; it is matter-of-fact, orderly, prim, and businesslike. Aside from its
unique trees, there is little to attract one, so we decide for once it would be
wise to engage a guide who can tell us something about the inhabitants of
the place, which otherwise promises to be rather dull.

Governor’s Palace and Public Gardens

Port of Spain, Trinidad

So we hunt up a crooked, stump-legged Portuguese gardener, by name

Manuel, who takes our heavy baskets, we following down a little glen
which grows at once quite dark and sweet and silent.
Through long, freshly cut bamboo poles, streams of water are being
carried hither and thither to special spots in the garden, and we stop to
watch the trickling, and dip our hands down into its pleasant coolness.
Away up through the dark leafage, a mighty royal palm with stern
aristocratic grace swings and rattles its great, dead, brown arms—the
skeleton of its last year’s growth—beneath the luxuriant crown of this
year’s green plumes.
In the thicket, we find the nutmegs, hiding among the delicate foliage of
a low-branching tree. Sister reaches among the leaves and pulls off some of
the fragrant fruit, and gathers many from the ground. A sense of rare
luxuriance comes over us. This gathering of the spices of life from the very
ground upon which we tread is intoxicating, and we just begin to
understand the causes back of those dark pages of West Indian history,
when man first partook of this delirium.
These large-leafed, upright little trees are the Madagascar coffee, and the
smaller and more graceful ones, the Java coffee—how they take us back to
those happy days and months among the coffee plantations, long ago!—and
near by is the friendly banana, so common an object that we pass its torn,
drooping leaves with scarcely a thought, but it is worth more than a passing
glance, for there is no plant in all the tropics more useful than the banana. It
has not only delicious fruit of many sizes and varieties, but it is also cooked
as a vegetable, and forms one of the chief sources of the native diet. It is
planted, on account of its heavy shade and quick growth, to shade the
coffee, while trees of slower growth and more permanent shade are
maturing, thus forming a necessary and temporary protection; it is also used
for the same purpose among the cacao trees. It is a sort of foster-mother to
the cacao, to care for the tender shrub until its real mother, “La Madre del
Cacao,” can assume permanent care of its charge. The banana takes so little
vitality from the ground that, as protection to the coffee and cacao, it is
indispensable. We had some very delicious, green-skinned bananas at
several places, and found the small apple banana everywhere.
Manuel leads us on, and stops under a spindling, tall tree, flowering with
dainty, pink buds of a delicious odour, and there’s one branch just low
enough for Little Blue Ribbons to reach on tiptoe. Does it seem possible
that the little brown cloves, rattling in my spice-box at home, could ever
have been so fresh and soft and pink? Poor little mummies!
And just see what we are coming to! Did you ever imagine there could
be such shade? It’s a tree from the Philippines. We stoop to get under the
black leaves, and there the shade is absolutely impenetrable. What an
adjustment of things there is in this grand old earth of ours!
 My thoughts fly back to our Northern woods. I see the sinuously
graceful elms, with the sunlight streaming through their wide open branches
upon an earth longing for warmth; and long shafts of white noonday
shooting through the interstices of basswood, maple, and ash; the woods are
not black and sunless; they are translucently green, quivering with light and
needed warmth. But here, where the sun is a ball of redundant flame the
year around, Nature bequeaths to her children a shaded forest, rigidly
trunked, stolidly formed, thick-leafed, which no blazing sun can penetrate
or sweeping hurricane desolate.

IV.
Quite as one strokes the head of a favourite animal, Manuel leads us to
an insignificant-looking tree, takes a branch caressingly in his hand, brings
out his clumsy knife, selects just the right spot, cuts off a bit, and hands us a
piece of camphor wood.
Into the dear St. Thomas basket it goes, with the leaves of coffee, the
pink and white clove blossoms, and a long spray of araucaria from the
Norfolk Islands,—a strange company, indeed!
Yonder long yellow avenues are cinnamon and spice groves with
reddish-yellow bark, smooth as wax, casting slender shadows in the golden
light. Here is the shaddock, entirely weaned from its Malayan home, and
farther on a clump of low bushes, in among the nutmeg trees and coffee,
with small satin-like leaf, brings us to the herb that “cheers but does not
inebriate,”—the tea.
Just see those glorious great lemons, glowing in the ever-splendid
sunlight, which transmits to every living object a radiance, a dazzling
brilliancy, in which life progresses and finally dissolves out of sheer
exhaustion from the exuberance of vitality.
Oh, to our starved eyes of the North; to our senses benumbed by dreary
days of darkened sky, hearts chilled by bitterness of wind and gray,
unyielding frost, this never-ending, unspeakable sunlight, filtering through
the yellow vistas of clove and cinnamon, comes like the actual presence of
Apollo, the Shining One! We may, in unguarded moments, in ungrateful
moments, maybe, consider his embrace too positive, and we may raise the
white umbrella, but we never quite lose our rejuvenated love for his golden
glory.
Manuel, but half-clad, looks as if he would dismember at any moment.
His trousers are hitched by a couple of old leathers, and his shirt looks as if
it wished it “didn’t have to,” and his old hat is only there on sufferance, and
his shoes—old flippety-flops—have dragged their ill-shaped existence
through many a weary mile. But Manuel doesn’t care; he loves his garden,
and the sunshine and the luscious fruit, all his children so well behaved and
so obedient to his voice. He takes a bamboo pole and gives one of the big,
juicy lemons a rap, and down it falls on Wee One’s head with such a thump!
Then Manuel is very sorry, and he apologises for his child’s misdemeanour
in his funny, mixed-up Portuguese-English-Spanish and the rest, and we
understand and don’t mind a bit; in fact, we wouldn’t care if more would
fall in the same way.
Once upon a time, in the far-off golden days, when the Divine in
Creation had not been quite forgot, there came to this shore a band of men,
—not faultless, no, not faultless—but great men “for a’ that,” who, with
glittering cross aloft, christened this fair land after the blessed Trinity. But
this was not her first sacrament. Deep in the eternal silence of the forest, the
dweller in the High Wood had sought expression of the divine through
beauty, and chose a name from out the radiant wilderness which would tell
for ever of its wonderment: “Iere,” the land of the humming-bird, they
called her—those dusky children of the High Wood—and to this day she
clings lovingly to her maiden name.
We look about us. Where are the birds once peopling these forests, like
myriads of rainbows? Oh, sisters! members of Humane Societies! Hunt up
your old bonnets and see the poor little stuffed carcasses ornamenting your
cast-off finery! So Trinidad has been bereft of her wonderful birds, and now
there is but a name, a sad-sounding, meaningless name—Iere—to tell of
days which knew not the pride and cruelty of women.
Think of it!—at one time, there averaged twenty thousand humming-
birds a year exported from Trinidad to England alone!
And now, well—there are none left to export. We must find new islands
to denude, to ravage, to desolate, for our adornment. But it’s too unpleasant,
—this seeing things as they are; we’ll hide the poor little innocent card
which the black woman gave us at the hotel; we’ll cover up the word “Iere”
with these coffee leaves. There, now the spray of araucaria, now the stick
of camphor, and I think the lemon will fit right in among the nutmegs.
Come along, Manuel, we are ready; and we follow through the birdless
paths, down where the Nux Vomica grows, and the pepper, and the lime and
the calabash, and the orange and breadfruit, and tamarind, and pineapple;
and we go on and wake up the comical lizards who scurry away like brown
flashes of whip-cord. What ridiculous creatures they are, and how
desperately frightened! Why, surely they must be fifteen inches long, and
fully four inches high, and what funny, nimble legs! They start off in the
same spasm-like way as do the toy lizards we buy for the youngsters.
Manuel brings us to the plant house where the great forest wonders of
the Far East are babied and loved into strength, and I could not but think of
Daudet’s dear old Tartarin of Tarascon, dreaming by the homesick little
baobab-tree, which grew in his window-garden; and of the long nights
under the mellow moon of sunny France; and how he fought great beasts
and achieved great fame in the land of sweet illusion.

On the Way to the Savannah

Port of Spain, Trinidad
Copyright. 1901, by Detroit Photographic Co.

Dream on, Tartarin, wherever you may be! The time will come when it
will all be true, and you, too, will rest under the yellow splendour of the
golden trees; and the earth, the great Mother Earth, will open her heart to
you and breathe upon you the spirit of limitless possibilities!
 Good-bye, Manuel! The basket is heavy to carry with its spoils of fruit
and flowers; and we take “turn about” across the savannah.
The races are on, and horses are dashing around the grassy turf, and the
Trinidadians are yelling, the cricket games are going, and the picnic parties
are gathering up their baskets for home; and the Hindoo girls clamour to
carry our basket, and we gladly give over the load to a tough little head; and
the merry-go-round wheezes out its squeaking tunes, and we pass through
the black crowd, and narrowly escape taking a cab, for the way to the quay
looks long, and we waver and weaken, and are just about to give in, when
up comes a tinkling tram, and we jump in, with a penny to the Hindoo girl,
and rumble away.
The man with the two monkeys, and the man with the green and blue
parrot, and the boy with the shells, are still waiting.
Alackaday! Where is the woman with the baskets?
 CHAPTER II.

ISLAND OF TRINIDAD. LA BREA

I.

W E were led to believe, through various accounts from former

travellers, that the excursion to the Pitch Lake would be attended with
considerable discomfort and some hardships.
After a run of about four hours from Port of Spain, Trinidad, we made
La Brea at two o’clock in the afternoon of a blistering hot day. Fully one-
third of the ship’s company were frightened off, while the rest of us made
ready for the much-anticipated expedition.
It was a funny-looking company that stood at the gangway, waiting for
the first boat ashore.
Handkerchiefs took the place of collars and ties; coats and vests were,
for the most part, discarded, and all endeavoured to make themselves as
light in wearing apparel as possible.
The Caribbean Sea, which had, until now, been ruffled only by the
regular sweep of the “trades,” was badly tossed by a strong wind, so that the
embarkation in the ship’s boat was to me unpleasantly exciting. The sea was
running so high that, in order to reach the boat without being wet through,
we had to gauge our time well and take the jump just as the boat was lifted
to the top of the wave. As we started down the ship’s ladder, with Little
Blue Ribbons tightly holding Daddy’s hand, Sister having gone before in
the whale-boat with friends, the ship’s mate begged us to leave the Wee One
with him. He said the sea was too rough and the landing too difficult; and
besides he would take such good care of her, and she should have ice
cream, and be a little queen all day,—if she would only stay. So, with some
tears, and disdain for ice-cream, Little Blue Ribbons remained on board; the
only time in the journey thus far when she was not one of the party.
Had it not been for the confident man, who likes the water, and the
absurdity of the thing, I should have begged to be taken back to the ship.
 We were in the second boat. The captain had arranged to have the launch
tow us ashore, but the launch—true to the traditions of “oil engines”—had
no intention of towing us ashore; it puffed and popped and made a great
fuss, but would not move an inch. The engineer lost his steerageway, and it
seemed every moment as if the great, clumsy thing would crash into us; and
there we lay, going up and down the side of the ship, rolling from side to
side, and bobbing from bow to stern, in a very disagreeable situation for
those who don’t like that sort of thing.
I know quite well that I was not the only one who would gladly have felt
himself safe on the solid decks of our ship. For once, the incessant talking
had ceased, and our boat-load of people sat there absolutely quiet, thinking
very hard.
After numerous unsuccessful attempts to make the launch behave, they
gave up the attempt, manned our life-boat with six round-faced, lubberly,
German “jackies,” each with a big oar, and went off independently.
I was heartily thankful not to have been assigned to the launch, for it
could not compare in sea-going qualities with the boat in which we were
placed.
As I said, it was a long row to the landing, but we finally reached smooth
water, and disembarked at the end of a long bridge-like pier; not, however,
without some difficulty.
We were still some distance from shore, which was reached by means of
a narrow board walk, carried along one side of the pier, and bridging over
the shoal water.
At the quay, a big “down-east” schooner (thank Heaven, there are a few
American merchant vessels left!), two barks, and one full-rigged ship, were
being loaded with pitch, by means of great steel buckets, travelling on an
endless wire cable, which went from the end of the pier, up an incline, to
the works on the hill, near to the great deposit of pitch beyond.
Welcome to our website – the perfect destination for book lovers and
knowledge seekers. We believe that every book holds a new world,
offering opportunities for learning, discovery, and personal growth.
That’s why we are dedicated to bringing you a diverse collection of
books, ranging from classic literature and specialized publications to
self-development guides and children's books.

More than just a book-buying platform, we strive to be a bridge

connecting you with timeless cultural and intellectual values. With an
elegant, user-friendly interface and a smart search system, you can
quickly find the books that best suit your interests. Additionally,
our special promotions and home delivery services help you save time
and fully enjoy the joy of reading.

Join us on a journey of knowledge exploration, passion nurturing, and

personal growth every day!

ebookbell.com