Praise for Introductory Statistics Using R:

An Easy Approach

“This text serves not only as a process guide for utilizing R for statistical analyses, but it
also contains a very clear presentation of research methods, and the foundational knowledge
needed to understand the analyses.”—Coleen Cicale, Southeastern Louisiana University

“An essential guide for learning statistics with R, this book offers clear explanations, practical
exercises, and valuable multimedia resources, making complex concepts accessible and engag-
ing for students of all levels.”—Nicole Farris, Texas A&M University–Commerce

“This is an essential guide for new R instructors and students aiming to proficiently grasp
statistical analysis. With its clear, step-by-step approach and practical emphasis on real-world
applications, this textbook empowers learners with the necessary skills to navigate statistical
analysis in R with confidence. I highly recommend it!”—Selye Lee, University of Arkansas at
Little Rock

“I highly appreciate this amazing book. It provides a crucial understanding for students of the
importance and application of statistical tools in practice. Part 1: Statistical Foundation sets
the book apart, as it uniquely emphasizes the foundational reasons behind using statistical
methods, something not commonly found in other texts.”—Qiwei Li, The University of Texas
at Dallas

“This book offers a step-by-step guide appropriate for college students new to both statistics
and the statistical software R. The author magically guides us through the journey using
approachable language, complemented by the activities and exercises that build on each other
throughout the textbook.”—Yoko Mimura, California State University, Northridge

“Finally! A truly straightforward introduction to statistics and R. Writing is approachable,

clear, and concise. Students and those interested in getting started with R will find this text-
book extremely useful and approachable.”—Kate Pok-Carabalona, CUNY-Lehman College

“This is an effective, easy to use textbook that guides students not only through the funda-
mentals of statistics, but also how to run the analysis and interpret the outputs.”—Karen L.
Robinson, Westcliff University
“It is rare that a textbook presents both parametric and nonparametric approaches to basic
statistics. The information presented in this textbook is accessible and adds the skill of using
R to students resumes.”—Brandt A. Smith, Columbus State University

“This text is a helpful tool for teaching and learning statistics with R.”—Michael Welker,
Franciscan University of Steubenville
Introductory Statistics Using R
For Dean
Introductory Statistics Using R

An Easy Approach

Herschel Knapp
Yeshiva University
FOR INFORMATION: Copyright © 2026 by Sage.

2455 Teller Road All rights reserved. Except as permitted by U.S.

Thousand Oaks, California 91320 copyright law, no part of this work may be reproduced
E-mail: order@[Link] or distributed in any form or by any means, or stored in
a database or retrieval system, without permission in
1 Oliver’s Yard
writing from the publisher.
55 City Road All third party trademarks referenced or depicted
London, EC1Y 1SP herein are included solely for the purpose of illustration
United Kingdom and are the property of their respective owners.
Reference to these trademarks in no way indicates any
Unit No. 323-333, Third Floor, F-Block relationship with, or endorsement by, the trademark
International Trade Tower owner.
Nehru Place, New Delhi – 110 019
India
Printed in the United States of America
ISBN: 978-1-0719-2900-1
18 Cross Street #10-10/11/12
China Square Central
Singapore 048423

Acquisitions Editor: Helen Salmon This book is printed on acid-free paper.

Content Development Editor: Jennifer 25 26 27 28 29 10 9 8 7 6 5 4 3 2 1
Milewski

Production Editor: Astha Jaiswal

Copy Editor: Gillian Dickens

Typesetter: diacriTech

Cover Designer: Scott Van Atta

Marketing Manager: Victoria Velasquez

 BRIEF CONTENTS

Preface xvii

Acknowledgments xxiii

About the Author xxv

PART I STATISTICAL FOUNDATION 1

Chapter 1 Research Principles 3
Chapter 2 Sampling 19
Chapter 3 Getting Started in R 39

PART II STATISTICAL TESTS 61

Chapter 4 Descriptive Statistics 63
Chapter 5 t Test and Welch Two-Sample t Test 93
Chapter 6 ANOVA—Tukey Test and Wilcoxon Multiple Pairwise
Comparisons Test 119
Chapter 7 Paired t Test and Paired Wilcoxon Test 145
Chapter 8 Correlation—Pearson Test and Spearman Test 167
Chapter 9 Chi-Square 195

Glossary 217

Index 223
 DETAILED CONTENTS

Preface xvii

Acknowledgments xxiii

About the Author xxv

PART I STATISTICAL FOUNDATION 1

Chapter 1 Research Principles 3
Overview—Research Principles 4
Rationale for Statistics 4
Levels of Measure and Types of Variables 5
Continuous Variables 5
Interval 6
Ratio 6
Categorical Variables 6
Nominal 7
Ordinal 7
Summary of Variable Types 7
Control and Treatment Groups 8
Random Assignment 11
Research Question and Hypothesis Formulation 12
H0: The Null Hypothesis 12
H1: The Alternate Hypothesis 12
Hypothesis Resolution 13
Asking and Answering Research Questions 13
Good Common Sense 15
Key Concepts 16
Practice Exercises 16

Chapter 2 Sampling 19
Overview—Sampling 20
Sampling Rationale 21
Time 21
Cost 21
Feasibility 21
Extrapolation 22
x Introductory Statistics Using R

Sampling Terminology 22
Population 22
Sample Frame 22
Sample 22
Representative Sample 24
Probability Sampling 24
Simple Random Sampling 24
Stratified Sampling 24
Proportionate and Disproportionate Stratified Sampling 26
Systematic Sampling 26
Area Sampling 27
Nonprobability Sampling 29
Convenience Sampling 29
Purposive Sampling 30
Quota Sampling 30
Snowball Sampling 31
Sampling Bias 33
Optimal Sample Size 34
Good Common Sense 35
Key Concepts 35
Practice Exercises 36

Chapter 3 Getting Started in R 39

Overview—R and RStudio 40
Setting Up Your RStudio Cloud Account 41
R Syntax Guide 42
Loading Packages 42
Dataset Structure 43
Codebook 44
Uploading a Dataset to R 45
Data File Types 46
First Statistical Run 47
Variable References 47
Exporting Results 49
Copying Graphs 50
Shortcuts 51
Clear the Console Window 52
Doing Math in R 52
Data Order Doesn’t Matter 53
Processing Your Own Data 54
 Detailed Contents xi

Excel Option 55
Text Editor Option 56
Logging Off 57
Good Common Sense 57
Key Concepts 58
Practice Exercises 58

PART II STATISTICAL TESTS 61

Chapter 4 Descriptive Statistics 63
Overview—Descriptive Statistics 64
Descriptive Statistics in Context 65
Statistical Reasoning 65
Applied Examples 65
When to Use This Statistic 65
Descriptive Statistics for Continuous and Categorical Variables 65
Continuous Variables 66
Categorical Variables 66
Descriptive Statistics: Continuous Variables (Score) 66
Analysis of the First Column of Scores 66
n 66
Mean (M) 66
Median 67
Mode 67
Mean vs. Mode 68
Standard Deviation (SD) 68
Variance 69
Minimum 70
Maximum 70
Range 70
Introducing R 70
Dataset 70
Descriptive Statistics: Continuous Variables (Overall) 71
Documenting Results (Continuous Variable, Overall) 74
Histogram (Overall) 74
Descriptive Statistics: Continuous Variables (Stratified by Group) 75
Documenting Results (Continuous Variable, Stratified) 76
Histogram (Stratified) 76
Normal Distribution 77
Interpreting the Shapiro–Wilk Test 77
Skewed Distribution 78
Descriptive Statistics: Categorical Variables (Hand) 79
Documenting Results (Categorical Variable) 80
xii Introductory Statistics Using R

Managing Data 81
Managing Plots 82
Moving Forward 83
Good Common Sense 84
Key Concepts 84
Practice Exercises 85

Chapter 5 t Test and Welch Two-Sample t Test 93

Overview—t Test 94
t Tests and Welch Two-Sample t Tests in Context 94
Statistical Reasoning 94
Applied Examples 95
When to Use This Statistic 95
Example 95
Groups 96
Procedure 96
Research Question 96
Hypotheses 96
Dataset 96
t Test Pretest Checklist: ☑ 1. Normality and ☑ 2. Homogeneity of Variance 97
Pretest Checklist: Criteria 1, Normality 98
Pretest Checklist: Criteria 2, Homogeneity of Variance 100
Test Run 103
Test Run: t Test 103
Test Run: Welch Two-Sample t Test 104
Test Run: Descriptive Statistics 104
Results 105
Hypothesis Resolution 106
Documenting Results 107
Statistics in an Imperfect World 107
p Value 108
Type I and Type II Errors 109
Type I Error 109
Type II Error 109
Good Common Sense 110
Key Concepts 111
Practice Exercises 111

Chapter 6 ANOVA—Tukey Test And Wilcoxon Multiple Pairwise

Comparisons Test 119
Overview—ANOVA Test 120
ANOVA Tests in Context 120
 Detailed Contents xiii

Statistical Reasoning 120

Applied Examples 121
When to Use This Statistic 121
Layered Learning 121
Example 122
Groups 122
Procedure 122
Research Question 122
Hypotheses 122
Dataset 122
ANOVA Test Pretest Checklist: ☑ 1. Normality and ☑ 2. Homogeneity of Variance 123
Pretest Checklist: Criteria 1, Normality 123
Pretest Checklist: Criteria 2, Homogeneity of Variance 125
Test Run 126
Test Run: t Test Omnibus Test 126
Test Run: Tukey Test/Descriptive Statistics for ANOVA Analysis 128
Results: Tukey Test/Descriptive Statistics for ANOVA Analysis 129
Test Run: Wilcoxon Multiple Pairwise Comparisons Test 132
Results: Wilcoxon Multiple Pairwise Comparisons Test 134
Hypothesis Resolution 135
Documenting Results 136
Good Common Sense 136
Key Concepts 137
Practice Exercises 138

Chapter 7 Paired t Test and Paired Wilcoxon Test 145

Overview—Paired t Test 146
Paired t Tests and Paired Wilcoxon Tests in Context 147
Statistical Reasoning 147
Applied Examples 147
When to Use This Statistic 148
Layered Learning 148
Example 148
Groups 148
Procedure 148
Research Question 149
Hypotheses 149
Dataset 149
Paired t Test Pretest Checklist: ☑ 1. Normality of Differences 150
Test Run 151
Test Run: Paired t Test 153
Test Run: Paired Wilcoxon Test 154
xiv Introductory Statistics Using R

Results 154
Hypothesis Resolution 155
Documenting Results 155
Good Common Sense 156
Key Concepts 156
Practice Exercises 157

Chapter 8 Correlation—Pearson Test and Spearman Test 167

Overview—Pearson Test 168
Correlation in Context 168
Statistical Reasoning 168
Applied Examples 169
When to Use This Statistic 169
More About Correlation 169
Example 173
Groups 173
Procedure 173
Research Question 173
Hypotheses 173
Dataset 174
Pearson Pretest Checklist: ☑ 1. Normality, ☑ 2. Linearity,
and ☑ 3. Homoscedasticity 174
Pretest Checklist: 1. Normality 175
Pretest Checklist: 2. Linearity 177
Pretest Checklist: 3. Homoscedasticity 178
Test Run 180
Test Run: Pearson Test 180
Test Run: Spearman Test 181
Results 182
Hypothesis Resolution 182
Documenting Results 183
Correlation Versus Causation 183
Good Common Sense 184
Key Concepts 185
Practice Exercises 185

Chapter 9 Chi-Square 195

Overview—Chi-Square Test 196
Chi-Square Tests in Context 197
Statistical Reasoning 197
Applied Examples 197
When to Use This Statistic 198
 Detailed Contents xv

Example 198
Research Question 198
Groups 198
Procedure 199
Hypotheses 199
Dataset 199
Chi-Square Pretest Checklist: ☑ n ≥ 5 Per Cell 200
Test Run: Chi-Square 200
Pass 1: Assess the Pretest Criterion; Observe the Ns 202
Pass 2: Assess Percentages and the p Value 202
Results 204
Hypothesis Resolution 204
Documenting Results 205
Test Run: Bar Chart 205
Good Common Sense 207
Key Concepts 207
Practice Exercises 207

Glossary 217

Index 223
 PREFACE

FOR INSTRUCTORS WHO ARE NEW TO R

P lease allow me to begin with a brief personal note that is pertinent to instructors who
have not yet used R: With years of experience conducting research and teaching statistics
in SPSS, I was offered an opportunity to teach statistics at a university using R, which I’d
only heard of from a few enthusiasts. I was concerned that I’d need to master R prior to the
classes beginning, enabling me to prepare lessons and proficiently answer questions. I rea-
soned that since I already knew statistics, all I’d need to learn was how to communicate with
the R processor and locate the numbers of interest in the output reports.
The professor who hired me took me on a 30-minute guided tour of R, demonstrating
how to set up R and how to load and process data by entering lines of R code. It all made sense.
Over the next few days, I read the relevant chapters and drafted reference notes consisting of
the lines of R code necessary to run each statistical test, enabling me to copy and paste lines
of R code from that document into R; by editing the file name and variable names, every
analysis ran perfectly. I presented the prototype document to this kind professor, who affir-
matively referred to it as a nice cheat sheet. I’ve evolved this document into the R Syntax Guide,
the essential copy and paste resource referenced throughout this book that I use to teach all of
my statistics courses.
While you may opt to step through this book starting from the beginning, since you
already understand statistics, you may consider a more rapid approach to learning R:

Expedited R Learning:

1. Download the media from the Sage website for instructors and students at https://
[Link]/knapp1e (specifically, the R Syntax Guide, tutorial videos,
datasets, and solutions).
2. View the video for Chapter 3 and follow the directions to establish and configure
your R Cloud account. Use the R Syntax Guide and the downloaded dataset(s) to run
the analyses demonstrated in the video.
3. For Chapters 4–9, watch the video for each chapter, pausing as needed to run the
analyses that are demonstrated using the downloadable datasets and R Syntax Guide.
Then, complete as many of the exercises at the end of the chapter as you need to
solidify your comprehension; you can check your results using the downloadable
Solutions documents.

xvii
xviii Introductory Statistics Using R

Using this expedited learning method, you can plausibly expect to become proficient in R
in a matter of hours or days, not months or years. As you work in R, you may share my obser-
vation that copying, pasting, and editing prepared lines of R code from the R Syntax Guide is
more efficient than setting parameters on multiple menus.
Per my own experience learning R from square one, I can confidently tell you what I tell
my students at the start of each term: You’ ll do fine . . . and they do.
As with any course, instructors may opt to teach all or selected chapters of the book.
Further, instructors have the privilege to resequence the teaching order of the chapters. For
example, some instructors may choose to teach the chapters in the order provided, while oth-
ers may opt for a different order (e.g., teach Chapter 3: Getting Started in R first).

MEDIA RESOURCES

This text includes an array of instructor teaching materials designed to save you time and to
help you keep students engaged. To learn more, visit [Link] or
contact your Sage representative at [Link]/findmyrep. You’ll find tutorial videos, datas-
ets for the exemplary demonstrations and the exercises at the end of the chapters, and solutions
for all of the odd-numbered exercises to check your comprehension.
You’ll also download the R Syntax Guide, which contains prepared lines of R code for each
analysis in blue. You’ll copy and paste the blue R code into the R Console window, make
some minor edits (specifying the FileName and VariableName), and then press Enter. This
copy-paste-edit process will enable you to produce quality statistical results involving num-
bers and graphs without having to learn a whole programming language or write and debug
lengthy programs.
The R Syntax Guide is provided in a standard document format, so you can easily custom-
ize the document with your own notes. One caution: Be careful not to alter the blue R code
in this document; only edit the R code after you’ve copied and pasted the line of code into R.

OVERVIEW OF BOOK

In recent decades, with the omnipresence of computers in businesses, schools, homes, and
backpacks, statistical software has continued to advance and has come to be an integral part
of the statistical realm, likely for two major reasons: (1) Calculating statistics manually (by
hand) would be time-consuming and essentially impractical for processing large samples of
data, which could involve thousands of calculations, and (2) humans aren’t perfect—run-
ning complex algebra equations repeatedly is likely to produce some errors. Instead, statistics,
at this level, generally involves handing our data over to a statistical program, selecting the
proper test to run, instructing the computer to run the statistic, and then interpreting and
documenting the results. Notice that this method takes the mathematical processing out of
your hands and gives it to the computer to run, hence making statistics the least mathy of
math classes.
 Preface xix

Further, this book is not about statistical theory, nor does it focus on teaching complex
equations or formulas; we’ll trust that the software that we’re handing our data over to was
programmed properly. Instead, this book is about getting you to your statistical answers as
efficiently as possible via applied (not theoretical) statistics, designed to answer the three prac-
tical questions that many students have about statistics:

1. Which statistic should I use to process these data?

2. How do I instruct the computer to run the statistic?
3. How do I interpret the results so I can document the outcome in plain English?

Brief Table of Contents

Part I: Statistical Foundation
These chapters provide the basis for understanding and running statistics.

Chapter 1. Research Principles explains the key statistical terms and concepts for con-
ducting statistical analyses and interpreting results.
Chapter 2. Sampling reveals that you needn’t gather data on everyone in the population
to make viable statistical determinations. Multiple sampling options are discussed.
Chapter 3. Getting Started in R provides step-by-step guidance on establishing and ini-
tializing your free R Cloud account. You’ll also learn how to load a variety of files (data-
sets) into R and how to copy, paste, and edit prepared lines of R code from the R Syntax
Guide into the R processor, to quickly and easily run analyses, without having to learn a
whole programming language.

Part II: Statistical Tests

These chapters explain when to use each statistic, how to order the calculations and plots
(graphs), and how to comprehend the results to draft meaningful reports. You’ll notice that
Chapters 5 through 8 each include two statistics; the first one is the parametric version of the
statistic, and the other is the nonparametric version, but not to worry—you’ll only be running
one for each analysis, not both. Before you run either of these two versions of the statistical
test, you’ll run some initial analyses (pretest checklist), which will direct you to which version
of the test is appropriate to run for your data—the parametric or the nonparametric formula.

Chapter 4. Descriptive Statistics explains how to reduce the contents of any variable to
a concise list of numbers and graphs.
Chapter 5. t Test and Welch Two-Sample t Test determines if one group substan-
tially outperformed another group (e.g., control group vs. treatment group). The chapter
xx Introductory Statistics Using R

includes instructions for selecting, running, and documenting the proper version of the
test: t test or Welch two-sample t test.
Chapter 6. ANOVA—Tukey Test and Wilcoxon Multiple Pairwise Comparisons
Test; these tests are similar to the t test, which compares the performance of two groups
to each other, but this chapter provides guidance on comparisons involving three or more
groups. The chapter includes instructions for selecting the proper version of the test:
Tukey test or Wilcoxon multiple pairwise comparisons test.
Chapter 7. Paired t Test and Paired Wilcoxon Test determines the extent to which a
variable changes from one time point to another. It is often used to measure the effective-
ness of an intervention by comparing the pretest score (acquired before the intervention)
to the posttest score (acquired after the intervention) for each person in a group. The
chapter includes instructions for selecting, running, and documenting the proper version
of the test: paired t test or paired Wilcoxon test.
Chapter 8. Correlation—Pearson Test and Spearman Test determines the strength
and direction of the relationship between two continuous variables. The chapter includes
instructions for selecting, running, and documenting the proper version of the test:
Pearson test or Spearman test.
Chapter 9. Chi-Square Test determines if there is an association between two categori-
cal variables. This is a nonparametric test; there’s no equivalent parametric alternative.

The following table is included on page 2 of the R Syntax Guide. Basically, if you can identify
the types of variables in a dataset (continuous/categorical), this table will guide you to selecting
the proper statistical test. The concept of variable types (continuous/categorical) is thoroughly
covered in Chapter 1: Research Principles.
You’ve probably also noticed that Chapters 5–8 contain two statistics; relax—this doesn’t
mean that there’s twice as much work in each chapter. This merely indicates that prior to run-
ning these statistics, you’ll first conduct a preliminary assessment of the data as specified by
a Pretest Checklist, generally referred to as statistical assumptions, which will point you to the
proper version of the test to run. For example, in Chapter 5: t Test and Welch Two-Sample t
Test, you’ll begin by processing the Pretest Checklist to assess the data. Based on that assess-
ment, you’ll know which version of the statistic to run: the t test or the Welch two-sample t
test.
To partially explain this, if the variables are normally distributed (which will be discussed
in Chapter 1: Research Principles), the data would be considered parametric, in which case,
the first statistical test (t test) is appropriate. Alternatively, if the data are not normally distrib-
uted, the data are considered nonparametric, in which case, the second statistic (Welch two-
sample t test) would provide more precise results. This principle pertains to Chapters 5–8.
 Preface xxi

Guide to Selecting a Statistical Test

Test Variable 1 Variable 2 Example

Chapter 4 Continuous What is the average score?

Descriptive statistics

Chapter 4 Categorical How many left-handed and right-

Descriptive statistics handed people are in this group?

Chapter 5 Categorical Continuous Do doctors and nurses have the same

t Test and Welch two-sample (two level of stress?
t test categories)

Chapter 6 Categorical Continuous Do doctors, nurses, and technicians

ANOVA—Tukey test and (three have the same level of stress?
Wilcoxon multiple pairwise or more
comparisons test categories)

Chapter 7 Continuous Continuous Did people improve from their

Paired t test and paired (pretest) (posttest) pretestscore (before treatment)
Wilcoxon test compared to their posttest score (after
treatment)?

Chapter 8 Continuous Continuous Is a person’s years of education

Correlation—Pearson test correlated with their income?
and Spearman test

Chapter 9 Categorical Categorical Do people who sit and people who

Chi-square test stand do the same things on public
transportation (phone, read, music,
talk, nothing, other)?

Parametric and Corresponding Nonparametric Tests

Chapter Parametric Test Nonparametric Test

5 t Test Welch two-sample t test

6 Tukey test Wilcoxon multiple pairwise comparisons test

7 Paired t test Paired Wilcoxon test

8 Pearson test Spearman test

Layered Learning
The notion of layered learning pertains to the order of the chapters. Specifically, the concepts
in Chapter 5: t Test and Welch Two-Sample t Test provide the foundation for the statistical
xxii Introductory Statistics Using R

tests in Chapter 6: ANOVA—Tukey Test and Wilcoxon Multiple Pairwise Comparisons Test
and Chapter 7: Paired t Test and Paired Wilcoxon Test. The logic is as follows:
Chapter 5: t Test and Welch Two-Sample t Test provides instructions for statistically
analyzing the results of a two-group design (e.g., Group 1 gets antidepressant talk therapy,
and Group 2 gets antidepressant medication); we can measure the level of depression of the
members in each group (1 = low depression . . . 10 = high depression), then use these tests to
determine if one group significantly outperformed the other.
Chapter 6: ANOVA—Tukey Test and Wilcoxon Multiple Pairwise Comparisons Test is
just like the statistics covered in Chapter 5, except instead of being limited to comparing only
two groups to each other, ANOVA can evaluate the performance of three or more groups (e.g.,
Group 1 gets antidepressant talk therapy, Group 2 gets antidepressant medication, and Group
3 gets antidepressant talk therapy with medication). Hence, if you’re confident in your under-
standing of Chapter 5 involving two groups, you’re about 90% there when it comes to analy-
ses involving three (or more) groups; this is the rationale for the layered learning principle.
Chapter 7: Paired t Test and Paired Wilcoxon Test is also similar to the two-group design
detailed in Chapter 5, but instead of comparing the performance of people in two separate
groups, this involves only one group, wherein each individual is assessed twice in a pretest/post-
test fashion (e.g., first, administer the pretest to assess each person’s level of depression, then
provide pet therapy, and finally, administer the posttest using the same depression instru-
ment); the statistic compares the pretest score to the posttest score to determine if there’s a
significant change (in depression) after the treatment.
 ACKNOWLEDGMENTS

S age and the author acknowledge and thank the following reviewers whose feedback con-
tributed to the development of this text:

Bryon C. Applequist, Texas A&M University–Corpus Christi

David Bugg, SUNY–Potsdam
Coleen Cicale, Southeastern Louisiana University
Nicole Farris, Texas A&M University–Commerce
Gary Giumetti, Quinnipiac University
Ross Gosky, Appalachian State University
Sara Hofmann, Eckerd College
Selye Lee, University of Arkansas at Little Rock
Qiwei Li, The University of Texas at Dallas
Lauren McClain, Western Kentucky University
Yoko Mimura, California State University Northridge
Kim Nguyen-Finn, University of Texas–Rio Grande Valley
Kate Pok-Carabalona, CUNY–Lehman College
Alessandro Quartiroli, University of Wisconsin–La Crosse/University of Portsmouth
Karen L. Robinson, Westcliff University
Heini Seo, University of Indianapolis
Vida K. Simonaitis, South University–Tampa Campus
Brandt A. Smith, Columbus State University
M. Dylan Spencer, Georgia Southern University
Alisia (Giac-Thao) Tran, Arizona State University
Michael Welker, Franciscan University of Steubenville
Sarah Wylie, Colorado State University–Pueblo

We extend special thanks to Adeline Yeh for her skillful technical proofreading, to better
ensure the precision of this text.
We also gratefully acknowledge the contribution of Dean Cameron, whose cartoons
enliven this book.

xxiii
 ABOUT THE AUTHOR

Herschel Knapp, PhD, MSSW, has more than

30 years of experience as a health care professional
in a variety of domains. In addition to his clini-
cal work as a psychotherapist, primarily in hospi-
tal settings, he has provided project management
for innovative interventions designed to improve
the quality of patient care via multisite, health sci-
ence implementations. He teaches master’s and
doctoral courses at Yeshiva University; he has also
taught at the University of California, Los Angeles;
University of Southern California; California State
University, Los Angeles; and California State, San
Bernardino. Dr. Knapp has served as the lead stat-
istician on a longitudinal cancer research project
and managed the program evaluation metrics for
a multisite, nonprofit children’s center. His clinical
work includes emergency/trauma therapy in hospital settings. Dr. Knapp has developed and
implemented innovative telehealth systems, using videoconferencing technology to facilitate
optimal health care service delivery to remote patients and to coordinate specialty consulta-
tions among health care providers, including interventions to diagnose and treat people with
HIV and hepatitis, with special outreach to the homeless.
He created and implemented a nurse research mentorship program, providing research
and analytic education, enabling nurses to design, implement, analyze, and publish the results
of their quality improvement projects.
Dr. Knapp has published numerous articles in peer-reviewed health science journals
and serves as a peer reviewer for more than 20 science journals. He is also the author of
other textbooks, including Intermediate Statistics Using SPSS (2018), Introductory Statistics
Using SPSS (2nd ed., 2017), Practical Statistics for Nursing Using SPSS (2017), Therapeutic
Communication: Developing Professional Skills (2nd ed., 2014), and Introduction to Social Work
Practice: A Practical Workbook (2010).

xxv
 PART

STATISTICAL
FOUNDATION I

1
 1 RESEARCH PRINCIPLES

Here’s the basis for statistical thinking:

• Statistical Realms

• Rationale for Statistics

• Treatment and Control Groups

• Random Assignment

• Research Questions

• Hypothesis Formulation

• Accept/Reject Hypothesis

• Levels of Measure

• Types of Variables

LEARNING OBJECTIVES

Upon completing this chapter, you will be able to:

1.1 Understand the various domains that utilize statistics
1.2 Understand the rationale for using statistics
1.3 Understand the two types of variables and four levels of measure
1.4 Determine the variable type: categorical or continuous
1.5 Understand the reasons for having treatment and control groups
1.6 Comprehend the rationale for random assignment
1.7 Construct research questions
1.8 Formulate hypotheses
1.9 Appropriately accept/reject hypotheses based on statistical outcomes

3
4 Part I • Statistical Foundation

OVERVIEW—RESEARCH PRINCIPLES

This chapter provides an introduction to using and understanding statistics. Consider this
analogy: When a chef learns to cook, it involves more than just a tabletop of fresh food. That
chef also needs to learn how to use the tools of the trade (e.g., pots, pans, spatulas, thermom-
eters, timers, knives, spoons, stove), know how and when to expertly use each object to get the
job done, and understand how food preparation fits into the larger realm of the dining experi-
ence. The terms and concepts presented in this chapter will provide the basis for comprehend-
ing how statistics fits into the process of conducting scientific research.
As you proceed through this text, observe the diversity of examples and exercises as you
process them, demonstrating the multiple broad applicability of the statistics that you’ll be
learning. A partial list of professions that utilize statistics includes actuary, agriculture, bank-
ing, biology, business, census, chemistry, clinical trials, communication, computer science,
data science, defense, ecology, economics, education, engineering, epidemiology, finance,
forestry, genetics, health care, insurance, law, machine learning, manufacturing, marketing,
medicine, meteorology, pharmacology, physics, psychology, public health, research scientist,
risk assessment, safety, science writing, social work, sociology, sports, survey science, telecom-
munications, transportation, urban planning, and zoology.

RATIONALE FOR STATISTICS

While there is a form of statistics known as single-subject design (SSD), which tracks the
progress of one person (e.g., “Prior to motivational interviewing intervention, Dusty was
walking an average of 5,100 steps per day; now Dusty is walking an average of 8,300 steps
per day”), this book focuses on a much more common form of statistics, which involves
comprehending the overall characteristics and behaviors of multiple individuals, some-
times arranged in groups. Just as chefs reach for multiple utensils to prepare meals, statisti-
cians have a variety of tests to select from to analyze and comprehend the phenomenon of
interest, based on the design and type of variables involved, which is why this book con-
tains more than one chapter:
Chapter 4 covers descriptive statistics, sometimes referred to as summary statistics, which
are the most fundamental of all statistics, making it possible to summarize the data in a vari-
able to a concise result. For example, if we had a list of data that was 1,000 pages, we could
use descriptive statistics to answer questions like “What was the average test score?” or “How
many left- and right-handed people took this test?”
While this kind of information is useful, Chapters 5 through 9 cover inferential statistics,
which goes beyond averages and headcounts, providing results designed to answer questions
that provide the basis for making more informed decisions.
For example, learning inferential statistics will enable you to ask and confidently answer
statistical questions such as “Do children whose parents are teachers do better in school?”
“What form of therapy is best for reducing depression: talk therapy, antidepressant medication,
or talk therapy combined with antidepressant medication?” “Do elementary students perform
 Chapter 1 • Research Principles 5

better on tests before or after lunch?” “Do people who have more siblings tend to be happier
than people from smaller families?” “Do people who took a test prep course pass an exam at the
same rate as those who didn’t take the course?” Despite the variety of research questions and
data that we may gather, statistics enables us to process the data and provide results that are uni-
versally accepted and understood, providing a common basis for communicating our findings.
These statistical processes provide the foundation for engaging in evidence-based practice
(EBP), meaning that we don’t just use our intuition to guess what might be an effective inter-
vention—instead, we turn to the numbers to help us make more objective determinations.
For example, statistics can help us see if one teaching method outperforms another or which
outreach process renders the most people opting for flu shots. To continue the example, sup-
pose a researcher discovers that an instructor has created a prep course that significantly
increases the passing rate of a certification exam. If that researcher publishes these results,
which would include not only the statistics but also the methods and materials that were
used to achieve these positive results, others could read this, and possibly other related manu-
scripts, and adopt or adapt this method to achieve the same results with others.
Although it may seem paradoxical, one might say that the first step in conducting evi-
dence-based practice research is not to launch a research project but rather to explore the
scientific literature to find out if your research question has already been asked and answered
by one or more other researchers. If not, then it may be time to design and implement a new
research project; however, often researchers and practitioners will find peer-reviewed quality
research publications that can be referenced and used as is or adapted to meet their needs.
Evidence-based practice sometimes involves creating a hybrid model, selectively combining
methods and metrics from more than one effective approach.
Without statistics, it can be difficult to know if well-intentioned interventions are having
a positive, negative, or neutral effect. For example, suppose a committee believes that regular
aerobic exercise will reduce stress, so the researchers at a workplace gather a group of 20 vol-
unteer participants and provide 30-minute exercise sessions Monday through Friday for 10
weeks. Without some form of statistical analysis, it would be difficult, if not impossible, to be
certain if this exercise program was actually effective in reducing stress.

LEVELS OF MEASURE AND TYPES OF VARIABLES

Some might say that half of the battle in statistics is knowing which statistical test to run. Part
of what you’ll need to know to select the proper statistic involves determining if a variable is
a continuous variable or a categorical variable. There are two types of continuous variables
(interval variable and ratio variable) and two types of categorical variables (nominal variable
and ordinal variable). We’ll proceed with definitions and examples of each of these.

Continuous Variables
Continuous variables are the kinds of variables that you’re probably most familiar with.
Continuous variables are the numbers that you are accustomed to using when counting or
6 Part I • Statistical Foundation

solving a math equation. Imagine a number line where 0 is in the middle; to the left of 0 are all
of the negative numbers, –1, –2, –3, and continue forever. To the right of 0 are all of the posi-
tive numbers, 1, 2, 3, and continue forever like numbers on a number line:

−5 −4 −3 −2 −1 0 1 2 3 4 5

Another way to think about this is a continuous number is any number that you can enter
into a calculator. There are two levels of continuous variables: interval and ratio.

Interval
Interval variables are any number from negative infinity to positive infinity (including all
fractions). The numbers are evenly spaced; for example, the distance between 3 and 4 is the
same as the distance between 4 and 5. Examples of interval variables include elevation where
sea level = 0; anything above sea level is a positive elevation, and anything below sea level is
a negative elevation. A bank balance is an interval variable, since a bank balance could be
negative (e.g., –$14.83) indicating debt, or positive (e.g., $65,780.99). Temperature is also an
interval variable as it can be negative or positive (e.g., –15° . . . 93°).

Ratio
Ratio variables are just like interval variables; however, ratio variables cannot be negative. For
example, the number of puppies in a box is a ratio variable, because it cannot be negative (e.g.,
5 puppies, 2 puppies, 1 puppy, 0 puppies); there’s no such thing as a box that contains –3 pup-
pies. Other examples of ratio variables include the number of minutes that one is engaged in
an activity, weight, height, age, IQ score, and heartbeats per minute.

Learning tip: The lowest possible value for a ratio variable is 0; notice that the word ratio
ends in o, which resembles the number 0.

Categorical Variables
Unlike continuous variables that contain numbers, categorical variables contain (nonnumeric)
lists. The items in these lists can be thought of as categories. For example, Employment is a cat-
egorical variable that contains two categories: Employed and Unemployed. DishwasherStatus
is a categorical variable that also contains two categories: Dirty and Clean. A categorical vari-
able can contain more than two categories; for example, the variable CarColor may contain
eight categories: Black, Blue, Brown, Gray, Green, Red, Silver, White. There are two levels of
categorical variables: nominal and ordinal.
 Chapter 1 • Research Principles 7

Nominal
Nominal variables contain categories that have no particular order. For example, the variable
EyeColor could contain the following categorical values: Amber, Blue, Brown, Gray, Green,
Hazel, Violet. Although this list is arranged alphabetically, there’s really no inherent order
to the variable EyeColor. For example, we could have presented the categories in some other
order such as Green, Hazel, Violet, Blue, Amber, Gray, Brown, or Brown, Gray, Hazel, Violet,
Blue, Amber, Green. Another example of a nominal variable is VoteStatus, where the categories
would be Voted, Did Not Vote. Alternatively, the categories could be sequenced the other way:
Did Not Vote, Voted.

Learning tip: The word nominal begins with the word no, as in no order.

Ordinal
Ordinal variables are categorical variables that have an inherent ranked order among the
categories. For example, the categorical variable Meal has the following categorical values:
Breakfast, Lunch, Dinner. It would be irrational to arrange the categories arbitrarily, such as
Lunch, Dinner, Breakfast. The sequence of the values in ordinal variables is typically arranged
in ascending order (from first to last or from lowest to highest); for example, for the ordinal
variable Size, it makes sense to arrange the categories in this order: Small, Medium, Large,
Extra Large. The ordinal variable Education could contain categories in this order: Did Not
Complete High School, High School Diploma, Associate’s Degree, Bachelor’s Degree, Master’s
Degree, Doctorate Degree.

Learning tip: The root of the word ordinal is order, indicating that there is an inherent
order to the categories.

Summary of Variable Types

The levels of categorical (nominal and ordinal) and continuous (interval and ratio) variables
are included for completeness, but when it comes to selecting, running, and documenting
statistics, you’ll need to proficiently determine if a variable is either continuous or categorical,
as detailed in Table 1.1.

Learning tip: Consider the acronym NOIR: Nominal, Ordinal, Interval, Ratio.
8 Part I • Statistical Foundation

TABLE 1.1 ■ Categorical and Continuous Variables

Type Level Example

Categorical Nominal Football, Hockey, Baseball

Ordinal Breakfast, Lunch, Dinner

Continuous Interval . . . –3, –2, –1, 0, 1, 2, 3 . . .

Ratio 0, 1, 2, 3 . . .

CONTROL AND TREATMENT GROUPS

You’ve likely heard of control and treatment groups in scientific settings. In short, the con-
trol group is given no intervention, and the treatment group is given the intervention that
is expected to improve some condition. This enables us to compare the performance of the
control group to that of the treatment group to determine how effective the treatment was.
Depending on the study design, the control group isn’t always given nothing; a variation of
this design involves providing the members of the control group treatment as usual (TAU),
and the treatment group gets a new intervention. For example, it could be considered unethi-
cal to give a placebo (e.g., a sugar pill) to a patient with a serious disease just because that per-
son is in the control group—instead, the researcher could provide them with the drug that’s
traditionally used to treat their condition, while those in the treatment group would be issued
the new drug that’s under investigation.
Initially, the notion of having a control group may seem useless: What’s the point of having
a group that we do absolutely nothing to—isn’t that a waste? Why not just get one group of people
and see if the treatment works on them—after all, isn’t that what we want to know?
Consider this series of five illustrated examples; the first involves only a treatment group
(no control group) to determine if classical music enhances plant growth. You plant a seed in
fertile soil and regularly provide the plant with water, sunlight, and 8 hours of classical music
per day for a year (Figure 1.1).

FIGURE 1.1 ■ Treatment Group Only (Positive Treatment Effect)

 Chapter 1 • Research Principles 9

In Figure 1.1, we observe that the plant that got the music produced 20 healthy leaves;
hence, we could conclude that classical music has a positive effect on plant growth, but the
results of this one-group (treatment-only) design may be questionable. Specifically, one may
reason: That plant got sunlight, good soil, water, and music, so how can we be sure that it was the
music that actually contributed to the plant’s growth? Maybe it would have grown up fine with just
the sunlight, soil, water, and no music.
To address this, we could implement a more robust model using a two-group design con-
sisting of a treatment group and a control group (Figure 1.2).

FIGURE 1.2 ■ Treatment Group and Control Group (Neutral Treatment Effect)

In Figure 1.2, we see a two-group design involving a treatment group and a control group.
The plant in the treatment group will be in Room 201, where the planted seed will get quality
soil, sunlight, water, and daily music. The plant in the control group will get the same seed,
soil, lighting, watering, and even the planter as the plant in the treatment group, but the plant
in the control group will be placed in Room 222, which is far down the hall where the music
cannot be heard. One year after planting, we observe that the plant in the treatment group
appears to have grown identically to the plant in the control group—both have 20 healthy
leaves; since the plant that got no music did as well as the plant that did get music, based on
this comparison, it appears that the music added nothing to the plant’s growth, and hence, we
might conclude that the music had a neutral effect on plant growth.
Before concluding this example, refer back to Figure 1.1, which involved only the treatment
group; in that case, we concluded that the plant growth was positive based on the 20 healthy
leaves, which may be attributable to the music. The two-group design in Figure 1.2 provides
us with more information; notice that the plant in the control group, which got no music, also
has 20 healthy leaves (just like in Figure 1.1). Further, when we compare the plant in the treat-
ment group to the plant in the control group, we change our opinion on the meaning of those 20
healthy leaves—since the plant that got no music in the control group grew equally to the plant
that got music in the treatment group, we can no longer plausibly claim that the music had a
positive effect on the plant’s growth; it seems that this is just normal growth for this plant.
10 Part I • Statistical Foundation

Figure 1.3 illustrates another two-group design, which tells a different story: The plant in
the treatment group that was exposed to music grew considerably more compared to the plant
in the control group. Since the only thing that’s different between these two groups is that the
treatment group got music and the control group did not, these results suggest that music had a
positive effect on the plant’s growth.

FIGURE 1.3 ■ Treatment Group and Control Group (Positive Treatment Effect)

In this example, Figure 1.4 illustrates that the music had a negative effect on plant growth;
clearly, the plant in the control group grew better without the music compared to the plant in
the treatment group. Had there only been the treatment group (and no control group), we might
conclude that the music helped the plant produce eight healthy leaves. Statistically comparing
the treatment group to the control group provides a stronger understanding of the effectiveness
(or ineffectiveness) of various treatments.

FIGURE 1.4 ■ Treatment Group and Control Group (Negative Treatment Effect)

To finalize this example, observe that Figure 1.5, depicting a plant that produced eight
leaves, with no control group to compare it to, might (mis)lead us to conclude that eight leaves
 Chapter 1 • Research Principles 11

FIGURE 1.5 ■ Treatment Group Only (Positive Treatment Effect)

are as much as we can expect from this plant, even with the music. The consequence of hav-
ing a treatment group only produces the same conundrum as we saw in Figure 1.1: Without
the control group for comparison, we might mistakenly conclude that the music had a positive
effect on this plant’s growth when, in fact, it may have had no effect or even a negative effect.
For visual clarity, this series of examples has involved one plant in each group, but to fur-
ther solidify the results of a study like this, it would be good practice to have more than one
plant in each group, just in case we might have unknowingly used a better seed for one plant
than the other—that would be an unfair comparison. To get more robust statistical results,
we could set up 30 plants in each group, and then after a year, we could calculate the average
number of leaves per plant in the treatment group and compare that to the average number of
leaves per plant in the control group to determine if one group significantly outperformed the
other; this will be covered in Chapter 5: t Test and Welch Two-Sample t Test.

RANDOM ASSIGNMENT

Continuing with our plant example, suppose you’re at the beginning of the process; you have
enough planters, soil, and seeds to prepare 60 plants—30 for the control group and 30 for the
treatment group. The seeds are provided in a clear container, and hence the seeds near the top
and sides have been exposed to more light than the rest of the seeds. You wonder if this might
have an effect on the plant’s growth, so you confer with a botanist, who confirms that when this
type of seed is exposed to light prior to planting, it slightly enhances growth. Since the seeds
can shift around in the bag during shipping, making it impossible to know precisely how much
light each seed has been exposed to, the best strategy is to randomize the seeds. This will help
to reduce the bias among the groups—if we mix up the seeds prior to planting and assigning
plants to (control/treatment) groups, then the light-exposed seeds have an equal chance of being
assigned to the control group and the treatment group. If we have a sufficient sample size, random
assignment should evenly distribute low- and high-exposed seeds among the groups; hence,
prior to beginning any treatment(s), random assignment to the groups helps neutralize bias. The
rationale and methods for randomization are covered in further detail in Chapter 2: Sampling.
12 Part I • Statistical Foundation

RESEARCH QUESTION AND HYPOTHESIS FORMULATION

A hypothesis is a provisional statement or statements that rationally address a research

question prior to gathering and analyzing data. For example, suppose we want to research
if working with a tutor improves academic performance. There are a variety of ways to form
the research question. For clarity, we’ll be using a unified approach throughout this book,
wherein the research question will be phrased as a yes/no question: Does working with a tutor
improve academic performance?
You can think of the hypotheses as the possible answers to the research question. Since
our research question has two possible outcomes, yes or no, we can write the two hypotheses,
anticipating the statistical results that will emerge: one for the yes answer and another for the
no answer. First, we’ll focus on the no answer:

H0: The Null Hypothesis

Realistically, we must recognize that not all interventions are going to be effective; some-
times, an intervention may fail to have the effect that we hoped for. In this case, our research
question is: Does working with a tutor improve academic performance? Hence, we first write the
hypothesis that corresponds to the no answer: Tutoring has no effect on academic performance.
This is the null hypothesis (H0), suggesting that the treatment had a null effect—tutoring had
no effect on the student’s academic performance. Essentially, we’re saying that the interven-
tion was ineffective.
This would be documented as follows:

H0: Tutoring has no effect on academic performance.

H1: The Alternate Hypothesis

Next, we consider the yes answer: Tutoring has an effect on academic performance. This is the
alternate hypothesis (H1), meaning that this is the alternative to the null hypothesis—tutoring
had an effect on the student’s academic performance.
This would be documented as follows:

H1: Tutoring has an effect on academic performance.

The research question and hypotheses could be presented as such:

Research question: Does working with a tutor improve academic performance?

H0: Tutoring has no effect on academic performance.

H1: Tutoring has an effect on academic performance.
 Chapter 1 • Research Principles 13

Notice that when we phrase the research question as a yes/no question, there’s only a one-word
difference between the null hypothesis (H0) and the alternate hypothesis (H1):

H0: Tutoring has no effect on academic performance.

H1: Tutoring has an effect on academic performance.

Hypothesis Resolution
After gathering the data and running the appropriate statistical analysis, you’ll then refer
back to the hypotheses (H0 and H1) and select the hypothesis that corresponds to the statisti-
cal results. The inferential statistics covered in Chapters 5 through 9 include guidance for
resolving the hypothesis. In other words, the statistical results will give you what you’ll need
to know for hypothesis resolution: identifying which hypothesis (H0 or H1) corresponds to
the statistical results. From there, you can concisely document your findings, citing selected
numbers that R will provide in the statistical results.

ASKING AND ANSWERING RESEARCH QUESTIONS

A statistician colleague of mine once said, “I want the numbers to tell me a story.” I’ve never heard
the mission of statistics expressed so elegantly. One way to think about statistics is to conceive
that the numbers in a dataset contain a story; the answers to your questions are hidden within the
data, and we can use statistics to filter the data, coaxing it to reveal its secrets. Admittedly, that
makes things sound a bit mystical, but in a way, that’s precisely what happens when we process
statistics—the formulas systematically organize and process the data to produce a concise set of
results that help us understand what we could not see just by gazing at the data table.
Chapters 4 through 9 are structured to answer different types of statistical questions
depending on the design and type(s) of variables involved.
Chapter 4: Descriptive Statistics reduces a long list of data into a short list of figures,
enabling you to comprehend and communicate the contents of any variable. For example, you
could answer questions like:

• How many left- and right-handed people are in our group?

• What’s the average score?
• What’s the youngest and oldest ages in a workplace?
• What percentage of our staff is part-time and full-time?
• What’s the average number of classes students are enrolled in?
• How many people in this group voted in the last election?
14 Part I • Statistical Foundation

Chapter 5: t Test and Welch Two-Sample t Test compares two conditions/treatments

to determine if one outperformed the other or if they both performed about the same. These
tests can statistically answer questions like:

• Do tutored students have the same test scores as nontutored students?

• Which drug is best for reducing hypertension (high blood pressure): Drug A or Drug B?
• Do people take the same amount of time to eat one scoop of chocolate ice cream compared
to one scoop of vanilla ice cream?

Chapter 6: ANOVA—Tukey Test and Wilcoxon Multiple Pairwise Comparisons

Test is very similar to Chapter 5: t Test and Welch Two-Sample t Test, except instead of being
limited to comparing only two groups to each other, the statistics in this chapter can process
comparisons involving three or more groups (notice that these examples are the same as those
in Chapter 5, but with additional groups):

• Do tutored students, nontutored students, students who use a learning app, and students
who engage in study groups have the same test scores?
• Which drug is best for reducing hypertension (high blood pressure): Drug A, Drug B, or
Drug C?
• Do people take the same amount of time to eat one scoop of chocolate ice cream, one scoop
of vanilla ice cream, and one scoop of strawberry ice cream?

Chapter 7: Paired t Test and Paired Wilcoxon Test can assess the effectiveness of a
treatment wherein one group of people is pretested, then given some intervention, and then
each person is given a posttest (which uses the same measurement as the pretest). If the post-
test score shows a significant improvement compared to the pretest score, this suggests that
the treatment was effective. This statistic can answer a variety of questions involving a single
group of people:

• To assess the effectiveness of Quick Coaching, a coach unobtrusively observes an archer

shooting 10 arrows at a target and records the score (center ring = 10 . . . outermost ring
= 1, misses the target = 0), and then the coach provides improvement recommendations,
after which, the archer shoots 10 more arrows. The coach compares the scores before and
after giving the feedback between rounds.
• To determine if food has an effect on mood, we ask each participant to score their mood
(1 = bad mood . . . 5 = good mood), and then we provide a slice of really good chocolate
cake; when they’re done with the cake, we ask them to report their current mood score
using the same 1 to 5 scale. Finally, we compare the mood scores before and after eating
the cake.
 Chapter 1 • Research Principles 15

• To find out if music helps people relax, we first take the person’s pulse rate, then we play 10
minutes of soft instrumental music, and then we take the person’s pulse rate again. We can
then compare the first pulse rate to the second pulse rate to discover if the pulse rate changes
significantly after listening to the music.

Chapter 8: Correlation—Pearson Test and Spearman Test can be used in a single-

group design to assess the relationship between two continuous variables. This statistic can
answer questions like:

• Is there a correlation between income and happiness?

• Is there a correlation between age and hours of sleep per night?
• Do students who spend more time completing a test tend to have higher scores or do
students who complete the test quicker tend to have higher scores?

Chapter 9: Chi-Square Test is used to determine if there’s an association between two

categorical variables. This statistic can be used to answer questions like:

• Do all political parties have the same proportion of members in favor of voting by mail?
• Do Collies, Keeshonds, and Cocker Spaniels have the same preference when it comes to
selecting a bone or a ball?
• Do the same percentage of degreed and nondegreed individuals pass a CPR certification class?

These examples demonstrate the diverse types of research questions that can be asked and
answered using a fairly concise set of statistical analyses.

GOOD COMMON SENSE

As you compute multiple statistics throughout this text, be aware that despite our diligence
for precision, statistics do not prove or disprove anything; rather, statistics are generally used
to help us to understand the characteristics of a group and to reduce uncertainty. Statistical
results are typically written tentatively (e.g., The results suggest that this intervention was effec-
tive in treating this problem.) as opposed to definitively (e.g., The results prove that this interven-
tion was effective in treating this problem.).
Also, recognize that statistical results reflect the characteristics or performance of the
group(s) of data (people) in our dataset and do not provide insights regarding any particular
individual. For example, suppose we analyze the ages of five people who are 8, 12, 22, 53,
and 60; the average age is 31, but it would be wrong to think that you could arbitrarily point
to any one person in that group and confidently proclaim, “You are 31 years old.” Although
the average is 31, it’s possible that no individual in the group is 31, and in this case, none of
16 Part I • Statistical Foundation

these people are in their 30s. Further, if we asked a group of people, “How many children did
your parents have?” and the average turns out to be 2.3, it would be foolish (and comically
gruesome) to think that all of these families have 2.3 children. In summary, statistical results
pertain to the overall group and are not intended to plausibly characterize or make predictions
about any specific individual(s).

KEY CONCEPTS

• Rationale and uses for statistics

• Level of data (continuous: interval, ratio; categorical: nominal, ordinal)
• Types of data (continuous, categorical)
• Treatment group
• Control group
• Random assignment
• Research question
• Hypotheses (H0: null, H1: alternate)
• Accepting/rejecting hypotheses

PRACTICE EXERCISES

For the following exercises, describe the basis for an experiment that would render data that
could be processed statistically.

Exercise 1.1
The director of a healing center wants to determine if 30 minutes of guided meditation
affects the resting pulse rate.
a. Indicate the groups (e.g., control, treatment).
b. State the research question.
c. State the hypotheses (H0 and H1).

Exercise 1.2
A pediatrician wants to find out if having a magician visit and perform for hospitalized chil-
dren at their bedside has an effect on their anxiety.
a. Indicate the groups (e.g., control, treatment).
b. State the research question.
c. State the hypotheses (H0 and H1).
 Chapter 1 • Research Principles 17

Exercise 1.3
A physical education instructor wants to know if seventh-graders and eighth-graders on a
team are about the same heights.
a. Indicate the groups (e.g., control, treatment).
b. State the research question.
c. State the hypotheses (H0 and H1).
Exercise 1.4
A professor wants to determine which version of a course produces the best grades: in-person
classroom or live online.
a. Indicate the groups (e.g., control, treatment).
b. State the research question.
c. State the hypotheses (H0 and H1).
Exercise 1.5
An office manager who oversees two sites wants to determine if providing a free lunch on
Friday to employees who are on time for the entire week affects lateness.
a. Indicate the groups (e.g., control, treatment).
b. State the research question.
c. State the hypotheses (H0 and H1).
Exercise 1.6
A school principal wants to find out if providing music education has an effect on math scores.
a. Indicate the groups (e.g., control, treatment).
b. State the research question.
c. State the hypotheses (H0 and H1).
Exercise 1.7
A nurse wants to find out if aromatherapy helps to reduce patient stress.
a. Indicate the groups (e.g., control, treatment).
b. State the research question.
c. State the hypotheses (H0 and H1).
Exercise 1.8
A dairy farmer wants to find out if classical music affects how much milk cows produce.
a. Indicate the groups (e.g., control, treatment).
b. State the research question.
c. State the hypotheses (H0 and H1).
18 Part I • Statistical Foundation

Exercise 1.9
A playground counselor wants to find out who’s better at jumping rope: children enrolled in
public school or children enrolled in private school.
a. Indicate the groups (e.g., control, treatment).
b. State the research question.
c. State the hypotheses (H0 and H1).

Exercise 1.10
A manager of a customer support call center that spans two separate rooms wants to see if
running classic cartoons on a big screen with the sound muted affects employee morale.
a. Indicate the groups (e.g., control, treatment).
b. State the research question.
c. State the hypotheses (H0 and H1).
 2 SAMPLING

You don’t need all the data to run quality statistics; just get a sample.
• Rationale for Sampling

• Sampling Terminology

• Representative Sample

• Probability Sampling

• Nonprobability Sampling

• Sampling Bias

• Optimal Sample Size

LEARNING OBJECTIVES

Upon completing this chapter, you will be able to:

2.1 Comprehend the rationale and advantages of sampling: time, cost, feasibility,
and extrapolation
2.2 Understand three-tier sampling terminology: population, sample frame, and
sample
2.3 Derive a representative sample to reduce threats to external validity
2.4 Understand the utility of probability sampling options
2.5 Understand the utility and constraints of nonprobability sampling options
2.6 Identify factors that could potentially bias a sample and techniques to reduce
such bias
2.7 Consider an appropriate sample size

19
20 Part I • Statistical Foundation

OVERVIEW—SAMPLING

Think about a simple statistic that you already know, like the average, where you total the
numbers and then divide by the amount of numbers that you have. For example, suppose
you had the following test scores: 76, 99, and 86. The total is 261. Since there are three
numbers, we’d then divide that total by 3: 261 ÷ 3 = 87; hence, we can say that the aver-
age is 87. It was fairly simple to process these data, consisting of three figures to derive the
average, but what if there were no data? Clearly, this would mean that we’d have nothing to
process; without data, there’s nothing to average, so the question naturally emerges: Where
do all these data come from? In cases where there’s a relatively small and manageable group of
data (e.g., What’s the average test score from a single class?), it’s possible to collect and process
the data on everyone. Other times, the group that we’re interested in is so large that it’s
simply not feasible to gather all of the data to answer our statistical question (e.g., What’s the
average income of everyone living in a city?). If you don’t have access to the income data for
every person in that city in an existing file, the population of that city may be too big for
you or your research team to gather data from everyone so that you could calculate the aver-
age income. In this case, feasibility is an issue; instead of trying to gather data on everyone,
we can gather a sample and then compute our statistics (average) based on that sample. This
process will give us some insights that we didn’t have before regarding the average income
of people in that city. If we do things right, the data gathered from this sample will admit-
tedly not include data from everyone in the city, but it can provide a viable estimate that
we can work with. If we use an appropriate sampling method to collect income data from
considerably less than half of the people, we’ll know more about the income level of those
living in that city than we did before acquiring the sample.
You’re likely already familiar with the notion of sampling. Perhaps you’ve tasted a sam-
ple of cheese at a store, and based on that small amount, you made a decision to buy or not
buy some quantity of it. At some point, you’ve likely had a blood sample taken consisting
of a small amount of blood, which can be sent to a lab for analysis. The findings from that
small sample can be used to understand the status of the rest of your blood (which was not
sent to the lab); further, that small amount of blood can suggest a diagnosis and treatment
strategy. You probably noticed that they didn’t draw your entire blood supply and send that
to the lab for analysis—that would be painful, time-consuming, expensive, and lethal.
The notion is that often in statistics, acquiring all of the data is not always a reasonable or
necessary goal. In statistics, we often make due with a sample, and in selected cases, we can
use what we learn from the analysis of the sample to guide us in understanding the larger
population that it was drawn from; this is known as external validity—taking what you’ve
statistically learned from the sample and plausibly generalizing it to the larger population
that it was drawn from. For example, if we’ve conducted our sample properly, the average
income, derived from a small sample, may reasonably represent the average income of all of
those employed in that community.
 Chapter 2 • Sampling 21

SAMPLING RATIONALE

The process of relying on data from a sample rather than the entire population is rooted in the
practical constraints of collecting quantities of data in the real world, specifically: time, cost,
feasibility, availability, and extrapolation.

Time
Time is a potent resource. We cannot change the pace of time, and nobody can store time to
use it later. Nor can we manufacture more time; all we can do is use it. Some processes are
serial, such as conducting research, wherein we (1) derive a research question and correspond-
ing hypotheses, (2) gather pertinent data, (3) compute statistical analyses on the data, (4) doc-
ument the results, and (5) derive a plan of action based on the results of the data. Considering
that these processes need to be carried out in order, if Step 2 involves gathering data from a
sizable population (e.g., 500,000 people), this could substantially slow the process, whereas
strategically gathering data from a sample would shorten the data collection process, enabling
us to gather data and act upon the findings more rapidly. Further, some decisions are time
dependent; you or others may need to make a decision within a specified time frame (e.g., by
the end of the week). Hence, there may simply not be enough time to gather a vast amount
of data, particularly if the data-gathering method involves administering lengthy interviews.

Cost
Aside from setup costs (e.g., developing a survey, acquiring computers, training staff, online
survey subscription fee, participant recruitment advertising expenses), there may be costs
associated with gathering data. This could involve paying trained interviewers to confer with
participants, participant compensation fees, photocopying questionnaires, postage, data
entry, and other per-participant expenses. Budgetary constraints often necessitate sampling as
opposed to attempting to gather data from the larger population.

Feasibility
The process of gathering data has challenges. Referring back to the example of the blood sam-
ple, it’d be deadly to draw a person’s entire blood volume for testing, plus it’s hard to conceive
of a laboratory equipped to conduct such testing. Consider another example wherein a school
board wants to have the parents of each student in a school district complete an online survey.
Some households may not have access to a computer with online capabilities. Also, if the
survey is not available in multiple languages, this may create language barriers. Those who
do have online access may opt to complete only part(s) of the survey, while others may simply
choose not to respond. If the research involves in-person encounters, geographical distance
may preclude remote regions.
22 Part I • Statistical Foundation

Extrapolation
Gathering data meaningfully can provide a sufficient representation of the population,
enabling us to analyze the sample that we acquire and reasonably assert that we now have a
better understanding of the population that it was drawn from—what we now know about
our small sample reasonably pertains to the larger population as well. Plausibly generalizing
the results of our sample to characterize the larger population is referred to as external validity.
Other times, the goal is not to extrapolate our findings from the sample to comprehend the
larger population—sometimes the goal is merely to understand those involved in our sample
or others who are similar to the unique characteristics of those in the sample that we analyzed;
this will be covered in the section on nonprobability sampling.

SAMPLING TERMINOLOGY

The process of deriving a sample involves a three-step process, wherein we move from the
largest realm (the population) to the smallest (the sample).

Population
The population refers to every person or data item in a specific domain. In terms of people, the
population of a country would include all of the people living in that country. On a smaller
scale, the population of an organization would be all of the members of that organization.
Regarding data items, the population would refer to all of the academic records for every stu-
dent at a school or all of the medical records of patients at a hospital. You’ve probably surmised
that when it comes to a population, the operative word is all; hence, the population is consid-
ered the largest potential pool of data in the data collection process. For example, suppose
20,000 students are enrolled in a college; the population is 20,000.

Sample Frame
The sample frame is a step down from the population. Whereas the population entails all of
the people or data records within a specified domain, the sample frame is a subset of the popu-
lation that you could potentially access. For example, suppose you want to email a survey to
the students enrolled at a college, and you locate a website that includes the email addresses
of the students, but students can freely configure their online profile: They can opt to have
their email listed on this public website or not. Suppose 8,000 of the 20,000 students in the
college population opt to have their email address posted on the website—this subset of 8,000
students that you could reach out to is the sample frame.

Sample
Whereas the sample frame constitutes the list of people (or data) that you could draw from, the
sample frame may be too large. Hence, we take one final step down and select a subset from
 Chapter 2 • Sampling 23

the sample frame; this constitutes the actual sample that we will work with. Continuing with
our university example, the sample frame consists of a pool of 8,000 students that we could
potentially contact. Suppose this research involves conducting a 30-minute individual inter-
view with each participant, and we provide $10 compensation for their time. If you gathered
data on all 8,000 students in the sample frame, this would require a budget of $80,000 just
for participant fees and a total of 4,000 hours of interviewing time. If you conducted back-
to-back interviews for 8 hours per day, it would take 500 days (more than a year) to gather
the data. Such a sizable project may not be plausible in terms of time, cost, and feasibility.
Alternatively, you could select a sample of 200 students from the sample frame of 8,000. This
would reduce the participant compensation fee to $2,000 and the data collection time to a
total of 100 hours (12.5 days). These 200 students would constitute the sample.
The population, sample frame, and sample involve a system of subsets (see Figure 2.1):

• The population includes all people or data records in a specified domain. In this
example, the population is the 20,000 students who are enrolled at a college.
• The sample frame is the portion of the population that you could potentially access.
In this example, the sample frame is 8,000, which is 40% of the population (8,000 ÷
20,000).
• The sample is the portion of the sample frame that you include in your research. In
this example, the sample is 200, which is 2.5% of the sample frame (200 ÷ 8,000) and
only 1% (200 ÷ 20,000) of the population.

FIGURE 2.1 ■ Three Sampling Tiers: Population, Sample Frame, and Sample

Population
(n = 20,000)

Sample Frame
(n = 8,000)

Sample
(n = 200)
24 Part I • Statistical Foundation

REPRESENTATIVE SAMPLE

You’ve probably already encountered a representative sample. In an ice cream parlor, you may
have requested a sample of a flavor that seemed promising. You’d then receive a very small
amount of the ice cream, and after you taste it, you’d decide if you wanted more of it. You
likely presumed that the small amount that you tasted was a representative sample, meaning
that you’d reasonably expect that the rest of the ice cream in the big container was the same
flavor, color, texture, and temperature as the small sample that you tasted.
Often, researchers make a deliberate effort to gather a representative sample, which can
facilitate external validity, meaning that what we (statistically) learn about the sample is reason-
ably generalizable to the larger population that the sample was drawn from. Probability sampling
methods have the potential for external validity. Alternatively, nonprobability sampling enables
the researcher to focus only on those in the sample, with no intention to generalize those find-
ings to the larger population, meaning that there would be no potential for external validity.
We’ll look at design options for probability sampling and nonprobability sampling.

PROBABILITY SAMPLING

Probability sampling is an equal-opportunity sampling technique, wherein every person or

data record has an equal chance of being selected for the sample. If a representative sample is
gathered, the results have the potential for external validity, meaning that what we know
about those in the sample can plausibly be generalized to comprehend the larger population
that the sample was drawn from. We’ll explore methods for probability sampling: simple
random sampling, systematic sampling, proportionate and disproportionate stratified sampling,
and area sampling.

Simple Random Sampling

The most fundamental form of sampling is simple random sampling, sometimes referred to
as SRS. This method involves randomly selecting people or data records. For a small group,
we could use a low-tech random selection method (e.g., coin flip, drawing names or num-
bers out of a hat); for larger groups, it would likely be more efficient to use a program with a
random-number generator to guide your selections. For example, suppose we had a sample
frame of 60 people to select from and we wanted to randomly choose 10 for our sample.
We could assign a number (1 . . . 60) to each person and then use one of these methods to
gather 10 numbers that would constitute our sample: 32, 20, 55, 12, 14, 51, 19, 25, 17, and 43
(Figure 2.2). If anyone in the selected sample opts out, then we could use the same random
selection process to replace them.

Stratified Sampling
In the prior example, the goal was simply to randomly select 10 people from a group
of 60, which produced 8 women and 2 men, but suppose you needed that sample to be
 Chapter 2 • Sampling 25

FIGURE 2.2 ■ Simple Random Sampling: The Researcher Randomly Selects 10 of

the 60 People

gender-balanced (5 women and 5 men); you could use a stratified sampling method. Instead
of everyone in the sample frame being in one group and then making random selections, we
could stratify this group based on gender, meaning that we would create two strata (think of
a stratum as a subset): one stratum for the women and another for the men. Since our goal is to
gather a sample of 10, we could then randomly select 5 from the women stratum and 5 from
the male stratum (Figure 2.3).

FIGURE 2.3 ■ Stratified Sampling: Split the Sample Frame Into Two (or More)
Strata, Then Randomly Select Five From Each Strata
Strata 1: Females Strata 2: Males
26 Part I • Statistical Foundation

Proportionate and Disproportionate Stratified Sampling

When working with stratified sampling, you may opt for proportionate stratified sampling
or disproportionate stratified sampling. Consider a different distribution that has a total of
40 people: 30 in the women stratum and 10 in the men stratum. We could gather a propor-
tionate sample, wherein we would select 10% from each stratum, meaning that we’d take the
same proportion (10%) from each group. The women stratum has 30 people; 10% of 30 is 3,
and hence, we’d randomly select 3 people from the women stratum. Next, we’d gather the
same proportion from the men stratum, which contains 10 men; 10% of 10 is 1, and hence,
we’d randomly select 1 person from the men stratum (Figure 2.4).

FIGURE 2.4 ■ Using Proportionate Stratified Sampling to Gather a 10% Sample

(From Each Stratum) Renders Three Women and One Man
Strata 1: Females Strata 2: Males
Sample = 3 (10%) Sample = 1 (10%)

In the prior example, notice that stratified sampling produced only one man (10% of
10 = 1). A technique for increasing the sample size when faced with one or more small strata is
to use disproportionate sampling, where instead of choosing a proportion (percentage) to select
from each stratum, we choose a set number of individuals (or data records) to select from each
stratum. For example, instead of randomly selecting 10% from each stratum, we may specify
that we’ll randomly select three from each stratum (Figure 2.5).
Although the number of people selected from each stratum is now the same (three women
and three men), this is disproportionate sampling since the proportion (percentage) selected
from each group is different. Randomly selecting 3 out of the 30 women constitutes a 10%
sample from that stratum, whereas randomly selecting 3 out of the 10 men is a 30% sample
from that stratum.

Systematic Sampling
You may have noticed that when using random sampling, it’s possible that the selections may
involve people (or data records) that are near each other or far apart. Sometimes you may want
 Chapter 2 • Sampling 27

FIGURE 2.5 ■ Using Disproportionate Stratified Sampling to Gather Three (From

Each Stratum) Renders Three (10%) Women and Three (30%) Men
Strata 1: Females Strata 2: Males
Sample = 3 (10%) Sample = 3 (30%)

to gather data in a random fashion, but you’d like the selections to be evenly spaced; system-
atic sampling provides a method for doing that. For example, suppose you’d like to select 15 of
the 60 people who will be attending a meeting. First, divide the number of people in the sam-
ple frame (60) by the target sample size (15); this will produce the “k” or skip term (k = 60 ÷ 15,
which equals 4). Next, identify the start point, which will be a random number between 1
and k (in this case, k = 4); suppose that the random start point is 3. This means that the first
selection for your sample would be the 3rd person who enters the meeting, along with every
4th person after that, so the 7th person who enters the meeting would be the second person in
your sample, then the 11th person, and so on. Based on this method, you’d get an even sample
of the meeting attendees in terms of their arrival times—early, on time, and late. This method
can also be used with lists of data records (Figure 2.6).

Area Sampling
Area sampling, also known as cluster sampling or multistage cluster sampling, is useful for
gathering data spanning a geographical region or when it’s not possible to attain a sample
frame. Considering that the characteristics of neighborhoods and residents can vary, area
sampling provides a method for gathering a (more) representative sample of the area of inter-
est. In this example, you want to gather data from 30 residents of Smalltown, consisting of 15
blocks. First, we’d number the blocks (Block 1, Block 2, Block 3 . . . Block 15). Since our goal
is to survey 30 households, and there are 15 blocks, simple arithmetic tells us that we should
randomly select and survey 2 households per block (30 samples ÷ 15 blocks = 2 samples per
block) (Figure 2.7).
This example presumes that there are an equal number of residences and people living on
each block. In reality, there may be substantial differences; for example, suppose this 15-block
neighborhood consists primarily of homes that are all about the same size, except Blocks 6,
7, 11, and 12 mostly have apartments with 8 to 12 units per building, and there is a high-rise
condominium on Block 9 with 100 units. In such cases, it would be appropriate to adjust the
28 Part I • Statistical Foundation

FIGURE 2.6 ■ Systematic Sampling Provides Periodic Selection From the Sample
Frame Beginning With a Randomly Derived Start Point

I’m #3, Skip every 4

start with me. to get some more.

Note. The skip term is k (k = sample frame ÷ target sample size), and the start point is a random number between
1 and k.

FIGURE 2.7 ■ Area Sampling: Identify Target Sample Size and Divide by Number
of Blocks to Derive Number of Samples to Randomly Select From
Each Block

Block Block Block Block Block

1 2 3 4 5

Block Block Block Block Block

6 7 8 9 10

Block Block Block Block Block

11 12 13 14 15

Note. Target sample size = 30 households; number of blocks = 15 blocks; number of samples to randomly select
from each block: 30 samples ÷ 15 blocks = 2 samples per block.
 Chapter 2 • Sampling 29

number of samples gathered from each block based on the population density of the block—
the more people per block, the larger the sample on that block.

NONPROBABILITY SAMPLING

Whereas the hallmark of probability sampling is that each person/data item has an equal chance
of being included in the sample, this is not the case when using nonprobability sampling meth-
ods. Sometimes, the sample frame is not readily accessible or may not exist (e.g., a list of all
left-handed people, a list of all undocumented immigrants, a list of all part-time students who
work full-time and provide care for a family member), which confounds the probability sam-
pling process. Since not everyone has an equal opportunity to be selected for the sample, it fol-
lows that the data gathered via nonprobability sampling are not expected to be a representative
sample. Further, since the sample does not proportionally represent the characteristics of the
larger population, research conducted via nonprobability sampling does not have external valid-
ity; what we learn about the sample cannot plausibly be generalized to our understanding of the
larger population.
Initially, it may seem as if nonprobability sampling would produce useless results since it
lacks external validity, but the goal of a study may be to gain a focused understanding of a
specific set of individuals, and the larger population is not the primary concern. This will
become clearer as we explore examples of various forms of nonprobability sampling techniques:
convenience sampling, purposive sampling, quota sampling, and snowball sampling.

Convenience Sampling
Convenience sampling, also known as availability sampling, is essentially what it sounds like:
The researcher gathers the sample from people or data that’s readily available (see Figure 2.8).
For example, suppose you wanted to investigate how many siblings people have; you could stand
at a busy ATM and survey the people who are waiting in line. Even if you were to gather a siz-
able sample, considering that not everyone in the community accesses this particular ATM, the
results would lack external validity, but this sample could still have value; the bank may be plan-
ning a promotional campaign directed at families who utilize their ATMs.

FIGURE 2.8 ■ Convenience/Availability Sampling Involves Recruiting Readily

Accessible Individuals or Data

Acme ATM

100
30 Part I • Statistical Foundation

Purposive Sampling
Purposive sampling is used when you’re interested in gathering information from a (very)
specific portion of the population that is considered low prevalence in the population. In
purposive sampling, the potential participants must meet one or several criteria. For example,
suppose there’s a new therapy that’s been designed to ease the symptoms of patients with can-
cer undergoing radiation treatments. Fortunately, most of the people in the population do not
meet these criteria, meaning that randomly selecting individuals from the population would
be an inefficient sampling method. Potential participants must meet all the following criteria:

• 18 to 65 years old
• Cancer diagnosis
• Scheduled for 5 to 10 radiation therapy treatments
• Willing to take an experimental drug or placebo
• Not using any unprescribed substances

The researcher may need to advertise for suitable participants or coordinate with appro-
priate health care providers or institutions to recruit a sufficient sample. Once again, we see
that those who would be in the sample for this study are unlike the people in the population;
hence, external validity is not plausible. Given the goal of this study (to ease the symptoms
of patients with cancer undergoing radiation treatments), this should serve to exemplify the
value of nonprobability sampling—sometimes the goal of the research is to gain an under-
standing of a selected portion of the population and not the overall population. Although tech-
nically the results would not have external validity, one could plausibly propose that what was
learned from studying this group of individuals may be applicable to others who meet these
same criteria (e.g., additional patients with cancer who are undergoing radiation treatment).

Quota Sampling
Quota sampling involves determining how many people you want to sample (e.g., 100 peo-
ple); the sampling process continues until the quota (of 100 people) is reached. Quota sam-
pling can be useful when a sample is needed promptly. For example, suppose you want to
sample 50 people as they exit a library, asking what hours they’d like the library to be open.
As soon as the 50th person responds, you stop the data collection process, even if there are
additional people willing to respond to your survey. From there, you can carry out statistical
analysis using the data that you gathered.
Quota sampling can also be used in a stratified data collection process. For example, sup-
pose the library is concerned about accessibility to left- and right-handed people and wants
to gather data from 20 left-handed people and 20 right-handed people; notice that we now
have two strata: left-handed people and right-handed people (Figure 2.9). You could stand
outside the library and ask people who are exiting, “Are you left or right handed?” and then
 Chapter 2 • Sampling 31

ask each person if they’d be willing to respond to a brief survey. As shown in Figure 2.9, you’ve
gathered data from 20 right-handed people and 17 left-handed people. It makes sense that
you’ve reached the 20-person quota for right-handed people first since most people are right-
handed. At that point, if the next person indicates that they are right-handed, you’d courte-
ously inform that person that you are currently only gathering data from left-handed people
(the right-handed part of this survey is now closed). You’d continue that process until you
recruit left-handed participants 18, 19, and 20. Also, the quotas don’t necessarily need to be
equal for each stratum; for example, our quotas might be to gather data from 15 left-handed
people and 40 right-handed people. In quota sampling, the rule is that you stop collecting data
(on a stratum) once the predetermined quota is met.

FIGURE 2.9 ■ Quota Sampling Specifies the Number of Participants Sought (in
Each Stratum)
Left-Handed Right-Handed

18 19 20

Snowball Sampling
Snowball sampling is based on the process for creating an actual snowball: Start by gathering
a handful of snow, then pack on more snow (that’s just like the snow that you initially picked
up), and continue packing on more and more snow until the snowball reaches the desired size.
When it comes to people, consider the following saying: Birds of a feather flock together. Since
people may know other similar people, snowball sampling can be useful when gathering a sam-
ple of people with characteristics that are uncommon or not readily identifiable; for example,
a musician may know other musicians, and a single parent may know other single parents,
and hence, referrals to others may be possible, which is precisely how snowball sampling works.
Suppose you’re interested in understanding how many hours per week dyslexic students
spend on their homework. Since dyslexia is a condition that doesn’t present with any overt
signs or symptoms, it could take some time to ask numerous passersby if they’re dyslexic
before you find your first potential participant for your sample. Referring to Figure 2.10, once
you find someone who is dyslexic and willing to participate in your study (Person 1 in the
figure), before dismissing her, you ask if she knows of anyone else who’s dyslexic who might
be willing to partake in your study. This first participant may contact her friend who’s also
dyslexic, and this could be your second participant. Upon completing your encounter with
this second person, you ask if she knows of anyone else who’s dyslexic, and she refers you to
32 Part I • Statistical Foundation

FIGURE 2.10 ■ Snowball Sampling: Requesting Referrals From Those Who Meet
the Research Criteria in Order to Recruit Others Who Meet the
Research Criteria

4 6

1 2 3

5 7 8

her brother (Person 3), who also chooses to partake in the study, and he provides referrals to
two more people (Persons 4 and 5), and so on.
Even if the participant characteristic that you are seeking is readily observable (e.g., any-
one who uses a wheelchair), considering that this is generally a low-prevalence condition,
snowball sampling may still be a valuable technique; a person who uses a wheelchair may know
of others who also use a wheelchair. Further, someone who uses a wheelchair may provide you
with other valuable referrals that may route you to places where people who use wheelchairs
can be found (e.g., rehabilitation center, wheelchair store/repair shop, hang-out areas, web-
sites, organizations).
Another possibility is that you’re interested in gathering a sample from an invisible or a hidden
population. Such individuals aren’t really transparent or necessarily in hiding, but they may pos-
sess a particular characteristic that you’d like to recruit in your sample, but unlike people who use
a wheelchair, the characteristic that you’re interested in is not readily observable, such as individu-
als who are a single parent, bisexual, fearful of thunderstorms (astraphobia), or chess enthusiasts.
Alternatively, some individuals may actively conceal their attributes, fearing consequences;
this may include people who are involved in illegal activities or individuals with potentially
embarrassing or uncommon characteristics such as a peculiar obsession, fetish, unpopular
belief system, or stigmatized condition or disease. If you’re fortunate enough to encounter
such a person, instead of asking for names and contact information of others, you may request
that participants pass your contact information along to other suitable individuals, enabling
those people the option to (anonymously or confidentially) reach out to you. When people
feel confident that disclosing sensitive information won’t cost them consequences, they may
provide honest disclosures of carefully guarded truths. To facilitate this process, keep an open
mind, a nonjudgmental attitude, and a professionally positive demeanor. Genuine respect for
your (potential) participants can plausibly convey that you are not a threat, which may gain
you valuable cooperation, additional participants, and (more) truthful information.
 Chapter 2 • Sampling 33

SAMPLING BIAS

Sampling bias can occur if individuals with a particular characteristic (e.g., high intelligence,
low socioeconomic status, youngest, tallest) are disproportionately recruited to partake in a
study. Consider the various ways that sampling bias could intentionally or unintentionally
occur:

Self-Selection Bias/Voluntary Response Bias—People with a unique characteristic

may be more likely to choose to partake in a study than others. For example, academi-
cally high-performing high school students may be more motivated to respond to a survey
focusing on college plans.
Nonresponse Bias—People may opt not to partake in a study, choose not to respond to
selected questions, or drop out of a study. For example, someone with a criminal past may
resist responding to topics involving illegal activities.
Undercoverage Bias—There may be too few or no individuals in the sample with a par-
ticular characteristic. For example, in a study involving handedness (left/right-handed), it
may be difficult to find (enough) individuals who are ambidextrous.
Advertising Bias—Depending on where and how participants are recruited can have an
impact on the results. For example, placing recruitment bulletins on public transit (only)
would exclude individuals who use other forms of transportation.
Time Bias—Responses can be affected by when a sample is attained. For example, sup-
pose a researcher wants to gather opinions regarding the community center in a public
park. Surveying people in the park between 2:00 and 4:00 PM would likely produce
very different results compared to data gathered from people who are in that same park
between 2:00 and 4:00 AM.
Survivorship Bias—Recruiting only people who meet a certain criterion can be prob-
lematic. For example, administering a customer satisfaction survey to people currently
using a bank and excluding those who opted to close their account(s) and take their busi-
ness elsewhere would likely produce incomplete findings.
Recall Bias—Considering that memory is imperfect, asking questions that involve
remote events or obscure information may produce erroneous results. For example, ask-
ing participants how many times they took aspirin in the last year is likely to result in an
estimate rather than an accurate number.
Exclusion Bias—Excluding selected individuals from participating can skew results. For
example, administering an online survey rules out people who do not have Internet access
and those who are not tech-savvy.
34 Part I • Statistical Foundation

Sampling bias may resonate through the research process all the way to the statisti-
cal results and possibly beyond if the findings are published. When the goal is to assess a
specialized portion of the population via a form of nonprobability sampling or compare
two different parts of the population, it’s important to make this clear throughout your
documentation. It would be appropriate to disclose such issues in terms of the limitations
of such a study and explicitly discuss how one or more forms of sampling bias may have
adversely impacted external validity.

OPTIMAL SAMPLE SIZE

You may be wondering: What’s the right sample size? The answer to this seemingly simple
question involves multiple factors, including the research design, the number of groups,
the type of variables, the type of statistical analysis, and the desired robustness (power)
of your findings. Researchers strive to attain an optimal sample size—not too few and not
too many.
If a researcher gathers data on too few participants, then the results may be found to
be underpowered, meaning that the sample was too small to produce robust/stable statistical
results; this could adversely impact the solidity of the findings. For example, if you surveyed
one person to determine if voting by mail is good or bad, clearly the sample (n = 1) is too small
to have confidence in the outcome. Imagine the results: “Our research revealed that 100% of
those surveyed favored (or opposed) mail-in ballots.” An underpowered study could be mis-
leading and may compromise the credibility of those involved.
Conversely, if a sample is too large (e.g., n = 1,000,000), the researcher has likely wasted
time and money, delayed the results, and potentially delayed action(s) that would be taken
based on the results.
Researchers can use power calculations to determine the proper sample size. Generally,
statistical power between 0.7 and 0.8 indicates a viable sample size. It’s possible to compute
power calculations at various points in the research process:

(1) Before starting a research project, power calculations can provide an estimate of how
many participants you’ll need to sample.
(2) During the research project, it’s appropriate to periodically compute the power
to determine if you have (already) acquired a sufficient sample size. For example,
suppose you are authorized to gather data on up to 300 people; after gathering data
on the 220th person, you find that you have attained a power of 0.8. Having achieved
sufficient power, you may opt to conclude the data collection process.
(3) At the conclusion of the sampling process, it is appropriate to run a final power
calculation to determine the level of power achieved in this study.
 Chapter 2 • Sampling 35

GOOD COMMON SENSE

The acronym GIGO (Garbage In, Garbage Out) is likely as old as computing. It implies that if
you input erroneous data into a computer, the computer will rapidly and meticulously process
the data per the instructions of the program, but you can expect the output to be erroneous
too. This would be like cooking with spoiled ingredients; even if you used proper cooking
techniques, you’d reasonably expect a bad meal to emerge. One can think of sampling as
the starting point of the data collection process; hence, the quality of the decisions and tech-
niques pertaining to sampling can have a substantial impact on the quality of the results.
For example, when deploying an online survey, unless comprehensive consideration is
given regarding sampling prior to the launch of the survey, the data collected could be polluted
with a multitude of possibly undetectable anomalies, including such incidents as responses
from people who do not meet the criteria for the study, individuals who revisit the survey mul-
tiple times and provide consistent findings in an effort to skew the results, (repeated) bogus/
random responses, or responses from automated software (e.g., bots). Clearly, a sample that’s
biased or not critically controlled can produce misleading results, prompting the researchers
to take potentially inappropriate (in)actions.

KEY CONCEPTS

• Rationale for sampling

• Time
• Cost
• Feasibility
• Extrapolation
• Population
• Sample frame
• Sample
• Representative sample
• External validity
• Probability sampling
• Simple random sampling
• Stratified sampling
• Proportional sampling
• Disproportional sampling
• Systematic sampling
• Area sampling
36 Part I • Statistical Foundation

• Nonprobability sampling
• Availability sampling
• Purposive sampling
• Quota sampling
• Snowball sampling
• Sampling bias
• Optimal sample size

PRACTICE EXERCISES

Exercise 2.1
Define the following terms and provide an example for each:
a. Population
b. Sample frame
c. Sample
d. Representative sample
e. Sampling bias

Exercise 2.2
Explain the difference(s) between a probability sample and a nonprobability sample.

Exercise 2.3
A community is considering turning a portion of a public park into a dog park. The city
planning commission has selected you to survey community members to gather opinions
about the project. Explain how you would use simple random sampling.

Exercise 2.4
An online video service has commissioned you to conduct a customer satisfaction survey;
they provide you with a list of 100,000 subscribers containing their name, cell phone num-
ber, and email address. Explain how you would gather surveys from 200 subscribers using
systematic sampling.

Exercise 2.5
At a community forum with 95 attendees, a community organizer wants to recruit 5 peo-
ple who have children and 5 people who do not have children to partake in a focus group.
Explain how you would recruit these 10 people using stratified sampling.

Exercise 2.6
Prior to building a new store in Anytown, Acme Corporation wants to conduct a survey of
240 local households. They provide you with a list of all of the addresses for each of the 80
 Chapter 2 • Sampling 37

blocks of Anytown. You are also informed that the blocks are evenly populated. Explain how
you would use area sampling.

Exercise 2.7
The board of directors of a mall wants to know how much money people spend in the food
court. Explain how you would use availability sampling.

Exercise 2.8
A tutoring service wants you to survey students who have a learning disability. Explain how
you would use snowball sampling.

Exercise 2.9
Acme Hospital has selected you to conduct a satisfaction survey of recently discharged
patients; they provide you with a list of patients who agreed to be contacted after their hos-
pitalization. You need to gather data from 50 minors (5 to 17 years old) and 100 adults (18
years or older). Explain how you would use quota sampling.

Exercise 2.10
A new afterschool program is starting up at Anytown Community Center, providing free
recreation and life skills classes (e.g., cooking, first aid, music). Participants must live in
Anytown, be between 7 and 17 years old, and be able to attend 2 days a week from 4:00–5:30
PM. Explain how you would use purposive sampling.
 3 GETTING STARTED IN R

Let’s get you going in R.

• R Overview
• Accessing RStudio
• R Syntax Guide
• Loading Packages
• Loading and Importing Files
• First Statistical Run
• Shortcuts
• Processing Your Own Data

LEARNING OBJECTIVES

Upon completing this chapter, you will be able to:

3.1 Set up your free account on the R Cloud
3.2 Efficiently utilize the R Syntax Guide to execute R commands
3.3 Load R packages
3.4 Comprehend dataset structures
3.5 Interpret dataset codebooks
3.6 Upload (import) datasets into R
3.7 Recognize data file types and file naming
3.8 Understand variable name references
3.9 Run a statistical analysis
3.10 Export R results
3.11 Use convenient shortcuts in R

39
40 Part I • Statistical Foundation

3.12 Calculate math in R

3.13 Process your own data in R
3.14 Log off of R

OVERVIEW—R AND RSTUDIO

We’ll start with a (very) short history of R and RStudio: R was first released in 1993 as an
open-source free statistical software system for public use with a traditional command line
interface, wherein the user types in programming instructions (similar to typing in com-
mands in DOS), and R would run the code and display the results.
RStudio, which we’ll be using, was released in 2016. RStudio is a third-party graphical
user interface (GUI) that’s linked with R, creating an integrated development environment
(IDE). RStudio provides a more user-friendly front-end to R and includes features like conve-
nient pull-down menus and checkboxes.
Unlike other popular statistical software (e.g., SAS, SPSS, Stata), R and RStudio are free.
At the time of this publication, the default membership level is Free, which provides 25 hours
a month of RStudio online access at no cost. The next step up is Cloud Plus, which provides 75
hours per month for $5.00 per month, with 10 cents per hour for additional hours. Considering
that hours, prices, and services are subject to change, it would be best to refer to the R Cloud
website for specific details on a variety of R Cloud access plans ([Link]
You can also download R and RStudio to run on Windows, Macintosh, and a variety of other
computing environments/operating systems for free. Using R and RStudio on your computer (not
the RStudio Cloud service) provides unlimited hours and free downloadable updates. The programs
are available at [Link] sure to select the software
that’s right for your computer (e.g., Windows, Macintosh). First, download and install R; next,
download and install RStudio; and finally, run RStudio. This software will run efficiently on your
computer, and it will look and operate just like the RStudio Cloud version with minor exceptions.
Occasionally, upon starting RStudio on your computer, the system may alert you that an update to
the software is available; you’ll be given the option to download and automatically install the new
software. Depending on your connection, the download could take some time, or you can pass on
the upgrade and continue using the existing software as is; the system will remind you about the
upgrade the next time you access the software, or you can click on Help/Check for Updates.
Considering how easy it is to set up the RStudio Cloud software, combined with its uni-
versality, meaning it can be used on any computer (Windows or Macintosh) that is equipped
with a browser and Internet access, this book uses the RStudio Cloud version of this system.
The RStudio Cloud also enables portability, meaning that you can access and process your
statistical data from any computer that has Internet access.
 Chapter 3 • Getting Started in R 41

VIDEO RESOURCE

The video for this chapter, Ch03 – Getting Started in R.mp4, provides step-by-step instruc-
tions demonstrating how to establish your R Cloud account, configure it, load files, and use
the R Syntax Guide (R Syntax [Link]) to run some basic analyses.

SETTING UP YOUR RSTUDIO CLOUD ACCOUNT

To setup your free RStudio Cloud account, follow these steps:

1. Go to the RStudio Cloud website: [Link]

2. On the R homepage, click on Sign Up or GET STARTED.
3. On the Plans page, click on Free.
4. On the Sign Up for Cloud Free page, click on Sign Up.
5. On the next Sign Up page, enter your Email, Password (create a password for
your new R Cloud account), your First name, and your Last name, and click on
Sign Up.
6. On the Verify Your Email screen, click on Continue.
7. Check your email and read the message from Posit Cloud (Subject: Please verify your
email address); click on Verify your email.
8. Return to R Cloud, and on the Verify Your Email screen, click on Continue.
9. Click on New Project.
10. Click on New RStudio Project (R will display Deploying Project while the processor
proceeds with your initial setup).
11. Click on Untitled Project and type in a name for your RStudio work (e.g., Stats Course
1) and press Enter.
12. To end a session, click the circle in the upper right corner with your initials or your
name.
13. Click on Log Out (NOTE: Be sure to Log Out when you’re done; this will
conserve your free 25 hours per month. Also, the R Cloud is set up to go
into “sleep” mode if you’re idle for about 15 minutes, which helps to conserve
your online time.)
42  Part I  •  Statistical Foundation

Fortunately, you only need to establish your R Cloud account (Steps 1 through 13) once.
To return to login to your R Cloud account, follow these steps:

1. Go to the (online) R Cloud: [Link]

2. Click on Log In or ALREADY A USER? LOG IN, and when prompted, enter your
Email and Password.
3. Click on the Project that you created (e.g., Statistics Course 1); this will take you into
the R Cloud environment; everything will be exactly as you left it.

R SYNTAX GUIDE

Whereas other statistical software involves navigating through multiple menus to specify the
statistical processing parameters, R uses lines of code, but don’t panic—you don’t need to
learn a whole programming language to run statistics in R. The downloadable R Syntax
Guide on the companion website contains every line of R code that you’ll need to run all of
the statistical tests covered in this book. Notice that some of the text in this document is not
only set in a different font and boldface but also blue1—that’s the R code that you’ll copy (Ctrl
C) from the R Syntax Guide and paste (Ctrl V) into the R Console window. Once the code is
in the R Console window, you’ll edit the words that are underlined in the R Syntax Guide (e.g.,
FileName, VariableName) and then press Enter, which will instantly produce results in the
form of numeric reports and plots (graphs).
Notice that anything that you type into the R Console window will be in blue; for clarity,
the R Syntax Guide uses the same shade of blue, so you’ll always know exactly what lines of
text to copy and paste.
As you’ll see, the R Syntax Guide is an integral part of this book; it is recommended that you
have the R Syntax Guide document open each time you access R. The R Syntax Guide is provided
as a regular Word document, not a PDF, enabling you to customize it to best suit your needs.

LOADING PACKAGES

R consists of the base system along with supplemental Packages, which are software add-ons
that provide further functions to R. As part of the setup process, we need to have R load some
additional Packages (yes, all of the Packages are free).

1. Go to page 1 of the R Syntax Guide and copy the names of the packages: car
gmodels Hmisc mvnormtest psych stringi (highlight all of this blue text and
press Ctrl C to Copy).
2. In R, click on Packages.
3. Click on Install.

1
If you are reading this present book via ebook or screen reader, the “blue” text that you will copy and paste into the R
console will be styled in boldface and in a different font than the surrounding text and can be differentiated that way.
 Chapter 3 • Getting Started in R 43

4. Click on the middle text box [Packages (separate multiple with space or comma):].
5. Press Ctrl V to Paste.
6. Click on Install, and you’ll see some red text scroll through the Console window as R
downloads and installs the specified Packages; this usually takes a few minutes.
7. When the installation process ends, scroll through the list of Packages (they’re in
alphabetical order), verifying that each Package (car gmodels Hmisc mvnormtest
psych stringi) is now on the list of Packages. If any of the Packages are missing,
repeat the above procedure, but only (re)install the package(s) that did not load on
your first attempt.

R is configured to present very small numbers, like p values (which we will cover in
Chapter 4: Descriptive Statistics) using scientific notation (e.g., p-value = 1.469e-3), which is
a bit difficult to interpret. We can instruct R to turn off scientific notation and display results
in a more familiar format (e.g., p-value = 0.001469). To turn off scientific notation, use the
options(scipen=999) line of code on page 1 of the R Syntax Guide:

1. Copy options(scipen=999) (highlight this blue text and press Ctrl C to Copy).
2. Put the cursor in the Console window (be sure to click).
3. When you see the flashing vertical cursor bar, press Ctrl V to Paste.
4. Press Enter (you won’t see anything happen).

You’ve now prepared R to process your statistics.

For visual clarity, anything you enter into the R Console window will be in blue text, and the
results that R replies with will be in black text. To help make the R Syntax Guide more intuitively
usable, the lines of R code that you’ll copy and paste into R are the same shade of blue that R uses.
The words that are underlined (e.g., FileName and VariableName) indicate the edits that you’ll
need to make after you’ve pasted the R code into the Console window. Finally, notice that R does
not carry the underlines into the Console window, but you’ll quickly become accustomed to edit-
ing the FileName and VariableName in R. In summary: Whatever’s in blue , that’s what you do.

DATASET STRUCTURE

To save you from having to spend hours of typing, you’ll be processing prepared downloadable
datasets—these are files that contain information arranged in a way that statistical programs
can import and process. These files are provided in csv (comma-separated values) format, mean-
ing that the dataset is a generic text file. The first file that we’ll import is Ch03_Demo.csv,
which contains the following information:

Score,Instructor
99,Jones
44 Part I • Statistical Foundation

86,Jones
77,Smith
82,Smith
77,Jones
91,Smith
78,Jones

Notice that the first line contains the names of the two variables in this dataset, sepa-
rated by a comma: Score,Instructor. As you might expect, a dataset can contain any num-
ber of variables; for example, a dataset may contain five variables: Score, Instructor, Age,
Course, Time.
Now, look at the second line: This line contains the information (data) pertaining to the
two variables for Person 1: This person has a Score of 99 and identifies Instructor as Jones. Notice
that the order of the data on line 2 (first Score, then a comma to separate the variables, followed
by Instructor) matches the order of the variable names on line 1. The next line contains the data
for Person 2 and so on. This dataset contains seven records (not counting the first line, which
contains the variable names). Naturally, a data file (dataset) can contain any number of records
(rows of data).

CODEBOOK

Inspecting a dataset, even when done carefully, can be ambiguous. In this case, Score could
indicate each person’s performance on a spelling test, or maybe Score reflects each person’s
evaluation of a chocolate soufflé (0 = terrible soufflé . . . 100 = perfect soufflé). As for the
second variable, Instructor, for all we know, the respondents may have been asked, “Who was
your kindergarten teacher?” or “Who taught you to bake a soufflé?” The codebook is the key
to resolving such ambiguities as it provides specific details for each variable in the dataset:

• Variable is the name of the variable (on the first row of the dataset).
• Definition provides an explanation of the variable.
• Type indicates if the variable is continuous or categorical and the values contained
within this variable.

Clearly understanding the variables in a dataset enables us to meaningfully select, run,

and document the appropriate statistics.
Here is the codebook for our first dataset, Ch03_Demo.csv, which contains two vari-
ables, Score and Instructor:
Codebook for Ch03_Demo.csv
 Chapter 3 • Getting Started in R 45

Variable: Score
Definition: Score in course (percentage)
Type: Continuous (0 . . . 100)

Variable: Instructor
Definition: Instructor for English 42 - Theory of Bad Poetry
Type: Categorical (Jones, Smith)

UPLOADING A DATASET TO R

You probably already know the term download, meaning that you’re bringing in a copy of a file
from some other computer or file server to reside on your computer. The opposite of a down-
load is an upload, meaning that you’re sending out a copy of a file from your computer to some
other computer or file server. For example, if you receive an email with an attachment, you can
download the attached file onto your computer, and if you send an email with an attachment,
you are uploading the attached file to the recipient.
In this case, you will download the multimedia resources (e.g., datasets) from the com-
panion website, and then you’ll upload the specified dataset(s) to the R Cloud for processing.
During the upload process, if the message below emerges (the phrasing may vary slightly
depending on the type of file you’re uploading), click Yes to enable the process (Figure 3.1).

FIGURE 3.1 ■ Installing Necessary Packages to Import Data

To upload a dataset to R, follow these steps:

1. Click on Files.
2. Click on Upload.
3. Click on Choose File.
4. Navigate to the proper file and click on that file (Ch03_Demo.csv).
5. Click on Open.
46 Part I • Statistical Foundation

6. Click on OK.
7. Click the file name in the lower right window (Ch03_Demo.csv).
8. Select Import Data set . . .
9. R will display a preview of the file.
10. Click on Import.
11. Click on the Environment tab.
12. Click on the file name (Ch03_Demo).

DATA FILE TYPES

The datasets provided with this book are csv files (or comma separated values files), but
R can rapidly and precisely import data from a variety of other popular statistical sources,
including Excel, SAS, SPSS, and Stata. One way to determine the kind of data contained
in a file is to look at the extension of the file name, which is the part that comes after the last
period in the file name. For example, the extension .mp3 (e.g., Big Bass Blues.mp3) signi-
fies that this is an audio file, whereas the extension .mp4 (e.g., Classic Cartoon.mp4) tells
us that this is a video file. R can import a variety of different types of datasets, as shown in
Table 3.1.

TABLE 3.1 ■ Dataset File Types That Are Compatible With R

Dataset Format File Name

Comma-separated values [Link]

Excel [Link]

R [Link]

SAS Experiment.sas7bdat

SPSS [Link]

Stata [Link]

Text [Link]

The three files below all contain the same data as Ch03_Demo.csv, but they are all in dif-
ferent formats (R, Excel, and SPSS); you can import these files into R just for practice:

• Ch03_Demo_RData.RData
• Ch03_Demo_Excel.xlsx
• Ch03_Demo_SPSS.sav
 Chapter 3 • Getting Started in R 47

FIRST STATISTICAL RUN

Now that R is configured and a dataset is loaded, we can run our first statistical analysis.
First, notice the “>” symbol in the Console window, which is next to the blinking bar cursor;
this is the prompt. The prompt indicates that the computer is standing by, ready for you to
type in your command. If the prompt vanishes, press the Escape key (Esc), then press Enter;
occasionally, you may need to do this more than once to get the prompt to reappear.

1. Open the R Syntax Guide (R Syntax [Link]) and go to page 3. At the top, you’ll see:

PACKAGES

☑ ggplot2

☐ Hmisc

☑ mvnormtest

☑ psych

2. Per these instructions, click on the Packages tab and click to check the box for
ggplot2, click to uncheck Hmisc, and click to check psych. Notice that the Packages are
in alphabetical order.
3. Find the segment that says:
Test Run: Descriptive Statistics for a Continuous Variable
Run descriptive statistics for a continuous variable:
describe(FileName$ContinuousVariable)

4. Highlight the blue text and press Ctrl C to copy it.

5. Move the cursor into the Console window in R and click; when you see the bar
cursor flashing, press Ctrl V to paste it. Notice that R pastes the code without the
underlines, which indicates the parameters that you’ll need to edit.

VARIABLE REFERENCES
Before making any edits, you’ll need to know how R addresses variables: Just as your name
can be written as two parts separated by a comma (e.g., Jones,Dusty), each variable name in R
has two parts separated by a dollar sign (e.g., FileName$VariableName).
The first part is the File Name, which is shown in the upper right window; in this case, the
file name is Ch03_Demo (Figure 3.2).
The second part of the (full) variable name is the Variable. In this case, we’ll process the
Score, as shown in the upper left window in R (Figure 3.3).
48 Part I • Statistical Foundation

FIGURE 3.2 ■ The File Name Is Ch03_Demo

FIGURE 3.3 ■ The Variable Name Is Score

Since the file name is Ch03_Demo and the variable that we want to analyze is Score, the
full name of the variable that we will analyze is Ch03_Demo$Score. We can now edit the
R code:

6. In the Console window, remove the FileName and replace it with ch0 —notice that
upon typing the third character, R provides autocomplete; select (or continue typing)
Ch03_Demo.

7. Remove ContinuousVariable and replace it with sco—again, notice that as soon as

you type the third character, R provides autocomplete; select (or continue typing) Score.
8. The line of R code should now look like this: describe(Ch03_Demo$Score), which
tells R to compute descriptive statistics for Score in the file Ch03_Demo (descriptive
statistics are covered thoroughly in the next chapter).
9. Press Enter.

After editing the file name and variable name, the R code and results should look like this:
 Chapter 3 • Getting Started in R 49

> describe(Ch03_Demo$Score)

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 7 84.29 8.32 82 84.29 7.41 77 99 22 0.61 -1.33 3.15

As mentioned earlier, for visual clarity, notice that anything you enter into the R Console
window is blue, and anything that R replies back is black.
To give some meaning to these results, notice that the n is 7; this tells us that the variable
Score contains seven cases (rows of data). Also, the mean is 84.29; this tells us that the average
Score is 84.29 (in statistics, mean is a synonym for average).
When editing the FileName and VariableName, be aware that R is case-sensitive, mean-
ing that uppercase and lowercase letters are relevant. For example, although this line of R
code may look fine: describe(Ch03_Demo$score), it would produce an error because the
proper name of the variable is Score, not score (yes, case matters a lot). Fortunately, the
autocomplete feature can help with precise spelling and uppercase/lowercase. To utilize
the autocomplete feature as efficiently as possible, consider this practice: First remove and
replace the FileName; most of the time, this will enable the autocomplete to assist when
you remove and replace the VariableName. Occasionally, autocomplete will not activate, in
which case, you’ll need to precisely type in the VariableName manually.

EXPORTING RESULTS

If you want to export anything in the Console window, use the regular copy and paste proce-
dure: Highlight the text that you want to copy, press Ctrl C to copy, and then put the cursor in
any application that can accept text and press Ctrl V to paste (or click the paste icon).
If a table is out of alignment after pasting, change the font of the pasted segment to Courier
or Courier New; here’s why this works: To help make text look better, most fonts use propor-
tional spacing, meaning that a character with a thin width (such as “i” or “.”) take up less space
than wider characters (such as “M” or “W”); as such, fonts that use proportional spacing will
produce misaligned tables. Alternatively, the Courier font uses fixed spacing, meaning that all
characters occupy the same width, which renders proper table alignment.
R can also draw graphs. To produce a bar chart for Instructor, showing how many students
had Jones and how many had Smith, follow these steps:

1. Refer to the R Syntax Guide, page 4, starting at the top:

PACKAGES

☑ ggplot2

☑ Hmisc

☐ psych
50 Part I • Statistical Foundation

2. Click on the Packages tab and check ggplot2 and Hmisc, and then uncheck psych.
3. Highlight the blue text and press Ctrl C to copy:
Test Run: Bar Chart for Categorical Variable
Draw a bar chart:
ggplot(FileName,aes(x=factor(CategoricalVariable)))+geom_bar()

4. Move the cursor into the Console window in R and click; when you see the bar cursor
flashing, press Ctrl V to paste it. Notice that it lost the underlines, which indicates
the parameters that you’ll need to edit.
5. In the Console window, replace FileName with Ch03_Demo and replace
CategoricalVariable with Instructor.

6. Press Enter.

After editing the file name and variable name, the R code and bar chart should look like
this (Figure 3.4):

FIGURE 3.4 ■ R Code and Bar Chart

> ggplot(Ch03_Demo,aes(x=factor(Instructor)))+geom_bar()

The bar chart represents the Instructor variable, representing the four students who had
Jones as their instructor and three students who had Smith.

COPYING GRAPHS

The graphs (Plots) that R produces are exportable, meaning that you can copy and paste
these images into other applications (e.g., Word, PowerPoint, Paint) using the following
steps:
 Chapter 3 • Getting Started in R 51

1. Click on the Plot tab.

2. Click on the Export tab.
3. Click on Copy to Clipboard . . .
4. Place the cursor on the plot and right-click.
5. Select Save image as.
6. The default file name is [Link]; [Link] extension stands for Portable Network
Graphic. You can edit the file name (e.g., change [Link] to Bar Chart [Link]), but do
not change the extension—[Link] part of the file name needs to stay the same. You
can also specify which directory (folder) you want to save the plot to.
7. Now that you have the plot file saved on your computer, you can import the file (Bar
Chart [Link]) into any program that can accept a graphic file using the drag-and-
drop method, or you can specify the graphic file name that you want to import into
the application.

SHORTCUTS

R has useful shortcut features that can help reduce the need for repetitive typing: R keeps a
list of every line of code that you enter into the Console window. While working in the Console
window, press the up-arrow key on your keyboard one time, and R will show you the last com-
mand that you entered. If you press Enter, R will run this line of code again. If you made a typo
that produced an error, you could press the up-arrow key to get that last line back, edit the line
to make the appropriate correction(s), and press Enter to run the corrected line of code.
Each time you press the up-arrow key, R steps backward through prior lines of code. The
down-arrow key takes you in the opposite direction. Essentially, using the up- and down-arrow
keys enables you to move through the list of commands that you’ve entered into the Console
window. As soon as you’ve found the line of interest, you can press Enter to (re)execute that
line, or you can edit that line and then press Enter to run it.
Whereas the up- and down-arrow keys present your prior lines of R code one at a time,
you can click on the History tab to view them in a window (Figure 3.5). You can scroll
through the History window to find the line of code of interest. When you find it, double-
click on the line, and it will be copied to the Console window, where you can edit it or press
Enter to run it as is.

FIGURE 3.5 ■ The History Tab Displays Lines of R Code That You’ve Executed
52 Part I • Statistical Foundation

You can also manage the list of commands in the History window. To delete line(s)
from the History window, highlight the line(s) that you want to remove and then click
the X icon. To remove all of the lines in the History window, click on the broom icon
(Figure 3.6).

FIGURE 3.6 ■ Significant Commands in the History Tab: X Icon Deletes

Highlighted Lines, Broom Icon Deletes All Lines

As you might expect, any lines that you delete via the History window will also be deleted from
the lines that you’ll see when you use the up- and down-arrow keys to view prior lines of R code.

CLEAR THE CONSOLE WINDOW

As you’ll see, R produces considerable results. To help reduce visual clutter, you can clear the
Console window; just click on the broom icon in the upper right corner of the Console window
(Figure 3.7).

FIGURE 3.7 ■ The Broom Icon Clears the Console Window

This will not affect anything else—all of the files and variables will remain intact. You
may find it useful to clear the Console window before beginning each new analysis. Another
way to clear the Console window is to press Control L (Ctrl L).

DOING MATH IN R

R can process math equations in the Console window. For example, you can enter 2 + 3 (the
spaces are optional) into the Console window the same way you would enter an equation in a
calculator and then press Enter; R will perform like a calculator and reply with 5.

> 2 + 3

[1] 5
 Chapter 3 • Getting Started in R 53

Some of the math symbols on a computer are different from those on a calculator. For example,
whereas calculators have a button, there’s no key on computer keyboards; instead, computers use
the slash (/) for division. Table 3.2 provides the protocol for entering math equations on a computer.

TABLE 3.2 ■ Math Symbols for Computer Processing

Rule Regular Notation Computer

+ is the same 8+2 8+2

− is the same 8−2 8−2

. is the same 8.2 8.2

Use the asterisk for multiplication 8×2 8*2

Use the slash for division 8÷2 8/2

Use the caret for exponent 8 2

8^2

Use round parentheses only [8 + 2] × {8 − 2} (8 + 2) * (8 − 2)

Do not include commas 5,150 5150

For example, to solve [459 × 888] ÷ 53, the R code is as follows:

> (459 * 888) / 5 ^ 3

[1] 3260.736

DATA ORDER DOESN’T MATTER

The order of the dataset that we used for this chapter is shown in Sequence A (Figure 3.8).
The data shown in Sequence B contain the same records (rows) but in a different order. As you
probably expect, the order of the records (rows) has no effect on the statistical results.

FIGURE 3.8 ■ Two Datasets Containing the Same Data With the Records (Rows) in
Different Order Produce the Same Statistical Results
Sequence A Sequence B
54 Part I • Statistical Foundation

Consider the results that emerged when we processed the dataset depicted in Sequence
A: It found seven records (n = 7) with a mean (average) Score of 84.29, and the Instructor bar
chart showed 4 for Jones and 3 for Smith. If we ran the same analyses on the Sequence B data-
set, we would see the same results: n = 7. As for calculating the average (mean), think about the
formula used to calculate an average; it doesn’t matter what order the computer adds up the
Scores and then divides that total by 7. As for the Instructor bar chart, the system would again
count four occurrences of Jones and three occurrences of Smith and produce the same graph.

PROCESSING YOUR OWN DATA

As you gain some experience processing statistics in R using the prepared datasets, you may
want to analyze some data of your own. Although R presents data in a table format similar to
Excel, unlike Excel, R does not include a data editor, meaning that you can’t establish variables
on the top row and then proceed to enter and edit data on the following rows in R. Fortunately,
there’s an easy workaround: You can create and edit/update a dataset outside of the R environ-
ment, save it on your system, and then seamlessly import the file into R for statistical processing.
Consider this example: To determine if role-play therapy enhances self-esteem, you recruit a
group of people, and when each person arrives, you assign each person a participant ID and have
them complete the Acme Self Esteem Scale, which produces a score between 1 and 30 (1 = low
self-esteem . . . 30 = high self-esteem); this will be the Pretreatment variable. Then you administer
a 1-hour role-play therapy session. After the intervention, you have each participant complete
the Acme Self Esteem Scale a second time; this will be the Posttreatment variable. To analyze the
data, you could use the paired t test or paired Wilcoxon test as detailed in Chapter 7 to deter-
mine if there is an improvement in self-esteem by comparing the Pretreatment variable to the
Posttreatment variable. The codebook and corresponding data would be as follows:

Codebook for [Link]

Variable: ParticipantID
Definition: Participant’s assigned number
Type: Continuous (101 . . . 999)

Variable: Pretreatment
Definition: Score on Acme Self Esteem Scale prior to treatment
Type: Continuous (1 = low self-esteem . . . 30 = high self-esteem)

Variable: Posttreatment
Definition: Score on Acme Self Esteem Scale after treatment
Type: Continuous (1 = low self-esteem . . . 30 = high self-esteem)
 Chapter 3 • Getting Started in R 55

ParticipantID Pretreatment Posttreatment

101 12 13

102 10 11

103 11 9

104 8 9

105 15 17

106 20 19

107 17 18

108 18 21

109 17 17

110 12 12

111 13 17

112 14 16

113 15 15

114 18 19

If you gathered the data on paper, you’d need to create a dataset in Excel or a text editor
(e.g., Notepad or TextEdit), which could then be imported into R.

Excel Option
The matrix layout in Excel is ideal for creating and maintaining a dataset that you can easily
import into R. Figure 3.9 shows an excerpt from a viable Excel dataset.

FIGURE 3.9 ■ Sample Dataset in Excel

Row 1 contains the three variable names (ParticipantID, Pretreatment, and Posttreatment).
When establishing the variable names and the file names, they’ll work best in R if there are no
spaces. When a file name or variable name contains more than one word, a standard convention
is to capitalize the first letter of each word instead of putting a space between words. For example,
instead of naming the file Self [Link], a more usable file name in R is [Link]; the same
pertains to the variables: Instead of Participant ID, ParticipantID will be more easily accessible in R.
56 Part I • Statistical Foundation

Also, each variable name must be unique (no duplicates). For example, even though we are
gathering two scores (pretest score and posttest score) from each participant using the same
instrument, we can’t use the variable name Score for both; instead, they each have unique
names: Pretreatment and Posttreatment. Be mindful when assigning variable names; this will
help with running statistics and reading the results reports, which include the variable names.
In this case, the variable names Pretreatment and Posttreatment are more comprehensible than
naming these variables X1 and X2.
Finally, save the Excel dataset using a meaningful file name (e.g., [Link]), which
you can then import into R for statistical processing.
If you make updates to the Excel dataset (e.g., provide corrections, include data from
additional participants), be aware that R is not dynamically linked to the source file. In other
words, R won’t know about any of the modifications that you made to the Excel file. If you
want to analyze the updated file, you’ll need to (re)import the most recent version of the Excel
dataset into R.

Text Editor Option

If you don’t have Excel, you can use a plain text editor (e.g., Notepad or TextEdit) to create
a csv file. The variable naming rules are the same, but instead of having the matrix arrange-
ment consisting of columns and rows of discrete cells, on the first line, you’ll list the vari-
able names with commas between each variable name. Starting at the second line, you’ll
have a separate line of data for each person with commas between the variables as in this
example:

ParticipantID,Pretreatment,Posttreatment
101,12,13
102,10,11
103,11,9
104,8,9
105,15,17

On a Windows computer, create a csv file:

1. Run the Notepad application.
2. On the first row, enter the variable names separated by commas.
3. On the next rows, enter the corresponding variable data separated by commas.
4. When you save the file, Notepad will save it as [Link] (text) file (e.g., [Link]).
5. When you exit Notepad, rename the file [Link] [Link] (e.g., rename [Link] to
[Link]).
 Chapter 3 • Getting Started in R 57

On a Macintosh, create a csv file:

1. Run the TextEdit application.
2. Before entering data, click on TextEdit, and select Preferences.
3. On the Preferences menu, select Plain text (not Rich text).
4. On the first row, enter the variable names separated by commas.
5. On the next rows, enter the corresponding variable data separated by commas.
6. When you save the file, TextEdit will save it as [Link] (text) file (e.g., [Link]).
7. When you exit TextEdit, rename the file [Link] [Link] (e.g., rename [Link] to
[Link]).

LOGGING OFF

Considering that the R Cloud system provides a specified amount of time per month, if you’re
idle for about 15 minutes, the system will automatically suspend your session so you don’t
mistakenly expire your online time. If the system goes into sleep mode, you can Resume and
continue where you left off.
To conserve your R Cloud monthly online time, it’s important to log off when you’re
done; just closing the Internet browser or shutting the computer would leave you logged on,
unnecessarily depleting 15 minutes or more from your monthly allotted online time. To log-
off from the R Cloud, follow these steps:

1. Click on your R Cloud user name or initials in the upper right corner.
2. Notice that the menu shows Compute hours, indicating how much of your monthly
time you’ve used up.
3. Click on Log Out to exit R Cloud.

GOOD COMMON SENSE

Consider what Albert Einstein said about computers: “Computers are incredibly fast, accu-
rate, and stupid. Human beings are incredibly slow, inaccurate, and brilliant. Together they
are powerful beyond imagination.” We can plausibly depend on R to rapidly provide accurate
calculations, effectively taking the tedious task of performing multiple math equations out of
our hands. This leaves us free to think about things that the computer cannot think about:
Which statistical test should I run? What do these results mean in the real world? How am I going
to use these results as part of my decision-making process?
The remaining chapters focus on applied (not theoretical) statistics; you’ll learn how to
select the appropriate statistic, how to run the analysis, and how to turn the numbers in the
results report into a clear and concise abstract that anyone can comprehend.
58 Part I • Statistical Foundation

If at some point you opt to run your own data, keep in mind the cautionary acronym
GIGO (Garbage In, Garbage Out)—if you enter erroneous data into a computer, the results
will be erroneous. Work carefully to uphold precision during each phase of the research pro-
cess: data collection, data entry, data analysis, and data reporting.
Finally, take care in securely managing your data, keeping your data system and data stor-
age (paper files, data backups) secure while maintaining confidentiality/anonymity protocols.

KEY CONCEPTS

• R Cloud setup
• Dataset structure
• Codebook
• File types
• File extensions
• Copy: Ctrl C
• Paste: Ctrl V
• Clear Console: Broom icon or Ctrl L
• Statistical run
• Copying results
• Copying graphics
• R shortcuts
• Doing math in R
• Processing your own data in R

PRACTICE EXERCISES

3.1 Get a screenshot from R showing the data in Ch03_Demo (in the upper left window).
3.2 Gather a set of screenshots from R showing that the following packages are installed:
car gmodels Hmisc mvnormtest psych stringi
3.3 What two keys should you press to Copy text?
3.4 What two keys should you press to Paste text?
3.5 How do you clear the screen in the Console window?
3.6 When viewing a statistical file outside of R (e.g., in Excel), what is on the first line?
(HINT: See the table in Exercise 3.7.)
 Chapter 3 • Getting Started in R 59

3.7 How many variables are in this dataset?

FirstName LastName Age Voted

Sandy Beach 16 No

Justin Thyme 33 Yes

Sarah Bellum 45 Yes

Robbin Banks 78 Yes

Melody Singer 51 No

Jack Hammer 50 Yes

3.8 What would you type into R to solve this equation?: {1,701 × 4,077} ÷ 422
3.9 What key(s) should you press to see prior lines of R code?
3.10 Explain what a file name extension is and give two examples.
 PART

STATISTICAL
TESTS II

61
 4 DESCRIPTIVE STATISTICS

You can summarize the contents of any variable with descriptive statistics.

Categorical Continuous

LEARNING OBJECTIVES

This chapter will enable you to:

4.1 Run and interpret descriptive statistics for continuous variables: n, mean,
median, mode, standard deviation, variance, minimum, maximum, and range
4.2 Load a dataset into R
4.3 Produce and export a histogram
4.4 Run and interpret stratified descriptive statistics for continuous variables
4.5 Produce and export a multigroup histogram
4.6 Comprehend and test for normal distribution
4.7 Recognize skewed distributions
4.8 Run and interpret descriptive statistics for categorical variables: n, %
4.9 Produce and export a bar chart
4.10 Manage data in R
4.11 Manage and export plots (graphics) in R
4.12 Utilize the 3P (Play, Pause, Practice) multimedia learning method

63
64 Part II • Statistical Tests

VIDEO RESOURCE

The video for this chapter is Ch04 – Descriptive Statistics.mp4. This video provides guid-
ance on running and interpreting descriptive (summary) statistics for continuous and cat-
egorical variables using the following dataset: Ch04_Demo.csv.

OVERVIEW—DESCRIPTIVE STATISTICS

Descriptive statistics, sometimes referred to as summary statistics, can be thought of as a

method for reducing a large list of data, often numbers, into a concise set of numbers, making
it more understandable. Table 4.1 serves as an example.

TABLE 4.1 ■ Data Containing 50 Records With Two Variables per Record:
Score and Hand

Score Hand Score Hand Score Hand Score Hand Score Hand

68 Right 82 Right 87 Left 96 Right 83 Right

87 Left 80 Right 81 Right 85 Right 87 Left

81 Right 86 Right 86 Right 75 Right 79 Left

96 Right 81 Right 82 Right 90 Left 88 Left

88 Left 73 Left 70 Left 67 Right 83 Right

76 Right 89 Left 77 Left 78 Right 88 Left

87 Left 96 Right 73 Left 76 Right 78 Left

91 Left 93 Left 87 Left 81 Right 73 Right

73 Right 81 Right 75 Right 82 Right 84 Right

79 Right 86 Right 83 Left 76 Right 94 Left

If someone asked you to explain the information that you’re looking at, you might
say: It looks like a list of scores for right- and left-handed people. There’s about half and half
right- and left-handed people, and their scores are mainly in the 70s and 90s. Since this is a
fairly short list of 50 people, we could consider this a reasonable estimate, but what if the
list spanned dozens or even hundreds of pages? Our minds simply can’t meaningfully con-
ceptualize that much information. It could take hours or even days to count the number of
entries, total them, and calculate an average. However, if we process the data using statisti-
cal software, we can promptly and precisely reduce a long list of data to a short list of figures
and graphs.
 Chapter 4 • Descriptive Statistics 65

DESCRIPTIVE STATISTICS IN CONTEXT

Statistical Reasoning
Descriptive statistics assess all of the data contained within a variable, no matter how big or
small the dataset is. The computer will produce a concise set of results that summarizes the
contents of the variable.
When you process a continuous variable, a variable that contains numbers, the descriptive sta-
tistics will count how many numbers were processed (the n), the average (the mean), the highest
and lowest (the maximum and minimum), and other figures that will be discussed in this chapter.
Descriptive statistics can also process a categorical variable, which, instead of numbers,
contains various categories (e.g., the categorical variable Flavor may contain three catego-
ries: Chocolate, Strawberry, and Vanilla). The descriptive statistics for this variable will simply
count how many times each flavor is listed (e.g., 35 Chocolate, 14 Strawberry, and 21 Vanilla);
these results are also provided in percentage form (e.g., 50% Chocolate, 20% Strawberry, and
30% Vanilla).

Applied Examples
We could conduct a student survey that asks how many courses the student is enrolled in
and the names of their instructors; descriptive statistics would tell us how many students
responded to the survey, the average number of courses that students are enrolled in, and also
how many students each instructor had.
We could also ask customers exiting a store how much they spent and note the time:
before or after noon; descriptive statistics can calculate the average amount spent and the
number of people who shopped before noon and how many shopped after noon.

When to Use This Statistic

Guidelines for Selecting Descriptive Statistics

Overview: This statistic summarizes the data contained within a single variable.

Variables: This statistic works with any variable (continuous or categorical).

Sample We tested 20 left-handed people and 30 right-handed people; the mean score was 82.34
Results: (SD = 7.17), spanning a 29-point range from 67 to 96; the most common scores were 81 and 87.

DESCRIPTIVE STATISTICS FOR CONTINUOUS

AND CATEGORICAL VARIABLES
Instead of looking at a list of data and trying to come up with your own words to describe it,
statisticians have come up with a concise set of descriptive statistics that are traditionally used
to reduce the data to a short list of results that together represent the whole dataset.
66 Part II • Statistical Tests

Continuous Variables
There are nine statistics used to describe the contents of a continuous variable: n, mean,
median, mode, standard deviation, variance, minimum, maximum, and range. You can also
order a histogram, which provides a bar chart to graphically visualize the data.

Categorical Variables
There are two statistics used to describe the contents of a categorical variable: n and percent.
You can also order a bar chart showing how many items are in each category.

DESCRIPTIVE STATISTICS: CONTINUOUS VARIABLES (SCORE)

Let’s see this in action. First, we’ll process the continuous variable Score (0 . . . 100). Then,
much later in the chapter, we’ll process the categorical variable Hand (Left-handed, Right-
handed). As we begin, for simplicity, we’ll use only the first column of numbers (Score) from
Table 4.1 (68, 87, 81, 96, 88, 76, 87, 91, 73, 79) to explain descriptive statistics for continuous
variables.

Analysis of the First Column of Scores

n
The n is the number of items in the variable. In this case, we have 68, 87, 81, 96, 88,
76, 87, 91, 73, and 79. To calculate the n, simply count how many items (numbers) are
contained in the variable; here we see that 68 is the first number, 87 is the second, 81 is
the third, and so on. Since the variable Score contains 10 numbers, the n is 10. Generally,
statisticians use the lowercase “n,” indicating that the numbers in the dataset were gath-
ered from a sample, not the entire population, which would be notated with the uppercase
“N.”

Mean (M)
For some reason, statisticians decided that the world needed a synonym for the word
“average.” The mean is the same formula as the average: Add up all the numbers and
divide by the n (the number of items contained in the variable). The abbreviation for the
mean is “M.”

68 + 87 + 81 + 96 + 88 + 76 + 87 + 91 + 73 + 79
M(Score) = __________________________________________
10
826
M(Score) = _
10

M(Score) = 82.6
 Chapter 4 • Descriptive Statistics 67

Median
The median is the middle number. Notice that the word median is similar to medium. To
find the median, arrange the numbers from lowest to highest. If the n is odd, there will be one
number in the middle. For this example, we’ll use the first nine numbers in our variable:

68, 73, 76, 79, 81, 87, 87, 88, 91

Since 81 is in the middle, the median is 81. Notice that there are four numbers that are less
than 81 and four numbers that are greater than 81.
The other possibility is that the n is even. Here we see 10 numbers, wherein two numbers
are tied for the middle position. When the n is even, the median is simply the mean (average)
of the two middle numbers.

68, 73, 76, 79, 81, 87, 87, 88, 91, 96

81 + 87
Median(Score) = _
2
168
Median(Score) = _
2

Median(Score) = 84

Mode
The mode is the most common number (notice that mode and most both begin with mo). Here
we see that each number occurs only once except for 87, which occurs twice, and hence the
mode is 87:

68, 73, 76, 79, 81, 87, 87, 88, 91, 96

Naturally, the numbers don’t need to be one-after-the-other in a dataset to be the mode;

they can be in any order. For example, if these same numbers were arranged in a different
order, the mode would still be 87:

79, 68, 73, 96, 87, 91, 81, 88, 87, 76

A variable may contain more than one mode, as in this case where there are two 68s and
two 87s:

68, 68, 76, 79, 81, 87, 87, 88, 91, 96

68 Part II • Statistical Tests

When there are two modes, this is referred to as bimodal and we would report: mode
(Score) = 68, 87. If there are more than two modes, this would be referred to as multimodal,
and you would report all of the modes.
Finally, if each number contained within a variable occurs only once (e.g., 16, 33, 45, 78),
the mode function will present all of the numbers (16 33 45 78) since they all occur an equal
number of times (once each). In terms of the mode, it would be appropriate to say that all of
the values (numbers) within this variable are unique.

Mean vs. Mode

Usually, the mean is preferred over the mode, but there are times when the mode is considered
the better choice. For example, in a dataset containing few numbers (low n), each number
can substantially alter the mean, whereas the mode would be more stable and less vulnerable
to such variability. For example, suppose we had a variable that contained 10,000 numbers;
changing any one number would have only a minor effect on the mean. Now consider a vari-
able containing only three numbers: 51, 54, 63; the mean is 56 and the mode is 54. Changing
the 51 to 21 would alter the mean substantially to 46, but the mode would keep stable at 54
since it remains the middle number.
Also, in cases where there are outliers (one or several unusually low or high numbers),
those few extreme numbers will impact the mean; in such cases, the median would provide a
more stable statistic. For example, consider a dataset consisting of five numbers: 5, 72, 78, 81,
84; the mean is 64, whereas the median is 78. Notice that the low outlier (5) affects the mean,
but since the median is the middle value in the sequence, the outlier (5) has no effect on it.
Inspecting these numbers, one could reasonably assert that the median of 78 is a better repre-
sentation of most of the values in this dataset compared to the mean of 64.

Standard Deviation (SD)

The standard deviation (SD) expresses how much variability there is within a variable. In
Figure 4.1, the heights of the three people in Group A are 67, 68, and 67 inches tall; the mean
height is 67.33. We see that their heights are fairly similar; there’s little variability among these
numbers. Notice that their heights don’t deviate (differ) much from the mean (M = 67.33)—
they’re each just slightly above or below the mean, resulting in a fairly low standard deviation
(SD = 0.58).
Now observe the heights of the three people in Group B: 86, 45, and 71 inches; coinciden-
tally, their mean height is also 67.33. Clearly, their heights are quite different from each other;
there’s more variability among these numbers. Notice that their heights deviate substantially
from the mean—the basketball player’s height (86) deviates substantially above the mean
(M = 67.33), and the little girl’s height (45) deviates substantially below the mean (M = 67.33).
These individual deviations (differences) from the mean are reflected in the higher standard
deviation for this group (SD = 20.74).
To summarize, the standard deviation helps us understand if there’s a lot or a little vari-
ability among the numbers contained within the variable. Notice that Group A and Group
B both have the same mean (67.33); without the standard deviation scores, we may get the
impression that since the means of the two groups are exactly the same, the heights of the
 Chapter 4 • Descriptive Statistics 69

FIGURE 4.1 ■ Less Diversity of Heights of the People in Group A Produces a

Low(er) Standard Deviation (SD = 0.58), Whereas Higher Diversity
of Heights of the People in Group B Produces a High(er) Standard
Deviation (SD = 20.74)

people in Group A are just like the heights of the people in Group B. It’s the low standard
deviation in Group A (SD = 0.58) that indicates that there’s less variability among the heights
of these people (their heights are fairly similar to each other—they don’t deviate much from
the mean), and the higher standard deviation in Group B (SD = 20.74) indicates that there’s
more variability among the heights of these people (their heights are very different from each
other—they deviate considerably more from the mean).
Another way to think about the standard deviation is that it (approximately) represents
the average distance of each number from the mean. If the numbers are close to the mean,
then there would be less difference between each number and the mean, which would render
a smaller standard deviation. Alternatively, if the numbers are further from the mean (above
or below the mean), the differences between those numbers and the mean are larger, which
increase the standard deviation.
When documenting statistical results, the standard deviation is generally written after the
mean as such: M = 67.33 (SD = 0.58).

Variance
The variance is simply the standard deviation squared (SD2). When it comes to documenting
statistical results, statisticians traditionally opt to include the standard deviation alongside the
mean instead of the variance, but the variance has other uses; many statistical formulas involve
the variance. Referring to Figure 4.1, since the standard deviation for Group A is 0.58, the
variance is 0.582 (0.58 × 0.58). Hence, the variance for Group A is 0.3364.
Although R does not include the variance in the descriptive statistics results, you can use
R to perform this simple calculation; just enter .58 * .58 or .58 ^ 2 (the spaces are optional)
in the R Console window, or enter .58 × .58 on any calculator.
70 Part II • Statistical Tests

> .58 * .58

[1] 0.3364

Minimum
The minimum is the lowest number (68).

68, 73, 76, 79, 81, 87, 87, 88, 91, 96

Maximum
The maximum is the highest number (96).

68, 73, 76, 79, 81, 87, 87, 88, 91, 96

When assessing minimum and maximum statistics, bigger is not necessarily better. It’s
important to consider the context of the variable; for example, a high bowling score is good,
whereas a low golf score is good.

Range
The range is the difference between the maximum and the minimum (just subtract them),
indicating the span of the numbers: 96 – 68 = 28; hence, the range is 28.

68, 73, 76, 79, 81, 87, 87, 88, 91, 96

Introducing R
For clarity, the examples used thus far have involved small lists of numbers. Now it’s time to
use R to process descriptive statistics using the entire dataset consisting of 50 records (rows of
data) with two variables (columns of data): Hand (indicating handedness: Left or Right) and
Score (their score on a test).

Dataset
This analysis uses the dataset: Ch04_Demo.csv.
To open the dataset:

1. Log on to your RStudio on the cloud account.

2. Click on your content (e.g., Stats Course 1 or Untitled Project).
3. Click on Files.
 Chapter 4 • Descriptive Statistics 71

4. Click on Upload.
5. Click on Choose File.
6. Navigate to the folder that contains the dataset: Ch04_Demo.csv.
7. Click on the file name: Ch04_Demo.csv.
8. Click on Open.
9. Click on OK.
10. Click on the file name: Ch04_Demo.csv.
11. Click on Import Dataset . . .
12. Click on Import.

If you see the message in Figure 4.2 during the upload process, click on Yes, which will
enable R to load in the appropriate conversion software to translate the non-R data into a form
that R can use.

FIGURE 4.2 ■ Select Yes to Update Package That Enables the Data Import

The codebook provides details about each variable in the dataset. The following is the
codebook for the dataset that we imported:
Codebook for Ch04_Demo.csv
Variable: Score
Definition: Score on test
Type: Continuous (0 . . . 100)

Variable: Hand
Definition: Handedness of the participant
Type: Categorical (Left, Right)

Descriptive Statistics: Continuous Variables (Overall)

There are two types of variables in this file: Score is a continuous variable, and Hand is a cat-
egorical variable. We’ll first process descriptive statistics for the continuous variable (Score);
later in this chapter, we’ll process the categorical variable (Hand), which produces a much
shorter list of results.
72 Part II • Statistical Tests

To process descriptive statistics for continuous variables, refer to the R Syntax Guide,
page 3; we’ll proceed in order starting from the top of the page.
First, click on the Packages tab, then check ggplot2, uncheck Hmisc, and check psych.

Packages
☑ ggplot2
☐ Hmisc
☑ mvnormtest
☑ psych

Next, run the descriptive statistics for a continuous variable.

Test Run: Descriptive Statistics for a Continuous Variable

Run descriptive statistics for a continuous variable:
describe(FileName$ContinuousVariable)

1. In the R Syntax Guide document, page 3, highlight: describe(FileName$Contin

uousVariable) and press Ctrl C (copy).

2. In R, place the cursor in the Console window and click (bar cursor will blink) and
press Ctrl V (paste).
3. Edit the line of code in R: Change the FileName to Ch04_Demo (notice that upon
typing the third letter, R helps with autocomplete).
4. Change ContinuousVariable to Score.
5. The edited line of code should look like this: describe(Ch04_Demo$Score).
6. Press Enter.

After editing the file name and variable name, the R code and results should look like this:

> describe(Ch04_Demo$Score)

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 50 82.34 7.17 82 82.28 7.41 67 96 29 -0.03 -0.6 1.01

Notice that anything you type into the R Console window is blue, which matches
the blue text in the R Syntax Guide, and the results that R produces are black text. Error
messages are red. Here we see that the n is 50, the mean (average) is 82.34 (SD = 7.17), the
median is 82, the minimum is 67, the maximum is 96, and the range is 29. Notice that the
variance and the mode are not included in this report. As mentioned earlier, most statistical
reports use the standard deviation, not the variance. The variance is commonly referenced in
 Chapter 4 • Descriptive Statistics 73

statistical formulas. It’s easy to compute the variance in R or on any calculator. The standard
deviation is 7.17; hence, the variance is simply the standard deviation squared; you can enter
7.17 * 7.17 or 7.17 ^ 2 (the spaces are optional) in the R Console window. Remember, com-
puters use the asterisk (*) for multiplication and the caret (^) (Shift 6 ) for the exponent. R will
calculate the variance (51.4089) as shown below:

> 7.17 * 7.17

[1] 51.4089

R does not currently have a built-in mode function, but we can easily create it.

Test Run: Establish Mode Function

To establish the mode function, copy the following four lines of code into R:
mode <- function(a)
{b <- unique(a)
tab <- tabulate(match(a,b))
b[tab == max(tab)]}

1. Highlight the above following four lines of R code and press Ctrl C (copy).

2. In R, place the cursor in the Console window and click (bar cursor will blink) and
press Ctrl V (paste).
3. Press Enter.

You only need to create the mode function once; now that you’ve established the mode
function, you can use it as many times as you want.

Test Run: Run the Mode

Run the Mode:
mode(FileName$ContinuousVariable)

1. Highlight: mode(FileName$ContinuousVariable) and press Ctrl C (copy).

2. In R, place the cursor in the Console window and click (bar cursor will blink) and
press Ctrl V (paste).
3. Edit the line of code in R: Change the FileName to Ch04_Demo (notice that when you
type the third letter, R will help you with autocomplete).
4. Change ContinuousVariable to Score.
5. Press Enter.
74 Part II • Statistical Tests

After editing the file name and variable name, the R code and results should look like this:

> mode(Ch04_Demo$Score)

[1] 81 87

The results indicate that Score has two modes: 81 and 87.

Documenting Results (Continuous Variable, Overall)

Instead of providing a list of numbers that may be incomprehensible to the reader, document
the results in the form of a concise abstract:

Abstract
We gathered test scores from a total of 50 participants. The mean score was 82.34 (SD = 7.17)
with a median of 82. The most common scores (modes) were 81 and 87. Scores spanned a
29-point range, from 67 to 96.

Histogram (Overall)
In addition to the numeric report, we can also order a histogram, which provides a graphical
representation of a continuous variable.

Test Run: Histogram for One Variable

Draw a histogram for one variable:
hist(FileName$ContinuousVariable)

After editing the file name and variable name, the R code and histogram should look like this:

> hist(Ch04_Demo$Score)
 Chapter 4 • Descriptive Statistics 75

The histogram provides further insight into the characteristics of a continuous variable—
a picture is indeed worth a thousand words. The heights of the bars show us that most of the
people earned a score between 80 and 90, there are some people who scored more than 90,
and there are progressively fewer who scored lower than 80.
Notice that this histogram is shaped like a fairly symmetrical mountain or volcano—it
peaks in the middle and tapers off at the ends. This shape is referred to as a normal distribu-
tion, and it’s often observed among continuous variables. This distribution is also referred to
as a normal curve or bell curve, since it’s shaped like a bell.
The histogram provides a graphical representation of the data, which we could use to
visually estimate if the data are normally distributed, but the Shapiro–Wilk statistic, which is
covered later in this chapter, provides a more precise way to make that determination.

Descriptive Statistics: Continuous Variables (Stratified by Group)

So far, we’ve run descriptive statistics for Score for the whole group (left- and right-handed
people combined), but suppose we wanted to know how left-handed people performed com-
pared to right-handed people. This is referred to as a stratified analysis; strata implies that the
list will be split into sections based on a categorical variable. Since Hand contains two values
(Left and Right), R will produce two separate copies of the descriptive statistics results for
Score: one for left-handed people and another for right-handed people.

Test Run: Descriptive Statistics for Each Group

Run descriptive statistics for each group:
describeBy(FileName$ContinuousVariable,FileName$Categorical)

After editing the file name and variable names, the R code and results should look like this:

> describeBy(Ch04_Demo$Score,Ch04_Demo$Hand)

Descriptive statistics by group

group: Left

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 20 84.45 6.99 87 84.94 5.19 70 94 24 -0.67 -0.88 1.56

----------------------------------------------------------------------

group: Right

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 30 80.93 7.05 81 80.5 6.67 67 96 29 0.39 0.18 1.29

This report should look familiar; it’s just like the prior descriptive statistics results, but
instead of providing a single set of results representing the scores of everyone (left- and right-
handed combined), notice that there are two sets of results; the results above the dashed line
76 Part II • Statistical Tests

pertain to the Score for those who are left-handed and the results below the dashed line pertain
to the Score for those who are right-handed.

Documenting Results (Continuous Variable, Stratified)

Abstract
We gathered test scores from a total of 50 participants. Among the 20 left-handed participants,
the mean score was 84.455 (SD = 6.99) with a median of 87.00. Scores spanned a 24-point
range, from 70 to 94. Among the 30 right-handed participants, the mean score was 80.93
(SD = 7.05) with a median of 81.00. Scores spanned a 29-point range, from 67 to 96.

Histogram (Stratified)
Whereas the histogram shown in the prior example is based on the Score of all of the individu-
als in the dataset (left- and right-handed combined), we can also produce separate histograms
for each category. For example, we could use the following code to produce one histogram of
Score for left-handed people and another for right-handed people.

Test Run: Histogram for Each Group

Draw a histogram for each group:
ggplot(FileName,aes(x=FileName$ContinuousVariable))+
geom_histogram()+facet_grid(FileName$CategoricalVariable~.)

After editing the file name and variable names, the R code and histograms should look
like this:

> ggplot(Ch04_Demo,aes(x=Ch04_Demo$Score))+geom_histogram()+

facet_grid(Ch04_Demo$Hand~.)
 Chapter 4 • Descriptive Statistics 77

Normal Distribution
The histogram in Figure 4.3 is an example of a normal curve; naturally, most histograms
aren’t this perfectly symmetrical. Most continuous variables are normally distributed, but as
you probably expect, there are exceptions.

FIGURE 4.3 ■ Illustration of a Normal Curve in Histogram Form

While it is possible to visually inspect a histogram to subjectively decide if it appears to be

normally distributed, the Shapiro–Wilk test can assess the normality of a variable objectively.
This test is easy to run and interpret. In this example, we’ll use the Shapiro–Wilk test to assess
the normality of the variable Score.

Test Run: Shapiro–Wilk Test for One Variable

> [Link](Ch04_Demo$Score)

Shapiro-Wilk normality test

data: Ch04_Demo$Score
W = 0.9819, p-value = 0.6342

Essentially, the Shapiro–Wilk test compares the distribution of your (continuous) vari-
able to a perfect normal distribution; this all happens numerically as internal calculations—
you won’t see any histograms emerge.

Interpreting the Shapiro–Wilk Test

If the Shapiro–Wilk p value is less than or equal to .05, this indicates that there is a statistically
significant difference between the distribution of your variable and a perfect normal distribu-
tion, meaning that your variable is not normally distributed.
Alternatively, if the Shapiro–Wilk p value is greater than .05, this indicates that there is
no statistically significant difference between the distribution of your variable and a perfectly
normal distribution, meaning that your variable is normally distributed.
78 Part II • Statistical Tests

To summarize the Shapiro–Wilk test results:

• If the Shapiro–Wilk p value is ≤ .05, then the variable is not normally distributed.
• If the Shapiro–Wilk p value is > .05, then the variable is normally distributed.

Since the Shapiro–Wilk p value for Score is .63, which is greater than .05, this indicates
that this variable is normally distributed.
Whereas the prior analysis analyzed the entire Score variable for normality, we can also
use the Shapiro–Wilk test to run stratified analyses. For example, we’ll use the Shapiro–Wilk
test to assess the normality of Score for Left (handed people) only:

Test Run: Shapiro–Wilk Test for Each Category (Left)

> [Link](subset(Ch04_Demo$Score,Ch04_Demo$Hand=="Left"))

Shapiro-Wilk normality test

data: subset(Ch04_Demo$Score, Ch04_Demo$Hand == "Left")

W = 0.88925, p-value = 0.02606

Since the p value (.02606) is less than or equal to .05, this indicates that the Score for the left-
handed people is not normally distributed. Next, run the same analysis, but change Left to Right:

Test Run: Shapiro–Wilk Test for Each Category (Right)

> [Link](subset(Ch04_Demo$Score,Ch04_Demo$Hand=="Right"))

Shapiro-Wilk normality test

data: subset(Ch04_Demo$Score, Ch04_Demo$Hand == "Right")

W = 0.93776, p-value = 0.07917

Since the p value (.07917) is greater than .05, this indicates that the Score for the right-handed
people is normally distributed.

Skewed Distribution
As with any rule, there are exceptions; not all histograms produce normally shaped curves. Depending
on the distribution of the data within the variable, the histogram may have a skewed distribution,
meaning that the distribution is shifted to one side or the other, as shown in Figures 4.4 and 4.5.
In Figure 4.4, we see that most of the data are on the right, between about 60 and 100, but
there is a small scattering of lower values (between 0 and about 30). These few low values that
substantially depart from the majority of the data are referred to as outliers. Typically, outliers
become apparent when graphing the data. We would say that the histogram in Figure 4.4 has
 Chapter 4 • Descriptive Statistics 79

FIGURE 4.4 ■ Negative (Left) Skew

FIGURE 4.5 ■ Positive (Right) Skew

outliers that are substantially lower than the majority of the values; hence, we see they’re to the
left of the histogram. This is skewed left, or negatively skewed.
Outliers can also be substantially higher than the majority of the values. Figure 4.5 shows
the outliers as a handful of high values scattered to the right on the histogram; this distribu-
tion would be referred to as being skewed right, or positively skewed. The notion of normality of
the data distribution will be discussed further in future chapters.

DESCRIPTIVE STATISTICS: CATEGORICAL VARIABLES (HAND)

Descriptive statistics for categorical variables are fairly concise; essentially, all you can do with
categorical variables is count how many entries there are in each category. In this case, we
could ask, How many left-handed people are there and how many right-handed people are there?
R answers this question for each category in two forms: Frequency (n) and Proportion (%).
To run descriptive statistics for a categorical variable, refer to the R Syntax Guide, page 4.
Click on the Packages tab, then check ggplot2 and Hmisc, and uncheck psych.

Packages
☑ ggplot2
☑ Hmisc
☐ psych
80 Part II • Statistical Tests

Next, proceed to run the descriptive statistics for categorical variables.

Test Run: Descriptive Statistics for Categorical Variable

Run descriptive statistics for a categorical variable:
describe(FileName$CategoricalVariable)

After editing the file name and variable name, the R code and results should look like this:

> describe(Ch04_Demo$Hand)

Ch04_Demo$Hand

n missing distinct

50 0 2

Value Left Right

Frequency 20 30

Proportion 0.4 0.6

The above report begins with the n (50 records in the file), with zero missing (null) entries
and two distinct values (Left and Right). The next row shows the Value(s) contained within
the variable Hand: Left and Right (remember: variables contain values). The next row is the
Frequency; this is the same as the n or headcount for each category; it shows that there are 20
Left (handed people) and 30 Right (handed people). The last row is the Proportion; this is the
percentage (just multiply the results by 100). The Proportion shows that Hand consists as 0.4
Left and 0.6 Right. To present these proportions as percentages, simply multiply them by 100:
40% of the people are left-handed and 60% are right-handed.

Documenting Results (Categorical Variable)

Abstract
Our sample consisted of 50 participants: 20 (40%) left-handed participants and 30 (60%)
right-handed participants.

We can also get a visualization of the data with a bar chart.

Test Run: Bar Chart for Categorical Variable

Draw a bar chart:
ggplot(FileName,aes(x=factor(CategoricalVariable)))+geom_bar()
 Chapter 4 • Descriptive Statistics 81

After editing the file name and variable name, the R code and bar chart should look like this:

> ggplot(Ch04_Demo,aes(x=factor(Hand)))+geom_bar()

MANAGING DATA

When you click on the Environment tab in the upper right window, you’ll see that R keeps a
list of all of the datasets that you’ve loaded via the Upload function. As you proceed to import
and analyze datasets, this list will grow; as it gets larger, the autocomplete feature may become
less efficient as it could provide more choices than you may want to see. There are two ways to
manage this list of datasets:
1. Remove selected dataset(s): To remove one or more specific datasets from the R
Environment, use the rm [remove] command.
To remove one dataset, enter the dataset name:
rm(Ch03_Demo)

To remove more than one dataset, enter each dataset name separated by a comma:
rm(Ch03_Demo,Ch04_Demo)

2. Remove all datasets: To remove all of the datasets from the R Environment, click
on the broom icon to sweep them all away (Figure 4.6). If you use this method, it will
also remove the mode function. To reestablish the mode function, use the process
detailed earlier in this chapter.

You can also remove one or more datasets from the R system completely. You may want
to use this function after completing each chapter to help reduce window clutter. If you
do this, the dataset file(s) will still exist on your storage device (e.g., computer drive, flash
82 Part II • Statistical Tests

FIGURE 4.6 ■ Click on the Broom Icon to Remove All Datasets From the R
Environment

drive), so if later you want to conduct further analysis on it, all you’d need to do is reupload
it to R:

1. Remove dataset(s): To remove one or more datasets from the R System, on the Files
menu, check the dataset(s) that you wish to delete, then click on Delete (Figure 4.7).

FIGURE 4.7 ■ Check the Dataset(s) That You Wish to Delete, Then Click on Delete

MANAGING PLOTS

Similar to dataset management, although R only presents your most recent plot (graphic), R
automatically retains all of your plots until you delete them. To view each plot, click on the
Plots tab and then click on the Left and Right Arrow icons (← →) Figure 4.8).

FIGURE 4.8 ■ On the Plots tab, Click on the Left and Right Arrows (← →) to Cycle
Through the Plots
 Chapter 4 • Descriptive Statistics 83

There are two ways to manage this list of datasets:

1. Remove one plot: To remove the plot that is currently on the screen, on the Plots
tab, click on the X icon (Figure 4.9). R will then present a dialog box to confirm the
deletion.

FIGURE 4.9 ■ On the Plots Tab, Click on the X Icon to Remove the Plot Currently
Showing

2. Remove all plots: To remove all of the plots, on the Plots tab, click on the broom icon
(Figure 4.10). R will then present a dialog box to confirm the deletion.

FIGURE 4.10 ■ On the Plots Tab, Click on the Broom Icon to Remove All of the Plots

The plots processor in R is good, but it’s not perfect (yet). If you find that your plots are
incomplete (e.g., missing an axis, missing titles) or otherwise erroneous, click on the remove
all plots broom; this seems to resolve a multitude of graphical problems.

MOVING FORWARD

To build your initial proficiency in R, this chapter contained numbered step-by-step instruc-
tions detailing how to import a dataset into R and how to copy, paste, and edit the blue
[remove] command. lines of code from the R Syntax Guide into R for processing. Now that
you know these skills, future chapters do not include those micro-details (e.g., Highlight the
line of code, Press Ctrl C, Put the cursor in the R console box, Click to see the blinking bar cursor,
Press Ctrl V ), which should conserve your reading time. Instead, the text indicates which
line(s) of code you should use from the R Syntax Guide; it’s presumed that you know how to
properly copy, paste, and edit R code in the Console window.
The remaining chapters provide instructions for using the R Syntax Guide to (1) select
and run appropriate statistical tests and plots, (2) interpret selected figures in the out-
put reports, and (3) mindfully document the results in the form of a concise readable
abstract.
84 Part II • Statistical Tests

Learning tip: To facilitate your learning, you can download and view the videos for each
chapter. Each video provides an overview of when to use the statistic, what the statistic
does, how to check if the assumptions (pretest criteria) are met, and how to run the
statistic. These videos use procedural learning—specifically, demonstration-based learn-
ing—providing narrated step-by-step guidance, enabling you to successfully reproduce
the steps to run each statistic using the 3P learning method:
(1) Play a segment of the video
(2) Pause the video
(3) Practice the segment that you just viewed

Repeat this three-step process until you’ve completed the statistical analysis.

GOOD COMMON SENSE

Instead of having to focus on memorizing multiple formulas and spending a lot of time rep-
etitiously calculating complex equations on large volumes of data, R, or any other statistical
program, can rapidly and precisely produce results in the form of numbers and graphics in
exchange for you entering some simple commands. This provides more time for you to con-
sider the meaning and implications of the results and compose comprehensive documenta-
tion in clear language that anyone can understand.
As you process statistics, keep in mind that the computer program is merely running
the data through a sequence of computational instructions; the computer isn’t sentient like
you are—it never really thinks about what it’s doing, and although it will produce accu-
rate results, it can’t contemplate the meaning of such results—it simply hands you the find-
ings. For example, a dataset may contain variables such as PhoneNumber, SerialNumber, or
ApartmentNumber; while it’s possible to run descriptive statistics on such variables, the results
would be meaningless. Consider the results that you’d get if you ran descriptive statistics for
the variable PhoneNumber; you’d know the mean PhoneNumber, but what would you do with
that number . . . call it?
In summary, R will carry out the calculations that you tell it to, but it’s up to you to
thoughtfully process and interpret the meanings and implications of your statistical results.

KEY CONCEPTS

• Descriptive statistics
• n (number)
• Mean (M)
• Median
• Mode
• Standard deviation (SD)
 Chapter 4 • Descriptive Statistics 85

• Variance
• Minimum
• Maximum
• Range
• Central tendency
• Histograms
• Skew
• Negative (left) skew
• Positive (right) skew
• Outliers
• Bar chart

PRACTICE EXERCISES

Download the prepared R datasets from the Sage companion website. Import the specified
datasets into R, and then process and document your findings for each exercise.

Exercise 4.1
You conducted a survey of Professor Campbell’s class and Professor Waddell’s class, asking
students: How many courses are you enrolled in?
Dataset: Ch04_Ex01.csv
Codebook:
Variable: Courses
Definition: Number of courses the student is enrolled in
Type: Continuous (1 . . . 6)

Variable: Professor (instructor)

Definition: Instructor
Type: Categorical (Campbell, Waddell)

Run and document the following analyses:

a. Descriptive statistics for Courses for the whole dataset
b. Histogram of Courses for the whole dataset
c. Descriptive statistics for Courses for each Professor
d. Histograms of Courses for each Professor
e. Descriptive statistics for Professor
f. Bar chart for Professor
86 Part II • Statistical Tests

Exercise 4.2
You ask customers exiting a grocery store: How much did you spend today? You then record
the total from the receipt and the time of day (AM/PM).
Dataset: Ch04_Ex02.csv
Codebook:
Variable: Bill
Definition: Amount of money spent shopping
Type: Continuous (0 . . . no limit)

Variable: Time
Definition: Before or after noon
Type: Categorical (AM, PM)

Run and document the following analyses:

a. Descriptive statistics for Bill for the whole dataset
b. Histogram of Bill for the whole dataset
c. Descriptive statistics for Bill for each Time
d. Histograms of Bill for each Time
e. Descriptive statistics for Time
f. Bar chart for Time

Exercise 4.3
You administer a 20-question math quiz to students in a fourth-grade class to assess students
who sit in the front (rows 1, 2, and 3) and students who sit in the back (rows 4, 5, and 6).
Dataset: Ch04_Ex03.csv
Codebook
Variable: QuizScore
Definition: Score on math quiz
Type: Continuous (0 . . . 20)

Variable: SeatPosition
Definition: Student’s seating position in classroom
(Front = Rows 1, 2, 3; Back = Rows 4, 5, 6)
Type: Categorical (Front, Back)

Run and document the following analyses:

a. Descriptive statistics for QuizScore for the whole dataset
b. Histogram of QuizScore for the whole dataset
 Chapter 4 • Descriptive Statistics 87

c. Descriptive statistics for QuizScore for each SeatPosition

d. Histograms of QuizScore for each SeatPosition
e. Descriptive statistics for SeatPosition
f. Bar chart for SeatPosition

Exercise 4.4
You are interested in how long it takes for two baristas to serve drinks. You use a stopwatch to
record how long it takes (in seconds) for Brad and Janet to serve the drink once they take the order.
Dataset: Ch04_Ex04.csv
Codebook
Variable: ServiceTime
Definition: Seconds required to serve the drink
Type: Continuous (0 . . . no limit)

Variable: Barista
Definition: Drink server
Type: Categorical (Brad, Janet)

Run and document the following analyses:

a. Descriptive statistics for ServiceTime for the whole dataset
b. Histogram of ServiceTime for the whole dataset
c. Descriptive statistics for ServiceTime for each Barista
d. Histograms of ServiceTime for each Barista
e. Descriptive statistics for Barista
f. Bar chart for Barista

Exercise 4.5
To find out how long parents and their children are spending on social media, you recruit a
group of individuals who allow you to load time-monitoring software on their devices for 1
week.
Dataset: Ch04_Ex05.csv
Codebook
Variable: SocialMediaTime
Definition: Minutes per day on social media
Type: Continuous (0 . . . 1440)
88 Part II • Statistical Tests

Variable: Participant
Definition: Role of study participant
Type: Categorical (Child, Parent)

Run and document the following analyses:

a. Descriptive statistics for SocialMediaTime for the whole dataset
b. Histogram of SocialMediaTime for the whole dataset
c. Descriptive statistics for SocialMediaTime for each Participant
d. Histograms of SocialMediaTime for each Participant
e. Descriptive statistics for Participant
f. Bar chart for Participant

Exercise 4.6
To determine which form of learning students prefer, live instruction or prerecorded video,
after each lesson, you arrange for students to respond to a one-question survey indicating
their satisfaction with the lesson (1 = very unsatisfied . . . 10 = very satisfied).
Dataset: Ch04_Ex06.csv
Codebook
Variable: Satisfaction
Definition: Satisfaction with teaching style
Type: Continuous (1 = very unsatisfied . . . 10 = very satisfied)

Variable: TeachingStyle
Definition: Style of instruction
Type: Categorical (Live, Prerecorded)

Run and document the following analyses:

a. Descriptive statistics for Satisfaction for the whole dataset
b. Histogram of Satisfaction for the whole dataset
c. Descriptive statistics for Satisfaction for each TeachingStyle
d. Histograms of Satisfaction for each TeachingStyle
e. Descriptive statistics for TeachingStyle
f. Bar chart for TeachingStyle
 Chapter 4 • Descriptive Statistics 89

Exercise 4.7
To assess the quality of life of adults, you recruit a group of working and retired people and
administer the Acme Quality of Life Scale (AQLS), which produces a score between 1 and
30 (1 = low quality of life . . . 30 = high quality of life).
Dataset: Ch04_Ex07.csv
Codebook
Variable: QualityOfLife
Definition: Score on the Acme Quality of Life Scale
Type: Continuous (1 = low quality of life . . . 30 = high quality of life

Variable: WorkStatus
Definition: Employment status
Type: Categorical (Working, Retired)

Run and document the following analyses:

a. Descriptive statistics for AQLS for the whole dataset
b. Histogram of AQLS for the whole dataset
c. Descriptive statistics for AQLS for each WorkStatus
d. Histograms of AQLS for each WorkStatus
e. Descriptive statistics for WorkStatus
f. Bar chart for WorkStatus

Exercise 4.8
To determine the effect that music may have on exercise, you recruit a group of individuals
who regularly walk for exercise and randomly give half of the participants an MP3 player
loaded with good walking music and a set of headphones; the others agree to walk without
music. You also give every participant a pedometer to count how many steps they take during
their walks over the course of 1 week.
Dataset: Ch04_Ex08.csv
Codebook
Variable: Steps
Definition: Total steps taken in 1 week
Type: Continuous (0 . . . no limit)

Variable: MusicStatus
Definition: Walked with or without music
Type: Categorical (Music, No Music)
90 Part II • Statistical Tests

Run and document the following analyses:

a. Descriptive statistics for Steps for the whole dataset
b. Histogram of Steps for the whole dataset
c. Descriptive statistics for Steps for each MusicStatus
d. Histograms of Steps for each MusicStatus
e. Descriptive statistics for MusicStatus
f. Bar chart for MusicStatus

Exercise 4.9
You’ve been asked to study the effectiveness of two tutoring programs: one where the tutor
works with one student at a time and the other where the tutor works with two students at a
time. At the end of each tutoring session, the students take a brief quiz to assess their skills.
The quiz scores are the percentage of correct answers (0% . . . 100%).
Dataset: Ch04_Ex09.csv
Codebook
Variable: QuizScore
Definition: Percentage of quiz correct after tutoring session
Type: Continuous (0 . . . 100)

Variable: Tutor
Definition: Tutor working with one student or two
Type: Categorical variable (one student, two students)

Run and document the following analyses:

a. Descriptive statistics for QuizScore for the whole dataset
b. Histogram of QuizScore for the whole dataset
c. Descriptive statistics for QuizScore for each Tutor (type)
d. Histograms of QuizScore for each Tutor (type)
e. Descriptive statistics for Tutor (type)
f. Bar chart for Tutor (type)

Exercise 4.10
To determine if coffee has an effect on mood, you recruit a group of participants and ran-
domly assign them to one of three groups: Group 1 will get regular coffee, Group 2 will get
decaffeinated coffee, and Group 3 will be given nothing to drink. Then you administer the
Acme Mood Instrument to all of the participants, which produces a score ranging from 1 to
10 (1 = very unhappy . . . 10 = very happy).
 Chapter 4 • Descriptive Statistics 91

Dataset: Ch04_Ex10.csv
Codebook
Variable: Mood
Definition: Score on Acme Mood Instrument
Type: Continuous (1 = very unhappy . . . 10 = very happy)

Variable: Beverage
Definition: Drink that participant was given
Type: Categorical (Coffee, Decaf, No Coffee)

Run and document the following analyses:

a. Descriptive statistics for Mood for the whole dataset
b. Histogram of Mood for the whole dataset
c. Descriptive statistics for Mood for each Beverage
d. Histograms of Mood for each Beverage
e. Descriptive statistics for Beverage
f. Bar chart for Beverage
 t TEST AND WELCH
5 TWO-SAMPLE t TEST

When you have two groups measured with a continuous variable, you can determine if one
group outperformed the other with a t test or Welch two-sample t test.

Group 1 Group 2

LEARNING OBJECTIVES

Upon completing this chapter, you will be able to:

5.1 Determine when to use the t test
5.2 Build a research question and corresponding hypotheses
5.3 Run and assess the pretest criteria: normality and homogeneity of variance
5.4 Select and run the proper version of the t test: the t test or the Welch two-
sample t test
5.5 Interpret the results
5.6 Comprehend the p and ɑ value
5.7 Resolve the hypotheses
5.8 Write a concise abstract detailing the research question and results
5.9 Understand the implications of Type I and Type II errors

93
94 Part II • Statistical Tests

VIDEO RESOURCE

The video for this chapter is Ch05 – t Test and Welch Two-Sample t Test.mp4. This video pro-
vides overviews of these tests, instructions for carrying out the pretest checklist, and running
and interpreting the results of this test using the following dataset: Ch05_Demo.csv.

OVERVIEW—t TEST

The t test is one of the most commonly used and versatile statistics in experimental and sur-
vey research. The t test is used when there are two groups (e.g., control group and treatment
group, left- and right-handed people, full-time students and part-time students, minors and
adults) wherein a continuous variable is gathered from each participant (e.g., age, height, level
of depression, test score, number of steps taken per day). The t test computes the mean for
each group and then indicates if there’s a statistically significant difference between the two
groups, meaning that one group substantially outperformed the other, or if there’s a statisti-
cally insignificant difference between the groups, meaning the groups performed about the
same.
When the pretest criteria are satisfied, the t test is the proper statistic; otherwise, the bet-
ter choice is the Welch two-sample t test, which is very similar to the t test. The Welch two-
sample t test is explained later in this chapter.

t TESTS AND WELCH TWO-SAMPLE t TESTS IN CONTEXT

Statistical Reasoning
When you have continuous data from two groups (e.g., the ages of people in Group 1 and the
ages of people in Group 2), we could calculate the mean (average) for each group. We reason-
ably expect that the mean age from Group 1 will not be exactly the same as the mean age from
Group 2; we expect them to be different, even if that difference is minuscule. The equations
for the t test and Welch two-sample t test primarily focus on the difference between the means
of the two groups (e.g., mean(Group 1) – mean(Group 2)). The formulas also involve the n
and the standard deviation from each group to arrive at a definitive statistical result: the p
value. If the p value is less than or equal to .05, this suggests that there is a statistically signifi-
cant difference between the mean of Group 1 compared to the mean of Group 2; otherwise,
if the p value is greater than .05, this suggests that even though the mean from Group 1 is dif-
ferent from the mean from Group 2, the difference is statistically insignificant, meaning that
there’s not a substantial difference between the ages of the people in Group 1 compared to the
ages of the people in Group 2.
 Chapter 5 • t Test and Welch Two-Sample t Test 95

Applied Examples
Suppose we wanted to know if it’s better to proofread on paper or video. To answer this
question, we could compose an essay that deliberately includes typographical errors. Then
we could recruit a group of people and assign half to Group 1: They would read the essay
on paper and circle the errors, and we’d record that number of errors that each person
detected. The other half would be assigned to Group 2: They would read the same essay
on a monitor and highlight the errors, and we’d record that number of errors that each
person identified. The t test would compute the mean number of errors that the people in
Group 1 identified and the mean number of errors that the people in Group 2 identified.
Finally, the t test would provide the p value. If that p value is less than or equal to .05, then
we’d conclude that one group found significantly more typos than the other group; if the
p value is greater than .05, then we’d conclude that even though the mean from Group 1 is
not exactly the same as the mean from Group 2, the difference between the two groups is
statistically insignificant.
An instructor could also use this statistic to answer the following question: Do online stu-
dents perform as well as students who take the course on campus in a traditional classroom? This
instructor would assemble a list of the final scores of online students and a list of the final
scores of classroom students. The t test would compute the mean score for each group and a
p value; if the p value is less than or equal to .05, this would suggest that one group statisti-
cally significantly outperformed the other. Alternatively, if the p value is greater than .05, that
would suggest that even though the mean score from the online students is different from the
classroom students, the difference is statistically insignificant, suggesting that both methods
of teaching are equivalent.

When to Use This Statistic

Guidelines for Selecting the t Test

Overview: This statistic compares two groups to determine if one group outperformed the other.

Variables: This test involves a categorical variable indicating the two groups (Labrador Retriever,
Beagle) and a continuous outcome variable (Stress).

Results: A t test revealed that those who had pet therapy with a Labrador Retriever had a statistically
significantly lower stress level (M = 10.64) compared to those who had pet therapy with a
Beagle (M = 11.91), p = .02.

EXAMPLE

To determine the effectiveness of two breeds of certified therapy dogs (Labrador Retriever
or Beagle) in helping stressed clients to relax, a pet therapy center offers stress management
sessions.
96 Part II • Statistical Tests

Groups
A researcher recruits a total of 101 participants who are experiencing stress; upon entering the
facility, each individual is randomly paired with a certified therapy dog—either a Labrador
Retriever or Beagle, thus constituting two groups.

Procedure
Each participant will be instructed that during their 30-minute session, they are welcome to
interact with the dog as they wish (e.g., hold, pet, groom, feed provided treats, snuggle). After
the pet therapy session, participants will be asked to complete the Acme Stress Index, a five-
question self-administered multiple-choice survey that produces a total score between 5 and
25 (5 = very low stress . . . 25 = very high stress); the staff member records the score along with
the breed of the therapy dog and then thanks the individual for their participation.

Research Question
There are several ways to form a research question; to simplify the process, we’ll be phrasing
the research questions as Yes/No questions:
Is there a difference in stress levels among individuals who had Labrador Retriever or Beagle
therapy dogs?

Hypotheses
Think of the hypotheses as the two possible answers to the (Yes/No) research question. The
null hypothesis (H0) is essentially the No answer to the research question: There is no differ-
ence in the stress levels between the two groups (Labrador Retriever or Beagle). The alternative
hypothesis (H1) is the Yes answer to the research question: There is a difference in the stress
levels between the two groups (Labrador Retriever or Beagle). Notice that these two hypotheses
are phrased identically, except for one word: H0 states: There is no difference . . . and H1 states:
There is a difference. . . . Document the hypotheses:

H0: There is no difference in the stress levels between the two groups (Labrador Retriever or
Beagle).
H1: There is a difference in the stress levels between the two groups (Labrador Retriever or
Beagle).
Dataset
Use the following dataset: Ch05_Demo.csv.

Codebook:

Variable: Stress
Definition: Score on Acme Stress Index
Type: Continuous (5 = very low stress . . . 25 = very high stress)
 Chapter 5 • t Test and Welch Two-Sample t Test 97

Variable: Breed
Definition: Dog breed
Type: Categorical (Labrador Retriever, Beagle)

To process a t test or Welch two-sample t test, refer to the R Syntax Guide, page 5; we’ll
proceed in order starting from the top of that page.
First, click on the Packages tab, then check ggplot2, mvnormtest, and psych, and uncheck
Hmisc.

Packages
☑ ggplot2
☐ Hmisc
☑ mvnormtest
☑ psych

t Test Pretest Checklist: ☑ 1. Normality and ☑ 2. Homogeneity of Variance

Inferential statistics, like the ones covered in Chapters 5–9, each have their own set of sta-
tistical assumptions, meaning that in order for the statistical formulas to produce precise
results, we need to verify that the data meet the criteria to run properly through the for-
mula. Perhaps assumption is not the best word for this, in that we need to do more than
passively assume that the data are in suitable shape for each statistical analysis. Actually, we
need to take a more deliberate approach in assessing the contents of the data before running
these analyses; hence, instead of using the term assumptions, this text refers to this process as
running the Pretest Checklist, the results of which will direct us to the best version of the
statistical test to run.
When the conditions on the pretest checklist (statistical assumptions) are satisfied, we can
consider the statistical results to be relatively robust. If there are minor deviations in these
criteria, one could still proceed with the analysis, but we would be a bit less confident in the
strength of our findings. In such instances, it would be appropriate to mention any such sta-
tistical shortcomings when discussing the results.
If the pretest checklist is satisfied, then the parametric version of the statistical test is the
best choice (implying that the data are normally distributed). If one or more of the specified
pretest criteria are substantially not satisfied, the better option is to use the nonparametric ver-
sion of the test (implying that the data are not normally distributed).
In this chapter, if the pretest criteria are met, then we’ll run the t test; otherwise, we’ll run
the Welch two-sample t test, which is covered later in this chapter.
There are two pretest criteria for running a t test: (1) normality and (2) homogeneity (pro-
nounced hoe-moe-juh-nay-it-tee) of variance.
98 Part II • Statistical Tests

Pretest Checklist: Criteria 1, Normality

In statistics, normality pertains to the most common distribution of data within a continu-
ous variable, wherein usually we find that there are few very low values, few very high values,
and most of the values are in the middle, as represented in Figure 5.1. Notice that a normal
distribution resembles a symmetrical hill that peaks in the middle and tapers off at both ends.

FIGURE 5.1 ■ Histogram of a Normal Distribution

Many statistical tests require that the data be normally distributed for the formulas to pro-
duce meaningful results; hence, we’ll see normality included in the pretest checklist (assump-
tions) for the t test and Welch two-sample t test (this chapter), ANOVA—Tukey test and
Wilcoxon multiple pairwise comparisons test (Chapter 6), paired t test and paired Wilcoxon
test (Chapter 7), and correlation—Pearson test and Spearman test (Chapter 8).
While we could generate histograms to visually estimate if the data are normally distrib-
uted, we can make more precise determinations regarding normality by using the Shapiro–
Wilk test. This test compares the data in your continuous variable to data that are normally
distributed to determine if they’re similar enough. This processing takes place internally, so
you won’t see any histograms emerge. Instead, this test reports a p value pertaining to that
(internally processed) comparison:
If the Shapiro–Wilk test p value is less than or equal to .05, this indicates that it found a
statistically significant difference between the distribution of your data and a normal distri-
bution, indicating that your data are not normally distributed.
Alternatively, if the Shapiro–Wilk test p value is greater than .05, this indicates that it
found no statistically significant difference between the distribution of your data and a nor-
mal distribution, indicating that your data are normally distributed.
To summarize the results of the Shapiro–Wilk (normality) test:

• If p ≤ .05, then your data are not normally distributed.

• If p > .05, then your data are normally distributed.
 Chapter 5 • t Test and Welch Two-Sample t Test 99

We need to assess the distribution of the continuous variable (Stress) for each category of
Breed (Beagle and Labrador Retriever) separately; hence, this will be a two-pass process (once
for each breed).
The following box is an excerpt from the R Syntax Guide. Copy and paste the blue R code
from the R Syntax Guide to the R Console window, and edit the underlined items: FileName,
ContinuousVariable, CategoricalVariable. and TargetCategory. Notice that the
underlines do not appear in the R Console window.

Pretest Checklist: Criteria 1, Normality

1. Run the Shapiro–Wilk test for each category:
[Link](subset(FileName$ContinuousVariable,FileName$CategoricalVar
iable=="TargetCategory"))

2. Repeat Step 1 for the other category.

After editing the file name, variable names, and the target category, the R code and results
should look like this:

> [Link](subset(Ch05_Demo$Stress,Ch05_Demo$Breed=="Beagle"))

Shapiro-Wilk normality test

data: subset(Ch06_Demo$Stress, Ch06_Demo$Breed == "Beagle")

W = 0.9794, p-value = 0.5815

Since the p value of .58 is greater than .05, this indicates that the Stress data for the Beagle
category are normally distributed.
We’ll now repeat this step to assess the Stress data for the Labrador Retriever category.
After editing the file name and variable names, the R code and results should look like this:

> [Link](subset(Ch05_Demo$Stress,Ch05_Demo$Breed=="Labrador Retriever"))

Shapiro-Wilk normality test

data: subset(Ch06_Demo$Stress, Ch06_Demo$Breed == "Labrador Retriever")

W = 0.97942, p-value = 0.4636

100 Part II • Statistical Tests

Since the second run of the Shapiro–Wilk test is so similar to the first, to save time, you
can press the up-arrow key until you see the line that begins with [Link] and change
Beagle to Labrador Retriever, then press Enter.
The autocomplete function will not help you when editing the TargetCategory, so be
sure that your entry matches the way the category is spelled in the database exactly (e.g., case,
spacing). Notice that there is one space between the words: Labrador Retriever). Hint: You
can copy and paste the words “Labrador Retriever” from the data table.
Although it is not required, you may opt to view the histograms for each category:

Optional Histograms
Draw a histogram for each group:
ggplot(FileName,aes(x=ContinuousVariable))+geom_histogram()+facet_grid
(CategoricalVariable~.)

After editing the file name and variable names, the R code and histograms should look
like this:

> ggplot(Ch05_Demo,aes(x=Stress))+geom_histogram()+facet_grid(Breed~.)

Pretest Checklist: Criteria 2, Homogeneity of Variance

Next, run the homogeneity of variance test.

Pretest Checklist: Criteria 2, Homogeneity of Variance

[Link](FileName$ContinuousVariable~FileName$CategoricalVariable)
 Chapter 5 • t Test and Welch Two-Sample t Test 101

The formula for the t test requires that the variances (standard deviations squared) of the
two groups can’t be too different from each other. If we run descriptive statistics to analyze
Stress for each group (Breed), we would see that the standard deviation for Stress is 2.89 in
the Beagle group and 2.58 in the Labrador Retriever group; the corresponding variances are
8.35 and 6.66. Instead of contemplating if these two variances are similar enough (homo-
geneous), we can get a more definitive answer by running the homogeneity of variance test to
determine if there is or is not a statistically significant difference between the variances of
these groups.
After editing the file name and variable names, the R code and results should look like
this:

> [Link](Ch05_Demo$Stress~Ch05_Demo$Breed)

F test to compare two variances

data: Ch05_Demo$Stress by Ch05_Demo$Breed

F = 1.2496, num df = 45, denom df = 54, p-value = 0.4311

alternative hypothesis: true ratio of variances is not equal to 1

95 percent confidence interval:

0.7147374 2.2172737

sample estimates:

ratio of variances

1.249649

To assess the results of the homogeneity of variance test, we look to the p value, which is
.43, and interpret it similar to the way that we used for the Shapiro–Wilk test:

• If p ≤ .05, this indicates that there is a statistically significant difference between the
variances.
• If p > .05, this indicates there is no statistically significant difference between the
variances.

Since .43 is greater than .05, this indicates that there is no statistically significant dif-
ference between the variances of the two groups, suggesting that the variances are similar
enough; hence, per the R Syntax Guide, the homogeneity of variance test is satisfied.
Sometimes the variances of the two groups are significantly different from each other; for
example, consider the histograms in Figure 5.2.
While both groups have a mean of 50, the variances are quite different. In Group 1 (the
top histogram), we see that the scores are fairly close to the mean, with a minimum of 40 to
102 Part II • Statistical Tests

FIGURE 5.2 ■ Two Histograms Demonstrating No Homogeneity of Variance

a maximum of 59, spanning a range of 19 points with a variance of 19.27. Next, observe
Group 2 (the bottom histogram); these scores depart substantially further from the mean
with a minimum of 13 to a maximum of 90, spanning a range of 77 points with a variance of
269.29. Graphically, we can see that the data distribution (spreads) of the scores from these
two groups is very different from each other; hence, it follows that the homogeneity of vari-
ance test renders a p value considerably less than .05, indicating the variances of these two
variables are quite different from each other, and hence the homogeneity of variance is not
satisfied.
Consider another example (Figure 5.3) where the distribution of the numbers within
Group 1 is very similar to the distribution of the numbers in Group 2.

FIGURE 5.3 ■ Two Histograms Demonstrating Homogeneity of Variance

In this case, the variance of Group 1 is 17.98 and the variance of Group 2 is 16.16;
the distributions look fairly similar. The homogeneity of variance test produces a p value of
.4684; since this p value is greater than .05, this tells us that there is no statistically significant
difference between the variances of these two groups, and hence, homogeneity of variance is
satisfied.
 Chapter 5 • t Test and Welch Two-Sample t Test 103

Test Run

SELECT THE t TEST OR WELCH TWO-SAMPLE t

TEST

Are all of these criteria met?

1. Normality: Shapiro–Wilk test p > .05 for both groups
2. Homogeneity of variance p > .05
If YES: Run the t test
If NO: Run the Welch two-sample t test

Test Run: t Test

Since the Shapiro–Wilk tests for both groups and the homogeneity of variance tests all produced
p values greater than .05, we can proceed with the t test:

Test Run: t Test

[Link](FileName$ContinuousVariable~FileName$CategoricalVariable,var.
equal=T)

After editing the file name and variable names, the R code and results should look like
this:

> [Link](Ch05_Demo$Stress~Ch05_Demo$Breed,[Link]=T)

Two Sample t-test

data: Ch05_Demo$Stress by Ch05_Demo$Breed

t = 2.3429, df = 99, p-value = 0.02114

alternative hypothesis: true difference in means between group Beagle and group
Labrador Retriever is not equal to 0

95 percent confidence interval:

0.1954381 2.3579216

sample estimates:

mean in group Beagle mean in group Labrador Retriever

11.91304 10.63636
104 Part II • Statistical Tests

Test Run: Welch Two-Sample t Test

If the pretest criteria were not satisfied for the t test, meaning that either of the groups pro-
duced a Shapiro–Wilk test p value that was less than or equal to .05 or if the homogeneity of
variance test produced a p value that was less than or equal to .05, then run the nonparametric
Welch two-sample t test:

Test Run: Welch Two-Sample t Test

[Link](FileName$ContinuousVariable~FileName$CategoricalVariable)

After editing the file name and variable names, the R code and results should look like
this:

> [Link](Ch05_Demo$Stress~Ch05_Demo$Breed)

Welch Two Sample t-test

data: Ch05_Demo$Stress by Ch05_Demo$Breed

t = 2.3196, df = 91.309, p-value = 0.02259

alternative hypothesis: true difference in means between group Beagle and group
Labrador Retriever is not equal to 0

95 percent confidence interval:

0.1834433 2.3699164

sample estimates:

mean in group Beagle mean in group Labrador Retriever

11.91304 10.63636

Unsurprisingly, the t test and the Welch two-sample t test produce the same results for the
Stress means: M(Beagle) = 11.91 and M(Labrador Retriever) = 10.64, but the p values are dif-
ferent: Whereas the t test produced a p value of .02114, the Welch two-sample t test produced
a p value of .02259. In both cases, the p values round to .02.

Test Run: Descriptive Statistics

If in addition to the p value and the means for the two groups, you wish to include further
descriptive statistics pertaining to each group in your documentation (e.g., n, standard devia-
tion), you can run descriptive statistics for each group.
 Chapter 5 • t Test and Welch Two-Sample t Test 105

Test Run: Descriptive Statistics for Each Group

describeBy(FileName$ContinuousVariable,FileName$CategoricalVariable)

After editing the file name and variable names, the R code and results should look like
this:

> describeBy(Ch05_Demo$Stress,Ch05_Demo$Breed)

Descriptive statistics by group

group: Beagle

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 46 11.91 2.89 12 11.95 2.97 6 18 12 -0.03 -0.44 0.43

------------------------------------------------------------------------

group: Labrador Retriever

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 55 10.64 2.58 11 10.67 2.97 5 16 11 -0.03 -0.58 0.35

In this demonstration, we ran both the (parametric) t test and the (nonparametric) Welch
two-sample t test to provide comprehensive learning, but in a nonacademic setting, we would
use the results of the pretest criteria to guide us to select the one test that we would run.
The remaining chapters provide instructions for assessing the pretest criteria, enabling you to
choose the appropriate version of the statistical test (parametric/nonparametric) along with
demonstrations of both tests.

Results
Since the pretest criteria were both satisfied, we’ll proceed using the results from the t test (not
the Welch two-sample t test). The t test produced three results that are of interest to us (from
this point forward, we will be rounding results to two decimal digits):

1. The mean Stress from the Beagle group = 11.91.

2. The mean Stress from the Labrador Retriever group = 10.64.
3. The p value = .02.
106 Part II • Statistical Tests

As expected, the mean Stress level of the two groups is different; the mean Stress score
for those who had pet therapy with a Beagle was 11.91, and the mean Stress score for those
who had pet therapy with a Labrador Retriever was 10.64. The remaining question is: Are
these means statistically significantly different from each other? In other words, Are the differ-
ences between these mean scores merely due to some random minor variability, or does this suggest
that those who had pet therapy with Labrador Retrievers have substantially different Stress scores
compared to those who had pet therapy with a Beagle? Fortunately, we don’t need to stare at these
numbers and try to make that determination; for the answer to that question, we turn to the
p value, which is .02, and apply the same protocol that we used when we ran the homogeneity
of variance test:

• If p ≤ .05, this indicates that there is a statistically significant difference between the
groups.
• If p > .05, this indicates there is no statistically significant difference between the
groups.

In this case, the t test produced a p value of .02; since this is less than or equal to .05, this
indicates that there is a statistically significant difference between the means of these two
groups. We can now proceed to resolve the hypotheses.

Hypothesis Resolution
The p value guides us in selecting which hypothesis to accept using the following rules:

• If p ≤ .05, accept the alternate hypothesis (H1).

• If p > .05, accept the null hypothesis (H0).

Per these rules, we can now revisit our hypotheses to make an informed selection:

REJECT H0: There is no difference in the stress levels between the two groups
(Labrador Retriever or Beagle).
ACCEPT H1: There is a difference in the stress levels between the two groups
(Labrador Retriever or Beagle).

Since the t test produced a p value of .02, and .02 is less than (or equal to) .05, we would
accept H1, which tells us that there is a difference in the stress levels between the two groups
(Labrador Retriever or Beagle). Considering the mean Stress levels of the two groups
(Beagle = 11.91 and Labrador Retriever = 10.64), we can be more specific, indicating that
Labrador Retrievers statistically significantly outperformed Beagles in reducing stress in pet
therapy.
 Chapter 5 • t Test and Welch Two-Sample t Test 107

Documenting Results
Although it is essential to comprehend the meaning of the key values in the statistical reports,
it would be inappropriate to simply present the figures in a results section without providing
a meaningful narrative. While all of the following figures are technically correct, try to avoid
documenting your findings as such:

Accurate but Inappropriate Numerical Results

Beagle: n = 46, M = 11.91 (SD = 2.89)
Labrador Retriever: n = 55, M = 10.64 (SD = 2.58)
p = .02 ∴ Labrador Retrievers significantly outperformed Beagles in stress reduction
NOTE: The ∴ symbol means therefore.

While the above results report is concise and accurate, it lacks contextual details that tell
the story of this research project. A better way to communicate results is to write a traditional
abstract. Journal abstracts are typically under 200 words, beginning with the research question,
the methods detailing the participants and how they were handled/tested, the statistical results,
and the implications of these findings. For exemplary purposes, these bold section labels are pre-
sented in the following abstract, but such headers are not always included in abstracts. Compare
this abstract, which is about 150 words, to the prior form of documentation:

Appropriate Statistical Abstract

[RESEARCH QUESTION] To determine which breed of therapy dog was most effective in
reducing stress, [METHODS] we recruited 101 participants who reported that they were expe-
riencing high stress. We randomly assigned participants to engage in 30 minutes of pet therapy
with either a Beagle (n = 46) or a Labrador Retriever (n = 55). After each session, participants
completed the Acme Stress Index, a five-question self-administered multiple-choice survey that
produces a total score between 5 and 25 (5 = very low stress . . . 25 = very high stress).
[STATISTICAL RESULTS] We conducted a t test, revealing that those who had pet
therapy with a Labrador Retriever had a statistically significantly lower average stress level (M
= 10.64) compared to those who had pet therapy with a Beagle (M = 11.91), p = .02.
[IMPLICATIONS] Based on these findings, we are considering including more Labrador
Retrievers in our program.

Statistics in an Imperfect World

Although it may seem like a surprising concept, the purpose of statistics is not to prove or dis-
prove anything with absolute confidence; rather, it’s a system designed to help reduce uncer-
tainty and, in some cases, make more informed decisions or predictions.
108 Part II • Statistical Tests

Let’s consider the results of the statistics that we just ran, which found that people
have lower stress levels after pet therapy with a Labrador Retriever compared to a Beagle.
Clearly, that’s what the numbers are showing us, but consider some other factors that
might have influenced the stress level: Maybe the Labrador Retriever had a name that
matched a benevolent public figure—this could have facilitated more positive bonding
with the dog. Alternatively, maybe people just liked the color of the Labrador Retriever’s
fur more than the Beagle’s. It’s also conceivable that despite randomly assigning par-
ticipants to the dog (Beagle or Labrador Retriever), it’s possible that more of the partici-
pants with lower levels of preexisting stress were unintentionally routed to the Labrador
Retriever group. If that’s the case, then we have unknowingly biased the groups; it may
be that the pet therapy is completely ineffective in lowering stress and that we are being
misled by the results.
Although research scientists and statisticians are challenged with designing and
implementing studies that have as few confounding factors as possible, it may not be
plausible to identify and resolve every possible condition; as such, statistical results are
generally documented using provisional, rather than definitive, language. For example,
instead of stating, These results prove that Labrador Retrievers are superior to Beagles when
it comes to reducing stress, a more tentative phrasing would be more appropriate: These
results suggest that Labrador Retrievers are a better option compared to Beagles when it comes
to reducing stress.

p Value
We have referred to the p value to guide us in determining if one breed of dog significantly
outperformed the other in relaxing participants. Essentially, the p value communicates the
likelihood that such results would occur by chance alone. In this case, the p value is .02,
indicating that there is a 2% probability that we would get this result (Beagles = 11.91 and
Labrador Retrievers = 10.64) just by (lucky) chance alone, which seems fairly unlikely.
The low p value of .02 suggests that there is probably something about the dog breed that
accounts for the observed differences in the Stress variable among the participants in the
two groups.
In this example, we’ve used .05 as the cutoff number for the p value; this is the alpha level
(or α level). In most statistical research, the alpha level is set to .05, which is the default in
most statistical software. We’ll be processing statistics using the .05 alpha level, meaning that
p values that are less than or equal to 0.05 are regarded as statistically significant, and p values
that are greater than .05 are regarded as statistically insignificant.
You might see very low p values in statistical reports (e.g., p = .0060609); while you
could document this lengthy number as is (p = .0060609), another option is to write it
as p < .01, which can be a better fit when assembling a results table. To be sure that you’re
seeing the full p value, be sure that you’ve switched R from the default (scientific nota-
tion) to standard notation using the options command on page 1 of the R Syntax Guide:
options(scipen=999), which will present the p values in a more readable format (without
the “e” notation).
 Chapter 5 • t Test and Welch Two-Sample t Test   109

TYPE I AND TYPE II ERRORS

Even with a quality research design, properly processed statistics, and a strong understanding
of the p value, it’s still possible to be misled by statistical results. The world is complex and
imperfect; despite our best efforts, errors can occur in virtually any decision-making realm no
matter how careful you are. Consider the two types of errors that can occur in a legal verdict:

Error I: The court finds the defendant guilty when, in fact, the defendant is actually not
guilty.
Error II: The court finds the defendant not guilty when, in fact, the defendant actually is
guilty.

These same two types of errors can happen in statistics. Consider this standard set of hypothe-
ses pertaining to an experiment involving two groups: A control group and a treatment group:

H0: There is no significant difference between the groups (p > .05; the treatment failed).
H1: There is a statistically significant difference between the groups (p ≤ .05; the treat-
ment worked).

Type I Error
When p ≤ .05, it is possible that the results may reflect a Type I error ([α] error). Although
the results indicate that there is a statistically significant difference between two variables (or
groups), on the whole, there actually is not, meaning that you would wrongly reject the null
hypothesis. You would conclude that the groups performed differently when, in fact, it only
occurred that way this time—some other factor(s) may have affected the results (e.g., out-
side influences may have affected the outcome, random assignment may have failed to create
appropriately balanced groups).
The p value indicates the probability that your statistically significant results may be
attributable to a Type I error. In this example, p = .02; since this is less than or equal to .05,
we conclude that those who had pet therapy with a Labrador Retriever had statistically signifi-
cantly lower stress than those who had pet therapy with a Beagle, but the p value of .02 indicates
that there is a 2% chance that we’re wrong about that determination—this would be a Type I
error. In summary, the lower the p value, the less likely it is that the results are attributable to
a Type I error, meaning that we can have more confidence in our findings with lower p values.
A Type I error can be thought of as a false positive such as a medical test indicating that the
patient has a disease when, in fact, the patient actually does not have the disease, or a court that
finds the defendant guilty when, in fact, the defendant actually is not guilty.

Type II Error
When p > .05, it is possible that the results may reflect a Type II error ([β] error). In this
example involving pet therapy, suppose the p value = .36; since this is greater than .05, we
110 Part II • Statistical Tests

would conclude that there is no statistically significant difference between the stress scores
of those who had pet therapy with a Labrador Retriever compared to a Beagle when, on
the whole, there actually is a difference, meaning that you would wrongly accept the null
hypothesis.
Sample size is inversely related to Type II errors, meaning that smaller samples are more
vulnerable to Type II errors; as such, one strategy for reducing the likelihood of committing a
Type II error is to attain a larger sample.
A Type II error, or false negative, is akin to a medical test indicating that the patient does
not have a disease when, in fact, the patient actually does have the disease, or a court that finds
the defendant not guilty when, in fact, the defendant actually is guilty.
There is no analysis to definitively tell you if your results contain a Type I or Type II error;
such errors are just endemic in the realm of statistical testing. The point to keep in mind
is that even if a statistical test produces a statistically significant p value (e.g., p ≤ .05), this
does not mean that you have solid evidentiary proof of anything; at best, you have reduced
uncertainty.
Since the p value never goes to zero, there is always some level of uncertainty in statisti-
cal findings. This concept is not limited to statistics; for example, no matter how careful you
are, there’s some chance that you could be involved in an automobile accident (it may even be
someone else’s fault), but the probability of you reaching your destination on the road safely is
never 100% certain no matter how careful you are.

GOOD COMMON SENSE

The t test is designed to detect statistically significant differences between the means of two
groups; it does not matter which group mean is higher and which is lower. In this case, the
goal was to decrease stress, but the t test is equally applicable where the goal is to increase a vari-
able, such as salary or number of words read per minute.
Considering the example that we processed, although we attained and interpreted the
results with competent precision, we don’t necessarily stop there. Often, statistical results
are not the only thing we consider when making informed decisions; rather, they are part of
a larger picture in the decision-making process. Continuing with the example that we pro-
cessed, our statistical results show us that Labrador Retrievers are the better choice for reduc-
ing stress compared to Beagles, but let’s consider some other factors: Suppose the animal
handler can provide the Beagles on a daily basis, but the Labrador Retrievers are only available
every other Thursday for 3 hours each day, or maybe there’s a prohibitively higher fee for the
Labrador Retrievers. While the statistical findings are relevant, they should be considered
alongside these and other real-world factors when making decisions regarding what breed of
therapy dogs this facility should opt for.
The point is that although we conduct and document statistical results precisely, we’re not
necessarily obliged to fully turn our decision-making process over to the computer. It’s often
the case that the statistics serve to better inform us in making a mindful decision along with
other relevant factors.
 Chapter 5 • t Test and Welch Two-Sample t Test 111

KEY CONCEPTS

• t test
• Pretest checklist
• Normality
• Homogeneity of variance
• α
• p
• Welch two-sample t test
• Hypothesis resolution
• Documenting results
• Type I (α) error
• Type II (β) error
• Good common sense

PRACTICE EXERCISES

Download the datasets from the Sage companion website to complete these exercises.

Exercise 5.1
To determine the best proofreading method, the director of a writing lab has created a five-page
document with 15 writing errors. Participants will be recruited and assigned to read the essay
on paper or computer screen. Participants will be asked to detect as many errors as they can.
Dataset: Ch05_Ex01A.csv
Codebook:
Variable: Errors
Definition: Number of writing errors detected
Type: Continuous (0 . . .15)

Variable: ReadingMethod
Definition: Media that the essay was provided on
Type: Categorical (Paper, Computer)

a. Write the research question.

b. Write the hypotheses.
c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
112 Part II • Statistical Tests

d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch05_Ex01B.csv.

Exercise 5.2
A therapist wants to determine the best method for reducing depression. Participants are
recruited and assigned to one of two groups: Talk therapy or Antidepressant medication
(Rx). After 10 weeks, all participants will complete the Acme Depression Instrument (1 =
low depression . . . 20 = high depression).
Dataset: Ch05_Ex02A.csv
Codebook:
Variable: Depression
Definition: Score on the Acme Depression Instrument
Type: Continuous (1 = low depression . . . 20 = high depression)

Variable: Therapy
Definition: Type of therapeutic intervention
Type: Categorical (Talk, Rx)

a. Write the research question.

b. Write the hypotheses.
c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch05_Ex02B.csv.

Exercise 5.3
A nurse who wants to determine the best treatment for reducing stress recruits a group of
participants and assigns them to one of two groups: Group 1 will sit quietly; Group 2 will
receive guided meditation. After 30 minutes, participants will complete the Acme Stress
Instrument (1 = very low stress . . . 7 = very high stress).
 Chapter 5 • t Test and Welch Two-Sample t Test 113

Dataset: Ch05_Ex03A.csv
Codebook:
Variable: Stress
Definition: Score on the Acme Stress Instrument
Type: Continuous (1 = very low stress . . . 7 = very high stress)

Variable: Treatment
Definition: Stress treatment
Type: Categorical (Sit quietly, Guided meditation)

a. Write the research question.

b. Write the hypotheses.
c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch05_Ex03B.csv.

Exercise 5.4
To determine the best teaching method, an educational researcher gathers the final scores of
students taking the same course two different ways: Traditional classroom or Online with a
live instructor.
Dataset: Ch05_Ex04A.csv
Codebook:
Variable: Final
Definition: Score on the final
Type: Continuous (0 . . . 100)

Variable: TeachingMethod
Definition: How the course was taught
Type: Categorical (Classroom, Online live)
a. Write the research question.
b. Write the hypotheses.
c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
114 Part II • Statistical Tests

d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch05_Ex04B.csv.

Exercise 5.5
To better inform class project planning, an art teacher visits a gallery and unobtrusively uses
a stopwatch to time how long people spend looking at two different artforms: Paintings and
Mobiles.
Dataset: Ch05_Ex05A.csv
Codebook:
Variable: Seconds
Definition: Time a person spent looking at an art piece
Type: Continuous (0 . . . no limit)

Variable: Artform
Definition: Type of art
Type: Categorical (Painting, Mobile)

a. Write the research question.

b. Write the hypotheses.
c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch05_Ex05B.csv.

Exercise 5.6
An industrial designer has created a desk that requires consumer assembly. To determine
the optimal form of assembly instructions, the designer recruits participants one at a time
and provides them with the desk parts in the box, the required screwdriver, and assembly
instructions in the form of a booklet with text instructions or a booklet containing text
with illustration instructions. The designer times how long it takes each person to assemble
the desk.
 Chapter 5 • t Test and Welch Two-Sample t Test 115

Dataset: Ch05_Ex06A.csv
Codebook:
Variable: Minutes
Definition: Number of minutes required to assemble the desk
Type: Continuous (0 . . . no limit)

Variable: Instructions
Definition: Type of assembly instructions provided
Type: Categorical (Text, Text with illustrations)

a. Write the research question.

b. Write the hypotheses.
c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch05_Ex06B.csv.

Exercise 5.7
An academic enrollment director wants to determine if employment status (full-time or part-
time) impacts the number of nonclassroom hours per week students spend on their classwork
(e.g., homework, studying, research, writing). Hours are rounded to the nearest quarter (e.g.,
4 hours and 15 minutes = 4.25 hours).
Dataset: Ch05_Ex07A.csv
Codebook:
Variable: Hours
Definition: Number of nonclassroom hours spent on classwork
Type: Continuous (0 . . . no limit)

Variable: Employment
Definition: Employment status
Type: Categorical (Full-time, Part-time)

a. Write the research question.

b. Write the hypotheses.
116 Part II • Statistical Tests

c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch05_Ex07B.csv.

Exercise 5.8
A farmer wants to determine the effect that soil nutrients have on lemon trees. In Orchard
1, the farmer waters the trees as usual, and in Orchard 2, the farmer provides water and
soil Nutrient A. At harvest time, the farmer will randomly select lemons from each of the
orchards and weigh them.
Dataset: Ch05_Ex08A.csv
Codebook:
Variable: Grams
Definition: Weight of the lemon in grams
Type: Continuous (1 . . . no limit)

Variable: Nutrient
Definition: Soil nutrient
Type: Categorical (None, Nutrient A)

a. Write the research question.

b. Write the hypotheses.
c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch05_Ex08B.csv.

Exercise 5.9
The board of directors of an amusement park wants to compare the popularity of two rides:
Haunted house and the Merry-go-round. They gather the daily number of riders from the
turnstile counters at the end of the day for 4 weeks (28 days).
 Chapter 5 • t Test and Welch Two-Sample t Test 117

Dataset: Ch05_Ex09A.csv
Codebook:

Variable: Riders
Definition: Number of daily riders
Type: Continuous (0 . . . no limit)

Variable: Attraction
Definition: Amusement park ride
Type: Categorical (Haunted house, Merry-go-round)

a. Write the research question.

b. Write the hypotheses.
c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch05_Ex09B.csv.

Exercise 5.10
The math department administers a common final to all students taking Statistics 210. The
department chair will compare the scores among the two instructors who taught the course
this term: Professor Anderson and Professor Baker.
Dataset: Ch05_Ex10A.csv
Codebook:
Variable: Final
Definition: Score on the final
Type: Continuous (0 . . . 100)

Variable: Professor
Definition: Professor who taught the course
Type: Categorical (Anderson, Baker)

a. Write the research question.

b. Write the hypotheses.
118 Part II • Statistical Tests

c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch05_Ex10B.csv.
 ANOVA—TUKEY TEST AND
6 WILCOXON MULTIPLE
PAIRWISE COMPARISONS TEST
When you have three or more groups measured with a continuous variable, you can deter-
mine which group outperformed which with an ANOVA Tukey test or Wilcoxon multiple
pairwise comparisons test.

Group 1 Group 2

Group 3

LEARNING OBJECTIVES

Upon completing this chapter, you will be able to:

6.1 Determine when to use the ANOVA test
6.2 Build a research question and corresponding hypotheses
6.3 Run and assess the pretest criteria: normality and homogeneity of variance
6.4 Select and run the proper version of the ANOVA test: the Tukey test or the
Wilcoxon multiple comparisons test
6.5 Interpret the results
6.6 Resolve the hypotheses
6.7 Write a concise abstract detailing the research question and results
6.8 Calculate the unique pairs formula

119
120 Part II • Statistical Tests

VIDEO RESOURCE

The video for this chapter is Ch06 – ANOVA – Tukey Test and Wilcoxon Multiple Pairwise
Comparisons Test.mp4. This video provides overviews of these tests, instructions for car-
rying out the pretest checklist, and running and interpreting the results of this test using the
following dataset: Ch06_Demo.csv.

OVERVIEW—ANOVA TEST

Whereas the t test processes two groups of continuous variables to determine if there’s a sta-
tistically significant difference between the group means, the ANOVA test does exactly the
same thing except it can handle three or more groups. The ANOVA test behaves very much like
a t test: It selects two groups at a time, computes the mean for each group, and then produces a
p value to indicate if there’s a statistically significant difference between the means of the two
groups. Basically, ANOVA processes a t test for every possible pair of groups in the dataset.
For example, if there are three groups, ANOVA will run a t test comparing Group 1 to Group
2, another t test comparing Group 1 to Group 3, and, finally, a t test comparing Group 2 to
Group 3.

ANOVA TESTS IN CONTEXT

Statistical Reasoning
The ANOVA (which an acronym for analysis of variance) test is very similar to the t test,
but whereas the t test is limited to comparing the means of two groups to each other, the
ANOVA can compare the means of more than two groups to each other. The process begins
with the omnibus test, which analyzes the variability between each of the groups and the
variability within each of the groups. If the omnibus test returns a p value that is greater than
.05, this indicates that there are no statistically significant differences among the means of
the groups; in other words, all of the groups performed about the same. Alternatively, if the
omnibus test returns a p value that is less than or equal to .05, this indicates that the mean
of at least one group is statistically significantly different from the mean of at least one
other group.
While this overall result is useful, it does not fully answer the following question:
Specifically, which group(s) statistically significantly outperformed which other group(s). To
answer that, we’d run the post hoc test—either the Tukey test or the Wilcoxon multiple pair-
wise comparisons test. The post hoc test will essentially run a t test on every possible (pair)
combination of groups and produce a p value for each comparison. Hence, if there are three
groups, the post hoc test would compute a p value comparing Group 1 to Group 2; next, it
would compute a p value comparing Group 1 to Group 3; and finally, it would compute a
 Chapter 6 • ANOVA—Tukey Test and Wilcoxon Multiple Pairwise Comparisons Test 121

p value comparing Group 2 to Group 3, thereby indicating which group(s) statistically signifi-
cantly outperformed which.

Applied Examples
A psychotherapist who’s interested in discovering the best approach for treating depression
may recruit a group of people diagnosed with depression and divide them into three groups:
Group 1 will get traditional talk therapy, Group 2 will get antidepressive medication, and
Group 3 will get talk therapy with antidepressive medication. At the final appointment,
each person completed a depression assessment instrument (e.g., 1 = low depression . . . 20
= high depression). The psychotherapist could use ANOVA to compare the three groups
to each other to determine if all of the treatment methods resulted in about the same level
of depression or if one group statistically significantly outperformed another group (or
groups).
An administrator has five different instructors teaching a course that involves a common
final. The administrator can use ANOVA to process the final exam scores gathered from each
of the five classes to determine if the students performed about the same in each class or if the
students in one (or more) class performs significantly differently from the others.

When to Use This Statistic

Guidelines for Selecting the ANOVA Test

Overview: This statistic is for designs that involve more than two groups to determine which
group(s) (if any) outperformed another.

Variables: This statistic requires two variables for each record: (1) a categorical variable to
designate the group (Labrador Retriever, Beagle, French Bulldog) and (2) a continuous
variable to contain the outcome score (Stress).

Results: ANOVA revealed that those who had pet therapy with a Labrador Retriever had a statistically
significantly lower stress level (M = 10.64) compared to those who had pet therapy with a
Beagle (M = 11.91), p = .04, or a French Bulldog (M = 11.99), p = .01.

LAYERED LEARNING

Considering how similar the t test and Welch two-sample t test (from Chapter 5) and the
ANOVA Tukey test and Wilcoxon multiple pairwise comparisons test are, if you’re proficient
with the t test statistics, you can consider yourself about 90% capable of running an ANOVA
test. The Wilcoxon multiple pairwise comparisons test is the nonparametric version of the
(parametric) Tukey test, which is used if the pretest criteria for the Tukey test are not met.
These two versions of the ANOVA test provide a different set of p values.
This chapter uses the same demonstration and exercises from Chapter 5 with a third
group included in each dataset, except for Exercise 9, which has four groups, and Exercise 10,
which has five groups.
122 Part II • Statistical Tests

EXAMPLE

To determine the effectiveness of three breeds of certified therapy dogs (Labrador Retriever,
Beagle, French Bulldog) in helping stressed clients to relax, a pet therapy center offers stress
management sessions.

Groups
A researcher recruits a total of 172 participants who are experiencing stress; upon entering
the facility, each individual is randomly paired with a certified therapy dog—a Labrador
Retriever, Beagle, or a French Bulldog, thus constituting three groups.

Procedure
Each participant will be instructed that during their 30-minute session, they are welcome
to interact with the dog as they wish (e.g., hold, pet, groom, feed provided treats, snug-
gle). After the pet therapy session, participants will be asked to complete the Acme Stress
Index, a five-question self-administered multiple-choice survey that produces a total score
between 5 and 25 (5 = very low stress . . . 25 = very high stress); the staff member records
the score along with the breed of the therapy dog and then thanks the individual for their
participation.

Research Question

Is there a difference in stress levels among individuals who had Labrador Retriever, Beagle,
or French Bulldog therapy dogs?

Hypotheses

H0: There is no difference in the stress levels between the three groups (Labrador
Retriever, Beagle, French Bulldog).
H 1 : There is a difference in the stress levels between the three groups (Labrador Retriever,
Beagle, French Bulldog).

Dataset
Use the following dataset: Ch06_Demo.csv.
Codebook:
Variable: Stress
Definition: Score on Acme Stress Index
Type: Continuous (5 = very low stress . . . 25 = very high stress)
 Chapter 6 • ANOVA—Tukey Test and Wilcoxon Multiple Pairwise Comparisons Test 123

Variable: Breed
Definition: Dog breed
Type: Categorical (Labrador Retriever, Beagle, French Bulldog)

To process an ANOVA—Tukey test or Wilcoxon multiple pairwise comparisons test,

refer to the R Syntax Guide, page 6; we’ll proceed in order starting from the top of that page.
First, click on the Packages tab; then check car, ggplot2, mvnormtest, and psych; and
uncheck Hmisc.

Packages
☑ car
☑ ggplot2
☐ Hmisc
☑ mvnormtest
☑ psych

ANOVA Test Pretest Checklist: ☑ 1. Normality and ☑ 2. Homogeneity of

Variance
These procedures are the same as we ran for the t test in Chapter 5, except these instructions
pertain to a design involving three groups, not two.

Pretest Checklist: Criteria 1, Normality

Pretest Checklist: Criteria 1, Normality

1. Run the Shapiro–Wilk test for each category:
[Link](subset(FileName$ContinuousVariable,FileName$Categorical
Variable==“TargetCategory”))

2. Repeat Step 1 for the other categories.

Since there are three groups (Beagle, Labrador Retriever, French Bulldog), you’ll run this
code three times, changing the TargetCategory each time.
After editing the file name, variable names, the R code and results should look like this:

> [Link](subset(Ch06_Demo$Stress,Ch06_Demo$Breed=="Beagle"))
124 Part II • Statistical Tests

Shapiro-Wilk normality test

data: subset(Ch06_Demo$Stress, Ch06_Demo$Breed == "Beagle")

W = 0.9794, p-value = 0.5815

> [Link](subset(Ch06_Demo$Stress,Ch06_Demo$Breed=="Labrador Retriever"))

Shapiro-Wilk normality test

data: subset(Ch06_Demo$Stress, Ch06_Demo$Breed == "Labrador Retriever")

W = 0.97942, p-value = 0.4636

> [Link](subset(Ch06_Demo$Stress,Ch06_Demo$Breed=="French Bulldog"))

Shapiro-Wilk normality test

data: subset(Ch06_Demo$Stress, Ch06_Demo$Breed == "French Bulldog")

W = 0.97562, p-value = 0.1818

After you’ve run the first analysis for the Beagle group, you may use the up- and
down-arrow keys to provide you with prior lines of code, wherein you can simply edit the
TargetCategory (for Labrador Retriever and French Bulldog) and then press Enter to run the
next [Link] command.
Since the Shapiro–Wilk test produced p values greater than .05 for each category, this
indicates that the data in each group are normally distributed, and hence the normality crite-
rion is satisfied.
Although it is not required, you may opt to view the histograms for each category:

Optional Histograms
Draw a histogram for each group:
ggplot(FileName,aes(x=ContinuousVariable))+geom_histogram()+facet_grid
(CategoricalVariable~.)

After editing the file name and variable names, the R code and histograms should look
like this:
 Chapter 6 • ANOVA—Tukey Test and Wilcoxon Multiple Pairwise Comparisons Test 125

> ggplot(Ch06_Demo,aes(x=Stress))+geom_histogram()+facet_grid(Breed~.)

Pretest Checklist: Criteria 2, Homogeneity of Variance

Next, run the homogeneity of variance test.

Pretest Checklist: Criteria 2, Homogeneity of Variance

Build a vector (variable):
x<-aov(FileName$ContinuousVariable~FileName$CategoricalVariable)

Run the Levene test for homogeneity of variance:

leveneTest(x)

After editing the file name and variable names (for the first line of code only), the R code
and homogeneity of variance test results should look like this:

> x<-aov(Ch06_Demo$Stress~Ch06_Demo$Breed)

> leveneTest(x)
Levene's Test for Homogeneity of Variance (center = median)
126 Part II • Statistical Tests

Df F value Pr(>F)

group 2 0.58 0.561

169

To assess the results of the homogeneity of variance test, which assesses the variances of all
groups, we look to the Pr result; this is the p value, which is .56 (rounded from .561). We’ll use
the same protocol for evaluating this p value as always:

• If p ≤ .05, this indicates that there is a statistically significant difference between the
variances.
• If p > .05, this indicates that no statistically significant differences have been
detected between any of the groups.

Since the p value of .56 is greater than .05, this indicates that there is no statistically sig-
nificant difference between the variances of the three groups, suggesting that the variances
are similar enough.

Test Run
Based on these findings, we can now select the appropriate version of the ANOVA test:

SELECT THE TUKEY TEST OR WILCOXON

MULTIPLE PAIRWISE COMPARISONS TEST

Are all of these criteria met?

1. Normality: Shapiro–Wilk test p > .05 for all groups
2. Homogeneity of variance p > .05
If YES: Run the Tukey test.
If NO: Run the Wilcoxon multiple pairwise comparisons test.

Since the Shapiro–Wilk tests for all of the groups and the homogeneity of variance tests all
produced p values greater than .05, we can proceed with the t test omnibus test.

Test Run: t Test Omnibus Test

Test Run: t Test Omnibus Test

Run the t test omnibus test (Pr = p value):
[Link](x)
 Chapter 6 • ANOVA—Tukey Test and Wilcoxon Multiple Pairwise Comparisons Test 127

After copying the R code for the t test omnibus test (no editing required), the results
should look like this:

> [Link](x)

Df Sum Sq Mean Sq F value Pr(>F)

Ch06_Demo$Breed 2 65.4 32.71 4.704 0.0103 *

Residuals 169 1175.4 6.95

---

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

To assess the results of the t test omnibus test, we look to the Pr result; this is the p value, which
is .01 (rounded from .0103). We’ll use the same protocol for evaluating this p value as always:

• If p ≤ .05, this indicates that a statistically significant difference has been detected
between at least one pair of the groups.
• If p > .05, this indicates that no statistically significant difference has been detected
between any of the groups.

Since the p value of .01 is less than .05 on the t test omnibus test, this indicates that we can
expect to see a statistically significant difference between at least one pair of groups, but the
question remains: Which group(s) performed differently from which group(s)? Before we pursue
that answer, consider the concept of pairwise combinations. Suppose we have a tandem bike
and two kids, Aaron and Blake (Figure 6.1).

FIGURE 6.1 ■ t Test Analogy: Two Groups Means a Single Possible Pair

Since there are two kids, only one pair can ride the tandem bike: Aaron and Blake can
ride together as a pair. This is akin to the t test: When there are two groups, there is only one
128 Part II • Statistical Tests

possible pairing for the t test to process. It will compare the mean of Group 1 to the mean of
Group 2 and produce a p value. However, in the ANOVA realm, we are dealing with three or
more groups. Continuing with our example, if instead of two kids (Aaron and Blake), what
if we had three kids (Aaron, Blake, and Clair)? With three kids, there are three ways that we
could pair up the kids to ride the tandem bike: Pair 1: Aaron and Blake, Pair 2: Aaron and
Clair, and Pair 3: Blake and Clair (see Figure 6.2).

FIGURE 6.2 ■ ANOVA Test Analogy: Three Groups Means Three Possible Pairs

This is essentially how the ANOVA test works; the ANOVA test will systematically select
two groups at a time and run a t test that produces a p value indicating if one group statisti-
cally significantly outperformed the other. It will repeat this process, analyzing every possible
paired of groups.

Test Run: Tukey Test/Descriptive Statistics for ANOVA Analysis

Unlike the t test, which produces the p value and the means for both groups, the ANOVA test
calculates the p values for each pair of groups, but it does not automatically display the means
for each group. As such, we’ll need to take two separate steps to run the ANOVA analysis: (1)
Run the Tukey test, which will provide the p values for each combination of pairs, and (2) run
the descriptive statistics for each group, which will give us the means for each group.

Test Run: Descriptive Statistics for Each Group and the Tukey Test
Run the Tukey test (p adj = p values for the group pairwise means comparisons tests):
TukeyHSD(x)

Run descriptive statistics for each group:

describeBy(FileName$ContinuousVariable,FileName$CategoricalVariable)
 Chapter 6 • ANOVA—Tukey Test and Wilcoxon Multiple Pairwise Comparisons Test 129

Copy the R code for the Tukey test (no editing required) and descriptive statistics for
each group; after editing the file name and variable names, the results should look like this:

> TukeyHSD(x)

Tukey multiple comparisons of means

95% family-wise confidence level

Fit: aov(formula = Ch06_Demo$Stress ~ Ch06_Demo$Breed)

$`Ch06_Demo$Breed`

diff lwr upr p adj

French Bulldog-Beagle 0.07287201 -1.107383 1.25312705 0.9883191

Labrador Retriever-Beagle -1.27667984 -2.522604 -0.03075614 0.0432297

Labrador Retriever-French Bulldog -1.34955186 -2.469675 -0.22942870 0.0136073

> describeBy(Ch06_Demo$Stress,Ch06_Demo$Breed)

Descriptive statistics by group

group: Beagle

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 46 11.91 2.89 12 11.95 2.97 6 18 12 -0.03 -0.44 0.43

---------------------------------------------------------------------------

group: French Bulldog

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 71 11.99 2.5 12 12.04 2.97 6 18 12 -0.18 -0.28 0.3

---------------------------------------------------------------------------

group: Labrador Retriever

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 55 10.64 2.58 11 10.67 2.97 5 16 11 -0.03 -0.58 0.35

Results: Tukey Test/Descriptive Statistics for ANOVA Analysis

Referring to the above output, notice that there are two reports: (1) The top of the report
contains results of the Tukey test, which gives us the p value (p adj) for each pair of groups. (2)
The bottom of the report (after the describeBy command) contains the descriptive statistics
for each group, which provides the mean of each group.
130 Part II • Statistical Tests

We’ll begin at the top section of the Tukey test: The first pairing compares French Bulldog
to Beagle; we see that this comparison produced a p value (p adj) of. 9883191, which we’ll
round to p = .99. Next, we look to the descriptive statistics at the bottom and find the relevant
means; French Bulldog had a mean of 11.99, and Beagle had a mean of 11.91. We’ll use these
results to begin building a concise summary report. After we’ve completed the analysis, this
table will serve as the basis for understanding the results of the ANOVA test, which will make
it easier to write the abstract:

Group Means p

M(French Bulldog) = 11.99 : M(Beagle) = 11.91 .99

Referring to this same results report, we move down one row in the Tukey test table and
see that the next pairwise comparison is between Labrador Retriever and Beagle. Here we see
that the p value (p adj) is. 0432297, which we will round to .04.

> TukeyHSD(x)

Tukey multiple comparisons of means

95% family-wise confidence level

Fit: aov(formula = Ch06_Demo$Stress ~ Ch06_Demo$Breed)

$`Ch06_Demo$Breed`

diff lwr upr p adj

French Bulldog-Beagle 0.07287201 -1.107383 1.25312705 0.9883191

Labrador Retriever-Beagle -1.27667984 -2.522604 -0.03075614 0.0432297

Labrador Retriever-French Bulldog -1.34955186 -2.469675 -0.22942870 0.0136073

> describeBy(Ch06_Demo$Stress,Ch06_Demo$Breed)

Descriptive statistics by group

group: Beagle
vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 46 11.91 2.89 12 11.95 2.97 6 18 12 -0.03 -0.44 0.43

---------------------------------------------------------------------------

group: French Bulldog

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 71 11.99 2.5 12 12.04 2.97 6 18 12 -0.18 -0.28 0.3

---------------------------------------------------------------------------
 Chapter 6 • ANOVA—Tukey Test and Wilcoxon Multiple Pairwise Comparisons Test 131

group: Labrador Retriever

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 55 10.64 2.58 11 10.67 2.97 5 16 11 -0.03 -0.58 0.35

Then we look to the descriptive statistics report at the bottom and see that Labrador
Retriever had a mean of 10.64 and Beagle had a mean of 11.91. We can now include these results
as the next line in our summary report. When documenting statistical results in tables, it’s
traditional to denote statistically significant results, where p < .05 is indicated with an asterisk:

Group Means p

M(French Bulldog) = 11.99 : M(Beagle) = 11.91 .99

M(Labrador Retriever) = 10.64 : M(Beagle) = 11.91 *.04

Finally, we’ll step down to the last row of the Tukey test, which compares the third pair:
Labrador Retriever to French Bulldog. Here we see that the p value (p adj) is 0. 0136073, which
we will round to 0.01.

> TukeyHSD(x)

Tukey multiple comparisons of means

95% family-wise confidence level

Fit: aov(formula = Ch06_Demo$Stress ~ Ch06_Demo$Breed)

$`Ch06_Demo$Breed`

diff lwr upr p adj

French Bulldog-Beagle 0.07287201 -1.107383 1.25312705 0.9883191

Labrador Retriever-Beagle -1.27667984 -2.522604 -0.03075614 0.0432297

Labrador Retriever-French Bulldog -1.34955186 -2.469675 -0.22942870 0.0136073

> describeBy(Ch06_Demo$Stress,Ch06_Demo$Breed)

Descriptive statistics by group

group: Beagle

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 46 11.91 2.89 12 11.95 2.97 6 18 12 -0.03 -0.44 0.43

---------------------------------------------------------------------------
132 Part II • Statistical Tests

group: French Bulldog

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 71 11.99 2.5 12 12.04 2.97 6 18 12 -0.18 -0.28 0.3

---------------------------------------------------------------------------

group: Labrador Retriever

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 55 10.64 2.58 11 10.67 2.97 5 16 11 -0.03 -0.58 0.35

Then we look to the descriptive statistics report at the bottom and see that French Bulldog
had a mean of 11.99 and Labrador Retriever had a mean of 10.64. We can now finalize our
summary table, which will provide us with what we’ll need to write an abstract.

Tukey Test Pairwise Comparisons

Group Means p

M(French Bulldog) = 11.99 : M(Beagle) = 11.91 .99

M(Labrador Retriever) = 10.64 : M(Beagle) = 11.91 *.04

M(Labrador Retriever) = 10.64 : M(French Bulldog) = 11.99 *.01

*Statistically significant difference (p ≤ .05).

Test Run: Wilcoxon Multiple Pairwise Comparisons Test

If any of the groups produced a Shapiro–Wilk test p value that was less than or equal to .05
or if the homogeneity of variance test produced a p value that was less than or equal to .05,
then run the nonparametric version of the ANOVA, the Wilcoxon multiple pairwise com-
parisons test:

Test Run: Wilcoxon Multiple Pairwise Comparisons Test

Run the Kruskal–Wallis omnibus test:
[Link](FileName$ContinuousVariable,FileName$CategoricalVariable)

After editing the file name and variable names, the R code and results should look like
this:

> [Link](Ch06_Demo$Stress,Ch06_Demo$Breed)
 Chapter 6 • ANOVA—Tukey Test and Wilcoxon Multiple Pairwise Comparisons Test 133

Kruskal-Wallis rank sum test

data: Ch06_Demo$Stress and Ch06_Demo$Breed

Kruskal-Wallis chi-squared = 8.9435, df = 2, p-value = 0.01143

To assess the results of the Kruskal–Wallis omnibus test, we look to the p value, which
is. 01143, which we’ll round to .01. We’ll use the same protocol for evaluating this p value as
always:

• If p ≤ .05, this indicates that a statistically significant difference has been detected
between at least one pair of the groups.
• If p > .05, this indicates that no statistically significant differences have been
detected between any of the groups.

Since the p value of .01 is less than .05 on the Kruskal–Wallis omnibus test, this indi-
cates that we can expect to see a statistically significant difference between at least one pair
of groups. We can now proceed as before; we’ll run the Wilcoxon multiple pairwise compar-
isons test, which will produce the (nonparametric) p values for each pair of groups, and then
we’ll run descriptive statistics for each group, which will produce the means for each group.

Test Run: Wilcoxon Multiple Pairwise Comparisons Test and Descriptive

Statistics for Each Group
Run the Wilcoxon multiple pairwise comparisons test:
[Link](FileName$ContinuousVariable,FileName$Categorical
Variable,[Link]="bonf")

Run descriptive statistics for each group:

describeBy(FileName$ContinuousVariable,FileName$CategoricalVariable)

After editing the file name and variable names, the R code and results should look like
this:

> [Link](Ch06_Demo$Stress,Ch06_Demo$Breed,[Link]="bonf")

Pairwise comparisons using Wilcoxon rank sum test with continuity correction
134 Part II • Statistical Tests

data: Ch06_Demo$Stress and Ch06_Demo$Breed

Beagle French Bulldog

French Bulldog 1.000 -

Labrador Retriever 0.080 0.013

P value adjustment method: bonferroni

> describeBy(Ch06_Demo$Stress,Ch06_Demo$Breed)

Descriptive statistics by group

group: Beagle

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 46 11.91 2.89 12 11.95 2.97 6 18 12 -0.03 -0.44 0.43

-------------------------------------------------------------
---------------
group: French Bulldog

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 71 11.99 2.5 12 12.04 2.97 6 18 12 -0.18 -0.28 0.3

---------------
-------------------------------------------------------------

group: Labrador Retriever

vars n mean sd median trimmed mad min max range skew kurtosis se

X1 1 55 10.64 2.58 11 10.67 2.97 5 16 11 -0.03 -0.58 0.35

Results: Wilcoxon Multiple Pairwise Comparisons Test

The only thing different in this report is the top part, detailing the results of the Wilcoxon
multiple pairwise comparisons test, which is arranged as a concise matrix containing the non-
parametric version of the p values: for the French Bulldog and Beagle comparison, p = 1.000;
for the Labrador Retriever and Beagle comparison, p = .080; and for the Labrador Retriever and
French Bulldog comparison, p = .013.
The bottom part of this report (after the describeBy command) consists of the
descriptive statistics for each group; notice that the means for each group are all the same
as the prior run. From these figures, we can build this summary report using the same
means as before and the new Wilcoxon multiple pairwise comparisons test (nonparamet-
ric) p values:
 Chapter 6 • ANOVA—Tukey Test and Wilcoxon Multiple Pairwise Comparisons Test 135

Wilcoxon Multiple Pairwise Comparisons

Group Means p

M(French Bulldog) = 11.99 : M(Beagle) = 11.91 1.00

M(Labrador Retriever) = 10.64 : M(Beagle) = 11.91 .08

M(Labrador Retriever) = 10.64 : M(French Bulldog) = 11.99 *.01

*Statistically significant difference (p ≤ .05).

Since the pretest criteria were satisfied, we’ll use the results from the Tukey test to resolve
the hypotheses and document the results, referring to the concise table that we assembled.

Tukey Test Pairwise Comparisons

Group Means p

M(French Bulldog) = 11.99 : M(Beagle) = 11.91 .99

M(Labrador Retriever) = 10.64 : M(Beagle) = 11.91 *.04

M(Labrador Retriever) = 10.64 : M(French Bulldog) = 11.99 *.01

*Statistically significant difference (p ≤ .05).

Hypothesis Resolution
The p values guide us in selecting which hypothesis to accept using the following rules:

• If p ≤ .05 for at least one pair of groups, accept the alternate hypothesis (H1).
• If p > .05 for all of the pairs of groups, accept the null hypothesis (H0).

Per these rules, we can now revisit our hypotheses to make an informed selection:

REJECT H0: There is no difference in the stress levels between the three groups
(Labrador Retriever, Beagle, French Bulldog).
ACCEPT H1: There is a difference in the stress levels between the three groups
(Labrador Retriever, Beagle, French Bulldog).

In preparation for documenting the results, we can refer to the descriptive statistics for each
group to identify the number of participants in each group: n(Beagle) = 46, n(French Bulldog)
= 71, and n(Labrador Retriever) = 55; there was a total of 172 participants in this study
(46 + 71 + 55 = 172).
136 Part II • Statistical Tests

Documenting Results

Abstract
To determine which breed of therapy dog was most effective in reducing stress, we recruited
172 participants who reported that they were experiencing high stress. We randomly assigned
participants to engage in 30 minutes of pet therapy with a Beagle (n = 46), a Labrador
Retriever (n = 55), or a French Bulldog (n = 71). After each session, participants completed
the Acme Stress Index, a five-question self-administered multiple-choice survey that produces
a total score between 5 and 25 (5 = very low stress . . . 25 = very high stress).
We conducted an ANOVA test, revealing that those who had pet therapy with a Labrador
Retriever had a statistically significantly lower average stress level (M = 10.64) compared to
those who had pet therapy with a Beagle (M = 11.91), p = .04, and those who had pet therapy
with a French Bulldog (M = 11.99), p = .01. We detected no statistically significant difference
in the stress levels of those who had pet therapy with a French Bulldog (M = 11.99) compared
to those who had pet therapy with a Beagle (M = 11.91), p = .99.
Based on these findings, we are considering including more Labrador Retrievers in our
program.

GOOD COMMON SENSE

We saw that in the t test, processing two groups involves comparing one pair of groups (Group
1: Group 2), but when we used the ANOVA test to analyze three groups, three pairwise com-
parisons emerged (Group 1: Group 2, Group 1: Group 3, and Group 2: Group 3). As you
might expect, increasing the number of groups increases the number of paired comparisons
that the ANOVA test will produce. For example, in an ANOVA test, a four-group design will
produce 6 unique pairs of results, and a five-group design will produce 10 unique pairs of
results.
When designing a multigroup study, you can easily calculate how many pairwise compar-
isons (combinations) an ANOVA test will produce based on the number of groups specified
in the design. At the bottom of the R Syntax Guide is R code to compute the unique pairs for-
mula: choose(Groups,2). For example, suppose you were considering implementing a five-
group study; you could run this code choose(5,2) and it would tell you how many ANOVA
pairwise comparisons to expect.

> choose(5,2)

[1] 10
 Chapter 6 • ANOVA—Tukey Test and Wilcoxon Multiple Pairwise Comparisons Test 137

The results indicate that a five-group implementation would produce 10 (unique)

ANOVA pairwise comparisons:

G1: G2 – – –

G1: G3 G 2: G 3 – –

G1: G4 G 2: G 4 G3: G4 –

G1: G5 G 2: G 5 G3: G5 G 4: G5

This warrants some consideration prior to implementing a five-group design: You may
want to consider if this study is practical in terms of the sample size; designs that involve more
groups require a larger number of participants. With that in mind, you need to consider other
feasibility factors: Do you expect that you’ll have access to enough participants to recruit for
this study? Do you have sufficient time, staff, and participant compensation fees to run this
five-group study? Finally, it may be a bit unwieldy to comprehend and document the results
involving 10 pairwise comparisons. This is not to say that researchers should inherently avoid
designing and conducting studies involving multiple groups.
Another consideration is statistical power, which pertains to acquiring a sufficient sample
size (n). For example, suppose you had access to 60 people who’ve agreed to participate in your
study. If you are using three groups, you could assign 20 participants to each group, which
could produce statistical results with adequate power, suggesting that you recruited enough
participants to render robust/stable statistics. However, if your design involved 10 groups,
you’d only be able to assign six participants to each group; this low group count would almost
assuredly produce low power, suggesting that the statistics could be regarded as unstable or
unusable due to undersampling. Generally speaking, larger sample sizes (per group) serve
to increase statistical power; hence, opting for designs that involve more groups necessitates
recruiting more participants to sufficiently populate those groups and ultimately produce
well-powered (stable) statistical results.

KEY CONCEPTS

ANOVA
• Pretest checklist
• Normality
• Homogeneity of variance
• t test omnibus test
• Kruskal–Wallis omnibus test
• Tukey test
138 Part II • Statistical Tests

• Wilcoxon multiple comparisons test

• Hypothesis resolution
• Documenting results
• Good common sense
• Combinations formula
• Statistical power

PRACTICE EXERCISES

Download the datasets from the Sage companion website to complete these exercises.

Exercise 6.1
To determine the best proofreading method, the director of a writing lab has created a
five-page document with 15 writing errors. Participants will be recruited and assigned to
read the essay on paper, computer screen, or computer screen with a text-to-speech (voice)
program reading the text aloud. Participants will be asked to detect as many errors as
they can.
Dataset: Ch06_Ex01A.csv
Codebook:
Variable: Errors
Definition: Number of writing errors detected
Type: Continuous (0 . . .15)

Variable: ReadingMethod
Definition: Media that the essay was provided on
Type: Categorical (Paper, Computer, Computer & voice)

a. Write the research question.

b. Write the hypotheses.
c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch06_Ex01B.csv.
 Chapter 6 • ANOVA—Tukey Test and Wilcoxon Multiple Pairwise Comparisons Test 139

Exercise 6.2
A therapist wants to determine the best method for reducing depression. Participants are
recruited and assigned to one of three groups: Talk therapy, Antidepressant medication (Rx),
and Talk therapy with antidepressant medication (Rx). After 10 weeks, all participants will
complete the Acme Depression Instrument (1 = low depression . . . 20 = high depression).
Dataset: Ch06_Ex02A.csv
Codebook:
Variable: Depression
Definition: Score on the Acme Depression Instrument
Type: Continuous (1 = low depression . . . 20 = high depression)

Variable: Therapy
Definition: Type of therapeutic intervention
Type: Categorical (Talk, Rx, Talk & Rx)

a. Write the research question.

b. Write the hypotheses.
c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch06_Ex02B.csv.

Exercise 6.3
A nurse who wants to determine the best treatment for reducing stress recruits a group of par-
ticipants and assigns them to one of three groups: Group 1 will sit quietly, Group 2 will receive
guided meditation, and Group 3 will do yoga with an instructor. After 30 minutes, partici-
pants will complete the Acme Stress Instrument (1 = very low stress . . . 7 = very high stress).
Dataset: Ch06_Ex03A.csv
Codebook:
Variable: Stress
Definition: Score on the Acme Stress Instrument
Type: Continuous (1 = very low stress . . . 7 = very high stress)

Variable: Treatment
Definition: Stress treatment
Type: Categorical (Sit quietly, Guided meditation, Yoga)
140 Part II • Statistical Tests

a. Write the research question.

b. Write the hypotheses.
c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch06_Ex03B.csv.

Exercise 6.4
To determine the best teaching method, an educational researcher gathers the final scores of
students taking the same course three different ways: Traditional classroom, Online with a
live instructor, and Online watching a prerecorded video.
Dataset: Ch06_Ex04A.csv
Codebook:
Variable: Final
Definition: Score on the final
Type: Continuous (0 . . . 100)

Variable: TeachingMethod
Definition: How the course was taught
Type: Categorical (Classroom, Online live, Online video)

a. Write the research question.

b. Write the hypotheses.
c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch06_Ex04B.csv.

Exercise 6.5
To better inform class project planning, an art teacher visits a gallery and unobtrusively uses
a stopwatch to time how long people spend looking at three different artforms: Paintings,
Mobiles, and Photographs.
 Chapter 6 • ANOVA—Tukey Test and Wilcoxon Multiple Pairwise Comparisons Test 141

Dataset: Ch06_Ex05A.csv
Codebook:
Variable: Seconds
Definition: Time a person spent looking at an art piece
Type: Continuous (0 . . . no limit)

Variable: Artform
Definition: Type of art
Type: Categorical (Painting, Mobile, Photograph)

a. Write the research question.

b. Write the hypotheses.
c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch06_Ex05B.csv.

Exercise 6.6
An industrial designer has created a desk that requires consumer assembly. To determine the
optimal form of assembly instructions, the designer recruits participants one at a time and
provides them with the desk parts in the box, the required screwdriver, and assembly instruc-
tions in the form of a booklet with text instructions, a booklet containing text with illustra-
tion instructions, or a website with a narrated video that they can control (e.g., pause, play,
rewind). The designer times how long it takes each person to assemble the desk.
Dataset: Ch06_Ex06A.csv
Codebook:
Variable: Minutes
Definition: Number of minutes required to assemble the desk
Type: Continuous (0 . . . no limit)

Variable: Instructions
Definition: Type of assembly instructions provided
Type: Categorical (Text, Text with illustrations, Website)

a. Write the research question.

b. Write the hypotheses.
142 Part II • Statistical Tests

c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch06_Ex06B.csv.

Exercise 6.7
An academic enrollment director wants to determine if employment status (full-time, part-
time, or unemployed) impacts the number of nonclassroom hours per week students spend
on their classwork (e.g., homework, studying, research, writing). Hours are rounded to near-
est quarter (e.g., 4.25 hours = 4 hours and 15 minutes).
Dataset: Ch06_Ex07A.csv
Codebook:
Variable: Hours
Definition: Number of nonclassroom hours spent on classwork
Type: Continuous (0 . . . no limit)

Variable: Employment
Definition: Employment status
Type: Categorical (Full-time, Part-time, Unemployed)

a. Write the research question.

b. Write the hypotheses.
c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch06_Ex07B.csv.

Exercise 6.8
A farmer wants to determine the effect that soil nutrients have on lemon trees. In Orchard
1, the farmer waters the trees as usual; in Orchard 2, the farmer provides water and soil
Nutrient A; and in Orchard 3, the farmer provides water and Nutrient B. At harvest time,
the farmer will randomly select lemons from each of the three orchards and weigh them.
 Chapter 6 • ANOVA—Tukey Test and Wilcoxon Multiple Pairwise Comparisons Test 143

Dataset: Ch06_Ex08A.csv
Codebook:
Variable: Grams
Definition: Weight of the lemon in grams
Type: Continuous (1 . . . no limit)

Variable: Nutrient
Definition: Soil nutrient
Type: Categorical (None, Nutrient A, Nutrient B)

a. Write the research question.

b. Write the hypotheses.
c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results, and
implications of your findings.
Repeat this exercise using dataset: Ch06_Ex08B.csv.

Exercise 6.9
NOTE: This exercise involves four groups.
The board of directors of an amusement park wants to compare the popularity of four rides:
Haunted house, Merry-go-round, Ferris wheel, and Bumper cars. They gather the daily
number of riders from the turnstile counters at the end of the day for 4 weeks (28 days).
Dataset: Ch06_Ex09A.csv
Codebook:
Variable: Riders
Definition: Number of daily riders
Type: Continuous (0 . . . no limit)

Variable: Attraction
Definition: Amusement park ride
Type: Categorical (Haunted house, Merry-go-round, Ferris wheel, Bumper
cars)

a. Write the research question.

b. Write the hypotheses.
144 Part II • Statistical Tests

c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch06_Ex09B.csv.

Exercise 6.10
NOTE: This exercise involves five groups.
The math department administers a common final to all students taking Statistics 210. The
department chair will compare the scores among the five instructors who taught the course
this term: Professor Anderson, Professor Baker, Professor Campbell, Professor Davis, and
Professor Ellis.
Dataset: Ch06_Ex10A.csv
Codebook:
Variable: Final
Definition: Score on the final
Type: Continuous (0 . . . 100)

Variable: Professor
Definition: Professor who taught the course
Type: Categorical (Anderson, Baker, Campbell, Davis, Ellis)

a. Write the research question.

b. Write the hypotheses.
c. Run each criterion of the pretest checklist (normality, homogeneity of variance) and
discuss your findings.
d. Run the t test and document your findings (ns, means, and p value) and hypothesis
resolution.
e. Write an abstract under 200 words detailing a summary of the study, the t test results,
and implications of your findings.
Repeat this exercise using dataset: Ch06_Ex10B.csv.
 7 PAIRED t TEST AND PAIRED
WILCOXON TEST

To detect change in a continuous variable from one time point (pretest score) to another
(future) time point (posttest score), use the paired t test or paired Wilcoxon test.

Pretest Posttest

Harold Harold
Maude Treatment Maude
Brad Brad
Janet Janet

LEARNING OBJECTIVES

Upon completing this chapter, you will be able to:

7.1 Determine when to use the paired t test
7.2 Build a research question and corresponding hypotheses
7.3 Run and assess the pretest criterion: normality of differences
7.4 Select and run the proper version of the paired t test: the paired t test or the
paired Wilcoxon test
7.5 Interpret the results
7.6 Calculate and document the Δ% formula
7.7 Resolve the hypotheses
7.8 Write a concise abstract detailing the research question and results

145
146 Part II • Statistical Tests

VIDEO RESOURCE

The video for this chapter is Ch07 – Paired t Test and Paired Wilcoxon Test.mp4. This video
provides overviews of these tests, instructions for carrying out the pretest checklist, and
running and interpreting the results of this test using the following dataset: Ch07_Demo.csv.

OVERVIEW—PAIRED t TEST

Whereas the t test processes two groups of continuous variables to determine if there’s a statis-
tically significant difference between the means of the two groups, there may be conditions
when it’s not possible or necessary to carry out a two-group design. You may be interested in
measuring how one group of individuals changes from one time point to the next; the paired
t test can carry out that analysis. This would be a pretest–posttest design, also known as a
simple time-series design, or O X O design, (O = observation wherein data is gathered and
X = treatment), as depicted in Figure 7.1

FIGURE 7.1 ■ Pretest–Posttest Design

Step 1 Step 2 Step 3

Pretest Treatment Posttest

This design involves three steps of pretest/treatment/posttest: (Step 1) The researcher

administers a pretest to gather a continuous variable (e.g., typing speed, mood, pulse rate).
(Step 2) Some form of treatment/intervention is administered, presumably designed to
improve the phenomenon of interest. (Step 3) When the treatment is complete, the researcher
administers the posttest, which is the same metric that was used for the pretest (Step 1).
Continuing with this example, notice that data are only gathered at Step 1 (pretest) and
Step 3 (posttest); technically, whatever happens at Step 2 (treatment) is irrelevant to the
statistical processor since no data are gathered at that point. As such, there doesn’t always
need to be a treatment at Step 2. For example, a store may want to detect if there’s a statisti-
cally significant change in the number of umbrellas sold during winter last year compared
to this year. Step 1 would involve gathering the number of umbrellas sold each day of the
first month of winter for last year. The next step would involve gathering the number of
umbrellas sold each day during the first month of winter for this year. Even though there’s
no treatment between the two data collection points, as you might expect, the formulas
will still work.
 Chapter 7 • Paired t Test and Paired Wilcoxon Test 147

PAIRED t TESTS AND PAIRED WILCOXON TESTS IN CONTEXT

Statistical Reasoning
To detect change in a variable from an earlier time point to a later time point, the paired t test
and the paired Wilcoxon test are the proper choices. For each participant, you’d gather the
same variable using the same instrument/metric twice: (1) the score at the first time point,
which could be a pretest score, and (2) the score using the same instrument/metric at the
second time point, which could be the posttest score. Often, there is some form of treatment
that’s administered in between these two scores. The formula for these statistical tests focuses
primarily on the difference between each person’s first score and their second score.
The formula will produce a mean for the first score (pretest), a mean for the second score
(posttest), and a p value. If the p value is less than or equal to .05, this tells us that there was
a statistically significant change from the first (pretest) score to the second (posttest) score,
suggesting that the treatment that was administered between these scores was effective. If the
p value is greater than .05, this indicates that the change from the first (pretest) score to the
second (posttest) score was statistically insignificant, suggesting that the treatment was not
effective (enough) to change the (posttest) score.

Applied Examples
To determine if sound can be used to help people relax, a researcher may recruit a group of
people and have each person sit in a comfortable chair and record the person’s pulse rate
(pretest), then play a quality recording of a light rain for 10 minutes, and then gather the per-
son’s pulse rate again (posttest). The statistical test would render the mean pulse rate before
listening to the rain (pretest) and the mean pulse rate after listening to the rain (posttest). If
the p value is less than or equal to .05, this would indicate that the mean posttest pulse rate is
significantly different from the mean pretest pulse rate, suggesting that the sound of the rain
influenced the pulse rate. If the p value is greater than .05, this would indicate that even if the
pretest mean and posttest mean are different from each other, the difference is not statistically
significant, suggesting that the rain sound had a negligible impact on the person’s pulse rate.
In some cases, there may be no deliberate treatment between taking the measurement at
the first time point and the second time point. Consider a fourth-grade teacher who wants
to determine if it’s best to teach math in the morning or in the afternoon. At 8:00 AM,
the teacher could hand out a 30-question math quiz and ask the students to solve the odd-
numbered questions. Then at 1:00 PM, the teacher has the students solve the even-numbered
questions. The teacher then records two scores for each student: (1) the number of correct
odd-numbered questions (gathered in the morning) and (2) the number of correct even-
numbered questions (gathered in the afternoon). The paired t test would produce the overall
mean from the 15 questions on the morning test, the overall mean from the 15 questions on
the afternoon test, and the p value. If the p value is less than or equal to .05, this would indi-
cate that there is a statistically significant difference when comparing the morning scores to
148 Part II • Statistical Tests

the afternoon scores. If the p value is greater than .05, this would indicate that there’s no sta-
tistically significant difference in the student’s morning math performance compared to their
afternoon math performance.

When to Use This Statistic

Guidelines for Selecting the Paired t Test

Overview: This statistic detects if a measurement changed from one time point to another.

Variables: This statistic involves two continuous variables for each record: (1) score at the initial
time point and (2) the score at a later time point, using the same metric at both time
points.

Results: An archery instructor observes each student shoot 10 arrows at a target with no coaching
and records the score, after which, the coach provides a 15-minute individualized coaching
session based on observations made during the first round. The student then shoots
another 10 arrows. On average, the archers scored 45.22 on the first round; after the
coaching, the average score was 49.56, which is a 9.6% increase. The paired t test found this
to be a statistically significant change (p < .01), suggesting that this form of coaching was
effective.

LAYERED LEARNING

If you’re skilled with the t test (Chapter 5), you’ll notice that the paired t test is quite similar:
Whereas the t test compares the means of two groups and produces a p value, the paired t test
analyzes one group and compares the means of two time points: before and after a treatment/
intervention and produces a p value.

EXAMPLE

An archery expert wants to evaluate a new form of coaching.

Groups
This is a one-group study involving 36 archers who are interested in improving their target-
shooting skills.

Procedure
The coach silently observes each participant shoot 10 arrows at a standard archery target
consisting of 10 concentric rings (10 points for hitting the center ring . . . 1 point for hitting
the outermost ring, 0 points for missing the target) and then records the total score for that
 Chapter 7 • Paired t Test and Paired Wilcoxon Test 149

first round. Next, the coach provides customized constructive advice based on observations
made during the first round. Then, the participant reclaims the 10 arrows and shoots a second
round. The coach records the precoaching and postcoaching score for each archer.

Research Question
Does providing feedback between rounds have an effect on archery scores?

Hypotheses
H0: Providing feedback between rounds has no effect on archery scores.
H1: Providing feedback between rounds has an effect on archery scores.

Dataset
Use the following dataset: Ch07_Demo.csv.
Codebook:
Variable: Archer
Definition: Name of the archery student
Type: Categorical (unique identification of each participant)

Variable: PreCoaching
Definition: Total score before coaching
Type: Continuous (0 = poor archery skills . . . 100 = excellent archery skills)

Variable: PostCoaching
Definition: Total score after coaching
Type: Continuous (0 = poor archery skills . . . 100 = excellent archery skills)

Notice that this dataset includes the variable Archer, which is the name of each student.
Although we have no intention to statistically process the Archer variable, it’s useful to include
in the dataset as it helps us to keep the PreCoaching and PostCoaching scores paired together
on the same row for each participant, as seen in Table 7.1.

TABLE 7.1 ■ Data Coded for Pretest–Posttest Design

Archer PreCoaching PostCoaching

Jessica 23 30

Chester 10 17
150 Part II • Statistical Tests

Since this statistic involves detecting the change from the first score (PreCoaching) to the
second score (PostCoaching) for each individual record (row), both scores must be entered
for each participant. If both scores are not present on a row, the processor will automatically
exclude that row of data from the analysis.
To process a paired t test or paired Wilcoxon test, refer to the R Syntax Guide, page 7;
we’ll proceed in order starting from the top of that page.
First, click on the Packages tab, then check mvnormtest and psych.

Packages
☑ mvnormtest
☑ psych

Paired t Test Pretest Checklist: ☑ 1. Normality of Differences

The paired t test has only one pretest criterion: normality of differences. This means that if we
were to subtract the pretest from the posttest (or subtract the posttest from the pretest), we would
get a column of numbers that measures the difference between the pretest and the posttest for
each pair of scores. This column of differences won’t appear on the screen as we route these
numbers directly to the Shapiro–Wilk test, which assesses the normality of a distribution.

Pretest Checklist: Normality of Differences

Run the Shapiro–Wilk test to check for normality of differences:
[Link](FileName$Pretest-FileName$Postest)

After editing the file name and variable names, the R code and results should look like this:

> [Link](Ch07_Demo$PreCoaching-Ch07_Demo$PostCoaching)

Shapiro-Wilk normality test

data: Ch07_Demo$PreCoaching - Ch07_Demo$PostCoaching

W = 0.96278, p-value = 0.2614

To interpret the results of the Shapiro–Wilk test, use the following protocol:

• If p ≤ .05, then the distribution is not normal.

• If p > .05, then the distribution is normal.
 Chapter 7 • Paired t Test and Paired Wilcoxon Test 151

The Shapiro–Wilk test produced a p value of .2614; since this is greater than .05, this indi-
cates that the distribution of differences is normal; hence, we can proceed with the paired t test.
Although the Shapiro–Wilk test determined that the differences between scores formed a
normal distribution, if you’d like to see the corresponding histogram, you can use the follow-
ing optional procedure:

Optional Histograms
Draw a histogram showing the difference between each pretest and posttest pair:
hist(FileName$Pretest-FileName$Posttest)

After editing the file name and variable names, the R code and histogram should look like
this:

> hist(Ch07_Demo$PreCoaching-Ch07_Demo$PostCoaching)

Test Run
The paired t test produces the p value to determine if there’s a statistically significant dif-
ference between the pretest mean and the posttest mean, but it does not display those two
means. Fortunately, we can get those means by running descriptive statistics for the pretest and
posttest variables.

Test Run: Descriptive Statistics for Pretest and Posttest Variables

Run descriptive statistics for Pretest and Posttest variables:
describe(FileName)
152 Part II • Statistical Tests

After editing the file name and variable names, the R code and results should look like
this:

> describe(Ch07_Demo)

vars n mean sd median trimmed mad min max range

Archer* 1 36 18.50 10.54 18.5 18.50 13.34 1 36 35

PreCoaching 2 36 45.22 20.32 43.5 45.40 22.24 8 81 73

PostCoaching 3 36 49.56 20.98 47.5 49.13 20.76 11 93 82

Note: Due to the line length of this report, the last three columns (skew, kurtosis, and se)
were eliminated from this exhibit.
These results show that the mean of the archers before they had their coaching session
(PreCoaching) is 45.22, and their mean after their coaching session (PostCoaching) is 49.56.
Set these figures aside; we’ll refer back to them when we document the results.
Next, calculate the ∆% (delta percent); the capital delta (Δ) is the Greek letter statistically
representing change.

Calculate the Δ%
(PosttestMean-PretestMean)/PretestMean*100

After editing the PretestMean and PosttestMean, the R code and results should look like
this (NOTE: You can copy and paste the means from the descriptive statistics results):

> (49.56 - 45.22) / 45.22 * 100

[1] 9.597523

For the Δ%, we’ll round the result 9.597523 to one decimal place: Δ% = 9.6, meaning that
there was a 9.6% increase from the archer’s mean score before coaching (M = 45.22) to their
mean score after coaching (M = 49.56). If the Δ% is negative, this signifies that there was a
decrease from the pretest to the posttest; for example, suppose we had a pretest mean of 99 and a
posttest mean of 86; the Δ% = –13.1, which we would document as such: . . . prior to coaching,
the mean was 99; after the coaching, the mean score dropped to 86, constituting a 13.1% decrease.

SELECTING PAIRED t TEST OR WILCOXON TEST

Is this criterion met?

1. Normality of differences: Shapiro–Wilk test of (Pretest – Posttest) p > .05
 Chapter 7 • Paired t Test and Paired Wilcoxon Test 153

If YES: Run the paired t test.

If NO: Run the Wilcoxon test.

Test Run: Paired t Test

Referring back to the results from the Shapiro–Wilk test, the p value was .2614; since this is
greater than .05, this indicates that the differences between the means of the pretest scores
and the posttest scores are normally distributed. Hence, we’ll run the paired t test.

Test Run: Paired t Test

If the pretest criterion is satisfied (Shapiro–Wilk test p > .05), then run the paired t test:
[Link](FileName$Pretest,FileName$Posttest,paired=T)

After editing the file name and variable names, the R code and results should look like
this:

> [Link](Ch07_Demo$PreCoaching,Ch07_Demo$PostCoaching,paired=T)

Paired t-test

data: Ch07_Demo$PreCoaching and Ch07_Demo$PostCoaching

t = -5.3009, df = 35, p-value = 0.00000645

alternative hypothesis: true mean difference is not equal to 0

95 percent confidence interval:

-5.992882 -2.673785

sample estimates:

mean difference

-4.333333

Since the p value of .00000645 is less than .05, this indicates that there is a statistically
significant difference between the mean of the pretest (PreCoaching) score and the mean of
the posttest (PostCoaching) score.
If you’re seeing an “e” in the p value, such as 6.45e-06, this means that R is using scientific
notation, which is the default mode. To switch to regular notation, run the following com-
mand (which is on page 1 of the R Syntax Guide): options(scipen=999) and press Enter.
Then press the up-arrow key (twice) until you see the paired t test code; press Enter to rerun
the analysis, which will now produce the more readable p value (without the “e”).
Finally, to identify the sample size (n), look to the df (degrees of freedom) result, and add 1:
df = 35; hence, the sample size (n) = 36.
154 Part II • Statistical Tests

Test Run: Paired Wilcoxon Test

If the Shapiro–Wilk test had produced a p value that was less than or equal to .05, then we’d
opt for the paired Wilcoxon test, which is the nonparametric version of the paired t test.

Test Run: Paired Wilcoxon Test

If the pretest criterion is not satisfied (Shapiro–Wilk test p ≤ .05), then run the paired
Wilcoxon test:
[Link](FileName$Pretest,FileName$Posttest,paired=T)

After editing the file name and variable names, the R code and results should look like this:

> [Link](Ch07_Demo$PreCoaching,Ch07_Demo$PostCoaching,paired=T)

Wilcoxon signed rank test with continuity correction

data: Ch07_Demo$PreCoaching and Ch07_Demo$PostCoaching

V = 79.5, p-value = 0.00006877

alternative hypothesis: true location shift is not equal to 0

Since the p value of .00006877 is less than .05, this indicates that there is a statistically
significant difference between the mean of the pretest (PreCoaching) score and the mean of
the posttest (PostCoaching) score.

Results
Since the pretest criterion was satisfied, we’ll proceed using the results from the paired t test,
not the paired Wilcoxon test, even though they both produced statistically significant p values.
The following summarizes the statistics that we’ve gathered:

1. The n (number of archers) = 36.

2. The mean pretest (PreCoaching) score = 45.22.
3. The mean posttest (PostCoaching) score = 49.56.
4. The Δ% = 9.6.
5. The p value = .00000645, which we can document as p = .00000645 or p < .01.

We use the same protocol for interpreting the p value as we’ve used before:

• If p ≤ .05, this indicates that there is a statistically significant difference between the
pretest score and the posttest score.
 Chapter 7 • Paired t Test and Paired Wilcoxon Test 155

• If p > .05, this indicates that there is no statistically significant difference between
pretest score and the posttest score.

In this case, the paired t test produced a p value of .00000645; since this is less than or equal
to .05, this indicates that there is a statistically significant difference between the means of the
PreCoaching score and the PostCoaching score. We can now proceed to resolve the hypotheses:

Hypothesis Resolution
The p value guides us in selecting which hypothesis to accept using the following rules:

• If p ≤ .05, accept the alternate hypothesis (H1).

• If p > .05, accept the null hypothesis (H0).

Per these rules, we can now revisit our hypotheses to make an informed selection:

REJECT H0: Providing feedback between rounds has no effect on archery scores.
ACCEPT H1: Providing feedback between rounds has an effect on archery scores.

Since the paired t test produced a p value of .00000645, which is less than or equal to .05,
we would accept H1, which tells us that providing feedback between rounds has an effect on
archery scores. We could plausibly conclude that these results suggest that this form of coach-
ing is effective in changing the scores of archers.

Documenting Results
Referring to our statistical results, we can now draft an abstract summarizing this study and
the outcome:

Abstract
An archery coach assessed a new form of coaching: The coach worked individually with 36
archers, requesting that each archer shoot 10 arrows at a standard target while the coach unob-
trusively observed the archer’s technique. After recording the total score from the first round
(0–100), the coach provided constructive criticism based on the coach’s observations, detailing
recommendations for improving their precision. Next, the coach instructed each archer to
reclaim their arrows and then to shoot a second round of 10 arrows, after which, the coach
recorded their second score.
Paired t test analysis revealed that in the first round, archers averaged 45.22; after the
coaching, the archers average score statistically significantly increased to 49.56 (p < .01). This
9.6% increase suggests that this form of coaching is effective for archers.
Based on these findings, this coach is considering assessing the effectiveness of this coaching
method for other sports involving precise individual performance (e.g., bowling, golf, weight
training, shotput).
156 Part II • Statistical Tests

GOOD COMMON SENSE

The last paragraph of the above results is plausible; the researcher may have contemplated
other domains that the research findings could be applied to, thus potentially expanding the
beneficence of an effective intervention/treatment to other areas.
The paired t test is quite versatile; it can be used to measure change in a variety of vari-
ables (e.g., attitude, pulse rate, exercise minutes). In the example processed in this chapter, the
coach provided a few minutes of coaching between the two archery rounds, but the amount of
time between the pretest and the posttest can vary considerably. For example, if we were mea-
suring academic performance, the pretest may be a test score gathered during the first week of
class, and then each student could be paired with a tutor; the posttest could be gathered weeks
or months later, at the end of the term.
Be aware that the pretest–posttest design, which is analyzed using the paired t test, is con-
sidered a quasi-experimental design, meaning that it approximates an experimental design, but
it’s not as robust when it comes to threats to internal validity (the influence that factors out-
side of the experiment could have on the outcome). Considering the tutoring example above,
it’s possible that multiple extraneous influences, other than the tutor, may have positively or
negatively affected the student’s performance over the course of the term (e.g., new romantic
interest, the presence of an annoying relative, change in health, discovery of online instruc-
tional videos, working collaboratively with a skillful classmate). Regardless of the outcome,
it’s appropriate to document the possibility of such limitations. This is part of our rationale for
using provisional language (e.g., These results suggest . . .), rather than definitive language (e.g.,
These results prove . . .), even when documenting results that involve a (very) low p value.

KEY CONCEPTS

• Paired t test
• Paired t test designs (synonyms)
• Pretest/treatment/posttest
• Pretest/posttest design
• Simple time-series design
• O-X-O design
• Histogram of differences
• Shapiro–Wilk test of normality
• Δ%
• Paired Wilcoxon test
 Chapter 7 • Paired t Test and Paired Wilcoxon Test 157

• Good common sense

• Internal validity
• Limitations
• Provisional conclusions

PRACTICE EXERCISES

Exercise 7.1
A fourth-grade teacher wants to determine the best time of day to teach math. At 8:00 AM,
the teacher hands out a sheet containing 30 math problems and instructs students to solve
the odd-numbered questions. At 1:00 PM, the teacher instructs students to solve the even-
numbered questions.
Dataset: Ch07_Ex01A.csv
Codebook:
Variable: Name
Definition: Student’s name
Type: Categorical (unique identification of each participant)

Variable: Morning
Definition: Score of the odd-numbered math questions
Type: Continuous (0 . . . 15)

Variable: Afternoon
Definition: Score of the even-numbered math questions
Type: Continuous (0 . . . 15)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality for posttest–pretest) and discuss
your findings.
d. Run the paired t test and document your findings (n, means, Δ%, and p value) and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the paired t
test results, and implications of your findings.
Repeat this exercise using dataset: Ch07_Ex01B.csv.
158 Part II • Statistical Tests

Exercise 7.2
To assess the effectiveness of a townhall meeting, an administrative assistant hands
out survey cards to each attendee. Before the meeting begins, the assistant makes an
announcement requesting that attendees answer the question on a card labeled “SIDE 1,”
which asks:

SIDE 1

Do you intend to vote for Candidate Smith in the upcoming election?

Please circle one number:

1 2 3 4 5

Absolutely no Probably no Undecided Probably yes Absolutely yes

At the conclusion of the meeting, the assistant makes a final announcement, requesting that
attendees flip the card over, which says “SIDE 2” at the top, followed by the same 5-point
scale as SIDE 1, and to drop the card into the sealed collection box as they exit.
Dataset: Ch07_Ex02A.csv
Codebook:

Variable: CardNo
Definition: Serial number printed on each survey card
Type: Categorical (unique identification of each participant)

Variable: Before
Definition: Score of Side 1 of the card
Type: Continuous (1 . . . 5)

Variable: After
Definition: Score of Side 2 of the card
Type: Continuous (1 . . . 5)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality for posttest–pretest) and discuss
your findings.
 Chapter 7 • Paired t Test and Paired Wilcoxon Test 159

d. Run the paired t test and document your findings (n, means, Δ%, and p value) and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the paired t
test results, and implications of your findings.
Repeat this exercise using dataset: Ch07_Ex02B.csv.

Exercise 7.3
To determine if taking a mood test on a computer produces the same score as taking the
same test on paper, a psychologist recruits a group of participants and has each person take a
mood test on paper, consisting of 20 multiple-choice questions (with three choices each), and
then after a brief break, each person is instructed to take the same mood test again, but this
time, on a computer.
Dataset: Ch07_Ex03A.csv
Codebook:

Variable: ID
Definition: Identification code assigned to each participant
Type: Categorical (unique identification of each participant)

Variable: Paper
Definition: Mood test score from paper
Type: Continuous (20 = low depression . . . 60 = high depression)

Variable: Computer
Definition: Mood test score from computer
Type: Continuous (20 = low depression . . . 60 = high depression)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality for posttest–pretest) and discuss
your findings.
d. Run the paired t test and document your findings (n, means, Δ%, and p value) and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the paired t
test results, and implications of your findings.
Repeat this exercise using dataset: Ch07_Ex03B.csv.
160 Part II • Statistical Tests

Exercise 7.4
To determine if reading an article a second time enhances comprehension, an English
teacher asks volunteers to read a three-page article and then answer 10 multiple-choice com-
prehension questions. Next, participants are instructed to reread the same article and answer
the same 10 questions a second time.
Dataset: Ch07_Ex04A.csv
Codebook:
Variable: ID
Definition: Identification code assigned to each participant
Type: Categorical (unique identification of each participant)

Variable: First
Definition: Score after reading the article the first time
Type: Continuous (0 = low comprehension . . . 10 = high comprehension)

Variable: Second
Definition: Score after reading the article the second time
Type: Continuous (0 = low comprehension . . . 10 = high comprehension)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality for posttest–pretest) and discuss
your findings.
d. Run the paired t test and document your findings (n, means, Δ%, and p value) and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the paired t
test results, and implications of your findings.
Repeat this exercise using dataset: Ch07_Ex04B.csv.

Exercise 7.5
The staff at a hospital are required to pass an annual CPR (re)certification course. The educa-
tion director wants to determine the participant’s CPR skill level before and after the train-
ing. Scores are based on the instructor’s observations of each participant demonstrating their
skills on a CPR dummy equipped with simulated pulse and sensors.
 Chapter 7 • Paired t Test and Paired Wilcoxon Test 161

Dataset: Ch07_Ex05A.csv
Codebook:
Variable: EmployeeNo
Definition: Employee number
Type: Categorical (unique identification of each participant)

Variable: Before
Definition: CPR skill level before the training
Type: Continuous (1 = low skills . . . 10 = high skills)

Variable: After
Definition: CPR skill level after the training
Type: Continuous (1 = low skills . . . 10 = high skills)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality for posttest–pretest) and discuss
your findings.
d. Run the paired t test and document your findings (n, means, Δ%, and p value) and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the paired t
test results, and implications of your findings.
Repeat this exercise using dataset: Ch07_Ex05B.csv.

Exercise 7.6
To assess the effectiveness of a video that encourages walking as exercise, a physical therapist
issues a pedometer to each participant and asks them to wear it for a week, walking as they
normally would. After 1 week, the physical therapist records the total number of steps taken
that week and resets the pedometer to zero steps. Next, the researcher shows each participant
a 15-minute video detailing the health benefits of walking and techniques for making the
process more fun (e.g., music, walking with a friend). After the second week, the physical
therapist reads each participant’s pedometer and records the total number of steps taken in
that week.
162 Part II • Statistical Tests

Dataset: Ch07_Ex06A.csv
Codebook:
Variable: PedSerial
Definition: Serial number of each pedometer
Type: Categorical (unique identification of each participant)

Variable: Before
Definition: Pedometer weekly step total before watching the video
Type: Continuous (0 . . . no limit)

Variable: After
Definition: Pedometer weekly step total after watching the video
Type: Continuous (0 . . . no limit)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality for posttest–pretest) and discuss
your findings.
d. Run the paired t test and document your findings (n, means, Δ%, and p value) and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the paired t
test results, and implications of your findings.
Repeat this exercise using dataset: Ch07_Ex06B.csv.

Exercise 7.7
To identify and remove underutilized post boxes, last September, a researcher counted the
number of items in a post box every day of the month. This September, the researcher repeats
that process. The analysis will reveal if there’s a significant change in its usage since last year.
Dataset: Ch07_Ex07A.csv
Codebook:
Variable: Day
Definition: Day number of September (1 = Sept. 1, 2 = Sept. 2, etc.)
Type: Continuous

Variable: SeptemberLastYear
Definition: Daily number of items in the post box in September from last year
Type: Continuous (1 . . . no limit)
 Chapter 7 • Paired t Test and Paired Wilcoxon Test 163

Variable: SeptemberThisYear
Definition: Daily number of items in the post box in September from this year
Type: Continuous (1 . . . no limit)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality for posttest–pretest) and discuss
your findings.
d. Run the paired t test and document your findings (n, means, Δ%, and p value) and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the paired t
test results, and implications of your findings.
Repeat this exercise using dataset: Ch07_Ex07B.csv.

Exercise 7.8
An instructor of a live online class records each session. At the end of the session, just prior to
logging-off, students are instructed to take an online quiz worth 20 points. The next day, stu-
dents are assigned to watch the videorecording of the session taught yesterday and at the end
of the video, they are instructed to take the online quiz again. The instructor will compare
the scores from both quizzes to determine if watching the video the next day affects learning.
Dataset: Ch07_Ex08A.csv
Codebook:
Variable: Student
Definition: Class and participant number
Type: Categorical (unique identification of each participant)

Variable: First
Definition: Quiz score after watching the video the first time
Type: Continuous (0 . . . 20)

Variable: Second
Definition: Quiz score after watching the video the second time
Type: Continuous (0 . . . 20)
164 Part II • Statistical Tests

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality for posttest–pretest) and discuss
your findings.
d. Run the paired t test and document your findings (n, means, Δ%, and p value) and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the paired t
test results, and implications of your findings.
Repeat this exercise using dataset: Ch07_Ex08B.csv.

Exercise 7.9
A human resources manager wants to determine if there’s a change in attentiveness in
employees who work 12-hour shifts. At the start of the shift, each employee is asked to com-
plete a peg-in-hole type puzzle as promptly as they can; the puzzle consists of 30 unique
geometric pieces (e.g., circle, square, triangle, rectangle, oval, star). The manager will use a
stopwatch to time how long it takes to complete the puzzle. The manager will administer the
same puzzle again at the end of each employee’s shift.
Dataset: Ch07_Ex09A.csv
Codebook:
Variable: EmployeeID
Definition: Employee number
Type: Categorical (unique identification of each participant)

Variable: Start
Definition: Seconds required to complete the puzzle at the start of the shift
Type: Continuous (1 . . . no limit)

Variable: End
Definition: Seconds required to complete the puzzle at the end of the shift
Type: Continuous (1 . . . no limit)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality for posttest–pretest) and discuss
your findings.
 Chapter 7 • Paired t Test and Paired Wilcoxon Test 165

d. Run the paired t test and document your findings (n, means, Δ%, and p value) and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the paired t
test results, and implications of your findings.
Repeat this exercise using dataset: Ch07_Ex09B.csv.
Exercise 7.10
To determine the effect that sound has on people, a researcher recruits a group of volunteers
and guides them to a comfortable chair. After resting for 10 minutes, the researcher records
their pulse rate using a pulse oximeter. Next, each person is given a pair of headphones and
told that they can close their eyes if they wish while they listen to the sound of light rain for
10 minutes, after which, the researcher gathers the pulse rate again.
Dataset: Ch07_Ex10A.csv
Codebook:
Variable: Participant
Definition: Last name and initial of first name
Type: Categorical (unique identification of each participant)

Variable: Before
Definition: Pulse rate before listening to rain
Type: Continuous (1 . . . no limit)

Variable: After
Definition: Pulse rate after listening to rain for 10 minutes
Type: Continuous (1 . . . no limit)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality for posttest–pretest) and discuss
your findings.
d. Run the paired t test and document your findings (n, means, Δ%, and p value) and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the paired t
test results, and implications of your findings.
Repeat this exercise using dataset: Ch07_Ex10B.csv.
 CORRELATION—PEARSON
8 TEST AND SPEARMAN TEST

To assess the strength and direction of the relationship between two (different) continuous
variables, run the Pearson test or the Spearman test.

LEARNING OBJECTIVES

Upon completing this chapter, you will be able to:

8.1 Determine when to use correlation
8.2 Comprehend the relationship between two continuous variables: negative/
positive and strength
8.3 Build a research question and corresponding hypotheses
8.4 Run and assess the pretest criteria: normality, linearity, and homoscedasticity
8.5 Select and run the proper version of correlation: the Pearson test or the
Spearman test
8.6 Interpret the direction and strength of a correlation

167
168 Part II • Statistical Tests

8.7 Order and comprehend a scatterplot with regression line

8.8 Interpret the results
8.9 Resolve the hypotheses
8.10 Write a concise abstract detailing the research question and results
8.11 Understand the three criteria for causation: correlation, temporality, and
nonspurious
8.12 Differentiate between correlation and causation

VIDEO RESOURCE

The video for this chapter is Ch08 – Correlation – Pearson Test and Spearman Test.mp4.
This video provides overviews of these tests, instructions for carrying out the pretest check-
list, and running and interpreting the results using the following dataset: Ch08_Demo.csv.

OVERVIEW—PEARSON TEST

The Pearson test is the parametric test for correlation. The word correlation consists of the
prefix co, meaning with or together (e.g., coworker is the other person that you work with), and
the suffix relation; hence, it follows that a correlation computes the nature of the relationship
that one continuous variable has with another continuous variable. This is also referred to
as bivariate correlation, which is symbolized using the lowercase r. Be sure to document the
correlation using the lowercase r; the uppercase R symbolizes multiple regression, which is a
very different statistic. The r ranges from –1 to +1, indicating the direction and strength of the
correlation.

CORRELATION IN CONTEXT

Statistical Reasoning
The correlation formula involves gathering two continuous variables from each participant:
X (e.g., home happiness) and Y (e.g., job satisfaction). The correlation formula first com-
putes the numerator: It totals the difference between each X number and the mean of X
and then multiplies that by the total difference between each Y number and the mean of Y.
The denominator is similar except the differences are squared as they are being totaled; after
 Chapter 8 • Correlation—Pearson Test and Spearman Test 169

they are fully totaled, the square root is applied, and these final numbers (for X and Y) are
multiplied together. This correlation calculation produces a number between –1 and 1 that
provides an overall summary of the trend (relationship) between the two variables (X and Y),
indicating the direction and strength of the correlation.

Applied Examples
A political scientist wants to determine if one’s preference to vote by mail is correlated with
age. This researcher administers a two-question survey asking each participant to provide
their age and their preference for vote-by-mail (1 = strongly dislike . . . 5 = strongly like). The
results would reveal if younger people prefer to vote by mail, if older people prefer to vote by
mail, or if age has nothing to do with vote-by-mail preference.
A sociologist may want to discover if there’s a correlation between the amount of time
a person accesses social media per day and self-esteem. This researcher could implement
a survey to collect two variables from each person: X = number of minutes spent on social
media in a day and Y = level of self-esteem (1 = very low self-esteem . . . 80 = very high self-
esteem). A positive correlation between these two variables would suggest that people who
spend more time on social media tend to have higher self-esteem, or it may produce a nega-
tive correlation, suggesting that people who spend more time on social media tend to have
lower self-esteem, or the researcher may find that there’s no correlation between these two
variables—the amount of time spent on social media has nothing to do with a person’s sense
of self-esteem.

When to Use This Statistic

Guidelines for Selecting Correlation

Overview: This statistic indicates if there is a correlation between two continuous variables.

Variables: This statistic requires two continuous variables for each record.

Results: To determine if there is a correlation between the quality of home life and job
satisfaction, we administered two surveys to each participant; the first survey
assessed home happiness (0 = low . . .30 = high); the second survey focused on
job satisfaction (0 = low . . .100 = high). Pearson analysis revealed a statistically
significant moderate correlation: r = .52 (p < .01); this positive correlation
suggests that those who were happy at home tended to be happy at work and vice
versa.

MORE ABOUT CORRELATION

Correlation direction pertains to the sign of the number (negative or positive). Positive
correlations (r = 0 . . . +1) occur when the two variables move in the same direction. For
example, consider the correlation between hours worked and money earned: Working only a
170 Part II • Statistical Tests

few hours is correlated with a small paycheck, whereas working more hours is correlated with
a larger paycheck. Since these variables move in the same direction (low hours correlates to
a low paycheck, and high hours correlates to a high paycheck), this is a positive correlation,
which would produce an r that’s between 0 and +1.
Negative correlations occur when the two variables move in different directions. For
example, consider the relationship between the amount of time spent practicing a musical
instrument and the number of performance mistakes: We would expect that when prac-
tice time is low, performance mistakes would be high, and when practice time is high, per-
formance mistakes would be low. Since these two variables move in opposite directions,
this is a negative correlation (r = –1 . . . 0). Table 8.1 summarizes positive and negative
correlations.

TABLE 8.1 ■ Correlation Direction (Positive or Negative)

Correlation r Variable Directions

Positive 0...1 X↑ Y↑ or X↓ Y↓

Negative –1 . . . 0 X↑ Y↓ or X↓ Y↑

Correlation strength, which ranges from –1 to +1, indicates the extent to which one vari-
able is associated with the other. For example, we might reasonably expect to find a strong
positive correlation between the amount of time spent doing homework and the course grade
(e.g., r = +.80), whereas we would expect to find a strong negative correlation between alcohol
consumption and the course grade (e.g., –.80). Considering the strength of these correlations,
we might consider variables such as study hours or alcohol consumption as reasonable predictors
of academic performance.
Not all correlations are strong; if we were to consider the correlation between shoe size
and grade, since these two variables seem unrelated, it’s implausible to assert that shoe size is
a viable predictor of a student’s grade. As such, we would expect to find a relatively weak cor-
relation between shoe size and grade (e.g., r = +.02 or –.02). Figure 8.1 shows that correlations
nearer to –1 or +1 are stronger than correlations nearer to 0. Table 8.2 provides guidelines for
interpreting the strength of the correlation (r).

FIGURE 8.1 ■ Correlation Strength (–1 . . . +1)

Strong Weak Strong

−1 0 +1
 Chapter 8 • Correlation—Pearson Test and Spearman Test 171

TABLE 8.2 ■ Interpreting Correlation Strength

r Strength

.80 to 1.00 or –1.00 to –.80 Very strong

.60 to .79 or –.79 to –.60 Strong

.40 to .59 or –.59 to –.40 Moderate

.20 to .39 or –.39 to –.20 Weak

.00 to .19 or –.19 to .00 Very weak

A scatterplot is used to visualize correlations, enabling us to see the direction and strength
of the correlation between two variables. Each point on a scatterplot represents the two vari-
ables gathered from each person. For example, suppose a person had spent 49 minutes taking
an exam and scored 87; this would result in a point being plotted at coordinates (49,87) (49 on
the horizontal X-axis, representing the number of minutes it took the student to complete the
exam, and 87 on the vertical Y-axis, representing the grade on the exam). In Figure 8.2, notice
that among the cloud of points in each scatterplot, there’s a straight line running through the
points; this is the regression line, which you can think of as the calculated average pathway
through the points. The regression line is a sort of summary of the individual points, suggest-
ing the trend of the data.
To assess the direction of a correlation, read the scatterplot from left to right, the same way
that you would read an English sentence. Moving from left to right, if the regression line is
increasing (moving uphill), this signifies a positive correlation, wherein both variables start low,
and both increase as you advance to the right; this concurs with the positive r. Alternatively, if
starting from the left and moving to the right, the regression line is decreasing (moving down-
hill); this signifies a negative correlation. As the values on the horizontal X-axis increase, the
values on the vertical Y-axis decrease, which concurs with the negative correlation (r).
To visually assess the strength of a correlation, notice the positions of the dots in relation
to the regression line. In Figure 8.2, observe that in the first two scatterplots, the dots are pre-
dominately clustered around the regression line, producing fairly strong correlations (r = .75
and r = –.75), whereas the dots in the third scatterplot depart substantially from the regression
line, which indicates a very low correlation (r = .02).
Before departing from the three scatterplots in Figure 8.2, we’ll explore each one individ-
ually. These scatterplots represent two variables: The number of minutes that a student spent
taking an exam is on the horizontal X-axis, and the exam score is on the Y-axis. Scatterplots
are more intuitively clear, with the independent variable on the X-axis and the dependent
(outcome) variable on the Y-axis.
The first scatterplot shows a strong positive correlation (r = .75), indicating that the two
variables moved in the same direction: Students who spent a low amount of minutes taking
172 Part II • Statistical Tests

FIGURE 8.2 ■ Three Scatterplots With Regression Lines: Positive, Negative, and
No Correlation

r = .75 r = -.75 r = .02

Positive Correlation Negative Correlation No Correlation

the exam had a low score, and students who spent a high amount of minutes taking the exam
had a high score. We could propose a speculative interpretation of these findings: This positive
correlation indicates that the longer a student spent taking the exam, the higher their score was,
suggesting that it takes more time to write more thorough, higher-quality answers, or perhaps more
knowledgeable students wrote more detailed responses, which takes longer.
The second scatterplot shows a strong negative correlation (r = –.75), indicating that the
two variables moved in opposite directions: Students who spent a low amount of minutes tak-
ing the exam had a high score, and students who spent a high amount of minutes taking the
exam had a low score. We could propose a speculative interpretation of these findings: This
negative correlation indicates that the less time a student spent taking the exam, the higher their
score was, suggesting that better-prepared students may have confidently responded to each ques-
tion and then promptly moved on to the next.
The third scatterplot shows a very low correlation (r = .02), indicating that the two vari-
ables moved independently of each other: Students who spent a low amount of minutes taking
the exam had a variety of scores ranging from low to high; the same can be said of all other stu-
dents. We could propose a speculative interpretation of these findings: This analysis indicates
that there is no correlation between the amount of time a student spends taking the exam and their
score. Apparently, the amount of time it takes for a student to complete the exam does not predict
their score.
The statistics that we have covered so far have involved a consistency in the continuous
variables. For example, the data that you would gather for a t test or an ANOVA test must be
consistent across the groups: If you’re measuring the participant’s pulse rate in one group, you
need to use that same metric in the other groups. Similarly, if you’re using a pretest–posttest
design, you would need to use the same metric to gather the pretest and posttest scores for a
paired t test. Essentially, these statistics compare apples to apples, whereas correlation enables
you to compare apples to oranges.
So far, in this chapter, we’ve considered some possible correlations between two very dif-
ferent variables—time, which is a measurement of minutes, and score, which is a measurement
 Chapter 8 • Correlation—Pearson Test and Spearman Test 173

of academic performance. In the demonstration analysis in this chapter, we’ll conduct a cor-
relational analysis between two different variables using two different metrics that use dif-
ferent scales. Specifically, we’ll process a correlation that analyzes the relationship between
each participant’s level of happiness at home (0 = very unhappy at home . . . 30 = very happy at
home) and their level of job satisfaction (0 = very unsatisfied at work . . . 100 = very satisfied at
work). Clearly, the scales are different; one ranges from 0 to 30, and the other ranges from 0
to 100. Also, notice that these two variables focus on two different domains—one involves
life at home, and the other is about the quality of the workplace. The exercises in this chapter
involve other correlational analyses to assess the relationship between different variables with
different scales (e.g., the correlation between self-esteem [1 = very low self-esteem . . . 80 =
very high self-esteem] and the number of minutes spent on social media per day [0 . . . 1,440],
the correlation between age [18 . . . 100] and vote by mail preference [1 = strongly dislike . . . 5
= strongly like]). These kinds of apples and oranges comparisons are fairly common among the
variables in correlational analyses.

EXAMPLE

A researcher wants to determine if there is a correlation between the level of happiness that a
person experiences at home and their job satisfaction.

Groups
This is a one-group design wherein we will gather two variables (home happiness and workplace
satisfaction) from each participant.

Procedure
A researcher recruits voluntary participants and asks them to complete two surveys, each
consisting of a series of scaled questions: The Acme Home Happiness Instrument (0 = very
unhappy at home . . . 30 = very happy at home) and the Acme Workplace Assessment Survey
(0 = very unsatisfied at work . . . 100 = very satisfied at work).

Research Question
Is there a correlation between the level of happiness at home and the level of workplace
satisfaction?

Hypotheses

H0: There is no correlation between happiness at home and workplace satisfaction.

H1: There is a correlation between happiness at home and workplace satisfaction.
174 Part II • Statistical Tests

Dataset
Use the following dataset: Ch08_Demo.csv.
Codebook:
Variable: Name
Definition: Name of the participant
Type: Categorical (unique identification of each participant)

Variable: HomeHappiness
Definition: Happiness at home measured by the Acme Home Happiness
Instrument
Type: Continuous (0 = very unhappy at home . . . 30 = very happy at home)

Variable: JobSatisfaction
Definition: Level of workplace satisfaction measure by the Acme Workplace
Assessment Survey
Type: Continuous (0 = very unsatisfied at work . . . 100 = very satisfied at
work)

Notice that the first variable is Name; since it is not a number, this is considered a categor-
ical variable, which we do not intend to statistically process, but including the name or some
other identification code in the dataset for each participant helps to keep the data pertaining
to that individual organized. Just as in the paired t test, in correlation, it’s important to keep
each participant’s two scores together and in order on the same line.
Another similarity that the paired t test has in common with correlation is that there must
be an entry for each of the two variables involved in the correlation (e.g., HomeHappiness and
JobSatisfaction) for each participant; if both entries are not present on a row, the processor will
automatically exclude that row of data from the analysis.
To process a Pearson (correlation) test or Spearman test, refer to the R Syntax Guide,
page 9; we’ll proceed in order starting from the top of that page.
First, click on the Packages tab, then check car, mvnormtest, and psych.

Packages
☑ car
☑ mvnormtest
☑ psych

Pearson Pretest Checklist: ☑ 1. Normality, ☑ 2. Linearity,

and ☑ 3. Homoscedasticity
For the Pearson Pretest Checklist, we will examine three criteria: normality, linearity (pro-
nounced: lin-ee-air-it-tee), and homoscedasticity (pronounced: hoe-moe-skuh-daz-tiss-it-tee).
 Chapter 8 • Correlation—Pearson Test and Spearman Test 175

Pretest Checklist: 1. Normality

Begin by checking the two continuous variables for normality (normal distributions):

Pretest Checklist: 1. Normality

Check for normal distribution:
[Link](FileName$ContinuousVariableX)

[Link](FileName$ContinuousVariableY)

After editing the file name and variable names, the R code and results should look like this:

> [Link](Ch08_Demo$HomeHappiness)

Shapiro-Wilk normality test

data: Ch08_Demo$HomeHappiness

W = 0.98417, p-value = 0.3767

> [Link](Ch08_Demo$JobSatisfaction)

Shapiro-Wilk normality test

data: Ch08_Demo$JobSatisfaction

W = 0.9845, p-value = 0.3946

To interpret the results of the Shapiro–Wilk test, use the following protocol:

• If p ≤ .05, then the distribution is not normal.

• If p > .05, then the distribution is normal.

The Shapiro–Wilk test produced a p value of .3767 for HomeHappiness and .3946 for
JobSatisfaction; since both of the p values are greater than .05, this indicates that both of these
176 Part II • Statistical Tests

variables are normally distributed; hence, this criterion is satisfied. If you’d like to see the cor-
responding histograms, you can use the following optional procedure:

Optional Histograms
Draw histograms for each variable:
hist(FileName$ContinuousVariableX)

hist(FileName$ContinuousVariableY)

After editing the file name and variable names, the R code and the (normally distributed)
histograms for HomeHappiness and JobSatisfaction should look like this:

> hist(Ch08_Demo$HomeHappiness)

> hist(Ch08_Demo$JobSatisfaction)

The remaining two pretest criteria, linearity and homoscedasticity, involve visually
inspecting the scatterplot with the regression line. If you can identify one variable as the
dependent variable and the other as the independent variable, it’s common practice to put
the dependent variable on the Y (vertical) axis and the independent variable on the X (hori-
zontal) axis.
 Chapter 8 • Correlation—Pearson Test and Spearman Test 177

Pretest Checklist: 2 and 3. Check for Linearity and Homoscedasticity

Draw a scatterplot with regression line:
scatterplot(FileName$ContinuousVariableY~FileName$ContinuousVariableX,
smooth=F,box=F)

After editing the file name and variable names, the R code and scatterplot should look like
this:

> scatterplot(Ch08_Demo$JobSatisfaction~Ch08_Demo$HomeHappiness,smooth=F,box=F)

Pretest Checklist: 2. Linearity

The term linearity pertains to a straight line. When it comes to linearity, observe the points on
the scatterplot in relation to the (straight) regression line. In Figure 8.3, the first scatterplot
shows that the points are arranged in a fairly straight line, concurrent with the regression line,
per the shaded rectangle; hence, we would say the criterion of linearity is satisfied (R does not
produce the superimposed shaded rectangle graphic; this overlay is for exemplary purposes).
Conversely, the points in the second scatterplot show a distinct curve—a substantial depar-
ture from the straight regression line; this would violate the criterion of linearity.
Figure 8.4 is the scatterplot for our analysis involving the variables HomeHappiness and
JobSatisfaction. Although the points span beyond the bounds of the shaded rectangle, they
are arranged in a fairly linear pattern; there are no unexpected bends or curves in the cloud of
points, and hence we would consider the criterion of linearity to be satisfied.
178 Part II • Statistical Tests

FIGURE 8.3 ■ Linear and Nonlinear Scatterplots

FIGURE 8.4 ■ Scatterplot of HomeHappiness and JobSatisfaction Shows Linearity

Pretest Checklist: 3. Homoscedasticity

In a homoscedastic scatterplot, we see the majority of the points arranged in an elliptical
shape, like a football. In Figure 8.5, the first scatterplot shows that the superimposed ellipse
(which R does not draw) covers most of the points in the scatterplot. Notice that this homosce-
dastic scatterplot is characterized by a high density of points toward the middle and the den-
sity tapers off in both directions, with very few points at both ends (of the football shape). The
second diagram in Figure 8.5 depicts a nonhomoscedastic distribution, as the highest density
of points is in the upper right of the scatterplot, not the middle.
Figure 8.6 is the scatterplot for our analysis involving the variables HomeHappiness
and JobSatisfaction. Notice that the superimposed ellipse covers most of the points on the
 Chapter 8 • Correlation—Pearson Test and Spearman Test 179

FIGURE 8.5 ■ Homoscedastic and Nonhomoscedastic Scatterplots

Homoscedastic Not Homoscedastic

FIGURE 8.6 ■ Scatterplot of HomeHappiness and JobSatisfaction Shows

Homoscedasticity

scatterplot; we see that most of the points are clustered toward the middle, and the density of
the points drops as we progress to either end of the regression line.
The criterion of homoscedasticity is not arbitrary; refer back to the histograms for the
two variables that comprise this scatterplot: HomeHappiness and JobSatisfaction. Notice that
both of these variables are normally distributed; the histograms reflect that most of the values
are in the middle, and there are progressively fewer lower and higher values, as shown in the
bell-shaped normal curves. Considering that both of these histograms follow this symmetri-
cal distribution, wherein most of the values are in the middle and progressively fewer at the
tails, it follows that if we were to combine these two (normally distributed) variables into a
180 Part II • Statistical Tests

single graph (scatterplot), we would expect to see a homoscedastic distribution, where most of
the points are in the middle region and fewer points at the ends.

Test Run
Considering that most of the points in this scatterplot fit within the elliptical area, this scat-
terplot satisfies the homoscedasticity criterion.

SELECT THE PEARSON TEST OR SPEARMAN TEST

Are all of these criteria met?

1. Normality: Shapiro–Wilk tests p > .05 for both variables.
2. Scatterplot is linear.
3. Scatterplot is homoscedastic.
If YES: Run the Pearson test.
If NO: Run the Spearman test.

Let’s summarize the results of three pretest criteria: (1) the Shapiro–Wilk test for both
variables produced p values that were greater than .05; hence, the variables are normally dis-
tributed. Additionally, the scatterplot revealed that the data are (2) linear and (3) homoscedas-
tic; since all of the pretest criteria are satisfied, we’ll run the Pearson test.

Test Run: Pearson Test

Test Run: Pearson Test

Run the Pearson test (r):
[Link](FileName$ContinuousVariableX,FileName$ContinuousVariableY)

After editing the file name and variable names, the R code and results should look like this:

> [Link](Ch08_Demo$HomeHappiness,Ch08_Demo$JobSatisfaction)

Pearson's product-moment correlation

data: Ch08_Demo$HomeHappiness and Ch08_Demo$JobSatisfaction

 Chapter 8 • Correlation—Pearson Test and Spearman Test 181

t = 5.5562, df = 84, p-value = 0.0000003176

alternative hypothesis: true correlation is not equal to 0

95 percent confidence interval:

0.344356 0.658009

sample estimates:

cor

0.518408

If there’s an “e” in the p value (e.g., 3.176e-07), this means that R is using scientific nota-
tion, which is the default mode. To switch to the more readable notation, run the follow-
ing command (which is on page 1 of the R Syntax Guide): options(scipen=999) and press
Enter. Then press the up-arrow key (twice) until you see the correlation test code; press Enter to
rerun the analysis, which will now produce the more readable p value (without the “e”).
The cor result is the correlation (r), which is 0.518408; round this to two decimal digits: r =
.52. We also see that the p value is .0000003176; since this is less than or equal to .05, this tells
us that this is a statistically significant correlation.
Finally, to identify the sample size (n), look to the df (degrees of freedom) result, and add 2:
df = 84, and hence n = 86. When it comes to the n, remember that only records that have both
entries (the X variable and the Y variable) will be included in the analysis.

Test Run: Spearman Test

If the pretest criteria (normality [of both variables], linearity, and homoscedasticity) are not all
satisfied, then run the nonparametric version of the Pearson test: the Spearman test.

Test Run: Spearman Test

Run the Spearman test (r):
[Link](FileName$ContinuousVariableY,FileName$ContinuousVariableX,
method="spearman")

After editing the file name and variable names, the R code and results should look like this:

> [Link](Ch08_Demo$HomeHappiness,Ch08_Demo$JobSatisfaction,method="spearman")

Spearman's rank correlation rho

data: Ch08_Demo$HomeHappiness and Ch08_Demo$JobSatisfaction

182 Part II • Statistical Tests

S = 62972, p-value = 0.0001056

alternative hypothesis: true rho is not equal to 0

sample estimates:

rho

0.4058938

Whereas the Pearson test produces the r (correlation) value, the Spearman test uses the
Greek letter rho (pronounced: row), symbolized as ρ. Per the results report above, we see a
Spearman’s rho of 0.4058938, which we round to rho = .41. Since the p value of .0001056 is
less than or equal to .05, this indicates that the rho of .41 is statistically significant.

Results
Since the pretest criteria were satisfied, we’ll proceed using the results from the Pearson test,
not the Spearman test, even though they both produced statistically significant p values.
We’ll use the same protocol for interpreting the p value as we’ve used before:

• If p ≤ .05, this indicates that there is a statistically significant correlation between the
two variables.
• If p > .05, this indicates that there is no statistically significant correlation between
the two variables.

In this case, the Pearson test produced a p value of .0000003176; since this is less than
or equal to .05, this indicates that there is a statistically significant correlation between
HomeHappiness and JobSatisfaction. We can now proceed to resolve the hypotheses.

Hypothesis Resolution
The p value guides us in selecting which hypothesis to accept using the following rules:

• If p ≤ .05, accept the alternate hypothesis (H1).

• If p > .05, accept the null hypothesis (H0).

Per these rules, we can now revisit our hypotheses to make an informed selection:

REJECT H0: There is no correlation between happiness at home and workplace

satisfaction.
ACCEPT H1: There is a correlation between happiness at home and workplace
satisfaction.
 Chapter 8 • Correlation—Pearson Test and Spearman Test 183

Since the Pearson test produced a p value of .0000003176, which is less than or equal to .05,
we would accept H1, which tells us that there is a correlation between happiness at home and work-
place satisfaction. This positive correlation suggests a general consistency in the person’s condi-
tion in both domains: Those who are happy at home tend to be satisfied at work, and those who
are unhappy at home tend to be unsatisfied at work.

Documenting Results
We can now draft an abstract summarizing this study and the outcome, wherein n = 86, r =
.52, and p = .0000003176 (which we can abbreviate as: p < .01):

Abstract
To determine if there’s a correlation between an individual’s happiness at home and their sat-
isfaction at work, we recruited 86 employed participants and asked them to complete two self-
administered surveys, each consisting of a series of scaled questions: The Acme Home Happiness
Instrument (0 = very unhappy at home . . . 30 = very happy at home) and the Acme Workplace
Assessment Survey (0 = very unsatisfied at work . . . 100 = very satisfied at work).
Pearson analysis revealed r = .52 (p < .01), indicating that there is a moderate positive
correlation between how happy a person is at home and how satisfied they are at work, sug-
gesting that there is a relative consistency in how people experience these two settings: Those
who reported being happy at home tended to be satisfied at work, and conversely, people who
are unhappy at home tended to be unsatisfied at work.

CORRELATION VERSUS CAUSATION

Although it can be tempting to take what’s learned from a correlational analysis and believe
that we’ve identified a causal relationship (change in one variable causes the change in the
other variable), we need to differentiate between correlation and causation.
To claim that we’ve identified a correlation, we only need to examine the statistical results
of a Pearson or Spearman test, which indicates the direction, strength, and significance of the
correlation, but to make the leap from correlation to causation, three conditions must be satis-
fied: (1) correlation, (2) temporality, and (3) nonspurious.

1. Correlation: In our example, the results revealed a statistically significant moderate

positive correlation (r = .52, p < .01); hence, Condition 1 for causation is satisfied.
2. Temporality: Temporality pertains to timing, wherein change in one variable
precedes change in the other variable. In our example, we can’t really determine if
people who are happy at home bring a joyful demeanor to the workplace or if it’s the
other way around—maybe people who had a very satisfying day at work come home
184 Part II • Statistical Tests

happy. In this case, temporality is ambiguous—we can’t really know what caused
what; hence, Condition 2 for causation is not satisfied.
3. Nonspurious: This condition is challenging to meet; it means that one variable, and
nothing else, influenced the change in the other variable. Naturally, it’s not easy to
rule out all other possible factors that could influence that second variable. In this
study, we don’t have enough information about the participants to account for factors
beyond the two tests that they were given. For example, there may be a third variable
such as personality type (pessimistic . . . optimistic) that accounts for the change(s)
observed in the variables; it may be that pessimistic people bring their negative
attitude into all settings (home and work) and optimistic people bring their positive
perspective into both settings, in which case, it wasn’t home happiness that influenced
workplace satisfaction or vice versa. Clearly, Condition 3 for causation is not satisfied.

This case demonstrates that although we’ve detected a statistically significant correla-
tion, wherein the two variables change in relation to each other, we cannot plausibly claim
causality—that change in one variable caused the change in the other.
Understanding the difference between correlation and causation will help you interpret
and document the results of Pearson and Spearman correlational analyses, without uninten-
tionally overstating your findings as causal, unless you can plausibly demonstrate that the
three causality conditions have been met.

GOOD COMMON SENSE

The Pearson and Spearman correlational tests are powerful statistical tools. It’s been said that
with great power comes great responsibility; as you conduct statistical analyses, you are likely
to encounter a multitude of continuous variables that could be paired and analyzed using
the Pearson or Spearman test. This is particularly true when conducting secondary analysis
(also known as data mining), wherein you may legitimately acquire a sizable dataset and cor-
responding codebook involving numerous variables that were gathered from another source.
While the Pearson and Spearman tests can promptly and accurately compute the correlation
of any two continuous variables, that does not necessarily mean that it would be wise to pair
any two continuous variables for such analysis.
For example, while it’s possible to compute the correlation between per capita consumption
of Mozzarella cheese and the number of lightning strikes in a remote rainforest, even if it resulted
in a statistically significant strong correlation (e.g., r = .92, p < .01, n = 1,000), on face value,
this variable pairing and the corresponding results are just silly. This would be an example of a
spurious correlation, wherein the statistical results appear noteworthy, but the real-world con-
nection between the two variables is essentially meaningless.
As discussed in Chapter 4: Descriptive Statistics, it is possible to use descriptive statistics
to compute the mean of a continuous variable like PhoneNumber, but the resulting average
phone number would be meaningless; it would be equally meaningless to run a correlation
 Chapter 8 • Correlation—Pearson Test and Spearman Test 185

between cheese consumption and electrical storms in a far-away land, unless you can cite a viable
linkage between these two variables.
Clearly, the computer will process anything that we order it to with no questions asked,
but as the thinking member of the computer–human partnership, it’s your responsibility
to balance curiosity with mindfulness when it comes to running and reporting meaningful
statistics.

KEY CONCEPTS

• Pearson correlation (r)

• Correlation
• Strength
• Direction
• Normality
• Linearity
• Homoscedasticity
• Scatterplot
• Regression line
• Spearman’s rho (ρ)
• Correlation vs. causation

PRACTICE EXERCISES

Exercise 8.1
Students have 30 days to submit an assignment. The instructor wants to determine if there is
a correlation between how long the student worked on the project and the score.
Dataset: Ch08_Ex01A.csv
Codebook:
Variable: Name
Definition: Last name of the student
Type: Categorical (unique identification of each participant)

Variable: Days
Definition: Number of days the student worked on the project
Type: Continuous (1 . . . 30)
186 Part II • Statistical Tests

Variable: Score
Definition: Score on the 40-point project
Type: Continuous (0 . . . 40)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality [for both variables], linearity,
homoscedasticity) and discuss your findings.
d. Run the correlation and document the n, r or rho, p value, scatterplot, and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the
correlation, and implications of your findings.
Repeat this exercise using dataset: Ch08_Ex01B.csv.

Exercise 8.2
A sociology student wants to determine if there’s a correlation between self-esteem and the
amount of time that people spend on social media. Participants agree to have a supplemen-
tal app temporarily loaded onto their cell phones that will track the total length of time the
person spends on social media. At the end of one week, the app will automatically administer
the Acme Self-Esteem Scale (ASES), after which, the app will transmit the social media time
data and ASES score to the researcher, and then the app will indicate that the study has con-
cluded and automatically delete itself.
Dataset: Ch08_Ex02A.csv
Codebook:
Variable: ParticipantID
Definition: First initial and last four digits of participant’s phone number
Type: Categorical (unique identification of each participant)

Variable: SocialMediaTime
Definition: Average minutes per day spent on social
Type: Continuous (0 . . . 1,440)

Variable: SelfEsteem
Definition: Score on the Acme Self-Esteem Scale
Type: Continuous (1 = very low self-esteem . . . 80 = very high self-esteem)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality [for both variables], linearity,
homoscedasticity) and discuss your findings.
 Chapter 8 • Correlation—Pearson Test and Spearman Test 187

d. Run the correlation and document the n, r or rho, p value, scatterplot, and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the
correlation, and implications of your findings.
Repeat this exercise using dataset: Ch08_Ex02B.csv.

Exercise 8.3
To determine if there’s a correlation between height and assertiveness, a sociologist recruits
a group of people who agree to take the Acme Assertiveness Survey, consisting of 50 yes/no
questions. They are also asked to indicate their height.
Dataset: Ch08_Ex03A.csv
Codebook:
Variable: Participant
Definition: Assigned participant code
Type: Categorical (unique identification of each participant)

Variable: Height
Definition: Height in inches
Type: Continuous (1 . . . no limit)

Variable: Assertiveness
Definition: Score on the Acme Assertiveness Survey
Type: Continuous (0 = low assertiveness . . . 50 = high assertiveness)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality [for both variables], linearity,
homoscedasticity) and discuss your findings.
d. Run the correlation and document the n, r or rho, p value, scatterplot, and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the
correlation, and implications of your findings.
Repeat this exercise using dataset: Ch08_Ex03B.csv.

Exercise 8.4
Upon being discharged from the hospital, a member of the Quality Improvement Department
visits each patient and asks them to complete a survey indicating how satisfied they were with
the care that they received. The employee also writes the patient’s age on the form to determine
if there’s a correlation between the patient’s level of satisfaction and their age.
188 Part II • Statistical Tests

Dataset: Ch08_Ex04A.csv
Codebook:
Variable: PatientID
Definition: Patient’s hospital identification number
Type: Categorical (unique identification of each participant)

Variable: Age
Definition: Age in years
Type: Continuous (1 . . . no limit)

Variable: PatientSatisfaction
Definition: Score on the Patient Satisfaction Survey
Type: Continuous (1 = very unsatisfied . . . 100 = very satisfied)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality [for both variables], linearity,
homoscedasticity) and discuss your findings.
d. Run the correlation and document the n, r or rho, p value, scatterplot, and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the
correlation, and implications of your findings.
Repeat this exercise using dataset: Ch08_Ex04B.csv.

Exercise 8.5
A political science student administered the following anonymous survey to a group of reg-
istered voters to determine if there’s a correlation between age and opinion of voting by mail:

1. How old are you? _____

2. How do you feel about voting by mail (please circle one number)?

1 2 3 4 5

Strongly dislike Dislike Neutral Like Strongly like

Please drop your completed card in the survey box.

Thank you.
 Chapter 8 • Correlation—Pearson Test and Spearman Test 189

Dataset: Ch08_Ex05A.csv
Codebook:
Variable: ID
Definition: Assigned identification code
Type: Categorical (unique identification of each participant)

Variable: Age
Definition: Age in years
Type: Continuous (18 . . . no limit)

Variable: VoteByMail
Definition: Opinion regarding voting by mail
Type: Continuous (1 = strongly dislike . . . 5 = strongly like)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality [for both variables], linearity,
homoscedasticity) and discuss your findings.
d. Run the correlation and document the n, r or rho, p value, scatterplot, and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the
correlation, and implications of your findings.
Repeat this exercise using dataset: Ch08_Ex05B.csv.

Exercise 8.6
To determine if sleep is correlated with solving skills, a researcher asks each individual how
many hours they slept last night, then uses a stopwatch to time how long it takes that person
to complete a 50-piece jigsaw puzzle.
Dataset: Ch08_Ex06A.csv
Codebook:
Variable: ParticipantCode
Definition: Assigned identification code
Type: Categorical (unique identification of each participant)

Variable: Sleep
Definition: Hours of sleep
Type: Continuous (0 . . . 24)
190 Part II • Statistical Tests

Variable: PuzzleTime
Definition: Seconds required to complete the puzzle
Type: Continuous (1 . . . no limit)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality [for both variables], linearity,
homoscedasticity) and discuss your findings.
d. Run the correlation and document the n, r or rho, p value, scatterplot, and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the
correlation, and implications of your findings.
Repeat this exercise using dataset: Ch08_Ex06B.csv.

Exercise 8.7
The researchers in a sleep lab want to find out if there’s a correlation between hours of sleep
per night and age by administering this survey to a group of volunteers:

1. Name ___________
2. How old are you? _____
3. What is your usual bedtime? _____
4. What time do you usually wakeup in the morning? _____

Please drop your completed card in the survey box.

Thank you.

Dataset: Ch08_Ex07A.csv
Codebook:
Variable: Name
Definition: Last name and first initial
Type: Categorical (unique identification of each participant)

Variable: Age
Definition: Age in years
Type: Continuous (1 . . . no limit)

Variable: SleepHours
Definition: Hours between bedtime and wakeup time
Type: Continuous (0 . . . 24)
 Chapter 8 • Correlation—Pearson Test and Spearman Test 191

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality [for both variables], linearity,
homoscedasticity) and discuss your findings.
d. Run the correlation and document the n, r or rho, p value, scatterplot, and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the
correlation, and implications of your findings.
Repeat this exercise using dataset: Ch08_Ex07B.csv.

Exercise 8.8
The manager of an ice cream parlor records the outdoor temperature (in Fahrenheit) and the
number of scoops of ice cream sold each day to determine if there is a correlation.
Dataset: Ch08_Ex08A.csv
Codebook:
Variable: Day
Definition: Day number
Type: Categorical (unique day number of the study)

Variable: Temperature
Definition: Average outdoor temperature during business hours
Type: Continuous (0 . . . no limit)

Variable: IceCream
Definition: The number of scoops of ice cream sold each day
Type: Continuous (0 . . . no limit)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality [for both variables], linearity,
homoscedasticity) and discuss your findings.
d. Run the correlation and document the n, r or rho, p value, scatterplot, and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the
correlation, and implications of your findings.
Repeat this exercise using dataset: Ch08_Ex08B.csv.
192 Part II • Statistical Tests

Exercise 8.9
A human resources manager administers a survey to employees to determine if there is a cor-
relation between commute time and employee satisfaction.
Dataset: Ch08_Ex09A.csv
Codebook:
Variable: EmployeeID
Definition: Employee number
Type: Categorical (unique identification of each participant)

Variable: CommuteTime
Definition: Minutes required to get from home to work
Type: Continuous (0 . . . no limit)

Variable: EmployeeSatisfaction
Definition: Level of workplace satisfaction
Type: Continuous (1 = very unsatisfied . . . 50 = very satisfied)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality [for both variables], linearity,
homoscedasticity) and discuss your findings.
d. Run the correlation and document the n, r or rho, p value, scatterplot, and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the
correlation, and implications of your findings.
Repeat this exercise using dataset: Ch08_Ex09B.csv.

Exercise 8.10
Theater students are required to select, memorize, and perform a 500-word monolog and
report how long it took them to memorize the script. The director will count the number
of mistakes made during the classroom performances to discover if there’s a correlation
between study time and recitation errors.
Dataset: Ch08_Ex10A.csv
Codebook:
Variable: StudentID
Definition: Identification number assigned to each student
Type: Categorical (unique identification of each participant)
 Chapter 8 • Correlation—Pearson Test and Spearman Test 193

Variable: StudyTime
Definition: Minutes spent learning the script
Type: Continuous (0 . . . no limit)

Variable: Errors
Definition: Number of recitation errors and omissions
Type: Continuous (0 . . . no limit)

a. Write the research question.

b. Write the hypotheses.
c. Run the criteria of the pretest checklist (normality [for both variables], linearity,
homoscedasticity) and discuss your findings.
d. Run the correlation and document the n, r or rho, p value, scatterplot, and
hypothesis resolution.
e. Write an abstract under 200 words detailing a summary of the study, the
correlation, and implications of your findings.
Repeat this exercise using dataset: Ch08_Ex10B.csv.
 9 CHI-SQUARE

To determine if there’s an association between two categorical variables, use the chi-square
test.

LEARNING OBJECTIVES

Upon completing this chapter, you will be able to:

9.1 Determine when to use the chi-square test
9.2 Build a research question and corresponding hypotheses
9.3 Run the chi-square test
9.4 Assess the pretest criteria: n ≥ 5 per cell
9.5 Order a bar chart
9.6 Interpret the results
9.7 Resolve the hypotheses
9.8 Write a concise abstract detailing the research question and results

195
196 Part II • Statistical Tests

VIDEO RESOURCE

The video for this chapter is Ch09 – Chi-Square Test.mp4. This video provides an overview
of this test, instructions for carrying out the pretest checklist, and running and interpreting
the results using the following dataset: Ch09_Demo.csv.

OVERVIEW—CHI-SQUARE TEST

The chi-square (chi is pronounced k-eye) test, sometimes written as X 2 (X is the Greek letter
chi), analyzes the association between categorical variables. Categorical variables can contain
as few as two categories (values); these are dichotomous (pronounced die-cot-uh-muss) vari-
ables, such as those shown in Table 9.1:

TABLE 9.1 ■ Examples of Dichotomous Categorical Variables

Variable Values

Elevator direction Up, Down

Voter status Voted, Did not vote

Climate Rain, No rain

Dishwasher status Clean, Dirty

Coin flip Heads, Tails

Categorical variables can also contain more than two categories (values); these are poly-
chotomous (pronounced poly-cot-uh-mus) variables (see table 9.2):

TABLE 9.2 ■ Examples of Polychotomous Categorical Variables

Variable Values

Appointment status On time, Late, Canceled, No-show, Reschedule

Marital status Single, Married, Separated, Divorced, Widowed

Visual aids None, Glasses, Contact lenses, Surgical correction

Transportation Car, Taxi, Motorcycle, Bicycle, Bus, Train, Plane, Walk

Movie type Adventure, Animation, Biography, Comedy, Crime, Documentary, Drama,

Family, Fantasy, Film-Noir, History, Horror, Musical, Mystery, Romance,
Sci-Fi, Sport, Thriller, War, Western
 Chapter 9 • Chi-Square 197

The chi-square statistic can process two categorical variables with the same number of
categories in both variables. For example, we might want to determine if Handedness (Left,
Right) is associated with winning a computer Game (Win, Lose). If the chi-square produces a
p value that’s less than or equal to .05, this would indicate that either left-handed people won
significantly more often than right-handed people or vice versa. Conversely, if the p value is
greater than .05, that suggests that both left- and right-handed players performed about the
same in terms of winning/losing.
As you might expect, chi-square also works when the two categorical variables contain
different numbers of categories. For example, suppose we wanted to know if Employment
was associated with level of Education; we could compare Employment, which contains three
categories (Full-time, Part-time, Unemployed), to Education, which contains six categories
(Did not complete high school, High school diploma, Associate’s degree, Bachelor’s degree, Master’s
degree, Doctorate). Further, you don’t need to specify how many categories are in each group;
chi-square will count that automatically.
Finally, since the chi-square test involves only categorical variables, and categorical vari-
ables do not produce a normal distribution, unlike prior chapters, this is exclusively a non-
parametric test; there is no parametric equivalent test.

CHI-SQUARE TESTS IN CONTEXT

Statistical Reasoning
The chi-square (X 2) statistic determines if there is a difference between the values that are
expected compared to the values that are actually observed among categorical variables.
Consider a bag of candy that contains three different colors (red, yellow, and green).
When we count them, we observe that 50% of the candies are red, 30% are yellow, and 20%
are green. Before we open a second bag of candy, we expect that this second bag would contain
the same proportions as the first bag. If after counting the candies in the second bag, we find
that the proportions that we observe are 49% red, 31% yellow, and 20% green, then we could
say that the observed proportions are close to the proportions that we expected. In this case, the
chi-square would produce a p value that’s greater than .05, indicating that there is no statisti-
cally significant difference in the color proportions between Bag 1 and Bag 2.
Alternatively, if we opened that second bag of candy and observed that it contained 5%
red, 5% yellow, and 90% green, this would be a substantial departure from the proportions
that we expected to see based on the percentages that we observed in the first bag of candy,
which was 50% red, 30% yellow, and 20% green. In this case, the chi-square would indicate
that there is a statistically significant difference in the proportions of these two bags, as indi-
cated by a p value that would be less than or equal to .05.

Applied Examples
Suppose the district supervisor of two libraries wants to determine if patrons at these
libraries value the same resources. The staff could set up a touchscreen survey at both
198 Part II • Statistical Tests

libraries (Northridge and Southside) at the checkout desk, asking which resource is their favor-
ite: Audiovisual, Books, Internet, Periodicals, or Storytime. If the data from both libraries were
combined, indicating the site (Northridge or Southside) and the resource selected (Audiovisual,
Books, Internet, Periodicals, or Storytime) for each entry, chi-square analysis would reveal the
proportions (percentages) selected at each site and a p value. If the p value is greater than .05, this
would indicate that there is no statistically significant difference in the patrons’ preferences at
the two libraries. Alternatively, if the p value was less than or equal to .05, this would indicate
that the preferences are statistically significantly different from each other at these two libraries.
Another example might involve a music teacher who wants to know if handedness (left-
handed or right-handed) is associated with the type of instrument that students opt to learn
(Brass, Keyboard, Percussion, Strings, or Woodwind). If the chi-square p value is greater
than .05, this would indicate that the proportion of instrument choices among left-handed
musicians is similar to the proportion of instrument choices among right-handed musicians.
However, if the p value is less than .05, this would indicate that left-handed students have sta-
tistically significantly different instrument preferences compared to right-handed students.

When to Use This Statistic

Guidelines for Selecting the Chi-Square Test

Overview: This statistic indicates if there is an association between two categorical variables.

Variables: This statistic requires two categorical variables for each record.

Results: To determine if the classroom and online (remote) versions of a course rendered
the same pass rate, we analyzed the grades of 33 classroom students and 122
online students on a pass/fail basis. Chi-square analysis revealed that 68% of the
students who learned in the classroom passed the course, statistically significantly
outperforming the 46% pass rate of those who took the course online (p = .03).

EXAMPLE

An educational designer wants to determine if classroom learning and online (remote) learn-
ing produce the same pass rate in a Safety Certification course.

Research Question
Do people who took the course in a classroom have the same pass rate as those who took the
class online?

Groups
We’ll assess the performance of two groups of learners: (1) classroom and (2) online.
 Chapter 9 • Chi-Square 199

Procedure
We sent all staff members an email indicating that they must complete a Safety Certification
course either in a traditional classroom setting with a live instructor alongside other students
or remotely online via a website that contains an instructional video. At the conclusion of the
1-hour training, participants must score at least 80% on a multiple-choice test to pass the
course.

Hypotheses

H0: There is no difference in the pass rate of classroom and online learners.
H1: There is a difference in the pass rate of classroom and online learners.

Dataset
Use the following dataset: Ch09_Demo.csv.
Codebook:
Variable: Staff
Definition: Name of the participant
Type: Categorical (unique identification of each participant)

Variable: TeachingMethod
Definition: Method of taking the class
Type: Categorical (Classroom, Online)

Variable: Outcome
Definition: Posttraining test result; minimum passing score is 80%
Type: Categorical (Pass, Fail)

To process a chi-square test, refer to the R Syntax Guide, page 10; we’ll proceed in order
starting from the top of that page.
First, click on the Packages tab, then check ggplot2 and gmodels.

Packages
☑ ggplot2
☑ gmodels
200 Part II • Statistical Tests

Chi-Square Pretest Checklist: ☑ n > 5 Per Cell

Test Run: Chi-Square
There is only one criterion for a chi-square test: There must be at least five entries per cell in the
cross-tabulation. This (cross-tabulation) table will emerge when we run the chi-square test,
so proceed with the chi-square run, and then we’ll check if the pretest criterion is satisfied.

Test Run: Chi-Square

Run cross-tabulation (chi-square):
CrossTable(FileName$IndependentVariable,FileName$DependentVariable,
chisq=T)

The variable that you enter as IndependentVariable will comprise the rows of the cross-
tabulation, and the DependentVariable will be the columns. It’s considered good practice to
assign the independent variable (IV) to the rows and the dependent (outcome) variable (DV)
to the columns. Typically, cross-tabulation reports are more intuitively readable when they’re
arranged this way.
After editing the file name and variable names, the R code and results should look like
this:

> CrossTable(Ch09_Demo$TeachingMethod,Ch09_Demo$Outcome,chisq=T)

Cell Contents

|-------------------------|

| N |

| Chi-square contribution |

| N / Row Total |

| N / Col Total |

| N / Table Total |

|-------------------------|

Total Observations in Table: 153

| Ch09_Demo$Outcome

Ch09_Demo$TeachingMethod | Fail | Pass | Row Total |

-------------------------|-----------|-----------|-----------|

Classroom | 10 | 21 | 31 |

| 1.893 | 1.868 | |
 Chapter 9 • Chi-Square 201

| 0.323 | 0.677 | 0.203 |

| 0.132 | 0.273 | |

| 0.065 | 0.137 | |

-------------------------|-----------|-----------|-----------|

Online | 66 | 56 | 122 |

| 0.481 | 0.475 | |

| 0.541 | 0.459 | 0.797 |

| 0.868 | 0.727 | |

| 0.431 | 0.366 | |

-------------------------|-----------|-----------|-----------|

Column Total | 76 | 77 | 153 |

| 0.497 | 0.503 | |

-------------------------|-----------|-----------|-----------|

Statistics for All Table Factors

Pearson's Chi-squared test

------------------------------------------------------------

Chi^2 = 4.716563 d.f. = 1 p = 0.02987339

Pearson's Chi-squared test with Yates' continuity correction

------------------------------------------------------------

Chi^2 = 3.883371 d.f. = 1 p = 0.04876655

Remember that in Chapter 4: Descriptive Statistics, we saw that the only thing you can
do with categorical variables is to simply count how many there are in each category (e.g., how
many blue-eyed people, how many brown-eyed people, how many green-eyed people), which
can be expressed in the form of n (the number of blue-eyed people, the number of brown-eyed
people, etc.) or percentage (the percentage of blue-eyed people, the percentage of brown-eyed
people, etc.).
Considering that chi-square analyzes two categorical variables, this cross-tabulation sim-
ply counted and totaled how many classroom students failed and passed and how many online
students failed and passed, and it presents the Ns and percentages for each pair of categories.
We’ll assess this cross-tabulation report in two passes: (1) First, we’ll observe the Ns to
determine if the pretest criterion is satisfied (n ≥ 5 per cell). (2) Next, we’ll focus on the per-
centages to comprehend the results of the chi-square.
202 Part II • Statistical Tests

Pass 1: Assess the Pretest Criterion; Observe the Ns

First, we need to determine if there are at least five in each pair of categories. Specifically,
there need to be at least five classroom students who passed, at least five classroom students who
failed, at least five online students who passed, and at least five online students who failed.
The small table at the top labeled Cell Contents provides the key for interpreting the five
lines of results in each cell. We see that the top row in each cell is the N, which is simply the
total count for each pair of categories.
Begin by looking at the Classroom row; among those who took the course in the classroom,
the table shows that 10 Classroom students Failed, and 21 Classroom students Passed. Next,
look to the Online row; among those who took the course online, 66 Online students Failed,
and 56 Online students Passed. Since all of these Ns are at least 5, the (one) pretest criterion is
satisfied.

Pass 2: Assess Percentages and the p Value

In our second pass through this cross-tabulation, start by referring to the Cell Contents
table at the top of this report; it indicates that the N / Row Total is on the third row of each
cell; this is the percentage that we’ll need to document our findings. Near the bottom of the
report, the Pearson’s chi-squared test p value (p = .02987339) will also be part of our results
documentation.

> CrossTable(Ch09_Demo$TeachingMethod,Ch09_Demo$Outcome,chisq=T)

Cell Contents

|-------------------------|

| N |

| Chi-square contribution |

| N / Row Total |

| N / Col Total |

| N / Table Total |

|-------------------------|

Total Observations in Table: 153

| Ch09_Demo$Outcome

Ch09_Demo$TeachingMethod | Fail | Pass | Row Total |

-------------------------|-----------|-----------|-----------|

Classroom | 10 | 21 | 31 |

| 1.893 | 1.868 | |
 Chapter 9 • Chi-Square 203

| 0.323 | 0.677 | 0.203 |

| 0.132 | 0.273 | |

| 0.065 | 0.137 | |

-------------------------|-----------|-----------|-----------|

Online | 66 | 56 | 122 |

| 0.481 | 0.475 | |

| 0.541 | 0.459 | 0.797 |

| 0.868 | 0.727 | |

| 0.431 | 0.366 | |

-------------------------|-----------|-----------|-----------|

Column Total | 76 | 77 | 153 |

| 0.497 | 0.503 | |

-------------------------|-----------|-----------|-----------|

Statistics for All Table Factors

Pearson's Chi-squared test

------------------------------------------------------------

Chi^2 = 4.716563 d.f. = 1 p = 0.02987339

Pearson's Chi-squared test with Yates' continuity correction

------------------------------------------------------------

Chi^2 = 3.883371 d.f. = 1 p = 0.04876655

Chi-square operates based on percentages; refer to the third line of each cell in the cross-
tabulation report, and multiply each number by 100 to derive the percentages (simply move
the decimal two places to the right). Table 9.3 succinctly organizes these key figures. You may
opt to include this concise table as part of your results report.
When it comes to comparing the pass rate of each group (Classroom and Online), we see
that 67.7% of the Classroom students passed, whereas 45.9% of the Online students passed;
since the p (.03) is less than or equal to .05, we would conclude that this is a statistically sig-
nificant finding. These findings indicate that when it comes to passing the course, students
who took the course in a Classroom significantly outperformed those who took it Online.
For exemplary purposes, observe Table 9.4, which expresses the pass rates in terms of N;
these results are drawn from the first row of each cell in the same cross-tabulation. Here we see
what appears to be a different outcome: Only 21 of the Classroom students passed, whereas 56
of the Online students passed, making it appear that Online outperformed Classroom learning,
204 Part II • Statistical Tests

TABLE 9.3 ■ Chi-Square Results in Percentages

Fail Pass

Classroom 32.3% 67.7%

Online 54.1% 45.9%

p = .03.

but notice the totals of each group: 21 classroom students passed, but that was out of a total
of 31 students, which is a 67.7% pass rate (21 31 = 67.7%). We also see that 56 of the Online
students passed, but that was out of a total of 122 Online students (56 122 = 45.9%). By using
percentages, we can get insightful results regardless of different group sizes since the percent-
age indicates the proportions on a 100-point (percent) scale, not just the N for each group.

TABLE 9.4 ■ Chi-Square Results in Ns

Fail Pass

Classroom 10 21

Online 66 56

p = .03.

Results
We’ll use the same protocol for interpreting the p value as we’ve used before:

• If p ≤ 0.05, this indicates that there is a statistically significant association between

the two variables.
• If p > 0.05, this indicates that there is no statistically significant association between
the two variables.

In this case, the chi-square test produced a p value of .02987339, which we can round to p
= .03; since this is less than or equal to .05, this indicates that there is a statistically significant
difference in the proportions (percentages) between the group outcomes. We can now pro-
ceed to resolve the hypotheses:

Hypothesis Resolution
The p value guides us in selecting which hypothesis to accept using the following rules:

• If p ≤ .05, accept the alternate hypothesis (H1).

• If p > .05, accept the null hypothesis (H0).
 Chapter 9 • Chi-Square 205

Per these rules, we can now revisit our hypotheses to make an informed selection:

REJECT H0: There is no difference in the pass rate of classroom and online learners.
ACCEPT H1: There is a difference in the pass rate of classroom and online learners.

Since the chi-square test produced p = .03, which is less than or equal to .05, we accept H1,
which tells us that there is a difference in the pass rate of classroom and online learners.

Documenting Results
We can now draft an abstract summarizing this study and the outcome, referring to the per-
centages in summary Table 9.3, the Ns in summary Table 9.4, and the p value from the cross-
tabulation report:

Abstract
As part of our annual training process, we emailed each staff member the option of taking the
required 1-hour Safety Certification course either in a classroom setting or remotely by watch-
ing a training video online. Of our 153 staff members, 31 (20.3%) opted to take the course in
a classroom with a live instructor along with their peers, and 122 (79.7%) chose to take the
course online. At the conclusion of the 1-hour training, participants must score at least 80%
on a multiple-choice test to pass the course.
Chi-square analysis revealed that 67.7% of those trained in the classroom passed the test,
statistically significantly outperforming those who took this training online, who had a 45.9%
pass rate (p = .03). Based on these findings, we will review and revise the contents of the
online training for this course, and we will assess the classroom/online performance of our
other courses.

Test Run: Bar Chart

In addition to analyzing and documenting the cross-tabulation report, you have the option to
produce a bar chart to better visualize the results of the chi-square.

Test Run: Bar Chart

Draw a bar chart:
ggplot(FileName,aes(x=CategoricalVariable1,fill=CategoricalVariable2))+
geom_bar(position=position_dodge())+theme_classic()
206 Part II • Statistical Tests

After editing the file name and variable names, the R code and bar chart should look like this:

> ggplot(Ch09_Demo, aes(x=TeachingMethod, fill=Outcome))+

geom_bar(position=position_dodge())+theme_classic()

This graph provides a visual representation of the data, but entering the variables in the
other order (exchanging the positions of TeachingMethod and Outcome) will use the same data
to produce a bar chart that’s arranged differently:

> ggplot(Ch09_Demo,aes(x=Outcome,fill=TeachingMethod))+
geom_bar(position=position_dodge())+theme_classic()
 Chapter 9 • Chi-Square 207

You may wish to include a bar chart to supplement your results report. After comparing
the two versions of the bar chart, select the one that you feel presents the clearest visualization.

GOOD COMMON SENSE

You’ve likely noticed a trend among the other inferential statistical tests, where we begin with
a parametric test, and then if the pretest checklist is not fully satisfied, we’d then opt for the
nonparametric version of that test. In the case of chi-square, it’s designed to process categori-
cal variables, which, by their nature, are nonparametric; they cannot produce a normal distri-
bution like continuous variables.
This means that if the pretest criterion (n ≥ 5 for each cell) is not fully satisfied, there’s
essentially nowhere (else) to go. If you encounter this, proceed with the chi-square statistic
but include a note detailing this limitation (e.g., We noted that two of the 12 cells contained an
n of less than 5, which is the minimum required to produce robust chi-square results). This hon-
est disclosure enables the reader the opportunity to consider the solidity of your findings in a
plausible light. Remember, even the most rigorous scientific studies have limitations.

KEY CONCEPTS

• Chi-square (X 2)
• Dichotomous variable
• Polychotomous variable
• n > .05 per cell
• Cross-tabulation
• Interpreting ns and percentages
• Bar charts

PRACTICE EXERCISES

Exercise 9.1
A public safety scientist is investigating driving records spanning the last 5 years in a com-
munity to determine if seniors (age 65 or older) tend to be involved in more automobile acci-
dents than younger drivers.
Dataset: Ch09_Ex01A.csv
Codebook:
Variable: DriverRecordID
Definition: Identity code of the driver
Type: Categorical (unique identification of each participant)
208 Part II • Statistical Tests

Variable: Age
Definition: Age of the driver
Type: Categorical (Under 65, 65 or older)

Variable: Accident
Definition: Was the driver involved in an automobile accident
Type: Categorical (Yes, No)

a. Write the research question.

b. Write the hypotheses.
c. Run the chi-square test and assess the criteria of the pretest checklist (n is at least 5 per
cell) and discuss your findings.
d. Per the results of the chi-square test, document the cell percentages, p value, hypothesis
resolution, and bar chart.
e. Write an abstract under 200 words detailing a summary of the study, the chi-square
test results, and implications of your findings.
Repeat this exercise using dataset: Ch09_Ex01B.csv.

Exercise 9.2
A college textbook manager wants to know if all students (Bachelors, Masters, Doctorate)
have the same preference in textbook form (paper book or e-book).
Dataset: Ch09_Ex02A.csv
Codebook:
Variable: StudentID
Definition: Identity code of the participant
Type: Categorical (unique identification of each participant)

Variable: Degree
Definition: Student’s degree program
Type: Categorical (Bachelors, Masters, Doctorate)

Variable: BookType
Definition: Student’s book preference
Type: Categorical (Paper, E-book)

a. Write the research question.

b. Write the hypotheses.
 Chapter 9 • Chi-Square 209

c. Run the chi-square test and assess the criteria of the pretest checklist (n is at least 5 per
cell) and discuss your findings.
d. Per the results of the chi-square test, document the cell percentages, p value, hypothesis
resolution, and bar chart.
e. Write an abstract under 200 words detailing a summary of the study, the chi-square
test results, and implications of your findings.
Repeat this exercise using dataset: Ch09_Ex02B.csv.

Exercise 9.3
The professor of an animal behavior laboratory course wants to determine if positive social-
ization has an effect on rats learning to run a maze. Students will train the rat for 15 minutes
a day for 3 weeks. Half of the students will be instructed to provide positive socialization
when interacting with their rat (e.g., handling the rat gently, petting the rat, talking kindly
to the rat, making eye contact with the rat); the other students will be instructed to handle
the rats in a safe but neutral fashion (e.g., no positive socialization).
Dataset: Ch09_Ex03A.csv
Codebook:
Variable: StudentName
Definition: Name of the participant
Type: Categorical (unique identification of each participant)

Variable: Socialization
Definition: Type of interaction students had with the rat
Type: Categorical (Neutral, Positive)

Variable: MazeCompletion
Definition: Rat run the maze in 60 seconds or less
Type: Categorical (Yes, No)

a. Write the research question.

b. Write the hypotheses.
c. Run the chi-square test and assess the criteria of the pretest checklist (n is at least 5 per
cell) and discuss your findings.
d. Per the results of the chi-square test, document the cell percentages, p value, hypothesis
resolution, and bar chart.
e. Write an abstract under 200 words detailing a summary of the study, the chi-square
test results, and implications of your findings.
Repeat this exercise using dataset: Ch09_Ex03B.csv.
210 Part II • Statistical Tests

Exercise 9.4
A supervisor at Acme Bank wants to determine if a person’s age (group) is associated with
their preferred bill-paying method. Each customer is handed a confidential self-administered
survey card:

Acme Bank is interested in how you prefer to pay your bills:

How old are you? Payment method:

18–25 Check
26–35 E-pay
36–55 Other
56–99

Please drop your completed card in the survey box.

Thank you.

Dataset: Ch09_Ex04A.csv
Codebook:
Variable: CardNo
Definition: Survey card serial number
Type: Categorical (unique identification of each participant)

Variable: Age
Definition: Age group of the customer
Type: Categorical (18–25, 26–35, 36–55, 56–99)

Variable: Payment
Definition: Payment method
Type: Categorical (Check, E-pay, Other)

a. Write the research question.

b. Write the hypotheses.
c. Run the chi-square test and assess the criteria of the pretest checklist (n is at least 5 per
cell) and discuss your findings.
d. Per the results of the chi-square test, document the cell percentages, p value, hypothesis
resolution, and bar chart.
e. Write an abstract under 200 words detailing a summary of the study, the chi-square
test results, and implications of your findings.
Repeat this exercise using dataset: Ch09_Ex04B.csv.
 Chapter 9 • Chi-Square 211

Exercise 9.5
A kindergarten teacher schedules two creative activity segments per day, one in the morning
and another in the afternoon, wherein students can freely select which activity they wish to
engage in. The teacher wants to determine if time of day (AM/PM) is associated with a stu-
dent’s choice during creative-play class time.
Dataset: Ch09_Ex05A.csv
Codebook:
Variable: Student
Definition: Name of the participant
Type: Categorical (unique identification of each participant)

Variable: Time
Definition: Before or after noon
Type: Categorical (AM, PM)

Variable: Activity
Definition: Student’s creative activity choice
Type: Categorical (Art, Blocks, Music)

a. Write the research question.

b. Write the hypotheses.
c. Run the chi-square test and assess the criteria of the pretest checklist (n is at least 5 per
cell) and discuss your findings.
d. Per the results of the chi-square test, document the cell percentages, p value, hypothesis
resolution, and bar chart.
e. Write an abstract under 200 words detailing a summary of the study, the chi-square
test results, and implications of your findings.
Repeat this exercise using dataset: Ch09_Ex04B.csv.

Exercise 9.6
Acme Confections makes three flavors of jellybeans: Raspberry, Lemon, and Lime (red, yel-
low, and green) and packages them in small bags and big bags. You want to determine if a
small bag has about the same proportion of flavors as a big bag.
Dataset: Ch09_Ex06A.csv
Codebook:
Variable: BagSize
Definition: Size of the package
Type: Categorical (Small, Large)
212 Part II • Statistical Tests

Variable: Flavor
Definition: Jellybean flavor
Type: Categorical (Raspberry, Lemon, Lime)

a. Write the research question.

b. Write the hypotheses.
c. Run the chi-square test and assess the criteria of the pretest checklist (n is at least 5 per
cell) and discuss your findings.
d. Per the results of the chi-square test, document the cell percentages, p value, hypothesis
resolution, and bar chart.
e. Write an abstract under 200 words detailing a summary of the study, the chi-square
test results, and implications of your findings.
Repeat this exercise using dataset: Ch09_Ex06B.csv.

Exercise 9.7
A public health official conducts a survey to determine if all age groups are equally receptive
to getting a flu shot using this survey card:

Flu Shot Survey

1. What is your age group?

1–17
18–64
65–100

2. Did you get a flu shot this year?

Yes
No

Thank you for your response.

Dataset: Ch09_Ex07A.csv
Codebook:
Variable: CardNo
Definition: Survey card number
Type: Categorical (unique identification of each participant)

Variable: AgeGroup
Definition: Respondent’s age group
Type: Categorical (1–17 [Pediatric], 18–64 [Adult], 65–100 [Senior])
 Chapter 9 • Chi-Square 213

Variable: FluShot
Definition: Respondent’s flu shot status
Type: Categorical (Yes, No)

a. Write the research question.

b. Write the hypotheses.
c. Run the chi-square test and assess the criteria of the pretest checklist (n is at least 5 per
cell) and discuss your findings.
d. Per the results of the chi-square test, document the cell percentages, p value, hypothesis
resolution, and bar chart.
e. Write an abstract under 200 words detailing a summary of the study, the chi-square
test results, and implications of your findings.
Repeat this exercise using dataset: Ch09_Ex07B.csv.

Exercise 9.8
The director of Library Services in Anytown conducted a survey to determine if patrons’
resource preferences are the same or different at the four public libraries. Upon exiting, peo-
ple at each library could answer a one-question survey as they check out using a convenient
touchscreen panel (the device records their response and automatically codes which library
the data were gathered from):

What is your favorite resource at this library?

(please touch one)

Audiovisual Books Internet

Periodicals Storytime

Dataset: Ch09_Ex08A.csv
Codebook:
Variable: ResponseNo
Definition: Response number
Type: Categorical (unique identification of each participant)

Variable: Library
Definition: Location of the library
Type: Categorical (Northridge, Southside)
214 Part II • Statistical Tests

Variable: FavoriteResource
Definition: Patron’s favorite resource
Type: Categorical (Audiovisual, Books, Internet, Periodicals, Storytime)

a. Write the research question.

b. Write the hypotheses.
c. Run the chi-square test and assess the criteria of the pretest checklist (n is at least 5 per
cell) and discuss your findings.
d. Per the results of the chi-square test, document the cell percentages, p value, hypothesis
resolution, and bar chart.
e. Write an abstract under 200 words detailing a summary of the study, the chi-square
test results, and implications of your findings.
Repeat this exercise using dataset: Ch09_Ex08B.csv.

Exercise 9.9
To help patients feel more comfortable at an infusion center, each person is offered the option
to listen to music, watch a video, or neither. The site manager observes and logs the choices
that patients who are under 65 make and the choices that patients who are 65 and older make
to discover if age is associated with this choice of diversion.
Dataset: Ch09_Ex09A.csv
Codebook:
Variable: PtID
Definition: Patient ID
Type: Categorical (unique identification of each participant)

Variable: Age
Definition: Patient’s age
Type: Categorical (Under 65, 65 & older)

Variable: Media
Definition: Choice of entertainment
Type: Categorical (Music, Video, Nothing)
a. Write the research question.
b. Write the hypotheses.
c. Run the chi-square test and assess the criteria of the pretest checklist (n is at least 5 per
cell) and discuss your findings.
 Chapter 9 • Chi-Square 215

d. Per the results of the chi-square test, document the cell percentages, p value, hypothesis
resolution, and bar chart.
e. Write an abstract under 200 words detailing a summary of the study, the chi-square
test results, and implications of your findings.
Repeat this exercise using dataset: Ch09_Ex09B.csv.

Exercise 9.10
To determine if there’s an association between handedness (left-handed/right-handed) and
the type of instrument that musicians play, a music instructor administers this survey to each
music student:

Musician Survey

1. Are you left- or right-handed?

Left
Right

2. What section is your (primary) instrument in?

Brass
Keyboard
Percussion
Strings
Woodwind

Thank you for your response.

Dataset: Ch09_Ex10A.csv
Codebook:
Variable: Name
Definition: Musician’s name
Type: Categorical (unique identification of each participant)

Variable: Handedness
Definition: Musician’s dominate hand
Type: Categorical (Left, Right)

Variable: Instrument
Definition: Musician’s instrument section
Type: Categorical (Brass, Keyboard, Percussion, Strings, Woodwind)
216 Part II • Statistical Tests

a. Write the research question.

b. Write the hypotheses.
c. Run the chi-square test and assess the criteria of the pretest checklist (n is at least 5 per
cell) and discuss your findings.
d. Per the results of the chi-square test, document the cell percentages, p value, hypothesis
resolution, and bar chart.
e. Write an abstract under 200 words detailing a summary of the study, the chi-square
test results, and implications of your findings.
Repeat this exercise using dataset: Ch09_Ex10B.csv.
 GLOSSARY

3P: A form of procedural Area sampling: A probability Cluster sampling: See Area
learning, specifically sampling technique typically sampling
demonstration-based learn- used to draw proportional ran-
Codebook: Documentation
ing, involving learning from a dom samples from multiple
detailing the variable defini-
video: (1) Play a segment of the domains (e.g., blocks spanning
tions in a dataset
video, (2) pause the video, (3) a community)
practice the procedure dem- Console: The lower-left
Availability sampling: See Con-
onstrated; repeat the process window in R
venience sampling
to completion
Continuous variable: A variable
Bar chart: A graphical rep-
.csv: See CSV file that contains a number along a
resentation of the numbers
continuum (e.g., Age = 0 . . . 100)
.dta: See Stata contained within a categorical
variable, consisting of a bar Control group: The group that
.RData: See R
chart receives either no treatment
.sas7bdat: See SAS or treatment as usual (TAU) to
Bell curve: See Normal curve.
serve as a comparison against
.sav: See SPSS
Bimodal: Two numbers are tied the group(s) that receive the
.txt: See Text file for the most common number (experimental) treatment
contained within a continuous
.xlsx: See Excel Convenience sampling: A non-
variable (both numbers are
probability sampling tech-
α error: See Type I error equally frequent within the
nique wherein those who are
variable)
α level: See Alpha level readily available are recruited
Bivariate correlation: Indicates as research participants
β error: See Type II error the direction and strength of
Correlation: Indicates the
the relationship between two
Δ%: See Delta % direction and strength of the
continuous variables gathered
relationship between two
Alpha level (α): The cutoff from each participant/data
continuous variables gathered
score for the p value; alpha is record
from each participant/data
typically set to .05, wherein p
Categorical variable: A variable record
values ≤ .05 suggest statisti-
that contains a discrete value
cally significant finding(s) Correlation direction: Positive
(e.g., Gender = Female/Male)
correlation direction (r > 0)
Alternate hypothesis (H1): The
Causation: Correlation dem- indicates that the variables
hypothesis that states that
onstrating that one variable move in the same direction (X↑
the treatment effect will be
influenced the outcome of and Y↑ or X↓ and Y↓); negative
significant; the score for the
another by meeting three correlation direction (r<0)
control group will be different
criteria: (1) association/corre- indicates that the variables
from the score for the treat-
lation, (2) temporality, and (3) move in the opposite direction
ment group
nonspurious relationship (X↑ and Y↓ or X↓ and Y↑)
ANOVA: Analysis of variance is
Chi-square: Indicates if there is Correlation strength: Cor-
similar to the t test, except it
a statistically significant dif- relations nearer to −1 or +1
compares all pairs of groups
ference between two categori- suggest stronger correlations
(G1:G2, G1:G 3, G2:G 3)
cal variables than correlations nearer to 0

217
218 Introductory Statistics Using R

Cross tabulation: A statistical the overall population from equally spaced and can be nega-
table that contains results which the sample was drawn tive (e.g., bank account balance)
based on column : row
GIGO: Acronym: “Garbage Linearity: Points on a scatter-
CSV files: Comma Separated In, Garbage Out” pertains to plot align in a (fairly) straight
Values datasets the necessity of entering and line
processing quality data to
Ctrl C: Copy Maximum: The highest number
produce quality results
contained within a continuous
Ctrl V: Paste
H0: See Null hypothesis variable
Ctrl L: Clear Console window
H1: See Alternate hypothesis Mean: The average of the
in R
numbers contained within a
Histogram: A graphical rep-
Datasets: Tables of alphanu- continuous variable
resentation of the numbers
meric information prepared
contained within a continuous Median: The center number
for statistical processing
variable, consisting of a bar contained within the sorted
Delta percent: Also represented chart list (lowest to highest) of a
as “Δ%,” expressing the variable
Homogeneity of variance: Simi-
change percentage in a vari-
larity of variance (SD2) among Minimum: The lowest number
able Δ% = (New − Old) ÷
two or more variables contained within a continuous
Old × 100
variable
Homoscedastic: See
Dependent variable: The vari-
Homoscedasticity Mode: The most common
able that contains the outcome
number contained within a
measurement (e.g., score) Homoscedasticity: The arrange-
continuous variable
ment of points on a scat-
Descriptive statistics: A sum-
terplot wherein most of the Multimodal: More than two
mary of a variable using
points are in the middle of the numbers are tied for the most
figures and graphs that can
distribution common number contained
characterize continuous or
within a continuous variable
categorical variables Hypothesis resolution: Using the
(the numbers are equally fre-
statistical results to deter-
Dichotomous: A categorical quent within the variable)
mine which hypothesis came
variable that contains two
true Multistage cluster sampling: See
values (e.g., Handedness:
Area sampling
Left/Right) Kruskal-Wallis omnibus test:
In ANOVA processing, this n: The total number (count) of
Disproportionate stratified sam-
test indicates if any group(s) items contained within a vari-
pling: A probability sampling
performed statistically sig- able for a sample
technique wherein the per-
nificantly differently from any
centage of items/participants N: The total number (count) of
other group(s)
selected from each stratum items contained within a vari-
does not match the percentage Import data: Transform data able for a population
in the population that were initially coded in a
Negative correlations: Among
foreign format to accurately
Excel: A spreadsheet program the specified pair of scores,
load into an application
that uses the dataset file one variable increases as the
extension: .xlsx Independent variable: The other decreases
variable that the researcher
Experimental group: See Treat- Nominal variable: A categorical
controls or an attribute of a
ment group variable wherein the values
participant (e.g., age)
have no sequence (e.g., Color =
External validity: The extent
Interval variable: A continuous Red, Green, Blue)
to which the results of the
variable wherein the values are
sample can be generalized to
 Glossary 219

Nonprobability sampling: A Packages: Software (subrou- Pretest/treatment/posttest:

sample wherein each item/ tines) that provide supplemen- Longitudinal design model,
participant does not have an tal functionality to R. typically using a single group,
equal chance of being selected wherein a pretest is adminis-
Paired t test: Indicates if there
to partake in the research tered, followed by the treat-
is a statistically significant
procedure ment, followed by the posttest,
difference between the
which involves (re)adminis-
Normal curve: The shape of the pretest and posttest (T1:T 2) for
tering the same instrument/
histogram of the normal dis- continuous variables
metric used at the pretest to
tribution. See Normal distribu-
Paired Wilcoxon test: Similar detect the effectiveness of the
tion, Normality, and Bell curve
to the paired t test, but used treatment
Normality: A symmetrical when the data distribution
Pretest checklist: Assumptions
distribution within a continu- (pretest–posttest) does not
regarding the characteris-
ous variable wherein most of meet normality criteria; the
tics of the data that must be
the values are near the mean, nonparametric version of the
assessed prior to running a
with a histogram resembling paired t test
statistical test
a symmetrical mountain. See
Pairwise combinations: The
Normal curve Probability sampling: A type
number of unique variable
of sampling wherein each
Normality of Differences: In pairings that ANOVA process-
item/participant has an equal
paired t test processing, this ing will produce (G = total
chance of being selected
criterion determines if the dif- number of groups); unique
to partake in the research
ferences between the pretest pairs = G! ÷ (2 x (G - 2)!)
procedure
score and posttest score form
Pearson test: Indicates the
a normal distribution Procedural learning: Observing
direction and strength of the
and then reproducing each
Null hypothesis (H0): The relationship between two
step of a process
hypothesis that states that the continuous variables gathered
treatment effect will be null; from each participant/data Proportionate stratified sam-
the score for the control group record pling: A probability sampling
will be the same as the score technique wherein the per-
Percentage (%): A method for centage of items/participants
for the treatment group
expressing a fraction in terms selected from each stratum
Omnibus test: In ANOVA pro- of 100 (% = Part ÷ Total × 100) matches the percentage in the
cessing, this test indicates if population
Polychotomous: A categorical
any group(s) performed statis-
variable that contains more Purposive sampling: A nonprob-
tically significantly differently
than two values (e.g., Meal: ability sampling technique
from any other group(s). See
Breakfast / Lunch / Dinner) wherein each potential par-
Kruskal-Wallis omnibus test.
Population: All of the mem- ticipant must meet multiple
O X O design: See Pretest/ criteria
bers/records (see Sampling)
treatment/posttest
Positive correlations: The Quota sampling: A nonprob-
Ordinal variable: A categorical ability sampling technique
specified pair of scores tends
variable wherein the values wherein the total number of
to increase or decrease
have a sequence (e.g., Meal = participants is specified prior
concurrently
Breakfast, Lunch, Dinner) to starting the data collec-
Power calculations: Formulas tion process; data collection
p value: A score generated by
that provide estimates speci- continues until the specified
inferential statistical tests
fying optimal sample size number of participants is
to indicate the probability
that the differences detected achieved
Pretest/posttest design: See
would emerge by chance alone Pretest/treatment/posttest
220 Introductory Statistics Using R

r: Symbol used with bivari- better comprehend the overall coefficient: See Spearman test
ate correlation (see Bivariate population, or a portion of
Spearman’s rho: See Spearman
Correlation) the population with specific
test
characteristics
R: A statistical program that
SPSS: A statistical program
uses the dataset file exten- Sampling bias: Any procedure/
that uses the dataset file
sion: .RData incident/factor that interferes
extension: .sav
with the process of gathering a
Random assignment: Ran-
representative sample Standard deviation (SD): A statis-
domly assigning members
tic that indicates the amount
to (control/experimental) Scatterplot: A graphical
of similarity/diversity among
groups reduces the likelihood representation of a bivariate
the numbers contained within
of creating biased/unbalanced correlation
a variable
groups
SD: See Standard deviation
Stata: A statistical program
Range: The maximum minus
Shapiro-Wilk test: A statistical that uses the dataset file
the minimum
test that assesses the normal- extension: .dta
Ratio variable: A continuous ity of a continuous variable
Stratified sampling: A prob-
variable wherein the values
Simple random sampling: A ability sampling technique
are equally spaced and cannot
probability sampling tech- wherein the sample frame is
be negative (e.g., Age)
nique wherein a set number split into two or more strata
Regression line: The line drawn of participants are randomly (lists) (e.g., Females/Males),
through a scatterplot that selected from a sample frame and then random selections
shows the average pathway are made from each stratum
Simple time-series design: See
through those points (list)
Pretest/treatment/posttest
Representative sample: A Summary statistics: See
Single-subject design (SSD):
sample that is proportionally Descriptive statistics
A statistical method that
equivalent to the population
measures the status of an Systemic sampling: A prob-
Research question: The inquiry individual repeatedly over a ability sampling technique
that forms the basis for the specified time frame wherein periodic selections of
hypotheses construction, items/participants are made
Skewed distribution: A nonnor-
analyses, and documentation
mal (asymmetrical) distri- t test: Indicates if there is a
of results
bution within a continuous statistically significant differ-
RStudio: A graphical user variable wherein most of the ence between the two groups
interface (GUI) for R numbers are either high or low (G1:G2) containing continuous
variables
Sample: A subset of the sample Snowball sampling: A nonprob-
frame or population specifying ability sampling technique Treatment group: The group(s)
those who will actually par- wherein the researcher that receives the (experi-
take in the research procedure requests each participant mental) treatment, which will
to provide referral(s) to be compared to the control
Sample frame: A subset of
other potentially suitable group
the population that could be
participants
accessed to comprise the Tukey test: A test used to detect
sample Spearman test: Similar to the pairwise score differences
Pearson test, but used when wherein the groups have equal
SAS: A statistical program that
data distribution does not ns; typically used as an ANOVA
uses the dataset file exten-
meet normality, linearity, and post hoc test
sion: .sas7bdat
homoscedasticity criteria; the
Type I error: Occurs when the
Sampling: The process of nonparametric version of the
findings indicate that there
gathering a (small) portion Pearson Correlation
is a statistically significant
of the population data to
 Glossary 221

difference between two erroneously accept the null Welch two-sample t test:
variables (or groups) (p ≤ hypothesis Similar to the t test, but
.05) when, in fact, on the used when data distribution
Unique pairs formula: Computes
whole, there actually is not, does not meet normality and
the total number of compari-
meaning that you would homogeneity of variance
sons that can be made when
erroneously reject the null criteria; the nonparametric
groups are gathered two at
hypothesis version of the t test
a time (G = total number of
Type II error: Occurs when the groups); unique pairs = G! ÷ Wilcoxon multiple pairwise
findings indicate that there is (2 × (G − 2)!) comparisons test: Similar to
no statistically significant dif- the Tukey test, but used when
Variables: A name assigned to a data distribution does not meet
ference between two variables
set of values normality and homogeneity of
(or groups) (p > .05) when, in
fact, on the whole, there actu- Variance: The standard devia- variance criteria; the nonpara-
ally is, meaning that you would tion squared (variance = SD2) metric version of the Tukey test
 INDEX

Alpha (α) Causation, 181–182 Control group, 8–11, 8–11

error, 107 Chi-square test (figure)
level, 106 applied examples, 195–196 Convenience sampling, 29, 29
Alternate hypothesis (H1), dataset, 197 (figure)
12–13, 104, 133, 153, dichotomous categorical Correlation. See also Pearson
180, 202 variables, 194 (table) test and Spearman
Analysis of variance (ANOVA), documenting results, test
118, 119, 124, 134 203–205 applied examples, 167
ANOVA—Tukey test and Wil- guidelines for selection, bivariate, 166
coxon multiple pair- 196 vs. causation, 181–182
wise comparisons hypothesis resolution, direction, 167–168, 168
test, 14 202–203 (table)
applied examples, 119 polychotomous categorical guidelines for selection, 167
descriptive statistics, variables, 194 (table) homoscedasticity, 172, 174,
126–130 pretest checklist, 198–202 176–178, 177 (figure)
documenting results, 134 results, 202 negative, 167–169, 168
guidelines for selection, 119 statistical reasoning, 195 (table), 170, 170
hypothesis resolution, 133 Cluster sampling. See Area (figure)
layered learning, 119 sampling positive, 167–169, 168
pretest checklist, 121–124 Codebook, 44–45 (table), 170, 170 (fig-
statistical reasoning, Comma-separated value (csv) ure), 181
118–119 files, 43–46, 71, 94 scatterplot, 169–170, 170
test run, 124–127, 130–132 Macintosh, 57 (figure), 174–177,
Area sampling, 27–29, 28 Windows computer, 56 176–177 (figure)
(figure) Console window, 52, 52 (figure) statistical reasoning,
Availability sampling. See Con- Continuous variables, 5–6, 8 166–167
venience sampling (table), 47 strength, 168, 168–169
descriptive statistics, 71–79 (table)
documenting results, 74, 76 Cross-tabulation, 198–201
Bar chart, 80–81, 203–205 histogram, 74–76 .csv. See Comma-separated
Bell curve, 75 maximum, 70 value (csv) files
Beta (β) error, 107–108 mean, 66 Ctrl C (copy), 42, 43, 47, 49, 50,
Bimodal, 68 mean vs. mode, 68 72, 73, 83
Bivariate correlation, 166 median, 67 Ctrl L, 52
minimum, 70 Ctrl V (paste), 42, 43, 47, 49, 50,
mode, 67–68, 73–74 72, 73, 83
Categorical variables, 6–7, 8 n, 66
(table), 50 normal distribution, 77, 77
bar chart, 80–81 (figure) Dataset
descriptive statistics, 66, range, 70 file types, 46, 46 (table)
79–81 skewed distribution, 78–79, order, 53–54
dichotomous, 194, 194 79 (figure) structure, 43–44
(table) standard deviation (SD), uploading, 45–46, 45 (figure)
polychotomous, 194, 194 68–69, 69 (figure) Delta (Δ) percent, 150–151
(table) variance, 69–70 Dependent variable, 174

223
224 Introductory Statistics Using R

Descriptive statistics Import data, 45–46, 45 (figure), Outliers, 78–79

ANOVA—Tukey test, 126–130 51, 55–56 O X O design, 144
applied examples, 65 Independent variable, 174, 198
categorical variables, 66, Interval variables, 6
79–81 Packages, 42–43
continuous variables, 66–79 Paired t test and paired Wil-
guidelines for selection, 65 Kruskal–Wallis omnibus test, coxon test
managing data, 81–82, 130–131 applied examples, 145–146
81–82 (figure) dataset, 147–148, 147 (table)
managing plots, 82–83, documenting results, 153
82–83 (figure) Linearity, 175, 176 (figure) guidelines for selection, 146
statistical reasoning, 65 hypothesis resolution, 153
3P learning method, 83 layered learning, 146
t test, 102–103 Math symbols, 52–53, 53 pretest checklist, 148–149
Dichotomous categorical vari- (figure) pretest–posttest design,
ables, 194, 194 (table) Maximum statistics, 70 144, 144 (figure), 154
Disproportionate stratified Mean (M), 66 results, 152–153
sampling, 26, 27 Mean vs. mode, 68 statistical reasoning, 145
(figure) Median, 67 test run, 149–152
.dta. See Stata Minimum statistics, 70 Pairwise combinations,
Mode, 67–68, 73–74 125–126
Multimodal, 68 Pearson test and Spearman
Evidence-based practice (EBP), Multistage cluster sampling. test, 15
5 See Area sampling documenting results, 181
Excel, 46, 54–56, 55 (figure) hypothesis resolution,
External validity, 20, 22, 24, 30, 34 180–181
n, 66, 198–199 pretest checklist, 172–178
N, 200–202 results, 180
Garbage In, Garbage Out Negative correlation, 167–169, test run, 178–180
(GIGO), 35, 58 168 (table), 170, 170 Percentage, 199–202, 202
Graphs, 50–51 (figure) (table)
Nominal variables, 7 Polychotomous categorical
Nonprobability sampling variables, 194, 194
H0. See Null hypothesis (H0) convenience, 29, 29 (figure) (table)
H1. See Alternate hypothesis (H1) purposive, 30 Population, sampling, 20, 22
Histogram, 74–77, 96, 96 (fig- quota, 30–31, 31 (figure) Positive correlation, 167–169,
ure), 98, 100 (figure), snowball, 31–32, 32 (figure) 168 (table), 170, 170
122–123, 149, 174 Normal curve, 77, 77 (figure), (figure), 181
Homogeneity of variance 177 Power calculations, 34
ANOVA test, 123–124 Normal distribution, 75, 77, 77 Pretest checklist
t test, 98–100, 100 (figure) (figure), 96 (figure) ANOVA—Tukey test,
Homoscedasticity, 172, 174, Normality, 96–98, 121–123, 121–124
176–178, 177 (figure) 173–174 chi-square test, 198–202
Hypothesis Normality of differences, paired t test, 148–149
alternate hypothesis (H1), 148–149 Pearson test, 172–178
12–13, 104, 133, 153, Null hypothesis (H0), 12–13, t test, 95–100
180, 202 94, 104, 107–108, 133, Pretest–posttest design, 144,
definition, 12 153, 180, 202 144 (figure), 154
null hypothesis (H0), 12–13, Pretest/treatment/posttest,
94, 104, 107–108, 133, 144, 144 (figure)
153, 180, 202 Omnibus test, 118, 124–125, Probability sampling
resolution, 13, 104, 133, 153, 130–131 area, 27–29, 28 (figure)
180, 202 Ordinal variables, 7 description, 24
 Index 225

proportionate and dispro- exporting results, 49–50, 50 Spearman test, 15, 172. See
portionate stratified, (figure) also Pearson test and
26, 26 (figure), 27 loading packages, 42–43 Spearman test
(figure) logging off, 57 correlation rho, 179–180
simple random, 24, 25 math equations, 52–53, 53 p values, 180
(figure) (figure) test run, 179–180
stratified, 24–25, 25 (figure) processing data, 54–57 SPSS, 40, 46
systematic, 26–27, 28 R Syntax Guide, 42 Standard deviation (SD), 68–69,
(figure) setting up, 41–42 69 (figure)
Procedural learning, 83 shortcuts, 51–52, 51–52 Stata, 40, 46
Proportionate stratified sam- (figure) Statistics, 4–5, 105–106. See
pling, 26, 26 (figure) statistical run, 47 also Descriptive
Purposive sampling, 30 uploading dataset, 45–46, 45 statistics
p value, 92, 106 (figure) Stratified sampling, 24–25, 25
variable references, 47–49, (figure)
48 (figure) disproportionate, 26, 27
Quasi-experimental design, R Syntax Guide, 42, 72, 79, 83, (figure)
154 95, 97, 134, 172, 197 proportionate, 26, 26 (figure)
Quota sampling, 30–31, 31 Summary statistics, 4, 64.
(figure) See also Descriptive
Sample, 20, 22–23 statistics
frame, 22 Systematic sampling, 26–27, 28
r, 166, 168–169, 168–169 (table) representative, 24 (figure)
R, 40, 70. See also RStudio Sampling
Cloud account bias, 33–34
math equations, 52–53, 53 description, 20 Temporality, 181–182
(figure) nonprobability, 29–32 Text, 43, 46, 72
uploading dataset, 45–46, 45 optimal sample size, 34 copy and paste, 49
(figure) probability, 24–29 editor, processing data,
Random assignment, 11 rationale, 21–22 56–57
Range, 70 representative sample, 24 proportional spacing, 49
Ratio variables, 6 terminology, 22–23 3P learning method, 83
.RData. See R tiers, 23 (figure) Treatment group, 8–11,
Regression line, 169, 170 (fig- SAS, 40, 46 8–11(figure)
ure), 174–175, 177 .sas7bdat. See SAS t test and Welch two-sample t
Representative sample, 24 .sav. See SPSS test, 14
Research principles Scatterplot, 169–170, 170 (fig- applied examples, 93
control and treatment ure), 174–177, 176–177 description, 92
groups, 8–11 (figure) descriptive statistics,
hypothesis, 12–13 SD. See Standard deviation 102–103
questions, 13–15, 94, 120, (SD) example, 93–106
147, 171, 196 Shapiro–Wilk test, 77–78, guidelines for selection, 93
random assignment, 11 96–99, 101–102, 148– hypotheses, 94
rationale for statistics, 4–5 149, 151–152, 173, 178 hypothesis resolution, 104
variable types, 5–8 Simple random sampling pretest checklist, 95–101
RStudio Cloud account (SRS), 24, 25 (figure) p value, 106
codebook, 44–45 Simple time-series design, statistical reasoning, 92
Console window, 52, 52 144 test run, 101
(figure) Single-subject design (SSD), 4 type I and II error, 107–108
copying graphs, 50–51 Skewed distribution, 78–79, 79 t test omnibus test, 124–125
data file types, 46, 46 (table) (figure) Tukey test. See ANOVA—Tukey
dataset order, 53–54 Snowball sampling, 31–32, 32 test
dataset structure, 43–44 (figure) .txt. See Text
226 Introductory Statistics Using R

Type I error, 107 interval, 6 Wilcoxon multiple pairwise

Type II error, 107–108 references, 47–49, 48 comparisons test,
(figure) 14, 118–119. See also
Variance, 69–70 ANOVA—Tukey test
Unique pairs, 134–135 and Wilcoxon multiple
pairwise comparisons
Welch two-sample t test. See test
Variables also t test and Welch p values, 132–133
categorical, 6–7, 8 (table), two-sample t test test run, 130–132
50 guidelines for selection, 93
continuous, 5–6, 8 (table), 47 statistical reasoning, 92
in dataset, 44–45 test run, 102–103 .xlsx. See Excel