For Mildred & Helen

Copyright © 2014 by SAGE Publications, Inc.

All rights reserved. No part of this book may be reproduced or utilized in any form or by any means, electronic or
mechanical, including photocopying, recording, or by any information storage and retrieval system, without permission
in writing from the publisher.

Printed in the United States of America

Library of Congress Cataloging-in-Publication Data
Knapp, Herschel.
Introductory statistics using SPSS / Herschel Knapp, University of California, Los Angeles.
pages cm
Includes index.
ISBN 978-1-4522-7769-1 (pbk.)
ISBN 978-1-4833-1310-8 (web pdf)
1. SPSS for Windows. 2. Social sciences—Statistical methods—Computer programs. I. Title.
HA32.K59 2014
005.5´5—dc23 2013024090
This book is printed on acid-free paper.

13 14 15 16 17 10 9 8 7 6 5 4 3 2 1
FOR INFORMATION:
SAGE Publications, Inc.
2455 Teller Road
Thousand Oaks, California 91320
E-mail: [email protected]
SAGE Publications Ltd.
1 Oliver’s Yard
55 City Road
London EC1Y 1SP
United Kingdom
SAGE Publications India Pvt. Ltd.
B 1/I 1 Mohan Cooperative Industrial Area
Mathura Road, New Delhi 110 044
India
SAGE Publications Asia-Pacific Pte. Ltd.
3 Church Street
#10-04 Samsung Hub
Singapore 049483

Acquisitions Editor: Helen Salmon

Assistant Editor: Katie Guarino
Editorial Assistant: Kaitlin Coghill
Production Editor: Libby Larson
Copy Editor: Gillian Dickens
Typesetter: C&M Digitals (P) Ltd.
Proofreader: A. J. Sobczak
Indexer: Will Ragsdale
Cover Designer: Edgar Abarca
Marketing Manager: Nicole Elliott
 Brief Contents
Preface
Acknowledgments
About the Author
Part I—Statistical Principles
1. Research Principles
2. Sampling
3. Working in SPSS
Part II—Statistical Processes
4. Descriptive Statistics
5. t Test
6. ANOVA
7. Paired t Test
8. Correlation and Regression
9. Chi-Square
Part III—Data Handling
10. Supplemental SPSS Operations
Part IV—Solutions to Odd-Numbered Exercises
INDEX
 Detailed Contents
Preface
Overview of Book
Organization
Part I—Statistical Principles
Part II—Statistical Processes
Overview of Statistical Functions
Part III—Data Handling
Part IV—Solutions to Odd-Numbered Exercises
Downloadable Learning Resources
Margin Icons
Acknowledgments
About the Author
PART I—STATISTICAL PRINCIPLES

1. Research Principles

Learning Objectives
Overview
Research Questions
Control and Experimental Groups
Rationale for Random Assignment
Hypothesis Formulation
Reading Statistical Outcomes
Accept/Reject Hypothesis
Levels of Measure
Nominal
Ordinal
Interval
Ratio
Types of Variables
 Categorical
Continuous
Key Concepts
Practice Exercises

2. Sampling

Learning Objectives
Overview
Rationale for Sampling
Time
Cost
Feasibility
Extrapolation
Sampling Terminology
Population
Sample Frame
Sample
Representative Sample
Probability Sampling
Simple Random Sampling
Systemic Sampling
Stratified Sampling
Disproportionate/Proportionate Sampling
Area Sampling
Nonprobability Sampling
Convenience Sampling
Snowball Sampling
Quota Sampling
Purposive Sampling
Sampling Bias
Key Concepts
 Practice Exercises

3. Working in SPSS

Learning Objectives
Digital Learning Resources
Overview
Two Views: Variable View and Data View
Variable View
Name
Type
Width
Decimals
Label
Values
Missing
Columns
Align
Measure
Role
Data View
Value Labels Icon
Codebook
Saving Data Files
Key Concepts
Practice Exercises
PART II—STATISTICAL PROCESSES

4. Descriptive Statistics

Learning Objectives
Digital Learning Resources
Overview
 Descriptive Statistics
Number (n)
Mean
Median
Mode
Standard Deviation
Variance
Minimum
Maximum
Range
SPSS—Loading an SPSS Data File
SPSS—Descriptive Statistics: Continuous Variables (Age)
Frequency Statistics Tables
Histogram With Normal Curve
Skewed Distribution
SPSS—Descriptive Statistics: Categorical Variables (Gender)
Frequency Statistics Tables
Bar Chart
SPSS—Descriptive Statistics: Continuous Variable (Age) Select by Categorical
Variable (Gender)—Females Only
SPSS—Descriptive Statistics: Continuous Variable (Age) Select by Categorical
Variable (Gender)—Males Only
SPSS—Selecting All Variables
Good Common Sense
Key Concepts
Practice Exercises

5. t Test

Learning Objectives
Digital Learning Resources
Overview
Example
 Research Question
Groups
Procedure
Hypotheses
Data Set
Pretest Checklist
Pretest Checklist Criterion 1—Normality
Pretest Checklist Criterion 2—Homogeneity of Variance
Pretest Checklist Criterion 3—n
Test Run
Results
Hypothesis Resolution
p Value
Document Results
Type I and Type II Errors
Good Common Sense
Key Concepts
Practice Exercises

6. ANOVA

Learning Objectives
Digital Learning Resources
Overview
Example
Research Question
Groups
Procedure
Hypotheses
Data Set
Pretest Checklist
Pretest Checklist Criterion 1—Normality
 Pretest Checklist Criterion 2—Homogeneity of Variance
Pretest Checklist Criterion 3—n
Test Run
Results
Comparison 1—Math as Usual vs. Z-Math
Comparison 2—Math as Usual vs. Flashcards
Comparison 3—Z-Math vs. Flashcards
Hypothesis Resolution
Document Results
Good Common Sense
Key Concepts
Practice Exercises

7. Paired t Test

Learning Objectives
Digital Learning Resources
Overview
Pretest/Posttest Design
Example
Research Question
Groups
Procedure
Hypotheses
Data Set
Pretest Checklist
Pretest Checklist Criterion 1—Normality of Differences
Test Run
Results
Hypothesis Resolution
Document Results
Good Common Sense
 Key Concepts
Practice Exercises

8. Correlation and Regression

Learning Objectives
Digital Learning Resources
Overview
Example
Research Question
Groups
Procedure
Hypotheses
Data Set
Pretest Checklist
Pretest Checklist Criterion 1—Normality
Test Run
Correlation
Scatterplot With Regression Line
Results
Scatterplot Points
Scatterplot Regression Line
Pretest Checklist Criterion 2—Linearity
Pretest Checklist Criterion 3—Homoscedasticity
Correlation
Hypothesis Resolution
Document Results
Good Common Sense
Correlation vs. Causation
Key Concepts
Practice Exercises

9. Chi-Square
 Learning Objectives
Digital Learning Resources
Overview
Example
Research Question
Groups
Procedure
Hypotheses
Data Set
Pretest Checklist
Pretest Checklist Criterion 1—n ≥ 5 per Cell Minimum
Test Run
Results
Finalizing Pretest Criteria
Hypothesis Resolution
Document Results
Good Common Sense
Key Concepts
Practice Exercises
PART III—DATA HANDLING

10. Supplemental SPSS Operations

Learning Objectives
Digital Learning Resources
Overview
Generating Random Numbers
Sort Cases
Select Cases
Recoding
Importing Data
 Importing Excel Data
Importing ASCII Data
Syntax
Good Common Sense
Key Concepts
Practice Exercises
PART IV—SOLUTIONS TO ODD-NUMBERED EXERCISES
INDEX
 Preface
Somewhere, something incredible is waiting to be known.

—Carl Sagan
OVERVIEW OF BOOK
This book covers the statistical functions used most frequently in social science
publications; this should not be considered a complete compendium of useful statistics,
however, since in other technological fields that you are likely already familiar with (e.g.,
word processing, spreadsheet, presentation software, etc.), you have probably discovered
that the “90/10 rule” applies: You can get 90% of your work done using only 10% of the
functions. For example, if you were to thoroughly explore each submenu of your word
processor, you would likely discover more than 100 functions and options; however, in
terms of actual productivity, 90% of the time, you are probably using only about 10% of
them to get your work done (e.g., save, load, copy, delete, paste, font, tab, center, print,
spell check). Back to statistics: If you can master the statistical processes contained in this
text, it is expected that this will arm you with what you need to effectively analyze the
majority of your own data and confidently interpret the statistical publications of others.
This is not a book about abstract statistical theory or the derivation of statistical
formulas; rather, this is a book about applied statistics. There are no formulas to memorize
in this text. This book is designed to provide you with practical answers to the following
questions: (1) If I have this kind of data, what statistical test should I use? (2) How do I set
up the data? (3) What parameters should I specify when ordering the test? and (4) How do
I interpret the results?
In terms of performing the actual statistical calculations, we will be using SPSS, a
sophisticated statistical processing software package. This facilitates speed and accuracy
when it comes to producing statistical results in the form of tables and graphs, but SPSS is
not an automatic program. In the same way that your word processor does not write your
papers for you, SPSS does not know what you want done with your data until you tell it.
Fortunately, those instructions are issued through clear menus. Your job will be to learn
what statistical procedure suits which circumstance, to configure the data properly, to
order the appropriate tests, and to mindfully interpret the output reports.
ORGANIZATION
This book is organized into four parts:
Part I—Statistical Principles
These chapters provide the basis for understanding the statistical concepts and SPSS
fundamentals that will be addressed in further detail in Part II.
Chapter 1—Research Principles
Chapter 2—Sampling
Chapter 3—Working in SPSS
Part II—Statistical Processes
These chapters contain the actual statistical procedures used to analyze data.
Chapter 4—Descriptive Statistics
Chapter 5—t Test
Chapter 6—ANOVA
Chapter 7—Paired t Test
Chapter 8—Correlation and Regression
Chapter 9—Chi-Square
After you have completed Chapters 4 through 9, the following table will help you
navigate this book to efficiently select the statistical test(s) best suited to your (data)
situation. For now, it is advised that you skip this table, as it contains statistical
terminology that will be covered thoroughly in the chapters that follow.
Overview of Statistical Functions

Part III—Data Handling

This chapter provides useful data handling techniques and SPSS tips and tricks to facilitate
efficient processing.
Chapter 10—Supplemental SPSS Operations
Part IV—Solutions to Odd-Numbered Exercises
This section provides fully developed solutions to the odd-numbered exercises for each
chapter.
DOWNLOADABLE LEARNING RESOURCES
In Part II, Exercises 1 through 5 have the codebook (description of each variable) and
corresponding data sets printed in the book. This will provide you with essential practice
in setting up SPSS data sets from the ground up. Mastering this skill will enable you to
analyze data from other sources. For convenience, larger data sets for Exercises 6 through
10 are available for download.
The website for this book is www.sagepub.com/knapp, which contains the following
resources:
SPSS data sets for each example
SPSS data sets for Part II, Exercises 6 through 10
SPSS files for Part III
SPSS tutorial videos
RESOURCES FOR INSTRUCTORS
Password-protected instructor resources are available on the website for this book at
www.sagepub.com/knapp and include:
Editable, chapter-specific PowerPoint slides
Solutions to even-numbered in-text questions
Sample course syllabi for semester and quarter courses
SPSS tutorial videos
SPSS data sets for all examples and exercises
MARGIN ICONS
This text includes icons to help you navigate (back) to critical points:
 Acknowledgments
SAGE and the author acknowledge and thank the following reviewers whose feedback
contributed to the development of this text:
Patricia F. Pearce, University of Alabama at Birmingham
Brenda K. Vollman, Loyola University New Orleans
Holly Johnson, University of Ottawa
Jamail D. Carter, Azusa Pacific University; Barstow Community College; United States
University
Timothy G. Ford, University of Louisiana at Monroe
Bruce H. Wade, Spelman College
Jo Brocato, California State University, Long Beach
Ellen P. McCann, George Mason University
Martha Vungkhanching, California State University, Fresno
Matthew Costello, University of Nebraska, Lincoln
Dazhi Yang, Boise State University
We also gratefully acknowledge the contribution of Dean Cameron, whose cartoons
enliven this book.
 About the Author
Herschel Knapp, PhD, MSSW, has 20 years of experience as a social
science researcher; for seven of those years he provided project
management for innovative implementations designed to improve the
quality of patient care throughout the Veterans Health Administration
via multisite, health science implementations. In addition to full-time
work at the VA, he teaches masters-level courses at the University of
California, Los Angeles and California State University, Los Angeles.
Dr. Knapp has served as the lead statistician on a longitudinal cancer
research project and managed the program evaluation metrics for a multisite, non-profit
children’s center. His clinical work includes emergency/trauma therapy in hospital
settings. The author of over 30 articles in peer-reviewed health science journals, he is also
the author of Therapeutic Communication: Developing Professional Skills, Second Edition
(SAGE, 2014) and Introduction to Social Work Practice: A Practical Workbook (SAGE,
2010). Dr. Knapp is currently involved in developing and implementing innovative
telehealth systems, utilizing videoconferencing technology to facilitate optimal healthcare
service delivery to remote patients, and to coordinate specialty consultations among
healthcare providers.
 PART I
Statistical Principles
I want the numbers to tell me a story.

—Anonymous
 CHAPTER 1
Research Principles

• Research Questions
• Control and Experimental Groups
• Rationale for Random Assignment
• Hypothesis Formulation
• Reading Statistical Outcomes
• Accept/Reject Hypothesis
• Levels of Measure
• Types of Variables
The scientific mind does not so much provide the right answers as ask the right questions.

—Claude Levi-Strauss
LEARNING OBJECTIVES
Upon completing this chapter, you will be able to:
• Identify various forms of research questions
• Differentiate between control and experimental groups
• Comprehend the rationale for random assignment
• Understand the basis for hypothesis formulation
• Understand the fundamentals of reading statistical outcomes
• Appropriately accept/reject hypotheses based on statistical outcomes
• Understand the four levels of measure
 • Determine the variable type: categorical or continuous
OVERVIEW
This chapter introduces statistical concepts that will be used throughout this book.
Applying statistics involves more than just processing tables of numbers; it involves being
curious and assembling mindful questions in an attempt to better understand what is going
on in a setting. As you will see, statistics extends far beyond simple averages and
headcounts. Just as a toolbox contains a variety of tools to accomplish a variety of diverse
tasks (e.g., screwdriver to place or remove screws, saw to cut materials, etc.), there are a
variety of statistical tests, each suited to address a different type of research question.
RESEARCH QUESTIONS
A statistician colleague of mine once said, “I want the numbers to tell me a story.” Those
nine words elegantly describe the mission of statistics. Naturally, the story depends on the
nature of the statistical question. Some statistical questions render descriptive (summary)
statistics, such as: How many minors visit a public park on weekends? How many cars
cross this bridge per day? What is the average age of students at a school? How many
accidents have occurred at this intersection? What percentage of people in a geographical
region has a particular disease? What is the average income per household in a
community? What percentage of students graduates from high school? Attempting to
comprehend such figures simply by inspecting them visually may work for a few dozen
numbers, but visual inspection of these figures would not be feasible if there were
hundreds or even thousands of numbers to consider. To get a reasonable idea of the nature
of these numbers, we can mathematically and graphically summarize them and thereby
better understand any amount of figures using descriptive statistics, as detailed in Chapter
4.
Another form of research question involves comparisons; often this takes the form of an
experimental outcome. Some questions may involve comparisons of scores between two
groups, such as: In a fourth-grade class, do girls or boys do better on math tests? Do
smokers sleep more than nonsmokers? Do students whose parents are teachers have better
test scores than students whose parents are not teachers? In a two-group clinical trial, one
group was given a new drug to lower blood pressure, and the other group was given a
placebo; is there a statistically significant difference between the blood pressure levels of
these two groups? These sorts of questions, involving the scores from two groups, are
answered using the t test, which is covered in Chapter 5.
Research questions and their corresponding designs may involve several groups. For
example, in a district with four elementary schools, each uses a different method for
teaching spelling; is there a statistically significant difference between spelling scores
from one school to another? Another example would be a clinical trial aimed at
discovering the optimal dosage of a new sleeping pill; Group 1 gets a placebo, Group 2
gets the drug at a 10-mg dose, and Group 3 gets the drug at a 15-mg dose; is there a
statistically significant difference between the groups in terms of number of hours of sleep
per night? Questions involving analyzing the scores from more than two groups are
processed using analysis of variance (ANOVA), which is covered in Chapter 6.
 Some research questions involve assessing the effectiveness of a treatment by
administering a pretest, then the treatment, then a posttest to determine if the group’s
scores improved after the treatment. For example, suppose it is expected that brighter
lighting may enhance mood. To test for this, the researcher administers a mood survey
under normal lighting to a group, which renders a score (e.g., 0 = very depressed, 10 =
very happy). Next, the lighting is brightened, after which that group is asked to reanswer
the mood test. The question is: According to the pretest and posttest scores, did the
group’s mood (score) increase significantly after the lighting was changed? Consider
another example: Suppose it is expected that physical exercise enhances math scores. To
test this, a fourth-grade teacher administers a multiplication test to each student. Next, the
students are taken out to the playground to run to the far fence and back three times, after
which the students immediately return to the classroom to take another multiplication test.
The question is: Is there a statistically significant difference between the test scores before
and after the physical activity? Questions involving before-and-after scores within a group
are processed with the paired t test, which is covered in Chapter 7.
Another kind of research question may seek to understand the (co)relation between two
variables. For example: What is the relationship between the number of homework hours
per week and grade point average (GPA)? We might expect that as homework hours go up,
GPA would go up as well. Similarly, we might ask: What is the relationship between
exercise and weight (if exercise goes up, does weight go down)? What is the relationship
between mood and hours of sleep per night (when mood is low, do people sleep less)?
Questions involving the correlation between two scores are processed using correlation
and regression, which are covered in Chapter 8.
Research questions may also involve comparisons between categories. For example: Is
there a difference in ice cream preferences (chocolate, strawberry, vanilla) based on
gender (male, female)—in other words, does gender have any bearing on ice cream flavor
selection? We could also investigate questions such as: Does the marital status of the
parents (divorced, not divorced) have any bearing on the child’s graduation from high
school (graduated, not graduated)? Questions involving comparisons among categories are
processed using chi-square (chi is pronounced k-eye), which is covered in Chapter 9.
As you can see, even at this introductory level, a variety of statistical questions can be
asked and answered. An important part of knowing which statistical test to reach for
involves understanding the nature of the question and the type of data at hand.
CONTROL AND EXPERIMENTAL GROUPS
Even if you are new to statistics, you have probably heard of control and experimental
groups. To understand the rationale for using this two-group system, consider an
investigation that uses an experimental group only. For prompt comprehension, we will
begin with a silly example and then move on to a more real-world example.
Example 1: The Acme Company has just released a new Monster Spray, guaranteed to
keep scary nighttime monsters from invading the bedroom all night. Clinical trials
involved issuing canisters of Monster Spray to 100 households; participants are instructed
to administer the Monster Spray every night for 1 month as directed and to journal the
number of scary nighttime monsters encountered each night. At the end of the month, the
data were gathered and processed, and based on statistical analysis, it was revealed that
zero monsters were encountered in these test households—hence, the Acme Company
marketing department proudly announces that scientific studies prove that Monster Spray
is safe and effective in warding off scary nighttime monsters from bedrooms.

Naturally, the question stands, “Can we really attribute the absence of monsters to the
use of Monster Spray?” In this single-group study design, we have no way of knowing
how many monsters would have invaded had the Monster Spray not been used. To assess
the effectiveness of Monster Spray in a more rigorous manner, we introduce a control
group. Control groups typically get no level of treatment, or sometimes a placebo, wherein
members of the control group are given something that appears to be a treatment but
actually has no therapeutic properties (e.g., a sugar pill). A placebo is often preferable to
administering absolutely no treatment because expectations can influence outcomes; to
clarify—if participants are given a placebo but believe that they have been given a viable
treatment designed to relieve their symptoms, some will begin to feel better through
mechanisms that we do not fully understand. From a statistical standpoint, we could test
the level of symptom relief experienced by individuals in the control group compared with
those in the experimental group. To call this treatment effective, we would need to see that
the experimental group statistically significantly outperforms the control group in terms of
symptom relief.
Getting back to our Monster Spray example, consider an enhanced design for the
experiment: 100 Monster Spray bottles are loaded with the newly formulated Monster
Repellant, and another 100 Monster Spray bottles are loaded with distilled water. Let us
also assume that the actual Monster Spray is indistinguishable from water—it produces a
clear, light, unscented mist. The bottles are then distributed to 200 households, but the
participants are not told which compound is in their spray bottle (the actual Monster Spray
or the distilled water); participants are instructed to use the Monster Spray nightly for 1
month and to record the number of creepy nighttime monster invasions in the journal for
each night. At the end of the month, the journals are collected, the data are entered and
processed, and we discover that there were zero monster invasions in the experimental
group (which got the actual Monster Repellant compound), which concurs with the initial
trial. We also learn that there were zero monster invasions in the control group (which got
the distilled water). Even without advanced statistical skills, it is becoming clear that the
Monster Spray did not outperform the water—hence, we would more confidently doubt
our initial conclusions that the absence of monsters is attributable to the use of Monster
Spray; this experiment shows that there would have been zero monsters anyway.
Now that you have the basic rationale for using an experimental group coupled with a
separate control group, consider a second, more realistic example.
Example 2: Suppose you conducted a one-group study and collected information only
on people who received a flu shot, and you discover that 95% of those individuals
remained healthy throughout the entire flu season. In isolation, this 95% finding seems
fairly impressive. Now consider an enhanced design, where in addition to surveying
people who had the flu shot, you also gather data on those who did not have a flu shot, and
you discover that 95% of those who did not get a flu shot remained healthy throughout the
flu season as well. Now suddenly that flu shot is starting to look considerably less
impressive. Having the second group (the control group) gives you a baseline to compare
with the experimental group. Intuitively, to determine the effectiveness of an intervention,
you are looking for substantial differences in the performance between the two groups—is
there a significant difference between the results of those in the experimental group
compared with the control group? The statistical tests detailed in Chapters 5 through 9
focus on different types of procedures for evaluating the difference(s) between groups
(experimental vs. control) to help determine the effect of the intervention–whether the
experimental group significantly outperformed the control group.
RATIONALE FOR RANDOM ASSIGNMENT
Understanding the utility of randomly assigning subjects to experimental/control groups is
best explained by example. Dr. Zinn and Dr. Zorders have come up with Z-Math, a
revolutionary system for teaching multiplication. The Z-Math package is shipped out to
schools in a local district to determine if it is more effective than the current teaching
method. The instructions specify that each fourth-grade class should be divided in half and
routed to separate rooms, with students in one room receiving the Z-Math teaching and
students in the other room getting their regular math lesson. At the end, both groups are
administered a multiplication test, with the results of both groups compared. The question
is: How should the class be divided into two groups? This is not such a simple question. If
the classroom is divided into boys and girls, this may influence the outcome; gender may
be a relevant factor in math skills—if by chance we send the gender with stronger math
skills to receive the Z-Math intervention, this may serve to inflate those scores.
Alternatively, suppose we decided to slice the class in half by seating—this introduces a
different potential confound; what if the half who sits near the front of the classroom are
naturally more attentive than those who sit in the back half of the classroom? Again, this
grouping method may confound the findings of the study. Finally, suppose the teacher
splits the class by age; this presents yet another potential confound—maybe older students
are able to perform math better than younger students. In addition, it is unwise to allow
subjects to self-select which group they want to be in; it may be that more proficient math
students, or students who take their studies more seriously, may systemically opt for the Z-
Math group, thereby potentially influencing the outcome.
Through this simple example, it should be clear that the act of selectively assigning
subjects to (control/experimental) groups can unintentionally affect the outcome of a
study; it is for this reason that we often opt for random assignment to assemble such
groups. In this example, the Z-Math instructions may specify that a coin flip be used to
assign students to each of the two groups: Heads assigns a student to Z-Math, and tails
assigns a student to the usual math teaching method. This random assignment method
ultimately means that regardless of factors such as gender, seating position, age, math
proficiency, and academic motivation, each student will have an equal (50/50) chance of
being assigned to either group. The process of random assignment will generally result in
roughly the same proportion of math-smart students, the same proportion of front- and
back-of-the-room students, and the same proportion of older and younger students being
assigned to each group. If done properly, random assignment helps to cancel out factors
endemic in subjects that may have otherwise tipped the findings one way or another.
HYPOTHESIS FORMULATION
Everyone has heard of the word hypothesis; hypotheses simply spell out each of the
anticipated possible outcomes of an experiment. In simplest terms, before we embark on
the experiment, we need one hypothesis that states that nothing notable happened (because
sometimes experiments fail). This would be the null hypothesis (H0), basically meaning
that the experiment had a null effect—nothing notable happened.
Another possibility is that something notable did happen (the experiment worked), so
we would need an alternate hypothesis that accounts for this (H1).
Continuing with the above example involving Z-Math, we first construct the null
hypothesis (H0); as expected, the null hypothesis states that the experiment produced null
results—basically, the experimental group (the group that got Z-Math) and the control
group (the group that got regular math) performed about the same; essentially, that would
mean Z-Math was no more effective than the traditional math lesson. The test hypothesis
(H1) is phrased indicating that the experimental (Z-Math) group outperformed the control
(regular math lesson) group. Hypotheses are typically written in this fashion:
H0: Z-Math and regular math teaching methods produced equivalent test results.
H1: Z-Math produced higher test results compared with regular teaching methods.
When the results are in, we would then know which hypothesis to reject and which to
accept; from there, we can document and discuss our findings.
In cases involving more test factors or groups, additional hypotheses (e.g., H2, H3, H4,
etc.) can be rendered; this will be demonstrated in Chapter 6—ANOVA.

Remember: In simplest terms, the statistics that we will be processing are designed to
answer the question: Do the members of the experimental group (that get the innovative
treatment) significantly outperform the members of the control group (who get no
treatment, a placebo, or treatment as usual)? As such, the hypotheses need to reflect each
possible outcome. In this simple example, we can anticipate two possible outcomes: H0
states that there is no significant difference between the experimental group and the
control group, suggesting that the treatment was ineffective. On the other hand, we need
another hypothesis that anticipates that the treatment will significantly outperform the
control condition; as such, H1 states that there is a significant difference in the outcomes
between the experimental and control conditions, suggesting that the treatment was
effective. The outcome of the statistical test will point us to which hypothesis to keep and
which to reject.
READING STATISTICAL OUTCOMES
Statistical tests vary substantially in terms of the types of research questions each are
designed to address, the format of the source data, their respective equations, and the
content of their results, which can include figures, tables, and graphs. Although there are
some similarities in reading statistical outcomes (e.g., means, alpha [α] level, p value),
these concepts are best explained in the context of working examples; as such, discussion
of how to read statistical outcomes will be thoroughly explained as each emerges in
Chapters 4 through 9.
ACCEPT/REJECT HYPOTHESIS
As is the case with reading statistical outcomes, the decision to accept or reject a
hypothesis depends on the nature of the test and, of course, the results: the alpha (α) level,
p value, and, in some cases, the means. Just as with reading statistical outcomes,
instructions for accepting/rejecting hypotheses for each test are best discussed in the
context of actual working examples; these concepts will be covered in Chapters 5 through
9.
LEVELS OF MEASURE
Selecting the proper statistical test partly depends on the type of variables that are
involved. There are four levels of measurement: nominal, ordinal, interval, and ratio.
Nominal
Nominal variables (also known as discrete or categorical variables) are used to represent
categories that defy ordering. For example, suppose you wish to code ice cream flavor
preference, and there are three choices: chocolate, strawberry, and vanilla. There is really
no way to put these in any order; for coding and computing purposes, we could assign 1 =
chocolate, 2 = strawberry, and 3 = vanilla. Since order does not matter among nominal
variables, these flavors could have just as well been coded 1 = vanilla, 2 = chocolate, and
3 = strawberry. Nominal variables may be used to represent categorical variables such as
gender (1 = female, 2 = male), agreement (1 = yes, 2 = no), religion (1 = atheist, 2 =
Buddhist, 3 = Catholic, 4 = Hindu, 5 = Jewish, 6 = Taoist, etc.), or marital status (1 =
single, 2 = married, 3 = separated, 4 = divorced, 5 = widow/widower). Since the numbers
are arbitrarily assigned to labels within a category, it would be inappropriate to perform
traditional arithmetic calculations on such numbers. For example, it would be foolish to
compute the average marital status (e.g., would 1.5 indicate a single married person?).
The same principle applies to other nominal variables such as religion or gender. There
are, however, appropriate statistical operations for processing nominal variables that will
be discussed in Chapter 4—descriptive statistics. In terms of statistical tests, nominal
variables are considered categorical.
Ordinal
Ordinal variables are similar to nominal variables in that numbers are assigned to
represent items within a category. Whereas nominal variables have no real rank order to
them (e.g., chocolate, strawberry, vanilla), the values in an ordinal variable can be placed
in a ranked order. For example, there is an order to educational degrees (1 = high school
diploma, 2 = associate’s degree, 3 = bachelor’s degree, 4 = master’s degree, 5 = doctorate
degree). Other examples of ordinal variables include military rank (1 = private, 2 =
corporal, 3 = sergeant, etc.) and meals (1 = breakfast, 2 = brunch, 3 = lunch, 4 = dinner, 5
= late-night snack). In terms of statistical tests, ordinal variables are considered categorical
variables.
Interval
Interval variables consist of numbers that have equal spacing between them, such as
numbers on a number line, ranging from −∞ to +∞; the distance between 1 and 2 is the
same as the distance between 2 and 3, which is the same as the distance between 3 and 4,
and so on. Some additional examples of interval variables are bank account balance
(which could be negative) and time (seconds, minutes, hours, days, etc.). Interval variables
are considered continuous variables.
Ratio
Ratio variables are similar to interval variables, except that interval variables can have
negative values, whereas ratio variables cannot be less than zero. Such variables include
measurements such as weight, distance, income, calories, academic grades (F = 0… A =
4), number of pets, number of siblings, or number of members in a group. Ratio variables
are considered continuous variables.

TYPES OF VARIABLES
When it comes to statistical processing, basically, variables fall into two types: categorical
and continuous. Comprehending the types of variables involved in a data set or research
design is essential when it comes to properly selecting, configuring, and running statistical
tests as detailed in Part II.
Categorical
Categorical variables involve assigning a number to an item in a category; the sequence of
the category may be irrelevant, as in nominal variables, wherein gender has no inherent
order and hence could be coded 1 = female, 2 = male or 1 = male, 2 = female. Ordinal
variables are also considered categorical; unlike nominal variables, ordinal variables do
have an inherent order to them, which should be preserved in the coding sequence, such as
1 = breakfast, 2 = lunch, and 3 = dinner.
Continuous
Continuous variables contain the kinds of numbers that you are accustomed to dealing
with. Continuous variables are interval- and ratio-type variables. Unlike categorical
variables, wherein numbers are assigned to represent categories, continuous variables
contain values that can stand up to actual arithmetic processing. For example, consider the
variable age; it is possible to compute the average (mean) age for a list of ages.

Key Concepts
• Research question
• Control group
 • Experimental group
• Random assignment
• Hypotheses
• Statistical outcomes
• Accepting/rejecting hypotheses
• Level of data (nominal, ordinal, interval, and ratio)
• Types of data (categorical and continuous)

Practice Exercises
Each of the following exercises describes the basis for an experiment that would render
data that could be processed statistically.
1. It is expected that aerobic square dancing during the 30-minute recess at an
elementary school will help fight childhood obesity.
a. State the research question
b. Identify the control and experimental group(s)
c. Explain how you would randomly assign participants to groups
d. State the hypotheses (H0 and H1)
e. Discuss the criteria for accepting/rejecting the hypotheses
2. Recent findings suggest that nursing home residents may experience fewer
depressive symptoms when they participate in pet therapy with certified dogs for 30
minutes per day.
a. State the research question
b. Identify the control and experimental group(s)
c. Explain how you would randomly assign participants to groups
d. State the hypotheses (H0 and H1)
e. Discuss the criteria for accepting/rejecting the hypotheses
3. A chain of retail stores has been experiencing substantial cash shortages in cashier
balances across 10 of their stores. The company is considering installing cashier
security cameras.
a. State the research question
b. Identify the control and experimental group(s)
c. Explain how you would randomly assign participants to groups
d. State the hypotheses (H0 and H1)
e. Discuss the criteria for accepting/rejecting the hypotheses
4. Anytown Community wants to determine if implementing a neighborhood watch
program will reduce vandalism incidents.
a. State the research question
b. Identify the control and experimental group(s)
c. Explain how you would randomly assign participants to groups
d. State the hypotheses (H0 and H1)
e. Discuss the criteria for accepting/rejecting the hypotheses
5. Employees at Acme Industries, consisting of four separate buildings, are chronically
late. An executive is considering implementing a get out of Friday free lottery; each
day an employee is on time, he or she gets one token entered into the weekly lottery.
a. State the research question
b. Identify the control and experimental group(s)
c. Explain how you would randomly assign participants to groups
d. State the hypotheses (H0 and H1)
e. Discuss the criteria for accepting/rejecting the hypotheses
6. The Acme Herbal Tea Company advertises that their product is “the tea that
relaxes.”
a. State the research question
b. Identify the control and experimental group(s)
c. Explain how you would randomly assign participants to groups
d. State the hypotheses (H0 and H1)
e. Discuss the criteria for accepting/rejecting the hypotheses
7. Professor Madrigal has a theory that singing improves memory.
a. State the research question
b. Identify the control and experimental group(s)
c. Explain how you would randomly assign participants to groups
d. State the hypotheses (H0 and H1)
e. Discuss the criteria for accepting/rejecting the hypotheses
8. Mr. Reed believes that providing assorted colored pens will prompt his students to
write longer essays.
a. State the research question
b. Identify the control and experimental group(s)
c. Explain how you would randomly assign participants to groups
 d. State the hypotheses (H0 and H1)
e. Discuss the criteria for accepting/rejecting the hypotheses
9. Ms. Fractal wants to determine if working with flashcards helps students learn the
multiplication table.
a. State the research question
b. Identify the control and experimental group(s)
c. Explain how you would randomly assign participants to groups
d. State the hypotheses (H0 and H1)
e. Discuss the criteria for accepting/rejecting the hypotheses
10. A manager at the Acme Company Call Center wants to see if running a classic
movie on a big screen (with the sound off) will increase the number of calls
processed per hour.
a. State the research question
b. Identify the control and experimental group(s)
c. Explain how you would randomly assign participants to groups
d. State the hypotheses (H0 and H1)
e. Discuss the criteria for accepting/rejecting the hypotheses
 CHAPTER 2
Sampling

• Rationale for Sampling

• Sampling Terminology
• Representative Sample
• Probability Sampling
• Nonprobability Sampling
• Sampling Bias

Ya gots to work with what you gots to work with.

—Stevie Wonder
LEARNING OBJECTIVES
Upon completing this chapter, you will be able to:
• Comprehend the rationale for sampling: time, cost, feasibility, and extrapolation
• Understand essential sampling terminology: population, sample frame, and
sample
• Derive a representative sample to facilitate external validity
• Select an appropriate method to conduct probability sampling: simple random
sampling, systemic sampling, stratified sampling, disproportionate sampling,
proportionate sampling, or area sampling
• Select an appropriate method to conduct nonprobability sampling: convenience
sampling, snowball sampling, quota sampling, or purposive sampling
• Understand techniques for detecting and reducing sample bias
OVERVIEW
Statistics is about processing numbers in a way to produce concise, readily consumable
information. One statistic that you are probably already familiar with is the average.
Suppose you wanted to know the average age of students in a classroom; the task would
be fairly simple—you could ask each person to write down his or her age on a slip of
paper and then proceed with the calculations. In the relatively small setting of a classroom,
it is possible to promptly gather the data on everyone, but what if you wanted to know the
age of all enrolled students or all students in a community? Now the mission becomes
more time-consuming, complex, and probably expensive. Instead of trying to gather data
on everyone, as in the U.S. census survey, another option is to gather a sample. Gathering
a sample of a population is quicker, easier, and more cost-effective than gathering data on
everyone, and if done properly, the findings from your sample can provide you with
quality information about the overall population.
You may not realize it, but critical decisions are made based on samples all the time.
Laboratories process thousands of blood samples every day. On the basis of the small
amount of blood contained in the test tube, a qualified health care professional can make
determinations about the overall health status of the patient from whom the blood was
drawn. Think about that for a moment: A few CCs (cubic centimeters) of blood are
sufficient to carry out the tests to make quality determinations; the laboratory did not need
to drain the entire blood supply from the patient, which would be time-consuming,
complicated, expensive, and totally impractical—it would kill the patient. Just as a small
sample of blood is sufficient to represent the status of the entire blood supply, proper
sampling enables us to gather a small and manageable bundle of data from a population of
interest, statistically process that data, and reasonably comprehend the larger population
from which it was drawn.
RATIONALE FOR SAMPLING
Moving beyond the realm of a statistics course, statistics takes place in the real world to
answer real-world questions. As with most things in the real world, gathering data
involves the utilization of scarce resources; key concerns involve the time, cost, and
feasibility associated with gathering quality data. With these very real constraints in mind,
it is a relief to know that it is not necessary to gather all the data available; in fact, it is rare
that a statistical data set consists of figures from the entire population (such as the U.S.
census). Typically, we proceed with a viable sample and extrapolate what we need to
know to better comprehend the larger population from which that sample was drawn. Let
us take a closer look at each of these terms.
Time
Some consider time to be the most valuable asset; time cannot be manufactured or stored
—it can only be used. Time spent doing one thing means that other things must wait.
Spending an exorbitant amount of time gathering data from an entire population precludes
the accomplishment of other vital activities. For example, suppose you are interested in
people’s opinions (yea/nay) regarding the death penalty for a paper that you are drafting
for a course. Every minute you spend gathering data postpones your ability to proceed
with the completion of the paper, and that paper has a firm due date. In addition, there are
other demands competing for your time (e.g., other courses, work, family, friends, rest,
recreation, etc.). Sampling reduces the amount of time involved in data gathering,
enabling you to statistically process the data and proceed with the completion of the
project within the allotted time.
Another aspect of time is that some (statistical) answers are time sensitive. Political
pollsters must use sampling to gather information in a prompt fashion, hence leaving
sufficient time to interpret the findings and fine-tune campaign strategies prior to the
election; they simply do not have time to poll all registered voters—a well-drawn sample
is sufficient.
Cost
Not all data are readily available (for free). Some statistical data may be derived from
experiments or interviews, which involves multiple costs, including a recruitment
advertising budget, paying staff to screen and process participants, providing reasonable
financial compensation to study participants, facility expenses, and so on. Surveys are not
free either; expenses may include photocopying, postage, website implementation charges,
telephone equipment, and financial compensation to study participants and staff.
Considering the costs associated with data collection, one can see the rationale for
resorting to sampling as opposed to attempting to gather data from an entire population.
Feasibility
Data gathering takes place in the real world—hence, real-world constraints must be
reckoned with when embarking on such research. Due to time and budgetary constraints, it
is seldom feasible or necessary to gather data on a population-wide basis; sampling is a
viable option. In the example involving the blood sample, clearly it is neither necessary
nor feasible to submit the patient’s entire blood supply to the lab for testing—quality
determinations can be made based on well-drawn samples. In addition, if a research
project focuses on a large population (e.g., all students in a school district) or a population
spanning a large geographical region, it may not be feasible to gather data on that many
people—hence, sampling makes sense.
Extrapolation
It turns out that by sampling properly, it is unnecessary to gather data on an entire
population to achieve a reasonable comprehension of it. Extrapolation involves using
sampling methods and sampling statistics to analyze the sample of data that was drawn
from the population. If done properly, such findings help us to (better) understand not only
the smaller (sample) group but also the larger group from which it was drawn.
SAMPLING TERMINOLOGY
As in any scientific endeavor, the realm of sampling has its own language and methods.
The following terms and types of sampling will help you comprehend the kinds of
sampling that you may encounter in scientific literature and provide you with viable
options for carrying out your own studies. We will begin with the largest realm (the
population) and work our way down to the smallest (the sample).
Population
The population is the entire realm of people (or items) that could be measured or counted.
A population is not simply all people on the planet; the researcher specifies the population,
which consists of the entire domain of interest. For example, the population may be
defined as all students who are currently enrolled at a specified campus. Additional
examples of populations could be all people who reside in a city, all people who belong to
a club, all people who are registered voters in an election district, or all people who work
for a company. As you might have surmised, the key word here is all.
Sample Frame
If the population that you are interested in is relatively small (e.g., the 5 people visiting the
public park, the 16 people who are members of a club, etc.), then gathering data from the
entire population is potentially doable. More often, the population is larger than you can
reasonably accommodate, or you may be unable to attain a complete list of the entire
population that you are interested in (e.g., all students enrolled in a school, every
registered voter in an election district, etc.). The sample frame, sometimes referred to as
the sampling frame, is the part of a population that you could potentially access. For
example, Acme University publishes a list of student names and e-mail IDs in the form of
a downloadable report on the school’s public website. If this list included every single
student enrolled, it would represent the entire population of the school; however, students
have the privilege to opt out of this list, meaning that each student can go to his or her
online profile and check a box to include or exclude his or her name and e-mail ID from
this public roster. Suppose the total population of the university consists of 30,000
enrolled students, and 70% chose to have their name appear on this list; this would mean
that the sample frame, the list from which you could potentially select subjects, consists of
21,000 students (30,000 × .70).
Sample
A sample is a portion of individuals selected from the sample frame. Certainly 21,000 is
considerably less that 30,000, but that may still be an unwieldy amount for your purposes.
Consider that your investigation involves conducting a 1-hour interview with participants
and that each participant will be compensated $10 for his or her time; the subject fee
budget for this study would be $210,000, and assuming you conducted back-to-back
interviews for 8 hours a day, 7 days a week, you would have your data set in a little over 7
years. Considering the constraints mentioned earlier (time, cost, and feasibility), you can
probably already see where this is going: (1) Is a $210,000 budget for subjects really
feasible? (2) Do you really have 7 years to gather your findings? (3) Most of the students
on this list will not be students 7 years from now. (4) After students graduate, their e-mail
IDs may change. In this case, accessing the entire sample frame is untenable, but the
sample frame is still useful; instead of attempting to recruit the 21,000 students, you may
choose to gather information from a subset of 100 students from this sample frame. These
100 students will constitute the sample. Selecting a sample of 100 students from the
sample frame of 21,000 means that your subject fee budget is reduced from $210,000 to
$1,000, and instead of taking more than 7 years to gather the data, using the same
interviewing schedule, you would have your complete data set in less than 2 weeks. In
terms of feasibility, sampling is clearly the way to go. As for how to select that sample of
100 students from among the sample frame of 21,000, there are a variety of techniques
covered in the sections on probability sampling and nonprobability sampling.
Just to recap, you can think of the population, sample frame, and sample as a hierarchy:

The population is the entire realm of those in a specified set (e.g., every person who
lives in a city, all members of an organization or club, all students enrolled on a
campus).
The sample frame is the list of those who could be potentially accessed from a
population.
The sample is the sublist of those selected from the sample frame who you will (attempt
to) gather data from.

REPRESENTATIVE SAMPLE
You may not realize it, but you already understand the notion of a representative sample.
Suppose you are at a cheese tasting party, and the host brings out a large wheel of cheese
from Acme Dairy. You are served a small morsel of the cheese and, based on that, decide
if you like it enough to buy a hunk of it or not. The assumption that you are perhaps
unknowingly making is that the whole rest of that big cheese wheel will be exactly like the
tiny sample that you tasted. You are presuming that the bottom part of the cheese is not
harder, that the other side of the cheese is not sharper, that the middle part of the cheese is
not runnier, and so on. Essentially, you are assuming that the sample of cheese that you
tasted is representative of the flavor and consistency of the whole wheel of cheese. This is
what a representative sample is all about: The small sample that you drew is
proportionally representative of the overall population (or big cheese) from which it was
taken. Often, it is the goal of researchers to select a representative sample, thereby
facilitating external validity—meaning that what you discover about the sample can be
viably generalized to the overall population from which the sample was drawn.
Sampling is often about gathering a manageable set of data so that you can learn
something about the larger population through statistical analysis. The question remains:
How do you get from the population, to the sample frame, to the actual representative
sample? Depending on the nature of the information that you are seeking and the
availability of viable participants/data, you may opt to employ probability sampling or
nonprobability sampling methods.
PROBABILITY SAMPLING
You can think of probability sampling as equal-opportunity sampling, meaning that each
potential element (person/data record) has the same chance of being selected for your
sample. There are several ways of conducting probability sampling.
Simple Random Sampling
Simple random sampling begins with gathering the largest sample frame possible and then
assigning sequential numbers to the items (1, 2, 3, . . ., 1,000). For this example, let us
assume that there are 1,000 items on this list and you want to recruit 30 participants; you
could use SPSS to generate 30 random numbers ranging from 1 to 1,000. It is not essential
that you perform this SPSS procedure at this time; Chapter 10 (Supplemental SPSS
Operations) has a section that provides step-by-step instructions for generating random
numbers to your specifications.
Systemic Sampling
Systemic sampling uses a simple periodic selection process to derive the sample. Suppose
you had a list of 1,000 (numbered) records and you decide that your target sample size
will be 10 items. Begin by dividing the total number of records by the target sample size
(1,000 ÷ 10 = 100); the solution (100) is the “k” or skip term. Next, you need to identify
the start point; this will be a random number between 1 and k; for this example, suppose
the randomly derived start point number is 71. The selection process begins with item 71
and then skips ahead k (100) items at a time to select the next item(s) in the sample, which
would be the following 10 items: 71, 171, 271, 371, 471, 571, 671, 771, 871, and 971.
Stratified Sampling
The sampling techniques described so far have used the entire sample frame from which
to draw the sample. Suppose you want to specifically control for the number of females
and males in your sample; stratified sampling would ensure that your sample is balanced
by gender. To draw a stratified sample based on gender, divide your sample frame into two
lists (strata): female and male. Next, decide on the number of items (n) that you would like
each list to consist of and use simple random sampling or systemic sampling to make
selections from each list. In this case, stratified sampling enables you to process the exact
number of females and males you wish to involve in your study as opposed to leaving
those counts to chance.
Disproportionate/Proportionate Sampling
Within stratified sampling, you can further specify if you want to work with a
disproportionate or proportionate sample. Continuing with the gender stratification
example, suppose you wished to conduct a student survey on a campus, and the student
population consists of 60% females and 40% males. For simplicity, the sample frame
consists of 10,000 records: 6,000 females and 4,000 males. The target sample size is 100.
The first step is to split the sample frame into two strata (lists): female and male. You
then have the option to draw a disproportionate sample or a proportionate sample. To draw
a disproportionate sample, you would randomly select 50% from the female stratum and
the other 50% from the male stratum.
Alternatively, if you wanted to gather a proportionate sample, which might better
represent the known gender proportions of this campus, then you would randomly select
60% of your sample from the female stratum and 40% from the male stratum.
The examples above used a two-stratum structure (gender: female, male), but you may
introduce any number of strata. For example, you may wish to process figures pertaining
not only to gender (female, male) but also to class ranking (freshman, sophomore, junior,
senior), which would involve eight strata (lists) to randomly select from, as shown in
Table 2.1.
Although there are no fixed limits when it comes to the number of strata that you can
establish, practicality is an issue: Every stratum that you specify divides the data into
smaller groups. When it comes to processing statistics, the solidity of the statistical tests
can be compromised if the sample size within the stratum becomes too low.
If you encounter a stratum that has a particularly low count, this could compromise the
power (solidity) of the statistical processes. Generally speaking, statistical results are
considered more robust when they are derived from larger samples. When the sample sizes
for one or more strata are low because they are proportionally low, it may be appropriate
to opt for disproportionate sampling to help increase the number of data elements in such
lacking strata. For instance, a data set consisting of four strata that would naturally render
proportions of 38%, 36%, 18%, and 8% could be rebalanced using disproportionate
sampling to compensate for the low 8% stratum, resulting in groups with 25% in each
stratum.
Table 2.1 Eight Strata Derived From Two Levels of Gender (Female, Male) and
Four Levels of Class Ranking (Freshman, Sophomore, Junior, Senior)

Area Sampling
Area sampling, also referred to as cluster sampling or multistage cluster sampling, is
typically used when it comes to gathering samples from a geographical region or when a
sample frame is not available. Since the characteristics of neighborhoods and their
residents can vary substantially from block to block, this form of sampling is useful when
it comes to appropriately capturing data spanning a geographical area.
This multistage process begins with acquiring a sample frame containing a list of
domestic addresses within a given geographical domain. This example involves the city of
Anytown, which consists of a population of about 60,000 people spanning 500 residential
blocks. Your goal is to gather a sample of data from 1,000 households. Begin by building
the block strata, which will consist of 500 strata (block 1, block 2, block 3… block 500).
Each stratum contains the addresses of each dwelling on that block. Next, randomly select
n number of blocks (strata) and sample everyone from the selected blocks or select a
randomly selected subset from those previously selected blocks, enough to meet your
target sample.
For clarity, this example presumed uniform population density across the 500 blocks of
Anytown—that each block contains about the same number of dwellings and about the
same number of people living in each dwelling. Naturally, prior to selecting the sample, it
would be wise to statistically check this assumption and consider adjusting the sampling
proportions to best represent the residents of Anytown. For example, if it is found that 5%
of the residents of Anytown live on one block, then it would be appropriate to draw 5% of
the overall sample from that single block.
NONPROBABILITY SAMPLING
In probability sampling, each item in the sample frame has an equal chance of being
selected when it comes to being included in the sample. In nonprobability sampling, either
the sample frame does not exist or the researcher does not consider it relevant to the
investigation at hand.
As you might expect, since the elements that will constitute the sample are not
randomly drawn from a sample frame, which would be drawn from the overall population,
external validity (the ability to viably generalize your sample findings to the overall
population) is the first casualty of nonprobability sampling. Whatever is learned from a
nonprobability sample, informative as it may be, cannot be generalized to the overall
population; the proportions would be off, and since we do not know the size or
characteristics of the source population, we would have no way of computing by how
much they are off. Still, much can be learned from nonprobability sampling.

Convenience Sampling
Convenience sampling, sometimes referred to as availability sampling, is exactly what it
sounds like; the researcher recruits whoever is relevant to the line of investigation and
readily accessible. For example, suppose a student researcher has limited time to conduct a
survey detailing students’ opinions of the campus computer labs (e.g., questions regarding
wait times, quality of equipment, skillfulness of attendants, recommendations, etc.). This
researcher has a scheduled break on Tuesdays from 10:00 to 11:00 a.m. The researcher
logs onto the school’s enrollment system and selects courses that meet on Tuesdays from
10:00 to 11:00 a.m., with more than 200 students per class. The researcher then contacts
each instructor, requesting permission to speak for one minute at the end of class to offer
students the opportunity to partake in this computer lab online survey. For classes wherein
the instructor consents, the researcher visits the class and provides a briefing of the project
and instructions for taking the online survey.
In terms of external validity, clearly this survey was not designed to characterize the
overall student population in terms of computer lab experience; instead, it was designed to
provide the computer labs with some valuable ideas as to how they may improve their
services.
Snowball Sampling
The term snowball sampling comes from the (cartoon) notion that if you roll a small
snowball downhill from the top of a mountain, it will pick up more snow as it goes,
ultimately building into a massive snowball. Actually, when it comes to snowball
sampling, it might be more apt to think of the proverb: “Birds of a feather flock together.”
Snowball sampling is useful when participants are scarce or difficult to readily identify.
For example, suppose you are conducting a study of people who use a wheelchair, and by
luck, you find someone in a wheelchair who is willing to participate in your study. After
administering your survey/experiment, you courteously ask this person if he or she knows
of anyone else who uses a wheelchair who might be interested in taking this survey. You
would then follow up with these referrals and progressively ask each following participant
for his or her list of referrals, and so on—hence, the sample snowballs up. Referrals may
lead to direct sources (e.g., their friends, family, colleagues, etc.) or indirect sources (e.g.,
wheelchair repair shop, wheelchair accessory website, rehab center, etc.).
Unlike the wheelchair example, the sample that you are interested in may need to be
drawn from an invisible or hidden population. This does not mean that these individuals
are literally invisible or in hiding; it merely means that the feature(s) that you are
interested in is not readily observable. For instance, such individuals may be invisible in
that upon casual observation, they simply possess no identifying characteristics that would
suggest that they meet your research criteria (e.g., single parent, bisexual, dyslexic,
ballroom dancer, etc.). Alternatively, some individuals deliberately hide the fact that they
would meet the criteria for your research because revealing such information could have
legal consequences (e.g., illegal aliens, involvement in criminal activities, etc.) or be
embarrassing (e.g., peculiar obsessions, fetishes, unpopular belief systems, uncommon
interests, etc.).
In any case, if you are fortunate enough to find one such person and are able to establish
a professional rapport with a genuine nonthreatening and nonjudgmental demeanor,
snowball sampling may lead you to other suitable subjects.
Quota Sampling
Quota sampling is typically used when a prompt response is needed from the population.
In quota sampling, one or several attributes are identified and quotas are set for each. For
example, a researcher is conducting exit polls, asking voters how they just voted on
selected issues. The researcher has specified the following quotas: 30 females and 30
males. Once the quotas are met for a gender, no further data will be collected for that
gender. Suppose after the first hour, the researcher has gathered exit polling data on 10
females and 30 males; the researcher would stop attempting to gather data from males,
even if they actively volunteered it, since that quota has been satisfied. The researcher
would continue efforts to gather data from an additional 20 female voters, at which point
the researcher could (immediately) process the findings.
Purposive Sampling
Purposive sampling is used when potential subjects must possess a (complex) set of
specific characteristics. For example, suppose a researcher is interested in the effects that a
particular drug has on patients undergoing radiation therapy. To be a viable participant in
this study, each individual must meet all of the following criteria: (1) between 18 and 65
years old, (2) diagnosed with cancer, (3) set to begin radiation therapy, (4) undergoing a
course of radiation therapy that consists of three to five treatments, (5) willing to take the
experimental drug (or placebo), and (6) not using any nonprescribed medications.
Purposive sampling may involve one or several criteria for subject selection. In this
case, all six criteria must be met in order for an individual to be a potential subject.
Clearly, it would be virtually impossible to encounter individuals who meet all of these
criteria simply by chance, as would be used in probability sampling. As you have likely
surmised by now, each time an additional criterion is added in purposive sampling, the
potential subject pool shrinks.

SAMPLING BIAS
Sampling bias occurs, perhaps unintentionally, when participants with certain
characteristics are more likely to be selected. Such bias can corrupt the external validity of
the findings. Depending on the methods used to identify potential participants or gather
the sample, sampling bias is a concern. For example, while it may seem reasonable to
administer a survey via the Internet, this method would preclude individuals who are not
computer savvy or do not have access to the Internet.
Recruitment location can also introduce survey bias; imagine the bias your data would
be subject to if you were to post recruitment flyers (only) in a women’s locker room, a
sports bar, the lobby of a technology company, a liquor store in an impoverished
neighborhood, and so forth. Such strategies would clearly be inappropriate unless you are
deliberately seeking to sample and comprehend individuals endemic to those domains.
It may not be possible to completely control for sample bias in every situation, but
awareness of this potential confound can be helpful when considering the credibility of the
research of others and in designing and implementing your own investigations.

Key Concepts
• Rationale for sampling (time, cost, feasibility, extrapolation)
• Population
• Sample frame
• Sample
• Representative sample
• External validity
• Probability sampling
Simple random sampling
Systematic sampling
 Stratified sampling
Area sampling
• Nonprobability sampling
Availability sampling
Snowball sampling
Quota sampling
Purposive sampling
• Sample bias

Practice Exercises
1. Explain the rationale for sampling in terms of
a. Time
b. Cost
c. Feasibility
d. Extrapolation
2. Define the following terms and provide an example for each:
a. Population
b. Sample frame
c. Sample
d. Representative sample
3. A school is considering building a new library, which would involve a 1% tuition
increase and take 2 years to complete. The school has selected you to conduct a
survey of students to gather opinions regarding funding the new library. You will
use simple random sampling.
a. Define the population
b. Define the sample frame
c. Explain how you would select the sample
d. Explain how you would gather the data
4. An Internet provider has commissioned you to conduct a customer satisfaction
survey, providing you with a list of all its subscribers containing their name, phone
number, and e-mail ID. You will use systematic sampling.
a. Define the population
b. Define the sample frame
c. Explain how you would select the sample
 d. Explain how you would gather the data
5. A public library wants to determine the research needs of children and adults who
access the reference section. You will use stratified sampling.
a. Define the population
b. Define the sample frame
c. Explain how you would select the sample
d. Explain how you would gather the data
6. Prior to building a factory in Cityville, Acme Corporation wants to conduct a survey
of the residents. They provide you with a list of the addresses covering the 300
blocks of Cityville. You will use area sampling.
a. Define the population
b. Define the sample frame
c. Explain how you would select the sample
d. Explain how you would gather the data
7. An amusement park wants to assess how much money its patrons intend to spend at
the park today (aside from the price of admission). You will use availability
sampling.
a. Explain how you would select the sample
b. Explain how you would gather the data
8. A learning lab has commissioned you to administer a survey of people with
dyslexia. You will use snowball sampling.
a. Explain how you would select the sample
b. Explain how you would gather the data
9. Acme Bus Company has selected you to conduct a survey of their riders. They want
data on 50 minors (younger than 18 years) and 100 adults (18 years or older). You
will use quota sampling.
a. Explain how you would select the sample
b. Explain how you would gather the data
10. A community tutorial program is recruiting students to participate in free
tutoring. Students must live within the school district, be younger than 18 years,
and be available 3 days a week from 3:30 to 5:00 p.m. You will use purposive
sampling.
a. Explain how you would select the sample
b. Explain how you would gather the data
 CHAPTER 3
Working in SPSS

• Data View
• Variable View
• Codebook
• Saving Data Files

Computers are useless. They can only give you answers.

—Pablo Picasso
LEARNING OBJECTIVES
Upon completing this chapter, you will be able to:
• Operate within the two primary views in SPSS: Variable View and Data View
• Establish or modify variable definitions on the Variable View screen: name, type,
width, decimals, label, values, missing, columns, align, measure, and role
• Use the value label icon to alternate between numeric and label displays
• Interpret and use a codebook to configure variables in SPSS
• Enter data into SPSS
• Save and identify SPSS data files
DIGITAL LEARNING RESOURCES
The tutorial video and data sets for this chapter are at www.sagepub.com/knapp
• SPSS setup video
• SPSS files
OVERVIEW
Based on what you have read thus far, you have probably figured out that when it comes
to statistics, larger sample sizes facilitate more robust statistical findings. Appropriately
large sample sizes are also important when it comes to gathering a representative sample,
which helps when it comes to generalizing the findings from your sample to the overall
population from which it was drawn (external validity).
Processing large samples can involve hundreds or thousands of calculations. For most
statistical formulas, the mathematical complexity does not go beyond simple algebra,
however; such formulas typically involve multiple mathematical operations on each
record. Attempting to process such data by hand would be inefficient in two ways: (1) It
would be very time-consuming, and (2) accuracy would be compromised. Performing a
set of calculations on a lengthy data set is bound to produce some errors along the way.
Even if each mathematical operation was correct, the data would be vulnerable to
cumulative rounding error.
With the advent of affordable, powerful computers and menu-driven statistical
programs, it is now possible to accurately perform a variety of statistical analyses in a
matter of seconds. This chapter will provide you with what you need to know to get
started using SPSS.
SPSS, which originally stood for “Statistical Program for the Social Sciences,” has gone
through some substantial evolution over time; some versions are now referred to as PASW
—“Predictive Analytics Software.” For the remainder of the text, the term SPSS will be
used. Regardless of the name, the SPSS functionality of the statistics covered in this text
has remained relatively stable across the evolution of the software.
TWO VIEWS: VARIABLE VIEW AND DATA VIEW
SPSS is laid out as two main screens: the Variable View, which is used for establishing the
characteristics of each variable, and the Data View, which contains the gathered data. We
will begin with the Variable View.
Variable View
The Variable View provides a screen for you to systematically set up the variables that will
contain your data. This is where you will assign the name and characteristics of each
variable that you will be including in the data set. To access the Variable View screen,
click on the tab at the bottom of the screen that says Variable View, as shown in Figure 3.1.
Basically, for each variable, you are telling SPSS the name of the variable and the kind
of data it will contain (e.g., regular numbers, dates, text, etc.), along with some other
properties (parameters). Once you have established each variable in the Variable View
screen, you can proceed to enter the data that you have gathered on the Data View screen,
which resembles a traditional spreadsheet.
The Variable View screen has 11 properties that you can set for each variable. Some
versions of SPSS may have a different amount of properties; you should be able to
proceed nonetheless. Naturally, you will use care when establishing variables on the
Variable View screen, but there is no need to be nervous; even after you have entered data
on the Data View screen, you can always return to the Variable View screen and make
changes (e.g., include more variables, delete variables, and, to some extent, reconfigure
the properties of existing variables).
Figure 3.1 SPSS Variable View Screen

The cursor is positioned in the Name column for the first variable; this is where the data
definition process begins.
Name
Each variable needs a unique name. The name can contain up to 64 letters and numbers;
the first character must be a letter. Some older versions of SPSS allow only eight
characters for variable names, so you may need to be imaginative when it comes to
assigning briefer variable names. Spaces are not allowed in the variable name, but you can
use the underscore (_) character instead. For your own convenience, try to assign
meaningful names (e.g., age, date_of_birth, first_name, last_name, gender, gpa,
question01, question02, question03, etc.). It is okay if you are unable to assign a perfect
variable name; this will be discussed in more detail when we look at the Label property.
The cursor is positioned in the first cell for the first variable. We will build a database
containing two variables: gender and age. Begin by entering gender in row 1 under Name.
When you press Enter, notice that SPSS automatically enters default values for all of the
remaining properties except for label. Each of these properties can be changed, but we will
accept the automatic defaults for some.
Type
The system needs to know what type of data the variable will contain. The system assigns
the default type as a numeric variable with a width of eight integer digits and two decimal
digits (which you can change), meaning that this variable will accommodate a number
such as 12345678.12 (Figure 3.2).
Figure 3.2 SPSS Variable Type Window
 To access the menu shown in Figure 3.2, click on the Type cell for that variable. The
options for variable type are fairly self-explanatory except for String; a string variable
contains nonmathematical data (e.g., name, note, comment, memo, etc.). A string variable
is useful for data that contain alphanumeric characters (letters, numbers, punctuation, or a
mixture of letters and numbers), such as an ID code, an address, or a name (e.g., APB-373,
852 S. Bedford Street, Dusty Jones, etc.); if your data are not a date or a numeric value,
then select String type. Obviously, it is not possible to perform mathematical or statistical
operations on string variables.
Width
The Width property is the same as the Width parameter on the Variable Type menu (Figure
3.2).
Decimals
The Decimals property is the same as the Decimals parameter on the Variable Type menu
(Figure 3.2).
Label
If the Label property is left blank, SPSS will use the variable name in all output reports;
otherwise, it will use whatever you specify as the label. For example, suppose the name of
the variable is dob, but in your reports, you want it to display as Date of Birth; in that
case, simply enter Date of Birth in the Label property. Notice that the Label can contain
spaces, but the Name cannot.
Values
The Values property provides a powerful instrument for assigning meaningful names to
the values (numbers) contained in categorical variables. For example, gender is a nominal
variable consisting of two categories (1 = female, 2 = male). When it comes to nominal
variables, SPSS handles categories as numbers (1, 2) as opposed to the textual names
(female, male). The Values property allows you to assign the textual name to each
category number, so even though you will code gender using 1s and 2s, the output reports
will exhibit these 1s and 2s as female and male. Here is how it works:
In the Name column, create a variable called gender; accept all the default values,
except change the Decimals property to 0.
Click on the Values cell for gender; this will bring up the Value Labels menu (Figure
3.3).
Assign the values one at a time; begin by entering 1 in Value and female in Label, then
click Add.
Do the same for the second category: Enter 2 in Value and male in Label, then click
Add.
To finalize these designations, click OK.
You will see the utility of this labeling system when you enter data on the Data View
screen and when you run your first reports.
Figure 3.3 SPSS Variable Labels Window

Missing
Sometimes, when the source data are either erroneous or missing, the cell is simply left
blank, in which case, the Missing property can remain blank as well. Other times, the
erroneous or missing data are represented by special codes; a common convention is to
code erroneous data as 888, and missing data are represented as 999—this conveys that a
blank cell is not an oversight. Consider the variable age; if the data contained an
erroneous entry (e.g., “I’m a kid”), or if the entry were left blank, the corresponding 888
or 999 codes would radically throw off the statistical (age) calculations. The Missing
property enables us to specify such codes (888 and 999) that we want SPSS to ignore so
that they will not be processed in the statistical calculations. Here is how it works:
Create a variable with the name age; accept all the default values, except change the
Decimals property to 0.
Click on the Select Discrete missing values and enter 888 and 999 (Figure 3.4).
 If you need to indicate more than three such values, you may opt for the Range plus one
optional discrete missing value function, which would enable you to specify a range of
values (e.g., Low: 888, High: 999—meaning that all values from 888 through 999
inclusive will be omitted from all statistical analysis for that variable). In addition, you
can specify one additional value (e.g., Discrete value: −1).
To finalize these designations, click OK.
The numbers 888 and 999 have been generally adopted as special values since they are
visually easy to recognize in a data set. Also, if these special values are not properly
designated as erroneous/missing values, statistical clues will begin to emerge, such as a
report indicating a mean age of 347 or a maximum height of 999 inches or centimeters.
Such extreme results alert you to check that the missing/erroneous designations have been
properly specified for a variable.
Figure 3.4 SPSS Missing Values Window

Columns
The Columns property allows you to change the column width on the Data View screen;
you can specify how many characters wide you want that column to be.
Align
The Align property lets you specify how you want the variable to be presented on the Data
View screen and in the output reports. Typically, Right alignment is used for numeric data
and Left alignment is used for string data or categorical variables with data labels assigned
to them. Center is also an option.
Measure
The Measure property pertains to the four levels of measures (nominal, ordinal, interval,
and ratio) covered in the level of measure section in Chapter 1. For variables that contain
continuous variables, select Scale. For the categorical variables, which may contain value
labels, select Nominal.
Role
Some versions of SPSS have the Role property; do not panic if the version that you are
using does not include this property—we will not be using it in this text. Role enables you
to define how the variable will be used in the statistical processes. If your version of the
software includes the Role property, just use the default setting: Input.
Use SPSS to set up the Variable View screen to establish the gender and age variables
as shown in Figure 3.5.
Figure 3.5 SPSS Variable View Screen

Data View
Now that the properties for each variable have been established on the Variable View
screen, the next step is to enter the actual data. To switch to the data entry mode, click on
the Data View tab at the bottom of the screen.
As you enter the data in Table 3.1 into the Data View screen, notice that for the gender
variable, you can access the pull-down menu in each cell to select female or male.
Alternatively, you can enter the corresponding numbers that you defined: 1 for female or 2
for male. Notice that SPSS will not allow you to type the words female or male in the
gender field; you will need to enter a number (in this case, 1 or 2) or use the pull-down
menu feature for this variable. The Data View screen should resemble Figure 3.6.
Value Labels Icon
When it comes to viewing your data, there will be times when you will want to see the
value labels (e.g., female, male) and other times when you will want to see the source
numbers (e.g., 1, 2). To toggle this display back and forth, from numbers to text (and
back), click on the Value Labels icon (with the 1 A on it), as shown in Figure 3.7.
Table 3.1 Gender and Age Source Data
 Figure 3.6 Data View Screen With Data Entered

Codebook
For learning purposes, the two-variable data set used in this chapter is admittedly simple.
Even so, without being told that for the variable gender, 1 stands for female and 2 stands
for male, this coding scheme would lead to confusion. Designations such as 1 = female
and 2 = male and other characteristics of each variable in a data set are traditionally
contained in the codebook, which is the companion to the data set. The codebook is
written by the person who develops the experiment or survey; it provides a list and a
description of each variable contained in a data set. This is particularly valuable in data
sets that contain numerous variables with arcane names. For example, suppose you came
across a variable named q105 and it contained dates. Without the codebook, we would
have no idea what any of this means; we would not know how this variable was gathered
or be able to assign any meaning to these dates (e.g., birth date, graduation date,
anniversaries, date of arrest, date admitted to a hospital, etc.); if you do not know the story
of a variable, that variable is virtually useless. Although there is no standard form for
codebooks, a quality codebook should indicate the information essential to understanding
each variable in the data set. Continuing with the q105 example, a reasonable codebook
entry for this variable might look like this:
Figure 3.7 The Value Labels Icon, Alternating the Display of Categorical
Variables From Text to Numeric Display (and Back)

Variable: q105 (Question 105)

Question: When did you graduate high school?
Type: Date
Format: MM/DD/YY
For our simple two-variable database detailed in Table 3.1, the codebook would look
like this:
Variable: gender
Question: What is your gender?
Type: Numeric (1 = Female, 2 = Male)
Format: 1 integer, 0 decimal
Error: 888
Missing: 999

Variable: age
Question: How old are you?
Type: Numeric
Format: 3 integer, 0 decimal
Error: 888
Missing: 999

Saving Data Files

The 50-record database that you entered earlier will be used in Chapter 4; to save the file,
click on the Save this document icon as shown in Figure 3.8. Use the file name Chapter 03
− Example 01 – Descriptives; SPSS automatically appends the .sav extension (suffix) to
the filename. The file on your system will be listed as Chapter 03 – Example 01 –
Descriptives.sav.
 Figure 3.8 The Save this document Icon

Key Concepts
Variable View
Name
Type
Width
Decimals
Label
Values
Missing
Columns
Align
Measure
Role
Data View
Value Labels icon
Codebook
Saving data files
Save this document icon
.sav files

Practice Exercises
Use the provided codebook in each exercise to establish the variables on the Variable
View screen, and then enter the data on the Data View screen.
To check your work, produce a variable list; click on Analyze, Reports, Codebook, as
shown in Figure 3.9.
Figure 3.9 Ordering a List of All Variables; Click on Analyze, Reports, Codebook

Next, select all the variables that you want to include in the codebook report; move the
variables from the left Variables window to the right Codebook Variables window (using
double-click, drag & drop, or the arrow button), then click OK, as shown in Figure 3.10.
Figure 3.10 Codebook Report Order Screen; Move Variables of Interest to Right
(Codebook Variables) Window

This will generate a Variable Information report showing the properties of all variables,
as shown in Table 3.2.
Table 3.2 Codebook Report Displaying the Variable Properties
 You can compare your Variable Information report with those in the solution section to
check for accuracy.
After each exercise, clear out the data; click on File, New, Data, as shown in Figure
3.11.
Figure 3.11 Clearing the Data, Click on File, New, Data

1. Codebook
Variable: enrolled
Question: Are you currently enrolled in school?
Type: Numeric (1 = Yes, 2 = No)
Format: 1 integer, 0 decimal
Variable: units
 Question: How many units are you taking?
Type: Numeric
Format: 2 integer, 0 decimal
Error: 888
Missing: 999

Variable: gpa
Question: What is your overall GPA?
Type: Numeric
Format: 1 integer, 2 decimal
Error: 888
Missing: 999
Data:
NOTE: Do not enter the data in the Rec. # column; this column corresponds to the
leftmost column in SPSS, indicating the record (row) number.

2. Codebook
Variable: id
Question: What is your ID number?
Type: String
Characters: 12
Variable: volunteer_hours
Question: How many hours per week are you willing to volunteer?
Type: Numeric
Format: 2 integer, 2 decimal
Data:
3. Codebook
Variable: degree
Question: What is the highest degree you’ve completed?
Type: Numeric (1 = Associate’s, 2 = Bachelor’s, 3 = Master’s, 4 = Doctorate)
Format: 1 integer, 0 decimal

Variable: pretest
Question: What was your pretest score?
Type: Numeric
Format: 3 integer, 0 decimal
Variable: posttest
Question: What was your posttest score?
Type: Numeric
Format: 3 integer, 0 decimal
Data:

4. Codebook
Variable: employment_status
Question: What is your current employment status?
Type: Numeric (1 = Unemployed, 2 = Temporary, 3 = Part-time, 4 = Full-
time)
Format: 1 integer, 0 decimal
 Variable: work_hours
Question: In an average week, how many hours do you work?
Type: Numeric
Format: 2 integer, 0 decimal

Variable: sleep_hours
Question: In an average day, how many hours do you sleep?
Type: Numeric
Format: 2 integer, 2 decimal
Error: 888
Missing: 999
Data:

5. Codebook
Variable: first_initial
Question: What is the first initial of your first name?
Type: String
Characters: 1

Variable: last_name
Question: What is your last name?
Type: String
Characters: 25

Variable: siblings
Question: How many siblings do you have?
Type: Numeric
Format: 2 integer, 0 decimal
Error: 888
Missing: 999
 Variable: adopted
Question: Are you adopted?
Type: Numeric (1 = Yes, 2 = No)
Format: 2 integer, 0 decimal
Error: 888
Missing: 999
Data:

6. Codebook
Variable: patient_id
Question: What is your patient ID number?
Type: Numeric
Format: 10 integer, 0 decimal

Variable: age
Question: How old are you?
Type: Numeric
Format: 2 integer, 0 decimal
Error: 888
Missing: 999

Variable: temperature
Question: What is your temperature (in Celsius)?
Type: Numeric
Format: 2 integer, 1 decimal
Error: 888
Missing: 999

Variable: flu_shot
 Question: Have you had a flu shot this season?
Type: Numeric (1 = Yes; 2 = No, and I don’t want one; 3 = Not yet, but I’d
like one)
Format: 1 integer, 0 decimal
Error: 888
Missing: 999

Variable: rx
Question: What medications are you currently taking?
Type: String
Characters: 100
Data:

7. Codebook
Variable: passport
Question: Do you have a valid passport?
Type: Numeric (1 = Yes, 2 = No, 3 = Decline to answer)
Format: 1 integer, 0 decimal

Variable: fired
Question: Have you ever been fired from a job?
Type: Numeric (1 = Yes, 2 = No, 3 = Decline to answer)
Format: 1 integer, 0 decimal

Variable: er
Question: Have you ever been treated in an emergency room?
Type: Numeric (1 = Yes, 2 = No, 3 = Decline to answer)
Format: 1 integer, 0 decimal

Variable: dob
Question: What is your birth date?
 Type: Date
Format: MM/DD/YYYY
Data:

8. Codebook
Variable: dogs
Question: I like dogs.
Type: Numeric (1 = Strongly disagree, 2 = Disagree, 3 = Neutral, 4 = Agree,
5 = Strongly agree)
Format: 1 integer, 0 decimal

Variable: cats
Question: I like cats.
Type: Numeric (1 = Strongly disagree, 2 = Disagree, 3 = Neutral, 4 = Agree,
5 = Strongly agree)
Format: 1 integer, 0 decimal

Variable: pets
Question: How many pets do you currently have?
Type: Numeric
Format: 2 integer, 0 decimal
Data:

9. Codebook
 Variable: blood_type
Question: What is your blood type (respond “?” if you don’t know)?
Type: Numeric (1 = A−, 2 = A+, 3 = B−, 4 = B+, 5 = AB−, 6 = AB+, 7 = O−,
8 = O+, 9 = don’t know)
Format: 1 integer, 0 decimal

Variable: gender
Question: What is your gender?
Type: Numeric (1 = Female, 2 = Male)
Format: 1 integer, 0 decimal

Variable: prior_donor
Question: Have you ever donated blood before?
Type: Numeric (1 = Yes, 2 = No)
Format: 1 integer, 0 decimal
Data:

10. Codebook
Variable: entree_food
Question: What entrée did you order?
Type: Numeric (1 = Fish, 2 = Chicken, 3 = Beef, 4 = Vegetarian)

Format: 1 integer, 0 decimal

Variable: entree_quality
Question: On a scale of 1 to 10 (10 being the best), how would you rate your
meal?
Type: Numeric
Format: 2 integer, 0 decimal

Variable: dessert
Question: On a scale of 1 to 10, how would you rate your dessert (if you did not
order a dessert, please mark “X”)?
Type: Numeric
Format: 2 integer, 0 decimal
Missing: 999
HINT: Code “X” as 999
Data:
 PART II
Statistical Processes
 CHAPTER 4
Descriptive Statistics

Whenever you can, count.

—Sir Francis Galton

LEARNING OBJECTIVES
Upon completing this chapter, you will be able to:
• Comprehend the meaning of each descriptive statistic: number (n), mean, median,
mode, standard deviation, variance, minimum, maximum, and range
• Load an SPSS data file
• Order and interpret descriptive statistics for continuous variables: frequency
statistics tables, histogram with normal curve, and skewed distribution
• Order and interpret descriptive statistics for categorical variables: frequency
statistics table and bar chart
• Select records to process/rule out
• Select all records to process
DIGITAL LEARNING RESOURCES
The tutorial video and data sets for this chapter are at: www.sagepub.com/knapp
• Descriptive statistics—continuous video
• Descriptive statistics—categorical video
• SPSS files
OVERVIEW
Statistics is about making sense of data, which could consist of one or many groups and
involve data sets of virtually any size. Take a look at the data set from Chapter 3 (Table
4.1); if someone were to ask you to describe this data set, you might say, “It looks like a
list of about half women and half men, mostly in their 20s,” which would not be wrong,
but we can be more precise than that. This list consists of (only) 50 records and 100 data
items, but how would you make a viable summary statement if the list consisted of 100,
1,000, or even 100,000 records? Such data would span multiple pages and defy cursory
visual inspection.
To better comprehend the nature of a data set, we use descriptive statistics, sometimes
referred to as summary statistics. Descriptive statistics enable us to concisely understand a
data set of any size using a handful of figures and simple graphs that serve to concisely
describe the contents of a variable.
Table 4.1 Data Set Containing 50 Records With Two Variables per Record:
Gender and Age

Continuous variables can be summarized using the nine descriptive statistics: number
(n), mean, median, mode, maximum, minimum, range, standard deviation, and variance.
Graphically, continuous variables can be depicted using a histogram (a special kind of bar
chart) with a normal curve.
Categorical variables can be summarized using number (n) and percent. Graphically,
categorical variables are depicted using a simple bar chart. You can see examples of a
histogram with a normal curve for a continuous variable and a bar chart for a categorical
variable on the first page of this chapter (next to the dog).
 We will begin with an explanation of the nine summary statistics used to analyze
continuous variables. For simplicity, we will work with a small data set—the first 10 ages
drawn from the second column of Table 4.1: 24, 25, 31, 19, 27, 20, 28, 23, 26, and 24.
DESCRIPTIVE STATISTICS
Number (n)
The most basic descriptive statistic is the number, represented as the letter n. To compute
the n, simply count the number of elements (numbers) in the sample; in this case, there are
10 elements—count them: 24 is the first, 25 is the second, 31 is the third… 24 (at the end)
is the tenth, so n = 10.
The lowercase n is the number of elements in a sample, whereas the uppercase N is the
number of elements in the (whole) population. Either way, n or N is an element count;
SPSS output reports always use the capital N. Since it is rare to be processing a data set
consisting of an entire population, it is considered good practice to use the lowercase n in
your documentation, as such: n(age) = 10.
Mean
In statistical language, the average is referred to as the mean. The calculation for the mean
is the same as the average: Add up all the elements and then divide that amount by the
total number of elements (n = 10):
Mean(age) = (24 + 25 + 31 + 19 + 27 + 20 + 28 + 23 + 26 + 24) ÷ 10
Mean(age) = 247 ÷ 10
Mean(age) = 24.7
The mean can be written using the lowercase Greek letter μ (pronounced m-you) or as
the variable with a horizontal bar over it—hence, the mean may be documented as such:
μ(age) = 24.7
= 24.7
For consistency throughout the rest of this text, the mean will be documented using the
μ(age) style.
Median
The median is the middle value of a variable. Think of the term median in terms of a street
—the median is in the middle; it splits the street in half. To find the median, arrange the
data in the variable from lowest to highest and then select the middle value(s).
When the n is even, as in the data set below (n = 10), there are two middle numbers: 24
and 25. The median is the mean of these two middle numbers:
 When the n is odd, as in this small data set below (n = 5), there is (only) one middle
number—hence, the median is simply the (one) middle number: 86.

Mode
The mode is the most common number in the data set. Notice that mode and most share
the first two letters. In this case, we see that each number in this data set is present only
once, except for 24, which occurs twice, and hence the mode is 24.

It is possible for a data set to have more than one mode; the example below has two
modes: 24 and 31 since both have the most (there are two 24s and two 31s; all the other
numbers appear just once). Such a variable would be referred to as bimodal—meaning
two modes.

Although it is relatively rare, a variable may have more than two modes, which would
be referred to as multimodal.
When SPSS detects more than one mode within a variable, it only reports the lowest
one and provides a footnote indicating that there is more than one mode.

The mean, median, and mode are referred to as measures of central tendency, as
together they suggest the center point of the variable.
Standard Deviation
The standard deviation describes the dispersion of the data within a variable. The core of
the standard deviation formula involves calculating the differences between each
individual number in the data set and the mean.
If the data set consists of relatively middle-of-the-road numbers, which are only slightly
lower or slightly higher than the mean, then this will result in a relatively low standard
deviation, indicating that overall, the numbers do not deviate very far from the mean.
Conversely, if the values in a data set are more diverse, meaning that the values deviate
substantially further from the mean (lower lows and higher highs), this will increase the
standard deviation.
We will begin with the 10-item sample of ages:
Data Set A: 19, 20, 23, 24, 24, 25, 26, 27, 28, 31
We know that the mean is 24.7, and we can see that the numbers in this data set do not
stray too far from that mean. The standard deviation for this variable is 3.592; this would
be written as SD(age) = 3.592. Now we will increase the diversity of these numbers; we
will subtract 15 from the lowest number and add 15 to the highest number, thus forcing
these numbers to deviate further from the mean, making for greater dispersion within the
variable:
Data Set B: 4, 20, 23, 24, 24, 25, 26, 27, 28, 46
After making this adjustment, the standard deviation for Data Set B is 10.144, which is
nearly three times as large as the standard deviation for Data Set A (3.592). What is going
on? In Data Set A, the end values (19 and 31) were fairly close to the mean (24.7); there
was only a 5.7 difference between 19 and the mean (24.7) and a 6.3 difference between 31
and the mean (24.7), but in Data Set B, we increased those differences by 15 (each); these
numbers now deviate further from the mean, which is reflected in the higher standard
deviation.
Another way to understand the standard deviation is to inspect the formula; we will
discuss a selected portion of the standard deviation formula, but we will not process it
manually.
The formula for the sample standard deviation is

but for now focus only on the (x − μx) term in the numerator. This expression says that
for each value in the variable x (in this case, age), calculate the difference between that age
value (x) and the mean age (μx). Very low and very high ages would deviate substantially
from the mean, producing a larger difference; these differences accumulate in the
numerator, which increases the value of the fraction, thus resulting in a higher standard
deviation—there is more deviation from the mean. Graphically, this would be depicted as
a wider, shorter bell curve on the histogram. Conversely, if the ages in this data set are all
fairly close together, then computing the difference between each age and the mean age
would produce relatively small(er) numbers accumulating in the numerator; the smaller
the numerator, the smaller the value of the fraction. Such a data set would produce a
smaller standard deviation—there is less deviation from the mean. Graphically, we would
see a thinner, taller normal curve on the histogram.
Variance
The variance is simply the standard deviation squared. For example, we will compute the
variance of age from the Data Set A:
Variance(age) = (Standard Deviation(age))2
Variance(age) = (3.592)2
Variance(age) = 3.592 × 3.592
Variance(age) = 12.902
The variance is seldom included in statistical reports; it is primarily used as a term
within other statistical formulas.
Minimum
The minimum is the smallest number in a variable. In the data set below, the minimum is
19.

Maximum
The maximum is the largest number in a variable. In the data set below, the maximum is
31.

Identifying the minimum and maximum values has some utility, but try not to bring
inappropriate suppositions to your interpretation of such figures—bigger is not necessarily
better. The meaning of the minimum and maximum values depends on the nature of the
variable. For example, high bowling scores are considered good, but low golf scores are
considered good, and high (or low) phone numbers or street addresses are considered
neither good nor bad.
Range
The range is the span of the data set; the formula for the range is maximum − minimum.
In the data set below, we would calculate: 31 − 19 = 12; the range is 12 (years).

SPSS—LOADING AN SPSS DATA FILE

For clarity, the examples used thus far have involved only 10 data items. Now it is time to
use SPSS to process descriptive statistics using the entire data set consisting of 50 records
and both variables (gender and age).
Run SPSS.
Use the Open Data Document icon (Figure 4.1) to load the file that you created (and
saved) in Chapter 3: Chapter 03 − Example 01 − Descriptives.sav.
Figure 4.1 Open Data Document Icon
SPSS—Descriptive Statistics: Continuous Variables (Age)
There are two types of variables in this SPSS file: Age is a continuous variable, and
gender is a categorical variable. In this section, we will process descriptive statistics for
the continuous variable (age); later in this chapter, we will process the categorical variable
(gender).

1. When the data have loaded, click on Analyze, Descriptive Statistics,

Frequencies… (Figure 4.2).
Figure 4.2 Running a Descriptive Statistics Report: Click on Analyze, Descriptive
Statistics, Frequencies…

2. SPSS will then prompt you to select the variable(s) that you would like to process.
Move the age variable to the Variable(s) window (Figure 4.3).
3. Click on the Statistics… button.
Figure 4.3 Move the Variable(s) to Be Analyzed (Age) From the Left Window to
the Right Variable(s) Window
Figure 4.4 Frequencies: Statistics Order Screen

4. Select the descriptive statistics indicated by the checkboxes (Figure 4.4): Mean,
Median, Mode, Std. deviation, Variance, Range, Minimum, and Maximum.
5. Click on the Continue button. This will take you back to the Frequencies screen
(Figure 4.3).
Figure 4.5 Frequencies: Charts Order Screen; Select Histograms and Show normal
curve on histogram
 6. On the Frequencies screen (Figure 4.3), click on the Charts… button.
7. Select Histograms and Show normal curve on histogram (Figure 4.5).
8. Click on the Continue button. This will take you back to the Frequencies screen
(Figure 4.3).
9. Click on the OK button on the Frequencies screen; this tells SPSS to process the
Frequencies report based on the parameters that you just specified. SPSS should
produce this report in under a minute (try to be patient).
Frequency Statistics Tables
The first part of the report is the Frequency Statistics table (Table 4.2), which shows the
summary statistical results as discussed earlier.
Table 4.2 Frequency Statistics Table Showing Summary Statistics for Age

The second part of the report shows the frequency of each value in the age variable
(Table 4.3). Focus on columns 1 and 2 of this table, which show that the numbers 18, 19,
20, and 21 each occur twice in the data set; 22 and 23 each occur six times; 24 occurs
eight times; 25 occurs nine times; and so on.
 Table 4.3 Frequency Statistics Table Showing the Frequency of Each Value in the
Age Variable

Histogram With Normal Curve

The third part of this report is the histogram of the age variable. The histogram is simply a
graphical representation of the frequency statistics. Basically, Figure 4.6 is a picture of the
data in Table 4.3. Notice that the first four bars are each two units tall; this is because the
first four numbers in the table (18, 19, 20, and 21) each occur two times in the data set.
Notice that the tallest bar is nine units tall; this is because the number 25 occurs nine times
in the data set.
In addition to understanding the nature of a variable via the descriptive statistics (mean,
median, mode, etc.), the histogram provides further insight into the characteristics of a
continuous variable—a picture is indeed worth a thousand words.
In addition to the bars, which constitute the histogram, it is also traditional to include a
normal curve—sometimes referred to as a “bell curve” because of its shape. The normal
curve is derived from the same source data as the bar chart; you can think of this normal
curve as the smoothed-out version of the bar chart. More often than not, we see this sort of
symmetrical distribution among continuous variables—most of the values are gathered
toward the middle, with the frequencies progressively dropping off as the values depart
(negatively or positively) from the mean.
Figure 4.6 Histogram of the Age Variable
 For example, if we were to measure the height of 100 randomly selected people, we
would expect to find that most people are moderate height (which would constitute the
tallness in the middle of the normal curve). We would also expect to find a few
exceptionally short people and about the same amount of exceptionally tall people, which
would account for the tapering off seen on the left and right tails of the normal curve. This
phenomenon of the bell-shaped distribution is so common that it is referred to as a normal
distribution, as represented by the normal curve. When assessing a histogram for
normality, focus primarily on inspecting the normal curve for symmetry; if we were to
slice the normal curve vertically down the middle, the left half of the normal curve should
resemble a mirror image of the right half. When inspecting a histogram with a normal
curve, we expect to see some variations/asymmetry among the heights of the bars.
Skewed Distribution
As with any rule, there are of course, exceptions; not all histograms produce normally
shaped curves. Depending on the distribution of the data within the variable, the histogram
may be skewed, meaning that the distribution is tilted, as shown in Figures 4.7 and 4.8.
Figure 4.7 Negative (Left) Skew
 Figure 4.8 Positive (Right) Skew

In Figure 4.7, we see that most of the data are on the right, between about 150 and 300,
but there is a small scattering of lower values (under 100), forcing the left tail of the curve
to be extended out. 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 4.7 has outliers to the left—hence, it is
skewed left, or negatively skewed.
Outliers are not always negative. Figure 4.8, which is a virtual mirror image of Figure
4.7, shows outliers scattered to the right; this distribution would be referred to as being
skewed right, or positively skewed. The notion of normality of the data distribution will be
discussed further in Chapters 5, 6, 7, and 8.
SPSS—DESCRIPTIVE STATISTICS: CATEGORICAL VARIABLES
(GENDER)
Descriptive statistics for categorical variables are derived using the same ordering screens
as for continuous variables, except you will be specifying different options.
1. Click on Analyze, Descriptive Statistics, Frequencies . . . (Figure 4.9).
Figure 4.9 Running a Descriptive Statistics Report: Click on Analyze, Descriptive
Statistics, Frequencies…

2. First, click on the Reset button; this will clear the parameters on the submenus that
you specified when running the summary statistics for age.
3. Next, move the gender variable from the left window to the right Variable(s)
window (Figure 4.10).
4. Click on the Charts… button.
Figure 4.10 Click on the Reset Button to Clear Prior Options, Then Move the
Variable(s) to Be Analyzed (Gender) From the Left Window to the
Right Variable(s) Window

5. On the Frequencies: Charts menu, there are two viable options: Bar charts or Pie
charts (Figure 4.11). In statistics, bar charts are used more often than pie charts. You
 can also choose to represent the numbers as frequencies (the actual counts) or
percentages. For this example, select Frequencies.
NOTE: After running this analysis as specified, feel free to return to this menu to rerun
this analysis using different settings (e.g., try selecting Pie chart, Percentages).
Figure 4.11 Frequencies: Charts Order Screen; Select Bar charts and Frequencies

6. Click on the Continue button; this will return you to the Frequencies menu.
7. Click on the OK button on the Frequencies screen; this tells SPSS to process the
Frequencies report based on the parameters that you just specified.
Frequency Statistics Tables
The first part of this frequency report shows the overall n (N)—the total number of entries
(records) in the variable: 50 valid records and 0 missing as shown (Table 4.4).
Table 4.4 Descriptive Statistics for Gender: n, Valid and Missing

The next part of the report provides more detailed information regarding the n,
indicating the frequency (actual number) and percent for each category within the gender
variable (female/male), as shown in Table 4.5. Incidentally, to calculate the percent, divide
the frequency for the category by the (valid) n, so for female, it would be 27 ÷ 50 = .54
(54%).
Table 4.5 Descriptive Statistics for Gender: Frequency and Percent
Bar Chart
Last, the report provides a bar chart representing the two gender categories (female/male)
(Figure 4.12).
Figure 4.12 Bar Chart of the Gender Variable

SPSS—DESCRIPTIVE STATISTICS: CONTINUOUS VARIABLE

(AGE) SELECT BY CATEGORICAL VARIABLE (GENDER)—
FEMALES ONLY
So far, we have processed the continuous variable, age, with both genders combined, but it
is also possible to produce separate reports for females only and males only, showing the
summary statistics and histograms for each. This technique not only satisfies curiosity
about what is going on in each category but also will be essential for running the pretest
checklist reports that will be covered in Chapters 5, 6, 7, and 8.
We will begin with processing the age summary statistics for females only, and then we
will repeat the process but select data for males only. The Select Case option allows you to
specify which cases (rows of data, also known as records) you would like to process;
SPSS will temporarily ignore all other cases—they will not be included in any statistical
computations until you choose to reselect them.
The following procedure will select only the cases where gender = 1 (female).
1. Click on the Select Cases icon (Figure 4.13).
Figure 4.13 The Select Cases Icon

2. This will bring you to the Select Cases menu (Figure 4.14). The default
Selection is All cases. Click on If condition is satisfied, then click on the
If… button.
Figure 4.14 The Select Cases Screen (Top Only)

3. This will bring you to the Select Cases: If screen (Figure 4.15).
Figure 4.15 The Select Cases: If Screen (Top Only)

4. Remember: SPSS handles categorical variables as numbers; earlier, we had

established that for the categorical variable gender, 1 = female and 2 = male.
Since we want statistical reports on females only, we enter the inclusion criteria,
gender = 1, in the big box at the top of the screen. Then click on the Continue
button.
5. This will return you to the Select Cases screen. Click on the OK button, and the
 system will process your selection criteria.
Figure 4.16 The Data View Screen After Select Cases (Gender = 1) Has Been
Executed (Section Only)

6. Switch back to the Data View screen. First, notice that the record (row) numbers
for each male is slashed out (Figure 4.16). You can think of all the data in the
slashed-out rows as being in a sort of penalty box—they are still part of the data
set, but they cannot play. Slashed-out data will not be included in any statistical
processing.
Also notice that SPSS has created the temporary variable Filter_$ in the last
column, which corresponds to the slashes in each row. If you click on the Value
Labels icon or go to the Variable View screen, you will see that the Filter_$
variable contains two categories: 0 = not selected and 1 = selected.
Since we selected only cases where gender = 1, this means that if we were to run
the descriptive statistics report again, the summary statistics and histogram
would reflect females only, as opposed to the earlier report which combined
females and males.
7. At this time, rerun the analysis for the age variable. The resulting statistical
report should resemble the data shown in Table 4.6.
Table 4.6 Frequency Statistics Table Showing Summary Statistics for Age for
Females Only
 8. Notice that the N has changed from 50, which included both females and males,
to 27, which is females only. Compared with the first report, all other statistics
have changed as well. Continuing our analysis of the females only, observe the
frequency statistics (Table 4.7) and corresponding histogram (Figure 4.17).
Table 4.7 Frequency Statistics Table Showing the Frequency of Each Value in the
Age Variable for Females Only

Figure 4.17 Histogram of the Age Variable for Females Only

SPSS—DESCRIPTIVE STATISTICS: CONTINUOUS VARIABLE
(AGE) SELECT BY CATEGORICAL VARIABLE (GENDER)—MALES
ONLY
9. As you can see, there is a lot to be learned by selecting the data and
examining statistics pertaining to females only. The next step is to run the same
reports for males only.
10. Repeat the steps to order the descriptive statistics report, except when you get to
Step 3, instead of specifying gender = 1, change that to gender = 2 (remember, we
established gender as 1 = female and 2 = male).
11. Upon (re)running the data for males, notice that the slashes, Filter_$, and output
reports all change to reflect males only (see Table 4.8, Table 4.9, and Figure 4.18).
Table 4.8 Frequency Statistics Table Showing Summary Statistics for Age for
Males Only
 Table 4.9 Frequency Statistics Table Showing the Frequency of Each Value in the
Age Variable for Males Only

Figure 4.18 Histogram of the Age Variable for Males Only

SPSS—SELECTING ALL VARIABLES

At this point, we have run three sets of descriptive statistics on age: (1) all records, (2)
females only, and (3) males only. Now, suppose we want to perform further analyses using
the entire data set again. There are several ways to do this:
On the Data View screen, click on the column header Filter_$ (which will highlight the
whole column) and press the Delete key.
On the Variable View screen, click on the corresponding row number—in this case, row
3 (which will highlight the whole row)—and press the Delete key.
Click on the Select Cases icon, then click on the All Cases button. Finally, click on the
OK button.
GOOD COMMON SENSE
The acronym GIGO (pronounced gig-oh) comes from the early days of computing; it
stands for garbage in, garbage out, and it is just as valid today. Basically, it means that if
you input junk data into a program, the output will be junk too.
Although SPSS is proficient at processing statistical data, keep in mind that the program
has no real intelligence per se—it just pushes the numbers through the functions using the
parameters that you specify. As such, it is up to you to make intelligent selections.
In the examples detailed in this chapter, we ordered a variety of statistical analyses on
the age variable, which makes sense; knowing the mean age can be useful. Larger
databases are likely to contain other numeric variables such as ID numbers, street
addresses, phone numbers, serial numbers, license numbers, and so on. Technically, you
could order SPSS to compute descriptive statistics on such variables; SPSS is not bright
enough to know that computing the mean for a series of phone numbers makes no sense;
SPSS (or any other statistical processing software) would mindlessly process the variable
and provide you with an average phone number, which would be useless. In summary, it is
up to you to proceed mindfully when using any such software.

Key Concepts
• Descriptive statistics
Number—N(variable) for population, n(variable) for sample
Mean—μ(variable) or
Median
Mode
Standard deviation—SD(variable)
Variance
Minimum
Maximum
Range
• Loading SPSS data files
• Histogram
• Normal curve
• Skew
Negative (left) skew
Positive (right) skew
Outlier
• Bar chart
 • Pie chart
• Select cases
• Good common sense

Practice Exercises
For Exercises 1 to 5:
• Configure SPSS per the specifications of the codebook.
• Enter the data set as shown in the table.
• Order descriptive statistics as indicated for each exercise.
For Exercises 6 to 10:
• Use the prepared SPSS data sets (download from www.sagepub.com/knapp).
• Load, process, and document your findings for each exercise.
1. A survey was conducted in Professor Lamm’s class and Professor Milner’s
class. The question that students responded to is, “How many siblings do you
have?”
Codebook—set up SPSS to contain two variables:
• class is a categorical variable (1 = Prof. Lamm, 2 = Prof. Milner)
• siblings is a continuous variable (number of siblings)
Data set—enter the data set:
 a. Run descriptive statistics and a histogram with a normal curve for siblings
for the whole data set.
b. Run descriptive statistics and a bar chart for class for the whole data set.
c. Run descriptive statistics and a histogram with a normal curve for siblings
for members of Prof. Lamm’s class only.
d. Run descriptive statistics and a histogram with a normal curve for siblings
for members of Prof. Milner’s class only.
2. While waiting in line to donate blood, donors were asked, “How many times
have you donated before?” The researcher recorded their gender and number of
prior donations.
Codebook—set up SPSS to contain two variables:
• gender is a categorical variable (1 = females, 2 = males).
• donated is a continuous variable (total number of blood donations given
before today).
 Data set—enter the data set:

a. Run descriptive statistics and a histogram with a normal curve for donated
for the whole data set.
b. Run descriptive statistics and a bar chart for gender for the whole data set.
c. Run descriptive statistics and a histogram with a normal curve for donated
for females only.
d. Run descriptive statistics and a histogram with a normal curve for donated
for males only.
3. You want to know if typing proficiency is associated with better spelling skills.
You administer a spelling test consisting of 20 words to the students in a
classroom. At the bottom of the sheet, there is a question: Can you type
accurately without looking at the keyboard?
Codebook—set up SPSS to contain two variables:
• looker (1 = looks at keyboard, 2 = doesn’t look at keyboard)
• spelling (score on spelling test)
Data set—enter the data set:
 a. Run descriptive statistics and a histogram with a normal curve for spelling
for the whole data set.
b. Run descriptive statistics and a bar chart for looker for the whole data set.
c. Run descriptive statistics and a histogram with a normal curve for spelling
for “looks at keyboard” only.
d. Run descriptive statistics and a histogram with a normal curve for spelling
for “doesn’t look at keyboard” only.
4. You are interested in the length of time it takes for individuals to complete their
transactions at an ATM. You use a stopwatch to record your unobtrusive
observations and gather two pieces of information on each person: gender and
the length of his or her ATM session (in seconds).
Codebook—set up SPSS to contain two variables:
• gender is a categorical variable (1 = female, 2 = male).
• atmsec is a continuous variable (number of seconds spent at ATM).
 Data set—enter the data set:

a. Run descriptive statistics and a histogram with a normal curve for atmsec
for the whole data set.
b. Run descriptive statistics and a bar chart for gender for the whole data set.
c. Run descriptive statistics and a histogram with a normal curve for atmsec
for female only.
d. Run descriptive statistics and a histogram with a normal curve for atmsec
for male only.
5. You are interested in finding out how many units students are enrolled in. You
conduct a survey of 40 students and record two pieces of information: the degree
(level) the student is working on (bachelor’s, master’s, doctorate) and total
number of units they are taking this term.
Codebook—set up SPSS to contain two variables:
• degree is a categorical variable (1 = bachelor’s, 2 = master’s, 3 =
doctorate).
 • units is a continuous variable (current number of enrolled units).
Data set—enter the data set:
 a. Run descriptive statistics and a histogram with a normal curve for units
for the whole data set.
b. Run descriptive statistics and a bar chart for degree for the whole data set.
c. Run descriptive statistics and a histogram with a normal curve for units
for Bachelor’s degree only.
d. Run descriptive statistics and a histogram with a normal curve for units
for Master’s degree only.
e. Run descriptive statistics and a histogram with a normal curve for units
for doctorate only.
6. You stand at a register in a hospital cafeteria; for each patron, you gather two
pieces of information: professional role (nurse, doctor, other), as indicated on
their badge, and the amount of their bill (as shown on the register).
Codebook—the SPSS file contains two variables:
• profrole (professional role) is a categorical variable (1 = nurse, 2 = doctor,
3 = other).
• bill is a continuous variable (total as shown on the register).
Data set—load Chapter 04 − Exercise 06.sav.
a. Run descriptive statistics and a histogram with a normal curve for bill for
the whole data set.
b. Run descriptive statistics and a bar chart for profrole for the whole data
set.
c. Run descriptive statistics and a histogram with a normal curve for bill for
nurse only.
d. Run descriptive statistics and a histogram with a normal curve for bill for
doctor only.
e. Run descriptive statistics and a histogram with a normal curve for bill for
other only.
7. You recruit a group of people who agree to report their total email counts (sent +
 received) for 30 days. Each participant also completed a survey regarding his or
her employment status (full-time, part-time, unemployed).
Codebook—the SPSS file contains two variables:
• employ is a categorical variable (1 = full-time, 2 = part-time, 3 =
unemployed).
• emails is a continuous variable (total number of emails).
Data set—load Chapter 04 – Exercise 07.sav.
a. Run descriptive statistics and a histogram with a normal curve for emails
for the whole data set.
b. Run descriptive statistics and a bar chart for employ for the whole data
set.
c. Run descriptive statistics and a histogram with a normal curve for emails
for full-time only.
d. Run descriptive statistics and a histogram with a normal curve for emails
for part-time only.
e. Run descriptive statistics and a histogram with a normal curve for emails
for unemployed only.
8. The members of an exercise walking group agree to partake in your study; you
randomly give half of the group walking music in a major key, and the others
are given walking music in a minor key. Each participant can walk as often and
for as long as he or she likes. The participants will record and submit the total
number of minutes that they walked in a week.
Codebook—the SPSS file contains two variables:
• musickey is a categorical variable (1 = major, 2 = minor).
• minwalk is a continuous variable (total number of minutes walked).
Data set—load Chapter 04 – Exercise 08.sav.
a. Run descriptive statistics and a histogram with a normal curve for
minwalk for the whole data set.
b. Run descriptive statistics and a bar chart for musickey for the whole data
set.
c. Run descriptive statistics and a histogram with a normal curve for
minwalk for major only.
d. Run descriptive statistics and a histogram with a normal curve for
minwalk for minor only.
9. The administrator of a two-ward hospital randomly selects one ward wherein the
nurses will be assigned to tend to two patients each; nurses in the other ward
will tend to four patients each. Over the course of a month, upon discharge, each
patient will complete a nursing care satisfaction survey, which renders a score
 ranging from 1 to 100 (1 = very unsatisfied… 100 = very satisfied).
Codebook—the SPSS file contains two variables:
• ward is a categorical variable (1 = 2 patients per nurse, 2 = 4 patients per
nurse).
• nsatisfy is a continuous variable (patient’s nurse satisfaction score).
Data set—load the data set: Chapter 04 – Exercise 09.sav.
a. Run descriptive statistics and a histogram with a normal curve for nsatisfy
for the whole data set.
b. Run descriptive statistics and a bar chart for ward for the whole data set.
c. Run descriptive statistics and a histogram with normal curve for nsatisfy
for the 2 patients per nurse ward only.
d. Run descriptive statistics and a histogram with a normal curve for nsatisfy
for the 4 patients per nurse ward only.
10. To determine if dancing enhances mood, you recruit 100 people who are
interested. You randomly select 50 and give them seven free dance lessons; the
other 50 get no dance lessons. After the seventh class, you administer the Acme
Happiness Scale Survey (AHSS) to all 100 individuals; this survey renders a
score ranging from 1 to 30 (1 = extremely unhappy… 30 = extremely happy).
Codebook—the SPSS file contains two variables:
• dance is a categorical variable (1 = dancer, 2 = nondancer).
• ahss is a continuous variable (Acme Happiness Scale Survey score).
Data set—load the data set: Chapter 04 – Exercise 10.sav.
a. Run descriptive statistics and a histogram with a normal curve for ahss for
the whole data set.
b. Run descriptive statistics and a bar chart for dance for the whole data set.
c. Run descriptive statistics and a histogram with a normal curve for ahss for
the dancers only.
d. Run descriptive statistics and a histogram with a normal curve for ahss for
the nondancers only.
 CHAPTER 5
t Test

The difference between a violin and a viola is that a viola burns longer.

—Victor Borge
LEARNING OBJECTIVES
Upon completing this chapter, you will be able to:
• Determine when it is appropriate to run a t test
• Verify that the data meet the criteria for t test processing: normality, homogeneity
of variance, and n
• Order a t test with graphics
• Interpret the test results
• Comprehend the α level and p value
• Resolve the hypotheses
• Document the results in plain English
• Understand the implications of Type I and Type II errors
• Apply techniques for reducing the likelihood of committing Type I and Type II
 errors
DIGITAL LEARNING RESOURCES
The tutorial video and data sets for this chapter are at www.sagepub.com/knapp
• t Test pretest checklist video
• t Test run video
• SPSS files
OVERVIEW
NOTE: From here forward, the “μ” character will be used to symbolize the mean.

The t test is one of the most common and versatile statistical tests in the realm of
experimental research and survey methodology. The t test is used when there are two
groups, each of which renders a continuous variable for the outcome (e.g., height, age,
weight, number of teeth, bank account balance, IQ score, score on a depression assessment
instrument, blood pressure, test score, typing speed, etc.).
In the most basic experimental setting, the design consists of two groups: a control
group, which gets a placebo or treatment as usual, and a treatment group, which gets the
innovative intervention that is the focus of the study.
We can compute the mean for each group, and we would not expect the two means to be
identical; they would likely be different. The t test answers the question, “Is there a
statistically significant difference between μ(Control) and μ(Treatment)?” In other words,
the result of the t test helps us to determine if one group substantially outperformed the
other or if the differences between the means are essentially incidental.
EXAMPLE
Dr. Zinn and Dr. Zorders arranged to offer Z-Math, their innovative method for teaching
students the multiplication table (1 × 1… 10 × 10), to fourth graders at Pico Street School.
RESEARCH QUESTION
Does Z-Math outperform traditional multiplication educational methods?
GROUPS
The names of all 60 fourth graders at the school are written on slips of paper and placed
into a hat. The principal randomly draws 30 names from the hat; these students will be
assigned to the control group. The remaining 30 students will be assigned to the Z-Math
group.
PROCEDURE
From 10:00 to 11:30 a.m., the usual time for the daily math lesson, the students in the
control group will go to Room 14, where they will receive the regular math lesson
(treatment as usual). The students in the Z-Math group will go to Room 8 for their Z-Math
lesson. The students in the Z-Math group are instructed not to disclose the Z-Math method
to the other students.
After 4 weeks, all 60 students will be given a test; they will have 10 minutes to write
out the entire multiplication table with the solutions on a blank sheet of paper (e.g., 1 × 1
= 1, 1 × 2 = 2… 10 × 10 = 100).
HYPOTHESES
H0: There is no statistically significant difference in the test scores between the two
groups.
H1: The Z-Math group statistically significantly outperformed the math-as-usual group.

DATA SET
Use the following data set: Chapter 05–Example 01–T Test.sav.

PRETEST CHECKLIST
The statistical pretest checklist is akin to looking both ways before you cross the street;
certainly you could cross the street without looking, but you would probably wind up in
much better shape if you looked first. In terms of statistical tests, certainly you could run
the statistical test without tending to the pretest checklist, but you may unknowingly wind
up with misleading findings.
The formulas that compose each statistical test require that the source data meet a
unique set of criteria in order for that test to operate properly. These criteria are referred to
as assumptions—we assume that the data meet the criteria specified by the test at hand.
Actually, we need to do more than just passively assume that the data are suitable for
processing; we need to actively assess the source data before proceeding with the test.

When the tests on the pretest checklist (statistical assumptions) are satisfied, we can
consider the statistical results relatively “robust.” When one or more of the pretest
checklist criteria are not satisfied, we still proceed with the analysis, but we would be a bit
less confident in the solidity of our findings. In the interest of proper scientific ethics and
the principles of full disclosure, it would be appropriate to mention any such (statistical)
shortcomings when discussing the results. This notion pertains to the unique pretest
checklists for the other tests covered in this text as well.
The pretest criteria for running a t test involve checking the data for (1) normality, (2)
homogeneity (pronounced hoe-moe-juh-nay-it-tea) of variance, and (3) an n quota. SPSS
is equipped to run all of these (pre)tests.
Pretest Checklist Criterion 1—Normality
Checking for normality involves producing a histogram with a normal curve for each of
the two groups. In this instance, you would click on the Select Cases icon to select the
records pertaining to the math-as-usual group; the selection criteria would be group = 1.
Next, run a histogram (with normal curve) on the variable score. Then repeat the process
for the Z-Math group (group = 2). For more details on this procedure, please refer to
Chapter 4 (“SPSS—Descriptive Statistics: Continuous Variable (Age) Select by
Categorical Variable (Gender)—Females Only”); see the star ( ) icon on page 71.
This will produce two histograms with normal curves—one for scores in the math-as-
usual group and the other for the scores in the Z-Math group. The histograms should
resemble the graphs shown in Figures 5.1 and 5.2.
As we read these two histograms, set aside the X,Y scales and the possible irregularities
among the bars; our attention is focused on the shape of the normal curve. We are looking
for normality (symmetry) in these two curves. Although the normal curve in Figure 5.1 is
shorter and fatter than the normal curve in Figure 5.2; in terms of normality, this is not an
issue. The critical thing to observe is that both normal curves are sufficiently symmetrical
—in other words, if you sliced this curve vertically down the middle, the left side would
resemble a mirror image of the right side; sometimes this normal curve is aptly referred to
by its characteristic shape as a “bell curve.” In this example, we see that both curves are
symmetrical; there is no notable skew in either curve. Hence, we would say that the
criteria of normality are satisfied for both the math-as-usual and the Z-Math groups.
Figure 5.1 Histogram of Score for G1: Math-as-Usual Group

Figure 5.2 Histogram of Score for G2: Z-Math Group

Pretest Checklist Criterion 2—Homogeneity of Variance
Homogeneity pertains to sameness; the homogeneity of variance criteria involve checking
that the variances of the two groups are not too different from each other. As a rule of
thumb, homogeneity of variance is likely to be achieved if the variance from one group is
not more than twice the variance of the other group. In this case, the variance (standard
deviation squared) for score in the math-as-usual group is 49.84, and the variance for
score in the Z-Math group is 77.96. Clearly, 77.96 is not more than twice the value of
49.84, so we would expect that the homogeneity of variance test would pass.
Take a look at Figures 5.3 and 5.4; these graphs are drawn from two theoretical data
sets, both of which have three things in common: n = 100, mean = 50, and normal
distributions. The most notable difference is the spread of the data—in Figure 5.3, the data
ranges from 27 to 70, with a variance of 80 (SD = 8.957), whereas the data in Figure 5.4
have a much broader range, from 7 to 92, with a variance of 313 (SD = 17.690). As
discussed in Chapter 4, as values deviate further from the mean, this increases the standard
deviation. Since the variance = the standard deviation squared, the variance reflects these
deviations in an even more pronounced way.
Figure 5.3 Smaller Variance (80)
 Figure 5.4 Larger Variance (313)

Considering that the variance from one data set (313) is more than double the variance
from the other data set (80), we would expect the homogeneity of variance test to render a
(p) value that is less than .05, indicating a statistically significant difference between the
variances of these two variables and a violation of that criterion.
In SPSS, the homogeneity of variance test is an option selected during the actual run of
the t test. If the homogeneity of variance test renders a significance (p) value that is greater
than .05, then this suggests that there is no statistically significant difference between the
variance from one group to the other group. This would mean that the data pass the
homogeneity of variance test. The notion of the p value will be discussed in detail in the
Results section in this chapter, when we examine the findings produced by the t test.
Pretest Checklist Criterion 3—n
Technically, you can process a t test with an n of any size in each group, but when the n is
at least 30 in each group, the result of the t test is considered more robust. We will see the
n reported for each group in the Results section, when we examine the findings produced
by the t test.

TEST RUN
To run a t test:
1. On the main screen, click on Analyze, Compare Means, One-Way ANOVA…
(Figure 5.5).
Figure 5.5 Running a t Test
 2. On the One-Way ANOVA menu, move the continuous variable that you wish to
analyze (score) into the Dependent List window, and move the variable that contains
the categorical variable that specifies the group (group) into the Factor window
(Figure 5.6).
NOTE: SPSS does offer a t test menu screen, but the ANOVA order menu (below) is
easier to load and produces an equivalent report. In addition, this same menu will be used
to run ANOVA tests, which will be covered in Chapter 6; it is easier to learn one menu
instead of two.
Figure 5.6 The One-Way ANOVA Menu

3. Click on the Options… button, and on the One-Way ANOVA: Options menu, check
Descriptive, Homogeneity of variance test, and Means plot, then click on the
Continue button (Figure 5.7).
Figure 5.7 The One-Way ANOVA: Options Menu
 4. This will take you back to the One-Way ANOVA menu (Figure 5.6); click on the OK
button, and the t test will process.

RESULTS

We will begin this section by completing the pretest checklist; we still need answers
regarding the homogeneity of variance and the ns for each group. The last column of the
homogeneity of variance table (Table 5.1) shows a significance (p) value of .433; since
this is greater than the α level of .05, this tells us that there is no statistically significant
difference between the variances in the score variable from the math-as-usual group
compared with the score variable from the Z-Math group. We would conclude that the
criteria of the homogeneity of variance have been satisfied.
Table 5.1 Homogeneity of Variance Test Results

Further details regarding these criteria will be discussed further, but before proceeding
with any further interpretation of the results report, it is necessary to discuss two statistical
concepts that pertain to the rest of the statistical tests contained in this text: p value and the
alpha (α) level, both of which will be covered shortly.
As for the final item on the pretest checklist, the n, we can see from the second column
in Table 5.2 that math as usual and Z-Math each has an n of 30. This satisfies the n
assumption, stating that each group should have an n of at least 30.
 At this point, you have probably noticed column 3 in Table 5.2, which shows that the
mean score for the Z-Math group (80.97) is higher than the mean score for the math-as-
usual group (75.77). You can also see these means plotted graphically (Figure 5.8). On the
basis of these means, you may hastily conclude that the school should just adopt the Z-
Math system since this group scored higher than the math-as-usual group, but in statistics,
the world is not so simple.
Statisticians recognize and actively acknowledge that we do not live in a perfect world;
no matter how hard we try to conduct quality investigations, the scientific process can be a
messy proposition littered with multiple confounding variables—conditions for which we
cannot fully control or account for.
Table 5.2 Descriptive Statistics of Score for Math-as-Usual and Z-Math Groups

Figure 5.8 Means Plot of Math-as-Usual and Z-Math Groups

In our simple example, judging by the mean test scores of the two groups, it looks like
the Z-Math group outperformed the math-as-usual group, and this may in fact turn out to
be the case, but other factors may have contributed to the differences observed between
the (mean) scores of these two groups. For example, maybe the random distribution
process unexpectedly routed more math-smart students to the Z-Math group; maybe the
class clown was routed to the math-as-usual group and created amusing distractions
during the lessons or the test; maybe the members of the math-as-usual group held
slumber parties where the Z-Math students were not invited, and there was very little
slumber—this lack of sleep may have adversely affected their concentration levels; maybe
the Z-Math group developed a sense of rivalry, which motivated them to work harder to
try to outperform the other group; maybe the members of the math-as-usual group were
displeased that they were not assigned to the Z-Math group and developed a negative
attitude toward their math lessons. Any number of these or other things that we do not
know about may have been going on over the course of this study.
As you can see, there is virtually no limit to the variety of confounding variables that
could potentially influence the outcome of a study. Since we cannot fully account for or
compensate for such confounds, we know we do not have a pure experimental setting.
Hence, we do not speak of our statistical findings with absolute certainty; rather, we speak
of how much confidence we have in our findings.
The key question in this case is, How certain can we be that the 5.2-point difference that
we detected between the group means (μ(math as usual) = 75.77 and μ(Z-Math) = 80.97)
is actually due to the superiority of the intervention (Z-Math) and not due to chance alone?
In other words, we want a number that will tell us how likely we would detect this result
(the 5.2-point difference in the means) if Z-Math was a sham—if it did nothing. This
number is known as the significance level, represented by the letter p.
Here is how the significance (p) value works: Look at the last column of Table 5.3; the
Sig.(nificance) score is .015; this is the p value. This tells us that we would expect to see
the 5.2-point difference in the group scores about 1.5% of the time if it were occurring by
(random) chance alone. In other words, based on the data gathered, if Z-Math in fact had a
null effect, we would see the Z-Math group outperform the math-as-usual group by 5.2
points about 1.5% of the time.
Since the p level tells us how often we would be accepting the intervention as effective,
when in fact it really is not, the lower the p level, the more significant the findings.
We will take this example one step further: Suppose the p level had been .001 in our
study; this would tell us that there is (only) 1 chance in 1,000 that we would be seeing
such results by chance alone (e.g., if Z-Math was completely ineffective), which would
lead us to wrongly believe that the differences in the scores that we are observing are
likely attributable to the effectiveness of the intervention (Z-Math).
Table 5.3 t Test Results Comparing Scores for Math-as-Usual and Z-Math Groups

In the next section, we will see how we use the p value to determine which hypothesis
to reject and which to accept.

HYPOTHESIS RESOLUTION
We need to have a way of using the p level to guide us in making decisions about our
pending hypotheses:
H0: There is no statistically significant difference in the test scores between the two
groups.
H1: The Z-Math group statistically significantly outperformed the math-as-usual group.

p Value
To do this, before we embark on our research process, we draw a somewhat arbitrary
numerical line in the sand, known as the alpha (α) level. In social sciences, the α level is
typically set to .05. Think of the α level as a sort of statistical significance threshold—any
p value that is .05 or less is considered statistically significant—hence, we reject H0,
which states that there is no significant difference between the groups. If the p value is
over .05, then the differences between the means are not considered statistically significant
—hence, we do not reject H0. This will guide us in making our decisions regarding the
hypotheses.

p Value Summary
α = .05 (typical for social sciences)
If p ≤ α, then there is a statistically significant difference; reject H0.
If p > α, then there is no statistically significant difference; do not reject H0.

Our investigation produced a p value of .015; since this is less than the α level (.05), we
consider the differences detected between the groups statistically significant, and as such,
we reject the null hypothesis (H0), which states that there is no statistically significant
difference between the groups, and opt for the alternate hypothesis (H1), which states that
the difference detected between the groups is attributable to the Z-Math intervention.
Alternatively, if our p value had been greater than .05 (e.g., .85), then we would have
accepted the null hypothesis (H0) and considered the differences observed an artifact of
random chance; furthermore, we would have rejected the alternate hypothesis (H1), which
states that it was the intervention (Z-Math) that accounted for the differences observed
between groups.
Another way of understanding the value of the p level is to consider running this
experiment repeatedly. The p level of .05 tells us that if we were to run this study 100
times and, in reality, Z-Math had no effect (students in the Z-Math group achieved about
the same scores as those in the math-as-usual group), then about 5% of the time, we would
see the Z-Math group outperform the math-as-usual group just by chance alone and not
because Z-Math is genuinely superior. So, as we encounter lower p levels, this suggests
that the difference in the (mean) scores between groups was more likely due to the
treatment and not simply attributable to random chance.
If the p level had been .85, then this tells us that if we ran this study 100 times, 85% of
the time we would expect to see the Z-Math group outperform the math-as-usual group
just by random chance and not because Z-Math is genuinely superior. Hence, the lower the
p level, the less likely that random chance is responsible for producing the observed
differences between the groups.
Furthermore, suppose we got the following results:
μ(Z-Math) = 97
μ (Math as usual) = 67
p = .01
Here is how we would interpret this: The Z-Math group outperformed the math-as-usual
group by 30 points (97 − 67 = 30). The p level of .01 tells us that if we ran this experiment
100 times and Z-Math had a null effect (meaning, it did not really work), then we would
see results like this just by random chance alone about 1% of the time. So, as the p level
gets lower, we can have more confidence that the (30-point) difference observed between
the groups is attributable to the treatment (Z-Math) and not merely an artifact of random
chance.
With the principles of the p (significance) value and the α level in mind, it is time to
revisit the results of the homogeneity of variance test (Table 5.4).
Table 5.4 Homogeneity of Variance Test Results (Same as Table 5.1)

The homogeneity of variance test checks to see that the variances of the groups
involved are similar enough to each other. Remember: The variance is the standard
deviation squared (SD scores derived from Table 5.2):
Variance for math as usual = 7.062 = 49.84
Variance for Z-Math = 8.832 = 77.96
The homogeneity of variance test compared 49.84 with 77.96 and computed a Sig.
(nificance) score (p value) of .433, which is greater than the α level of .05. This tells us
that there is no statistically significant difference between these two variances, meaning
that the variances are similar enough to say that the homogeneity of variance criteria are
satisfied.
DOCUMENT RESULTS
Although it is essential to comprehend the meaning of key values in the statistical reports,
it would be inappropriate to simply present the figures in a results section without
providing a concise narrative. While all figures below are technically correct, try to avoid
documenting your findings as such.

Accurate but Inappropriate Numerical Statistical Abstract

Math as usual: n = 30, μ = 75.77, SD = 7.06
Z-Math: n = 30, μ = 80.97, SD = 8.83
p = .015, α = .05, therefore, use Z-Math

While the above data may be useful in assembling a table, it is important that you
become proficient at translating your methodology and numerical findings into a brief
textual abstract detailing the story of the study specifying the research question, along
with an overview of how you got from the research question to the results.

Appropriate Verbose Statistical Abstract

A group of 60 fourth graders were randomly assigned to one of two groups to
determine if Z-Math was superior to traditional teaching methods when it comes to
learning the multiplication table.
After 1 month, students in the Z-Math group learned an average of 80.97% of the
multiplication table, significantly outperforming students taught using traditional
methods, who, on average, learned 75.77% of the multiplication table (p = .015, α =
.05).

In addition to the full manuscript, scientific journals also require authors to submit an
abstract that tells the overall story of the study and key findings; usually the limit for the
abstract is about 200 words. While initially it can be a challenge to write technical
information so concisely, this is a worthy skill to develop. The above abstract is under 100
words.

NOTE: In the example processed in this chapter, we saw that the t test assessed the means
of the two groups and revealed that the mean for the treatment group (Z-Math) was
statistically significantly higher than the mean for the control group (math as usual). As
you will see in the exercises for this chapter, the t test is equally effective in detecting
statistically significant differences when the mean of the treatment group is lower than the
mean of the control group. For example, instead of the treatment (Z-Math) that is designed
to increase a test score, in Exercise 1 in the Practice Exercises, the treatment (meditation)
is designed to lower resting pulse rate. Remember: The t test is designed to simply detect
statistically significant differences between the means of two groups; it does not matter
which group mean is higher and which is lower.

TYPE I AND TYPE II ERRORS

The world is imperfect; despite all best efforts, errors can occur in virtually any realm no
matter how careful you are. For starters, consider the two types of errors that can occur in
a legal verdict:
• (Error 1) The court finds the defendant not guilty when, in fact, he or she actually is
guilty.
• (Error 2) The court finds the defendant guilty when, in fact, he or she is actually not
guilty.
These same two types of errors can happen in statistics:
• Type I Error: A Type I error (α error) occurs when the findings indicate that there is
a statistically significant difference between two variables (or groups) when, in fact, on the
whole, there actually is not, meaning that you would erroneously reject the null
hypothesis. The consequence is that you would conclude the treatment was effective
when, in fact, on the whole, it was not. This is connected with the p value; a p value of .05
means that there is a 5% chance that you have committed a Type I error—hence, the lower
the p value, the less likely that you have committed a Type I error.
• Type II Error: A Type II error (β error) occurs when the findings indicate that there
is no statistically significant difference between two variables (or groups) when, in fact, on
the whole, there actually is, meaning that you would erroneously accept the null
hypothesis. The consequence is that you would conclude the treatment was ineffective
when, in fact, on the whole, it was. Sample size is inversely related to Type II errors—the
higher the sample size, the lower the likelihood of committing a Type II error.
There is no formal metric that you can run that will tell you if you have a Type I or Type
II error on hand; they are just characteristics 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 level (e.g., p < .05), this does not mean that you have solid evidentiary proof of
anything; at best, you have reduced uncertainty. Essentially, a p value of .05 means that if
the effect of the treatment group were the same as the control group (null intervention), we
would see this (anomalous) statistical outcome (where the treatment group outperforms
the control group, just by chance) about 5% of the time. Since p never goes to zero, there
is always some level of uncertainty in statistical findings.
Remember: Statistics is not about proving or disproving anything; statistics is about
reducing uncertainty—there is always some margin of error, no matter how small. The
notion of Type I and Type II errors pertains to all other tests covered in the chapters that
follow.
GOOD COMMON SENSE
Clearly, it is essential that you comprehend key figures of the statistical reports that you
order, but when it comes to using these findings to make decisions in the real world, other
considerations must be taken into account.
For example, suppose you measured the results of a new treatment, wherein Group 1
was the control group (received no treatment) and Group 2 received the treatment. Your
analysis revealed that on a 100-point scale, μ(Group 1) = 88.1, μ(Group 2) = 88.3, and p =
.01. Your first reaction might be to leap to the conclusion that since p = .01, and this is less
than the α level of .05, Group 2 statistically significantly outperformed Group 1, so we
should adopt the treatment that was used for Group 2. Statistically speaking, you would be
right, but consider some other real-world factors: Group 2 outperformed Group 1 by
(only) .2 points; one might wonder if, in the real world, a .2-point difference is of any
practical significance—can anyone really detect the difference between an outcome of
88.1 and 88.3?
Also, scarce resources may be an issue; the intervention used for Group 2 might have
been very costly, time-consuming, labor intensive, or complex to carry out—it would be
reasonable to consider if the cost, time, labor, and inconvenience involved are really worth
the .2-point improvement.
Another concern might be the initial goal of the intervention: If the goal was to raise the
score to at least 80, then clearly both groups achieved the desired effect—in this light, the
innovative treatment would seem less worthy.
The lesson at hand is that when interpreting statistical findings, it is important that you
not only tend to the numbers but also mindfully comprehend that those numbers are only a
part of the picture when it comes to making practical decisions in the real world.

Key Concepts
• t Test
• Pretest checklist
Normality
Homogeneity of variance
n
• α
• p
• Hypothesis resolution
• Documenting results
• Good common sense
• Type I (α) error
• Type II (β) error

Practice Exercises
For Exercises 1 to 5:
• Configure SPSS per the specifications of the codebook.
• Enter Data Set A as shown in the table.
• Process and document your findings for Data Set A.
• Repeat each exercise using Data Set B (which will produce different results).
• Be sure to save the data sets since they will be used again in Chapter 6.
For Exercises 6 to 10:
• Use the prepared SPSS data sets (download from www.sagepub.com/knapp).
• Load, process, and document your findings for Data Set A.
• Repeat each exercise using Data Set B (which will produce different results).
1. You want to determine if meditation can reduce resting pulse rate. Participants
were recruited and randomly assigned to one of two groups: Members of Group
1 (the control group) did not meditate; members of Group 2 (the treatment
group) meditated for 30 minutes per day on Mondays, Wednesdays, and Fridays
over the course of 2 weeks. At the end, you gathered the resting pulse rates for
each participant.
Codebook—set up SPSS to contain two variables:
• group is a categorical variable (1 = no meditation, 2 = meditates 3 days).
• pulse is a continuous variable (number of beats per minute).
Data set—enter Data Set A.
• Save Data Set A as Chapter 05 − Exercise 01 A.sav; it will be used again
for Chapter 6, Exercise 1.
a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the t test and document your findings (ns, means, and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the t
test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Data Set B.
• Save Data Set B as Chapter 05 − Exercise 01 B.sav; it will be used
 again for Chapter 6, Exercise 1.
2. You want to determine if pairing an incoming freshman with a sophomore in a
protégé-mentor relationship will enhance the freshman’s grade point average
(GPA). You recruit sophomores who are willing to mentor a student in their
major for their first term. You then recruit freshmen who are interested in having
a mentor. Freshmen who apply to this program will be assigned to the mentor
group (and will be designated a mentor) or to the control group (no mentor) on
an alternating basis. Those in the mentor group are to meet in person once a
week at a time of their choosing. All freshmen, in both groups, agree to submit
their transcript at the conclusion of the term.
Codebook—set up SPSS to contain two variables:
• group is a categorical variable (1 = no mentor, 2 = in-person mentor).
• gpa is a continuous variable (the first-term GPA for each freshman).
Data set—enter Data Set A.
• Save Data Set A as Chapter 05 − Exercise 02 A.sav; it will be used again
for Chapter 6, Exercise 2.
a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the t test and document your findings (ns, means, and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the t
test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Data Set B.
 • Save Data Set B as Chapter 05 − Exercise 02 B.sav; it will be used
again for Chapter 6, Exercise 2.
3. The Acme Company claims that its new reading lamp increases reading speed;
you want to test this. You will record how long (in seconds) it takes for subjects
to read a 1,000-word essay. Half of the subjects will be randomly assigned to the
control group; they will read the essay using regular room lighting. The other
half of the subjects will read the same essay using the Acme reading lamp.
Codebook—set up SPSS to contain two variables:
• group is a categorical variable (1 = room lighting, 2 = Acme lamp).
• seconds is a continuous variable (the time it takes to read the essay).
Data set—enter Data Set A.
• Save Data Set A as Chapter 05 – Exercise 03 A.sav; it will be used again
for Chapter 6, Exercise 3.
a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the t test and document your findings (ns, means, and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the t
test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Data Set B.
• Save Data Set B as Chapter 05 – Exercise 03 B.sav; it will be used
 again for Chapter 6, Exercise 3.
4. You want to determine if chocolate enhances mood. Subjects will be recruited
and randomly assigned to one of two groups: Those in the control group will eat
their regular diet, and those in the experimental group will also eat their usual
meals and have one piece of chocolate at breakfast, lunch, and dinner over the
course of a week. At the end of the week, all subjects (in both groups) will
complete the Acme Mood Scale (1 = extremely bad mood… 100 = extremely
good mood).
Codebook—set up SPSS to contain two variables:
• group is a categorical variable (1 = no chocolate, 2 = chocolate [1 per
meal]).
• mood is a continuous variable (score on the Acme Mood Scale).
Data set—enter Data Set A.
• Save Data Set A as Chapter 05 – Exercise 04 A.sav; it will be used again
for Chapter 6, Exercise 4.
a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the t test and document your findings (ns, means, and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the t
test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Data Set B.
• Save Data Set B as Chapter 05 – Exercise 04 B.sav; it will be used
 again for Chapter 6, Exercise 4.
5. You want to find out if music enhances problem solving. Subjects will be
recruited and randomly assigned to one of two groups: Those in the control
group will be given a standard 100-piece jigsaw puzzle to solve in a quiet room.
Participants in the experimental group will be given the same puzzle to
assemble, but instead of silence, there will be classical music playing at a soft
volume (30 decibels [dB]) in the room. You will record the time it takes for each
person to complete the puzzle.
Codebook—set up SPSS to contain two variables:
• group is a categorical variable (1 = no music, 2 = music at 30 dB).
• seconds is a continuous variable (how long it took to finish the puzzle).
Data set—enter Data Set A.
• Save Data Set A as Chapter 05 – Exercise 05 A.sav; it will be used again
for Chapter 6, Exercise 5.
 a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the t test and document your findings (ns, means, and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the t
test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Data Set B.
• Save Data Set B as Chapter 05 – Exercise 05 B.sav; it will be used
again for Chapter 6, Exercise 5.
6. You want to determine if watching a video of a comedy with a laugh track is
more enjoyable than watching without it. Subjects will be recruited and
 randomly assigned to one of two groups: Those in the control group will watch
the video without the laugh track, and those assigned to the treatment group will
watch the same video with the sound(s) of a 50-person audience included in the
soundtrack. Each participant will watch the video individually; no others will be
present in the room. Immediately following the video, each participant will be
asked to rate how enjoyable the show was on a scale of 1 to 5 (1 = not very
enjoyable… 5 = very enjoyable).
Codebook—set up SPSS to contain two variables:
• group is a categorical variable (1 = no laugh track, 2 = laugh track at 50).
• enjoy is a continuous variable (1 = not very enjoyable… 5 = very
enjoyable).
Data set—load Chapter 05 – Exercise 06 A.sav.
a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the t test and document your findings (ns, means, and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the t
test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Chapter 05 – Exercise 06 B.sav.
7. In an effort to determine the effectiveness of light therapy to alleviate
depression, you recruit a group of subjects who have been diagnosed as
depressed. The subjects are randomly assigned to one of two groups: The
control group will receive no light therapy, and the treatment group will get light
therapy for 1 hour on even-numbered days over the course of 1 month, at which
time, all participants will complete the Acme Mood Scale, consisting of 10
questions; this instrument renders a score between 1 and 100: 1 = extremely bad
mood… 100 = extremely good mood.
Codebook—the data set contains two variables:
• group is a categorical variable (1 = no light therapy, 2 = light therapy:
even days).
• mood is a continuous variable (1 = extremely bad mood… 100 =
extremely good mood).
Data set—load Chapter 05 – Exercise 07 A.sav.
a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the t test and document your findings (ns, means, and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the t
 test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Chapter 05 – Exercise 07 B.sav.
8. It is thought that exercising early in the morning will provide better energy
throughout the day. To test this idea, subjects are recruited and randomly
assigned to two groups: For 30 days, members of the experimental group will
walk from 7:00 to 7:30 a.m., Monday through Friday; members of the control
group will do no walking. At the conclusion of the study, each subject will
answer the 10 questions on the Acme End-of-the-Day Energy Scale. This
instrument produces a score between 1 and 100 (1 = extremely low
energy… 100 = extremely high energy).
Codebook—the data set contain two variables:
• group is a categorical variable (1 = no walking, 2 = walking: 30 minutes).
• energy is a continuous variable (1 = extremely low energy… 100 =
extremely high energy).
Data set—load Chapter 05 – Exercise 08 A.sav.
a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the t test and document your findings (ns, means, and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the t
test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Chapter 05 – Exercise 08 B.sav.
9. Clinicians at a nursing home facility want to see if giving residents a plant to
tend to will help to lower depression. To test this idea, the residents are
randomly assigned to two groups: Each member of the treatment group will be
issued a small bamboo plant along with a card detailing care instructions;
members of the control group will receive no plant. After 90 days, all
participants will complete the Acme Depression Scale, a 10-question instrument
that renders a score between 1 and 100 (1 = low depression… 100 = high
depression).
Codebook—the data set contain two variables:
• group is a categorical variable (1 = no plant, 2 = bamboo).
• depress is a continuous variable (1 = low depression… 100 = high
depression).
Data set—load Chapter 05 – Exercise 09 A.sav.
a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
 c. Run the t test and document your findings (ns, means, and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the t
test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Chapter 05 – Exercise 09 B.sav.
10. During flu season, the administrators at a walk-in health clinic want to
determine if providing patients with a pamphlet will increase their receptivity to
flu shots. Once escorted to the exam room, patients who had an odd-numbered
service ticket were given a flu shot information pamphlet; patients with even
tickets were given nothing. At the end of the day, the charts were reviewed and
two entries were made in the database: total number of flu shots given to
patients who received a pamphlet and total number of flu shots given to patients
who were not given flu shot pamphlets.
Codebook—the data set contain two variables:
• group is a categorical variable (1 = nothing, 2 = flu shot pamphlet).
• shots is a continuous variable (number of flu shots given in a day [for
each group]).
Data set—load Chapter 05 – Exercise 10 A.sav.
a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the t test and document your findings (ns, means, and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the t
test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Chapter 05 – Exercise 10 B.sav.
 CHAPTER 6
ANOVA

Three is a magic number.

—Bob Dorough
LEARNING OBJECTIVES
Upon completing this chapter, you will be able to:
• Determine when it is appropriate to run an ANOVA test
• Verify that the data meet the criteria for ANOVA processing: normality,
homogeneity of variance, and n
• Order an ANOVA test with graphics
• Select an appropriate ANOVA post hoc test: Tukey or Sidak
• Derive results from the multiple comparisons table
• Compute a unique pairs formula
• Resolve the hypotheses
 • Document the results in plain English
DIGITAL LEARNING RESOURCES
The tutorial video and data sets for this chapter are at www.sagepub.com/knapp
• ANOVA pretest checklist video
• ANOVA test run video
• SPSS files
OVERVIEW
The ANOVA (analysis of variance) test is similar to the t test, except whereas the t test
compares two groups of continuous variables with each other, the ANOVA test can
compare three or more groups.
The t test and ANOVA are so similar that this chapter will use the same example and the
same exercises used in Chapter 5; the only difference is that the data sets have been
enhanced to include a third or fourth group. If you are proficient with the t test, you are
already more than halfway there to comprehending ANOVA. The only real differences
between the t test and ANOVA are in ordering the test run and interpreting the test results;
several other minor differences will be pointed out along the way.
That being said, let us go into the expanded example, drawn from Chapter 5, which
involved Group 1 (math taught as usual), Group 2 (Z-Math), and now a third group: Group
3 (flashcards).
The ANOVA test will reveal which (if any) of these groups statistically significantly
outperforms the others.
EXAMPLE
Dr. Zinn and Dr. Zorders arranged to offer Z-Math, their innovative method for teaching
students the multiplication table (1 × 1… 10 × 10), to fourth graders at Sepulveda Street
School. The principal is willing to try the Z-Math method, but she is also curious about
how effective multiplication flashcards with a math coach might be.
RESEARCH QUESTION
Which is the best method for teaching multiplication: (1) the usual teaching method, (2) Z-
Math, or (3) flashcards with a math coach?
GROUPS
The names of all 90 fourth graders at the school are written on slips of paper and placed in
a hat. The principal randomly draws 30 names from the hat; these students will be
assigned to the control group, the next 30 names drawn will be assigned to the Z-Math
group, and the remaining 30 students will be assigned to the flashcard group.
PROCEDURE
From 10:00 to 11:30 a.m., the usual time for the daily math lesson, the students in the
control group will go to Room 14, where they will receive the regular math lesson
(treatment as usual). The students in the Z-Math group will go to Room 8 for their Z-Math
lesson. Finally, the students in the flashcard group will go to Room 20, where each student
will be paired with a community volunteer to work with the student using flashcards. The
students in the Z-Math group are instructed not to disclose the Z-Math method to the other
students, and the students in the flashcard group are instructed not to share their flashcards
with other students.
After 4 weeks, all 90 students will be given a test; they will have 10 minutes to write
out the entire multiplication table with the solutions on a blank sheet of paper (e.g., 1 × 1
= 1, 1 × 2 = 2… 10 × 10 = 100).
HYPOTHESES
H0: There is no statistically significant difference in the test scores among the three
groups.
H1: The Z-Math group outperforms the math-as-usual group.
H2: The flashcards group outperforms the math-as-usual group.
H3: The Z-Math group outperforms the flashcards group.
H4: The flashcards outperforms the Z-Math group.
Notice that as more groups are introduced, more comparisons are possible, and hence
the supplemental hypotheses: H2 addresses the comparison between the math-as-usual and
flashcards groups, and H3 and H4 involve comparing the Z-Math and flashcards groups.
As there are more groups, there are more comparisons to analyze, which is what ANOVA
is all about.
DATA SET
Use the following data set: Chapter 06 – Example 01 – ANOVA.sav.
Notice that this data set has 90 records; the first 60 records (rows) are the same as the t
test example data set used in Chapter 5:
Records 1 to 30 are for Group 1: math as usual.
Records 31 to 60 are for Group 2: Z-Math.
Records 61 to 90 are new; they contain the data for Group 3: Flashcards.
NOTE: The data are grouped this way just for visual clarity; you do not need to load your
data sequenced by group.
If you go to the Variable View and open the Values menu for group, you will see that
the label Flashcards for the third group has been assigned to the value 3 (Figure 6.1).
Figure 6.1 Value Labels for a Three-Group ANOVA Analysis
PRETEST CHECKLIST
The statistical pretest checklist for the ANOVA is similar to the t test, except instead of
processing two groups, you will process three or however many groups there are.
Pretest Checklist Criterion 1—Normality
Check for normality by inspecting the histogram with a normal curve for each of the three
groups. Begin by using the Select Cases icon to select the records pertaining to the math-
as-usual group; the selection criteria would be group = 1. Next, run a histogram (with
normal curve) on the variable score. For more details on this procedure, please refer to
Chapter 4 (“SPSS—Descriptive Statistics: Continuous Variable (Age) Select by
Categorical Variable (Gender)—Females Only”); see the star ( ) icon on page 71.
Then repeat the process for the Z-Math group (group = 2), and finally, repeat the
process a third time for the flashcards group (group = 3).
This will produce three histograms with normal curves—one for the scores in the math-
as-usual group, a second for the scores in the Z-Math group, and a third for the scores in
the flashcard group. The histograms should resemble the graphs shown in Figures 6.2, 6.3,
and 6.4.
Figure 6.2 Histogram of Score for G1: Math-as-Usual Group
 Figure 6.3 Histogram of Score for G2: Z-Math Group

Figure 6.4 Histogram of Score for G3: Flashcards Group

As we read these three histograms, our focus is on the normality of the curve, as
opposed to the characteristics of the individual bars. Although each curve is unique in
terms of height and width, we see that each is bell-shaped and shows good symmetry with
no substantial skewing. On the basis of the inspection of these three figures, we would
conclude that the criteria of normality are satisfied for all three groups.
Next, you need to (re)activate all records for further analysis; you can either delete the
temporary variable Filter_$ or click on the Select Cases icon and select the All cases
button. For more details on this procedure, please refer to Chapter 4 (“SPSS—Selecting
All Variables”); see the star ( ) icon on page 76.
Pretest Checklist Criterion 2—Homogeneity of Variance
Since we process the ANOVA with the same menu as the t test, we will select the
homogeneity of variance test when we order the ANOVA test and read the findings as part
of the results. The homogeneity of variance rule of thumb for the ANOVA test is just like
the t test: None of the groups should have a variance (standard deviation2) that is more
than twice the variance of any other group. In other words, if Group 1 had a variance of
20.1, Group 2 had a variance of 24.7, and Group 3 had a variance of 90.6, we would
expect the homogeneity of variance criteria to fail since 90.6 is clearly more than twice as
large as 20.1 or 24.7.
The results of the homogeneity of variance test will be discussed in the Results section
when we examine the findings produced by the ANOVA test.
Pretest Checklist Criterion 3—n
Again, as with the t test, technically, you can run an ANOVA test with an n of any size
in each group, but when the n is at least 30 in each group, the ANOVA is considered more
robust. We will see the n reported for each group in the Results section when we examine
the findings produced by the ANOVA test.

TEST RUN
To run an ANOVA test (for the most part, this is t test déjà vu time):
1. On the main screen, click on Analyze, Compare Means, One-Way ANOVA… (Figure
6.5).
Figure 6.5 Running an ANOVA Test
2. On the One-Way ANOVA menu, move the continuous variable that you wish to
analyze (score) into the Dependent List window, and move the variable that contains
the categorical variable that specifies the group (group) into the Factor window
(Figure 6.6).
Figure 6.6 The One-Way ANOVA Menu

3. Click on the Options… button, and on the One-Way ANOVA: Options menu, check
Descriptive, Homogeneity of variance test, and Means plot, then click on the
Continue button (Figure 6.7).
Figure 6.7 The One-Way ANOVA: Options Menu

This will take you back to the One-Way ANOVA menu.

4. Click on the Post Hoc… button.
5. This will take you to the One-Way ANOVA: Post Hoc Multiple Comparisons menu
(Figure 6.8).
Figure 6.8 The One-Way ANOVA: Options Menu
 6. If you were to run the ANOVA test without selecting a post hoc test, then all it
would return is a single p value; if that p is statistically significant, then that would
tell you that somewhere among the groups processed, the mean for at least one
group is statistically significantly different from the mean of at least one other
group, but it would not tell you specifically which group is different from which.
The post hoc test produces a table comparing the mean of each group with that of
each other group, along with the p values for each pair of comparisons. This will
become clearer in the Results section when we read the post hoc table.
As for which post hoc test to select, there are a lot of choices. To keep things simple, we
will focus on only two options: Tukey and Sidak. Tukey is an appropriate selection when
each group has the same ns; in this case, each group has an n of 30, so check the Tukey
checkbox, then click on the Continue button (this will take you back to the One-Way
ANOVA menu [Figure 6.6]). If the groups had different ns (e.g., n(Group1) = 40,
n(Group2) = 55, n(Group3) = 36), then you would select the Sidak post hoc test. If you do
not know the ns for each group in advance, just select either Tukey or Sidak and observe
the ns on the resulting report; if you guessed wrong, you can always go back and rerun the
analysis using the appropriate post hoc test.

ANOVA Post Hoc Summary

• If all groups have the same ns, then select Tukey.
• If the groups have different ns, then select Sidak.

7. On the One-Way ANOVA menu (Figure 6.6), click on the OK button, and the
ANOVA test will process.

RESULTS
Now that we have the results for the ANOVA test, we can complete the pretest checklist:
We still need answers regarding the homogeneity of variance and the ns for each group.
As we can see in Table 6.1, the homogeneity of variance test produced a significance (p)
value of .752; since this is greater than the α level of .05, this tells us that there are no
statistically significant differences among the variances of the score variable for the three
groups analyzed. In other words, the variances for score are similar enough among the
three groups: math as usual, Z-Math, and flashcards. We would conclude that the criteria
of the homogeneity of variance have been satisfied.
Table 6.1 Homogeneity of Variance Test Results

As for the final item on the pretest checklist, the n, we can see from the second column
in Table 6.2 that each group has an n of 30. This satisfies the n assumption, indicating that
the ANOVA test becomes more robust when the n for each group is at least 30.
Next, we look at the ANOVA table (Table 6.3) and find a significance (p) value of .029;
since this is less than the α level of .05, this tells us that there is a statistically significant
difference between the (three) group means for score, but unlike reading the results of the t
test, we are not done yet.
Table 6.2 Descriptive Statistics of Score for Math-as-Usual, Z-Math, and
Flashcard Groups

Table 6.3 ANOVA Test Results Comparing Score of Math-as-Usual, Z-Math, and
Flashcards Groups
 Remember that in the realm of the t test, only two groups are involved, so when it
comes to interpreting the p value, there is no question as to which group is different from
which—clearly, the mean from Group 1 is statistically significantly different from the
mean of Group 2, but when there are three or more groups, we need more information as
to which group is different from which; that is what the post hoc test answers.
Consider this: Suppose you have two kids, Ariel and Blake; you are in the living room,
and someone calls out from the den, “The kids are fighting again!” Since there are only
two kids, you immediately know that the fight involves Ariel and Blake. Now suppose
you have three kids—Ariel, Blake, and Cary—and the voice calls out, “The kids are
fighting again!” You can no longer simply know that the fight is between Ariel and Blake;
you need more information. Instead of just one possibility, there are now three possible
combinations of fighters:
Ariel vs. Blake
Ariel vs. Cary
Blake vs. Cary
Back to our example: The statistically significant p value observed in the ANOVA table
(Table 6.3) tells us that there is a statistically significant difference in the score variable
somewhere among the three groups (“The kids are fighting!”); the post hoc table will tell
us precisely where that difference is coming from (which pair of kids is fighting).
This brings us to the (Tukey post hoc) multiple comparisons table (Table 6.4). As with
the three kids fighting, in this three-group design, there are three possible pairs of
comparisons that we can assess in terms of (mean) score for the groups.
Comparison 1—math as usual vs. Z-Math
Comparison 2—math as usual vs. flashcards
Comparison 3—Z-Math vs. flashcards
We will use Table 6.2 (descriptives) and Table 6.4 (multiple comparisons) to analyze the
ANOVA test results. Table 6.2 lists the mean score for each of the three groups: μ(math as
usual) = 75.77, μ(Z-Math) = 80.97, and μ(flashcards) = 80.63. We will assess each of the
three pairwise score comparisons separately.
Comparison 1—Math as Usual vs. Z-Math
Table 6.4 first compares the mean score for the math-as-usual group with the mean score
for the Z-Math group, which produces a Sig.(nificance) (p) of .045. Since the p is less than
the .05 α level, this tells us that for score, there is a statistically significant difference
between math as usual and Z-Math.
Table 6.4 ANOVA Post Hoc Multiple Comparisons Table

Comparison 2—Math as Usual vs. Flashcards

The second score comparison in Table 6.5 is between math as usual and flashcards, which
produces a Sig.(nificance) (p) of .066. Since the p is greater than the .05 α level, this tells
us that for score, there is no statistically significant difference between math as usual and
flashcards.
Table 6.5 ANOVA Post Hoc Multiple Comparisons Table

Comparison 3—Z-Math vs. Flashcards

The third score comparison in Table 6.6 is between Z-Math and flashcards, which
produces a Sig.(nificance) (p) of .987. Since the p is greater than the .05 α level, this tells
us that for score, there is no statistically significant difference between Z-Math and
flashcards.
Table 6.6 ANOVA Post Hoc Multiple Comparisons Table
 This concludes the analysis of the multiple comparisons (post hoc) table. You have
probably noticed that we skipped analyzing half of the rows because there is a double
redundancy among the figures in the Sig. column. This is the kind of double redundancy
that you would expect to see in a typical two-dimensional table. For example, in a
multiplication table, you would see two 32s in the table because 4 × 8 = 32 and 8 × 4 = 32.
In the same sense, comparing math as usual with Z-Math renders the same p level (.45) as
comparing Z-Math with math as usual (Table 6.7). As you would expect, the Sig. column
in Table 6.8 also contains two .066 entries (math as usual vs. flashcards and flashcards vs.
math as usual) and two .987 entries (Z-Math vs. math as usual and math as usual vs. Z-
Math).
Table 6.7 ANOVA Post Hoc Multiple Comparisons Table Containing Double-
Redundant Sig. Figures

The ANOVA test can process any number of groups, provided the pretest criteria are
met. As the number of groups increases, the number of (multiple) comparisons increases
as well (see Table 6.8).
Table 6.8 Increasing Groups Substantially Increases ANOVA Post Hoc Multiple
Comparisons
 If you have a design that exceeds four groups, you can easily calculate the number of
(unique) pairwise comparisons the post hoc test will produce:

Unique Pairs Formula

G = Number of groups
Number of ANOVA post hoc unique pairs = G! ÷ (2 × (G − 2)!)

HYPOTHESIS RESOLUTION
To simplify the hypothesis resolution process, it is helpful to organize the findings in a
table and use an asterisk to flag statistically significant difference(s) (Table 6.9).
NOTE: SPSS does not generate this table directly; you can assemble this table by
gathering the means from the descriptives table (Table 6.2) and the p values from the Sig.
column in the multiple comparisons table (Table 6.4).
Table 6.9 Results of ANOVA for Score

*Statistically significant difference(s) (α = .05).

With this summary assembled, we can make clear decisions about our pending
hypotheses, which focus on determining if there is a better way to teach the multiplication
tables. Resolving these hypotheses will tell us which hypotheses came true:
H0: There is no statistically significant difference in the test scores among the three
groups.
H1: The Z-Math group outperforms the math-as-usual group.
H2: The flashcards group outperforms the math-as-usual group.
H3: The Z-Math group outperforms the flashcards group.
H4: The flashcards group outperforms the Z-Math group.
Since there is a statistically significant difference between at least two groups, we
would reject H0.
The Z-Math group did outperform the math-as-usual group, and that difference is
 statistically significant, so we accept H1.
The flashcards group did outperform the math-as-usual group, but that difference is not
statistically significant, so we would reject H2.
The Z-Math group did outperform the flashcards group, but that difference is not
statistically significant, so we would reject H3.
The flashcards group did not outperform the Z-Math group, but that difference is not
statistically significant, so we would reject H4.

DOCUMENT RESULTS
The data arranged in Table 6.10 could be used as a supplement in documenting the results
of this study:

A group of 90 fourth graders was randomly assigned to one of three groups to

determine which method of teaching the multiplication table is most effective: math
as usual, Z-Math, or flashcards with a volunteer math coach.
After 1 month, students in the Z-Math group learned an average of 80.97% of the
multiplication table, significantly outperforming students in the math-as-usual group;
those who received math as usual learned an average of 75.77% of the multiplication
table, and those who used flashcards learned an average of 80.63% (see Table 1).
Table 1 Results of ANOVA for Score

*Statistically significant difference(s) (α = .05).

Considering the above results report, you may feel compelled to comment that the .066
p level is approaching statistical significance. While the optimism may be commendable,
this semantic error is a common mistake. The term approaching wrongly implies that the
p level is a dynamic value—that it is somehow in motion and on its way to crossing the
.05 finish line, but this is not at all the case. The .066 p value is a static variable, meaning
that it is not in motion—the .066 p value is no more approaching .05 than it is
approaching .07. Think of the .066 p value as parked; it is not going anywhere, in the
same way that a parked car is neither approaching nor departing from the car parked in
front of it, no matter how close those cars are parked to each other. At best, one could state
that it (the .066 p value) is close to the .05 α level and that it would be interesting to
consider monitoring this variable should this experiment be repeated at some future point.
Here is a simpler way to think about this: 2 + 2 = 4, and 4 is not approaching 3 or 5; it is
just 4, and it is not drifting in any direction.
GOOD COMMON SENSE
When carrying the results of an ANOVA test into the real world, there are some practical
considerations to take into account. Using this example, suppose the goal for this school
district was to get multiplication test scores over 80%; the school principal may opt to set
aside the p values and simply reference the mean scores rendered by Z-Math and
flashcards, each of which exceeded 80%.
Cost-effectiveness may be another influential factor when it comes to applying these
findings in the real world. Suppose it costs $75.00 per student to implement Z-Math but
only $2.00 per student to provide each with a full set of multiplication flashcards. The
principal of the school, who is responsible for the budget, may inspect the means and p
values and make the following determination:
According to the p value, Z-Math clearly outperformed math as usual, but there is no statistically significant
difference (p = .987) between Z-Math (μ = 80.97) and flashcards (μ = 80.63). It would cost $6,750.00 ($75.00 × 90)
to provide Z-Math to our 90 fourth graders, but it would only cost $180.00 to provide them with flashcards. The
cost difference of $6,570.00 ($6,750.00 − $180 = $6,570.00) is not worth the .34-point improvement (80.97 − 80.63
= .34) in multiplication test scores that would be gained by using Z-Math over flashcards. Also, after the students
achieve 100% scores on their multiplication tests, the flashcards could be returned to the school and reused
(repeatedly) for future fourth graders, thereby reducing future math education costs.

Another issue involves the capacity of the ANOVA model. Table 6.8 and the
combinations formula (Unique pairs = G! ÷ (2 × (G − 2)!)) reveal that as more groups are
included, the number of ANOVA post hoc paired comparisons increases substantially. A 5-
group design would render 10 unique comparisons, 6 groups would render 15, and a 10-
group design would render 45 unique comparisons along with their corresponding p
values. While SPSS or any statistical software would have no problem processing these
figures, there would be some real-world challenges to address. Consider the pretest criteria
—in order for the results of an ANOVA test to be considered robust, there should be a
minimum n of 30 per group. Hence, for a design involving 10 groups, this would require
an overall n of at least 300. In addition, a 10-group study would render 45 unique pairwise
comparisons in the ANOVA post hoc table, which, depending on the nature of the data,
may be a bit unwieldy when it comes to interpretation and overall comprehension of the
results.

Key Concepts
• ANOVA
• Pretest checklist
Normality
Homogeneity of variance
n
• Post hoc tests
 Tukey
Sidak
• Hypothesis resolution
• Documenting results
• Good common sense

Practice Exercises
For Exercises 1 to 5:
• If you saved the data sets from Chapter 5 (Chapter 05 − Exercise 01 A.sav through
Chapter 05 – Exercise 05 B.sav), you will have less data entry to do; you will find
that the data for Groups 1 and 2 from the exercises in Chapter 5 are the same as in
this chapter. All you need to do is load the file(s) that you saved from the exercises
in Chapter 5, enter the data from the table below pertaining to additional group(s)
starting at the bottom of your loaded data set, and add the corresponding value
label(s) (on the Variable View screen) to account for the new group(s).
NOTE: The data sets for Questions 3 and 5 involve four groups.
• After finalizing Data Set A, repeat the exercise using Data Set B, which will
produce different findings.
For Exercises 6 to 10:
• Use the prepared SPSS data sets (download from www.sagepub.com/knapp).
• Load, process, and document your findings for Data Set A.
• Repeat each exercise using Data Set B (which will produce different results).
1. You want to determine if meditation can reduce resting pulse rate. Participants
were recruited and randomly assigned to one of three groups: Members of
Group 1 (the control group) will not meditate; members of Group 2 (the first
treatment group) will meditate for 30 minutes per day on Mondays,
Wednesdays, and Fridays over the course of 2 weeks; and members of Group 3
(the second treatment group) will meditate for 30 minutes a day 6 days a week,
Monday through Saturday. At the end, you gathered the resting pulse rates for
each participant.
Codebook—set up SPSS to contain two variables:
• group is a categorical variable (1 = no meditation, 2 = meditates 3 days, 3
= meditates 6 days).
• pulse is a continuous variable (number of beats per minute).
Data set—enter Data Set A.
a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the ANOVA test and document your findings (ns, means, and Sig. [p
value]).
d. Write an abstract under 200 words detailing a summary of the study, the
ANOVA test results, hypothesis resolution, and implications of your
findings.
 • Repeat this exercise using Data Set B.
2. You want to determine if pairing an incoming freshman with a sophomore in a
protégé-mentor relationship will enhance the freshman’s grade point average
(GPA). You recruit sophomores who are willing to mentor a student in their
major for their first term. You then recruit freshmen who are interested in having
a mentor. Freshmen who apply to this program will be sequentially assigned to
one of three groups: Group 1 will be the control group (no mentor), Group 2 will
be the in-person mentor group, and Group 3 will be the e-mentor group. Those
in the in-person mentor group are to meet in person once a week at a time of
their choosing, and those in the e-mentor group will communicate digitally at
least once a week. All freshmen, in each group, agree to submit their transcript
at the conclusion of the term.
Codebook—set up SPSS to contain two variables:
• group is a categorical variable (1 = no mentor, 2 = in-person mentor, 3 =
e-mentor).
• gpa is a continuous variable (the first-term GPA for each freshman).
Data set—enter Data Set A.
 a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the ANOVA test and document your findings (ns, means, and Sig. [p
value]).
d. Write an abstract under 200 words detailing a summary of the study, the
ANOVA test results, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Data Set B.
3. The Acme Company claims that its new reading lamp increases reading speed;
you want to test this. You will record how long (in seconds) it takes for
participants to read a 1,000-word essay. Participants will be randomly assigned
to one of four groups: Group 1 will be the control group; they will read the essay
using regular room lighting. Those in Group 2 will read the essay using the
Acme lamp. Those in Group 3 will read the essay using a generic reading lamp.
Those in Group 4 will read the essay using a flashlight.
Codebook—set up SPSS to contain two variables:
• group is a categorical variable (1 = room lighting, 2 = Acme lamp, 3 =
generic lamp, 4 = flashlight).
• seconds is a continuous variable (the time it takes to read the essay).
Data set—enter Data Set A.
 a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the ANOVA test and document your findings (ns, means, and Sig. [p
value]).
d. Write an abstract under 200 words detailing a summary of the study, the
ANOVA test results, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Data Set B.
4. You want to determine if chocolate enhances mood. Subjects will be recruited
and randomly assigned to one of three groups: Those in Group 1 will be the
control group and will eat their regular diet. Those in Group 2 will eat their
usual meals and have a piece of chocolate at breakfast, lunch, and dinner over
the course of a week. Those in Group 3 will eat their meals as usual and have
two pieces of chocolate at breakfast, lunch, and dinner over the course of a
week. At the end of the week, all participants will complete the Acme Mood
Scale (1 = extremely bad mood… 100 = extremely good mood).
Codebook—set up SPSS to contain two variables:
• group is a categorical variable (1 = no chocolate, 2 = chocolate [1 per
meal], 3 = chocolate [2 per meal]).
• mood is a continuous variable (score on the Acme Mood Scale).
Data set—enter Data Set A.
 a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the ANOVA test and document your findings (ns, means, and Sig. [p
value]).
d. Write an abstract under 200 words detailing a summary of the study, the
ANOVA test results, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Data Set B.
5. You want to find out if music enhances problem solving. Subjects will be
recruited and randomly assigned to one of four groups: Those in Group 1 will
serve as the control group and will be given a standard 100-piece jigsaw puzzle
to solve in a quiet room. Participants in Group 2 will be given the same puzzle
to assemble but instead of silence, there will be classical music playing at a soft
volume (30 decibels [dB]) in the room. Participants in Group 3 will be given the
same puzzle to assemble using the same classical music, but the music will be
played at a moderate volume (60 dB). Participants in Group 4 will be given the
same puzzle to assemble using the same classical music, but the music will be
played at a loud volume (90 dB). You will record the time (in seconds) that it
takes for each person to complete the puzzle.
Codebook—set up SPSS to contain two variables:
• group is a categorical variable (1 = no music, 2 = music at 30 dB, 3 =
music at 60 dB, 4 = music at 90 dB).
• seconds is a continuous variable (how long it took to finish the puzzle).
Data set—enter Data Set A.
 a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the ANOVA test and document your findings (ns, means, and Sig. [p
value]).
d. Write an abstract under 200 words detailing a summary of the study, the
ANOVA test results, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Data Set B.
6. You want to determine if watching a video of a comedy with a laugh track
enhances enjoyment. Subjects will be recruited and randomly assigned to one of
three groups: Those in Group 1 (the control group) will watch the video without
the laugh track, those assigned to Group 2 will watch the same video with the
sound(s) of a 50-person audience included in the soundtrack, and those assigned
to Group 3 will watch the same video with the sound(s) of a 100-person
audience included in the soundtrack. Each participant will watch the video
individually; no others will be present in the room. Immediately following the
video, each participant will be asked to rate how enjoyable the show was on a
scale of 1 to 5 (1 = not very enjoyable… 5 = very enjoyable).
Codebook—The data set contains two variables:
• group is a categorical variable (1 = no laugh track, 2 = laugh track at 50, 3
= laugh track at 100).
• enjoy is a continuous variable (1 = not very enjoyable… 5 = very
enjoyable).
Data set—load Chapter 06 – Exercise 06 A.sav.
a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
 c. Run the ANOVA test and document your findings (ns, means, and Sig. [p
value]).
d. Write an abstract under 200 words detailing a summary of the study, the
ANOVA test results, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Chapter 06 – Exercise 06 B.sav.
7. In an effort to determine the effectiveness of light therapy to alleviate
depression, you recruit a group of subjects who have been diagnosed with
depression. The subjects are randomly assigned to one of three groups: Group 1
will be the control group—members of this group will receive no light therapy.
Members of Group 2 will get light therapy for 1 hour on even-numbered days
over the course of 1 month. Members of Group 3 will get light therapy every
day for 1 hour over the course of 1 month. After 1 month, all participants will
complete the Acme Mood Scale, consisting of 10 questions; this instrument
renders a score between 1 and 100 (1 = extremely bad mood… 100 = extremely
good mood).
Codebook—the data set contains two variables:
• group is a categorical variable (1 = no light therapy, 2 = light therapy:
even days, 3 = light therapy: every day).
• mood is a continuous variable (1 = extremely bad mood… 100 =
extremely good mood).
Data set—load Chapter 06 – Exercise 07 A.sav.
a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the ANOVA test and document your findings (ns, means, and Sig. [p
value]).
d. Write an abstract under 200 words detailing a summary of the study, the
ANOVA test results, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Chapter 06 – Exercise 07 B.sav.
8. It is thought that exercising early in the morning will provide better energy
throughout the day. To test this idea, subjects are recruited and randomly
assigned to one of three groups: Members of Group 1 will constitute the control
group and not be assigned any walking. Members of Group 2 will walk from
7:00 to 7:30 a.m., Monday through Friday, over the course of 30 days. Members
of Group 3 will walk from 7:00 to 8:00 a.m., Monday through Friday, over the
course of 30 days. At the conclusion of the study, each subject will answer the
10 questions on the Acme End-of-the-Day Energy Scale. This instrument
produces a score between 1 and 100 (1 = extremely low energy… 100 =
 extremely high energy).
Codebook—the data set contains two variables:
• group is a categorical variable (1 = no walking, 2 = walking: 30 minutes,
3 = walking: 60 minutes).
• energy is a continuous variable (1 = extremely low energy… 100 =
extremely high energy).
Data set—load Chapter 06 – Exercise 08 A.sav.
a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the ANOVA test and document your findings (ns, means, and Sig. [p
value]).
d. Write an abstract under 200 words detailing a summary of the study, the
ANOVA test results, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Chapter 06 – Exercise 08 B.sav.
9. Clinicians at a nursing home facility want to see if giving residents a plant to
tend to will help lower depression. To test this idea, the residents are randomly
assigned to one of three groups: Those assigned to Group 1 will serve as the
control group and will not be given a plant. Members of Group 2 will be given a
small bamboo plant along with a card detailing care instructions. Members of
Group 3 will be given a small cactus along with a card detailing care
instructions. After 90 days, all participants will complete the Acme Depression
Scale, a 10-question instrument that renders a score between 1 and 100 (1 = low
depression… 100 = high depression).
Codebook—the data set contains two variables:
• group is a categorical variable (1 = no plant, 2 = bamboo, 3 = cactus).
• depress is a continuous variable (1 = low depression… 100 = high
depression).
Data set—load Chapter 06 – Exercise 09 A.sav.
a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the ANOVA test and document your findings (ns, means, and Sig. [p
value]).
d. Write an abstract under 200 words detailing a summary of the study, the
ANOVA test results, hypothesis resolution, and implications of your
findings.
 • Repeat this exercise using Chapter 06 – Exercise 09 B.sav.
10. During flu season, the administrators at a walk-in health clinic want to
determine if providing patients with a pamphlet or a video will increase their
receptivity to flu shots. Each patient will be given a ticket at the check-in desk
with a 1, 2, or 3 on it; the tickets will be issued in (repeating) sequence. Once
escorted to the exam room, patients with a number 1 ticket will serve as control
participants and will not be offered any flu shot informational material. Patients
with a number 2 ticket will be given a flu shot information pamphlet. Patients
with a number 3 ticket will be shown a brief video covering the same
information as contained in the pamphlet. At the end of the day, the charts were
reviewed and three entries were made in the database: total number of flu shots
given to patients in Group 1, total number of flu shots given to patients in Group
2, and the total number of flu shots given to patients in Group 3.
Codebook—the data set contains two variables:
• group is a categorical variable (1 = nothing, 2 = flu shot pamphlet, 3 = flu
shot video).
• shots is a continuous variable (number of flu shots given in a day [for
each group]).
Data set—load Chapter 06 – Exercise 10 A.sav.
a. Write the hypotheses.
b. Run each criterion of the pretest checklist (normality, homogeneity of
variance, and n) and discuss your findings.
c. Run the ANOVA test and document your findings (ns, means, and Sig. [p
value]).
d. Write an abstract under 200 words detailing a summary of the study, the
ANOVA test results, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Chapter 06 – Exercise 10 B.sav.
 CHAPTER 7
Paired t Test

Things never happen the same way twice.

—C. S. Lewis
LEARNING OBJECTIVES
Upon completing this chapter, you will be able to:
• Determine when it is appropriate to run a paired t test
• Verify that the data meet the criteria for paired t test processing: normality of
differences
• Order a paired t test
• Interpret test results
• Resolve the hypotheses
• Document the results in plain English
• Calculate and document the ∆% formula
DIGITAL LEARNING RESOURCES
The tutorial video and data sets for this chapter are at www.sagepub.com/knapp
• Paired t test pretest checklist video
 • Paired t test run video
• SPSS files
OVERVIEW
The t test and ANOVA test were appropriate for conducting research using a classic
experimental model, which involves random assignment of participants to a control group
and at least one other (treatment) group. There will be times when such rigorous designs
are not feasible due to limited resources (e.g., low n, limited staff, lack of facilities, budget
constraints, etc.). The paired t test provides an alternate approach that can be used to test
the effectiveness of an implementation using a single group that does not require a sizable
n.
Pretest/Posttest Design
The design associated with the paired t test is typically referred to as a pretest/posttest
design, also known as a simple time-series design, or O-X-O design (O = observation, X =
treatment) (Figure 7.1).
This design consists of one group and three steps:
Step 1—Pretest: Begin by gathering a quantitative metric, and attach each participant’s
name or ID to the score. The score needs to be a continuous variable. This could be an
existing score or a test that you administer. This will be the pretest score, sometimes
referred to as the baseline score, indicating the level that each participant was at prior to
exposing him or her to the treatment. Essentially, each subject acts as his or her own
control group.
Figure 7.1 Pretest/Posttest Design

Step 2—Treatment: Execute the treatment (e.g., intervention, change in condition,

training, etc.).
Step 3—Posttest: Administer the same test that was used in Step 1 (pretest), or gather
the (same) metric and record the participant’s name or ID with the score.
The paired t test analysis presents three critical pieces of information: (1) the mean
pretest score, (2) the mean posttest score, and (3) the p value. If the p value is less than the
specified α level (.05), then this indicates that there is a statistically significant difference
between the pretest score and the posttest score, suggesting that the treatment made an
impact.
EXAMPLE
Acme Industries has 10 people on its production staff. The supervisor is looking for an
affordable way to boost productivity and is considering offering unlimited free coffee to
the staff.
RESEARCH QUESTION
The question is: Will providing the staff with unlimited free coffee increase productivity?
GROUPS
As stated earlier, part of the utility of the paired t test is that it can function with just one
group. The (one) group will consist of the 10 members of the production staff.
PROCEDURE
Step 1—Pretest: First, the manager will count the total number of units that each staff
member processed last week; this will constitute the pretest score.
Step 2—Treatment: The manager will announce that the coffee machine has been set
to free access; the staff can have as much coffee as they want at no charge.
Step 3—Posttest: At the end of the week, the manager will gather the total number of
units that each staff member processed (that week); this will constitute the posttest
score.
HYPOTHESES
H0: Providing the staff with free coffee will have no impact on productivity.
H1: Providing the staff with free coffee will enhance productivity.

DATA SET
Use the following data set: Chapter 07 – Example 01 – Paired T Test.sav.

PRETEST CHECKLIST
Pretest Checklist Criterion 1—Normality of Differences
To run a paired t test, you must satisfy only one pretest criterion: The difference between
the pretest scores and the posttest scores must be normally distributed. This process
involves two steps:
(1) We will have SPSS compute a new variable (diff), which will contain the
difference between the pretest scores and the posttest scores (diff = posttest –
pretest).
(2) We will run a histogram with a normal curve for diff and inspect the curve for
normality.
Begin by selecting Transform, Compute Variable… (Figure 7.2).
Figure 7.2 Select Transform, Compute Variable…
 This will take you to the Compute Variable menu (Figure 7.3). Enter diff in the Target
Variable box. Enter posttest – pretest in the Numeric Expression box. You can type in the
variables posttest and pretest, double-click on them, or use the arrow key to copy them
from the left box to the right box.
NOTE: For this test, you can enter posttest – pretest or pretest – posttest; either is fine.
Click the OK button to process this menu.
At this point, let us review the data on the Data View screen (Figure 7.4):
Column 1 contains the name of each staff member; this is a string (alphanumeric)
variable.
Columns 2 and 3 contain the pretest and posttest scores for each staff member.
Figure 7.3 Compute Variable Menu

Column 4 contains the diff variable; this variable, as well as the data in it, was created
as we specified on the Compute Variable menu (diff = posttest – pretest).
It is now time to generate the histogram with a normal curve for the diff variable. This is
the same procedure used as part of the pretest checklist for the t test and ANOVA. For
more details on this procedure, please refer to Chapter 4 (“SPSS—Descriptive Statistics:
Continuous Variables (Age)”); see the star ( ) icon on page 61.
Figure 7.4 Data View
Alternatively, the following steps will produce a histogram with a normal curve for diff:
1. From the main screen, select Analyze / Descriptive Statistics / Frequencies…; this
will take you to the Frequencies menu.
2. On the Frequencies menu, move diff from the left window to the right (Variables)
window.
3. Click on the Charts… button; this will take you to the Charts menu.
4. Click on the Histograms button, and check the Show normal curve on histogram
checkbox.
5. Click on the Continue button; this will return you to the Frequencies menu.
6. Click on the OK button, and the system will produce a histogram with a normal
curve for the diff variable (Figure 7.5).
Figure 7.5 Histogram With Normal Curve for diff
 The normal curve for diff (Figure 7.5) presents as a reasonably symmetrical bell shape
—hence, we would say that the difference between the pretest and posttest scores meets
the criteria of normality.

TEST RUN
To run the paired t test, from the main screen, click on Analyze, Compare Means, Paired-
Samples T Test… . This will take you to the paired-samples t test (Figure 7.6).
Figure 7.6 Run Paired-Samples t Test

Copy the pretest variable from the left window to the right window (under Variable1),
then copy the posttest variable from the left window to the right window (under
Variable2), and then click on the OK button.
RESULTS
The results of the paired t test are read from two tables: The paired-samples statistics test
(Table 7.1) reports the means for each group: μ(pretest) = 36.40 and μ(posttest) = 40.10.
The table also shows the corresponding ns and standard deviations.
The paired-samples test (Table 7.2) focuses on the difference between the pretest and
posttest scores. Column 2 shows that there is a (−)3.700-point difference between the
means (36.40 − 40.10 = −3.700). The last column (Sig.) shows that the p value is .030 for
this comparison; since .030 is less than the specified α level of .05, we would conclude
that there is a statistically significant difference between the pretest and posttest scores.
Table 7.1 Paired Samples (Summary) Statistics for Pretest and Posttest

Table 7.2 Paired Samples Test Results

HYPOTHESIS RESOLUTION
H0: Providing the staff with free coffee will have no impact on productivity.
H1: Providing the staff with free coffee will enhance productivity.
Since the p value (.030) is less than the specified α level (.05), this suggests that the 3.7-
point increase in productivity (from 36.4 to 40.1) is statistically significant. In terms of the
hypotheses, we would reject H0 and not reject H1.

DOCUMENT RESULTS
Since the paired t test is a fairly straightforward process, the documentation is typically
concise:
We compared weekly productivity figures the week before and after implementing a free unlimited coffee program.
Our 10 staff members processed an average of 3.7 additional cases per week with the coffee (up from 36.4 to 40.1),
which is statistically significant (p = .03, α = .05).

Another useful way of documenting such change is the Δ% formula (Δ is the Greek
letter delta, which symbolizes change). The Δ% formula is simple, yet expressive; you can
run it on any calculator by plugging in two variables: the old value (pretest mean = 36.4)
and the new value (posttest mean = 40.1) (Table 7.3).
Table 7.3 Δ% Formula Computes Change Percentage

In terms of documentation, you could include the Δ% = 10.16 as-is or express it

verbosely: “. . . we observed a 10.16% increase in weekly productivity, from an average of
36.4 cases per week to 40.1.”
GOOD COMMON SENSE
For simplicity, the example used in this chapter focused exclusively on weekly
productivity, but you may opt to gather additional pre/post data (e.g., job satisfaction
survey scores, attendance scores, minutes late, etc.) and conduct separate paired t tests on
those metrics as well. Such findings may shed light on other relevant variables; since this
intervention involves introducing free unlimited coffee, it may be wise to include
additional pre/post metrics that gather data on other possibly relevant topics such as
irritability, coworker relations, workplace satisfaction, and so on.
When opting for the pretest/posttest design and corresponding paired t test, the potential
for historical confounds must be considered. In a classic experimental design, the control
group and treatment group(s) are being processed during the same timeframe, but when
using the paired t test, the pretest (control/baseline) data and posttest (treatment) data are
collected at different time points. This time difference may involve minutes, hours, weeks,
or even months, depending on the study design. During this time, history does not stand
still; things outside your investigation can change over time either in the individual or in
the environment (e.g., personal issues, sociopolitical events, season, etc.), which could
potentially affect the pretest or posttest scores.
In addition, in some cases, there is a chance that participants may show improvement in
the posttest, partly because they are familiar with it—they remember it from the pretest
phase.

Key Concepts
• Paired t test
• Paired t test designs (synonyms):
Pretest/treatment/posttest
Pre/post design
Simple time-series design
O-X-O design
• Histogram with a normal curve
• Paired t test with multiple metrics
• Historical confound
• Δ%

Practice Exercises
1. The Acme Company produces Monster Spray, a pleasant fragrance that parents can
use to help relax children who are afraid of nighttime monsters. A group of parents
who use the spray has been recruited to find out how effective the spray is in
calming monster-anxious children. Prior to spraying, the parent will ask the child to
rate his or her anxiety on a 1 to 10 scale (1 = not afraid at all… 10 = very afraid);
the parent will ask the same anxiety scale question after spraying. The parent will
record both the before and after (spraying) numbers.
Codebook—set up SPSS to contain three variables:
• ID is an alphanumeric (string) variable (the participant’s ID).
• pretest is a continuous variable (level of anxiety before spraying).
• posttest is a continuous variable (level of anxiety after spraying).
Data set—enter Data Set A.
 a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality for posttest − pretest) and
discuss your findings.
c. Run the paired t test and document your findings (means and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
paired t test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Data Set B.
2. Prior to a Heart Health presentation, you administer a survey asking participants to
indicate how many times they used the stairs (as opposed to the elevator) in the past
week. A week after the lecture, you resurvey the attendees.
Codebook—set up SPSS to contain three variables:
• ID is an alphanumeric (string) variable (the participant’s ID).
• pretest is a continuous variable (number of times per week the participant
used the stairs before the lecture).
• posttest is a continuous variable (number of times per week the participant
used the stairs after the lecture).
Data set—enter Data Set A.
 a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality for posttest − pretest) and
discuss your findings.
c. Run the paired t test and document your findings (means and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
paired t test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Data Set B.
3. An English teacher recognizes that students already know how to spell some of the
words assigned for the weekly spelling test. This teacher wants to discover how
many new words students learn. The usual assignment is 25 words per week. At the
beginning of the month, the teacher administers a spelling test consisting of all 100
words that will be assigned over the month. The teacher then administers the same
100-word test at the end of the month.
Codebook—set up SPSS to contain three variables:
• ID is an alphanumeric (string) variable (the student’s ID).
• pretest is a continuous variable (number of words spelled correctly on the
test given at the beginning of the month).
• posttest is a continuous variable (number of words spelled correctly on the
test given at the end of the month).
Data set—enter Data Set A.
 a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality for posttest − pretest) and
discuss your findings.
c. Run the paired t test and document your findings (means and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
paired t test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Data Set B.
4. The staff at a mental health clinic wants to determine if their current form of short-
term therapy substantially reduces depression. Prior to treatment, each patient will
be asked to complete the Acme Depression Inventory (ADI), which renders a score
from 0 to 75 (0 = low depression… 75 = high depression). Patients will also be
asked to complete the same instrument at the conclusion of their final appointment.
Codebook—set up SPSS to contain three variables:
• ID is an alphanumeric (string) variable (the patient’s initials).
• pretest is a continuous variable (ADI score before therapy).
• posttest is a continuous variable (ADI score after therapy).
Data set—enter Data Set A.
 a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality for posttest − pretest) and
discuss your findings.
c. Run the paired t test and document your findings (means and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
paired t test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Data Set B.
5. The staff of the Physical Education Department wants to know if providing a single
15-minute individual coaching session with an expert bowler will enhance students’
bowling scores. Each participant will bowl one game, during which time the coach
will unobtrusively observe his or her bowling style. Then, the coach provides the
15-minute coaching session. Finally, the student bowls a second game. The scores
from both games are recorded.
Codebook—set up SPSS to contain three variables:
• ID is an alphanumeric (string) variable (the bowler’s first name).
• pretest is a continuous variable (score from first game).
• posttest is a continuous variable (score from second game).
Data set—enter Data Set A.
 a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality for posttest − pretest) and
discuss your findings.
c. Run the paired t test and document your findings (means and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
paired t test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Data Set B.
6. The administrators of a school want to determine if issuing students laptop
computers will enhance their grades. At the beginning of the academic term, each
student in a class was issued a laptop computer. At the end of the term, the grade
point average (GPA) for each student was gathered at two time points: the GPA for
the term before issuing the laptop and the current GPA.
Codebook—the data set contains three variables:
• ID is an alphanumeric (string) variable (the student’s ID).
• pretest is a continuous variable (GPA for term before issuing computers).
• posttest is a continuous variable (GPA for term after issuing computers).
Data set—load Chapter 07 − Exercise 06 A.sav.
a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality for posttest – pretest) and
 discuss your findings.
c. Run the paired t test and document your findings (means and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
paired t test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Chapter 07 – Exercise 06 B.sav.
7. Acme Brand allergy medicine claims that its product reduces allergy-related
sneezing. To test this claim, you recruit a group of allergy sufferers who are not
currently taking any medications for their allergies and ask them to count the
number of times they sneeze per day. The next day, each participant takes the Acme
allergy medicine in the morning as directed and keeps a (separate) sneeze tally for
that day too.
Codebook—the data set contains three variables:
• ID is an alphanumeric (string) variable (the participant’s ID).
• pretest is a continuous variable (number of sneezes per day before taking
allergy medicine).
• posttest is a continuous variable (number of sneezes per day after taking
allergy medicine).
Data set—load Chapter 07 – Exercise 07 A.sav.
a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality for posttest – pretest) and
discuss your findings.
c. Run the paired t test and document your findings (means and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
paired t test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Chapter 07 – Exercise 07 B.sav.
8. An herbalist interested in natural remedies wants to assess the effectiveness of the
Tutsle root (a fictitious plant) in reducing fever. The investigator recruits a group of
patients who have fever and records the temperature for each person. Next, each
participant is given one cup of Tutsle root tea. An hour later, the investigator takes
each participant’s temperature again.
Codebook—the data set contains three variables:
• ID is an alphanumeric (string) variable (the participant’s ID).
• pretest is a continuous variable (participant’s temperature before drinking
Tutsle root tea).
• posttest is a continuous variable (participant’s temperature 1 hour after
drinking Tutsle root tea).
Data set—load Chapter 07 – Exercise 08 A.sav.
 a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality for posttest – pretest) and
discuss your findings.
c. Run the paired t test and document your findings (means and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
paired t test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Chapter 07 – Exercise 08 B.sav.
9. In an effort to discover ways to boost morale, an investigator wants to assess the
effects that chocolate has on attitude. The researcher recruits a group of participants
and has them complete the Acme Attitude Survey (AAS), which renders a score
ranging from 0 to 100 (0 = very bad attitude… 100 = very good attitude). After
gathering the (pre)tests, the researcher serves each participant a generous slice of
chocolate fudge cake. One hour later, the researcher administers the AAS a second
time.
Codebook—the data set contains three variables:
• ID is an alphanumeric (string) variable (the participant’s ID).
• pretest is a continuous variable (AAS score before eating chocolate).
• posttest is a continuous variable (AAS score 1 hour after eating chocolate).
Data set—load Chapter 07 – Exercise 09 A.sav.
a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality for posttest – pretest) and
discuss your findings.
c. Run the paired t test and document your findings (means and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
paired t test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Chapter 07 – Exercise 09 B.sav.
10. A political consultant wants to judge the impact of a speech. The consultant recruits
a group of registered voters and asks them to indicate their voting intentions on a 1
to 7 scale (1 = absolutely will not vote for this candidate… 7 = absolutely will vote
for this candidate). Next, the candidate delivers the speech. Finally, the consultant
readministers the one-question survey to each participant.
Codebook—the data set contains three variables:
• ID is an alphanumeric (string) variable (the participant’s ID).
• pretest is a continuous variable (voter’s score before speech).
• posttest is a continuous variable (voter’s score after speech).
Data set—load Chapter 07 – Exercise 10 A.sav.
a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality for posttest – pretest) and
discuss your findings.
c. Run the paired t test and document your findings (means and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
paired t test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Chapter 07 – Exercise 10 B.sav.
 CHAPTER 8
Correlation and Regression

He who laughs most, learns best.

—John Cleese
LEARNING OBJECTIVES
Upon completing this chapter, you will be able to:
• Determine when it is appropriate to run regression and correlational analyses
• Interpret the direction and strength of a correlation
• Verify that the data meet the criteria for running regression and correlational
analyses: normality, linearity, and homoscedasticity
• Order a regression analysis: correlation and scatterplot with regression line
• Interpret the test results
• Resolve the hypotheses
• Document the results in plain English
 • Understand the criteria for causation: association/correlation, temporality, and
nonspurious
• Differentiate between correlation and causation
DIGITAL LEARNING RESOURCES
The tutorial video and data sets for this chapter are at: www.sagepub.com/knapp
• Regression and correlation pretest checklist video
• Regression and correlation test run video
• SPSS files
OVERVIEW
Regression involves assessing the correlation between two variables. Before proceeding,
let us deconstruct the word correlation: The prefix co means two—hence, correlation is
about the relationship between two things. Regression is about assessing the correlation
between two continuous variables.
Correlation involving two variables, sometimes referred to as bivariate correlation, is
notated using the lowercase r and has a value between −1 and +1. Correlations have two
primary attributes: direction and strength.
Direction is indicated by the sign of the r value: − or +. Positive correlations (r =
0… +1) emerge when the two variables move in the same direction. For example, we
would expect that low homework hours would correlate with low grade point average
(GPA), just as we would expect that high homework hours would correlate with high
GPA. Negative correlations (r = −1… 0) emerge when the two variables move in different
directions. For example, we would expect that high alcohol consumption would correlate
with low GPA, just as we would expect that low alcohol consumption would correlate
with high GPA (see Table 8.1).

Table 8.1 Correlation Direction Summary

Figure 8.1 Correlation Strength Summary

 Strength is indicated by the numeric value. A correlation wherein the r is close to 0 is
considered weaker than those nearer to −1 or +1 (see Figure 8.1). Continuing with the
prior example, we would expect to find a strong negative correlation between alcohol
consumption and GPA (e.g., r = −.81); conversely, we would expect to find a strong
positive correlation between homework hours and GPA (e.g., r = +.81). However, we
would not expect that a variable such as height would have much to do with academic
performance, and hence we would expect to find a relatively weak correlation between
height and GPA (e.g., r = +.02 or r = −.02).
The concepts of correlation direction and strength will become clearer as we examine
the test results, specifically upon inspecting the graph of the scatterplot with the regression
line in the Results section.
EXAMPLE
One might intuit that there is a correlation between the number of hours a student spends
doing homework per week and his or her GPA. An instructor wants to investigate this.
RESEARCH QUESTION
Is there a statistically significant correlation between homework hours per week and GPA?
GROUPS
Bivariate regression/correlation involves only one group, but two different continuous
variables are gathered for each participant: In this case, the variables are (1) homework
hours per week and (2) GPA.
Notice that in correlation analysis, you can mix apples and oranges; homework hours is
a measure of time, whereas GPA is a measure of academic performance. The only
constraints in this respect are that the two metrics must both be continuous variables, and
of course, the comparison needs to inherently make sense.
PROCEDURE
The instructor will record two values for each student for a week: total number of hours
spent on homework (as reported by the student’s parent) and the GPA for that week based
on a multisubject test given on Friday afternoon.
HYPOTHESES
H0: There is no correlation between homework hours and GPA
H1: There is a positive correlation between homework hours and GPA.
 H2: There is a negative correlation between homework hours and GPA.

DATA SET
Use the following data set: Chapter 08 − Example 01 – Correlation & Regression.sav.
For clarity, notice that the name of each participant is coded to demonstrate that
homework hours and GPA data are recorded for each subject.

PRETEST CHECKLIST
The pretest criteria for running a correlation/regression involve checking the data for (1)
normality, (2) linearity, and (3) homoscedasticity (pronounced hoe-moe-skuh-daz-tis-city).
Pretest Checklist Criterion 1—Normality
The two variables involved in the correlation/regression each need to be inspected for
normality. To do this, generate separate histograms with normal curves for homework and
gpa (this is similar to the steps used to check for normality when using the t test and
ANOVA).
For more details on this procedure, please refer to Chapter 4 (“SPSS—Descriptive
Statistics: Continuous Variables (Age)”); see the star ( ) icon on page 61. Instead of
processing age, load the two variables: homework and gpa. Alternatively, the following
steps will produce a histogram with a normal curve for homework and GPA:
1. From the main screen, select Analyze / Descriptive Statistics / Frequencies… ; this
will take you to the Frequencies menu.
2. On the Frequencies menu, move homework and gpa from the left window to the
right (Variables) window. With the data in this configuration, you can order both
histograms at the same time.
3. Click on the Charts… button; this will take you to the Charts menu.
4. Click on the Histograms button, and check the Show normal curve on histogram
checkbox.
5. Click on the Continue button; this will return you to the Frequencies menu.
Figure 8.2 Histogram With a Normal Curve for Homework
 Figure 8.3 Histogram With a Normal Curve for GPA

6. Click on the OK button, and the system will produce (two) histograms with normal
curves for homework and GPA (Figures 8.2 and 8.3).
The curves on each histogram are reasonably symmetrically bell shaped; there is no
notable skewing, and hence these criteria are satisfied.
Linearity and homoscedasticity are both graphical in nature; they are read off the
scatterplot with the regression line. We will observe and discuss the findings for these two
remaining criteria in the Results section.

TEST RUN
The test run for correlation and regression is a two-part process; first we will process the
correlation table, which will render the correlation value (r) and the corresponding p
value. Next, we will order a scatterplot, which will provide a clear graph showing the
paired points from both variables plotted on an X,Y-axes along with the regression line,
sometimes referred to as a trend line, which can be thought of as the average pathway
through the points.
Correlation
1. To run a correlation, starting from the main screen click on Analyze (Figure 8.4).
 2. Click on Correlate.
3. Click on Bivariate… ; this will take you to the Correlate menu (Figure 8.5).
Figure 8.4 Accessing the Correlation Menu: Analyze, Correlate, Bivariate…

Figure 8.5 Accessing the Correlation Menu: Analyze, Correlate, Bivariate…

4. On the Correlate menu (Figure 8.5), move the GPA and homework variables from
the left window to the right (Variables) window.
5. Click the OK button, and the correlation will process. For now, set aside the
correlations table that is produced; we will interpret it in the Results section.

Scatterplot With Regression Line

NOTE: SPSS graphics processing menus tend to differ across versions. If these
instructions do not fit your version of the software, use the Help menu to guide you to
order a scatterplot with the regression line. Indicate that you want the homework variable
on the X-axis and the GPA variable on the Y-axis.
6. To order a scatterplot with a regression line, from the main menu, click on Graphs
 (Figure 8.6).
7. Click on Chart Builder… . This will bring you to the Chart Builder menu (Figure
8.7).
Figure 8.6 Accessing the Chart Builder Menu: Graphics, Chart Builder…

Figure 8.7 Chart Builder Menu

8. In the Choose from: list, click on Scatter/Dot (Figure 8.7).

9. Double-click on the (circled) first choice, or click and drag this icon to the
Chart preview uses example data window.
10. Click and drag homework from the Variables window to the X-Axis? box (Figure
8.8).
11. Click and drag gpa from the Variables window to the Y-Axis? box.
12. Click on the OK button, and the system will produce the scatterplot.
Figure 8.8 Chart Builder Menu—Assign homework to X-axis and gpa to Y-axis
 13. When the scatterplot emerges, you will need to order the regression line: In the
Output window, double-click on the scatterplot. This will bring you to the Chart
Editor (Figure 8.9).
Figure 8.9 Chart Editor Menu—Click on Add Fit Line at Total Icon to Include the
Regression Line on the Scatterplot

14. Click on the Add Fit Line at Total icon to include the regression line on the
scatterplot.
15. When you see the regression line emerge on the scatterplot, close the Chart Editor
and you will see that the regression line is now included on the scatterplot in the
Output window.

RESULTS
In this section, we will begin by explaining the two elements present on the scatterplot: the
points and the regression line. Next, we will finalize the two remaining pretest criteria
(linearity and homoscedasticity), and finally, we will discuss the overall meaning of the
scatterplot and correlation findings.
Scatterplot Points
The coordinates of each point on the scatterplot is derived from the two variables:
homework and GPA for each record (individual).
Figure 8.10 Source Data for Scatterplot
 The first record of the data set shows that Madeline did homework for 1.5 hours and her
GPA was 2.11 (Figure 8.10). When we ordered the histogram, we placed homework on the
X-axis and gpa on the Y-axis—hence, Madeline’s point on the scatterplot is at coordinates
(1.50, 2.11), Mark’s point on the scatterplot is at (3.50, 2.22), and so on.
Scatterplot Regression Line
The simplest way to conceive the regression line, without presenting the formula, is to
think of it as the average straight-line pathway through the cloud of points, based on their
positions. Just as the descriptive statistics provide a summary of a single variable, the
regression line provides a sort of graphical summary of two variables—in this case,
homework and GPA.
NOTE: We already covered Pretest Checklist Criterion 1—normality of distribution of the
two variables, homework and GPA, via the histograms with normal curves; now we will
assess the remaining pretest checklist criteria: (2) linearity and (3) homoscedasticity.

Pretest Checklist Criterion 2—Linearity

The points on the scatterplot should form a relatively straight line (Figure 8.11); the
regression line should take a middle-of-the-road path through the cloud of points. If the
overall shape of the points departs into some other shape(s) that is not conducive to
drawing a straight (regression) line through (Figure 8.12), then this would constitute a
violation of the linearity assumption.
Figure 8.11 Linearity Satisfied
 Figure 8.12 Linearity Violated

Pretest Checklist Criterion 3—Homoscedasticity

Homoscedasticity pertains to the density of the points along the regression line. The
criterion of homoscedasticity is satisfied when the cloud of points is densest in the middle
and tapers off at the ends (Figure 8.13). The rationale for this distribution of points on the
scatterplot follows the same notion as the shape of the normal curve of the histogram—the
majority of the values are gathered around the mean, which accounts for the height of the
normal bell-shaped curve on the histogram, whereas the tapered tails signify that there are
considerably fewer very low and very high values. The positions of the points on the
scatterplot are derived from the same data that rendered the normally distributed
histograms for the two variables (Figures 8.13 and 8.14), so it follows that the middle of
the cloud should contain considerably more points (and be denser) than the ends.
Figure 8.13 Homoscedasticity Satisfied
 Figure 8.14 Homoscedasticity Violated

Correlation
Table 8.2 shows a positive correlation (r = .881) between homework and GPA, with a p
value of .000. Actually, there is no such thing as a p value that equals zero; in this case, p
= .0000000001385. When the p value is less than .01, it is typically notated as p < .001.
Since the p value is less than the α level of .05, we would say that there is a statistically
significant correlation between homework (hours) and GPA. The positive correlation (r =
.881) pertains to the positive slope of the regression line.
Table 8.2 Correlations Between Homework and GPA
 Notice that the correlations table is double-redundant; there are two .881s and two .000s
in the table. This is because the correlation between homework and GPA is the same as the
correlation between GPA and homework. A larger image of the corresponding graph
(scatterplot with regression line) is shown in Figure 8.15.
Figure 8.15 Scatterplot With Regression Line for the Homework-GPA Correlation

The points clearly plot the intersection between each student’s homework hours and his
or her corresponding GPA. The regression line can be thought of as the average pathway
through the cloud of points, or a linear summary of the points.
Another way to think about this graph is to consider the regression line as a predictive
pathway. For example, if you drew a vertical line from 8.00 on the homework (X) axis up,
you would see that it would intersect the regression line at about 3.0 on the GPA (Y) axis.
Since we know that this is a statistically significant correlation, we can use this chart as a
viable predictor—we would predict that for a student who did homework for about 8
hours a week, we would expect to see a corresponding GPA of about 3.0.

HYPOTHESIS RESOLUTION
Since the correlation calculation produced a p (p < .001) that is less than the specified .05
α level, we would say that there is a statistically significant (positive) correlation between
homework and GPA. As such, we would reject H0 and H2 but not reject H1.

DOCUMENT RESULTS
We were interested in discovering if there was a correlation between the number of hours
of homework a student did in a week and his or her GPA for that week. Correlation
analysis revealed a strong positive correlation between these two variables (r = .881),
which was statistically significant (p < .001) using an α level of .05.
GOOD COMMON SENSE
Even though a statistically significant, strong positive correlation was found between
homework and GPA, it is presumptuous to simply claim that homework hours (and
nothing else) caused the GPA. An underlying unaccounted-for factor that would not be
revealed by correlation analysis may be responsible for affecting homework hours, which,
in turn, may have affected GPA. For example, an adverse factor (e.g., household stress,
health problem, adverse social issue, etc.) may be cutting into the student’s ability to
engage in longer homework hours; conversely, a pleasurable factor may be detracting
from homework time (e.g., great social circle, multiple extracurricular activities,
compelling videogames, etc.).
Alternatively, an overarching factor may affect both homework and GPA, such as an
undiagnosed learning disability, depression, exhaustion, and so forth.
The point is that correlation, no matter what the r or the p, is just that—correlation; try
to avoid jumping to conclusions regarding causation.

Correlation vs. Causation

Correlation only means that two variables appear to move in a predictable fashion with
respect to each other (e.g., when one goes up, the other goes up; when one goes down, the
other goes down; or when one goes up, the other goes down), but keep in mind, this is not
necessarily due to causation, which would involve the change in one variable causing the
change in the other. To make the leap from correlation to causation, three criteria must be
met: (1) association/correlation, (2) temporality (meaning timing), and (3) nonspurious
(meaning authentic) (Table 8.3):
Admittedly, the criteria to claim causation are strict, but without this rigor, numerous
spurious correlations could be wrongly attributed to causality, leading to inappropriate
conclusions and potentially misguided interventions.
Table 8.3 Three Criteria for Satisfying Causality
 For example, one might find a positive correlation between chocolate milk consumption
and automobile theft—as chocolate milk sales go up, so do car thefts. Instead of
concluding that chocolate milk causes people to steal cars or that car theft causes one to
crave chocolate milk, anyone reasonable would continue their investigation and probably
discover that population may be a variable worth consideration: In a town with a
population of 2,000, we would find low chocolate milk sales and few car thefts, whereas
in a city with a population of 2,000,000, chocolate milk sales and car thefts would both be
considerably higher. In this case, we would be free to notice the positive correlation
between chocolate milk consumption and car theft, but the causal criteria between these
two variables clearly break down at all three levels.

Key Concepts
• Correlation
Strength
Direction
• Normality
• Linearity
• Homoscedasticity
• Bivariate correlation
• Scatterplot
• Regression
• r
• p
• Correlation vs. causation

Practice Exercises
1. An exercise advocate wants to determine the effect that walking rigorously has on
weight loss. The researcher recruits participants to engage in a weeklong study. The
researcher instructs participants to take a brisk walk as many days of the week as
possible for as long as they can. Participants will record the following data: weight
prior to engaging in the walking regimen, the amount of time walked each day, and
 their weight at the end of the week. Participants will submit their data to the
researcher at the end of the week. The researcher will preprocess the data to derive
the total number of hours walked (walkhrs) and the change in weight for each
participant (wtloss = weight at the end of the week − weight at the beginning of the
week).
Codebook—set up SPSS to contain two variables:
• walkhrs is a continuous variable (number of hours walked in a week).
• wtloss is a continuous variable (number of pounds lost in a week).
Data set—enter Data Set A.
NOTE: In Data Set A, Record 3, notice that the weight loss (wtloss) is −1.00; this
indicates that the participant gained 1 pound. Data Set B, Record 16 also signifies a half-
pound weight gain for that participant.

a. Write the hypotheses.

b. Run the criteria of the pretest checklist (normality [for both variables],
linearity, homoscedasticity) and discuss your findings.
c. Run the bivariate correlation and the scatterplot with regression line and
document your findings (r and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
 bivariate correlation, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Data Set B.
2. A social scientist has noticed that people seem to be spending a lot of nonwork
hours on computers and wants to determine if this may, in some way, be associated
with social relationship satisfaction (satisfaction derived from interacting with
others). To determine if there is a correlation between nonwork computer hours and
social satisfaction, the scientist recruited a group of participants and asked them to
indicate (about) how many nonwork hours they spend on the computer each week.
Next, each participant was given the Acme Social Satisfaction Inventory (ASSI);
this self-administered instrument renders a score between 0 and 80 (0 = very low
social satisfaction… 80 = very high social satisfaction).
Codebook—set up SPSS to contain two variables:
• comphrs is a continuous variable (number of nonwork hours spent on
computer per week).
• assi is a continuous variable (Acme Social Satisfaction Inventory).
Data set—enter Data Set A.
 a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality [for both variables],
linearity, and homoscedasticity) and discuss your findings.
c. Run the bivariate correlation and the scatterplot with regression line and
document your findings (r and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
bivariate correlation, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Data Set B.
3. A social scientist and an economist working together want to discover if there is a
correlation between income and happiness. The researchers recruit a group of
participants and ask them to complete a confidential survey. This self-administered
survey asks for the participant’s annual income; it also includes the Acme Life
Happiness Scale (ALHS), which renders a score between 0 and 100 (0 = very
unhappy… 100 = very happy).
Codebook—set up SPSS to contain two variables:
• income is a continuous variable (annual income in dollars rounded to nearest
thousand).
• alhs is a continuous variable (Acme Life Happiness Scale).
Data set—enter Data Set A.
 a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality [for both variables],
linearity, and homoscedasticity) and discuss your findings.
c. Run the bivariate correlation and the scatterplot with regression line and
document your findings (r and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
bivariate correlation, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Data Set B.
4. A political scientist wants to find out if there is a correlation between listening to a
newscast and an individual’s mood. This researcher recruits a group of participants
and has them listen to a newscast that was recorded earlier that morning.
Participants are instructed to listen for as long as they want; when they are done
listening, the researcher writes down the listening duration and then asks each
participant to complete the Acme Mood Report (AMR), a self-administered
instrument that renders a score between 0 and 100 (0 = very bad mood… 100 = very
good mood).
Codebook—set up SPSS to contain two variables:
• minnews is a continuous variable (number of minutes of news listened to).
 • amr is a continuous variable (Acme Mood Report).
Data set—enter Data Set A.

a. Write the hypotheses.

b. Run the criteria of the pretest checklist (normality [for both variables],
linearity, and homoscedasticity) and discuss your findings.
c. Run the bivariate correlation and the scatterplot with regression line and
document your findings (r and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
bivariate correlation, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Data Set B.
5. An educational scientist wants to examine the correlation between years of
education and job satisfaction. To address this question, the scientist recruits a group
of participants and has each complete a self-administered survey; the first question
asks how many years of education the participant has (e.g., 12 = high school
diploma, 14 = associate’s degree, 16 = bachelor’s degree, 18 = master’s degree).
The remaining questions consist of the Acme Job Satisfaction Index (AJSI), which
produces a score between 0 and 60 (0 = very unsatisfied with job… 60 = very
satisfied with job).
Codebook—set up SPSS to contain two variables:
• yearsed is a continuous variable (number of years of education).
 • ajsi is a continuous variable (Acme Job Satisfaction Index).
Data set—enter Data Set A.

a. Write the hypotheses.

b. Run the criteria of the pretest checklist (normality [for both variables],
linearity, and homoscedasticity) and discuss your findings.
c. Run the bivariate correlation and the scatterplot with regression line and
document your findings (r and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
bivariate correlation, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Data Set B.
6. A dietician wants to discover if there is a correlation between age and number of
meals eaten outside the home. The dietician recruits participants and administers a
two-question survey: (1) How old are you? and (2) How many times do you eat out
(meals not eaten at home) in an average week?
Codebook—the data set contains two variables:
• age is a continuous variable (age of participant).
 • mealsout is a continuous variable (number of meals out participant eats per
week).
Data set—load Chapter 08 − Exercise 06 A.sav.
a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality [for both variables],
linearity, and homoscedasticity) and discuss your findings.
c. Run the bivariate correlation and the scatterplot with regression line and
document your findings (r and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
bivariate correlation, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Chapter 08 – Exercise 06 B.sav.
7. A social scientist wants to determine if a person’s height might be correlated with
his or her sense of self-confidence. To explore this, the scientist recruits a group of
participants and gathers two metrics: First the researcher administers the Acme Self-
Confidence Instrument (ASCI), a self-administered survey that produces a score
between 0 and 50 (0 = very low self-confidence… 50 = very high self-confidence).
Second, the scientist measures the height of each participant.
Codebook—the data set contains two variables:
• height is a continuous variable (height of participant).
• asci is a continuous variable (Acme Self-Confidence Instrument score).
Data set—load Chapter 08 – Exercise 07 A.sav.
a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality [for both variables],
linearity, and homoscedasticity) and discuss your findings.
c. Run the bivariate correlation and the scatterplot with regression line and
document your findings (r and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
bivariate correlation, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Chapter 08 – Exercise 07 B.sav.
8. A sociologist has learned from a prior study that there is a strong positive
correlation between time spent playing a video game and the score the player earns
on that game (practice makes perfect). Since achieving such proficiency is time-
consuming, this sociologist expects that there may be a (negative) correlation
between game score and GPA. To determine if there is such an inverse correlation,
the sociologist recruits a group of participants to play a popular video game for 15
minutes, at which time the researcher records the score. Participants will also be
asked to provide a copy of their most recent transcript.
 Codebook—the data set contains two variables:
• score is a continuous variable (score on video game).
• gpa is a continuous variable (GPA for the term).
Data set—load Chapter 08 – Exercise 08 A.sav.
a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality [for both variables],
linearity, and homoscedasticity) and discuss your findings.
c. Run the bivariate correlation and the scatterplot with regression line and
document your findings (r and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
bivariate correlation, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Chapter 08 – Exercise 08 B.sav.
9. During a staff meeting, members of the English department discussed some of the
differences they have observed in students (e.g., demeanor, participation,
punctuality, etc.) enrolled in the freshman English course in the morning classes
compared with those who attend the same course later in the day. The committee
chair extracts a sample of data from the grade database for students who took
freshman English over the past year and gathers two variables: time of class and
grade earned.
Codebook—the data set contains two variables:
• time is a continuous variable (time of freshman English class: 800 =8:00
a.m… . 1600 = 4:00 p.m.).
• grade is a continuous variable (grade earned in freshman English class: 0 =
F… 4 = A).
Data set—load Chapter 08 – Exercise 09 A.sav.
a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality [for both variables],
linearity, and homoscedasticity) and discuss your findings.
c. Run the bivariate correlation and the scatterplot with regression line and
document your findings (r and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
bivariate correlation, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Chapter 08 – Exercise 09 B.sav.
10. A sociologist read an old study asserting that the number of children that an adult
has tends to match the number of siblings they grew up with; the data were gathered
from adults who were 35 years of age and older. To determine if this finding still
holds true, the sociologist recruited a group of individuals who were at least 35
years old and asked each to respond to two questions: (1) How many brothers and
sisters did you grow up with? and (2) How many children do you have?
Codebook—the data set contains two variables:
• siblings is a continuous variable (number of siblings the participant grew up
with).
• children is a continuous variable (number of children the participant has).
Data set—load Chapter 08 – Exercise 10 A.sav.
a. Write the hypotheses.
b. Run the criteria of the pretest checklist (normality [for both variables],
linearity, and homoscedasticity) and discuss your findings.
c. Run the bivariate correlation and the scatterplot with regression line and
document your findings (r and Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the
bivariate correlation, hypothesis resolution, and implications of your
findings.
• Repeat this exercise using Chapter 08 – Exercise 10 B.sav.
 CHAPTER 9
Chi-Square

Some people are passionate about aisles, others about window seats.

—Terry Jones
LEARNING OBJECTIVES
Upon completing this chapter, you will be able to:
• Determine when it is appropriate to run a chi-square test
• Identify dichotomous and polychotomous variables
• Verify that the data meet the criteria for running a chi-square test: n ≥ 5 per cell
• Order a chi-square test: table and bar chart
• Interpret the test results
• Resolve the hypotheses
• Document the results in plain English
• Calculate the % formula
DIGITAL LEARNING RESOURCES
The tutorial video and data sets for this chapter are at www.sagepub.com/knapp
• ANOVA test run video (provides details for pretest checklist)
• SPSS files
OVERVIEW
As we have seen, variables can be continuous, such as temperature, distance, weight, bank
account balance, mood score, height, and grade point average (GPA). Alternatively,
variables may be categorical (nominal), such as gender, race, religion, blood type, and
marital status.
A categorical variable may consist of any number of categories. Variables that contain
two categories are dichotomous (pronounced die-cot-uh-muss), such as
• Gender: female/male
• Voter status: registered/not registered
• Opinion: yes/no
• Attendance: present/absent
• Dwelling: house/apartment
• Grade: pass/fail
A categorical variable that contains more than two categories is polychotomous
(pronounced poly-cot-uh-muss):
• Appointment status: on time/late/canceled/no-show/rescheduled
• Marital status: single/married/separated/divorced/widowed
• Ice cream flavor preference: chocolate/strawberry/vanilla
• Visual aids: none/glasses/contact lenses/surgical correction
• Transportation: train/plane/car/bus/taxi/motorcycle/bicycle/walk
• Blood type: A+/A−/B+/B−/AB+/AB−/O+/O−
The statistical function for comparing categorical variables to each other is the chi-
square (chi is pronounced k-eye), sometimes written as χ2 (χ is the Greek letter chi).
Fortunately, the chi-square does not need you to specify how many categories are in each
variable; it will derive that as part of the process. Chi-square can efficiently handle a mix
of dichotomous and polychotomous variables.
The chi-square organizes the data from each categorical variable into a grid, compares
the categories to each other, and produces a p value. If the chi-square produces a p value
that is less than α (.05), this indicates that there is a statistically significant difference
among the categories; alternatively, a p value greater than α (.05) indicates that no
statistically significant differences among the categories exist.
Chi-square can be used to answer questions involving categorical variables, such as, Is
gender (female/male) correlated with ice cream preference (chocolate/strawberry/vanilla)?
In other words, do girls and boys tend to select the same ice cream flavors, or are there
differences in their preferences? If the p value is less than α (.05), then we would say that
there is a statistically significant difference between girls’ ice cream preference and boys’
ice cream preference (girls like different ice cream flavors from boys). Alternatively, if the
p value is greater than the α (.05), then we would say that there is no statistically
significant difference between the genders when it comes to ice cream selection (chocolate
can be pretty compelling to both genders).

To recap, chi-square can be used to compare categorical variables with the same number
of categories in each variable, such as gender (female/male) to opinion (yes/no), which
would render a 2 × 2 chi-square table. Chi-square can also analyze categorical variables
that have different category counts without having to specify any additional processing
parameters, such as gender (female/male) to blood type (A+/A−/B+/B−/AB+/AB−/O+/O
−). This chi-square test would produce a 2 × 8 or an 8 × 2 chi-square table, depending on
how you choose to load the variables into rows and columns—either way, the analysis
would produce equivalent results.
EXAMPLE
A political analyst is seeking to determine if gender is associated with political party
affiliation.
RESEARCH QUESTION
Does gender have a bearing on political party affiliation, or are women and men fairly
evenly distributed among the political parties?
GROUPS
When it comes to chi-square, it is not so much a matter of groups as categories within the
variables. This inquiry involves two categorical variables: gender, which has two
categories, and the party, which has three categories (notice that gender is dichotomous
and party is polychotomous):
• Gender: female/male
• Party: Democrat/Republican/Other
PROCEDURE
The researcher will recruit a group of participants and ask each person to complete the
following self-administered anonymous survey card:

Please check one answer for each question:

1. What is your gender:
Female
Male
 2. What is your political party affiliation:
Democrat
Republican
Other
When completed, please drop this card in the response box.
Thank you for participating in this survey.

HYPOTHESES
H0: There is no correlation between gender and political party affiliation.
H1: There is a correlation between gender and political party affiliation.

DATA SET
Use the following data set: Chapter 09 − Example 01 – Chi-Square.sav.
This data set contains two categorical variables representing the data gathered from the
participants’ survey cards: gender (1 = female, 2 = male) and party (1 = Democrat, 2 =
Republican, 3 = Other).

PRETEST CHECKLIST
Pretest Checklist Criterion 1—n ≥ 5 per Cell Minimum
The chi-square will organize the categorical data from the variables into a table. It is easy
to anticipate the dimensions of the table simply by multiplying the number of categories in
each variable. In this case, gender has two categories (female/male), and party has three
categories (Democrat/Republican/Other); hence, the chi-square table will consist of (2 × 3
=) 6 cells (Table 9.1).
Table 9.1 Chi-Square Table Basic Structure for Gender and Party Contains Six
Cells

The pretest checklist rule for chi-square states that each cell should have at least 5
entries; initially, one might anticipate that the total n for this study should be 30 (6 cells ×
5 per cell = 30). Actually, the total n will need to be more than 30, since a total n of 30
would presume that participants’ responses will fill the six cells evenly (5 per cell). Since
this is implausible, we should consider 30 as the minimum total n; we will require a total n
of more than 30.
The chi-square report will show these counts for each cell, and hence we will be able to
verify these criteria when we inspect the table in the Results section.

TEST RUN
1. From the main screen, click on Analyze, Descriptive Statistics, Crosstabs… (Figure
9.1); this will bring you to the Crosstabs menu (Figure 9.2).
Figure 9.1 Run the Chi-Square Analysis; Click on Analyze, Descriptive Statistics,
Crosstabs . . .

Figure 9.2 Load Variables Into Row(s) and Column(s) Windows and Check
Display clustered bar charts

2. On the Crosstabs menu, move gender from the left window to the Row(s) window,
and move party from the left window to the Column(s) window.
3. Check the Display clustered bar charts checkbox.
4. Click on the Statistics… button; this will take you to the Crosstabs: Statistics menu
(Figure 9.3).
Figure 9.3 Crosstabs: Statistics Menu: Check the Chi-square Checkbox
 5. Check the Chi-square checkbox.
6. Click on the Continue button. This will take you back to the Crosstabs menu.
7. Click the OK button, and the chi-square will process.

RESULTS

Finalizing Pretest Criteria

We begin by inspecting the (six) circled cells in the Crosstabulation table (Table 9.2) and
note that each cell has a count (n) of at least 5; hence, the pretest criteria are satisfied.
Next, observe the Sig. (p) value in the chi-square tests table (Table 9.3) on the Pearson
chi-square row; it indicates a Sig. (p) value of .023; since this is less than the specified .05
α level, we conclude that there is a statistically significant difference among genders when
it comes to political party affiliation.
Table 9.2 Gender * Political Party Crosstabulation

Table 9.3 Chi-Square Tests Results: Sig. (p) = .023

 The Gender * Political Party Crosstabulation (Table 9.2) provides clear enough results
detailing how many of each gender are affiliated with each political party; even a cursory
overview of this table helps us to see that the numbers for females look very different from
those of males, which is confirmed by the .023 p value (Table 9.3), but to get a more
intuitive grasp of these data, inspect the corresponding bar chart (Figure 9.4).
The better you can conceptualize the data, the better prepared you will be to write a
cogent Results section detailing your findings. In addition to discussing that the p value
indicates a statistically significant difference in party affiliation among the genders, the
bar chart can help provide additional meaning to the discussion. You may include
narrative explaining that the differences detected in this sample show that females tend to
be Democrats and males tend to be Republicans, whereas Other seems to attract about the
same number of males and females.

HYPOTHESIS RESOLUTION
The chi-square produced a p of .023, which is less than the specified .05 α level, which
indicates that there is a statistically significant difference between the genders with respect
to political party affiliation. As such, we would reject H0 and not reject H1.

DOCUMENT RESULTS
According to a survey of 60 adults, we detected a statistically significant difference (p =
.023, α = .05) in political party affiliation based on gender. Of the 31 females surveyed,
most were Democrats (16 Democrats, 9 Republicans, and 6 Other), whereas the 29 men in
our sample were predominately Republicans (6 Democrats, 18 Republicans, and 5 Other).

Another way of documenting chi-square results involves discussing percentages. With

respect to chi-square, percentages are fairly easy to calculate; just divide the part (smaller
number) by the total (larger number) and multiply by 100 (Table 9.4). Consider this
documentation variation (the bracketed portions below are included to clarify the
percentage calculations but would not typically be included in the write-up):
 In an effort to discover if gender is associated with political party affiliation, we recruited 60 adults (31 females and
29 males); 52% [16 ÷ 31 = .5161] of the women surveyed identified as Democrats, whereas 62% [18 ÷ 29 = .6207]
of men identified as Republicans. Chi-square analysis revealed a statistically significant difference (p = .023, α =
.05) between genders when it comes to political party affiliation.

GOOD COMMON SENSE

It can be useful to run the chi-square test a second time; try swapping the variables
(rows/columns) (Figure 9.2). The statistical results will be the same, but exchanging the
variables in this way will arrange the data differently in the table and bar chart, which may
offer an alternate perspective on the results.
Figure 9.4 Bar Chart for Gender * Political Party

*Table 9.4 % Formula

You may find it useful to order that the chi-square crosstabulation data be presented
using percentages instead of (just) the n. To try this, go to the Crosstabs menu (Figure 9.2)
and click on the Cells… button. This will take you to the Crosstabs: Cell Display menu
(Figure 9.5), where you can order percentages for row, column, and total.
Figure 9.5 Crosstabs: Cell Display—Check Percentages
 Occasionally, a continuous variable can be reduced to a categorical variable, thereby
facilitating chi-square analyses. For example, age is a continuous variable (ranging from
0–100), but age could be reduced to two categories: 1 = juvenile and 2 = adult. SPSS
includes an easy-to-use Recode feature that helps automate such processes.
Recoding would leave the continuous variable age as is, but based on the age, we could
generate values for a new categorical variable, AgeClass (1 = juvenile, 2 = adult), using
the following criteria:
If age is less than 18, then AgeClass = 1.
If age is 18 or greater, then AgeClass = 2.
The procedure for recoding variables in this way is detailed in Chapter 10, specifically
the section on recoding.

Key Concepts
• Continuous variables
• Categorical variables
• Chi-square (χ2)
• Pretest checklist (n ≥ 5 per cell)
• Overall n actually required
• Crosstabs
• Percentage calculation

Practice Exercises
1. Acme Creamery wants to create advertisements showing boys and girls enjoying
Acme Ice Cream; they have consulted with you to discover if boys and girls have
the same or different ice cream flavor preferences. You recruit a group of volunteers
and gather two pieces of information for each participant using this self-
administered survey card:

Ice Cream Preference Survey

Please check one answer for each question:
1. Gender:
Female
Male
2. What is your favorite ice cream flavor:
Chocolate
Strawberry
Vanilla
Please drop this card in the collection box.
Thank you for participating in our survey.

Codebook—set up SPSS to contain two variables:

• gender is a categorical variable (1 = female, 2 = male).
• icecream is a categorical variable (1 = chocolate, 2 = strawberry, 3 = vanilla).
Data set—enter Data Set A.
a. Write the hypotheses.
 b. Run the criteria of the pretest checklist (n is at least 5 per cell in the
Crosstabs) and discuss your findings.
c. Run the chi-square test and document your findings (ns and/or percentages,
Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the chi-
square test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Data Set B.
2. You notice that some bus riders wear headphones while others do not, but with so
many passengers on so many different buses, it is hard to estimate if gender is a
factor when it comes to wearing or not wearing headphones. To address this
question, you spend the day riding city buses as you unobtrusively record your
observations of each passenger on the following data sheet:

Codebook—set up SPSS to contain two variables:

• gender is a categorical variable (1 = female, 2 = male).
• headphones is a categorical variable (1 = headphones, 2 = no headphones).
Data set—enter Data Set A.
 a. Write the hypotheses.
b. Run the criteria of the pretest checklist (n is at least 5 per cell in the
Crosstabs) and discuss your findings.
c. Run the chi-square test and document your findings (ns and/or percentages,
Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the chi-
square test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Data Set B.
3. The clinicians at Anytown Health Clinic want to determine how useful the flu shot
is in their community. The researcher approaches patients as they exit the clinic;
those who are willing to partake in this study are asked to sign an informed consent
document and complete the following card:

Flu Shot Survey

 1. Have you had a flu shot this season?
Yes
No
2. Phone number or e-mail ID: ____________________.
A member of our staff will contact you in 60 days.
Please drop this card in the collection box.
Thank you for participating in our survey.

FOR ADMINISTRATIVE USE ONLY

Got sick with flu
Did not get sick with flu

The researcher will contact each participant in 60 days to ask if he or she contracted the
flu in the past 60 days and mark the bottom of each card accordingly.
Codebook—set up SPSS to contain two variables:
• flushot is a categorical variable (1 = had flu shot, 2 = did not have flu shot).
• flusick is a categorical variable (1 = got sick with flu, 2 = did not get sick
with flu).
Data set—enter Data Set A.
a. Write the hypotheses.
b. Run the criteria of the pretest checklist (n is at least 5 per cell in the
Crosstabs) and discuss your findings.
c. Run the chi-square test and document your findings (ns and/or percentages,
Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the chi-
square test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Data Set B.
4. The administrative staff of Acme College wants to tune the availability of student
resources (e.g., website content, library hours, support staffing, etc.) to better fit the
needs of students. You have been asked to determine if the degree students are
working on (bachelor’s vs. master’s) is associated with the type of learning (in
classroom vs. remote learning) students have opted for; you are given a sample
drawn from the student enrollment database to analyze.
Codebook—set up SPSS to contain two variables:
• degree is a categorical variable (1 = bachelor’s, 2 = master’s).
• location is a categorical variable (1 = in classroom, 2 = remote learning).
Data set—enter Data Set A.
 a. Write the hypotheses.
b. Run the criteria of the pretest checklist (n is at least 5 per cell in the
Crosstabs) and discuss your findings.
c. Run the chi-square test and document your findings (ns and/or percentages,
Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the chi-
square test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Data Set B.
5. To determine if how data are gathered has any bearing on responses to a question
involving substance abuse (Have you ever used an illegal drug?), you recruit
willing participants and randomly assign them to one of three groups: Those in
Group 1 will be asked the question via face-to-face interview, those in Group 2 will
respond using a standard pencil-and-paper mail-in survey, and those in Group 3 will
be directed to an online survey; no names or identifying information will be
gathered.
Codebook—set up SPSS to contain two variables:
• media is a categorical variable (1 = face-to-face interview, 2 = mail-in
survey, 3 = online survey).
• drug is a categorical variable (1 = yes, 2 = no).
Data set—enter Data Set A.
a. Write the hypotheses.
b. Run the criteria of the pretest checklist (n is at least 5 per cell in the
Crosstabs) and discuss your findings.
c. Run the chi-square test and document your findings (ns and/or percentages,
Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the chi-
square test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Data Set B.
6. In an effort to better accommodate students, Acme University wants to find out if
students pursuing different degrees (bachelor’s/master’s/doctorate) have the same or
different preferences when it comes to class time (day/night). To determine this, you
are commissioned to administer the following survey to a sample of the students
currently enrolled.

Course Schedule Preference Survey

Please check one answer for each question:
1. When do you prefer to take the majority of your courses?
Day (8:00 a.m.–5:00 p.m.)
Night (5:00 p.m.–10:00 p.m.)
2. What degree are you currently working on?
Bachelor’s
Master’s
Doctorate
Please drop this card in the collection box.
Thank you for participating in our survey.

Codebook—the data set contains two variables:

• time is a categorical variable (1 = day, 2 = night).
• degree is a categorical variable (1 = bachelor’s, 2 = master’s, 3 = doctorate).
Data set—load Chapter 09 – Exercise 06 A.sav.
a. Write the hypotheses.
b. Run the criteria of the pretest checklist (n is at least 5 per cell in the
Crosstabs) and discuss your findings.
c. Run the chi-square test and document your findings (ns and/or percentages,
Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the chi-
square test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Chapter 09 – Exercise 06 B.sav.
7. A political scientist expects that how a person votes (or does not vote) may be
associated with his or her age. To investigate this, the scientist gathers a
convenience sample, asking voluntary participants to anonymously complete the
following card:

Voter Survey
 Please check one answer for each question:
1. How old are you?
18–35
36–64
65 and older
2. How did you vote in the last election?
I voted in person at a polling precinct.
I voted by mail.
I did not vote.
Please drop this card in the collection box.
Thank you for participating in our survey.

Codebook—The data set contains two variables:

• age is a categorical variable (1 = 18−35, 2 = 36−64, 3 = 65 and older).
• vote is a categorical variable (1 = vote in person, 2 = vote by mail, 3 = not
vote).
Data set—load Chapter 09 − Exercise 07 A.sav.
a. Write the hypotheses.
b. Run the criteria of the pretest checklist (n is at least 5 per cell in the
Crosstabs) and discuss your findings.
c. Run the chi-square test and document your findings (ns and/or percentages,
Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the chi-
square test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Chapter 09 – Exercise 07 B.sav.
8. The Acme Veterinary Science Center wants to find out if its three dog foods appeal
to all dogs equally or if breed is a factor in a dog’s food preference. Per the research
criteria specified, you use quota sampling to recruit 90 pets: 30 Cocker Spaniels, 30
Beagles, and 30 Keeshonds. Owners are asked not to feed their pets for 4 hours
prior to the test. Each dog is tested individually; the dog is placed 5 feet (1.5 meters)
away from three clear bowls of dog food, all with equal weights. On cue, the leash
is removed and the dog is free to eat from any bowl(s). After dismissing each
participant, you weigh the bowls; the lightest bowl wins (meaning that the dog ate
the most food from that bowl). In case of a tie, the winning bowl is the one that the
dog went to first.
Codebook—the data set contains two variables:
 • dog is a categorical variable (1 = Cocker Spaniel, 2 = Beagle, 3 = Keeshond).
• food is a categorical variable (1 = Food A, 2 = Food B, 3 = Food C).
Data set—load Chapter 09 – Exercise 08 A.sav.
a. Write the hypotheses.
b. Run the criteria of the pretest checklist (n is at least 5 per cell in the
Crosstabs) and discuss your findings.
c. Run the chi-square test and document your findings (ns and/or percentages,
Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the chi-
square test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Chapter 09 – Exercise 08 B.sav.
9. Each year, the Department of Education in Anytown publishes a report of college-
bound high school seniors. You have been recruited to compare data gathered from
seniors at the Acme Academy, a local private school, with seniors at Anytown High
School, a public institution.
Codebook—the data set contains two variables:
• school is a categorical variable (1 = Acme Academy, 2 = Anytown High
School).
• college is a categorical variable (1 = not attending university, 2 = attending
university).
Data set—load Chapter 09 – Exercise 09 A.sav.
a. Write the hypotheses.
b. Run the criteria of the pretest checklist (n is at least 5 per cell in the
Crosstabs) and discuss your findings.
c. Run the chi-square test and document your findings (ns and/or percentages,
Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the chi-
square test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Chapter 09 – Exercise 09 B.sav.
10. A political analyst is conducting a survey aimed at tuning campaign messages; the
question of interest is, “Do women and men tend to vote the same or differently
when it comes to electing the mayor?” The analyst gathered responses from willing
participants using self-administered survey cards:

Voter Survey—City Mayor

Please check one answer for each question:
1. What is your gender?
Female
Male
2. Who will you vote for?
Smith
Jones
Undecided
Please drop this card in the collection box.
Thank you for participating in our survey.

Codebook—the data set contains two variables:

• gender is a categorical variable (1 = female, 2 = male).
• vote is a categorical variable (1 = Smith, 2 = Jones, 3 = undecided).
Data set—load Chapter 09 – Exercise 10 A.sav.
a. Write the hypotheses.
b. Run the criteria of the pretest checklist (n is at least 5 per cell in the
Crosstabs) and discuss your findings.
c. Run the chi-square test and document your findings (ns and/or percentages,
Sig. [p value]).
d. Write an abstract under 200 words detailing a summary of the study, the chi-
square test results, hypothesis resolution, and implications of your findings.
• Repeat this exercise using Chapter 09 – Exercise 10 B.sav.
PART III
Data Handling
 CHAPTER 10
Supplemental SPSS Operations

• Generating Random Numbers

• Sort Cases
• Select Cases
• Recoding
• Importing Data
• Syntax

Never trust a computer you can’t throw out a window.

—Steve Wozniak
LEARNING OBJECTIVES
Upon completing this chapter, you will be able to:
• Perform extended SPSS operations to enhance your capabilities, versatility, and
data-processing efficiency
• Generate a list of random numbers to your specifications
• Perform single and multilevel sorting
• Select cases using multiple criteria
• Recode variables
• Import data from external sources: Excel and ASCII files
• Comprehend how to generate SPSS Syntax code automatically using the Paste
function
• Understand how to customize SPSS Syntax code
 • Save, load, and execute SPSS Syntax code
• Practice safe data storage protocols
DIGITAL LEARNING RESOURCES
The data sets for this chapter are at www.sagepub.com/knapp
• SPSS data files
• SPSS Syntax file
• Excel file
• ASCII file
OVERVIEW
The data sets that have been provided thus far has been crafted to work as is in the SPSS
environment, but as you become more statistically proficient, your research curiosity and
scientific creative thinking are likely to further develop. You may want to analyze data of
your own, examine data from other non-SPSS sources, or run more elaborate statistical
analyses. This chapter explains some of the most useful supplemental SPSS features and
functions to help you work more productively.
GENERATING RANDOM NUMBERS
In Chapter 2, the “Simple Random Sampling” section discussed the need to randomly
select 30 individuals from a sample frame of 1,000 potential participants. Flipping a coin
to make these selections is clearly out of the question for this task; instead, we can use
SPSS to generate this list of random numbers:
1. On the Variable View screen, create a numeric variable to contain the random
numbers that we will have SPSS generate; we will call it randnum (Figure 10.1).
Figure 10.1 Create a Numeric Variable (randnum) to Contain the Random
Numbers

2. Switch to the Data View screen and put a 0 (or any number) in the randnum column
at record 30 (Figure 10.2).
Figure 10.2 Enter Any Number at Record 30 for randnum (So SPSS Will Know
How Long the List Should Be)
3. Click on Transform/Compute Variable… (Figure 10.3). This will take you to the
Compute Variable menu.
Figure 10.3 Click on Transform/Compute Variables…

4. On the Compute Variable menu, in the Target Variable window, enter randnum; in
the Numeric Expression window, enter rnd(rv.uniform(1,1000)) (Figure 10.4). This
tells SPSS to place the random values in the randnum variable. Now to demystify
the rnd(rv.uniform(1,1000)) expression:
• rnd means round the result to the nearest integer; if you wanted the random
numbers to include decimal digits, you could enter rv.uniform(1,1000).
• rv.uniform means “random values, uniform,” wherein each number has an equal
chance of being selected.
• (1,1000) specifies the minimum and maximum values.
Figure 10.4 On the Compute Variable Menu, in the Target Variable Window, Enter
randnum; in the Numeric Expression Window, Enter
rnd(rv.uniform(1,1000))
 5. Click on the OK button. If you are then asked if you wish to Change existing
variable?, click on the OK button.
6. The Data View screen should now show 30 random numbers in the randnum
column (Figure 10.5).

NOTE: Naturally, your results will produce a different set of random numbers.
The random number generator does not keep track of repeats among these numbers;
hence, you may want to order more random numbers than you actually need so that you
can ignore duplicates.
SORT CASES
As you have probably noticed, the order of the data on the Data View screen has no effect
on the statistical outcomes, but at times you may find it useful to inflict some order on the
data. In data processing, the term sort is akin to alphabetizing; you can sort the data to
help you make sense of them. You might be interested in those who scored highest or
lowest on a particular variable to better conceptualize the data set; inflicting such order
may help you identify patterns or trends within the data set that may not have been evident
otherwise—you can then follow your curiosity with additional statistical tests.
Figure 10.5 Data View Screen With Resulting Random Numbers for randnum
 SPSS supports multilevel sorting. This means that you could specify the first level to
sort by name and the second level to sort by age (you can specify as many levels as you
need). So if two or more people have the same name, the system would then look to age to
finalize the sorting sequence (Table 10.1).
Table 10.1 Data Sorted by Name (Level 1) and Then by Age (level 2)

The default is to sort the variable at each level in ascending order (from lowest to
highest); alternatively, you can specify that you want to sort in descending order (from
highest to lowest). For example, if you specify that Level 1 is name ascending and Level 2
is age descending, the system will sort the data with names from A to Z, but if there is a
tie at Level 1 (name), it will subsort those records by age, from highest to lowest (Table
10.2).
Table 10.2 Data Sorted Ascending by Name (Level 1), Then Descending by Age
(Level 2)
1. Load the following data file: Chapter 10 – Example 01 – Sort, Select, Recode.sav.
This data set contains bowling league information. Notice that initially, the records
are in no particular order.
Codebook—the data set contains five variables:
• name is a string variable.
• gender is a categorical variable (1 = female, 2 = male).
• age is a continuous variable.
• bowlavg is the bowler’s average score to date.
• team is a categorical variable (1 = Strike Force, 2 = Lane Surfers, 3 = 7-10
Squad, 4 = Pinbots, 5 = Bowled Over, 6 = The Pin Boys).
2. Suppose you want to identify the top bowler in each team; this would involve a two-
level sort:
Level 1: team (ascending)
Level 2: bowlavg (descending)
3. Click on Data/Sort Cases… (Figure 10.6). This will take you to the Sort Cases
menu (Figure 10.7).
4. On the Sort Cases menu (Figure 10.7), move team from the left window to the Sort
by window, and click on Ascending.
5. Move bowlavg from the left window to the Sort by window and click on
Descending.
Figure 10.6 To Sort Data, Click on Data/Sort Cases…

6. Notice that once a variable is in the Sort by window, you can click on a variable to
drag it up and down the list, thereby altering the sorting levels. You can also change
the Sort Order for a variable by clicking on the variable, then selecting Ascending or
Descending.
7. Click on OK, and the system will sort the cases.
8. Observe the order of the data on the Data View screen (Figure 10.8).
 Figure 10.7 Sort Cases Menu Indicates a Two-Level Sort: First by team (in
Ascending Order), Then by bowlavg (in Descending Order)

Figure 10.8 Data View Screen Showing Labels (Not Values)

Notice that the data are grouped by team, then by highest to lowest bowling average
(bowlavg), but the teams are not arranged alphabetically. This is because team is not really
a string (alphanumeric) variable like name; team is actually a numeric variable, with
labels assigned. If you click on the Value Labels icon, you will see team switch to the
numeric values for that variable (Figure 10.9), which will show that the sort actually did
work as ordered—per the codebook, the label “Strike Force” is coded as 1, “Lane Surfers”
is coded as 2, and so on.
Feel free to sort the data other ways and observe the resulting sequences.
SELECT CASES
In prior chapters, we have used the Select Cases (icon) to run the pretest checklist for
various statistical tests, which has enabled us to process statistics for one group at a time.
The Select Cases function is also capable of isolating data using more complex selection
criteria.
 For example, you may wish to perform statistical analyses only on members of the
bowling league who are (1) females and (2) have a bowling average of 120 or higher.
1. Load the following data file: Chapter 10 – Example 01 – Sort, Select, Recode.sav.
2. Click on the Select Cases icon.
Figure 10.9 Data View Screen Showing Values (Not Labels)

3. In the Select group, click on the If condition is satisfied button.

4. Click on the If… button. This will take you to the Select Cases: If menu; in the large
box at the top, enter the following selection criteria: gender = 1 and bowlavg >= 120
(Figure 10.10). Note: These criteria include gender = 1 since the gender variable has
the following labels assigned to it: 1 = female, 2 = male.
Figure 10.10 Select If Menu, Specifying Two Selection Criteria

5. Click on the Continue button; this will take you back to the Select Cases menu.
6. Click on the OK button.
7. Go to the Data View screen.
8. Notice that all records are slashed out except for the females with bowling averages
that are at least 120. Now you can proceed with whatever statistical analysis you
wish to perform on the selected (not slashed-out) records.
Try some other case selection criteria and observe which records are affected on the
Data View screen:
• team > 2 and team <= 5 and bowlavg > 150
 • team = 3 or team = 5
• name < “G”
• name > “A” and name < “F”
• age >= 17 and age < 31 and bowlavg < 170

You have probably surmised some of the coding syntax from the examples above; just
to clarify a few things:
• Single and compound and/or logic is supported.
• Commonly used logical relationships are symbolized as
< less than
> greater than
= equal to
<= less than or equal to
>= greater than or equal to
• Although the system does support not logic, negative logic can be confusing; try to
build your selection criteria using and/or parameters.
• When your selection criteria involve string variables, be sure to wrap quotation
marks around your parameter(s) (e.g., name < “G”); otherwise, the processor will
think you are referring to a variable (G).
RECODING
Occasionally, you may wish to change the way a variable is presented in a data set. For
example, in the current database, age is a continuous variable that ranges from 16 to 41,
but suppose you wanted to use a t test to compare bowling averages of minors versus
adults; you would need a categorical variable to designate which group (minor vs. adult)
each record belonged to based on age. This is accomplished via recoding. We will leave
age (a continuous variable) intact, but we can use the recode function to create the new
variable, age2 (a categorical variable), which will be based on age using the following
(two) criteria:
• If age < 18, then age2 = 1.
• If age >= 18, then age2 = 2.
Remember: “>=” notation is computer language for “greater than or equal to.” After the
recode function generates the values for age2, we will assign the following value labels to
age2 to provide clarity:
• 1 = “Minor”
• 2 = “Adult”
 1. Load the following data file: Chapter 10 – Example 01 – Sort, Select,
Recode.sav.
2. Click on Transform/Recode into Different Variables… (Figure 10.11).
Figure 10.11 To Begin Recoding, Click on Transform/Recode into Different
Variables… .

3. On the Recode into Different Variables menu, move age from the left
window into the Numeric Variable Output Variable window (Figure 10.12).
4. Enter age2 in the Output Variable, Name box.
5. Click on the Change button.
Figure 10.12 Recode into Different Variables Menu

6. So far, you have indicated that the (continuous) variable age will be recoded
into the new (categorical) variable age2. Now you need to indicate how age
will be recoded into age2. Click on the Old and New Values… button.
7. Notice that there are variety of recoding options; we will use a simple
method. In the Old Value area, select Range, and enter 1 through 17 (Figure
10.13). In the New Value area, select Value and enter 1. Then click on the Add
button.
This tells the processor to look for age, and for any record with an age between 1 and
17 (inclusive), write a 1 in that record in the age2 variable (and do not change the contents
of the age variable).
Figure 10.13 Recode into Different Variables: Old and New Values Menu,
Recoding Ages 1–17 to age2 as 1
 8. In the Old Value area, select Range, and enter 18 through 99. In the New
Value area, select Value and enter 2. Then click on the Add button (Figure
10.14).
This tells the processor to look for age, and for any record with an age between 18 and
99 (inclusive), write a 2 in that record in the age2 variable (and do not change the contents
of the age variable).
9. At this point, the Old New window shows the recoding criteria (Figure
10.15).
10. Click on the Continue button; this will return you to the Recode into Different
Variables menu.
11. Click on the OK button, and the recoding will process.
Figure 10.14 Recode into Different Variables: Old and New Values Menu,
Recoding Ages 18–99 to age2 as 2

Figure 10.15 Excerpt From Recode into Different Variables: Old and New Values
Menu: Criteria Used to Recode age to age2
 12. Go to the Data View screen, and notice that the new variable age2 is populated
appropriately based on the age for each record.
13. To finalize the process, go to the Variable View screen and specify the
corresponding Value Labels for age2 (1 = minor, 2 = adult) (Figure 10.16).
In this example, notice that we opted to Recode into Different Variables… as opposed to
Recode into Same Variables… . This choice preserved the original (continuous) age
variable, so that we could perform further analyses on it. Had we used Recode into Same
Variables… , each original age would have been overwritten with 1s and 2s as we
specified (signifying minors and adults); you would not be able to get back to the original
age(s). In addition, if an error had occurred during the recoding process, the age variable
itself would have been altered, and hence it would not be possible to retry the recoding
procedure. A reasonable rule of thumb is this: Recode all you want, but always keep your
source data intact.
Figure 10.16 Assign Value Labels to age2 (1 = Minor, 2 = Adult)

IMPORTING DATA
So far, all data used have been prepared to operate properly in SPSS, but, as you might
expect, there is a world of worthy data not necessarily in SPSS format. When the data are
only available on paper, naturally you will have to enter the data manually. Fortunately,
more and more data are available in a digital form; even if the data are not in SPSS format,
SPSS is equipped with some fairly versatile data import features designed to promptly
load non-SPSS data into the SPSS environment for processing.
The two most common forms of non-SPSS data are Microsoft Excel and ASCII
(pronounced ask-key) files; once you see how to import data from these two sources, you
should be able to reason your way through importing other data formats.
 The import data feature in SPSS tends to vary somewhat from version to version. If
there is a discrepancy between the instructions in this section and how your version of
SPSS operates, then consult the Help menu in your software and search for import or
import data.
Importing Excel Data
The Excel worksheet in this example contains 101 rows; the first row contains the variable
name for each column (ID, Age, Score), followed by 100 records, each with three
variables (columns):
1. ID is a string variable consisting of six letters and numbers.
2. Age is a numeric variable consisting of two digits.
3. Score is a numeric variable consisting of two digits.
The first row of the Excel file that you will be importing has the variable names at the
top of each column; this will be useful when it comes to the import process. If these names
were not present, you could still proceed with the import, but you would need a codebook
to know how to label the variables after the file has been imported.
1. Click on File/Open/Data… (Figure 10.17). This will take you to the Open File
menu (Figure 10.18).
Figure 10.17 To Begin the Import Process, Click on File/Open/Data…

2. On the Open Data menu (Figure 10.18) for Files of type, select Excel; this will
narrow the file list to Excel worksheets only.
Figure 10.18 Open Data Menu: For Files of type, Select Excel; for File name,
Select Chapter 10 – Example 02 – Excel Data.xls
3. In the large file list window, select Chapter 10 – Example 02 – Excel Data.xls and
click on the Open button. This will take you to the Open Excel Data Source menu
(Figure 10.19).
Figure 10.19 Opening Excel Data Source Menu

4. In this case, in the Opening Excel Data Source menu (Figure 10.19), the defaults are
correct: Since the Excel worksheet has the variable names at the top of each column,
the corresponding checkbox (Read variable names from the first row of data) is
checked. If the variable names are not included as the first row in the Excel sheet,
then uncheck that box.
5. The input utility also identified the worksheet and cells involved correctly: Sheet1
[A1:C101]. Click on the OK button.
6. SPSS will process the import; notice that the system loaded the Excel file and the
variable names have been assigned accordingly (Figure 10.20).
Figure 10.20 Data View Screen After Excel File Import

7. To further verify that the import worked properly, switch to the Variable View
screen (Figure 10.21). Notice that ID has been brought in as a string variable since it
contains alphanumeric characters, and age and score have been correctly configured
as numeric variables.
Figure 10.21 Variable View Screen After Excel file Import
 8. You can now proceed with statistical analyses. When you are ready to save the file,
the system will write it out as an SPSS file unless you specify otherwise.
Importing ASCII Data
ASCII stands for American Standard Code for Information Interchange; it is basically a
generic, plain text file, not associated with any particular software package. ASCII file
names typically have a .txt (text) or, less often, an .asc (ASCII) suffix (e.g.,
Experiment18.txt, DistrictA.asc). The data in such files are traditionally arranged with one
record per row; the variables within each row are usually separated by a delimiter
character, such as a comma or other symbol (Figure 10.22). Alternatively, some files do
not use delimiters to separate variables; instead, they use a fixed number of characters per
variable, producing columns of data padded with spaces (Figure 10.23).
Figure 10.22 Comma-Delimited ASCII Data

Figure 10.23 Fixed-Column ASCII Data

Since delimited files are more common, this example will involve a comma-delimited
ASCII file.
1. Click on File/Open/Data… (Figure 10.24). This will take you to the Open File
menu (Figure 10.24).
Figure 10.24 To Begin Import Process, Click on File/Open/Data…
2. In the Open Data menu (Figure 10.25) for Files of type, select Text; this will narrow
the file list. If the file name has a suffix other than .txt, then at the Files of type
option, select All files.
3. In the large file list window, select Chapter 10 – Example 03 – Comma delimited
data.txt and click on the Open button. This will take you to the Text Import Wizard –
Step 1 of 6 menu (Figure 10.26).
Figure 10.25 Open Data Menu: For Files of type, Select Text; for File name,
Select Chapter 10 – Example 03 – Comma Delimited Data.txt

Figure 10.26 Text Import Wizard – Step 1 of 6 Menu

4. In the Text Import Wizard – Step 1 of 6 menu, click on the Next > button. This will
take you to the Text Import Wizard – Step 2 of 6 menu (Figure 10.27).
Figure 10.27 Text Import Wizard – Step 2 of 6 Menu

5. In the Text Import Wizard – Step 2 of 6 menu, since this is a comma-delimited data
set, for the How are your variables arranged? question, select Delimited. For the
Are variables names included at the top of your file? question, select Yes. Then,
click on the Next > button; this will take you to the Text Import Wizard – Step 3 of 6
menu.
6. In the Text Import Wizard – Steps 3 – 5 menus, the defaults are all appropriate. Click
on the Next > button for each of these menus until you reach the Text Import Wizard
– Step 6 of 6 menu.
7. In the Text Import Wizard – Step 6 of 6 menu, click on the Finish button.
8. SPSS will process the import; notice that the system loaded the ASCII comma-
delimited text file, with the variable names assigned accordingly (Figure 10.28).
Figure 10.28 Data View Screen After Text File Import

9. To further verify that the import worked properly, switch to the Variable View
screen (Figure 10.29). Notice that ID has been brought in as a string variable since it
contains alphanumeric characters, and age and score have been correctly configured
as numeric variables.
Figure 10.29 Variable View Screen After Text File Import
 10. You can now proceed with statistical analyses. When you are ready to save the file,
the system will write it out as an SPSS file unless you specify otherwise
If the data that you are importing contain categorical variables coded as numbers (e.g., 1
= yes, 2 = no), it would be to your advantage to gather as much codebook information as
you can, so you can create the appropriate data labels for each such variable.

SYNTAX
First, a bit of history: SPSS was initially developed and implemented long, long ago, in
the dark ages of computing, before mankind had discovered the ability to point and click.
In those days, the only way to communicate with SPSS was to type in commands and
parameters using the SPSS Syntax language. Despite the advent of elegant menu systems,
the SPSS Syntax language is still alive and well, lurking silently in the background of
every SPSS menu and beyond. End of history lesson.
So far, we have used the SPSS menu system to run analyses, which is optimal for
procedures that you intend to run only once, but in cases where you will be running the
same analyses repeatedly on a data set over time, it is possible to record and save your
series of menu selections and parameter settings so you can rerun the entire analysis
without having to click your way through multiple menus.
The good news is, you do not have to learn the entire SPSS Syntax programming
language to use it; you can use the Paste function to generate these programs for you.
Here is how it works:
1. Load the following file: Chapter 10 – Exercise 04 – Syntax.sav (this data set may
look familiar; it is a copy of Chapter 06 – Example 01 – ANOVA.sav).
2. Order the traditional descriptive statistics with a histogram for the variable score,
BUT instead of clicking on the OK button at the end to execute the analysis, click
on the Paste button. It will initially seem as if nothing happened (but keep reading).
NOTE: For more details on this procedure, please refer to Chapter 4 (“SPSS—Descriptive
Statistics: Continuous Variables (Age)”); see the star ( ) icon on page 71, but instead of
age, process the score variable.
3. Whereas clicking on the OK button runs the analysis, clicking on the Paste button
instead produces the block of code shown in Figure 10.30, which is the programmed
 equivalence for producing the summary statistics and histogram with a normal curve
that you specified via the menus:
4. To run this block of code, click on Run/All (Figure 10.30), and it will process as if
you had clicked on the OK button.
5. When working with larger Syntax programs, you may choose to run only a section
of the code—in that case, highlight the lines of code that you want to execute, then
click on the Selection icon (the triangle icon just to the right of the binoculars icon).
Figure 10.30 Variable View Screen After Text File Import

Each time you click on the Paste button on a menu, SPSS will assess the parameters
that you specified on the menu involved, convert that to SPSS Syntax code, and add it to
the bottom of the accumulating Syntax file—essentially, you are building a program one
block of code at a time.
When you save the Syntax code, it will be assigned the .sps suffix to the file name; you
have probably already noticed that SPSS data files have the .sav file name suffix.
To better comprehend the power and convenience of using SPSS Syntax programs,
imagine that you are involved in a long-term study wherein data are collected regularly,
and each week, you need to run an ANOVA report. If you opted for an SPSS Syntax
program to carry out your tests, you could use the following three-step procedure:
1. Load Chapter 10 – Example 04 – Syntax.sav (if it is not already loaded).
2. Load Chapter 10 – Example 04 – Syntax.sps (this is the Syntax program).
3. In the Syntax editor window, click on Run/All.
This SPSS Syntax program, which was derived using the Paste function, automatically
carries out all tests, specifying each parameter along the way to run the three-group
ANOVA, as opposed to manually going to multiple menus and entering all corresponding
parameters to carry out the following steps (note that Steps 1–6 pertain to the pretest
checklist):
1. Select group 1.
2. Run a histogram with a normal curve for score.
3. Select group 2.
4. Run a histogram with a normal curve for score.
 5. Select group 3.
6. Run a histogram with a normal curve for score.
7. Select all groups.
8. Run an ANOVA specifying that the factor is group and the dependent list is score,
with a homogeneity of variance test, descriptive statistics, and a Tukey post hoc test.

Take a moment to scroll through the Syntax file—examine the code. Notice the
comments; these comments are manually entered into the Syntax file by the user to help
make the code more readable. It is good practice to use comments at the top of the
program to identify who wrote the program, the version of the program, the data file that it
is associated with, and a brief synopsis of what the program does. If you want to better
appreciate the value of comments, try deleting them and see how readable the code looks;
it becomes less comprehensible and, in turn, less maintainable.
There are a few further points regarding the use of comments:
• The processor will ignore everything after the word comment until it encounters a
period, so be sure to put a period at the end of each comment. Without the period at
the end of the comment, the processor will ignore all further code until a period is
encountered.
• Comments do not slow down the processing; the system simply ignores them. As a
rule, comments help to make your programs clearer to you, making the code easier
to comprehend and, thereby, maintain and edit.
• You can also use as many blank lines as you want to separate blocks of code; this
will help you see where one procedure ends and the next begins, which can help
with debugging.
The SPSS Syntax language is a fully developed programming language that extends
well beyond the blocks of code that you can generate by clicking on the Paste button and
the comment function. Most but not all menus in SPSS contain the Paste button; hence, if
you opt for the SPSS Syntax language to run your jobs, occasionally you may need to do
some manual coding.
For a more comprehensive list of the commands and functions available in the SPSS
Syntax language, search the SPSS Help menu for Syntax.

GOOD COMMON SENSE

It is good practice to save an unedited master copy of your source data sets (e.g.,
Filename[MASTER].sav) and only work on a copy of the file (e.g.,
Filename[WORK].sav). Even for the most careful person, data loss or data corruption is
just an accidental keystroke away.
 Try to avoid deleting records; a better practice is to use Select Cases to rule them out of
the analyses.
If you have source data (e.g., survey cards, written records, recordings, etc.), keep them
intact and stored in a secured fashion. Do not destroy or discard source media just because
the information has been entered into a database; you may need to refer to such materials
over the course of your investigation (e.g., to resolve discrepancies, verify that some
anomalous data are coded properly, etc.). Occasionally, such references to source materials
take place during the analytic process as outliers, and other reasonably questionable data
begin to emerge in reports or in visual inspection of the coded data.
All hardware fails at some point; therefore, it is good practice to keep secured/encrypted
up-to-date backups of data files in at least two safe locations (e.g., a copy at home and
another copy at work), unless your research protocol prohibits this. The rational for two
sites is to offer additional protection against unforeseen adverse events at a site (e.g., fire,
water damage, natural disaster, theft, electrical problems, etc.).
Some research protocols specify how long the research team is required to retain the
source data (e.g., surveys, field notes, recordings, etc.). The time frame may range from
days to years. The research protocol may also indicate how such data are to be disposed of
on the specified expiration date (e.g., shredding, purging media, transfer to a secured
storage site, etc.) so as to facilitate participant confidentiality. As a responsible researcher,
be sure you fully understand and adhere to these protocols.

Key Concepts
• Generating random numbers
• Sorting cases
Ascending
Descending
• Selecting cases
• Recoding
• Importing data
• Syntax
Paste
Manual edits
• Master file
• Work file
• Data handling
• Data storage
• Data disposal
 Practice Exercises
1. You have been given a data set reflecting the baseline test results of a group
of people administered a pretest prior to a training.
Data set—load Chapter 10 – Exercise 01.sav.
Codebook—the data set contains three variables:
• name is a string variable (participant’s name).
• score is a continuous variable (participant’s pretest score).
• skill is a categorical variable (1 = novice, 2 = intermediate, 3 = expert).
a. Sort the data alphabetically by name.
b. Sort the data from highest to lowest score; in cases where the score is a
tie, subsort alphabetically by name.
2. You have been given a sample frame with record numbers ranging from 852
through 5723; the research team wants to gather a 2% sample (n = 97). Use SPSS to
generate this list of random numbers.
3. You have been requested to compute statistics using only certain records
within a data set.
Data set—load Chapter 10 – Exercise 03.sav (this is the same data set as Chapter
10 – Exercise 01.sav).
Codebook—the data set contains three variables:
• name is a string variable (participant’s name).
• score is a continuous variable (participant’s pretest score).
• skill is a categorical variable (1 = novice, 2 = intermediate, 3 = expert).
Compute descriptive statistics and histograms with a normal curve for score for:
a. participants with a skill level that is intermediate.
b. participants with a score between 20 and 65.
4. You have been given two non-SPSS data sets to import into SPSS and
process:
a. Data set—import the Excel file: Chapter 10 – Exercise 04 A.xls into SPSS.
NOTE: The import utility does not know what value labels are involved in the
(categorical) group variable; it will just bring in the 1s and 2s.
Codebook—the data set contains two variables:
• group is a categorical variable (1 = control, 2 = treatment).
• score is a continuous variable (participant’s score).
Compute a t test on score using group as the grouping variable.
b. Data set—import the ASCII file: Chapter 10 – Exercise 04 B.txt into SPSS.
Codebook—the data set contains three variables:
• ID is a numeric variable (the participant’s ID).
• pretest is a continuous variable (pretest score).
• posttest is a continuous variable (posttest score).
Compute a paired t test using pretest and posttest.
5. Prior to a training, those who enrolled were given a pretest to determine their
baseline knowledge of the subject. Those who scored 75 or higher on a pretest will
be issued a pass and will be excused from the training session.
Data set—load Chapter 10 – Exercise 05.sav (this is the same data set as Chapter
10 – Exercise 01.sav).
Codebook—the data set contains three variables:
• name is a string variable (participant’s name).
• score is a continuous variable (participant’s pretest score).
• skill is a categorical variable (1 = novice, 2 = intermediate, 3 = expert).
a. Use the Recode function to create a new variable (based on score) that
identifies those who passed and those who failed.
b. Run a t test comparing the scores of those who passed with those who
failed.
6. Acme Research Labs gathers data on an ongoing basis; you have been asked
to automate the analytic process for a data set that is updated each week.
Data set—load Chapter 10 – Exercise 06.sav.
Codebook—the data set contains four variables:
• id is a string variable (participant’s ID).
• reading is a continuous variable (reading score).
• writing is a continuous variable (writing score).
• math is a continuous variable (math score).
a. Use the Paste function to create an SPSS Syntax file that performs the
following operations:
• Performs a multilevel sort:
1. reading (ascending)
2. writing (descending)
3. math (ascending)
• Runs descriptive statistics with histograms and normal curves for reading
and math
 b. Save the SPSS Syntax file as Weekly Report – 3 Subject Test.sps (this
SPSS Syntax file will be used in Exercise 7).
7. Open the SPSS Syntax file created in Exercise 6 (Weekly Report – 3 Subject
Test.sps) and (manually) add in comments detailing:
• Appropriate header information (name of source data set, name of
programmer, version number, and summary of what this SPSS Syntax
program does)
• Notes explaining what each block of code does
8. Explain what is meant by a master file and the rationale for safely preserving
it.
9. Explain what is meant by a work file and the rationale for performing
analyses on it.
10. Data safety and confidentiality are essential in the realm of research and analysis.
Discuss the rationale and techniques for appropriate data handling, data storage, and
data disposal.
 PART IV
Solutions to Odd-Numbered Exercises
NOTE: Your statistical findings and figures should match with the solutions; however, it is
expected that your written responses will vary somewhat from the solutions, but they
should concur conceptually.
 CHAPTER 1
Research Principles
(1a) Will 30 minutes of square dancing, 5 days a week, help to reduce pediatric
weight?
(1b) This would be a two-group study. The control group would have recess as usual
with no structured activities; the kids can do whatever they want (except
participate in the square dancing). The experimental group would participate in
the square dancing.
(1c) The briefing sheet that would be distributed to teachers would instruct each
teacher to go through his or her roll list and flip a coin one time for each student:
Heads assigns the student to the square dancing, and tails assigns the student to
have regular recess.
(1d) H0: Aerobic square dancing 30 minutes a day, 5 days a week facilitates no weight
loss among elementary school students.
H1: Aerobic square dancing 30 minutes a day, 5 days a week facilitates weight
loss among elementary school students.
(1e) At the conclusion of the study, the school nurse will weigh each student. The
statistician will compare the weights of those who participated in the aerobic
square dancing against those who had regular recess. If there is no statistically
significant difference between the weights of these two groups, then we would
accept H0; otherwise, we would reject H0 in favor of H1.
(3a) Will placing a security camera on cashiers reduce cash shortages?
(3b) This will be a two-group study. The control group will have no cameras
installed. The experimental group will have a camera installed focused on each
cashier, and all cashiers will be notified that their actions are now being recorded.
(3c) The name of each of the 10 stores will be written on a chip and placed into a bag.
The bag will be sealed, shaken, and then opened; a staff member will reach into
the bag and, without looking, withdraw five chips. The stores indicated on these
five chips will constitute the experimental group; the other five stores will
constitute the control group. [NOTE: Since this is a two-group design, one could
have simply used the coin-flip method; the chip selection is one potential
alternative to consider when there are more than two groups, as will be the case
in Chapter 6.]
(3d) H0: Video recording will have no effect on cashier balances.
H1: Video recording will reduce cashier losses.
(3e) The statistician will gather data from all 10 stores and compare cash register
losses from the stores with no cameras to the stores with cameras. If there is no
 statistically significant difference between the cash losses of these two groups,
then we would accept H0; otherwise, we would reject H0 in favor of H1.
(5a) Will entering on-time employees in a weekly get out of Friday free lottery reduce
morning lateness?
(5b) Two of the four buildings should participate in the get out of Friday free lottery
—these sites will serve as the experimental groups. The other two should
continue business as usual, serving as the control groups.
(5c) A staff member writes the name or address of each of the four buildings on four
identical cards and places them in a box. Without looking, the staff member
draws two cards out of the box, one at a time. These two cards will constitute the
experimental groups; the other two will serve as the control groups.
(5d) H0: Rewarding on-time arrivals with the chance to win a free day off will have no
effect on lateness.
H1: Rewarding on-time arrivals with the chance to win a free day off will reduce
lateness.
(5e) After running this lottery program for a month, the statistician will gather and
analyze the time cards from each of the four buildings and compare the minutes
late from those in the two buildings that had no day-off lottery (in the control
groups) with those who did participate in the day-off lottery (in the experimental
group). If there is no significant difference in the minutes late, comparing the
employees in the control group with those in the experimental group, then we
would accept H0; otherwise, we would reject H0, in favor of H1.
(7a) Does singing 1 hour a day improve memory?
(7b) Members in the control group will sit quietly for an hour and then take a memory
test. Those in the experimental group will sing well-known songs for 1 hour
(karaoke style) and then take the memory test.
(7c) Professor Madrigal will recruit 30 participants for this experiment. Prior to the
arrival of the first participant, the professor flipped a coin (heads for control
group, tails for experimental group)—the coin flip rendered tails, and hence the
first participant who shows up will be assigned to the experimental group. Each
remaining subject would be assigned to the control or experimental group on an
alternating basis, upon arrival, thereby keeping the groups balanced.
(7d) H0: Singing has no effect on memory.
H1: Singing enhances memory.
(7e) Upon the departure of the last subject, the professor compares the memory test
scores of those in the control group with those in the experimental group. If there
is no statistically significant difference in the scores, then this supports H0;
otherwise, the professor would reject H0 in that the data support H1.
(9a) Do flashcards help students memorize the multiplication table?
(9b) Members in the experimental group will each be issued a set of 100 flashcards (1
× 1… 10 × 10) and will be instructed to work with them for 30 minutes a day for
1 month. Members in the control group will use the usual multiplication teaching
method (only). Members of both groups will be tested once a week; they will be
given 10 minutes to answer 100 multiplication problems (1 × 1 = ___… 10 × 10
= ___).
(9c) Ms. Fractal calls each student to the front of the room one at a time; she holds up
a coin and asks the student to guess heads or tails. She then flips a coin; if the
student guessed correctly, then that student is assigned to the experimental group
and given a set of multiplication flashcards; otherwise, the student is assigned to
the control group and receives no flashcards.
(9d) H0: Flashcards do not help students learn the multiplication table.
H1: Flashcards help students learn the multiplication table.
(9e) Ms. Fract al will grade the tests. If the experimental group’s overall score is
statistically significantly higher than the control group’s score, then she would
reject the null hypothesis (H0), in favor of the experimental hypothesis (H1);
otherwise, the findings suggest support for the null hypothesis (H0).
 CHAPTER 2
Sampling
(1a) In terms of time, gathering a sample of an entire population could take weeks,
months, or possibly years depending on the size of the population, geographical
area involved, and complexity of the data to be gathered. The point of research is
to get the answer to a meaningful question, presumably to solve a problem or
improve a situation. Such answers can be expedited by sampling, as opposed to
gathering data on the entire population. Attempting to sample the entire
population would be so time-consuming that by the time you had acquired your
full data set, the nature of the initial problem may have changed substantially,
thereby making your findings virtually irrelevant.
(1b) No matter what method of gathering data one uses (e.g., in-person surveys,
phone surveys, postal/e-mail contact, experimental designs, tests, etc.), costs are
involved. When it comes to data collection, subject participation fees are
customary, data collection teams need to be paid, and there are administrative
costs involved (photocopying, office space, equipment, office supplies, etc.).
Considering the volume of individuals that would be involved in studying an
entire population, the costs would be prohibitive.
(1c) Beyond the lengthy time and exorbitant costs that would be involved in studying
an entire population, it is seldom feasible to gather data on so many people. The
population may span a broad geographical region, requiring lengthy travel
among the research staff. Within a population, one would likely encounter a
variety of languages; it may not be possible to translate verbal or written
instructions to attain a full population sample. In addition, some studies may
involve experiments that can be administered only at a special facility; it would
be impossible to arrange a round-trip commute for every member of a population
to that facility to participate.
(1d) Extrapolation involves working diligently to gather a representative sample in
sufficient quantities to facilitate stable statistical processing. If the data gathered
constitute a representative sample, then one could (more) confidently extrapolate
the findings to (better) comprehend the overall population from which it was
drawn; this is also known as external validity.
(3a) The population is the entire student body currently enrolled (e.g., 22,000
students).
(3b) The sample frame is the available list of students currently enrolled (e.g., 16,000
students).
(3c) Fortunately, the list of participants in the sample frame was already numbered
from 1 to 16,000. I would want to gather data on 2% of the sample frame (16,000
× .02 = 320); I would use SPSS to generate 320 random numbers between 1 and
16,000—this list would indicate which students in the sample frame to recruit.
 Actually, I would probably generate more than 320 random numbers just in case
SPSS produces some duplicates.
(3d) I would build (and test) an online survey. Next, I would send an e-mail to each of
the 320 students in the sample set; the e-mail would concisely explain the nature
of the study and courteously request that they voluntarily click on the link to the
online survey.
(5a) The population is all of the people who enter the library.
(5b) The sample frame consists of those who enter the Reference Room.
(5c) I would approach each person accessing the reference section throughout the day
and ask if they would be willing to spend a few minutes answering some simple
questions about how they use the library.
(5d) I would use a clipboard with a prepared questionnaire. To satisfy the strata
(children/adults), the first question would be, “How old are you?”
(7a) I wouldn’t want to bother people who are walking someplace or who are
engaged in conversations with friends/family, so I would approach people
waiting in line (for food or an attraction) since they’d probably be a bit bored just
waiting in line.
(7b) I would ask them their first name, just to be friendly, and then tell them that I
work for Acme Research Firm (I’d point to my badge) and tell them that I’m
gathering some information about how much money they intend to spend today
and ask them if they wouldn’t mind estimating that amount for me. Whether they
choose to respond or not, I’d thank them.
(9a) I would ride the bus starting at 8:00 a.m. At first I’d approach anyone. If they
consented to talk to me, I would begin by asking them their age. Once I’d hit my
quota for a group (50 minors, 100 adults), I’d stop approaching that group and
proceed with the other group. When I’d gathered all 150 surveys, I’d stop the
sampling process completely and move on to the analysis (off the bus).
(9b) I would approach potential participants, identify myself as a staff member of the
Acme Bus Company (I’d point to my badge), and ask the person if I could ask
them a few questions about their impressions of their bus riding experience(s). I
would record their responses on a clipboard, which would contain the printed
survey. The first question I would ask is “What is your age?” which I would
record on the survey form; if they didn’t want to tell me their age, I would ask
“Are you 18 or over?” unless it was obvious that the person was clearly either a
child or an adult. This way I could keep a running count of how many minors
and how many adults I had surveyed.
 CHAPTER 3
Working in SPSS
(1)
(3)
(5)
(7)
(9)
 CHAPTER 4
Descriptive Statistics
(1a)

(1b)
(1c)
(1d)

(3a)
(3b)

(3c)
(3d)
(5a)

(5b)
(5c)
(5d)

(5e)
(7a)
(7b)

(7c)
(7d)
(7e)

(9a)
(9b)
(9c)

(9d)
 CHAPTER 5
t Test
1. Data Set B
(1a) H0: Practicing meditation for 30 minutes a day, 3 days a week for 2 weeks has no
effect on resting pulse rate.
H1: Practicing meditation for 30 minutes a day, 3 days a week for 2 weeks affects
resting pulse rate.
(1b) Histograms with normal curve plots show a normal distribution of pulse for both
groups, as shown in the two figures below; hence, the pretest criterion of
normality is satisfied.
Normal distribution for pulse in Group 1 (no meditation)

Normal distribution for pulse in Group 2 (meditated 30 minutes a day, 3 days per week)
 The homogeneity of variance score shows a significance (p) of .766; since this is greater
than the α level of .05, this suggests that there is no statistically significant difference
between the variances of the two groups; hence, this pretest criterion passes.
The n for each group, as shown in the Descriptives table below, is 35 for each group;
since the ns are greater than 30, this criterion passes also.
(1c) The t test revealed the following:

The mean pulse rate for Group 1 (no meditation) is 97.40, whereas the mean pulse rate
for Group 2 (meditated 30 minutes, 3 days per week) is 92.20. This 5.2-point difference is
statistically significant since the significance (p) is .003 (which is less than the .05 α
level).
(1d) This study analyzed the effects that meditation had on resting pulse rates. The
subjects were randomly assigned to one of two groups: the group that did not
 meditate and the other group that meditated for 30 minutes on Monday,
Wednesday, and Friday for 2 weeks. Results revealed a mean resting pulse rate of
97.40 for those who did not meditate and 92.20 for those who meditated. Using a
.05 α level, the p value of .003 suggests that meditation facilitates a significant
reduction in resting pulse rate; hence, we reject H0. These findings suggest
support for H1—specifically, that practicing meditation for 30 minutes a day, 3
days a week for 2 weeks affected the mean resting pulse rate among these
participants.
1. Data Set B
(1a) H0: Practicing meditation for 30 minutes a day, 3 days a week for 2 weeks has no
effect on resting pulse rate.
H1: Practicing meditation for 30 minutes a day, 3 days a week for 2 weeks affects
resting pulse rate.
(1b) Histograms with normal curve plots show a normal distribution of pulse for both
groups, as shown in the two figures below; hence, the pretest criterion of
normality is satisfied.
The homogeneity of variance score for pulse shows a significance (p) of .576; since this
is greater than the α level of .05, this suggests that there is no statistically significant
difference between the variances of the two groups; hence, this pretest criterion passes.
The n criterion is satisfied as both groups have an n of more than 30 (see Descriptives
table below).
Normal distribution for pulse in Group 1 (no meditation)

Normal distribution for pulse in Group 2 (meditated 30 minutes a day, 3 days per
week)
 (1c) The t test revealed the following:

The mean pulse rate for Group 1 (no meditation) is 95.15, whereas the mean pulse rate
for Group 2 (meditated 30 minutes, 3 days per week) is 93.54. Even though the mean
pulse rate for the meditation group is 1.61 points lower than the control group, this
difference is not considered statistically significant since the significance (p) is .302
(which is greater than the .05 α level).
(1d) This study analyzed the effects that meditation had on resting pulse rates. The
participants were randomly assigned to two groups: those who did not meditate
and those who meditated for 30 minutes on Monday, Wednesday, and Friday for
2 weeks. Results revealed a mean resting pulse rate of 95.15 for those who did
not meditate and 93.54 for those who did meditate. This study showed a 1.61
reduction in the pulse rate of those who meditated, but since the p value of .302
is greater than the .05 α level, we would conclude that this difference is not
statistically significant; hence, we do not reject H0. For the subjects studied,
 meditation did not significantly affect resting pulse rate (we would reject H1).
3. Data Set A
(3a) H0: The Acme reading lamp is no different from regular room lighting when it
comes to reading speed.
H1: The Acme reading lamp facilitates faster reading speed than regular room
lighting.
(3b) Histograms with normal curve plots show a normal distribution of seconds for
both groups, as shown in the two figures below; hence, the pretest criterion of
normality is satisfied.
Normal distribution for seconds in Group 1 (room lighting)

The homogeneity of variance score for seconds shows a significance (p) of .643; since
this is greater than the α level of .05, this suggests that there is no statistically significant
difference between the variances of the two groups; hence, this pretest criterion passes.
The room lighting group had an n of 25, and the Acme lamp group had an n of 28 (see
Descriptives table below); these figures are close to the minimal quota of 30 per group.
The findings of the t test would be more robust if the ns were slightly higher for these
groups.
Normal distribution for seconds in Group 2 (Acme reading lamp)
 (3c) The t test revealed the following:

The mean read time (seconds) rate for Group 1 (room lighting) is 436 seconds
(rounded), whereas the mean read time for Group 2 (Acme reading lamp) is 406 seconds
(rounded). This 30-second difference is statistically significant since the significance (p) is
.003 (which is less than the .05 α level).
(3d) This study analyzed the effects that the Acme reading lamp had on reading
speed. The 53 subjects were randomly assigned to one of two groups; one group
read a 1,000-word essay using regular room lighting, and the other group read
the same essay using the new Acme reading lamp. Results revealed that on
average, those who read using the Acme reading lamp completed the essay 30
seconds earlier than those who used regular room lighting (406 seconds vs. 436
seconds, respectively). Using a .05 α level, the p value of .003 suggests that the
Acme reading lamp facilitates prompter reading speeds; hence, we reject H0.
 These findings suggest support for H1—specifically, that the Acme reading lamp
enhances reading rates.
3. Data Set B
(3a) H0: The Acme reading lamp is no different from regular room lighting when it
comes to reading speed.
H1: The Acme reading lamp facilitates faster reading speed than regular room
lighting.
(3b) Histograms with normal curve plots show a normal distribution of seconds for
both groups, as shown in the two figures below; hence, the pretest criterion of
normality is satisfied.
The homogeneity of variance score for seconds shows a significance (p) of .380; since
this is greater than the α level of .05, this suggests that there is no statistically significant
difference between the variances of the two groups; hence, this pretest criterion passes.
The n for each group is 25 (see Descriptives table below), which is close to the minimal
quota of 30 per group. The findings of the t test would be more robust if the ns were
slightly higher.
Normal distribution for seconds in Group 1 (room lighting)

Normal distribution for seconds in Group 2 (Acme reading lamp)

 (3c) The t test revealed the following:

The mean read time (seconds) rate for Group 1 (room lighting) is 416.44 seconds,
whereas the mean read time for Group 2 (Acme reading lamp) is 415.60 seconds. This
.84-second difference is statistically insignificant since the significance (p) is .935 (which
is greater than the .05 α level).
(3d) This study analyzed the effects that the Acme reading lamp had on reading
speed. The 50 subjects were randomly assigned to one of two groups; half read a
1,000-word essay using regular room lighting, and the other half read the same
essay using the new Acme reading lamp. Results revealed that on average, those
who read using the Acme reading lamp completed the essay about 1 second (.86
seconds) earlier than those who used regular room lighting (415.60 seconds vs.
416.44 seconds, respectively). Using a .05 α level, the p value of .935 suggests
 that the Acme reading lamp does not facilitate significantly prompter reading
speeds; hence, we do not reject H0, but we do reject H1. The claim that the Acme
Company made, that this lamp increases reading speed, is not supported by these
findings.
5. Data Set A
(5a) H0: Classical music has no effect on problem-solving skills.
H1: Classical music enhances problem-solving skills.
(5b) Histograms with normal curve plots show a normal distribution of seconds for
both groups, as shown in the two figures below; hence, the pretest criterion of
normality is satisfied.
Normal distribution for seconds in Group 1 (no music)

The homogeneity of variance score for seconds shows a significance (p) of .095; since
this is greater than the α level of .05, this suggests that there is no statistically significant
difference between the variances of the two groups; hence, this pretest criterion passes.
The n for each group is 40 (see Descriptives table below), which satisfies the 30 per
group minimum criterion.
Normal distribution for seconds in Group 2 (music at 30 dB)
 (5c) The t test revealed the following:

The mean puzzle completion time for Group 1 (no music) is 620 seconds (rounded),
whereas the subjects in Group 2 (music at 30 dB) had a mean completion of 594 seconds
(rounded). This 26-second difference is statistically significant since the significance (p) is
.008 (which is less than the .05 α level).
(5d) This experiment was designed to determine if classical music facilitates problem
solving. We recruited 80 subjects; half were randomly assigned to solve a 100-
piece jigsaw puzzle in a silent room, and the other half solved the same puzzle in
a room with soft classical music playing (at 30 dB). Those in the music group
solved the puzzle an average of 26 seconds faster than those who worked in
silence (594 seconds vs. 620 seconds). Using a .05 α level, the p value of .008
suggests that this is a statistically significant difference; hence, we reject H0.
 These findings support H1—that classical music facilitates problem solving.
5. Data Set B
(5a) H0: Classical music has no effect on problem-solving skills.
H1: Classical music enhances problem-solving skills.
(5b) Histograms with normal curve plots show a normal distribution of seconds for
both groups, as shown in the two figures below; hence, the pretest criterion of
normality is satisfied.
The homogeneity of variance score for seconds shows a significance (p) of .235; since
this is greater than the α level of .05, this suggests that there is no statistically significant
difference between the variances of the two groups; hence, this pretest criterion passes.
The n for each group is over 30: 43 for the no music and 39 for music at 30 dB (see
Descriptives table below), which satisfies the 30 per group minimum criterion.
Normal distribution for seconds in Group 1 (no music)

Normal distribution for seconds in Group 2 (music at 30 dB)

 (5c) The t test revealed the following:

The mean puzzle completion time for Group 1 (no music) is 609 seconds (rounded),
whereas the subjects in Group 2 (music at 30 dB) had a mean puzzle completion time of
606 seconds (rounded). This difference is statistically insignificant since the significance
(p) is .323 (which is greater than the .05 α level).
(5d) This experiment was designed to determine if classical music aids in problem
solving. We recruited 82 participants and randomly assigned them to one of two
groups: those who assembled a 100-piece jigsaw puzzle in a silent room and
those who assembled the same puzzle in a room with soft classical music playing
(at 30 dB). On average, participants in the music group completed the puzzle 2.6
seconds faster than those who worked in silence (606.33 seconds vs. 608.93
 seconds), but using a .05 α level, the p value of .323 suggests that this difference
is not statistically significant; hence, we do not reject H0 but reject H1. For those
tested, it appears that classical music is not a substantial factor when it comes to
problem solving.
7. Data Set A
(7a) H0: Light therapy has no effect on depression.
H1: Light therapy is effective in reducing depression.
(7b) Histograms with normal curve plots show a normal distribution of mood for both
groups, as shown in the two figures below; hence, the pretest criterion of
normality is satisfied.
Normal distribution for mood in Group 1 (no light therapy)

The homogeneity of variance score for mood shows a significance (p) of .194; since this
is greater than the α level of .05, this suggests that there is no statistically significant
difference between the variances of the two groups; hence, this pretest criterion passes.
The n for each group is 80, which satisfies the 30 per group minimum criterion (see
Descriptives table below).
Normal distribution for mood in Group 2 (light therapy: even days)
 (7c) The t test revealed the following:

The mean mood level for those in the light therapy group was 61.88, which is 2.75
points higher than the mean score of those in the control group (59.13). In light of the
significance (p) score of .025 (which is less than the .05 α level), this difference is
considered statistically significant.
(7d) To determine if light therapy is a viable supplement to treating depression, 160
subjects diagnosed with depression were randomly assigned to one of two
groups: Half received 1 hour of light therapy every other day for a month; the
other half received no light therapy. After 30 days, all participants completed the
Acme Mood Scale, a 10-question survey that renders a score from 1 to 100 (1 =
extremely bad mood… 100 = extremely good mood). Those who received light
therapy showed a small but statistically significant improvement in mood; light
therapy participants had an average mood score of 61.88, which is 2.75 points
higher than those who did not have the light therapy, who scored an average of
 59.13 on the mood test. Using a .05 α level, the p value of .025 suggests that this
difference is statistically significant; hence, we reject H0. For those tested, it
appears that light therapy provided some relief from depressive symptoms, hence
supporting H1.
7. Data Set B
(7a) H0: Light therapy has no effect on depression.
H1: Light therapy is effective in reducing depression.
(7b) Despite the spike in the histogram for the no light therapy group (the tall bar at
around 42), the histograms with normal curve plots show a normal distribution of
mood for both groups, as shown in the two figures below; hence, the pretest
criterion of normality is satisfied.
The homogeneity of variance score for mood shows a significance (p) of .731; since this
is greater than the α level of .05, this suggests that there is no statistically significant
difference between the variances of the two groups; hence, this pretest criterion passes.
The n for the groups are 48 and 56 (see Descriptives table below); these figures exceed
the 30 per group minimum criterion.
Normal distribution for mood in Group 1 (no light therapy)

Normal distribution for mood in Group 2 (light therapy: even days)

 (7c) The t test revealed the following:

The mean mood level for those in the control group (no light therapy) was 45.44,
whereas those in the treatment group (light therapy: even days) had a mean mood level of
44.86. Unexpectedly, the mood level for those who received no light therapy was .58
points higher than that of those who did, but since the significance (p) is .430 (which is
greater than the .05 α level), this difference is not considered statistically significant.
(7d) To determine if light therapy is a viable supplement to treating depressed
individuals, 104 participants with a diagnosis of depression were randomly
assigned to one of two groups: In addition to their regular care, 56 subjects
received light therapy for 1 hour every other day; the other 48 participants
received none. After 30 days, all participants completed the Acme Mood Scale, a
10-question survey that renders a score from 1 to 100 (1 = extremely bad
 mood… 100 = extremely good mood). Contrary to expectations, on average,
those in the light therapy group scored 44.86, about a half point (.58) lower than
those who had no light therapy, who had an average mood score of 45.44. Using
a .05 α level, the p value of .430 suggests that this difference is not statistically
significant; hence, we do not reject H0, but we do reject H1. For those tested, it
appears that light therapy did not have a significant impact on their mood.
9. Data Set A
(9a) H0: Tending to a plant has no effect on depressive mood.
H1: Tending to a plant reduces depressive mood.
(9b) Despite the few low-score outliers shown in the histogram for Group 2, the
histograms with normal curve plots show a normal distribution of the depress
variable for both groups, as shown in the two figures below; hence, the pretest
criterion of normality is satisfied.
Normal distribution for depress in Group 1 (no plant)

The homogeneity of variance score for mood shows a significance (p) of .060; since this
is greater than the a level of .05, this suggests that there is no statistically significant
difference between the variances of the two groups; hence, this pretest criterion passes.
The n for each group is 60 (see Descriptives table below), which satisfies the 30 per
group minimum criterion.
Normal distribution for depress in Group 2 (bamboo)
 (9c) The t test revealed the following:

The mean depression level for those in the control group (no plant) was 19.60, whereas
those in the treatment group (bamboo) had a mean depression level of 18.05. Since the
significance (p) is .001 (which is less than the .05 α level), the 1.55-point mood
improvement in those who were given plants is considered statistically significant. As
such, we would reject H0 in favor of H1.
(9d) We hypothesized that empowering nursing home residents with an opportunity to
provide nurturance would help to reduce depression. To test this hypothesis, 120
residents were randomly assigned to one of two groups: The 60 people in the
treatment group were each given a small bamboo plant to tend to along with a
card providing care instructions; the 60 members of the control group were given
no plant. After 90 days, we administered the Acme Depression Scale (1 = low
 depression… 100 = high depression) to members of both groups. We found that
those who were given the bamboo plant scored an average of 18.05; their
depression level was 1.55 points lower than those in the control group, who had
an average depression score of 19.60. This improvement in depression, although
small, produced a statistically significant p value of .001, using a .05 α level;
hence, we rejected H0. For those involved in this study, it appears that having a
plant reduced depression, thereby supporting H1.
9. Data Set B
(9a) H0: Tending to a plant has no effect on depressive mood.
H1: Tending to a plant reduces depressive mood.
(9b) The histograms with normal curve plots show a normal distribution of the
depress variable for both groups, as shown in the two figures below; hence, the
pretest criterion of normality is satisfied.
The homogeneity of variance score for mood shows a significance (p) of .156; since this
is greater than the α level of .05, this suggests that there is no statistically significant
difference between the variances of the two groups; hence, this pretest criterion passes.
The n for the no-plant group is 58, and the n for the bamboo group is 51 (see
Descriptives table below). These ns exceed the 30 per group minimum criterion.
Normal distribution for depress in Group 1 (no plant)

Normal distribution for depress in Group 2 (bamboo)

 (9c) The t test revealed the following:

The mean depression level for those in the control group (no plant) was 15.69, whereas
those in the treatment group (bamboo) had a mean depression level of 16.37. Even though
the mean depression score for those who received the bamboo plant was .68 points higher
than those in the control group, ultimately, this difference is not statistically significant
since the significance (p) is .364 (which is greater than the .05 α level). As such, we would
not reject H0, but we would reject H1.
(9d) We hypothesized that empowering nursing home residents with an opportunity to
provide nurturance would help to reduce depression. To test this hypothesis, 109
residents were randomly assigned to one of two groups: 51 of the residents were
given a small bamboo plant to tend to along with a card providing care
instructions; the remaining 58 residents received no plant. After 90 days, we
administered the Acme Depression Scale (1 = low depression… 100 = high
depression) to members of both groups; we found that on average, those who
were given no plant were less depressed (15.69) than those who were given the
bamboo plant (16.37), but in light of the p value of .364, using a .05 α level, this
.68-point difference in the average depression scores is not considered
statistically significant. Accordingly, we did not reject H0.
 CHAPTER 6
ANOVA
1. Data Set B
(1a) H0: Practicing meditation has no effect on resting pulse rate.
H1: Practicing meditation for 30 minutes a day, 3 days a week for 2 weeks affects
resting pulse rate.
H2: Practicing meditation for 30 minutes a day, 6 days a week for 2 weeks affects
resting pulse rate.
(1b) Histograms with normal curve plots show a normal distribution of pulse for the
groups, as shown in the three figures below; hence, the pretest criterion of
normality is satisfied.
Normal distribution for pulse in Group 1 (no meditation)

Normal distribution for pulse in Group 2 (meditated 30 minutes a day, 3 days per week)
 Normal distribution for pulse in Group 3 (meditated 30 minutes a day, 6 days per
week)

The homogeneity of variance score shows a significance (p) of .920; since this is greater
than the α level of .05, this suggests that there is no statistically significant difference
among the variances of the three groups; hence, this pretest criterion passes.
The n for each group, as shown in the Descriptives table below, is 35 for each group;
since the ns are greater than 30, this criterion passes also.
(1c) The ANOVA revealed the following:
 The Tukey post hoc test was used since the ns for each group were the same (35 each).
NOTE: Since the ANOVA test renders results involving multiple comparisons, it may be
helpful to organize the findings as shown in the table below. SPSS does not generate this
table directly, but you can construct it manually by copying the group names and means
from the Descriptives table, as well as the p values from the Sig. column in the Multiple
Comparisons table.

Practicing meditation for 2 weeks was effective in statistically significantly reducing

resting pulse rate, but there was no statistically significant difference in resting pulse rate
when comparing those who meditated for 30 minutes three times a week with those who
meditated for 30 minutes six times a week.
On the basis of these findings, we reject H0, but we do not reject H1 or H2.
(1d) This study analyzed the effects that meditation had on resting pulse rates. The
subjects were randomly assigned to one of three groups. Members of the control
 group did not meditate; those in the second group meditated for 30 minutes on
Monday, Wednesday and Friday; and members of the third group meditated for
30 minutes Monday through Saturday. After 2 weeks, those who meditated
(3×/wk, μ = 92.20; 6×/wk, μ = 91.40) showed a statistically significant reduction
(p = .007 and p = .001, respectively) in resting pulse rate compared with those
who did not meditate (μ = 97.40) using a .05 α level. We found no statistically
significant difference in the resting pulse rates between those who meditated 3
days per week compared with those who meditated 6 days per week (p = .881).
1. Data Set B
(1a) H0: Practicing meditation has no effect on resting pulse rate.
H1: Practicing meditation for 30 minutes a day, 3 days a week for 2 weeks affects
resting pulse rate.
H2: Practicing meditation for 30 minutes a day, 6 days a week for 2 weeks affects
resting pulse rate.
(1b) Histograms with normal curve plots show a normal distribution of pulse for the
three groups, as shown in the three figures below; hence, the pretest criterion of
normality is satisfied.
Normal distribution for pulse in Group 1 (no meditation)

Normal distribution for pulse in Group 2 (meditated 30 minutes a day, 3 days per
week)
 Normal distribution for pulse in Group 3 (meditated 30 minutes a day, 6 days per week)

The homogeneity of variance score shows a significance (p) of .786; since this is greater
than the α level of .05, this suggests that there is no statistically significant difference
among the variances of the three groups; hence, this pretest criterion passes.
The ns for each group, as shown in the Descriptives table below, are 33, 35, and 31;
since the ns are greater than 30, this criterion passes also.
(1c) The ANOVA revealed the following:
 The Sidak post hoc test was used since the ns for each group were not all the same (ns =
33, 35, 31).

Per the table above, practicing meditation for 2 weeks produced no statistically
significant reduction in resting pulse rate using a .05 α level.
On the basis of these findings, we would not reject H0, but we would reject H1 and H2.
(1d) This study analyzed the effects that meditation had on resting pulse rates. The
subjects were randomly assigned to one of three groups. Members of the control
group did not meditate; those in the second group meditated for 30 minutes on
Monday, Wednesday, and Friday; and members of the third group meditated for
30 minutes Monday through Saturday. After 2 weeks, resting pulse rates were
recorded for each participant. Those who did not meditate had a mean resting
pulse rate of 95.15, which was slightly higher than those who meditated (3 days
per week: μ = 93.54; 6 days per week: μ = 92.68), but we detected no statistically
significant differences among any of the three groups using a .05 α level,
suggesting that the meditation schedules tested were not effective in reducing
 resting pulse rates.
3. Data Set A
(3a) H0: The Acme reading lamp is no different from any other lighting source when it
comes to reading speed.
H1: The Acme reading lamp facilitates faster reading speed than regular room
lighting.
H2: The Acme reading lamp facilitates faster reading speed than a generic
reading lamp.
H3: The Acme reading lamp facilitates faster reading speed than a flashlight.
(3b) Histograms with normal curve plots show a normal distribution of seconds for
all groups as shown in the four figures below; hence, the pretest criterion of
normality is satisfied.
Normal distribution for seconds in Group 1 (room lighting)

Normal distribution for seconds in Group 2 (Acme reading lamp)

 Normal distribution for seconds in Group 3 (generic reading lamp)

Normal distribution for seconds in Group 4 (flashlight)

 The homogeneity of variance score for seconds shows a significance (p) of .219; since
this is greater than the α level of .05, this suggests that there is no statistically significant
difference between the variances among the four groups; hence, this pretest criterion
passes.
The generic lamp group was the only group that had the minimal n of 30. The room
lighting, Acme lighting, and flashlight groups had ns of 25, 28, and 24, respectively (see
Descriptives table below). The findings of the ANOVA test would be more robust if the ns
were slightly higher for these three groups.
(3c) The ANOVA test revealed the following:
Means rounded to nearest second.
*Statistically significant difference (α = .05).

Per the table above, since the mean reading time in the Acme reading lamp group is
statistically significantly lower than that of those who read using room lighting and by
flashlight, we reject H0. For the same reason, we would not reject H1 and H3. Since there
was no statistically significant difference in the reading times from those in the Acme
reading lamp group compared with those who used a generic reading lamp, we would
reject H2.
(3d) This study analyzed the effects that the Acme reading lamp had on reading speed
compared with other light sources. The 107 participants were randomly assigned
to one of four groups. One group read a 1,000-word essay using regular room
lighting, the second group read the same essay using the new Acme reading
lamp, the third group read using a generic reading lamp, and the fourth group
read using a flashlight. Results revealed that on average, those who read using
the Acme reading lamp read significantly faster (μ = 406) than those who read
using a flashlight (μ = 441, p = .008), or regular room lighting (μ = 436, p =
.032), using an α level of .05. Incidentally, those who used a generic reading
lamp (μ = 410) finished reading the essay significantly faster than those who read
 by flashlight (μ = 441, p = .032). We also discovered that those who read using
an Acme reading lamp (μ = 406) completed the essay faster than those who used
the generic reading lamp (μ = 410), but this difference was not statistically
significant (p = .999).
3. Data Set B
(3a) H0: The Acme reading lamp is no different from any other lighting source when it
comes to reading speed.
H1: The Acme reading lamp facilitates faster reading speed than regular room
lighting.
H2: The Acme reading lamp facilitates faster reading speed than a generic
reading lamp.
H3: The Acme reading lamp facilitates faster reading speed than a flashlight.
(3b) Histograms with normal curve plots show a normal distribution of seconds for
all groups, as shown in the four figures below; hence, the pretest criterion of
normality is satisfied.
Normal distribution for seconds in Group 1 (room lighting)

Normal distribution for seconds in Group 2 (Acme reading lamp)

Normal distribution for seconds in Group 3 (generic reading lamp)

Normal distribution for seconds in Group 4 (flashlight)

 The homogeneity of variance score for seconds shows a significance (p) of .328; since
this is greater than the α level of .05, this suggests that there is no statistically significant
difference between the variances among the four groups; hence, this pretest criterion
passes.
The groups each had an n of 25 (see Descriptives table below). The findings of the
ANOVA test would be more robust if the ns were at least 30 per group.
(3c) The ANOVA test revealed the following:
Means rounded to nearest second.
*Statistically significant difference (α = .05).

Per the table above, since the mean reading time in the Acme lamp group (μ = 416) is
statistically significantly lower than that of those who read using a flashlight (μ = 449, p =
.022), based on the .05 α level, we reject H0. For the same reason, we would not reject H3.
Since comparing the Acme lamp (μ = 416) with the generic lamp (μ = 414) produced a
p value of .998, we see that there is no statistically significant difference between these
groups; we would reject H2.
Comparing the Acme lamp (μ = 416) with room lighting (μ = 416) produced a p value
of 1.000; hence, we reject H1.
In summary, it appears that all lighting facilitated about the same reading speed, except
for the flashlight; those who read using a flashlight took (statistically) significantly longer
than those in the other three groups.
(3d) This study analyzed the effects that the Acme reading lamp had on reading
speed. The 100 participants were randomly assigned to one of four groups. One
group read a 1,000-word essay using regular room lighting, the second group
read the same essay using the new Acme reading lamp, the third group read
using a generic reading lamp, and the fourth group read using a flashlight. There
 was no statistically significant difference in reading times among those who used
the Acme reading lamp (μ = 416), room lighting (μ = 416), or the generic lamp
(μ = 414), using an α level of .05. All three of these groups read statistically
significantly faster than the fourth group, who read using a flashlight (μ = 449); p
values ranged from .013 to .027.
5. Data Set A
(5a) H0: Classical music does not enhance problem-solving skills.
H1: Classical music at 30 dB enhances problem-solving skills.
H2: Classical music at 60 dB enhances problem-solving skills.
H3: Classical music at 90 dB enhances problem-solving skills.
(5b) Histograms with normal curve plots show a normal distribution of seconds for
all groups, as shown in the four figures below; hence, the pretest criterion of
normality is satisfied.
Normal distribution for seconds in Group 1 (no music)

Normal distribution for seconds in Group 2 (music at 30 dB)

 Normal distribution for seconds in Group 3 (music at 60 dB)

Normal distribution for seconds in Group 4 (music at 90 dB)

 The homogeneity of variance score for seconds shows a significance (p) of .314; since
this is greater than the α level of .05, this suggests that there is no statistically significant
difference between the variances among the four groups; hence, this pretest criterion
passes.
The n for each group is 40 (see Descriptives table below), which satisfies the 30 per
group minimum criterion.
(5c) The ANOVA test revealed the following:
Means rounded to nearest second.
*Statistically significant difference (α = .05).

Participants who assembled the puzzle with classical music at 30 dB (μ = 594)

completed the puzzle statistically significantly faster than those who had no music (μ =
620) (p = .047, α = .05).
Based on these findings, the hypotheses would be resolved as such:
Reject H0 and do not reject H1 since those who were exposed to classical music at 30
dB solved the puzzle an average of 26 seconds faster than those who worked in silence.
Reject H2 since there was no statistically significant difference between the mean
completion time of those who worked with classical music at 60 dB (μ = 596) compared
with those who worked in silence (μ = 620) (p = .077, α = .05).
Reject H3 since there was no statistically significant difference between the mean
completion time of those who worked with classical music at 90 dB (μ = 626) compared
with those who worked in silence (μ = 620) (p = .924, α = .05).
(5d) This experiment was designed to determine if classical music facilitates problem
solving. We recruited 160 people and randomly assigned participants to one of
four groups: Members of each group were given a 100-piece jigsaw puzzle to
 solve. Those in Group 1 worked in silence, and those in Groups 2, 3, and 4
solved the puzzle with classical music playing at different volumes in each
group: soft (30 dB), medium (60 dB), and loud (90 dB), respectively. We found
that those who listened to the soft classical music (at 30 dB) completed the
puzzle (statistically) significantly faster (μ = 594 seconds) than those who
worked in silence (μ = 620) (p = .047, α = .05), but it seems that as the music
was played louder, the beneficial effect was lost: Those who listened at a medium
volume completed the puzzle in 596 seconds, and those who were exposed to the
loud volume took 626 seconds. It seems that while the soft classical music helped
to facilitate the problem solving, louder volumes may have been distracting.
5. Data Set B
(5a) H0: Classical music does not enhance problem-solving skills.
H1: Classical music at 30 dB enhances problem-solving skills.
H2: Classical music at 60 dB enhances problem-solving skills.
H3: Classical music at 90 dB enhances problem-solving skills.
(5b) Histograms with normal curve plots show a normal distribution of seconds for
all groups, as shown in the four figures below; hence, the pretest criterion of
normality is satisfied.
Normal distribution for seconds in Group 1 (no music)

Normal distribution for seconds in Group 2 (music at 30 dB)

Normal distribution for seconds in Group 3 (music at 60 dB)

Normal distribution for seconds in Group 4 (music at 90 dB)

 The homogeneity of variance score for seconds shows a significance (p) of .284; since
this is greater than the α level of .05, this suggests that there is no statistically significant
difference between the variances among the four groups; hence, this pretest criterion
passes.
The n for each group is over 30 (see Descriptives table below), which satisfies the 30
per group minimum criterion.
(5c) The ANOVA test revealed the following:
Means rounded to nearest second.
*Statistically significant difference (α = .05).

There was no statistically significant difference in the participants’ time to complete the
puzzle among those who worked in silence (μ = 609) compared with those who listened to
soft classical music (μ = 606) (p = .895); hence, we reject H1.
Similarly, there was no statistically significant difference in completion time between
those who worked with no music (μ = 609) compared with those who listened to classical
music at a medium volume (μ = 608) (p = 1.000); hence, we reject H2.
Those who listened to loud classical music took (statistically) significantly longer (μ =
617) to complete the puzzle compared with those who worked in silence (μ = 609) (p =
.015); hence, we would reject H0 and H3.
(5d) This experiment was designed to determine if classical music facilitates problem
solving. We recruited 153 subjects and randomly assigned them to one of four
groups. Those in Group 1 worked on a 100-piece jigsaw puzzle in silence; those
in Groups 2, 3, and 4 solved the same puzzle with classical music playing at
different volumes in each group: soft (30 dB), medium (60 dB), and loud (90
dB), respectively. We found no statistically significant difference in puzzle
solving time among those who worked in silence (μ = 609 seconds) compared
with those who had classical music played at a soft volume (μ = 606, p = .895, α
= .05) or moderate volume (μ = 608, p = 1.000, α = .05). However, participants
who worked on the puzzle while listening to loud classical music (90 dB) took
significantly longer (μ = 617) to solve the puzzle compared with those who
worked in silence (μ = 509, p = .015, α = .05), suggesting that classical music has
 no effect on this sort of problem solving, unless it is played loudly, in which case,
the music appears to act as a distraction.
7. Data Set A
(7a) H0: Light therapy has no effect on depression.
H1: Light therapy given every other day is effective in reducing depression.
H2: Light therapy given daily is effective in reducing depression.
(7b) Histograms with normal curve plots show a normal distribution of mood for all
groups as shown in the three figures below; hence, the pretest criterion of
normality is satisfied.
Normal distribution for mood in Group 1 (no light therapy)

Normal distribution for mood in Group 2 (light therapy: even days)

Normal distribution for mood in Group 3 (light therapy: every day)

 The homogeneity of variance score for mood shows a significance (p) of .246; since this
is greater than the α level of .05, this suggests that there is no statistically significant
difference between the variances of the three groups; hence, this pretest criterion passes.
The n for each group is 80, which satisfies the 30 per group minimum criterion (see
Descriptives table below).
(7c) The ANOVA test revealed the following:
Means rounded to one decimal digit.
*Statistically significant difference (α = .05).

After 1 month of treatment, participants who received light therapy for 1 hour a day
scored an average of 62.7 on a mood test compared with 59.1 among those who had no
light therapy; this 3.6-point difference in their scores is statistically significant (p = .008, α
= .05). Those who received 1 hour of light therapy every other day showed a better mood
score (61.9) than those who received no light therapy, but the improvement is not
considered statistically significant (p = .053) using the .05 α level. Incidentally, comparing
the mean mood score of those who received light therapy for every-other-day use (61.9)
and daily use (62.7) revealed no statistically significant difference between these two
groups (p = .770, α = .05).
H0: Light therapy has no effect on mood.
H1: Light therapy given every other day enhances mood.
H2: Light therapy given daily enhances mood.
Comparing the mood score from those who received no light therapy (μ = 59.1) with
those who received light therapy daily (μ = 62.7) produced a statistically significant
difference (p = .008); since this is less than the specified α level of .05, we reject H0.
Since there is no statistically significant difference (p = .053) between receiving no light
therapy (μ = 59.1) and receiving light therapy every other day (μ = 61.9), we reject H1.
Comparing the mood level of those who received no light therapy (μ = 59.1) with those
who received light therapy daily (μ = 62.7) renders a statistically significant difference (p
= .008, α = .05); hence, we do not reject H2.
(7d) To determine if light therapy is a viable supplement to treating depression, 240
subjects diagnosed with depression were randomly assigned to one of three
groups: The control group received no light therapy, the second group received 1
hour of light therapy every other day for a month, and the third group received
 light therapy for 1 hour every day for a month. After 30 days, all participants
completed the Acme Mood Scale, a 10-question survey that renders a score from
1 to 100 (1 = extremely bad mood… 100 = extremely good mood). The group
average scores on the mood test were similar among the two light therapy
groups: 61.9 for the every-other-day group and 62.7 for the daily group.
Although these differences were not statistically significantly different from each
other (p = .770, α = .05), only the group that received light therapy on a daily
basis (μ = 62.7) showed a statistically significant improvement in mood
compared with those who received no light therapy (μ = 59.1) (p = .008, α = .05).
7. Data Set B
(7a) H0: Light therapy has no effect on depression.
H1: Light therapy given every other day is effective in reducing depression.
H2: Light therapy given daily is effective in reducing depression.
(7b) Histograms with normal curve plots show a normal distribution of mood for all
groups per the three figures below; hence, the pretest criterion of normality is
satisfied.
Normal distribution for mood in Group 1 (no light therapy)

Normal distribution for mood in Group 2 (light therapy: even days)

 Normal distribution for mood in Group 3 (light therapy: every day)

The homogeneity of variance score for mood shows a significance (p) of .262; since this
is greater than the α level of .05, this suggests that there is no statistically significant
difference between the variances among the three groups; hence, this pretest criterion
passes.
The ns for these groups are 48, 56, and 37, which satisfies the 30 per group minimum
criterion (see Descriptives table below).
(7c) The ANOVA test revealed the following:
 After 1 month of treatment, participants who received light therapy for 1 hour a day
scored an average of 45.54 on a mood test, those who had light therapy every other day
scored an average of 44.86, and those who had no light therapy had an average score of
45.44. The ANOVA table reports a Sig. (p) value of .657, which is greater than the .05 α
level, indicating that there are no statistically significant differences among the groups.
This is confirmed by the Sig. column in the Multiple Comparisons table, wherein the p
values range from .805 to .999, which is well above the .05 α level. Hence, we can
conclude that for this sample, the light therapy protocol produced no statistically
significant improvement in mood.
H0: Light therapy has no effect on mood.
H1: Light therapy given every other day enhances mood.
H2: Light therapy given daily enhances mood.
The Sig. (p) level in the ANOVA table is .657; since this is greater than the .05 α level,
this tells us that there are no statistically significant differences among the groups
involved; hence, we would not reject H0.
Those who had light therapy every other day had a mean mood score of 44.86, and
those who had no light therapy had a mean mood score of 45.44, which renders a p value
of .843. Since the p value is greater than the α level of .05, we rule that there is no
statistically significant difference between these groups; light therapy given every other
day does not enhance mood; hence, we reject H1.
Those who had light therapy daily had a mean mood score of 45.54, and those who had
no light therapy had a mean mood score of 45.44, which renders a p value of .999. Since
the p value is greater than the α level of .05, we rule that there is no statistically significant
difference between these groups; light therapy given every day does not enhance mood;
hence, we reject H2.
(7d) To determine if light therapy is a viable supplement to treating depression, 141
subjects diagnosed with depression were randomly assigned to one of three
groups: The control group received no light therapy, the second group received 1
hour of light therapy every other day for a month, and the third group received
light therapy for 1 hour every day for a month. After 30 days, all participants
completed the Acme Mood Scale, a 10-question survey that renders a score from
1 to 100 (1 = extremely bad mood… 100 = extremely good mood). The group
average scores on the mood test were similar: 45.44 for those who received no
light therapy, 44.86 for those who had light therapy every other day, and 45.54
for those who had light therapy daily. The p values among these three groups
ranged from .805 to .999; using an α level of .05, we conclude that this schedule
of light therapy did not produce a statistically significant improvement in mood
for these participants.
9. Data Set A
(9a) H0: Tending to a plant has no effect on depressive mood.
H1: Tending to a bamboo plant reduces depressive mood.
H2: Tending to a cactus plant reduces depressive mood.
(9b) Despite the few low-score outliers shown in the histogram for Group 2, the
histograms with normal curve plots exhibit a normal distribution of the depress
variable for all groups as shown in the three figures below; hence, the pretest
criterion of normality is satisfied.
Normal distribution for depress in Group 1 (no plant)
 Normal distribution for depress in Group 2 (bamboo)

Normal distribution for depress in Group 3 (cactus)

 The homogeneity of variance score for mood shows a significance (p) of .133; since this
is greater than the α level of .05, this suggests that there is no statistically significant
difference between the variances among the three groups; hence, this pretest criterion
passes.
The n for each group is 60 (see Descriptives table below), which satisfies the 30 per
group minimum criterion.
(9c) The ANOVA test revealed the following:
*Statistically significant difference (α = .05).

There is no statistically significant difference in the average scores between those who
received no plant (μ = 19.60) and those who received a cactus (μ = 19.63) (p = .997, α =
.05). The mean depression score for those who received a bamboo plant (μ = 18.05) was
statistically significantly lower than the scores for those who received no plant (p = .003)
and those who received a cactus (p = .003).
H0: Tending to a plant has no effect on depressive mood.
H1: Tending to a bamboo plant reduces depressive mood.
H2: Tending to a cactus plant reduces depressive mood.
Since those in the bamboo group had a statistically significantly lower depression score
than those who received no plant, we would reject H0; by that same reasoning, we would
not reject H1.
There was no statistically significant difference between the depression score for the
group that received no plant and the group that received a cactus; in summary, tending to a
cactus did not reduce depression, and hence we would reject H2.
(9d) We hypothesized that empowering nursing home residents with an opportunity to
provide nurturance would help reduce depression. To test this hypothesis, 180
residents were randomly assigned to one of three groups: Those in Group 1
served as the control group and were given no plant; Group 2 members were
given a small bamboo plant, and those in Group 3 were given a small cactus.
After 90 days, we administered the Acme Depression Scale (1 = low
depression… 100 = high depression) to members of all three groups. Participants
in the bamboo group scored an average of 18.05, which was statistically
significantly lower than the no-plant group (μ = 19.60, p = .003) and the cactus
group (μ = 19.63, p = .003) using a .05 α level. The cactus group had a slightly
 higher mean depression score (μ = 19.63) compared with those who received no
plant (μ = 19.60), but this was not a statistically significant difference (p = .997).
This suggests that tending to a plant has the potential to reduce depression in
nursing home residents, but the cactus, which requires very little tending, did not
provide the desired effect, whereas the bamboo, which required more monitoring
and watering, did.
9. Data Set B
(9a) H0: Tending to a plant has no effect on depressive mood.
H1: Tending to a bamboo plant reduces depressive mood.
H2: Tending to a cactus plant reduces depressive mood.
(9b) The histograms with normal curve plots exhibit a normal distribution of the
depress variable for all groups, as shown in the three figures below; hence, the
pretest criterion of normality is satisfied.
Normal distribution for depress in Group 1 (no plant)

Normal distribution for depress in Group 2 (bamboo)

 Normal distribution for depress in Group 3 (cactus)

The homogeneity of variance score for mood shows a significance (p) of .333; since this
is greater than the α level of .05, this suggests that there is no statistically significant
difference variances among the three groups; hence, this pretest criterion passes.
The ns for the groups are 58, 51, and 65 (see Descriptives table below), which satisfies
the 30 per group minimum criterion.
(9c) The ANOVA test revealed the following:
 Inspection of the Sig. (p) figure in the ANOVA table (p = .186, which is greater than the
.05 α level) tells us that there are no statistically significant differences detected in the
depression scores among any of the groups.
Upon reviewing the comparisons presented in the Multiple Comparisons table, we see
that this finding is confirmed: The p level for each pair of depression scores is greater than
the specified .05 α level, and hence there are no statistically significant differences among
any of the groups, as shown in the table above.
H0: Tending to a plant has no effect on depressive mood.
H1: Tending to a bamboo plant reduces depressive mood.
H2: Tending to a cactus plant reduces depressive mood.
Since there are no statistically significant differences between the mean depression
score of those who received no plant and those who received any plant, we would not
reject H0.
There was no statistically significant difference between the depression score for the
group that received no plant and the group that received a bamboo plant; apparently,
tending to a bamboo did not significantly reduce depression, and hence we would reject
H1.
There was no statistically significant difference between the depression score for the
group that received no plant and the group that received a cactus; in summary, tending to a
cactus did not significantly reduce depression, and hence we would reject H2.
(9d) We hypothesized that empowering nursing home residents with an opportunity to
provide nurturance would help to reduce depression. To test this hypothesis, 174
residents were randomly assigned to one of three groups: Those in Group 1
served as the control group and were given no plant; Group 2 members were
given a small bamboo plant to tend to along with a card providing care
instructions, and those in Group 3 were given a small cactus along with a card
providing care instructions. After 90 days, we administered the Acme Depression
Scale (1 = low depression… 100 = high depression) to members of both groups.
We found that those who were given the cactus scored an average of 17.03, those
who were given a bamboo plant had an average score of 16.37, and those who
were given no plant scored an average of 15.69; using a .05 α level, we found no
statistically significant differences among any of these groups (p ranged from
.188 to .766). We found that these plants were not helpful in reducing depression
among these nursing home residents.
 CHAPTER 7
Paired t Test
1. Data Set B
(1a) H0: Acme Monster spray has no effect on children’s monster anxiety.
H1: Acme Monster spray reduces children’s monster anxiety.
(1b) After computing the difference between the pretest score and posttest score (diff
= posttest – pretest), a histogram with normal curve was plotted for this
difference (diff). The graph below presents a symmetrical (bell-shaped) normal
curve for diff, thus satisfying this criterion.
Posttest – pretest (diff) renders a normal curve

(1c) The paired t test revealed the following:

Prior to using the Monster Spray, the children’s mean anxiety level was 6.75; after using
the spray, that level dropped to 4.08. This 2.67-point reduction in reported anxiety is
statistically significant since the p value of .001 is less than the specified α level of .05.
On the basis of these findings, we would reject H0 but not reject H1.
(1d) To help children rest easier at night, 12 children were asked to rate their bedtime
anxiety due to fear of nighttime monsters on a 1 to 10 scale (1 = not afraid at
all… 10 = very afraid), and then Acme Monster Spray, an inert sweet-smelling
mist, was administered by their parents, after which children were asked to rate
their anxiety level. Upon spraying, children reported a 2.67-point average drop in
anxiety (6.75 before spraying, down to 4.08 after spraying). Paired t test analysis
revealed this change to be statistically significant (p = .001) using a .05 α level,
suggesting that the (placebo) effect of this spray may help children rest more
comfortably.
1. Data Set B
(1a) H0: Acme Monster spray has no effect on children’s monster anxiety.
H1: Acme Monster spray reduces children’s monster anxiety.
(1b) After computing the difference between the pretest score and posttest score (diff
= posttest – pretest), a histogram with a normal curve was plotted for this
difference (diff). The graph below presents a symmetrical (bell-shaped) normal
curve for diff, thus satisfying this criterion.
Posttest – pretest (diff) renders a normal curve

(1c) The paired t test revealed the following:

 Prior to using the Monster Spray, the children’s mean anxiety level was 5.70; after using
the spray, that level dropped to 5.40. This .30-point reduction in reported anxiety is not
statistically significant since the p value of .081 is greater than the specified α level of .05.
On the basis of these findings, we would not reject H0 but reject H1.
(1d) To help children rest easier at night, 10 children were asked to rate their bedtime
anxiety due to fear of nighttime monsters on a 1 to 10 scale (1 = not afraid at
all… 10 = very afraid), and then Acme Monster Spray, an inert sweet-smelling
mist, was administered by their parents, after which children were asked to rate
their anxiety level. Upon spraying, children reported a .30-point average drop in
anxiety (5.70 before spraying, down to 5.40 after spraying). Paired t test analysis
revealed a p level of .081; using a .05 α level, this suggests that this difference in
scores is not statistically significant; hence, parents will need to find a different
way to console their children at nighttime.
3. Data Set A
(1a) H0: Students do not significantly improve their spelling skills over the course of a
month.
H1: Students improve their spelling skills over the course of a month.
(1b) After computing the difference between the pretest score and posttest score (diff
= posttest – pretest), a histogram with a normal curve was plotted for this
difference (diff). The graph below presents a symmetrical (bell-shaped) normal
curve for diff, thus satisfying this criterion.
Posttest – pretest (diff) renders a normal curve
 (1c) The paired t test revealed the following:

At the start of the month, students were administered a 100-word spelling test to find
out how many words on the prescribed spelling list for that month students could already
spell correctly. Students spelled an average of 62.05 words correctly on this pretest. At the
end of the month, students were readministered this 100-word test and spelled an average
of 69.55 words correctly. The 7.5-point increase in score constitutes a statistically
significant improvement since the p level is .019, which is less than the specified α level
of .05. On the basis of these findings, we would reject H0 but not reject H1.
(1d) To assess if students substantially advance their spelling proficiency on a
monthly basis, students were given a 100-word spelling test at the beginning of
the month; after studying 25 words per week from the list, students took the same
100-word test 30 days later. At the end of the month, students spelled an average
of 69.55 of the words correctly compared with 62.05 at the beginning of the
month. Using an α level of .05, this 7.5-point improvement in score was found to
be statistically significant (p = .019).
3. Data Set B
 (1a) H0: Students do not significantly improve their spelling skills over the course of a
month.
H1: Students improve their spelling skills over the course of a month.
(1b) After computing the difference between the pretest score and posttest score (diff
= posttest – pretest), a histogram with a normal curve was plotted for this
difference (diff). The graph below presents a symmetrical (bell-shaped) normal
curve for diff, thus satisfying this criterion.
Posttest – pretest (diff) renders a normal curve

(1c) The paired t test revealed the following:

At the start of the month, students were administered a 100-word spelling test to find
out how many words on the prescribed spelling list for that month students could already
spell correctly. Students spelled an average of 86.14 words correctly on this pretest. At the
end of the month, students were readministered this 100-word test and spelled an average
of 87.59 words correctly. The 1.45-point increase in score is not a statistically significant
improvement since the p level is .283, which is greater than the specified α level of .05.
On the basis of these findings, we would not reject H0 but reject H1.
(1d) To find out if students substantially advance their spelling proficiency on a
 monthly basis, students were given a 100-word spelling test at the beginning of
the month; after studying 25 words per week from the list, students took the same
100-word test 30 days later. At the end of the month, students spelled an average
of 87.59 of the words correctly compared with 86.14 at the beginning of the
month. Using an α level of .05, this 1.45-point improvement in score is not
statistically significant (p = .283).
5. Data Set A
(1a) H0: A 15-minute individual session with a bowling coach has no effect on
bowling scores.
H1: A 15-minute individual session with a bowling coach enhances bowling
scores.
(1b) After computing the difference between the pretest score and posttest score (diff
= posttest – pretest), a histogram with a normal curve was plotted for this
difference (diff). The graph below presents a symmetrical (bell-shaped) normal
curve for diff, thus satisfying this criterion.
Posttest – Pretest (diff) renders a normal curve

(1c) The paired t test revealed the following:

 Fifteen students bowled an initial game and scored a mean of 151.60, and then a coach
provided 15 minutes of instructions based on observations made during the students’ first
game. Students’ mean score in the second game was 157.73. This 6.13-point increase
produced a p level of .003; since this is less than the α level of .05, this is considered a
statistically significant difference. On the basis of these findings, we would reject H0 but
not reject H1.
(1d) To assess the effectiveness of short-term coaching, 15 students bowled one game
with a coach observing unobtrusively. Next, the coach provided 15 minutes of
individual coaching, after which each student bowled a second game. On
average, scores were 6.13 points higher in the second game (157.73, up from
151.60 in the first game). This finding is considered statistically significant (p =
.003, α = .05). On the basis of these findings, we will continue to use this
coaching method for bowling, and we will be evaluating the utility of providing
this style of coaching for other sports.
5. Data Set B
(1a) H0: A 15-minute individual session with a bowling coach has no effect on
bowling scores.
H1: A 15-minute individual session with a bowling coach enhances bowling
scores.
(1b) After computing the difference between the pretest score and posttest score (diff
= posttest – pretest), a histogram with a normal curve was plotted for this
difference (diff). The graph below presents a symmetrical (bell-shaped) normal
curve for diff, thus satisfying this criterion.
Posttest – pretest (diff) renders a normal curve

(1c) The paired t test revealed the following:

 Eight students bowled an initial game and scored a mean of 72.50, and then a coach
provided 15 minutes of instructions based on observations made during the students’ first
game. Students’ mean score in the second game was 74.25. This 1.75-point increase
produced a p level of .334; since this is greater than the α level of .05, this is not
considered a statistically significant difference. On the basis of these findings, we would
not reject H0 but reject H1.
(1d) To assess the effectiveness of short-term coaching, eight students bowled one
game with a coach observing unobtrusively. Next, the coach provided 15 minutes
of individual coaching, after which the students bowled a second game. On
average, students’ scores were 1.75 points higher in their second game (74.25, up
from 72.50 in their first game). This finding is not considered statistically
significant (p = .334, α = .05). On the basis of these findings, we are considering
evaluating if this form of coaching may provide more promising results for
different individual sports (e.g., golf, archery, weight training).
7. Data Set A
(1a) H0: Acme allergy medicine has no effect on sneezing.
H1: Acme allergy medicine reduces sneezing.
(1b) After computing the difference between the pretest score and posttest score (diff
= posttest – pretest), a histogram with a normal curve was plotted for this
difference (diff). The graph below presents a symmetrical (bell-shaped) normal
curve for diff, thus satisfying this criterion.
Posttest – pretest (diff) renders a normal curve
 (1c) The paired t test revealed the following:

Eighteen people who were experiencing allergy symptoms were recruited and asked to
count their total number of sneezes for a day. The next day, they took the Acme allergy
medicine as directed and counted their total number of sneezes for that (second) day. This
resulted in an average of 2.778 fewer sneezes (an average of 15.39 on the first day and
12.61 on the second day). Using a .05 α level, this difference is considered statistically
significant (p = .002). As such, we would reject H0 but not reject H1.
(1d) To determine if Acme allergy medicine helps reduce sneezing among those
experiencing allergy symptoms, 18 participants were recruited and instructed to
count their total number of sneezes for 1 day. The next morning, each took the
Acme allergy medicine as directed and counted their total sneezes for that day.
On average, participants sneezed a total of 15.39 times the day before taking the
medication and 12.61 times after taking the medication, an average of 2.778
fewer sneezes, which equates to an 18%* reduction in sneezing. This difference
is considered statistically significant (p = .002, α = .05), suggesting that the
medicine provided some allergy symptom relief.
*NOTE: The 18% figure was calculated using the ∆% formula on page 165:
∆% = (New – Old) ÷ Old × 100.
7. Data Set B
(1a) H0: Acme allergy medicine has no effect on sneezing.
H1: Acme allergy medicine reduces sneezing.
(1b) After computing the difference between the pretest score and posttest score (diff
= posttest – pretest), a histogram with a normal curve was plotted for this
difference (diff). The graph below presents a symmetrical (bell-shaped) normal
curve for diff, thus satisfying this criterion.
Posttest – pretest (diff) renders a normal curve

(1c) The paired t test revealed the following:

Twenty-one people who were experiencing allergy symptoms were recruited and asked
to count their total number of sneezes for a day. The next day, they took the Acme allergy
medicine as directed and counted their total number of sneezes for that (second) day. This
resulted in an average of .15 fewer sneezes (an average of 4.10 on the first day and 3.95 on
the second day). This rendered a p of .817; since the p value is greater than the designated
.05 α level, this difference is not considered statistically significant. As such, we would
not reject H0 but reject H1.
(1d) To determine if Acme allergy medicine helps reduce sneezing among those
experiencing allergy symptoms, 21 participants were recruited and instructed to
count their total number of sneezes for 1 day. The next morning, they took the
Acme allergy medicine as directed and counted their total sneezes for that day.
On average, participants sneezed a total of 4.10 times the day before taking the
medication and 3.95 times after taking the medication, amounting to an average
of .15 fewer sneezes, which is about a 3.6%* reduction in sneezing. This
reduction is not considered statistically significant (p = .817, α = .05), suggesting
that this medicine was not particularly helpful for these individuals.

*NOTE: The 3.6% figure was calculated using the formula on page 165:
∆% = (New – Old) ÷ Old × 100.
9. Data Set A
(1a) H0: Chocolate has no effect on attitude.
H1: Chocolate enhances attitude.
(1b) After computing the difference between the pretest score and posttest score (diff
= posttest – pretest), a histogram with a normal curve was plotted for this
difference (diff). The graph below presents a symmetrical (bell-shaped) normal
curve for diff, thus satisfying this criterion.
Posttest – pretest (diff) renders a normal curve
 (1c) The paired t test revealed the following:

Seventy-five people were recruited to determine if chocolate helps to enhance attitude.

We began by administering the Acme Attitude Survey (AAS) (0 = very bad attitude…
100 = very good attitude), resulting in a mean score of 81.91. After eating a piece of
chocolate fudge cake, participants answered the AAS again; the mean of the results
dropped by 1.06 points to 80.85. This change in the AAS score produced a p level of .318,
which, using a .05 α level, is not statistically significant. This suggests that the chocolate
cake did not make a statistically significant impact on attitude; hence, we would not reject
H0 but reject H1.
(1d) To better comprehend the effects that chocolate might have on attitude, 75
participants were recruited; we began by administering the Acme Attitude
Survey (AAS), which renders a score ranging from 0 to 100 (0 = very bad
attitude… 100 = very good attitude). Next, each participant was served a
generous slice of chocolate fudge cake. After the cake, participants were asked to
complete a second AAS. The pretest revealed a mean attitude score of 81.91;
contrary to our expectations, the post-chocolate AAS score dropped an average
of 1.06 points, to 80.85, rendering a p value of .318. Using a .05 α level, this drop
 in the AAS is not considered statistically significant; hence, in this case,
chocolate had no impact on mood.
9. Data Set B
(1a) H0: Chocolate has no effect on attitude.
H1: Chocolate enhances attitude.
(1b) After computing the difference between the pretest score and posttest score (diff
= posttest – pretest), a histogram with a normal curve was plotted for this
difference (diff). The graph below presents a symmetrical (bell-shaped) normal
curve for diff, thus satisfying this criterion.
Posttest – pretest (diff) renders a normal curve

(1c) The paired t test revealed the following:

Seventy-five people were recruited to determine if chocolate helps to enhance attitude.

We began by administering the Acme Attitude Survey (AAS) (0 = very bad attitude… 100
= very good attitude), resulting in an average score of 81.17. After eating a piece of
chocolate fudge cake, participants answered the AAS again; the mean results rose by 2.10
points to 83.27. This change in the AAS score produced a p level of .027, which, using a
.05 α level, is considered statistically significant. This suggests that the chocolate cake
made a statistically significant impact on attitude; hence, we would reject H0 but not reject
H1.
(1d) To better comprehend the effects that chocolate might have on attitude, 75
participants were recruited; we began by administering the Acme Attitude
Survey (AAS), which renders a score ranging from 0 to 100 (0 = very bad
attitude… 100 = very good attitude). After collecting the AAS forms, each
participant was served a generous slice of chocolate fudge cake. After the cake,
participants were asked to complete a second AAS. The pretest revealed a mean
attitude score of 81.17, and after dining on the chocolate cake, the mean AAS
score went up an average of 2.10 points, to 83.27. These figures rendered a p
value of .027. Using a .05 α level, this increase in the AAS score suggests that
chocolate may help to enhance attitude.
 CHAPTER 8
Correlation and Regression
1. Data Set A
(1a) H0: There is no significant correlation between hours of walking and weight loss.
H1: There is a significant correlation between hours of walking and weight loss.
(1b) Histograms with normal curve plots show a normal distribution for walkhrs and
wtloss as shown in the two figures below; hence, the pretest criterion of
normality is satisfied.

The scatterplot below shows that the straight regression line reasonably fits the field of
points, and hence the criterion of linearity is satisfied. The scatterplot also shows that the
points are primarily clustered toward the center of the regression line with substantially
fewer points toward the ends, and hence the criterion of homoscedasticity is satisfied.
Scatterplot with regression line for walkhrs and wtloss is linear and homoscedastic
 (1c) The correlations table below shows a positive correlation (r = .624) between
walkhrs and wtloss; p = .006, which is less than the specified .05 α level,
indicating that this is a statistically significant correlation.

(1d) In a study examining the effects of brisk walking on weight loss, 18 adults were
recruited and instructed to walk as many days of the week as possible for as long
as they could. The researcher gathered each participant’s daily walking time, as
well as his or her weight at the beginning and end of the week, and found a
positive, statistically significant correlation between the amount of walking time
and the amount of weight lost (r = .624, p = .006; α = .05). In other words, on the
whole, the more the participants walked, the more weight they lost. These
findings suggest that moderate daily exercise may be effective in facilitating
weight loss. As such, we reject H0 but do not reject H1.
NOTE: You might also inspect the descriptive statistics for walkhrs and wtloss and
consider including the supplement in your findings: “Participants averaged 4.7 hours of
walking per week and had an average weight loss of 1.5 pounds.”
1. Data Set B
(1a) H0: There is no significant correlation between hours of walking and weight loss.
H1: There is a significant correlation between hours of walking and weight loss.
(1b) Histograms with normal curve plots show a normal distribution for walkhrs and
wtloss, as shown in the two figures below; hence, the pretest criterion of
normality is satisfied.
 The scatterplot below shows that the straight regression line reasonably fits the field of
points, and hence the criterion of linearity is satisfied. The scatterplot also shows that the
points are primarily clustered toward the center of the regression line with substantially
fewer points toward the ends, and hence the criterion of homoscedasticity is satisfied.
Scatterplot with regression line for walkhrs and wtloss is linear and homoscedastic

(1c) The correlations table below shows a positive correlation (r = .250) between
walkhrs and wtloss, but since p = .317, which is greater than the specified .05 α
level, this indicates that this is not a statistically significant correlation.
 (1d) In a study examining the effects of brisk walking on weight loss, 18 adults were
recruited and instructed to walk as many days of the week as possible for as long
as they could. The researcher gathered each participant’s daily walking time, as
well as his or her weight at the beginning and end of the week, and found a
positive correlation between the amount of walking time and the amount of
weight lost (r = .250). Although the participants experienced some weight loss in
this study, the correlation between walking and weight loss was not statistically
significant (p = .317; α = .05). As such, we do not reject H0 but reject H1.
NOTE: You might also inspect the descriptive statistics for walkhrs and wtloss and
consider including the following supplement in your findings: “Participants averaged 5.9
hours of walking per week and had an average weight loss of 1.5 pounds.”
3. Data Set A
(3a) H0: There is no significant correlation between income and happiness.
H1: There is a significant correlation between income and happiness.
(3b) Histograms with normal curve plots show a normal distribution for income and
alhs, as shown in the two figures below; hence, the pretest criterion of normality
is satisfied.
 The scatterplot below shows that the straight regression line reasonably fits the field of
points, and hence the criterion of linearity is satisfied. The scatterplot also shows that the
points are primarily clustered toward the center of the regression line with substantially
fewer points toward the ends, and hence the criterion of homoscedasticity is satisfied.
Scatterplot with regression line for income and alhs is linear and homoscedastic

(3c) The correlations table below shows a negative correlation (r = –.026) between
income and alhs, but because p = .917, which is greater than the specified .05 α
level, this is not a statistically significant correlation, as illustrated in the above
scatterplot; note the flatness of the regression line along with the distance of the
points from the regression line.

(3d) In a study exploring the relationship between money and happiness, 18 adults
were recruited and given a self-administered survey that asked their annual
income; the survey also included the Acme Life Happiness Scale (ALHS), a 10-
question survey that renders a score from 0 to 100 (0 = very unhappy… 100 =
very happy). Our findings revealed a slight negative correlation (r = −.026),
meaning that earning more money was slightly associated with less happiness,
 but this (negative) correlation was not statistically significant (p = .917; α = .05).
As such, we do not reject H0 but reject H1.
3. Data Set B
(3a) H0: There is no significant correlation between income and happiness.
H1: There is a significant correlation between income and happiness.
(3b) Histograms with normal curve plots show a normal distribution for income and
alhs, as shown in the two figures below; hence, the pretest criterion of normality
is satisfied.

The scatterplot below shows that the straight regression line reasonably fits the field of
points, and hence the criterion of linearity is satisfied. The scatterplot also shows that the
points are primarily clustered toward the center of the regression line with substantially
fewer points toward the ends, and hence the criterion of homoscedasticity is satisfied.
Scatterplot with regression line for income and alhs is linear and homoscedastic

(3c) The correlations table below shows a negative correlation (r = .468) between
income alhs; since p = .032, is less than the specified .05 α level, this indicates
that this is a statistically significant correlation.
 (3d) In a study exploring the relationship between money and happiness, 21 adults
were recruited and given a self-administered survey that asked their annual
income; the survey also included the Acme Life Happiness Scale (ALHS), a 10-
question survey that renders a score from 0 to 100 (0 = very unhappy… 100 =
very happy). Our findings revealed a positive correlation (r = .468), meaning that
earning more money was associated with higher levels of happiness; this
(positive) correlation was found to be statistically significant (p = .032; α = .05).
As such, we reject H0 but do not reject H1.
5. Data Set A
(5a) H0: There is no significant correlation between years of education and job
satisfaction.
H1: There is a significant correlation between years of education and job
satisfaction.
(5b) Histograms with normal curve plots show a normal distribution for yearsed and
ajsi, as shown in the two figures below; hence, the pretest criterion of normality
is satisfied.
 The scatterplot below shows that the straight regression line reasonably fits the field of
points, and hence the criterion of linearity is satisfied. The scatterplot also shows that the
points are primarily clustered toward the center of the regression line with substantially
fewer points toward the ends, and hence the criterion of homoscedasticity is satisfied.
Scatterplot with regression line for yearsed and ajsi is linear and homoscedastic

(5c) The correlations table below shows a positive correlation (r = .679) between
income and ajsi; p = .004, which is less than the specified .05 α level, indicating
that this is a statistically significant correlation, as illustrated in the scatterplot
above; note the positive slope of the regression line and that most of the points
are clustered fairly close to the regression line.

(5d) To assess the possible correlation between years of education and job
satisfaction, 16 adults were recruited and given a self-administered survey that
asked how many years of education they had; the survey also included the Acme
Job Satisfaction Inventory (AJSI), a six-question survey that renders a score from
0 to 60 (0 = very dissatisfied with one’s job… 60 = very satisfied). The
participant’s education level ranged from 11th grade through master’s degree.
Our findings revealed a relatively strong positive correlation (r = .679),
suggesting that those with more education tended to get more satisfaction from
their jobs; this correlation was statistically significant (p = .004; α = .05). As
such, we reject H0 but do not reject H1.
5. Data Set B
(5a) H0: There is no significant correlation between years of education and job
satisfaction.
H1: There is a significant correlation between years of education and job
satisfaction.
(5b) Histograms with normal curve plots show a normal distribution for yearsed and
ajsi, as shown in the two figures below; hence, the pretest criterion of normality
is satisfied.

The scatterplot below shows that the straight regression line reasonably fits the field of
points, and hence the criterion of linearity is satisfied. The scatterplot also shows that the
points are primarily clustered toward the center of the regression line with substantially
fewer points toward the ends, and hence the criterion of homoscedasticity is satisfied.
Scatterplot with regression line for yearsed and ajsi is linear and homoscedastic

(5c) The correlations table below shows a relatively small positive correlation (r =
.196) between income and ajsi; however, p = .484, which is greater than the
specified .05 α level, indicating that this is not a statistically significant
correlation, as illustrated in the scatterplot above; note the slight positive slope of
the regression line and that most of the points are clustered fairly far from the
 regression line.

(5d) To assess the possible correlation between years of education and job
satisfaction, 15 adults were recruited and given a self-administered survey that
asked how many years of education they had; the survey also included the Acme
Job Satisfaction Inventory (AJSI), a six-question survey that renders a score from
0 to 60 (0 = very dissatisfied with one’s job… 60 = very satisfied). The
participant’s education level ranged from 11th grade through master’s degree.
Our findings revealed a relatively mild positive correlation (r = .196), suggesting
that those with more education tended to get somewhat more satisfaction from
their jobs, but this correlation was not statistically significant (p = .484; α = .05).
As such, we do not reject H0 but reject H1.
7. Data Set A
(7a) H0: There is no significant correlation between height and self-confidence.
H1: There is a significant correlation between height and self-confidence.
(7b) Histograms with normal curve plots show a normal distribution for height and
asci, as shown in the two figures below; hence, the pretest criterion of normality
is satisfied.
 The scatterplot below shows that the straight regression line reasonably fits the field of
points, and hence the criterion of linearity is satisfied. The scatterplot also shows that the
points are primarily clustered toward the center of the regression line with substantially
fewer points toward the ends, and hence the criterion of homoscedasticity is satisfied.
Scatterplot with regression line for height and asci is linear and homoscedastic

(7c) The correlations table below shows a relatively small negative correlation (r =
−.024) between height and asci; p = .882, which is greater than the specified .05
α level, indicating that this is not a statistically significant correlation, as
illustrated in the above scatterplot; note the relative flatness of the regression line
and that most of the points are clustered fairly far from the regression line.

(7d) To evaluate the possible correlation between height and self-confidence, 41

adults were recruited and given a self-administered survey that asked them to
indicate their height; the survey also included the Acme Self-Confidence Index
(ASCI), a 10-question survey that renders a score from 0 to 50 (0 = very low
self-confidence… 50 = very high self-confidence). Our findings indicated a
slight negative correlation between the variables (r = −.024), suggesting that
taller people were slightly less self-confident than shorter people. Further
analysis revealed that this correlation is not statistically significant (p = .882; α =
.05). As such, we do not reject H0 but reject H1.
7. Data Set B
 (7a) H0: There is no significant correlation between height and self-confidence.
H1: There is a significant correlation between height and self-confidence.
(7b) Histograms with normal curve plots show a normal distribution for height and
asci, as shown in the two figures below; hence, the pretest criterion of normality
is satisfied.

The scatterplot below shows that the straight regression line reasonably fits the field of
points, and hence the criterion of linearity is satisfied. The scatterplot also shows that the
points are primarily clustered toward the center of the regression line with substantially
fewer points toward the ends, and hence the criterion of homoscedasticity is satisfied.
Scatterplot with regression line for height and asci is linear and homoscedastic

(7c) The correlations table below shows a relatively small positive correlation (r =
.098) between height and asci; however, p = .492, which is greater than the
specified .05 α level, indicating that this is not a statistically significant
correlation, as illustrated in the above scatterplot; note the relative flatness of the
regression line and that most of the points are clustered fairly far from the
regression line.
 (7d) To evaluate the possible correlation between height and self-confidence, 51
adults were recruited and given a self-administered survey that asked their
height; the survey also included the Acme Self-Confidence Index (ASCI), a 10-
question survey that renders a score from 0 to 50 (0 = very low self-
confidence… 50 = very high self-confidence). Our findings suggest that although
there was a slight positive correlation between the variables (r = .098), implying
that taller people tend to have more self-confidence than shorter people, it was
determined that this correlation is not statistically significant (p = .492; α = .05).
As such, we do not reject H0 but reject H1.
9. Data Set A
(9a) H0: There is no significant correlation between class time and freshman English
grade.
H1: There is a significant correlation between class time and freshman English
grade.
(9b) Histograms with normal curve plots show a normal distribution for time and
grade, as shown in the two figures below; hence, the pretest criterion of
normality is satisfied.

The scatterplot below shows that the straight regression line reasonably fits the field of
points, and hence the criterion of linearity is satisfied. The scatterplot also shows that the
points are primarily clustered toward the center of the regression line with substantially
fewer points toward the ends, and hence the criterion of homoscedasticity is satisfied.
Scatterplot with regression line for time and grade is linear and homoscedastic

(9c) The correlations table below shows a relatively small negative correlation (r =
−.234) between time and grade; p = .019, which is less than the specified .05 α
level, indicating that this is a statistically significant correlation, as illustrated in
the above scatterplot; note the negative slope of the regression line.

(9d) To evaluate the possible correlation between the time of day that students have
their freshman English course and the associated grade, 100 students were
surveyed and asked to present their transcript indicating the grade they received
in freshman English; the transcript also indicated the day and time that the course
met. Our findings revealed a negative correlation (r = −.234) between time of
day and the grade received, suggesting that students enrolled in earlier classes
outperformed those who took the same course later in the day. This correlation
was determined to be statistically significant (p = .019, α = .05). As such, we
reject H0 but do not reject H1.
9. Data Set B
(9a) H0: There is no significant correlation between class time and freshman English
 grade.
H1: There is a significant correlation between class time and freshman English
grade.
(9b) Histograms with normal curve plots show a normal distribution for time and
grade, as shown in the two figures below; hence, the pretest criterion of
normality is satisfied.

The scatterplot below shows that the straight regression line reasonably fits the field of
points, and hence the criterion of linearity is satisfied. The scatterplot also shows that the
points are primarily clustered toward the center of the regression line with substantially
fewer points toward the ends, and hence the criterion of homoscedasticity is satisfied.
Scatterplot with regression line for time and grade is linear and homoscedastic

(9c) The correlations table below shows a relatively small positive correlation (r =
.056) between time and grade; however, p = .619, which is greater than the
specified .05 α level, indicating that this is not a statistically significant
correlation, as illustrated in the above scatterplot; note the slight positive slope of
the regression line.
(9d) To evaluate the possible correlation between the time of day that students have
their freshman English course and the associated grade, 80 students were
surveyed and asked to present their transcript indicating the grade they received
in freshman English; the transcript also indicated the day and time that the course
met. Our findings revealed a slight positive correlation (r = .056) between time
of day and the grade received, suggesting that students enrolled in classes held
later in the day outperformed those who took the same course earlier in the day;
however, this correlation was not statistically significant (p = .619, α = .05). As
such, we do not reject H0 but reject H1.
 CHAPTER 9
Chi-Square
1. Data Set A
(1a) H0: There is no significant difference in ice cream preference in terms of gender.
H1: There is a significant difference in ice cream preference in terms of gender.
(1b) Per the Crosstabulation table below, each cell contains an n of at least 5; hence,
the pretest criterion is satisfied.

(1c) The Chi-Square Tests table below shows a Sig. (p) value of .023, which is less
than the specified .05 α level, indicating that there is a statistically significant
difference among genders when it comes to ice cream preference.
 (1d) To determine if gender is associated with ice cream flavor preference, 48
participants (23 females, 25 males) were individually asked to identify which
flavor of ice cream they prefer: chocolate, strawberry, or vanilla. Among the
females, 11 of the 23 (48%*) selected chocolate, whereas 15 of the 25 (60%*)
male participants favored strawberry. Our χ2 (chi-square) findings revealed that
there was a statistically significant difference in flavor preference with respect to
gender (p = .023, α = .05). As such, we reject H0 but do not reject H1.

*NOTE: The percentages were calculated using the formula on page 207:
% = Part ÷ Total × 100.
1. Data Set B
(1a) H0: There is no significant difference in ice cream preference in terms of gender.
H1: There is a significant difference in ice cream preference in terms of gender.
(1b) Per the Crosstabulation table below, each cell contains an n of at least 5; hence,
the pretest criterion is satisfied.
 (1c) The Chi-Square Tests table below shows a Sig. (p) value of .984, which is
greater than the specified .05 α level, indicating that there is no statistically
significant difference among genders when it comes to ice cream preference.

(1d) To determine if gender is associated with ice cream flavor preference, 46

participants (24 females, 22 males) were individually asked to identify which
flavor of ice cream they prefer: chocolate, strawberry, or vanilla. Among the
females, 10 of the 24 (42%) selected chocolate; similarly, 9 of the 22 (41%) male
participants also favored chocolate. Both genders ranked strawberry as second
and vanilla as third. Our χ2 (chi-square) findings revealed that there was no
statistically significant difference in flavor preference with respect to gender (p =
.984, α = .05). As such, we do not reject H0 but reject H1.
3. Data Set A
(3a) H0: The flu shot does not help prevent the flu.
 H1: The flu shot helps prevent the flu.
(3b) Per the Crosstabulation table below, each cell contains an n of at least 5; hence,
the pretest criterion is satisfied.

(3c) The Chi-Square Tests table below shows a Sig. (p) value of .296, which is
greater than the specified .05 α level, indicating that there is no statistically
significant difference in flu sickness comparing those who got a flu shot with
those who did not.
 (3d) To determine if there were significantly fewer cases of flu among those who had
a flu shot compared with those who did not, we analyzed the health status of 30
participants. Sixty days after recruiting participants, our researcher contacted
each one to find out of they had contracted the flu. We found that 12 had gotten
flu shots and 18 had not. Among those who got the flu shot, 41% reported that
they had gotten sick with the flu compared with 61% among those who did not
get a flu shot. Even though there were proportionally fewer cases of flu among
those who had a flu shot, chi-square analysis indicates that this difference is not
statistically significant (p = .296, α = .05). As such, we do not reject H0 but reject
H1.
3. Data Set B
(3a) H0: The flu shot does not help prevent the flu.
H1: The flu shot helps prevent the flu.
(3b) Per the Crosstabulation table below, each cell contains an n of at least 5; hence,
the pretest criterion is satisfied.
(3c) The Chi-Square Tests table below shows a Sig. (p) value of .023, which is less
than the specified .05 α level, indicating that there is a statistically significant
difference in flu sickness comparing those who got a flu shot with those who did
not.

(3d) To determine if there were significantly fewer cases of flu among those who had
a flu shot compared with those who did not, we analyzed the health status of 36
participants. Sixty days after recruiting participants, our researcher contacted
each one to find out of they had contracted the flu. We found that 15 had gotten
flu shots, and 21 had not. Among those who got the flu shot, 33% reported that
 they had gotten sick with the flu compared with 71% among those who did not
get a flu shot. Chi-square analysis indicates that this difference is statistically
significant (p = .023, α = .05). As such, we reject H0 but do not reject H1.
5. Data Set A
(5a) H0: Responses are the same across media (face-to-face interview, mail-in survey,
online survey) when it comes to asking about substance abuse.
H1: Responses vary significantly across media (face-to-face interview, mail-in
survey, online survey) when it comes to asking about substance abuse.
(5b) Per the Crosstabulation table below, each cell contains an n of at least 5; hence,
the pretest criterion is satisfied.

(5c) The Chi-Square Tests table below shows a Sig. (p) value of .729, which is
greater than the specified .05 α level, indicating that there is no statistically
significant difference in responses across the three media tested: face-to-face
 interview, mail-in survey, and online survey.

(5d) To determine if media make a difference in responses when it comes to inquiries

regarding illegal substance use, we recruited 46 subjects and asked each to
respond to one question: “Have you ever used illegal drugs?” However, our
method of inquiry varied. Subjects were randomly assigned to one of three
groups: The 15 participants in Group 1 were asked the question via a face-to-face
interview, the 17 people assigned to Group 2 responded via standard pencil-and-
paper mail-in survey, and the 14 in Group 3 were directed to an online survey
website. The results revealed that 33% of those who were asked the question
face-to-face replied that they had used illegal drugs (at least once) compared with
29% in the mail-in group and 43% in the online group. Although there is some
response variability among these media, chi-square analysis revealed that these
differences are not statistically significant (p = .729, α = .05). Hence, we do not
reject H0 but do reject H1.
5. Data Set B
(5a) H0: Responses are the same across media (face-to-face interview, mail-in survey,
online survey) when it comes to asking about substance abuse.
H1: Responses vary across media (face-to-face interview, mail-in survey, online
survey) when it comes to asking about substance abuse.
(5b) Per the Crosstabulation table below, each cell contains an n of at least 5; hence,
the pretest criterion is satisfied.
(5c) The Chi-Square Tests table below shows a Sig. (p) value of .045, which is less
than the specified .05 α level, indicating that there is a statistically significant
difference in responses across the three media tested: face-to-face interview,
mail-in survey, and online survey.

(5d) To determine if media make a difference in responses when it comes to inquiries

regarding illegal substance use, we recruited 53 subjects and asked each to
respond to one question: “Have you ever used illegal drugs?” However, our
method of inquiry varied. Subjects were randomly assigned to one of three
groups: The 18 participants in Group 1 were asked the question via a face-to-face
interview, the 18 people assigned to Group 2 responded via standard pencil-and-
paper mail-in survey, and the 17 in Group 3 were directed to an online survey
website. The results revealed that 33% of those who were asked the question
face-to-face replied that they had used illegal drugs (at least once) compared with
72% in the mail-in group and 64% in the online group. Chi-square analysis
 revealed that these differences are statistically significant (p = .045, α = .05).
Hence, we would reject H0 but not reject H1. These findings suggest that when it
comes to sensitive issues, participants who are interviewed face-to-face seem to
“sanitize” or alter their responses to reveal less negative disclosure than those
responding by less identifiable means (mail-in/online surveys).
7. Data Set A
(7a) H0: Age is not associated with voting practices.
H1: Age is associated with voting practices.
(7b) Per the Crosstabulation table below, each cell contains an n of at least 5; hence,
the pretest criterion is satisfied.

(7c) The Chi-Square Tests table below shows a Sig. (p) value of .802, which is
greater than the specified .05 α level, indicating that there is no statistically
significant difference in voting practices across age groups.
 (7d) To determine if voting practices vary significantly across age groups, we
surveyed a total of 117 participants spanning three age groups—26 were 18 to
35, 58 were 36 to 64, and 33 were 65 or older—and asked each about how they
voted in the last election: voted in person, voted by mail, or did not vote. The
majority of those surveyed in each category indicated they voted by mail: 54% in
the 18 to 35 group and 48% in each of the other two groups. In-person voting
ranged from 23% in the 18 to 35 group to 19% in the 36 to 64 group. Finally,
when it comes to not voting, 36- to 64-year-olds topped the list at 33%, followed
by the 65 or older group and the 18 to 35 group: 24% and 23%, respectively.
Despite the moderate variability in voting practices observed among these
groups, chi-square analysis indicates that these differences are not statistically
significant (p = .802, α = .05). Hence, we do not reject H0 but reject H1. Among
these groups, mail-in voting appears to be the preferred option.
7. Data Set B
(7a) H0: Age is not associated with voting practices.
H1: Age is associated with voting practices.
(7b) Per the Crosstabulation table below, each cell contains an n of at least 5; hence,
the pretest criterion is satisfied.
 (7c) The Chi-Square Tests table below shows a Sig. (p) value of .003, which is less
than the specified .05 α level, indicating that there is a statistically significant
difference in voting practices based on age groups.

(7d) To determine if voting practices vary significantly across age groups, we

surveyed a total of 60 participants spanning three age groups—25 were 18 to 35,
26 were 36 to 64, and 41 were 65 or older—and asked each about how they
voted in the last election: voted in person, voted by mail, or did not vote. Each
group indicated a different majority: 48% of those in the 18 to 35 group did not
vote, 58% of those in the 36 to 64 group voted by mail, and 49% of those 65 or
older voted in person. Chi-square analysis revealed that there is a statistically
significant difference among age groups when it comes to voting practices (p =
.003, α = .05). Hence, we rejected H0 but did not reject H1.
9. Data Set A
(9a) H0: There is no significant difference in university admissions based on high
 school (public vs. private).
H1: There is a significant difference in university admissions based on high
school (public vs. private).
(9b) Per the Crosstabulation table below, each cell contains an n of at least 5; hence,
the pretest criterion is satisfied.

(9c) The Chi-Square Tests table below shows a Sig. (p) value of .652, which is
greater than the specified .05 α level, indicating that there is no statistically
significant difference in university admissions based on school (public vs.
private).
 (9d) We evaluated college admission rates from two schools: the Acme Academy, a
private school with a graduating class of 49 students, and Anytown High School,
a public school with a graduating class of 225 students. We found that 55% of
those who graduated from the Acme Academy were admitted to a university
compared with 51% of the students who graduated from Anytown High School.
In terms of total students, Acme Academy launched 27 students into higher
education compared with 116 students from Anytown High School. Despite the
variability among these figures, chi-square analysis indicates that the
(proportional) difference is not statistically significant (p = .652, α = .05). Hence,
we do not reject H0 but reject H1.
9. Data Set B
(9a) H0: There is no significant difference in university admissions based on high
school (public vs. private).
H1: There is a significant difference in university admissions based on high
school (public vs. private).
(9b) Per the Crosstabulation table below, each cell contains an n of at least 5; hence,
the pretest criterion is satisfied.
(9c) The Chi-Square Tests table below shows a Sig. (p) value of .001, which is less
than the specified .05 α level, indicating that there is a statistically significant
difference in university admissions based on school (public vs. private).

(9d) We evaluated college admission rates from two schools: the Acme Academy, a
private school with a graduating class of 33 students, and Anytown High School,
a public school with a graduating class of 150 students. We found that 85% of
those who graduated from the Acme Academy were admitted to a university
compared with 52% of the students who graduated from Anytown High School.
In terms of total students, Acme Academy launched 28 students into higher
education compared with 78 students from Anytown High School. Despite the
higher total number of students admitted to a university from the public school,
chi-square analysis indicates that proportionally, the Acme Academy
significantly outperformed Anytown High School when it comes to college
admissions (p = .001, α = .05). Hence, we reject H0 but do not reject H1.
 C H A P T E R 10
Supplemental SPSS Operations
(1a)
(1b)
(3a)
(3b)
(5a)
(5b)

(7)
comment Program: Weekly Report − 3 Subject Test.sps.
comment Programmed by H Knapp.
comment Revision 1.2.
comment Data file: Chapter 10 − Exercise 06.sav.
comment This program runs the weekly report for the reading, writing, and math scores.
comment It performs a three-level sort on the data by reading (ascending), by writing
(descending), and by math (ascending).
 comment Finally, it runs descriptive statistics with histograms for reading and math
scores.
comment - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
-.
comment Sort the data using a three-level sort: (1) reading [ascending], (2) writing
[descending], (3) math [ascending].
DATASET ACTIVATE DataSet8.
SORT CASES BY reading(A) writing(D) math(A).
comment Run descriptive statistics (mean, median, mode, maximum, minimum, range,
standard deviation, and variance) for reading and math; also include histograms with
normal curves for those variables.
FREQUENCIES VARIABLES=reading math
/FORMAT=NOTABLE
/STATISTICS=STDDEV VARIANCE RANGE MINIMUM MAXIMUM MEAN
MEDIAN MODE
/HISTOGRAM NORMAL
/ORDER=ANALYSIS.
comment end of Syntax file.
(9)
The purpose of a work file is to preserve the source (master) data set in its original form
to protect against operator or system errors. Before beginning to work on a data set, it is
good practice to make two copies of it: one to save in its original state and another to work
on. For example, if the original file is called Project18.sav, I would make two copies of
the data: Project18[MASTER].sav and Project18[WORK].sav.
I would work on only the Project18[WORK].sav file and store Project18[MASTER].sav
someplace safe. If somehow the Project18[WORK].sav file became altered or corrupted, I
could easily create a new Project18[WORK].sav work file from the
Project18[MASTER].sav file.
 Index
Align property in SPSS, 35
Alpha (α) level, 9, 96, 99
Analysis of variance (ANOVA), 119–120
common sense considerations, 133–134
comparison with t test, 120
documenting results, 132
example results, 127–132
example scenario (math as usual, Z-math and flashcards), 120–121
homogeneity of variance, 124, 127
hypothesis resolution, 132
post hoc tests, 126–127, 129–131
pretest checklist, 122–124, 127
research questions and, 4
SPSS t test procedure, 94
using SPSS, 124–127
ANOVA. See Analysis of variance
Area sampling, 22
.asc (ASCII) files, 243
Ascending order sort, 232
ASCII data, importing, 243–247
Assumptions, 91. See also Pretest checklists
Availability sampling, 23
Average (mean), 16, 57
Backup data files, 249–250
Bar charts, 69, 70
Bell curves, 65–66. See also Normal curves
Bias, sampling, 24–25
Bimodal, 58
Bivariate correlation, 176
Categorical variables, 9, 10, 200
 chi-square for comparing, 200. See also Chi-square
descriptive statistics in SPSS, 67–69
recoding of continuous variables, 208, 236–240
selecting continuous variables by, 71–75
Causation, correlation and, 187–188
Chart Builder, 181–183
Chi-square, 199–201
documenting results, 206
example results, 205–206
example scenario (gender and party affiliation), 201–202
hypothesis resolution, 206
pretest checklist, 202–203
research questions and, 5
using percentages, 206, 207–208
using SPSS, 203–205
Cluster sampling, 22
Codebook, 37–39
Columns property in SPSS, 35
Comma-delimited ASCII data, 243–244
Comments, 249
Common sense considerations
applying ANOVA findings to real world, 133–134
correlation and causation, 187
interpreting statistical findings, 103
pretest/posttest designs, 165
saving and storing source data, 249–250
using software, 76–77
Confounding variables, 97–98
Continuous variables, 11, 200
descriptive statistics in SPSS, 61–65
recoding to categorical variable, 208, 236–240
selecting by categorical variable, 71–75
 summarizing using descriptive statistics, 57
Control groups, 5–7, 90
random assignment to, 7–8, 158
Convenience sampling, 23
Correlation, 175–176
causation and, 187–188
common sense considerations, 187
direction and strength of, 176–177
documenting results, 186
example results, 183–186
example scenario (homework hours and GPA), 177–178
hypothesis resolution, 186
pretest checklist, 178–179, 184–185
research questions and, 5
scatterplots and regression lines, 179, 181–184
using SPSS, 179–183
Cost, rationale for sampling, 17
Data entry. See Data View in SPSS
Data files
loading, 61
saving, 39
Data import for SPSS, 240
ASCII data, 243–247
Excel data, 240–243
Data saving and storing, 249–250
Data View in SPSS, 30, 36–39
Value labels, 36
Decimals property in SPSS, 33
Delimiter, 243
Delta% (Δ %) formula, 165
Descending order sort, 232
Descriptive statistics, 57–58
 mean (µ), 57–58
median, 58
minimum and maximum, 60
mode, 58
number (n), 57
range, 61
research questions and, 4
standard deviation, 59–60
variance, 60
Descriptive statistics, using SPSS, 61–64
categorical variables (gender), 67–69
continuous variables (age), 61–65
Frequency Statistics tables, 65, 69–70
histogram with normal curve, 65–66
Select Cases, 71–76
Dichotomous variables, 200
Direction of correlation, 176–177
Disproportionate/proportionate sampling, 21–22
Documenting results
ANOVA example, 132
chi-square example result, 206
correlation/regression example, 186–187
paired t test example, 164–165
t test example, 101–102
Excel data, importing, 240–243
Experimental groups, 5–7
random assignment, 7–8
External validity, 19, 30
nonprobability sampling and, 22, 23
sampling bias and, 24
Extrapolation and sampling, 17
Feasibility, rational for sampling, 17
Frequency Statistics tables, 65, 69–70
Generalization, 19, 22, 30
GIGO (garbage in, garbage out), 76
Hidden populations, 23
Histograms, 65–66
assessing for normality, 66, 92–93, 122–123, 160–161, 178–179. See also Normality,
assessing histograms for
normal curves, 65–66, 92
skewed distributions, 66–67
Homogeneity of variance, 93–94, 124, 127
test results, 96, 100–101
Homoscedasticity, 178–179, 184–185
Hypothesis formulation, 8
Hypothesis resolution, 129–131
ANOVA example result, 132
chi-square example result, 206
correlation/regression example result, 186
hypothesis rejection, 9
paired t test example result, 164
t test example result, 99–100
Importing data, 240
ASCII data, 243–247
Excel data, 240–243
Interpreting statistical findings, common sense considerations, 103
Interval variables, 10, 11
Label property in SPSS, 33
Levels of measurement, 9–10
Linearity, criteria for correlation/regression, 178, 179, 184
Loading a data file, 61
Logical relationships, 236
Master copy of source data, 249–250
Maximum, 60
Mean (µ), 57–58, 90
Measurement levels, 9–10
Measure property in SPSS, 35
Measures of central tendency, 58
Median, 58
Minimum, 60
Missing data in SPSS, 34–35
Mode, 58
Multimodal, 58
Multistage cluster sampling, 22
Name ascending sort, 232
Names for variables in SPSS, 31–32
Negative correlation, 176
Negative (left) skew, 67
Nominal variables, 9, 10. See also Categorical variables
Nonprobability sampling, 22–24
Normal curves, 65–66, 92, 160–163
Normality, assessing histograms for, 66
ANOVA pretest criteria, 122–123
correlation/regression pretest checklist, 178–179
paired t test pretest checklist, 160–163
t test pretest checklist, 92–93
Null hypothesis (H0), 8
rejecting/accepting based on p value, 99–100
Type I and Type II errors, 102
Number (n), 57
ANOVA pretest criteria, 124
chi-square pretest criteria, 202–203
One-Way ANOVA
SPSS ANOVA procedure, 124–127
SPSS t test procedure, 94–96
Ordering data (sorting), 230–234
Ordinal variables, 10
Outliers, 67
Paired t test, 157–158
common sense considerations, 165
documenting results, 164–165
example results, 163–164
example scenario (coffee and productivity), 159
hypothesis resolution, 164
pretest checklist, 160–163
pretest/posttest design, 158–159
research questions and, 5
using SPSS, 163
Paste function, 247–249
PASW (Predictive Analytics Software), 30
Percentages, documenting chi-square results, 206, 207–208
Pie charts, 69
Polychotomous variables, 200
Population, 16, 18, 19
generalizing from sample, 19, 22, 30
N, 57
Positive correlation, 176
Positive (right) skew, 67
Post hoc tests, 126–127, 129–131
Pretest checklists, 91
for ANOVA, 122–124, 127
for chi-square, 202–203
for correlation and regression, 178–179, 184
for paired t test, 160–163
for t test, 92–94, 96
See also Normality, assessing histograms for
Pretest/posttest design, 5, 158–159
common sense considerations, 165
See also Paired t test
Probability sampling, 20–22
Proportionate/disproportionate sampling, 21–22
Purposive sampling, 24
p value, 9, 96, 99
ANOVA without post hoc test, 126
as static value, 133
chi-square results, 200–201, 205–206
correlation results, 185
hypothesis resolution using, 99–100, 132, 164, 186
interpreting for multiple groups in ANOVA, 128–129
paired t test results, 159
Type I errors and, 102
Quota sampling, 24
Random assignment, 7–8, 158
Random number generation, 20, 228–230
Random sampling, 20
random number generation, 20, 228–230
Range, 61
Ratio variables, 10, 11
Reading statistical outcomes, 9
Rec. # column, 43
Recoding of variables, 208, 236–240
Regression, 5, 176
pretest checklist, 178–179, 184–185
using SPSS, 179–183
See also Correlation
Regression line, 179, 181–184
Reporting statistical results. See Documenting results
Representative sample, 19–20, 30
Research questions, 4–5
Role property in SPSS, 35–36
Rounding error, 30
Sample, 18–19
representative sample, 19–20, 30
Sample frame, 18, 19
Sample size, 30
disproportionate/proportionate sampling and, 22
Type II errors and, 102
Sampling, 15–16
area, 22
bias, 24–25
convenience, 23
disproportionate/proportionate, 21–22
nonprobability sampling, 22–24
probability sampling, 20–22
purposive, 24
quota, 24
rationale for, 16–17
simple random, 20
snowball, 23–24
stratified, 20–22
systemic, 20
terminology, 18–19
Saving and storing source data, 249–250
Saving data files, 39
Scatterplot, 179, 181–184
points, 183
regression lines, 179, 181–184
Select Cases, 71–76, 234–236
Sidak post hoc test, 126–127
Significance level, 98–99
Type I and Type II errors, 102
See also p value
Simple random sampling, 20
Simple time-series design, 158. See also Paired t test
Skewed distributions, 66–67
Snowball sampling, 23–24
Sorting cases, 230–234
Source data, saving and storing, 249–250
SPSS, 29–30
ANOVA procedure, 124–127
chi-square procedure, 203–205
common sense considerations, 76–77
computing normality of pretest/posttest difference, 160–163
correlation/regression procedure, 179–183
data import features, 240–247
Data View, 30, 36–39
descriptive statistics, categorical variables, 67–69
descriptive statistics, continuous variables, 61–65
importing data, 240–243
loading a data file, 61
paired t test procedure, 163
PASW (Predictive Analytics Software), 30
random number generation, 20, 228–230
recoding, 208, 236–240
saving data files, 39
scatterplots and regression lines, 179, 181–184
Select Cases, 71–76, 234–236
sorting cases, 230–234
supplemental operations, 227–228
syntax programming using Paste, 247–249
t test procedure, 94–96
variable recoding, 208
Variable View, 30–36
See also Data View in SPSS; Variable View in SPSS
Standard deviation (SD), 59–60
Statistical abstracts, 101–102. See also Documenting results
Statistical assumptions, 91. See also Pretest checklists
Statistical findings, common sense considerations for interpreting, 103
Statistical outcomes, reading, 9
Statistical significance level, 98–99
Type I and Type II errors, 102
See also p value
Stratified sampling, 20
disproportionate/proportionate, 21–22
Strength of correlation, 176–177
String variables, 32–33, 236
Summary statistics, 56. See also Descriptive statistics
Syntax programming using Paste, 247–249
Systemic sampling, 20
Time, rationale for sampling, 16–17
Treatment groups, 90
Trend line, 179
See also Regression line
t test, 89–90
comparison with ANOVA, 120
documenting results, 101–102
example results, 96–99
example scenario (Z-math), 90–91
hypothesis resolution, 99–100
pretest checklist, 92–94, 96
research questions and, 4
using SPSS, 94–96
See also Paired t test
Tukey post hoc test, 126–127, 129–131
.txt (text) files, 243
Type I errors, 102
Type II errors, 102
Types of data in SPSS, 32–33
Value labels, 36
Values property in SPSS, 33
Variable descriptions, in codebook, 37–39
Variable recoding, 208
Variable types, 10–11. See also Categorical variables; Continuous variables
Variable View in SPSS, 30–36
Align property, 35
Columns property, 35
Decimals property, 33
Label property, 33
Measure property, 35
Missing property, 34–35
Name property, 31–32
Role property, 35–36
Type property, 32–33
Values property, 33
Width property, 33
Variance, 60
homogeneity of, 93–94, 96, 100–101, 124, 127
Width property in SPSS, 33