SSD for R

SSD for R
An R Package for Analyzing Single- Subject Data
Second Edition

CHARLES AUERBACH AND WENDY ZEITLIN

 Oxford University Press is a department of the University of Oxford. It furthers the University’s
objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a
registered trade mark of Oxford University Press in the UK and certain other countries.
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America.
© Oxford University Press 2022
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or
transmitted, in any form or by any means, without the prior permission in writing of Oxford
University Press, or as expressly permitted by law, by license, or under terms agreed with the
appropriate reproduction rights organization. Inquiries concerning reproduction outside the scope of
the above should be sent to the Rights Department, Oxford University Press, at the address above.
You must not circulate this work in any other form and you must impose this same condition on any
acquirer.
Library of Congress Cataloging-in-Publication Data
Names: Auerbach, Charles, editor. | Zeitlin, Wendy, editor.
Title: SSD for R : an R package for analyzing single-subject data /
Charles Auerbach & Wendy Zeitlin.
Description: Second edition. | New York, NY : Oxford University Press, [2022] |
Includes bibliographical references and index.
Identifiers: LCCN 2021025228 (print) | LCCN 2021025229 (ebook) |
ISBN 9780197582756 (paperback) | ISBN 9780197582770 (epub) |
ISBN 9780197582787
Subjects: LCSH: Social sciences—Research. | Single subject research.
Classification: LCC H62 .A8493 2022 (print) | LCC H62 (ebook) |
DDC 001.4/2—dc23
LC record available at https://lccn.loc.gov/2021025228
LC ebook record available at https://lccn.loc.gov/2021025229
DOI: 10.1093/oso/9780197582756.001.0001
Contents

Introduction: Single-Subject Research Designs in the Behavioral

and Health Sciences
1. Getting Your Data Into SSD for R
2. Overview of SSD for R Functions
3. Analyzing Baseline Phase Data
4. Comparing Baseline and Intervention Phases: Visualizing Your
Findings and Descriptive Statistics
5. Statistical Tests of Type I Error
6. Analyzing Group Data
7. Meta-Analysis in Single-Subject Evaluation Research
8. Using RMarkdown to Present Your Findings
9. Building Support for Practice Research

Appendix A: Entering and Editing Data Directly in R

Appendix B: SSD for R Quick Functions Guide
Appendix C: Decision Trees
Appendix D: Bibliography of Additional Resources
References
Index
Introduction
Single-Subject Research Designs in the Behavioral and Health
Sciences

Introduction
This introduction provides background information on single-subject
research and its use in the behavioral and health sciences. Here you will
find a brief history of the use of this type of research design to provide a
backdrop for its current use and a discussion of future directions in the use
of these designs. The purpose of this is to provide contextual information
for the introduction of SSD for R, a visual and statistical software package
that is easily accessible and useful in the analysis of single-subject data.
SSD for R is a package written in R, a free and open-source statistical
programming language (The R Project for Statistical Computing, n.d.). In
this Introduction, we also go over the contents of this book, including
updates in the second edition.

Description and Usage of Single-Subject Research

Single-subject research is also referred to in the literature as “n = 1
research,” “interrupted time series research,” and “single-case research.” By
whatever name, single-subject research designs are substantively different
from the more commonly published group research designs; instead of
examining aggregate data for multiple research subjects simultaneously,
single-subject research is concerned with the empirical examination of a
single research subject over time. While a single research subject could
mean an individual, it could be any subject that could be conceptualized to
be a single unit. For example, single-subject research could just as easily
use a couple, family, therapeutic group, community, or even a larger
population as its unit of analysis.
 Single-subject research has some unique characteristics that differentiate
it from other types of research designs. First, the commonest use of this is
in the evaluation of interventions (Smith, 2012). In single-subject research,
however, the subject serves as his or her own control or comparison. To
accommodate this, single-subject research designs typically collect data
using frequent repeated measures over time, with collected data being
assigned to phases (e.g., baseline, intervention) with the baseline, or
preintevention, phase often used as the control or comparison. Therefore,
single-subject research is a type of time series in which the series may be
interrupted with the introduction or withdrawal of one or more
interventions.
Single-subject research, although quantitative, has some of the
characteristics of narrative case studies that focus on individuals. Like case
studies, single-subject research often contains a detailed description of the
subject and, unlike group designs, also contains a detailed description of the
intervention or experimental condition (Kazdin, 2011). However, with its
inclusion of repeated measures, single-subject research adds a level of
methodological rigor not found in case studies. To this end, single-subject
research falls into the realm of scientific, quasi-experimental research,
while narrative case studies are considered prescientific (Miller, n.d.).
Unlike other designs, single-subject research can have two distinct, but
equally important, purposes: for formal research such as studies published
in peer-reviewed journals and the less formal, but equally important,
evaluation of practice. In terms of formal research, single-subject studies
have been published in a wide variety of social science, health, and allied
health fields, including social work, psychology, education, speech therapy,
rehabilitation medicine, occupational therapy, physical therapy, and
physiology. The advantages to utilizing single-subject research designs
include the ability to examine conditions that are rare or to examine unique
client/patient populations (Janosky, Leininger, Hoerger, & Libkuman,
2009). They can also be used to introduce a new intervention or practice
modality into the professional literature.
With regard to practice evaluation, single-subject research techniques can
be used to empirically evaluate client progress over time, simply and
dynamically. Techniques used to capture and evaluate client data in order to
do this type of analysis are easily learned, inexpensive, and simple to
implement and can be built into typical practice. The benefit of including
this type of evaluation into ordinary practice is the ability to more
accurately and empirically evaluate client progress longitudinally, which
can then be used to inform the practitioner’s work with individual client
systems. That is, practitioners can readily evaluate their clients’ progress
over time to adapt their work to meet the specific needs of those they serve.

The History of Single-Subject Research

Single-subject research has long been part of the repertoire of inquiry in the
behavioral and health sciences. Some of the earliest known research using
these designs were the physiological studies conducted by Mueller and
Bernard in the 1830s (Krishef, 1991). From the late 1880s through the early
1900s, a good deal of experimental psychology focused on only one or a
few subjects in a given study. These included Pavlov’s quantitative
analyses, Breuer’s qualitative cases studies, and Ebbinghaus’s memory
studies (Kazdin, 2011; Krishef, 1991).
In the early to mid-1900s, there was a general shift in research to
studying groups instead of individuals. This has been attributed to the
development and popularity of group statistical techniques such as the t test.
Group analysis, then, gained favor as it is difficult to attain statistical
significance with small sample sizes, including the typical number of small
observations found in most single-subject research designs. Therefore,
experimental group designs became more popular due to their scientific
rigor (Kazdin, 2011; Krishef, 1991).
Learning and behavioral analysis, such as the type studied by Skinner,
became the forerunner of modern single-subject research designs (Kazdin,
2011). In those studies, research was used to develop and refine theory (J.
D. Smith, 2012). Additionally, many measures were behaviorally based,
often noting the frequency of a target behavior. This allowed for the close
examination of the impact on an individual of an intervention that simply
could not be detected using group analysis. This research focused less on
statistical analysis and more on observed change over time (Kazdin, 2011).
In the early 1970s, Bergin and Strupp noted the inherent limitations of
group research designs. These included the high costs associated with
conducting group research, the ethical issues related to assigning subjects to
control conditions, the difficulty of applying results from group analyses to
individual clients, and the difficulty in explaining why some subjects would
improve with treatment while others did not (Krishef, 1991). More recently,
questions have continued to be posed about the utility of group findings
from randomized control trials to actual practice settings. This concern has
resulted in a renewed and growing interest in the area of single-subject
research designs (J. D. Smith, 2012).

Single-Subject Research Today

In recent years, the scientific community has seen renewed interest in
single-subject research, and it is also increasing in relevance due to
professional organizations that espouse high practice standards of its
membership. For example, in 2008 and again in 2015, the Council on Social
Work Education (CSWE), the accrediting body of schools of social work in
the United States, specifically identified two areas in which single-subject
research would be appropriate in its Educational Policy and Accreditation
Standards. Competency 4 states that accredited professional social work
programs should contain curriculum that builds student competency in
“engag[ing] in research-informed practice and practice informed research.”
Additionally, Competency 9 directs programs to teach social worker
students to “evaluate practice with individuals, families, groups,
organizations, and communities” (CSWE, 2015). Similarly, the National
Association of Social Workers’ (NASW) Code of Ethics requires social
workers to evaluate their own practices under Ethical Standard 5.02—
Evaluation and Research. Specifically, “social workers should monitor and
evaluate policies, the implementation of programs, and practice
interventions,” and “social workers should critically examine and keep
current with emerging knowledge relevant to social work and fully use
evaluation and research evidence in their professional practice” (NASW,
2017). While single-subject research is not the only way to evaluate clinical
practice, the techniques available in these designs are easy to learn and
congruent with the goals set forth by both CSWE and NASW.
Additionally, the evidence-based practice movement has attempted to
bring more rigor into practice settings by incorporating scientific research
into clinical work with the hope of improved service to clients (Thyer &
Myers, 2011). To this end, single-subject research designs have been used
to build evidence to the effective treatment of problems such as attention
deficit hyperactivity disorder and other behavioral disorders, schizophrenia,
and depersonalization disorder (Nathan & Gorman, 2002; Schudrich, 2012).
One of the advantages of doing this type of research is the ability to
evaluate interventions as they are delivered in practice settings with actual
clients, which differs dramatically from group intervention designs
(Beddoe, 2011). In this way, practice-based research can be used to help
bridge the translational research gap between group research designs and
actual practice.
Single-subject research designs are considered for inclusion in systematic
reviews by respected groups such as the Cochrane and Campbell
Collaborations, provided they are scientifically rigorous (Higgins &
Thomas, 2020; Schlosser & Wendt, 2008; Scruggs & Mastropieri, 2013;
Wendt, 2009). With systematic replication of single-subject research,
studies can uncover causal relationships between an intervention and the
dependent variable (Schlosser & Wendt, 2008).
In order to help elevate what constitutes scientifically rigorous single-
subject studies, various groups valuing this type of research have developed
research standards. These groups include the What Works Clearinghouse
from the U.S. Department of Education, the National Institute of Child
Health and Human Development, and several divisions of the American
Psychological Association: Division 12 (clinical psychology), Division 53
(clinical child and adolescent psychology), and Division 16 (evidence-based
practice in school psychology) (Smith, 2012; Thyer & Myers, 2011).
Common standards found across some of these groups include

• Use of a minimum number of data points per phase (e.g., 3 to 5 in

the baseline)
• More than one judge to visually analyze data to increase interrater
reliability
• The use of validated measurement instruments
• The use of multiple independent measures
• The use of multiple single-subject research studies
• The requirement for alternating treatment designs (e.g., ABAB)
• Reaching levels of statistical significance between the treatment and
test condition (Chambless et al., 1998; Kratochwill et al., 2013;
Vannest, Davis, & Parker, 2013)

The Future of Single-Subject Research

Researchers and others interested in single-subject research designs have
been increasingly productive in better understanding existing statistical
methods and developing new techniques. For example, in 2014, a group of
psychologists, special educators, and statisticians collaborated to introduce
new methodologies and present reactions to these from leaders in the
behavioral and social sciences (Kratochwill & Levin, 2014).
More recently, Evidence-Based Communication Assessment and
Intervention published a special issue, Advances in Statistical Analysis and
Meta-analysis of Single-Case Experimental Designs (Wendt & Rindskopf,
2020). This volume highlighted newer statistical techniques that allow for
aggregating multiple single-subject studies into one, which is more rigorous
than analyzing one case at a time, as well as newer measures of effect size,
which are all discussed further in this text.
In our work developing SSD for R, we have taken many of these
advances into consideration, and this second edition describes new
functions that were not available when we published the original text.

Visual Analysis? Statistical Analysis?

Historically, visual analysis has been the preferred mode of examining
single-subject research data, and the depiction of graphs showing data
across phases has been one of the hallmarks of published single-subject
research studies. For the most part, this preference for visual interpretation
of data continues to exist; however, there are situations in which statistical
analysis may be used to supplement visual analysis. For example, statistical
analysis is appropriate in cases when treatment “effects are small; the
variation is large; the effects of the intervention do not appear immediately
as phases change; [or] there is no clear trend in the data within phases”
(Auerbach & Schudrich, 2013, p. 346).
There are two additional concerns that have been raised with regard to
solely using visual analysis to interpret single-subject research data. First,
autocorrelation, or serial dependence, can greatly impact visual
interpretation, and there is no way to detect autocorrelation without
statistical analysis (Auerbach & Schudrich, 2013; Smith, 2012). The other
issue has to do with the mechanics of visual analysis. Studies have shown
that there is inconsistency in how graphs are interpreted, even when skilled
raters have received the same training (Janosky et al., 2009; Kazdin, 2011;
Nourbakhsh & Ottenbacher, 1994). Therefore, there can be multiple
interpretations for the same data when visual analysis is the only tool that is
used.
While there has been consensus that there is a need for statistical analysis
in at least some single-subject research studies, there is an ongoing
discussion of how this should be done. That is, it is not always clear which
statistical techniques should be applied to single-subject data. However, as
in group research designs, it is important to consider that the types of
statistical treatments applied to single-subject research designs should be
based on the characteristics of the data (J. D. Smith, 2012).
Throughout this book, then, you will notice that we often present
methods to analyze data both visually and statistically, and we provide
examples of how you would interpret and present your findings based on
both visual and statistical analysis. These examples are applicable whether
you are using these techniques for the purposes of evaluating your own
practice or for the sake of producing publishable research findings.

Tools for Evaluating Single-Subject Research Data

While single-subject research has existed for many years, there has been a
dearth of software to help analyze this type of data. This has been
problematic because statistical software that has been intended for group
research designs is often not appropriate for analyzing single-subject data.
In terms of visual analysis, it appears as if many researchers use
Microsoft Excel or similar spreadsheet packages to draw line graphs. J. G.
Orme and Combs-Orme (2011), for example, published a text that uses
Excel macros to produce graphs specifically for the visual interpretation of
single-subject data.
Another group of researchers has improvised a way of statistically
analyzing visually presented data by using several software packages
consecutively. First, graphs are scanned into Ungraph to digitize x-y
coordinates, then SPSS is used to prepare the data for import and eventual
analysis in HLM (Nagler, Rindskopf, & Shadish, 2008). While this is an
interesting proposition, a couple of obvious issues arise. First, researchers
have to have access to all three propriety software packages, which can be
expensive. Additionally, the analytical capabilities of this combination, as
described by Nagler and colleagues, are limited.
 Simulation Modeling Analysis was designed for analyzing single-subject
data in the healthcare arena (Borckardt, 2008). While this has been a useful
tool, it also has limited visual and statistical functionality.
Pustejovsky and Swan (2018) have created a web-based effect size
calculator that can be accessed through their website or from the
Comprehensive R Archive Network under the package SingleCaseES. Their
software enables users to enter data for two phases and then compute a
number of effect sizes with additional output and graphs.
Some additional limited resources for analyzing single-subject data have
been developed, and these are listed in Appendix D, Bibliography of
Additional Resources.

What This Book Is and What This Book Isn’t

This book is designed to be a helpful tool in the use of SSD for R for
visually and statistically analyzing single-subject data. SSD for R has the
most comprehensive functionality specifically designed for the analysis of
single-subject research data currently available (Auerbach & Zeitlin
Schudrich, 2013). The aim of this text is to introduce readers to the various
functions available in SSD for R and to step them through the analytical
process based on the characteristics of the collected data.
SSD for R has numerous graphing and charting functions available to
conduct robust visual analysis. For example, besides the ability to create
simple line graphs, additional features are available to add mean, median,
and standard deviation lines to help better visualize change over time.
Graphs can also be annotated with text.
Additionally, graphing functions are available to depict several non-
overlapping effect sizes. Some of these, such as the Percentage of All Non-
overlapping Data (PAND) and the Improvement Rate Difference (IRD),
have been more difficult to calculate with larger datasets previously;
however, introduction of software to graph these effect sizes has made it
considerably easier to actually calculate these numeric values. This is
particularly important when publishing single-subject research as calculable
effect sizes are necessary in order to include these studies in meta-analyses
(Morgan, 2008; Parker, 2006; Schlosser & Wendt, 2008; Scruggs &
Mastropieri, 2013).
 SSD for R also contains a host of statistical process control (SPC) charts
that can be used to visually detect changes in the dependent variable
between phases. The usage of the various available SPC charts is explained
in detail in this book so that appropriate methods can be selected based on
the characteristics of your data.
This book provides a thorough explanation of traditional statistical tests
that have been applied to single-subject research and teaches readers how
and when to apply these to their own research or practice evaluation
situations. These include descriptive statistics as well as tests of statistical
significance, such as t tests and the conservative dual criteria. Didactic
material is provided on the interpretation of statistical and supporting visual
output from SSD for R for each of these functions.
Following both the traditions and current trends in single-subject
research, most functions in SSD for R contain both visual and statistical
output. We believe that this is one of the major strengths of SSD for R, as a
combined approach is likely to provide support for the most pragmatic
interpretations of this primarily intervention-based type of research. For
example, in your own research, you may discover that your study yielded
statistically significant findings that may not be clinically significant or vice
versa. We believe that a combination of both visual and statistical analysis
is likely to lead practitioners and researchers to make the most appropriate
decisions for their clients.
This book contains numerous examples using sample data that is
available to you. We recommend that as you read through the book, you
download these files and replicate the examples yourself in order to learn
and master the SSD for R commands.
In addition, each chapter contains exercises that are cumulative
throughout this book. In this way, skills that are learned in previous
chapters are reinforced through repetition and application in further
chapters.
This book, however, is not a book on single-subject research
methodology. Therefore, this text does not provide an overview of topics
such as different types of single-subject research designs, an in-depth
discussion of measurement, or discussions of reliability and validity. There
are many excellent texts and resources available to help you better
understand the broader field of single-subject research methods, and many
of these are listed in Appendix D. We suggest referring to these to gain a
better understanding of single-subject research in general and to assist you
in designing your own studies.

The Structure of This Book

We have designed this book to help novice users of SSD for R walk through
the process of data collection and analysis. The first section of this book
consists of two chapters and provides a general introduction to both single-
subject research in general and SSD for R more specifically. Chapter 1
discusses data collection and walks through an example of creating a
spreadsheet collecting hypothetical data. In this chapter, we also introduce
three practice senarios that will be used to illustrate practice evaluations
throughout the rest of the book. Chapter 2 provides an overview of SSD for
R, including how to access it, open files, and begin working with this
package.
The next section of this book consists of three chapters that will guide
your data analysis. Chapter 3 covers the analysis of baseline data, including
addressing issues specific to single-subject research data. Chapter 4 then
covers visual comparisons of phase data, while Chapter 5 introduces readers
to statistical comparisons of phase data.
The remaining section of this text contains chapters in special topics
related to single-subject research data analysis. Chapter 6 covers analyzing
group data. This is particularly useful if the client system under
consideration is made up of more than one individual. Chapter 7 contains a
discussion of aggregating multiple single-subject research studies using
meta-analysis. Chapter 8 covers presenting findings by using RMarkdown, a
tool readily and freely available to R users. Finally, Chapter 9 covers how
organizations can build research capacity by supporting research activities
and reducing barriers to building evidence.
Finally, this text contains a number of appendices to make using this text
and SSD for R easier. Appendix A provides instruction on how to enter and
edit directly in R. This is useful if you do not have access to spreadsheet
software such as Excel or Google Sheets. Appendix B provides detail about
each of the SSD for R functions grouped by the type of analysis you would
want to do. This appendix can be used if you are unsure of which function
to select. Appendix B contains a series of decision trees. These flowcharts
will help you select functions for specific analyses based on the
characteristics of your data. Finally, Appendix C provides resources that
can be used to enhance your understanding of single-subject research.
These include books, websites, and other resources that you might find
helpful in addition to this text.

Conclusion
Single-subject research has and continues to have an important position in
the areas of both research and practice evaluation. With the current trend
toward evidence-based practice, we believe that the demand for and
reliance on this type of research is only going to increase in the future. SSD
for R is a software package available to help researchers and practitioners
analyze single-subject research data.
This text is designed to help readers learn to use both the visual and
statistical functions in SSD for R. Examples, screenshots, and instructions
for when and how to use each of these functions help to make this book a
useful tool in the interpretation of single-subject data analysis in general
and in the use of SSD for R in particular.
1
Getting Your Data Into SSD for R

Introduction
In this chapter you will learn how to measure target behaviors and use Excel
or other software to record and edit client data. You will then be able to
import these data into R and use the SSD for R functions to analyze them.
The first part of this chapter focuses on the types of data you will want to
record and some common issues related to collecting these. While an
overview of this material is covered in this chapter, additional resources that
include these topics in depth are listed in Appendix D. The second part of
this chapter shows you how to use Excel or another spreadsheet program to
quickly and effectively record the data.

What to Measure
Single-subject research designs rely on repeated measures of target
behaviors that are related to client problems. You can measure one or more
target behaviors using Excel or any software package whose files can be
saved as comma-separated values. This data can then be analyzed with SSD
for R. There are several types of measurements you can use to accomplish
this: direct behavioral observations, standardized scales, individualized
rating scales, and logs.

Direct Behavioral Observations

Direct behavioral observations refer to target behaviors that are easily
observed by the client or someone else. For example, Mary is an elderly
client who is experiencing social isolation. Mary or someone else, such as a
family member or caregiver, could easily report how many times she
participates in social activities at the local senior center each week.
When measuring behaviors directly, whoever is actually collecting the
information is typically counting the number of times actual events occur.
Depending on the type of behavior you are measuring and the importance of
time in assessing the behavior, you may want to eventually record these data
as proportions. For example, let’s look at the example of our socially isolated
client. Data collection lasted 4 weeks while Mary was being assessed, during
which time she was asked to report how many times a week she went to the
senior center. The results are shown in the table below.
Week # of Days Mary Goes to the Senior Center Each Week
1 0
2 2
3 1
4 1

While you could record these data in Excel, you, as the social worker,
realize that during Week 4, the senior center was only open 5 days instead of
its customary seven due to a holiday schedule. To improve the validity of
your measurement, then, you may want to record the proportion of the week
that Mary goes to the senior center by dividing the number of days she
attends by the possible number of days that she could attend. Your data for
this time period then would be reported in Excel differently:
Week Proportion of Days Mary Went to the Senior Center
1
; record “0”
2
; record “0.29”

3
; record “0.14”

4
; record “0.20”

Standardized Scales
Standardized scales are instruments that have typically been developed for
use in either practice or research to measure a particular problem or
phenomenon. Standardized scales usually have a consistent method for
scoring, and scores can be compared across individuals (Bloom et al., 2009;
Orme & Combs-Orme, 2011).
You may choose to measure one or more target behaviors with
standardized scales; however, like other types of measures, there are
advantages and disadvantages to using these. Some advantages include the
ability to measure multiple aspects of a complex construct, the availability of
reliability and validity information in many cases, and ease of use.
Disadvantages may include the appropriateness of a given standardized scale
for a particular practice situation and the length of time the scale may take to
complete.
To more closely understand these advantages and disadvantages, let’s look
at the example of Mary a bit further. You, as the social worker, suspect that
Mary’s social isolation may be a result of depression. There are several ways
to try to assess possible depression in this client, but one way would be to
use the Beck Depression Inventory (BDI) (Beck, Ward, Mendelson, Mock,
& Erbaugh, 1961). This is a 13- or 21-question survey that is considered
both a reliable and a valid assessment of depression in adults and includes
questions about various facets of depression, including mood, pessimism,
and guilt (Thombs, Ziegelstein, Beck, & Pilote, 2008). While this, then,
would seem like an obvious measure of depression to use with Mary, it is
fairly lengthy, and it would not make sense to have her complete it daily or
even weekly. With this in mind, it might make sense clinically to use the
BDI to evaluate her depression and track Mary’s progress over time, but for
the purposes of a single-subject practice evaluation, you would probably
want to use this measure in conjunction with others as you might want more
frequent measures to aid in the analysis of your work with this client.
There are many sources for locating standardized scales, but one place to
start is with the work of Orme and Combs-Orme (2011), who included scales
and references for online and published standardized scales in two
appendices.

Individualized Rating Scales

The individualized rating scales are created by the practitioner or researcher
specifically to meet the needs of an individual client. Typically, these scales
are brief and may consist of only one or two items. These can be used when
no standardized scales are available that adequately assess the target
behavior or you would prefer to use a brief measure of a target behavior.
 The advantage of using individualized rating scales is that they can be
used with almost any client since they are designed to specifically meet the
needs of an individual. They are typically easy to administer and score, they
are appropriate for measuring the intensity of a target behavior, and they are
good for measuring beliefs, thoughts, and feelings. When well designed,
individualized rating scales are usually considered reliable and valid when
used repeatedly to measure a target behavior with the same individual
(Bloom et al., 2009; Orme & Combs-Orme, 2011).
Care should be taken in designing self-anchoring scales for use in practice
evaluation and/or single-subject research. Several excellent resources for
designing these are listed in Appendix D.
To better understand how individual rating scales can be used in single-
subject research, let’s return to the example of Mary, our socially isolated,
possibly depressed, senior citizen. The social worker has decided that she
would like to use an individualized rating scale to assess Mary’s mood and
comfort with going out in public. She creates an individualized rating scale
made up of two questions:
1. On a scale of 1 to 5 with 1 being the most uncomfortable you have ever been and 5 being the
most comfortable you have been, how did you feel the last time you went to the senior center?
2. On a scale of 1 to 5 with 1 being the saddest you have ever felt and 5 being the happiest you
have ever felt, what has your mood been like, on average, since we last met?

During Mary’s assessment period, which has lasted 4 weeks, you collect the
following information:
Week Comfort at Senior Center Mood
1 1 3
2 4 2
3 2 2
4 2 1

Note that Mary could very easily be asked these questions when she meets
with her social worker, and the responses gathered could actually be used
therapeutically during the session in which she reports the information.

Logs
Logs can be referred to as client-annotated records, critical incident
recordings, or self-monitoring, but no matter how they are referred to, logs
can be used to provide a qualitative dimension to quantitative data you
collect using other means. In general, a log is an organized journal of events
related to client problems. The log usually includes the clients’ perceptions
of these and/or the circumstances under which they occur. While we might
ask a client to use a log, they can also be completed by professionals and
significant others to get a deeper, more thorough understanding of the
overall problem we are seeking to address. Logs, then, are personal and
specific to the individual client and can be used to help formulate client-
specific interventions.
The format of a log is up to the preferences of the practitioner and client.
Some logs are very structured and can be in the form of a table that the client
is asked to fill in. Some are very unstructured and can simply be made up of
a notebook, pad of paper, or mobile device that the client uses for the
purposes of recording events and his or her reactions to these.
When using logs with clients, you will want to decide whether you want
them used at preset times or if they are to be open ended. When the client’s
problem occurs regularly, you may want to suggest that logs be used to
record events and reactions at preset times. When this is done, logs can
generate a lot of information due to their frequent use; however, if the
problem is sporadic, regular use of logs can generate large amounts of
useless information. In this case, you may suggest that your client use his or
her log only when the problem occurs.
In order to maximize the reliability of logs, you should consider
discussing the importance of accurate reporting with your clients and assure
them that the use of these falls within the bounds of confidentiality set up in
other aspects of your work. Creating a structured format for logs may help
minimize difficulties in clients using them. You will also want to be sure that
your clients understand what they are to record and how often. If you are
asking your clients to use logs when an event occurs, it is advisable to have
them record the event and their reactions as close in time to the event
occurring as possible. In all cases, you will want to encourage your clients to
be as accurate in their reporting as possible without inadvertently
encouraging inaccurate reports of improvement.
Both Bloom, Fischer, and Orme (2009) and Orme and Combs-Orme
(2011) provide more detailed information about the use of logs in single-
subject designs, and both texts provide templates for logs that can be tailored
to meet your clients’ needs.
Using Spreadsheets to Record Client Data
Once you determine what you are going to measure, how you are going to
measure it, and how often you are going to measure, you will want to collect
and record these data so that it can eventually be imported and analyzed in
SSD for R. In this section, we walk you through the steps necessary in order
to do this accurately.
One of the simplest ways to bring your client data into SSD for R for
analysis is by entering it into Excel, Google Sheets, Numbers, or any other
program that can create “.csv” files. Because Excel is the most commonly
used spreadsheet program, this chapter shows you how to enter data in
Excel. Other programs used for entering data will use a method similar to,
although not exactly the same as, Excel.
In some cases, you may not be able to use Excel or another program to
enter your data. It is possible to enter your data directly into R for analysis
into SSD for R. This is explained in detail in Appendix A, “Entering and
Editing Data Directly in R.”
If you are using Excel or another program, you will need to create a
separate file for each client/group; however, you can track multiple
behaviors for each client/group in a single file.
Creating a file that can successfully be imported into SSD for R has to be
done in a particular manner. To do this, complete the following steps:
1. Create a folder that will be used to store your data and give it a name you will remember, such
as ssddata.
2. Open Excel.
3. On the first row (labeled “1”), enter names for each behavior (i.e., variables) across the
columns, beginning with column A.

HELPFUL HINT: For each behavior you are measuring, you will need to create both a behavior
variable and a corresponding phase variable. The behavior variable will measure the target
behavior, and the phase variable will indicate in which phase, baseline or intervention, the
measurement occurred. In order to do this systematically, we recommend giving the behavior a
meaningful name, and the associated phase variable name should be similar to the behavior
variable name. For example, if you are measuring crying behavior, your behavior variable
could be named “cry” and your phase variable could be named “pcry.”

4. Starting in row 2, begin entering your data for each occurrence the behavior is measured.

IMPORTANT NOTE: When phases change, you will need to enter “NA” (do not enter “N/A” or
“na”) into the row between a change in phases for both the behavior and phase variable.
 To look at a relevant example, let’s return to the example of Mary,
discussed previously in this chapter. You will note that the social worker
ultimately collected three measurements during her assessment: Mary’s
attendance at the senior center, her comfort with going to the senior center,
and her mood. In order to record these data in Excel, we will need to create
three behavior variables and three corresponding phase variables. We will
call the behavior variables “attendance,” “comfort,” and “mood.” We will
call the corresponding phase variables “pattendance,” “pcomfort,” and
“pmood.” In addition, we will also create a variable called “notes” to record
any critical incidents that might have occurred.
In Figure 1.1, note that we have entered baseline, or preintervention, data
for Mary. If we had begun our intervention, we would continue entering that
data on Excel row 7; however, the phase variable would change as “A”
generally denotes baseline data, while “B” indicates data collected during
the intervention.

Figure 1.1 Baseline data entered into Microsoft Excel

5. Once your data are entered into Excel, you will need to save it as a “.csv (Comma delimited)”
or “.csv (Comma Separated Values)” file in your ssddata directory. To do this, click SAVE AS
and choose a name for your file. Do NOT click SAVE, but instead select one of the .csv options
from the drop-down menu for SAVE AS TYPE or FORMAT, as shown in Figure 1.2 (see p.
16). After you finish this, you should click SAVE and close Excel. You may receive several
warnings, but you can accept all of these by selecting CONTINUE.

Once you enter data into Excel, you can import it into SSD for R and
begin your analysis. In the next chapter, you will learn how to access R, the
SSD for R software, and RStudio user interface to complete this analysis.
Figure 1.2 Saving a Microsoft Excel file in “.csv” format

Conclusion
In this chapter, you learned about different ways to measure target behaviors
associated with your clients’ identified problems. Four different types of
measurement exist, each of which has its own strengths and weaknesses.
These include direct behavioral observations, standardized scales, individual
rating scales, and logs. When selecting one or more methods of measuring a
target behavior, you will want to consider the specific needs of your client,
identified problem, and practice or research situation.
As you collect your client data, Excel is an outstanding tool for recording
this as Excel files saved in .csv format can be imported directly into R for
analysis using the SSD for R package. This chapter showed you step-by-step
instructions to accurately and easily format and save these Excel files. A
method for entering and editing data directly in R is detailed in Appendix A.

Chapter 1
Assignment 1.1—Entering Data
Brenda is a developmentally delayed client with oppositional behavior. She
was in danger of not being accepted into a residential program because she
didn’t get along with others. Let’s examine one aspect of her behavior,
oppositional behavior in school.
This was important to measure because each episode typically lasted 20–
30 minutes and was disruptive to the school day.
For this assignment, you will create a Excel or Google Sheet with
Brenda’s baseline data. BE SURE TO STORE THIS SPREADSHEET IN A
PLACE YOU CAN ACCESS LATER, as you will use it for future
homework assignments. The data you need for this is in the following table:

Day # of Oppositional Episodes

1 2
2 1
3 1
4 1
5 3
6 2
7 1

1. Create the spreadsheet and name it “Brenda—FirstLast” with FirstLast being YOUR first and
last name. For example, the file would be titled “Brenda—WendyZeitlin.”
2
Overview of SSD for R Functions

Introduction
SSD for R is a software package composed of a set of statistical functions
that are designed for the analysis of single-subject research data using R, a
free, open-source statistical programming language (R Core Team, 2013). R,
in general, is becoming popular. There are numerous readily available, free
resources to help people get started with basic and advanced use and
programming; interest in R has virtually exploded in recent years. In this
chapter, you are given step-by-step instructions on how to access the
software necessary to use this package. You also are presented with a brief
overview of the capabilities of the SSD for R package. How and when to use
these packages are expanded on in subsequent chapters.

Getting Started
SSD for R is a set of functions that was designed for single-subject research
in the behavioral and health sciences and is written in R, a programming
language designed for statistical analysis and graphics. R provides an
environment where statistical techniques can be implemented (The R Project
for Statistical Computing, n.d.). Its capabilities can be extended through the
development of functions and packages. Fox and Weisberg stated, “One of
the great strengths of R is that it allows users and experts in particular areas
of statistics to add new capabilities to the software” (2010, p. xiii). SSD for R
is a software package that extends the capacity of R to analyze single-system
research data (Auerbach & Zeitlin, 2021; Auerbach & Zeitlin Schudrich,
2013).
 Throughout this book, we demonstrate statistical procedures using the
SSD for R package. In order to get started using SSD for R yourself, you will
need to download three things: R; RStudio, which we use as a user interface
for R; and the SSD for R package (Auerbach & Zeitlin Schudrich, 2013; Free
Software Foundation, Inc., 2012; R Core Team, 2013). R and RStudio can be
downloaded either directly or through links from The Single-System Design
Analysis website (https://www.ssdanalysis.com). Once R and RStudio have
been downloaded, the SSD for R package can be installed using a simple R
command.
Begin by downloading R and RStudio free of charge from the Single-
System Design Analysis website (https://www.ssdanalysis.com or from the
companion site to this text at Oxford University Press). When you click the
links for each of these, you will be taken to external sites. Both R and
RStudio are completely free and are considered safe and stable downloads.
RStudio has developed a cloud version as an alternative to running
RStudio on your computer. All you need to run the cloud version is a web
browser and an RStudio Cloud account. You will also find instructions on
using the RStudio Cloud version on both the Single-System Design Analysis
website and the companion site to this text at Oxford University Press.
Throughout the rest of this book, we use sample datasets that you can
download from https://www.ssdanalysis.com or from the companion site to
this text at Oxford University Press. To download these, click on the
“Datasets” tab from the home page and follow the instructions. These files
include stability.csv, jenny.csv, jennyab.csv, ed.csv, and
social_skill_group.csv. R scripts for each chapter are also provided on the
website.
As you are getting started, you may want to create a single folder on your
computer to hold your R data files. These downloaded files should be stored
there.
Once these are installed, open RStudio. When you open it, your screen
should look like Figure 2.1.
Figure 2.1 RStudio

From within RStudio, type the following R command in the Console:

>install.packages(“SSDforR”) and press <ENTER>.

After the SSD for R package is installed, type the following in the Console:
>require(SSDforR) and press <ENTER>.

Another way to load the SSD for R package is to click the Packages tab in
the lower right pane of RStudio. Click “Install Packages,” and a dialogue box
will appear. Select “Repository (CRAN)” as the location to “Install from:”
Under “Packages,” type “SSDforR.” Select this package and then click
“Install.” That package will automatically be installed, and the Packages tab
will be refreshed. Now, check the box next to SSDforR to make it available
for your use. You will only need to install SSD for R once but will need to
check the box for each new R session.
Your screen should now look like what is displayed in Figure 2.2.
Important Note: The package only needs to be installed once using
install.packages(). This command can also be used to install other R
packages and upgrades. The require(SSDforR) command has to be run once
at the beginning of each R session.
Figure 2.2 RStudio with SSD for R installed and required

Getting Help
You can obtain a list of the SSD for R functions by typing the following in
the Console:
>SSDforR()

The list of functions will appear in the Console on the left. You can obtain
help for any function in the list by typing a question mark (?) followed by
the function title. For example, if you need more information on using the
ABrf2() function, you would enter the following in the Console:
>?ABrf2

Help for the command will appear in the bottom right window under the
Help tab. Your screen should look as it does in Figure 2.3 (see p. 21).
Figure 2.3 RStudio with functions listed and help file displayed

All help files have the same structure. They begin with a brief description
of what the function does. The “Usage” section shows the required syntax
for the function, and the “Arguments” section describes what is required in
the syntax. The “Authors” section lists the authors of the function, and the
“References” section provides additional information. At the bottom of each
help file is an “Examples” section that can be copied and pasted into the
Console and run. For example, type the following in the Console:
>?ABplot

When you scroll to the bottom of this help file, you will see an example.
Copy the example in the help file and then paste it into the Console. Remove
the number sign (#) and press <ENTER>. Your screen should look like what
is displayed in Figure 2.4 (see p. 22):
Figure 2.4 Using examples in ABplot() help file

You just created your first graph in SSD for R, and you now see in the
bottom of the Console the proper syntax for the ABplot() command.
Alternatively, you can access hyperlinks for all SSDforR functions by
clicking the hyperlink for SSDforR in the Packages tab in the lower right
pane of RStudio.

A Word About the Functions

Each of the SSDforR functions is a statistical test and/or graph that is
appropriate for single-subject data analysis. This book will teach you when
and how to use each function.
When you use help to look at the SSDforR functions, you will note that
each function is written in a similar format. Each begins with the function
name and is followed by a series of words, which we call “arguments,” in
parentheses. The way in which these functions are written will need to be
repeated in the Console in the exact format it is shown in the Help file.
Let’s begin by going over the meanings of the majority of the arguments
that are specified inside the parentheses in the functions:
 • behavior—This argument refers to the specific behavior variable that
you want to analyze.
• phasex—This argument refers to the phase variable associated with
the behavior variable you want to analyze.
• v—However it is written (v, v1, v2, etc.), this argument refers to the
value in the phase variable that you want to include in your analysis.
For example, if you want to look at baseline data, it is likely that your
value for v would be “A”. Every time you enter a function that
includes v, that value should be put in quotation marks.
• ABxlab, ABylab, ABmain—These arguments refer to how you want
your graphs labeled. ABxlab refers to the x-axis label; ABylab refers
to the y-axis label; ABmain refers to the main title of the graph. Every
time you enter a function that includes any one of these, those values
should be put in quotation marks.
• statx—This argument refers to the statistic you would like to obtain
for your data. Acceptable choices for this include “mean,” “median,”
and “sd.” Every time you enter a function that includes statx, that
value should be put in quotation marks.
• textx—This argument refers to any text that you would like to insert
on a graph. Every time you enter a function that includes textx, that
value should be put in quotation marks.

Note: Most, although not all, function names begin with a capital letter.
When you type the functions in the Console, you must type it EXACTLY as
it appears in the help file.

Example Data: A Case Study of Jenny

There are three different example case studies referred to throughout the
text: Jenny, Gloria, and Anthony. Jenny is a first-grade student at Woodside
Elementary School. Her teacher, because of Jenny’s disruptive behavior in
class, has referred her to the school social worker. The social worker, on
evaluation of Jenny, has learned that Jenny repeatedly yells at her teachers
and peers, cries in class, and interrupts the class by calling out of turn
frequently.
To get a better understanding of Jenny’s behavior, the school social
worker collects baseline data to aid in her evaluation by asking the
classroom teacher to tally Jenny’s troublesome behaviors. She uses Excel to
record information about how many times a day Jenny yells, cries, and calls
out in class. These data are shown in Figure 2.5 (see p. 24).

Figure 2.5 Excel spreadsheet with Jenny’s baseline data

The column labeled “yell” shows the number of times that Jenny yelled
during the school day. The label “pyell” is a phase variable associated with
the variable “yell.” Since there has been no intervention done at this point,
the “A” for each measurement indicates that these are baseline data. The
“cry” variable indicates how many times Jenny cried in school each day, and
“pcry” is the phase variable associated with “cry.” The “callout” indicates
how many times Jenny called out in class without raising her hand, while
“pcallout” is the phase variable associated with “callout.” Note that each
behavior variable has an associated phase variable, which is required in
order to analyze these data using SSD for R. While we have given each
phase variable a prefix of “p,” you could actually name these phase variables
anything meaningful to you.
 To understand this, we can see that on the first day of collecting data
Jenny yelled twice, called out of turn three times, and cried once. On the
second day, she yelled twice, cried twice, and called out three times.
The “notes” column is optional but allows the social worker to put in his
or her comments and note critical incidences that could impact any of the
data. Note that the social worker mentioned that Jenny’s mother was out of
town one day, and, on another day, Jenny got sick while she was in school
and needed to go home early. On the day Jenny went home sick, she cried 10
times, which seems to be a lot, even for her.
This Excel data file can be imported into RStudio so it can be analyzed
using SSD for R. In order to do that, however, remember that the spreadsheet
needs to be saved as a “CSV (Comma delimited)” file.
To begin your analysis, you need to load the SSD for R package. To do
this type the following in the Console:
>require(SSDforR)

To import Jenny.csv into RStudio, type the following in the Console:

>Getcsv()

This command is one of the functions in SSD for R. Once you hit
<ENTER>, you will be prompted to select a file. Select Jenny.csv and, after
opening the file, type the following in the Console:
>attach(ssd)

As displayed in Figure 2.6 (see p. 25), to view your file in the top left pane
in RStudio, follow these steps: Click Environment in the top right pane and
then in the top right pane click ssd in the Global Environment window.
Jenny’s data will appear in the top left pane.
Figure 2.6 Displaying ssd spreadsheet in Rstudio

To obtain a list of variables, type the following in the Console:

>listnames()

Your RStudio window should look like what is displayed in Figure 2.7 (see
p. 25).
Figure 2.7 Using listnames() to view variable names in alphabetical order

Note how all the variables in the Jenny.csv Excel spreadsheet are now
listed in the Console in alphabetical order.
Alternatively, you may want to view the variables you created in the order
in which you created them. To do that, enter the following command in the
Console:
>names(ssd)

Note how the order in which the variables appear in Figure 2.8 is the same as
in your Excel spreadsheet (see Figure 2.5).
Figure 2.8 Using names(ssd) to view variable names in creation order

If you want to see all the values for a given variable, simply type the name
of the variable into the prompt in the Console. For example, to see the values
for “yell,” simply type yell at the prompt, which is displayed in Figure 2.9
(see p. 27).
Figure 2.9 Displaying all values for the variable yell

Note that the values for “yell” are the same regardless of whether you
view them in R Studio or in Excel (see Figure 2.5).
Once you import your data file, you will be able to use the SSD for R
functions to analyze your data visually and statistically. As you continue
going through this book, you will learn more about Jenny’s presenting
problems by analyzing data at different points in her treatment process.
Gloria a 62-year-old widow described as having a major acute depressive
disorder that had its present onset after her husband’s sudden death from a
heart attack. Her symptoms are feeling sadness, tearfulness, lack of interest
in most daily activities, insomnia, and suicidal ideations.
To better understand Gloria’s degree of depression, her worker
administered a depression scale daily. Also measured was the number of
hours Gloria slept daily. These data are shown in Figure 2.10 (see p. 28). The
column labeled “depress” shows Gloria’s daily depression score. The
“pdepress” is a phase variable associated with the variable “depress.” Since
there has been no intervention done at this point, the “A” for each
measurement indicates that this is baseline data. The “sleep” variable
indicates how many hours Gloria slept each day, and “psleep” is the phase
variable associated with “sleep.”

Figure 2.10 Excel spreadsheet with Gloria’s baseline data

 Anthony, age 38, was referred to therapy by his employee assistance
program. Anthony reported that he was compulsively repeating some acts in
his daily life. He is occupied by thoughts of repeating actions that lead to the
repetition of some of his daily life acts. For example, he worries about
leaving appliances on or doors unlocked; he expresses discomfort about
things being out of order. To better understand Anthony’s behavior, his
worker recommended he keep a log of the number of times daily he checked
his appliances and rearranged items. These data are presented in Figure 2.11
(see p. 29). The column labeled “checking” indicates the number of times
Anthony exhibited checking behavior daily. “pchecking” is a phase variable
associated with the variable “checking.” Since there has been no intervention
done at this point, the “A” for each measurement indicates that these are
baseline data. The “arrange” variable indicates how many times Anthony
rearranged items each day, and “parrange” is the phase variable associated
with “sleep.”
Figure 2.11 Excel spreadsheet with Anthony’s baseline data

Summary of Functions
SSD for R supports the full range of analysis that can be conducted with
single-subject research data. How and under what conditions are each of
these functions used is discussed in detail in subsequent chapters, but a brief
overview of these functions is described here. You can also refer to
Appendix B, SSD for R Functions Quick Guide, to see all of these functions
grouped by usage.
• Basic graphing functions—These include a set of commands that will
allow you to make and label a variety of graphs, draw vertical lines
between phases, and note critical incidences. Graphs include, among
other things, simple line graphs (with or without mean or median
lines), standard deviation graphs, and graphs showing interquartile
ranges. These graphs can be exported in a number of different
formats.
• Functions associated with descriptive and basic statistics—These
include basic descriptive statistics, including mean, median, standard
deviations, and quartiles.
• Functions associated with statistical process control (SPC) charts—
These include X-R charts, R charts, p charts, X-mr charts, and c
charts. Again, all charts and graphs can be exported in a variety of
formats.
• Functions associated with autocorrelation and data transformations—
These are appropriate for calculating and dealing with issues of
autocorrelation, which are unique to time series research designs.
• Functions associated with effect sizes—These enable users to
quantify the degree of change observed between phases.
• Functions associated with hypothesis testing—These enable users to
test whether there are statistically significant differences between two
or more phases. Available tests include t-tests, chi-square, analysis of
variance (ANOVA), binomial, and conservative dual criteria.
• Functions appropriate for analyzing group data—These are suitable
for analyzing group data, including variation of individuals’
behaviors within a group.
• Functions appropriate for conducting meta-analysis—These enable
the user to conduct meta-analysis utilizing omnibus and random-
effect and fixed-effect models.
• Functions appropriate for RMarkdown—These are suitable for
producing MS Word and PDF documents for communication of
results produced by SSDforR.
• Functions suitable for community or system-wide data—These
functions are used to examine large amounts of data across phases
and are typically used in time series research that may go beyond the
scope of one client or client group.
Chapter Summary
In combination, R, RStudio, and SSD for R provide a robust way to analyze
single-system research data. This chapter showed you how to download the
necessary software and provided an overview of the visual and statistical
capability available with SSD for R. In the following chapters, you will learn
how to use these to analyze and interpret single-subject research data.

Chapter Exercises
Assignment 2.1—Opening and Displaying an SSD for R File
1. Load SSDforR and use the Getcsv() function to open the Jenny baseline data, attach them, and
create a line graph of the yelling behavior.
2. Display the spreadsheet in the top left console of RStudio by double clicking ssd in the global
environment in the top right panel. Take a screenshot of RStudio with the spreadsheet open.
3. You will need to submit a document (Word or Google Doc) with a copy of the graph and the
screenshot.
3
Analyzing Baseline Phase Data

Introduction
In this chapter you will learn about methodological issues in analyzing your
baseline. There may be times, however, when you may want to do additional
analysis and compare, for example, one intervention phase to another. In
these cases, you would also want to consider conducting the analyses we
introduce in this chapter to those phases as well.
This chapter is not intended to provide a comprehensive discussion of
how to design a baseline; there are a number of excellent texts listed in
Appendix D that provide in-depth discussion of single-subject research
designs. Rather, in this chapter we focus on assessing baseline data. The
baseline is the phase in single-subject research in which data are collected on
the target behavior prior to any intervention. As a result, the baseline serves
as the basis for comparison with information collected during any
intervention that might take place. This comparison allows you to determine
the target behavior is changing and how it is changing. Well done, the
baseline lets you know what would be expected to continue if there were no
intervention at all (Bloom et al., 2009).
There are two different types of baselines, concurrent and reconstructed.
In a concurrent baseline, data are collected while other assessment activities
are being conducted. A reconstructed baseline is an attempt to approximate
the naturally occurring behavior based on the memories or documentation of
the client or others; case records can also be utilized to produce a
reconstructed baseline. The reconstructed baseline can be utilized when a
concurrent one is not possible. In some cases, a combination of
reconstructed and concurrent data is used to form the baseline.
Analyzing Your Baseline
In general, analyzing baseline data is simply a matter of describing it.
Traditionally, single-subject research uses visual analysis to accomplish this,
and descriptive statistics, such as the mean, standard deviation, median, and
quantiles, are used to provide more in-depth information. SSD for R can be
used to accomplish this.
The other issue that is relevant in the analysis of baseline data is that of
autocorrelation, also known as serial dependency. Data in any phase that
have high levels of autocorrelation may need to be treated differently from
data that do not have this problem when comparing phases as most analytical
methods are based on the notion that observations are independent of one
another. Therefore, it is important to assess the degree to which phase data
are autocorrelated in order to better understand how comparisons to other
phases can be made at other times, and this issue is discussed more at the
end of the chapter.

Visual Analysis
Determining the stability of a baseline is an important first step in the
analysis of single-system data. A stable baseline is one in which there are no
wide fluctuations or obvious cycles. Stable baselines provide the ability to
estimate what might happen to the behavior in the future if no intervention
were provided. A lot of variability in the data makes it difficult to determine
how the behavior is affecting the client system and can lead to incorrect
conclusions about the impact of the intervention (Matyas & Greenwood,
1990).
Ideally, the baseline should continue until stability is achieved; however,
there are situations in which this is not possible or practical (Bloom et al.,
2009; Logan, Hickman, Harris, & Heriza, 2008; Portney & Watkins, 2008).
In some cases, it may be necessary to intervene with the client before
stability in the baseline occurs because of pressing needs of the client, and it
may be possible to get a good understanding of the behavior by combining
concurrent and retrospective baseline data. This issue is often the case when
single-subject research is part of practice evaluation. If it is not possible to
continue collecting baseline data until that phase is stable, it might be
feasible to select another target behavior to assess.
 Illustrations of baselines with varying degrees of stability follow next.
Figure 3.1 is an illustration of a stable baseline.

Figure 3.1 Example of a stable baseline

Note that there is only a slight upward or downward direction in the data
as all baseline values range between 3 and 4. This is a desirable pattern
because there is a consistent occurrence of the behavior over time. Changes
in the behavior during the intervention phase would be readily apparent.
Figure 3.2 provides an example of an increasing and stable baseline.
Figure 3.2 Example of an increasing stable baseline

In this example, the baseline is stable because we can use the obvious
trend to predict future values; however, the outcome being measured is
consistently increasing. If the goal were to decrease or stabilize the behavior,
a reversing or flattening trend in the intervention phase would indicate that.
If, however, the goal was for the behavior to continue increasing, a
successful intervention may appear to be a change in the rate of the increase.
Figure 3.3 shows an example of a semistable baseline.

Figure 3.3 Example of a semi-stable baseline

 Although a future trend can be predicted, the wide differences between
data points make it difficult to draw any conclusions with certainty. Visually
comparing this semi-stable baseline to future intervention data is difficult.
Finally, Figure 3.4 (see p. 34) illustrates an unstable baseline.

Figure 3.4 Example of an unstable baseline

This is an extremely variable baseline with no ability to predict the future.

Ideally, the best way to deal with this type of baseline is to continue to
collect data until a pattern stabilizes. Similar to the semistable baseline of
Figure 3.3, visually comparing this unstable baseline to intervention data
may be difficult.

Visual Examination of Your Baseline Data: An Example

To provide a more in-depth look at analyzing baseline data, we now look at a
comprehensive example that examines multiple indicators of a client
problem. To begin, use Getcsv() to open the jenny.csv data file, which
contains only baseline data. After the file is open, type the following in the
Console:
>attach(ssd)

This dataset contains information about Jenny’s externalizing behaviors in

school, including yelling, crying, and calling out of turn as described in the
previous chapter. We use these variables to assess the extent of her
classroom behavior problems.
We can begin exploring Jenny’s baseline yelling behavior by creating a
box plot and displaying descriptive statistics, shown in Figures 3.5 (see p.
35) and 3.6 (see p. 36), by entering the following in the Console:
>ABdescrip(yell, pyell)

In the Plots pane, you will find the box plot for these data. In the Console,
find a host of descriptive statistics, including the sample size (n), values for
the mean, the 10% trimmed mean, median, sd, range, interquartile range
(iqr), and quantiles. The box plot in Figure 3.5 shows data with considerable
variation with a minimum value of one and a maximum of five.

Figure 3.5 Boxplot of Jenny’s yelling during baseline phase

Figure 3.6 Console output of descriptives for Jenny’s yelling during baseline phase

We can also illustrate this with a simple line graph displaying the number
of yelling incidents using the following command, as shown in Figure 3.7
(see p. 37):
>ABplot(yell, pyell,“time”,“amount of yelling”, “Jenny’s yelling”)

Figure 3.7 Line graph of Jenny’s yelling during baseline phases

Note: If the error “figure margins too large”, occurs, you will have to
increase the size of the plot pane. If this does not solve the issue, reset the
plot pane by issuing the following R command in the Console: dev.off(). Be
mindful that any graphs currently in the pane will be removed.
To begin our visual analysis, it is important to determine if the behavior
displayed in the graph is typical (Bloom et al., 2009). One way to do this is
to determine how the events vary around a measure of central tendency, such
as the mean or median.
Use the following set of commands, entered into the Console one at a
time, to enhance the simple line graph you created in Figure 3.7 to
accomplish this, which is displayed in Figure 3.8 (see p. 38):
>ABstat(yell,pyell,“A”,“mean”)
>ABtext(“mean”)

Figure 3.8 Annotated line graph of Jenny’s yelling during baseline phases

Note that when you enter the ABstat() and ABtext() functions, you will
receive an additional prompt in the Console instructing you to place the
mouse where you want your lines and text to appear. For the ABstat()
command, place the cursor at the beginning left of the phase where the line
should appear to ensure that the statistic line covers all the phase data. Since
we only have baseline data at this point, hover your cursor in the graph
above the 0 on the x-axis. For the ABtext() function, center the mouse over
the area where you want the text. After the text or statistic line appears, if
you are satisfied with its appearance on the graph, simply type “y” and
<ENTER>. If you are not satisfied, type “n” and <ENTER>. In this case,
you will have to reenter the command to continue annotating your graph.
You can easily do this by using your up and down arrow keys to re-create
previous commands.
The addition of a measure of central tendency provides valuable
information. We can now see that the points in the graph vary widely from
the mean and median, and the baseline looks pretty unstable. The sixth point
is the furthest below the mean and median while points 10, 12, and 13 are
the furthest above.
Another way to enhance your visual analysis is to examine how the data
fall around standard deviation bands, which display the mean along with
lines depicting the bounds of either one or two standard deviations around
the mean.
Approximately two thirds of all data in a normal distribution fall between
one standard deviation (sd) above and one standard deviation below the
mean. Bloom et al. (2009) suggested that, depending on the nature of the
behavior, data falling outside of one standard deviation can be defined as
desirable or undesirable, depending on the behavior being measured. This
can be illustrated, as shown in Figure 3.9 (see p. 38), using Jenny’s yelling
behavior with the following command:
>sd1bandgraph(yell,pyell,“A”,“time”,“amount of yelling”, “Jenny’s yelling (1-SD)”)

The actual values for the means and standard deviations are displayed in the
Console:
“SD=” “1.26”
“+1sd=” “4.06”
“mean=” “2.8”
“-1SD=” “1.54”

and we can add this to the band graph by adding a key to your graph in an
unobtrusive place:
>ABtext(“KEY: mean=2.8, +1 SD=4.06, -1 SD=1.54”)
Figure 3.9 1-sd band graph of Jenny’s yelling during baseline phase

Notice that 11 of the 15 data points in the baseline are within one standard
deviation, which is 73% of the data. As this is above two thirds of the data
expected in a normal distribution, we can assume that variability of this
baseline is not excessively high.
Using two standard deviations would be considered even more rigorous
because approximately 95% of all scores in a normal distribution are within
plus or minus two standard deviations of the mean. This is illustrated in
Figure 3.10, which is annotated with the values of each band and can be
demonstrated using the following function:
Figure 3.10 2-sd band graph of Jenny’s yelling during baseline phase

>sd2bandgraph(yell, pyell,“A”,“time”,“amount of yelling”, “Jenny’s yelling (2-SD)”)

As the graph displays, none of the points are above or below two standard
deviations. Therefore, if we were using plus or minus two standard
deviations as the criterion for typical behavior, none of the values would be
considered unusual (i.e., desirable or undesirable). The actual values for the
mean and standard deviations are displayed in the Console and were used in
the annotation.
“SD=” “1.26”
“+2sd=” “5.33”
“mean=” “2.8”
“–2SD=” “0.27”

Trending
As we continue our analysis of the baseline, it is important to assess if the
data have a significant trend and to note the direction of the trend (i.e.,
whether it is increasing or decreasing). The trend can also be completely flat
or irregular, as discussed previously. The goal of an intervention may be to
reverse the trend in a negative behavior or increase it for positive behavior. It
is important to detect a trend early on as a strong trend will impact the type
of data analysis we can do in comparing phases.
As an example, let’s examine Jenny’s yelling behavior to see if we can
detect a trend. Invoke the following command in the Console to display what
is illustrated in Figures 3.11 and 3.12 (see p. 41):

Figure 3.11 Console output of assessment for trending of Jenny’s yelling during baseline phase
Figure 3.12 Visual assessment for trending of Jenny’s yelling during baseline phase

>Aregres(yell, pyell,“A”)

The graph displays a slight trend in the data, with the dotted regression line
displaying the slope; however, note that the individual data points are not
clustered tightly around the regression line. In addition to the graph,
statistical output is produced in the Console.
These values show that the degree of change can be quantified by the
estimate for x1 of 0.13214, which is the slope of the regression line. This can
be interpreted as follows: For each unit increase in time, there is an
estimated 0.13214 unit increase in the yelling behavior. The column labeled
“t value” is the calculated value for the statistical significance of the slope
and constant. The last column labeled “Pr(>|t|)” is the probability of making
a Type I error (i.e., testing the hypothesis that the coefficient for the slope is
greater than zero). Because the probability is greater than the commonly
accepted 0.05 threshold (p = .0791), the slope is not considered statistically
significant. Despite an insignificant p value, however, we may still want to
consider the fit of the data around the regression line because significance
may be hard to achieve, particularly with small sample sizes. If this is the
case, we might want to look at the value for Adjusted R-squared, which
explains the proportion of the variance in the outcome, in this case, Jenny’s
yelling, explained by the predictor variable time. In this example, we see the
adjusted R-squared is 0.1581, indicating that approximately 16% of yelling
is predicted by time.
A visual inspection of the graph suggests that the data are not linear and
outliers at points 10, 12, and 13 have an impact on the slope by pulling it
upward.

Autocorrelation
All statistical tests rely on various assumptions. When these assumptions are
not met, incorrect conclusions about calculated values can be made, such as
deciding that observed differences between phases are meaningful when
they, in fact, are not (i.e., Type I error). On the other hand, we may
erroneously not detect a difference between phases when those differences
do exist (i.e., Type II error).
An assumption in the analysis of data in many types of analyses is that
each observation is independent of the others. This means that observations
cannot be predicted from one another. This is often the case when many
research subjects are included in a study, and each individual is observed
separately; however, this is less often the case when subjects may have close
relationships to one another or we measure individuals repeatedly, as is the
case in single-system research designs.
When data lack independence, they are considered correlated, and a
special consideration in single-system research is serial dependency (Bloom
et al., 2009). Serial dependency can be measured by testing for a correlation
known as autocorrelation (Gast & Ledford, 2010). When visual analysis is
used, high levels of autocorrelation combined with variability in the baseline
increase the likelihood of a Type I error (Matyas & Greenwood, 1990).
Measuring autocorrelation using the rf2 (Huitema & McKean, 1994) is
preferable when the number of observations are small, usually considered
less than six (Bloom et al., 2009). The following command entered into the
Console is needed to test for lag-1 autocorrelation for Jenny’s yelling
behavior:
>ABrf2(yell,pyell,“A”)

A lag-1 autocorrelation is the degree of dependence of an observation on

the previous one; a lag-2 autocorrelation is the degree of dependence of an
observation on the observation that occurred two time periods prior, and so
on. While there can be an autocorrelation at any lag, the lag-1
autocorrelation is considered to be the most crucial to assess (Bloom et al.,
2009).
The results of the test are shown in Figures 3.13 (see p. 43) and 3.14 (see
p. 43).

Figure 3.13 Console output of assessment for autocorrelation of Jenny’s yelling during baseline phase
Figure 3.14 Visual assessment for auotocorrelation of Jenny’s yelling during baseline phase

The rf2 is a measure of the degree of autocorrelation. The magnitude of

autocorrelation is what is of interest here, regardless of whether the
calculated value is positive or negative. The output labeled sig of rf2 is the
chances of making a Type I error testing the hypothesis that the observed
autocorrelation is significantly different from zero. Because the sig of rf2 is
above 0.05 in our example with a value of 0.432, what autocorrelation we do
observe is not statistically significant. As a result, we can conclude that
whatever autocorrelation exists may not be problematic, and we can assume
independence of observations.
There is, however, an important caveat: Because of the small number of
observations in most single-system research studies, a large autocorrelation
is necessary to obtain statistical significance. It has been suggested that if the
number of cases is small, you should assume the data are autocorrelated
regardless of whether the rf2 is significant or not (Bloom et al., 2009). In
these cases, an observed rf2 greater than 0.20 is considered undesirable and
independence of observations should not be assumed (Bloom et al., 2009).
When assessing for autocorrelation, testing should be conducted
separately for each phase. If data in any phase are autocorrelated or the
number of observations is small, care should be taken in making visual,
descriptive, and statistical comparisons between phases.
 If you cannot assume independence of observations in any, you could
attempt to transform the data in all phases in order to reduce the effect of
autocorrelation. If the data have a high degree of variation, using a moving
average could smooth it. The moving average is simply the mean of two
adjacent observations: ([point 1 + point 2]/2). The following function
entered into the Console will perform the transformation:
>ABma(yell, pyell,“A”)

You will note in Figure 3.15 that the graph does display a decrease in the
degree of variation for the transformed data, which could indicate a
reduction in the degree of autocorrelation.

Figure 3.15 Moving average plot for Jenny’s yelling during baseline phase

When you use the ABma() function, you have an option to save the
transformed data for future analysis. Once you save this, you will want to
test it again for autocorrelation. If it appears as if transforming the data has
reduced variation, you will want to save use these data, and not your
original, for further analysis.
If the data are not independent and have a significant trend, transformation
using differencing is recommended. Differencing simply calculates the
difference of the value of one data point to the one immediately preceding it.
As an example, begin by examining Jenny’s calling out behavior at school,
shown in Figures 3.16 (see p. 45) and 3.17 (see p. 45), using the following
function:
>Aregres(callout, pcallout, “A”)

Figure 3.16 Console output for trending of Jenny’s calling out during baseline phase
Figure 3.17 Visual assessment for trending of Jenny’s calling out during baseline phase

As both the graph and output in the Console indicate, there is a strong
trend in the data. Next we test for autocorrelation of the calling out behavior
using the following function, the results of which are displayed in Figures
3.18 (see p. 46) and 3.19 (see p. 46):
>ABrf2(callout, pcallout,“A”)
Figure 3.18 Console output of assessment for autocorrelation of Jenny’s calling out during baseline
phase
Figure 3.19 Visual assessment for autocorrelation of Jenny’s calling out during baseline phase

The data are autocorrelated with rf2 = 1.497 and a significance of rf2 of
less than 0.05. To try to reduce the effects of this autocorrelation, these data
can be transformed using differencing with the following command (a graph
of this function is displayed in Figure 3.20) (see p. 47):
>diffchart(callout, pcallout,“A”)
Figure 3.20 Differencing plot for Jenny’s calling out during baseline phase

Since differencing seems to smooth out the line, we will save the
transformed data by typing “y” when prompted.
Regardless of how you transform your data (i.e., either through the
moving average or through differencing), after you have saved the
transformed data, use the Getcsv() and attach(ssd) commands to open and
attach the newly created dataset. Once this is accomplished, you will need to
test the transformed data for autocorrelation using the following command
(results are shown in Figures 3.21 and 3.22) (see p. 48):
>ABrf2(diff,phase,“A”)
Figure 3.21 Console output of assessment for autocorrelation of differenced calling out during
baseline phase
Figure 3.22 Visual assessment for autocorrelation of differenced calling out during baseline phase

In the case of Jenny’s calling out behavior, the transformation has reduced
the rf2, and it is no longer statistically significant. It should be noted that
transformed data should always be tested for autocorrelation, and all other
phases will need to be transformed in order to compare them.
There are situations in which transforming data does not lower levels of
serial dependency. In these cases, checking for autocorrelation will yield
high values for rf2 and/or significance values less than 0.05. When this
happens, it will likely be best to continue working with the original,
untransformed data as the comparison between the phases will be easier to
interpret than having to account for transformed, highly autocorrelated data.
You can calculate the autocorrelation for any lag using the ABautoacf()
function. Because of small sample size, we strongly recommend the use of
the rf2 method to calculate the lag-1 autocorrelation. Load and attach Jenny’s
baseline data for an example of how to produce the autocorrelation for a lag-
5 autocorrelation for her yelling behavior. Output for this is shown in
Figures 3.23 (see p. 49) and 3.24 (see p. 49):
>ABautoacf(yell, pyell, “A”, 5)
When entering this command, note that the integer “5” is the lag number
requested. You can see from the correlogram in Figure 3.24 that none of the
autocorrelations in lags 1 through 5 are above the upper blue line or below
the lower blue line, indicating nonsignificance. The Box-Ljung test statistic
is 2.9987, and the p value is above .05, further indicating that the
autocorrelations for the phase are not statistically significant.

Figure 3.23 Console output for assessment of lag-5 autocorrelation in Jenny’s yelling during baseline
Figure 3.24 Visual assessment of lag-5 autocorrelation in Jenny’s yelling during baseline

Another Example
Now, let’s consider Anthony’s baseline checking behavior. If you recall,
Anthony is a client who appears to have obsessive-compulsive disorder, and
he has been asked to count the number of times he checks that his appliances
are turned off or unplugged each day. At his next appointment, he reports his
checking behavior for the previous week. To work with these data, use the
Getcsv() and attach(ssd) functions to load Anthony.csv. Then, describe it as
follows:
>ABdescrip(checking, pchecking)

The output displayed in the Console and illustrated in Figure 3.25 indicates
that the mean for Anthony’s checking is four times per day (sd = 0.894),
with the behavior ranging from three to five times per day. The box plot,
displayed in Figure 3.26 (see p. 51), shows the narrow range of this
behavior, with no outliers.
Figure 3.25 Console output of descriptives for Anthony’s checking during baseline phase

Figure 3.26 Boxplot of Anthony’s checking during baseline phase

With this in mind, you can create a line plot by invoking the following
command:
>ABplot(checking, pchecking, “day”, “frequency”, “Anthony Checking”)

Since there were no obvious outliers in the box plot, you might choose to
annotate this line graph with a labelled mean line, being sure to accept your
annotations:
>ABstat(checking,pchecking,“A”,“mean”)
>ABtext(“mean=4”)

This annotated graph is displayed in Figure 3.27.

Figure 3.27 Annotated line graph of Anthony’s checking during baseline phases

From this initial analysis, it looks like the baseline is fairly stable, but we
can check this out with a one standard deviation band graph, which can be
annotated to display the values for the mean and the standard deviation
bands:
>sd1bandgraph(checking, pchecking, “A”, “day”, “frequency”, “Anthony Checking (1-
SD)”)
>ABtext(“KEY: mean=4, +1 SD=4.89, -1 SD=3.11”)

This annotated graph is displayed in Figure 3.28.

Figure 3.28 1-sd band graph of Anthony’s checking during baseline phase

Now we see that the majority of the data do not fall within one standard
deviation since only two of six data points are within the bands. By
expanding the graph to two standard deviations, it is hoped we will find
stability:
>sd2bandgraph(checking, pchecking, “A”, “day”, “frequency”, “Anthony Checking (2-
SD)”)
>ABtext(“KEY: mean=4, +2 SD=5.79, -1 SD=2.21”)

As illustrated in Figure 3.29 (see p. 53), all baseline data fall within two
standard deviations, so we have less concern about instability in this phase.
Figure 3.29 2-sd band graph of Anthony’s checking during baseline phase

Before comparing Anthony’s baseline checking to future phases, we will

want to ascertain the level of trending and serial dependency in the baseline.
We can check for trending with the following command:
>Aregres(checking, pchecking, “A”)

The output in the Console, shown in Figure 3.30 (see p. 53), indicates that
whatever trend is observed, and there does not appear to be one from looking
at any of the line graphs we produced so far, is not statistically significant
since the significance level for the regression line is 0.2103.
Figure 3.30 Console output for assessment of trending in Anthony’s checking during baseline

Now we can see if there is problematic serial dependency by entering the

following command:
>ABrf2(checking, pchecking, “A”)

The output displayed in the Console and in Figure 3.31 shows a

nonsignificant level of autocorrelation with a significance value of 0.745 and
a relatively low level of observed autocorrelation with rf2.
Figure 3.31 Console output for assessment of autocorrelation in Anthony’s checking during baseline

With no significant trend and low levels of serial dependency, we do not

have to make any special considerations when comparing these baseline data
to future phases at this point.

Chapter Summary
Early analysis of baseline data in single-subject research entails decision-
making that informs the duration of this phase as well as how other phases
may be compared to the baseline in the future. For example, visual analysis
of the baseline early on may show an unstable or semistable trend that may
indicate the need to collect additional baseline data in the hopes of creating
stability. This analysis is critical for ensuring the integrity of the baseline for
future analysis.
SSD for R enables users to gain a thorough understanding of baseline data
through both visual and statistical inspection. The ability to visualize and
statistically understand how data are clustered around the mean or median of
the baseline (or any other phase for that matter) will give you a greater sense
of the variability of the data and, in a practice evaluation setting, may guide
you in setting concrete goals with clients.
In all cases, however, trending and autocorrelation of phase data need to
be considered since disregard for these issues could lead to erroneous
comparison of phases. This chapter provided guidance on assessment of
trending and autocorrelation in small samples, which are typical of single-
subject research. Additionally, SSD for R has the capability to transform and
save phase data for which serial dependency is an issue.

Chapter Exercises
Assignment 3.1—Annotating a Line Graph
For this assignment you will continue assessing Brenda’s baseline data using
the file you created in Chapter 1.
1. Obtain descriptive statistics for Brenda’s oppositional behaviors. In SENTENCE form, provide
information about the number of observations there are in the baseline, the average number of
oppositional episodes per day, and the standard deviation.
2. Create a line graph (like you did in the previous module) AND annotate it with a line denoting
the mean and text, appropriately placed, to display information about the mean. Be sure to label
your graph with your name in the main label.

Assignment 3.2—Baseline Analysis: Part 1

For this assignment, you will use Brenda’s baseline data. Be sure to include
your initials in your main labels for your band graphs. As well, be sure to
answer the questions in your narrative explanation of your findings.
1. Test for baseline stability/outliers using the sd1bandgraph() and sd2bandgraph() functions.
In your narrative, include the plots for each of these along with your interpretation. What
percentage of data falls within each band graph? Based on this, do you think you have
sufficient baseline data or would you attempt to collect more if you were able?
2. Test for a trend in your baseline data by using the Aregres() function. What is the slope of the
trend line, and what does this mean? Is the trend significant? Would you need to consider
baseline data trending in your comparison of the baseline to the intervention phase?
Assignment 3.3—Baseline Analysis: Part 2
For this assignment, you will use Brenda’s baseline data ONLY.
1. Test for baseline autocorrelation using the ABrf2() functions. In your narrative, report the
MAGNITUDE of autocorrelation (the rf2) and whether the observed autocorrelation is
STATISTICALLY SIGNIFICANT (sig of rf2). Add to your narrative whether you think the
observed autocorrelation would be problematic when eventually comparing your baseline to the
intervention.
2. IF you determine autocorrelation to be problematic, what function(s) might you use to
transform your data? Support your decision.
4
Comparing Baseline and Intervention Phases
Visualizing Your Findings and Descriptive Statistics

Introduction
In this chapter you will learn about methodological issues to consider in
analyzing the success of your intervention and how to conduct your visual
analysis. We begin with a discussion of descriptive statistics that can aid in
the visual analysis of your findings by summarizing patterns of data across
phases. Four common forms of descriptive statistics are explained: central
tendency, variation, trends, and effect size. We then continue to a discussion
of testing for autocorrelation in the intervention phase. We conclude with a
discussion of the goal line, which provides a visual method to quantify
effectiveness of an intervention when a specific decision is required to attain
a specific goal.

Descriptive Statistics
A good way to begin comparing phases is by producing descriptive statistics
for central tendency and variation for all phases. As an example, let’s use the
Jennyab dataset by invoking the Getcsv() function. These data are a
continuation of our study of Jenny, but now includes both baseline and
intervention data.
Once the dataset is open, type attach(ssd) in the Console to make the
dataset available for analysis. Once the dataset is attached, you can produce
descriptive statistics and view a box plot for both phases of Jenny’s crying
behavior by entering the following command in the Console:
>ABdescrip(cry,pcry)
The output, shown in Figures 4.1 (see p. 58) and 4.2 (see p. 58), includes
descriptive statistics, and a box plot is displayed:

Figure 4.1 Boxplot comparing crying behavior from baseline to intervention

Figure 4.2 Descriptive statistics for crying behavior for all phases

The statistics displayed in Figure 4.2 are explained as you review this
chapter, but we begin with a discussion of the median. The median (Md) is
the value for which 50% of the observations fall both above and below; it is
the middle value if all the data were listed in increasing value. The median
can be used to express the typical value in a given phase. In the Jenny
example, the median number of daily episodes of crying was five during the
baseline (the “A” phase) compared to one during the intervention (the “B”
phase). The box plot in Figure 4.1 graphically displays the median, the
range, and the interquartile range for both phases. The thick black line in the
boxes represents the median score for the phase. The median is visibly lower
in the intervention phase, which is desirable in this case. The box plot also
shows that there is considerably more variation in the baseline than in the
intervention as noted by the upper and lower bounds of the box plots and the
sizes of the actual boxes.
Now that we have a general sense of how the data are distributed in each
phase, we can get additional information by examining a simple line graph.
Consider Figure 4.3, which was produced in part by entering the following
in the Console:

Figure 4.3 Simple line graph of Jenny’s crying behavior

>ABplot(cry, pcry,“Days”,“Crying Episodes”, “Jenny’s Crying”)

We can add a vertical line to separate the phases by entering the following in
the Console:
>ABlines(cry)

After you type this command in the Console, click the mouse in the gap
between the phases. You will be queried with a “Y/N” request to accept the
line. If you are satisfied with the placement of the line, enter “Y.” If you
enter “N,” you will then need to reenter ABlines() to place the line in the
correct position.
Looking at this line graph, we note that there is an extreme value in the
baseline, and that there is an associated note with that data point. In order to
make this clear, we can make a notation on the graph. To do this we labeled
the extreme score by entering the following command in the Console:
>ABtext(“Goes Home Sick”)

This is helpful in explaining why the observation is not typical. Entering the
following in the Console produces an arrow extending from the outlying
data point to your text:
>ABarrow()

Similar to drawing a line on the graph, you will be prompted to place the
ends of the arrow in the desired locations.
Figure 4.4 illustrates that adding lines that represent the median in each of
the phases can enhance the graph, especially if there are extreme, or atypical,
values. To add a median line to the baseline phase, enter the following
command in the Console:

Figure 4.4 Jenny’s crying behavior with medians displayed

 >ABstat(cry, pcry,“A”,“median”)

Now click the mouse at zero on the x-axis. To add a median line to the
intervention phase, enter the following in the Console:
>ABstat(cry,pcry,“B”,“median”)

The lines were labeled using the ABtext() function and clicking the mouse
in the appropriate location. The addition of the median lines visually
demonstrates that the typical amount of crying decreased after the
intervention. The median is a good choice to use instead of the mean when
there are observations that are very different from the typical value, or
“outliers.” The last observation in the baseline of 10 episodes could be
considered an outlier.
The mean is another measure of central tendency that is helpful in
describing the typical observation. Figure 4.5 displays a simple line graph,
but this time, the means for each phase of Jenny’s crying behavior are
displayed. Unlike the median, the mean can be strongly impacted by outliers.
This is especially true when there are a small number of observations, which
is common in single-subject research.

Figure 4.5 Jenny’s crying behavior with means displayed

 To produce this graph, a simple line graph was first constructed using the
ABplot() function, described previously, then the mean line was added to the
baseline by entering the following command in the Console:
>ABstat(cry, pcry,“A”,“mean”)

A similar command, ABstat(cry, pcry,“B”,“mean”) was entered to produce

the mean line in the intervention phase.
The trimmed mean (tM) can also be used when there are outliers because
this measure of central tendency can dilute their impact (Bloom et al., 2009;
Verzani, 2004). The trimmed mean removes a percentage of the highest and
lowest values when calculating the mean. A common practice is to exclude
10% of the lowest and highest values. Figure 4.6 (see p. 62) provides an
example. The trimmed mean can be added to the baseline and intervention
by entering the following commands in the Console:

Figure 4.6 Jenny’s crying behavior with means and trimmed means displayed

>Trimline(cry,pcry,“A”)
>Trimline(cry,pcry,“B”)

The text was added to baseline and intervention with the following
commands:
 >ABtext(“tM=4.45”) and ABtext(“tM=1.44”)

As Figure 4.6 illustrates, there is very little difference between the mean and
the trimmed mean in this example. The trimmed mean is generally preferred
over the median as a measure of central tendency because it uses more
information about the data than the median.

Measures of Variation
Whenever a measure of central tendency (e.g., mean or median) is reported
on a set of data, it is necessary to describe the degree to which scores deviate
from it. Variation can be defined as the degree to which scores fluctuate
around a measure of central tendency. As a result, measures of variation are
necessary to fully describe a set of data. Two measures of variation that are
commonly used in single-system data analysis are the range and standard
deviation (SD) (Bloom et al., 2009).

Range
The range is the simplest measure of variation. It is simply the difference
between the highest and lowest values in a set of observations. The range in
the baseline of Jenny’s crying behavior is nine, and it is five for the
intervention. A more useful form of this measure is the interquartile range
(IQR); the IQR is the difference between the third (75th percentile) and first
(25th percentile) quartiles. These percentiles are displayed in the box plot in
Figure 4.1, with the 25th percentile represented by the bottom of each box
and the 75th percentile represented by the top of each box. For the Jenny
data, the IQR is 3 for both the baseline (6 minus 3) and intervention (3 minus
0) phases. The IQR can be graphed using the following set of functions,
which will produce Figure 4.7:
Figure 4.7 Jenny’s crying with IQR ranges and medians in baseline and intervention phases

This figure is helpful in comparing the typical amount of crying in the

baseline compared to the intervention. Figure 4.7 shows that typical values
were between 3 and 6 in the baseline and between 0 and 3 during the
intervention. This is another way to view data displayed in the box plot that
is the output of the ABdescrip() function.
Command Description
>ABplot(cry,pcry,“Days”,“Crying Episodes”, “Jenny’s Produces basic AB graph
Crying”)
>ABlines(cry) Draws line between phases
>IQRline(cry,pcry,“A”) Adds IQR lines to baseline
>IQRline(cry,pcry,“B”) Adds IQR lines to intervention
>ABstat(cry,pcry,“A”,“median”) Adds baseline median line
>ABstat(cry,pcry,“B”,“median”) Adds intervention median line
>ABtext(“Goes Home Sick”) Adds text
>ABarrow() Adds arrow from point 10 to
text
>ABtext(“Median = 5”) Adds text
>ABtext(“IQR= 3 to 6”) Adds text
>ABtext(“Median = 1”) Adds text
>ABtext(“IQR= 0 to 3”) Adds text

Standard Deviation
The SD is the most widely used measure of variation and is frequently used
in conjunction with the mean. Like the mean, it takes into account the impact
of all observations, thus using all available information. The SD is the
average amount of differences of the scores from the mean. In a normal
distribution, the SD provides the percentage of scores above and below the
mean. In fact, if the scores are normally distributed, 68.2% of them will fall
between 1 SD above and 1 SD below the mean, and 95% will fall between 2
SDs above and below the mean. “Typical” behavior can be defined as values
between ±1 SD in the baseline.
Lines representing 1 or 2 SDs above and below the mean can be displayed
on a graph, which provides a view of the variability between phases. Figure
4.8 displays a band graph displaying ±1 SD for the baseline going through
both the baseline and intervention. The following commands will produce
the graph, and output in the Console will provide values for the SD bands
and the mean:

Figure 4.8 Jenny’s crying with 1 SD band based on baseline

Command Description
>SD1(cry,pcry,“A”,“Days”,“Crying Produces basic graph with ±1 SD in baseline
Episodes”,“Jenny’s Crying”) extending across phases
>ABlines(cry) Draws line between phases
>ABstat(cry,pcry,“B”,“mean”) Draws mean line in intervention
>ABtext(“Baseline Mean = 4.6”) Adds text
>ABtext(“Intervention mean = 1.6”) Adds text
>ABtext(“+1SD 7.08”) Adds text
>ABtext(“-1SD 2.15”) Adds text

Looking at the graph in Figure 4.8, we see that a larger percentage of the
scores are below 1 SD during the intervention as compared to the baseline.
Since decreasing values are desirable in this example, we can assume that
the intervention is making a positive difference for Jenny’s crying.

How to Measure Trends

Single-subject data analysis is impacted by trends in the data. Each phase
should be assessed for its presence prior to proceeding with additional
inquiry (Kratochwill et al., 2013). When observations in any phase are
trending significantly, either decreasing or increasing, the ability for a
measure of central tendency to accurately represent the typical response is
diminished as the direction in which the data are trending becomes more
important. Therefore, when a trend is present, it is strongly recommended
that the mean or median not be used (Bloom et al., 2009). Thus, it is a good
idea to compute a trend line to help select the proper type of analysis to
conduct with your data.
A trend can be defined by the slope of the best-fitting line within a phase
(Kratochwill et al., 2010), also known as the ordinary least squares
regression. The slope is a measure of the steepness of a line; a positive or
upward slope indicates an increasing trend, while a negative or downward
slope indicates a decreasing trend. With single-subject data, the slope is the
average degree of change in the target behavior between time points.
Figures 4.9 and 4.10 (see p. 66) provide an example of output from
entering the following command into the Console:
Figure 4.9 Assessing trend in baseline and intervention for crying behavior
Figure 4.10 ABregres() statistical output for crying behavior

>ABregres(cry, pcry,“A”,“B”)

Note that a separate graph is produced for the two phases. The ABregres()
function has the capability to compare any two phases of a design. If, for
example, there was a B1 phase, it could be compared to the baseline with the
following command:
>ABregres(cry,pcry,“A”,“B1”)

Figure 4.9 displays an increasing trend in the baseline (i.e., the number of
crying episodes is generally increasing from one time point to the next)
while decreasing during the intervention phase (i.e., the amount of crying is
generally decreasing from one time point to the next).
In the Console to the left of your graphs, you will find some important
measures that quantify what is presented graphically. Here we see that the
slope is positive (0.4121) compared to the intervention, which is negative (–
0.1790). A decreasing trend in the number of episodes is desired during the
intervention in this case as we would hope to see a reduction in the number
of crying episodes. In this example, the trend is a modest one. The multiple
R-squared values are 0.4229 for the baseline and 0.4783 for the intervention.
The closer this value is to 1, the greater the trend and the closer to 0 is the
weaker the trend. Additionally, the p values for the slopes of these are both
statistically significant.
Now consider Figures 4.11 and 4.12 (see p. 68), which examine the trend
of Jenny’s calling out behavior in both the baseline and intervention phases.
To generate these figures, the following command was entered in the
Console:
>ABregres(callout,pcallout,“A”,“B”)

In this example, the R squared in the baseline is 0.9685 for the baseline and
0.9388 in the intervention. In this instance, the trend line almost perfectly
fits the observations. This would be considered a strong trend with slopes of
0.90476 and –0.62238 for the baseline and intervention phases respectively.
Figure 4.11 Assessing trend in baseline and intervention for calling out behavior
Figure 4.12 ABregres() statistical output for calling out behavior

Now let’s consider Figures 4.13 (see p. 68) and 4.14 (see p. 69), where we
examine the trend for Jenny’s yelling behavior by entering the following
command in the Console:
>ABregres(yell,pyell,“A”,“B”)

In this example, we see that the baseline has a weak trend, and the
intervention has almost no trend, with R-squared values of 0.2183 in the
baseline and 0.03175 in the intervention. The slopes are 0.13214 and –
0.01961, respectively, which indicate fairly flat regression lines. When a
trend line is flat, it indicates that the behavior is not changing over time.

Figure 4.13 Assessing trend in baseline and intervention for yelling behavior
Figure 4.14 ABregres() statistical output for yelling behavior

Using Effect Size to Describe Change

Thus far we have described visual techniques combined with measures of
central tendency and variation to describe changes in behavior across phases.
Effect size quantifies the amount of change between phases. Measuring
effect size has become a common method for assessing the degree of change
between phases as it gives a sense of practical, or clinical, significance
(Ferguson, 2009; J. G. Orme, 1991). When effect size is used to describe
data, it has advantages over tests of statistical significance that are discussed
in the next section as the focus is on the magnitude of change between
phases rather than whether differences are statistically significant (Kromrey
& Foster-Johnson, 1996). Effect size is also relatively easy to interpret and is
understandable to laypeople. There is some controversy, however, regarding
how to calculate effect size (Bloom et al., 2009). SSD for R calculates four
basic forms of effect size: ES, d-index, Hedge’s g, and the g-index.
The ES (Glass, McGaw, & Smith, 1981) is simply the difference between
the intervention and baseline means divided by the standard deviation of the
baseline. The index can indicate no change, improvement, or deterioration.
This index should be used only when there is no trend in either phase
(Kromrey & Foster-Johnson, 1996). A positive ES represents improvement
when an increase in the target is desirable during the intervention; a negative
ES signifies improvement when a decrease in the target behavior is
desirable. The ES can be interpreted in terms of the size of the difference
between the phases in SD units. An ES of +1 signifies a 1-SD increase
between the baseline and intervention. Conversely, a value of –1 indicates a
decrease of 1-SD between the phases. The ES value can be substituted for a
z score, which allows you to estimate the degree the behavior changed in the
intervention over the baseline as a percentage (Bloom et al., 2009).
The d-index is similar to ES except that its calculation is more complex
because it uses a pooled SD, which improves the accuracy of the index.
Therefore, the d-index is appropriate to use when the variation between the
phases differs. The interpretation of the d-index is the same as the ES;
however, unlike the ES, the d-index is not directional. Like the ES, the d-
index should not be used when there is trend in the data.
Hedge’s g is very similar to the d-index when the number of observations
is greater than 20. Hedge’s g has less bias than the d-index when the number
of observations is less than 20.
The SSD for R Effectsize() function automatically displays the percentage
change for the ES, d-index, and Hedge’s g calculations of effect size for any
two phases. Bloom et al. (2009) suggested that an absolute value for ES, d-
index, or Hedge’s g that is less than 0.87 is a small effect size, values of
0.87–2.67 are medium effect sizes, and scores above 2.67 are large effect
sizes (Bloom et al., 2009). The d-index and Hedge’s g are considered more
dependable than ES, which, as previously stated, is influenced by unequal
variation between phases.
As an example, let’s examine the effect size between baseline and
intervention for Jenny’s yelling behavior. We can use these measures of
effect size for yelling because we saw from our previous analysis that yelling
had no trend in either the baseline or the intervention. To do this, we enter
the following command in the Console:
>Effectsize(yell,pyell,“A”,“B”)

Figure 4.15 displays the results of this.

Figure 4.15 Effectsize() function for yelling behavior

Notice the request to choose one of the following: (s)ave, (a)ppend, or

(n)either results? (s/a or n), select n. The other two choices are used to store
the effect size results to a database for use in meta-analysis. This option is
discussed more fully in the meta-analysis chapter. The ES (1.46953), d-index
(1.94711), and Hedge’s g (1.89795) indicate a moderate degree of change.
The percentages of change for the ES, d-index, and Hedge’s g are similar,
with values of –42.92%, 47.42%, and 47.11%, respectively. Consider that
except for ES, the 95% confidence intervals range from a medium to a large
effect. In other words, there is 95% confidence that the true effect size is
between a medium and large effect. The finding provides some evidence that
the intervention is having the desired effect.
 As another example, let’s examine the effect size between baseline and
intervention for Anthony’s obsessive checking behavior (Figure 4.16 [see p.
72]). Open the AnthonyAB file by typing Getcsv() in the Console. We can
use these measures of effect size for checking because, as with the previous
example, checking had no trend in either the baseline or the intervention. To
do this, we enter the following command in the Console:
>Effectsize(checking,pchecking,“A”,“B”)

Figure 4.16 Effectsize() function for checking behavior

The ES (0.44721), d-index (0.59161), and Hedge’s g (0.55924) indicate a

small degree of change. The percentages of change for the ES, d-index, and
Hedge’s g are similar with values of 17.26%, –22.29%, and –21.2%,
respectively. Except for ES, the 95% confidence intervals range from a small
to a medium effect. The finding provides some evidence of a small effect but
does not provide clear evidence the intervention has the desired effect of
reducing Anthony’s checking behavior, at least at this point in time.
As mentioned, when there is a trend in any phase, the ES, d-index, and
Hedge’s g are not appropriate measures of effect size. The g-index (Cohen,
1988) is an alternative measure that can be utilized in these cases, although it
is also suitable for use when there is not a trend. The SSD for R function for
the g-index generates a graph that displays the regression line, the mean, and
the median for the baseline (Figure 4.17 [see p. 73]). When there is a trend in
the data, you will want to pay attention to the regression, and in the absence
of a trend, it is more appropriate to focus on the mean or median. If an
increase in behavior is desirable, then a higher proportion of scores above
the line of focus in the intervention compared to the baseline is desirable. On
the other hand, if lower scores are desirable, then a higher proportion of
scores below the line would be desirable.

Figure 4.17 Gindex() function for calling out behavior

 Consider Figures 4.11 and 4.12, where we examine the effect size for
Jenny’s calling out behavior. If you recall from our previous examination of
trend, there is a strong trend in both phases for calling out. Therefore, it is
appropriate to use the regression line to calculate the g-index. The index is
calculated using the proportion of scores in the desired zone, in this example
below the regression line, for the baseline and intervention. Then the
baseline average is subtracted from the intervention average. To complete
this example yourself, type the following command in the Console:
>Gindex(callout,pcallout,“A”,“B”)

To the left in the Console, you will see the presentation in Figure 4.18 (see p.
74).
The output in Figure 4.18 displays calculations below and above for the
mean, median, and regression line. In this case, the desired zone is below the
regression line (G Regression line). Because the g-index is a positive
number, 0.375, improvement is noted during the intervention. A positive g-
index indicates improvement, while a negative value denotes deterioration.
Figure 4.18 Statistical output from the Gindex() function for calling out behavior

The absolute value for the g-index can be interpreted as follows: Scores
between 0.1 and 0.3 are considered a small effect, scores between 0.31 and
0.5 are considered a medium effect, and scores 0.51 and higher are
considered a large effect (Cohen, 1988). Therefore, we can interpret the
results to indicate that in this case there was moderate improvement between
phases.
Now let’s look at the use of the g-index with Jenny’s yelling, which you
may recall only contained a weak trend in both the baseline and intervention.
Figures 4.19 (see p. 75) and 4.20 (see p. 75) show results for this, which
were created by entering the following command in the Console:
>Gindex(yell,pyell,“A”,“B”)
Figure 4.19 Gindex() function for yelling behavior
Figure 4.20 Statistical output from the Gindex() function for yelling behavior

Because there is some variation in the baseline, the median can be used to
define the desired zone. Since lower scores are desirable, the desired zone is
below the median. The g-index based on scores below the median is 0.816,
which indicates a large degree of improvement in the behavior.
A word of caution about the use of effect sizes in general. While these
calculations indicate the amount of change between phases, observed
changes cannot be attributed to causality. That is, effect sizes alone, even
large effects, do not prove that the intervention was the cause of the
observed change.
Special Cases of Effect Size: Non-overlap Methods
In this section, we discuss a type of effect size known as non-overlapping
data effect size measures. The methods discussed in this section can be used
in lieu of traditional effects sizes (e.g., ES) in a number of situations. For
instance, when data are not distributed normally, there is a great deal of
variation between or within phases or there are multiple phases to be
compared, non-overlap methods may be more appropriate to use (Bloom et
al., 2009; Parker, Vannest, & Davis, 2011; Vannest et al., 2013).
These types of effect size are calculated by using the percentage of data
points in the intervention/treatment phase that is above or below a notable
point in the baseline phase (Lenz, 2012; Scruggs & Mastropieri, 1998,
2013). Five different forms of non-overlap methods are presented:
Percentage of Non-overlapping Data (PND), Percentage of Data Exceeding
the Median (PEM), Percentage of all Non-overlapping Data (PAND),
Improvement Rate Difference (IRD), and the Non-overlap of All Pairs
(NAP). The strengths and weaknesses of these methods will also be
discussed.

Percentage of Non-overlapping Data

The PND is based on the percentage of data points either below the lowest or
above the highest data point in the baseline, or reference, phase. The data
point used is based upon the direction of desired change. If the desired
change is to increase the behavior, then the highest baseline value is used.
On the other hand, if the desired change is a decrease in behavior then the
lowest baseline value is used. The PNDabove() function is used for desired
increasing behavior, and the PNDbelow() is used for desired decreasing
behavior.
Now let’s look at the use of the PND with Jenny’s yelling, which, you
may recall, only contained a weak trend in both the baseline and intervention
phases. Figures 4.21 (see p. 77) and 4.22 (see p. 77) show results for this;
they were created by entering the following command in the Console:
>PNDbelow(yell, pyell,“A”,“B”)
Figure 4.21 PND values below the lowest data point for Jenny’s yelling behavior

Figure 4.22 Statistical output from PNDbelow() for Jenny’s yelling behavior
The PND graphs can be annotated using the ABlines() and ABtext()
commands. The results of the PND indicate that 18% of the data points are
below the lowest baseline point. According to the criteria presented in Figure
4.22, a PND of 0.18 would indicate an ineffective intervention (Scruggs &
Mastropieri, 1998). This finding, however, does not confirm what we see
visually. One of the weaknesses of the PND is that it is based on a single
value in the baseline and is, therefore, susceptible to the influence of outliers
(Lenz, 2012). Therefore, the presence of outliers increases the likelihood of a
small effect, which may not be an accurate representation of actual change.

Percentage of Data Exceeding the Median

The PEM procedure offers a method to adjust for the influence of outliers in
the baseline phase that were observed in the PND example (Lenz, 2012; Ma,
2009). Let’s look at the use of the PEM with Jenny’s yelling behavior to see
if using the median provides a more accurate representation of change.
Figures 4.23 (see p. 78) and 4.24 (see p. 78) show results for this; they were
created by entering the following command in the Console:
>PEMbelow(yell, pyell,“A”,“B”)

Figure 4.23 PEM values below the median for Jenny’s yelling behavior
Figure 4.24 Statistical output from PEMbelow() for Jenny’s yelling behavior

The PEM graphs can be annotated using the ABlines() and ABtext()
commands. The results of the PEM indicate that 88% of the data points are
below the baseline median of two. According to the criteria presented in
Figure 4.24, a PEM of 0.88 would indicate a moderate effect size and is
more consistent with what we see visually.

Percentage of All Non-overlapping Data

The PAND (Parker, Hagan-Burke, & Vannest, 2007) provides an alternative
to PND and PEM, both of which stress a single data point in the baseline,
but to differing degrees. The PAND uses a ratio based on non-overlap of data
between phases (Lenz, 2012). Unlike PND and PEM, the PAND uses all
available data from both phases (Parker et al., 2007).
Let’s continue to use Jenny’s yelling behavior to look at the use of the
PAND. In this example, because decreasing the degree of yelling behavior is
desired, the PNDbelow() function is utilized. Figures 4.25 (see p. 79) and
4.26 (see p. 79) show results for this; these were created by entering the
following command in the Console:
>PANDbelow(yell,pyell,“A”,“B”)
Figure 4.25 PAND values below the lowest data point for Jenny’s yelling behavior

Figure 4.26 Statistical output from PANDbelow() for Jenny’s yelling behavior

The effect size resulting from the PANDbelow() is 0.90625, which is

considered a very effective intervention based on the criteria presented in the
output.
Improvement Difference
The IRD is defined as the difference in the rate of desired behavior between
the baseline and intervention phases (Parker, Vannest, & Brown, 2009).
Using Jenny’s yelling behavior as an example, we can see how this can be
done using the IRDbelow() function with yelling behavior. Here is an
example:
>IRDbelow(yell, pyell,“A”,“B”)

The graph in Figure 4.27 will appear.

Figure 4.27 IRDbelow() graph for Jenny’s yelling behavior with reference line

Figure 4.28 displays the statistical output from the Console. The results
indicate an 81.18% improvement from baseline to intervention. A percentile
chart is presented that indicates the percentile the IRD is in (Parker et al.,
2009). This chart is based on a nonrandom study of data from 166 A-B
single-subject design studies (Parker et al., 2009). An IRD of 81.57% would
be considered moderately high since only about 35% of IRD values obtained
from the sample of studies reported by Parker et al. (2009) would be larger.

Figure 4.28 Statistical output from IRDbelow() for Jenny’s yelling behavior

Non-overlap of All Pairs

The NAP effect size summarizes data overlap between each phase baseline
data point and each intervention data point. A non-overlapping pair will have
a phase intervention data point in the desired direction (lower or higher) than
its paired baseline phase data point. Tied pairs receive half a point. NAP
equals the number of comparison pairs showing no overlap divided by the
total number of comparisons (Parker et al., 2011; Pustejovsky, 2019).
Using Jenny’s yelling behavior as an example, we can see how this can be
done using the NAPbelow() function with yelling behavior. Here is an
example:
>NAPbelow(yell, pyell,“A”,“B”)

Figure 4.29 displays the statistical output from the console.

Figure 4.29 Statistical output from NAPbelow() for Jenny’s yelling behavior

Notice the prompt in the Console to choose one of the following: (s)ave,
(a)ppend, or (n)either results? (s/a or n), select n. The other two choices are
used to store the NAP results to a database for use in meta-analysis. This
option is discussed more fully in the meta-analysis chapter.
A chart of the degree of effectiveness is presented alongside the computed
effect size. In this case, the calculated value of 0.9411765 displayed below
“Est” would be considered very effective (Parker & Vannest, 2009). Unlike
the previous non-overlap methods discussed, dependable 95% confidence
intervals (CIs) can be calculated. In this example, the CI ranges from
0.7773566 to 0.9850379, which indicates that there is 95% confidence that
the impact of the intervention is at least a moderate one.

Use of Non-overlap Methods

The PND, PEM, PAND, IRD, and NAP have a number of advantages over
the ES, d-index Hedge’s g, and g index. A normal distribution or equality of
variance between phases is not required, which makes their use more
practical. Another advantage is that they can be used to compare studies
containing multiple phases; for example, A can be compared to B1, and A
can be compared to B2 (Bloom et al., 2009; Vannest et al., 2013).
If autocorrelation is problematic, its effects can be removed by
transforming the data prior to analysis. Parker (2006), however, found that
the impact of autocorrelation on effect size was minimal (Parker, 2006).
Manolov (2011) found that NAP is the most desirable test when
autocorrelation is present. Archer (2019), on the other hand, found that the
NAP was impacted by autocorrelation (Archer, Azios, Müller, &
Macatangay, 2019). As a result, we strongly recommend testing for
autocorrelation, and if the data are autocorrelated, to consider transforming
the data (Manolov, et al., 2011). As mentioned, outliers in the data can lead
to an increase in the likelihood of small effects. As a result, when there are
outliers, the PEM is more appropriate than the PND.

The Goal Line

The goal line provides a visual method to quantify effectiveness of an
intervention when a specific decision is required to determine if a specific
objective of an outcome measured has been attained. Let’s consider Gloria,
who is being treated for depression at an inpatient hospital. While waiting to
start a dialectical behavior therapy (DBT) program, Gloria’s depression is
assessed with a standardized depression scale. Conventional cutoffs for the
scale are 0–9 for normal range, 10–18 for mild-to-moderate depression, 19–
29 for moderate-to-severe depression, and 30–63 for severe depression. The
objective is to have Gloria’s score on the depression scale to at least fall
within the mild-to-moderate range (below 19).
As an example, let’s use the GloriaAB dataset by invoking the Getcsv()
function. These data include both baseline and intervention data.
Once the dataset is open, type attach(ssd) in the Console to make the
dataset available for analysis. Once the dataset is attached, you can produce
an ABplot() plot for Gloria’s outcomes on the depression scale by entering
the following command in the Console:
>ABplot(depress,pdepress,“Time”, “Behavior”, “Goal Line”)

and the graph will display in the Plots pane.

Next a goal line is placed on the graph by typing Gline() in the console.
The following message will appear in the console: Y ordinate for your goal
line. Type 19 and press <ENTER>. A goal line will be placed on the graph.
We can add a line to separate the phases by entering the following in the
Console:
>ABlines(depress)

The phases were labeled using the ABtext() function and clicking the mouse
in the appropriate location. Figure 4.30 displays the ABplot in the Plots
pane.

Figure 4.30 Goal line graph for Gloria’s depression score

Figure 4.31 displays a substantial decrease in Gloria’s score on the

depression scale. Yet, only measures on the 13th and 14th day are within the
mild and moderate range of depression. Let’s examine the findings more
closely applying the ABdescrip() function by typing the following example:
>ABdescrip(depress pdepress)

Figure 4.31 displays a box plot in the Plots pane. The thick black line in each
box represents the median for each phase (Mda = 37 and Mdb = 26).
Although an 11-point decrease in the median from the baseline to the
intervention is substantial, it can lead to a different conclusion than observed
in Figure 4.30, that the DBT is not efficacious.
Figure 4.31 Box plot of Gloria’s depression score

Autocorrelation
As stated in the previous chapter, autocorrelation should be tested in each
phase as it can impact the results of tests of significance.
We can, for example, look at autocorrelation in the intervention phase for
Jenny’s crying behavior by typing the following command in the Console:
>ABrf2(cry, pcry,“B”)

Statistical results are displayed in Figures 4.32 and 4.33 (see p. 85).
Figure 4.32 ABrf2() for crying behavior in the intervention phase

Figure 4.33 Statistical output from ABrf2() for crying behavior

 The rf2, the measure of the degree of autocorrelation, is high and
statistically significant, thus indicating that the observed autocorrelation may
be problematic. This will have to be considered in any decisions made to test
for statistical differences between phases.
We can try to transform the data to reduce the degree of autocorrelation.
As mentioned in the chapter on the baseline, there are two alternatives for
transforming data, first differencing or moving average.
IMPORTANT NOTE: If you transform data in one phase, you must use
the same method of transformation, either moving average or differencing,
to transform data in subsequent phases. Otherwise, an accurate comparison
of phases, visually or statistically, is not possible.
When there is a trend in the data, the first difference method is preferred.
To do this in SSD for R, use the following command:
>diffchart(cry, pcry,“B”)

This function produces the graph displayed in Figure 4.34. As presented

in this figure, the differencing does smooth out the data. The scores represent
the degree of change between adjacent observations; thus, values can be
positive, negative, or zero. In Figure 4.34, the dotted line represents the
degree of difference between adjacent observations. The solid line represents
the original data. Notice that the shape of both lines is very similar;
removing the trend and autocorrelation in the dotted line, there is more
confidence in what we observe. Because this appears to have possibly taken
care of the autocorrelation problem, you will want to save the transformed
data by responding appropriately to the prompt in the Console.
Figure 4.34 First order differencing for crying behvaior in the intervention phase

After you have saved the transformed data, use the Getcsv() function to
open it. Once the dataset is open, type attach(ssd). Once the dataset is
attached, you can test it for autocorrelation by entering the following
command:
>ABrf2(diff, phase,“B”)

The statistical output of this is shown in Figure 4.35.

Figure 4.35 Statistical results for ABrf2() for transformed intervention data

As we can see, differencing reduced the autocorrelation from an rf2 of

1.019 to an rf2 value of 0.209; additionally, what autocorrelation exists is no
longer significant (sig of rf2 = 0.434). The slope is now 0.02597, indicating
that the trend has also been removed.
Figure 4.36 (see p. 87) displays the graphical output from the function you
entered. Notice that the regression line is flat, providing visual evidence that
the trend was removed.
Figure 4.36 Graphical output from ABrf2() function of transformed data

In the chapter on analyzing the baseline, you used the first difference to
transform your baseline data. Now, both the phases have been transformed,
and the autocorrelation has successfully been removed.
Another option for transforming data is with the ABma() function. As
stated in Chapter 3, it is an appropriate function to use in trying to transform
autocorrelated data when the data in a phase is highly variable.
In the next chapter on statistical significance, you will learn how to use
the Append() command in SSD for R to combine transformed datasets in
multiple phases (e.g., baseline and intervention) when testing for statistical
significance.

Conclusion
In this chapter, we discussed how you might use descriptive statistics to
compare two phases in a single-subject study. These include measures of
central tendency, variation, trend, and effect size. Using Jenny as an
example, you learned the commands necessary to calculate and interpret
these statistics in SSD for R. Additionally, these commands generate
supporting graphs, including box plots, SD band graphs, and line charts
showing the mean, median, and the trimmed mean for any phase that you
can use to compare any two phases.
 SSD for R also provides three standard methods for computing effect size,
which were discussed in detail. Additionally, four methods of evaluating
effect size using non-overlap methods were examined. The chapter
concluded with a discussion of autocorrelation in the intervention phase and
how to consider dealing with this issue in light of the possibility of
autocorrelation in a comparative (e.g., baseline) phase.

Chapter Exercises
Assignment 4.1—Working With Two Phases of Data
For this assignment, you will modify the spreadsheet you created in Chapter
1 with Brenda’s baseline data. You will add data to continue tracking her
behavior into the intervention phase. Add the following data to your
EXISTING file:
Day # of Oppositional Episodes
8 2
9 1
10 1
11 2
12 0
13 1
14 0
15 1
16 1
17 0
18 0
19 0
20 0
21 0
22 0
23 0

Once you add these data to your Google Sheet, do the following:
 1. Import it into RStudio with a different file name than you used before.
2. Create a line graph labeling it appropriately. In the main title, include YOUR initials. For
example, my main title could read, “Brenda’s Oppositional Behavior—WZ”
3. Annotate the line graph to put a vertical line between phases.
4. Label the baseline phase “A” and the intervention phase “B”
5. Export the graph to a Word document and add your answer to the following questions: Based
on the line graph alone, do you think that the social worker’s intervention with Brenda is
having the desired effect? Why or why not?

Assignment 4.2—Visual Comparison of Two Phases

For this assignment, you will continue to use the file you created in
assignment 4.1 with baseline and intervention data for Brenda.
1. Re-create the graph you produced for the previous assignment by creating and labeling a line
graph. Make the main title contain your own initials.
2. Add a vertical line between phases and label each phase as “A” or “B.”
3. Add a mean line for each phase and label the line with the mean value for each phase.
4. Produce descriptive statistics for each phase by using the ABdescrip() function.

Then report the following:

1. Include the line graph in your report.
2. Analyze each phase of the data by reporting the phase, number of observations in the phase, the
mean, sd, median, and range of data in the phases.
3. Visually compare phases: Does the problem appear to be accelerating, decelerating, or zero-
celerating? What makes you think this?
4. Based on this analysis, if you were working with Brenda, what would you do at this point?

Assignment 4.3—Effect Sizes

1. Create a line graph of Brenda’s tantrums, separating the phases with a vertical line. Take an
educated guess about whether your intervention is “not effective,” is “debatably effective,”
“moderately effective,” or “very effective.”
2. Consider the qualities of Brenda’s tantrums (i.e., trending and autocorrelation in each phase)
and determine which type(s) of effect sizes would be appropriate to use.
3. Calculate those effect sizes. Were you right in your visual analysis? Why do you think you
might have been right or wrong?
4. Now, take what you have learned and make a practice decision. What would you do if Brenda
was your client?
5
Statistical Tests of Type I Error

Introduction
In this chapter, we discuss tests that can be used to compare your data across
phases. These are all tests of statistical significance that are used to
determine whether observed outcomes are likely the result of an intervention
or, more likely, the result of chance. Used correctly, statistical tests can help
you make sounder decisions about your data (Bloom et al., 2009).
Usually, the first step in significance testing is to form a hypothesis of no
difference, referred to as the null hypothesis. In single-system design, the
typical null hypothesis, described as H0, is that the target behavior did not
change after the introduction of the intervention. For example, we could
state the null hypothesis that Jenny’s self-esteem did not change after
participation in a social skills group. An alternative hypothesis, denoted as
H1 or HA, is that the target behavior changed after the introduction of the
intervention. This is a nondirectional hypothesis because, stated as such,
change can be positive (i.e., desired) or negative (i.e., undesired). Continuing
the Jenny example, we could test the hypothesis that Jenny’s self-estemm did
change after participation in the school’s social skills group. A directional
null hypothesis would be that the target behavior was either the same or
worse after the intervention, such as Jenny’s self-esteem not improving or
deteriorating after participating in social skills. The alternative directional
hypothesis would be that the behavior improved after the intervention, and
we would see an improvement in Jenny’s self-esteem after participating in
the social skills group. We recommend the use of a nondirectional
hypothesis because it allows you to establish if the change is positive or
negative.
 The purpose of a statistical test is to determine how likely it is that we are
making an incorrect decision by rejecting the null hypothesis and accepting
the alternative one. In statistics, this is referred to as a Type I error. In more
common parlance, this is a “false positive.” By using probability, tests of
statistical significance quantify the likelihood of making a Type I error.
Typically, in the social sciences, we are willing to accept the alternative
hypothesis if the chance of making a Type I error is 5%, or a probability of
.05, or less. This is typically shown as output in statistical programs as a p
value (e.g., p ≤ .05) or a sig value (e.g., sig ≤ .05).

Statistical Significance Versus Practical Significance

It should be noted that if you observe statistical significance between phases,
this does not guarantee that the observed differences will be noticeable in a
real-life or practical setting. It is possible to compute statistically significant
differences, but observe a small magnitude of change that is, perhaps,
imperceptible to you or your client. Conversely, you could observe moderate
or large clinically meaningful differences between the baseline and
intervention phases that are not statistically significant due to the way those
computations are calculated.
Therefore, we recommend, and best practice dictates, that if you are going
to conduct hypothesis tests, you should holistically consider both the results
of your hypothesis test and effect size calculations when making practice
decisions.
Consider the following example: You compare phases with a hypothesis
test, but the difference between phases is not statistically significant. This
could lead you to the conclusion that there is no difference in the client’s
behavior after the intervention, and you decide to discontinue the
intervention. If, however, you continue your analysis, you may notice that
there is a moderate change in the desired direction, it is possible that your
intervention is having the desired impact, and more time and data are needed
with the current intervention to observe statistically significant findings.
A number of tests of significance are presented in this chapter: statistical
process charts (SPCs), proportion/frequency, chi-square, the conservative
dual criteria (CDC), robust conservative dual criteria, the t test, and analysis
of variance (ANOVA). How and when to use each of these tests is also
discussed.
Statistical Process Control Charts
As Orme and Cox (2001) pointed out, SPC charts, also known as Shewhart
charts or control charts, are very familiar to social work research. This
section is based on the work of Orme and Cox (2001), who did an excellent
job of describing the use of SPC charts, which come out of the discipline of
industrial quality control (Orme & Cox, 2001). A number of these charts
were described at length by Bloom et al. (2009), and, in recent years, these
charts have become popular in healthcare quality improvement, which
requires making changes in processes of care and service delivery. In the
social and health sciences, SPC charts are used to measure if process
changes are having a desired beneficial effect (Benneyan, Lloyd, & Plsek,
2003; Mohammed & Worthington, 2012; Polit & Chaboyer, 2012; Smith et
al., 2012; Tasdemir, 2012; Woodall, 2006).
Since a stable process will follow a normal distribution, we expect
approximately 95% of collected data to fall within two standard deviations
and approximately 99% to be within three standard deviations (Benneyan et
al., 2003). The probability that any one observation would fall above or
below three standard deviations is .0027, which would be considered
statistically significant because it is less than .05; however, it should be
noted that the standard deviation for SPC charts is derived differently from
the form we discussed in the section on variation.
In the examples we provide in this chapter, we use the more conservative
three standard deviation band approach in examining SPC charts; however,
two standard deviation bands have been used in prior research and are also
considered acceptable. Using two standard deviations, as opposed to three,
reduces the chances of making a Type II error, or a false negative, but
increases the chances of making a Type I error (0.025 compared to 0.0027
for a two-tailed test). In SSD for R, you have the option of choosing SPC
charts with one, two, or three standard deviations.
With the enactment of the Patient Protection and Affordable Care Act
(2010), there has been an increased need for accountability. SSD for R
produces a number of SPC charts that can be used specifically for this
purpose.
The SPC charts have a number of advantages in single-system data
analysis. Polit and Chaboyer (2012) described a number of these. First, they
can be used with data with a variety of characteristics. They can be grouped
by a time-based variable such as days or weeks. Additionally, the impact of
an intervention can be detected even as data are being collected. Because
SPC charts are easy to understand, they are an excellent form of
communicating change over time to untrained audiences (Polit & Chaboyer,
2012).
There are a number of SPC charts presented in this chapter, and not all
charts are appropriate for every situation. Decision trees are available in
Appendix C to help you select the chart(s) that are appropriate for your data
and reporting needs (Orme & Cox, 2001).
All SPC charts presented in this section could be impacted by
autocorrelation, so it is always a good idea to test for it prior to generating
any of these. According to Wheeler (2004), SPC charts work well when
autocorrelation is modest (Wheeler, 2004). The adequacy of these
procedures is diminished when the lag-1 autocorrelation is high (i.e., greater
than or equal to 0.6).

X-bar Range Chart (X-R Chart)

The X-bar range chart (X̄-R chart) can be used when there are multiple
observations per sample (Orme & Cox, 2001). For example, in hospitals,
social work services are often extended to the emergency department (ED) to
decrease the rate of unnecessary hospital admissions due to nonmedical-
related factors. The director of social work could track the daily percentage
of admissions for nonmedical reasons. A retrospective baseline could, for
example, be acquired for 6 weeks prior to the implementation of social work
services and 6 weeks postintervention. If data were collected on a daily
basis, there would be 12 samples (6 weeks of baseline plus 6 weeks of
intervention), with seven observations (days) per sample. Using the mean of
these samples increases the accuracy of our measure (Mohammed &
Worthington, 2012; Orme & Cox, 2001).
Figure 5.1 displays the X̄-R chart for this example of ED data. To begin
creating this chart yourself, you will need to use data from the ed.csv dataset
by entering the following command in the Console:
Figure 5.1 Example of X-bar R chart

>Getcsv()

Once the dataset is open, attach it by entering the following command in the
Console:
>attach(ssd)

To begin, it is important to consider each variable in the dataset. Note that

when you open this dataset you can view the contents by clicking on the
spreadsheet icon to the right of the file in the Environment pane. When you
do this, you will note that there are three variables listed: admits, admitweek,
and padmit. The admits variable contains the data on the percentage of
nonmedical admissions for each day, admitweek is the grouping variable for
the eight samples (i.e., weeks), and padmit is the phase variable with “A”
representing the baseline and “B” representing the intervention.
 To create the annotated X-R chart, enter the following SSD for R
commands in the Console. After each annotation, you will be prompted to
accept the alteration to the plot. Once you accept each one by entering y in
the Console, note that you will not be able to change that annotation again.
Command Purpose
XRchart(admits,admitweek, 3, “weeks”,“mean % of Creates SPC chart using three
admits”,“Social Admits”) standard deviations
ABlines(admits) Adds line between phases
ABtext(“73.94”) Enters mean text
ABtext(“79.32”) Enters Uband text
ABtext(“68.56”) Enters Lband text
SPClegend() Creates a legend

Note that once you create a legend, you will no longer be able to alter the
graph. Therefore, we recommend that you add this as a last step in creating
SPC charts. Be sure that your graph looks exactly as you would like it before
adding the legend.
In the Console, you will see the values for the upper band (Uband), mean,
and lower band (Lband). These values were used to label the X̄-R chart.
In this case, the desired zone would be below the lower band (68.56)
because the hospital’s goal is to reduce the number of nonmedical
admissions. The last 3 weeks of the intervention are in the desired zone,
showing that the intervention has some promise. The hospital administrator
would probably want to continue measuring this process over some time
because it seems that the intervention is going in the desired direction. It
would seem that the process of a high level of nonmedical admissions might
have been interrupted by the intervention. To increase confidence, continued
tracking would be recommended.

R Chart
The R chart is designed to detect changes in variation over time. There are
two forms of this chart available with SSD for R. One version of the R chart,
which is recommended when the sample size is small (i.e., less than 10),
uses the mean range of the samples to track variation. Similar to the X̄-R
chart, using the mean range improves confidence in the measure because it
takes into account all available data. If the sample size is larger than 10, use
of the standard deviation is recommended over the range (Orme & Cox,
2001). Although our hypothetical data have a sample size of 12, for
illustration purposes we have presented an R chart using the range.
Figure 5.2 was created with data from the ed.csv file by entering the
following commands in the Console:

Figure 5.2 Example of R chart based on mean range

Command Purpose
Rchart(admits, admitweek, 3,“weeks”,“mean Creates SPC chart using three standard deviations
range of admits”,“Social Admits”)
SPCline() Adds line between phases; click twice to indicate
the top and bottom of the line
ABtext(“24.69”) Enters mean text
ABtext(“12.83”) Enters Uband text
ABtext(“0.97”) Enters Lband text
SPClegend() Creates a legend

Figure 5.2 shows that all sample mean ranges for the 12 weeks are within
the two bands, indicating that the variation for this process remained stable.
The commands in the following table can be used to create an R chart
using the standard deviation as the measure of process:
Command Purpose
Rchartsd(admits, admitweek, 3, “weeks”, “mean Creates SPC chart using three standard
SD of admits”,“Social Admits”) deviations
SPCline() Adds line between phases; click twice to indicate
the top and bottom of the line
ABtext(“8.61”) Enters Uband text
ABtext(“4.48”) Enters mean text
ABtext(“0.34”) Enters Lband text
SPClegend() Creates a legend

Figure 5.3 shows virtually the same trend displayed in Figure 5.2. More
importantly, all the weekly standard deviations are within the two bands,
demonstrating that the variation for this process remained stable. As we
described previously, extreme variation influences the stability of a measure.

Figure 5.3 R chart based on standard deviation bands

Proportion Chart
Often clinicians need to track a client system to determine the extent to
which a set of tasks has been completed. When the target behavior has a
binary outcome (e.g., completing or not completing a task), comparing the
proportion of task completion over time or between phases can be measured.
Attendance (vs. nonattendance) in a group, child welfare workers
completing records on time (vs. not completing on time), workers
completing home visits (vs. not completing home visits), whether a patient is
compliant with taking medication (vs. noncompliant), and if a patient with
hearing loss wears their hearing aid (vs. not wearing the hearing aid) are all
examples of task completion that can be assessed using proportion charts (P
charts). For a more in-depth discussion of this procedure, we suggest you
refer to J. G. Orme and Cox (2001) and Mitra (2008).
As an example, open the Jennyab.csv dataset, attach it, and view the
variables by clicking on the spreadsheet icon to the right of the ssd file in the
Environment pane. In this example, we use the following three variables:
group, wgroup, and pgroup. The group variable is a binary measure of group
attendance, with “0” indicating that Jenny did not attend the daily group and
“1” indicating that she did attend. The variable wgroup is the week of
attendance, and pgroup is the phase variable indicating whether the measure
was obtained during the baseline or intervention. To improve attendance,
after the fifth week an intervention was introduced; children who had perfect
weekly attendance were given an extra 20 minutes of free play on Friday
afternoons.
These data have 15 samples (i.e., weeks) and five observations (i.e., days)
per sample. Entering the following commands will create the p chart in
Figure 5.4.
Figure 5.4 P chart based on 3-sd bands

Command Purpose
Pchart(group,wgroup, 3, “weeks”, “proportion of attendance”, Creates SPC chart with three
“Jenny’s Group Attendance”) standard deviations
ABlines(group) Adds line between phases
ABtext(“0.32”) Enters mean text
ABtext(“1.00”) Enters Uband text
ABtext(“–0.36”) Enters Lband text
SPClegend() Creates a legend

Figure 5.4 shows that all the samples during the baseline are within the
upper and lower bands. After the intervention, the proportion of attendance
increases, and, during the last 5 weeks, Jenny attended all group sessions.
The findings from the chart would provide evidence to continue the
intervention.

X-Moving-Range Chart
Like the previous SPC charts, the X-moving-range chart (X-mR chart) can
be used to detect changes within and between phases. This chart should not,
however, be used when there is a trend in the data. It can be used, though,
when you are not using samples of data, but individual data points, such as
what we have seen with the majority of Jenny’s data. As with the previous
charts, large unexpected changes in the undesired zone may indicate the
need to modify the intervention. Additionally, slow or no movement into the
desired zone may also indicate the need to modify the intervention. During
the baseline, abrupt change, specifically undesirable change, may signal a
need to begin the intervention.
Figure 5.5 provides an example of hypothetical data on self-esteem
measured on a scale of 1 to 5, with 1 being the lowest level of self-esteem
and 5 being the highest. Because we want Jenny’s self-esteem to increase,
the desired zone, then, would be above the upper band.

Figure 5.5 X-mR chart based on Jenny’s self-esteem during baseline phase

To create this chart yourself, enter the following using data from the
Jennyab dataset:
Command Purpose
>Xmrchart(esteem, pesteem, “A”, 3, “weeks”,“self-esteem Creates SPC chart with three
score”,“Jenny’s Self-Esteem”) standard deviations
>ABlines(esteem) Adds line
>ABlines(esteem) Adds line
>ABtext(“3.12”) Adds mean text
>ABtext(“5”) Adds upper band text
>ABtext(“2”) Adds lower band text
>ABtext(“A”) Adds text
>ABtext(“B”) Adds text
>ABtext(expression(A[1])) Adds text with subscript
>SPClegend() Adds legend

Note that, in this example, we have three phases: a baseline (A), an

intervention (B), and a return to baseline (A1), as the intervention was
removed after the 24th week. In this case, the intervention is to encourage
Jenny’s parents to be more responsive to her at home. After the intervention,
Jenny’s level of self-esteem increased and was sustained when the
intervention was removed (A1).

C Chart
It is suggested that when the outcome is a count of behavior the C chart
should be utilized. Examples of behaviors that are counts would be the
number of times a child displays aggressive behavior, the number of times a
child leaves his or her seat, or the number of times a child yells out an
answer.
As an example, you can re-create Figure 5.6 (see p. 100) by entering the
following commands in the Console using the Jennyab dataset:
Figure 5.6 C chart based on Jenny’s yelling during baseline phase

Command Purpose
>Cchart(yell,pyell,“A”, 3,“days”, “Count”, “Count of Creates SPC chart with three standard
Yelling”) deviations
>ABlines(yell) Adds line between phases
>ABtext(“2.8”) Enters mean text
>ABtext(“8”) Enters Uband text
>ABtext(“–2”) Enters Lband text
>ABtext(“A”) Enters text for the A phase
>ABtext(“B”) Enters text for the B phase
>SPClegend() Creates a legend for the graph

As Figure 5.6 illustrates, all the points are within the upper and lower
bands. Although there are no unusual observations during either the baseline
or the intervention, the yelling was consistently lower after the intervention
was initiated.

Additional Notes on SPC Charts

Once again, all the charts presented in our examples used three standard
deviation bands. Bloom et al. (2009) pointed out that using the bands with
three standard deviations creates a condition where the probability of making
a Type I error is low (0.0027); therefore, they posited that the X-mR chart
will be reliable even under conditions of autocorrelation.
On the other hand, Mohammad and Worthington (2012) suggested
attribute charts, like the p chart and C chart, can at times be misleading.
Therefore, they suggested that using more than one SPC chart can reduce
flawed interpretations of findings. Under conditions of normal variation, the
X-mR chart will be in close agreement with the p chart and C chart. If this is
not the case, it could signal some underlying problem that could lead to
misinterpretation (Mohammed & Worthington, 2012).

Other Tests of Type I Error

In addition to the numerous SPC charts described in the preceding section,
there are other tests of Type I error that are appropriate for use in single-
system research. SSD for R has functions to conduct these tests; however, the
decision to use a particular test of Type I should be based on the nature of
the data in each phase. See Appendix C for decision trees to help you select
the most appropriate test to use.

Proportion/Frequency or Binomial Test

The proportion/frequency procedure compares typical patterns between the
baseline and the intervention. In this statistical test, the binomial distribution
is used to test for statistical significance between phases (Bloom et al.,
2009).
Based on the properties of the normal distribution, we know that 68.2% of
all scores should fall within one standard deviation of the mean. We can,
therefore, use this knowledge to define desired and undesired zones. If
higher values signify improvement, then the desired zone will be above the
upper band. On the other hand, if lower values indicate improvement, the
desired zone will be below the lower band. Two standard deviations, which
encompass 95% of scores in a normal distribution, could also be used to
define the desired zone; however, the probability of a score falling above or
below two standard deviations is only 0.025.
Let’s use Jenny’s yelling behavior as an example. We begin by
constructing a standard deviation band graph that extends the mean and
standard deviation bands calculated for the baseline into the intervention
phase. In this case, the desired zone would be below the lower band because
we want the instances of yelling to decrease. Any other location on the graph
would be considered undesirable. Figure 5.7 (see p. 102) displays this graph.
The following commands with the Jennyab file open produced the figure:

Figure 5.7 Line graph of Jenny’s yelling with 1-sd bands based on baseline phase

Command Purpose
SD1(yell,pyell,“A”,“days”, “Count”, “Count of Yelling”) Creates graph
ABlines(yell) Adds line between phases
ABtext(“+1sd = 4.06”) Enters text between quotes
ABtext(“–1sd=1.54”) Enters text between quotes
ABtext(“mean=2.8”) Enters text between quotes

As the figure illustrates, many more of the observations of the target

behavior are in the desired zone in the intervention phase, which is below the
lower standard deviation band, compared to the baseline. This shows that the
yelling behavior dropped sharply during the intervention.
We can test if the proportion of scores in the desired zone during the
intervention was significantly different from the baseline using the
ABbinomial() function. The first step to doing this is to count the number of
observations in the desired zone for both the baseline and intervention,
which are 1 and 15, respectively.
The following command entered in the Console will produce the output in
Figure 5.8:
>ABbinomial(pyell,“A”,“B”, 1, 15)
Let’s dissect this function: Here pyell is the phase variable for Jenny’s
yelling behavior; “A” is the baseline phase or the phase that is to serve as
the basis for comparison, “B” is the intervention phase, 1 denotes the
number of observations in the desired zone for the baseline phase, and 15 is
the number of observations in the desired zone during the intervention. With
the ABbinomial() function, any two phases can be compared, but the letter
entered after the phase variable needs to be the phase to which the second
phase is compared.

Figure 5.8 Console output for binomial test of Jenny’s yelling across phases

The null hypothesis is that there is no difference between the rates of

success, as defined by the number of points in the desired zone, in the two
phases. The alternative hypothesis is that the rates are different. The p value
in Figure 5.8 is displayed in scientific notation, denoting that it is very low.
If you want to view the entire decimal value, there are two options. The first
turns off scientific notation for your entire R session. Enter the following in
the Console:
options(scipen=999)

Alternatively, you can use the following SSDforR function:

>SN(2.73e-16)

As we discussed in the beginning of this chapter, we need a significance

value of .05 or less to reject the null hypotheses and therefore accept the
alternative. Because the p value, or significance level, is less than .05 we can
reject the null. Because 15 of 17 observations during the intervention were in
the desired zone during the intervention, the rate of success is 0.8823529.
This simply means that 88% of the observations were in the desired zone
during the intervention. We can compare this to the rate of success during
the baseline, which is listed next to true probability of success is not equal to
in the Console. In this case, only 1 observation out of 15, or 6.7%, was in the
desired zone during the baseline.

Another Example
Proportion/frequency can also be used to help track task completion and can
be used as an alternative to SPC p charts. When the target behavior has a
binary outcome (i.e., completing or not completing a task), comparing the
proportion of tasks completed over time or between phases can be measured.
One example is a client’s attendance in a group (attendance vs.
nonattendance).
Let’s use Jenny’s group attendance as an example. The behavior variable
group is coded as “1” for attended and “0” for did not attend. We can use the
R table() function to compare the frequency of attendance during the
baseline to the intervention. To do this, enter the following in the Console:
>table(group, pgroup)

Figure 5.9 is displayed in the Console.

Figure 5.9 Console output of table()function

From the output in the table in Figure 5.9, we can see that Jenny attended
the group 8 out of 25 (17 + 8) sessions during the baseline compared to 40 of
50 (10 + 40) sessions during the intervention.
The ABbinomial() function can now be utilized to test if the rate of
attendance during the intervention was statistically different from the
baseline. In this example, we would enter the following into the Console
with the output displayed in Figure 5.10:
>ABbinomial(pgroup,“A”,“B”, 8, 40)
>SN(3.934e-12)

The significance value is less than .05, and the null hypothesis can therefore
be rejected. The output shows that there was an increase from a 32% (0.32)
attendance success rate during the baseline to 80% (.8) during the
intervention.

Figure 5.10 Console output for binomial test of Jenny’s group attendance across phases

Using Proportion/Frequency With Client Goals

Recall that Gloria is experiencing depression, and you would like to assess
whether her sleeping improves with the introduction of cognitive behavioral
therapy (CBT). Since clients with depression can sleep too much or too little,
you work with Gloria to determine a healthy range of sleep, between 6 and 8
hours per night.
You can use the Gline() function to create boundaries around your desired
zone in a line graph. Begin by opening and attaching the GloriaABC file.
> ABplot(sleep, psleep, “night”, “hours”,“Gloria’s sleeping”)

You will want to separate your phases with a vertical line:

 > ABlines(sleep)

Then create your desired zones with the addition of Gline(), specifying lines
at both 6 and 8 hours of sleep. Follow the prompts in the Console until your
plot looks like the one illustrated in Figure 5.11.

Figure 5.11 Line graph of Gloria’s sleeping with defined goal lines

Notice that by using the Gline() function you are able to customize what
is considered desirable. This function, then, can be used in collaboration
with work with clients to help assess the degree to which they are reaching
goals that are meaningful to them.
We can now count two desirable values in the baseline and six in the
intervention. With this in mind, we can run the ABbinomial() function:
> ABbinomial(psleep, “A”, “B”, 2, 6)

The output found in the Console is illustrated in Figure 5.12 (see p. 106).
Figure 5.12 Console output for binomial test of Gloria’s sleep across phases

The output illustrates that Gloria slept a healthy amount 40% of the time
during the baseline and 6 out of 10 times, or 60% of the time, during the
intervention. While we see an improvement, the p value of .2126 is greater
than .05, so we are unable to reject the null hypothesis.
This is disappointing, but remember that the number of observations in
each phase impacts p values, so do not give up just yet. Perhaps with more
time, statistical significance could be achieved.

Chi-Square
The chi-square (χ2) is a widely used statistic in the helping professions. The
chi-square is a nonparametric statistic used to test for Type I error when data
are categorical. Because it is a nonparametric test, we assume independence;
therefore, the chi-square should not be used when autocorrelation is
problematic. Before deciding to use chi-square or any other test for Type I
error, you should test both phases to see if autocorrelation is a problem.
In this example, let’s examine Jenny’s yelling behavior more closely. You
can test two phases using the ABrf2() function to test for autocorrelation of
Jenny’s yelling behavior, by entering the following commands in the
Console, the results of which are displayed in Figure 5.13 (see p. 107):
>ABrf2(yell, pyell, “A”)
>ABrf2(yell, pyell, “B”)
Figure 5.13 Console output assessing autocorrelation of yelling in two phases

The rf2 values in Figure 5.13 display small nonsignificant rf2 values of
0.267 and 0.383, respectively, for the baseline and intervention phases.
Because we met the assumption that the data in both phases are sufficiently
independent, we can proceed with a chi-square test.
Because the chi-square is used for categorical data, a method to create
frequencies is necessary. Creating a desired zone similar to what was done
for the proportion/frequency can be used. Five methods for doing this are
available in SSD for R: the median, the mean, the trimmed mean, the
ordinary least squares (OLS) regression line, and the robust regression line.
You should use whichever method is most appropriate for your situation.
The median or trimmed mean is appropriate when there are outliers and
no trend. If there were a trend, but not severe outliers in the data, it would be
more appropriate to use the OLS regression method. If there were outliers
and a trend, it would be preferable to use the robust regression method.
Finally, if there is no trend and there are no outliers, it would be appropriate
to use the mean. To help you choose which method is most appropriate for
your situation, refer to Appendix C.
To better illustrate the general use of the chi-square methods, we can look
more closely at Jenny’s yelling behavior, the results of which are displayed
in Figures 5.14 (see p. 108) and 5.15 (see p. 109). This example uses the
baseline mean to form the desired zone. We chose this method because there
are no outliers in the data, and there is not a significant trend in either phase.
Because we want the number of yelling episodes to decrease, the desired
zone is below the regression line. Enter the following command in the
Console:
>meanbelow(yell, pyell,“A”,“B”)
Figure 5.14 Console output of chi-square test comparing Jenny’s yelling across phases
Figure 5.15 Visual output of chi-square test comparing Jenny’s yelling across phases

The first part of the output in Figure 5.14 displays the frequencies and
percentages. The “[,1]” is the baseline phase, and “[,2]” is the intervention
phase. “FALSE” is above the reference line and represents what is not
desirable, while “TRUE” is below the mean line, or in the desired zone. For
the baseline, eight observations (53.3%) are in the desired zone, and seven
(46.7%) are in the undesired zone. For the intervention, all 17 observations
(100%) are in the desired zone.
The next section shows the frequencies are the percentages for all
observations across all phases. All of the undesired scores (100%) occurred
in the baseline. Over two thirds (68%) of the observations in the desired
zone occurred in the intervention.
The p value for the chi-square is .001439. Because the p value is less than
.05, the null hypothesis can be rejected. Because the entire table is 2 × 2
(two columns by two rows), Fisher’s exact test is calculated. This test
provides a more precise calculation of significance in small tables with small
cell sizes. In this case, the p value = .001912. Based on this result, we would
conclude that there was a statistically significant higher frequency of scores
in the desired zone during the intervention compared to the baseline.
 The plot shown in Figure 5.15 (see p. 109) illustrates both baseline and
intervention phases with the horizontal line placed at the mean for the
baseline, denoting the boundary for the desired zone.
If the desired zone were above the mean line, the command would be as
follows: meanabove (yell,pyell,“A”,“B”). If the median was to be used to
determine the desired zone, the commands would be medbelow(yell,pyell,
“A”,“B”) or medabove(yell,pyell,“A”,“B”). If the OLS regression line was
used, the commands would be regbelow(yell,pyell,“A”,“B”) or
regabove(yell,pyell, “A”,“B”). If the trimmed mean was to be used, the
commands would be trimdbelow(yell,pyell,“A”,“B”) or
trimabove(yell,pyell,“A”,“B”). If you were using robust regression and the
desired zone was above the line, the command would be robregabove(yell,
pyell, “A”, “B”). If you were using robust regression and the desired zone
was below the line, the command would be robregbelow(yell, pyell, “A”,
“B”).

Conservative Dual Criteria

The CDC is a test of Type 1 error that works well when data have relatively
high autocorrelation provided there are between 5 and 20 intervention data
points (Fisher et al., 2003; Morgan, 2008; Stewart et al., 2007; Swoboda et
al., 2010; Tate & Perdices, 2019). The CDC uses two lines to define a
desired zone: the mean and the regression line of the comparison phase,
usually the baseline. If the preferred outcome is an increase in the target
behavior, then the desired zone would be above both lines, and the
CDCabove() function should be used. On the other hand, if lower scores are
preferred, then the desired zone would be below both lines, and the
CDCbelow() function should be utilized.
In our example of proportion/frequency, we considered whether there was
a significant change in Gloria’s sleeping prior to and after the introduction of
CBT; however, that was based on a goal set between Gloria and her social
worker. Let’s consider whether there were any statistically significant
differences in her sleeping from baseline to intervention. Begin by assessing
the degree to which there is autocorrelation in both phases:
>ABrf2(sleep, psleep, “A”)
>ABrf2(sleep, psleep, “B”)
 The output in the Console is shown in Figure 5.16. Notice that there is a
moderate amount of autocorrelation in both phases even though it is not
significant in either phase. To be on the safe side, we will not assume
independence of observations, particularly since there are very few
observations (n = 5) in the baseline.
Figure 5.16 Console output assessing autocorrelation of sleep in two phases

We can use the CDC to determine whether there were significant changes
between baseline and intervention. Since Gloria’s goal was to increase her
sleep, we use the CDCabove() function:
>CDCabove(sleep, psleep, “A”, “B”)

As Figure 5.17 (see p. 111) shows, during the intervention, there were six
observations above both the adjusted regression line and the adjusted mean
line. In the Console, you see that noted as the TRUE TRUE condition. The
output in the Console tells you that eight observations are needed in this
desired zone in order to achieve statistical significance, and we are short by
two. While things seem to be going in the right direction, this test reveals
that there are no significant differences between the phases at this point.

Figure 5.17 Output results of CDC test comparing sleep across phases

There is a robust version of the CDC, RobustCDCbelow() and

RobustCDCabove(), which replaces the mean with the trimmed mean and
the OLS regression line with the robust regression line. Brossart, Parker, and
Castillo (2011) made the case for the use of robust regression because it is
less susceptible to the influence of outliers; therefore, it would be advisable
to use a robust form of the CDC function if there are outliers in either phase.
It should be noted that, unlike some of the other tests you have read about
in this section, there is no p value or significance value presented. You
simply are provided the minimum number of data points that have to be in
the desired zone in the intervention in order to achieve statistical
significance.

t-test
The proportion/frequency test, the chi-square, and the CDC are based on
comparing the number of observations in some desired zone in one phase to
another. The t test compares the differences in the means between phases.
The null hypothesis is that the difference in means between the phases is
zero (i.e., there is no change between phases). If the t-test is statistically
significant (i.e., p ≤ .05), then we accept the alternative that the mean
differences are greater than zero. The difference, however, could be
improvement or deterioration in the behavior. If higher values are desired,
the difference should be positive, and if lower values are desired, the
difference should be negative.
There are some caveats for the use of the t test. It should not be used if
there is a trend in the data or if either phase has problematic autocorrelation
(Auerbach & Schudrich, 2013).
The t test also assumes equal variation between the phases. If the
variances are unequal, then a t test for unequal variances should be used.
SSD for R produces results for both equal and unequal variances. The
ABttest() function allows you to test the differences in variance between
phases and provides output for both versions of the t test.
As an example, let’s use a simple t test to look at Jenny’s yelling behavior
across phases. Previously, we determined that there was no significant trend
in either phase, and independence of observations can be assumed. Enter the
following command in the Console to get the output that is displayed in
Figure 5.18 (see p. 113) and Figure 5.19 (see p. 114):
>ABttest(yell, pyell,“A”,“B”)
Figure 5.18 Console output results of t-test comparing yelling across phases

Figure 5.19 Visual comparison of mean yelling across phases

There are three tests displayed in the Console panel to the left of the
graphical output: a t test for equal variances (Two Sample t-test), a test for
equality of variances (F test to compare two variances), and a t test for
unequal variances (Welch Two Sample t-test). Because the p value for
equality variances is less than .05 (.002407), we can reject the null
hypothesis that the variances are not different and accept the alternative that
they are not equal. Therefore, we use the t-test results for unequal variances
to determine whether there are statistically significant differences between
the baseline and the intervention. Looking at the results in this case, the null
hypothesis is rejected because the p value is less than .05 (4.685e-05 or
.00004685). Therefore, we accept the alternative that the mean difference
between the phases is greater than zero. We also observe that the mean went
down from 2.8 in the baseline to 0.9411765 in the intervention. We can
therefore conclude that the introduction of the intervention was associated
with an improvement in behavior.
The bar plot depicting this is shown in the Plots pane to the right of the
Console.
Note that any two phases can be compared with t tests. If, for example,
there were a second intervention (C), you could compare the mean of that
phase against the baseline or against the first intervention (B), but not both.
One-way ANOVA is a better choice if you want to compare more than two
phases simultaneously.

One-Way Analysis of Variance

The one-way ANOVA is an extension of the t test. The assumptions are the
same as required for the t test (i.e., no autocorrelation, no trend, equal
variances, and normal distribution). However, this test is different because
more than two phases can be compared simultaneously.
If we analyzed a behavior with three phases—a baseline (A), an
intervention (B1), and the introduction of a more intense intervention (B2)—
three separate t tests would be needed: A compared to B1, B1 compared to
B2, and A compared to B2. A one-way ANOVA offers a single test to
determine whether the three phases differ from one another. The use of the
ANOVA in this case, then, is advantageous over multiple t tests as multiple
tests of significance are more likely to lead to spurious results.
Let’s consider Gloria’s case, in which her depression was measured prior
to intervention, after the introduction of an intervention, and again when the
intervention was intensified by doubling the number of sessions she was
having each week.
We can proceed with conducting the one-way ANOVA because there were
no significant trends in any phase, and autocorrelation in each phase was low
and not significant. As a refresher from Chapter 4, we can see the descriptive
statistics for Gloria’s depression by entering the following in the Console.
>ABdescrip(depress, pdepress)

Output displayed in the Console reminds us that Gloria’s baseline mean

depression was 37.8 (sd = 5.85). After the introduction of the intervention,
her mean depression dropped to 24.8 (sd = 4.92), and when the intervention
was intensified, her mean depression score dropped further to 18.0 (sd =
5.12).
With Gloria’s dataset open and attached, type the following command in
the Console:
>ABanova(depress,pdepress)

Figures 5.20 (see p. 115) and 5.21 (see p. 115) display the results of the one-
way ANOVA.
Figure 5.20 Console output results of one-way ANOVA comparing depression across phases
Figure 5.21 Visual comparison of mean depression across phases

The results of the one-way ANOVA indicate that the means between the
three phases are statistically different, and the null hypothesis can be
rejected. Note that the significance value is in scientific notation (2.65e-06).
Next to this value are three asterisks, indicating the p value is less than .001.
Note that the difference in the phase means are reported in alphabetical
order. The mean drops 13 points from the baseline to the intervention, is
reduced another 6.8 points on intensifying the intervention, and drops a total
of 19.8 points overall.
Because the one-way ANOVA is a single test, it is not known which mean
differences are accounting for the significant finding (B1-A, B2-A, and/or
B2-B1). The Tukey multiple-comparison test accomplishes this. The Tukey
test compensates for the increase in the chance of a Type I error by adjusting
the p values. The results indicate all pairs are significantly different from one
another. By examining the means, we can conclude that Gloria’s depression
improved during the intervention phase and improved further once the
intensity of the intervention increased.
Tests of Type I Error When Autocorrelation Is High
As previously stated, prior to conducting tests of Type I error, we
recommend testing for autocorrelation in both phases. As an example of a
method for dealing with the issue of autocorrelation when comparing phases,
we look more closely at Jenny’s crying behavior.
Begin by using the ABrf2() function to assess for autocorrelation in both
the baseline and intervention phases by entering the following into the
Console after opening and attaching Jenny’s file:
>ABrf2(cry,pcry, “A”)
>ABrf2(cry,pcry, “B”)

The resulting rf2 values are 0.302 (sig = .425) and 1.019 (sig = .00),
respectively. The results indicate that the data for the intervention are highly
autocorrelated, so it would be advisable to see if transforming the data for
this phase reduces or eliminates the issue. Before deciding how to deal with
this, check whether the data trend by entering the following in the Console:
>ABregress(cry, pcry, “A”, “B”)

Since there is also a significant trend in both the baseline and intervention, it
is advisable to use differencing to transform the data.
The steps for accomplishing this are in the following table:
Command Purpose
>diffchart(cry,pcry, “B”) Runs first difference transformation on
intervention data.
Select Y when prompted in the Console. This saves the transformed data in the place
of your choosing.
Enter “tjennyb” as a file name and save it in a location Saves transformed intervention data.
you will remember.
>Getcsv() Open transformed data tjennyb.
>attach(ssd) Attaches data.
>ABrf2(diff, phase, “B”) Runs autocorrelation on transformed
intervention data.

NOTE: When you transform data, the behavior variable is renamed diff, and
the phase variable is renamed phase.
The output from the ABrf2() on the transformed data shows that the rf2 =
0.209 and sig = .434. It appears as if transforming the data was helpful in
decreasing the autocorrelation. In order to proceed we now need to transform
the baseline data using the same method we used for transforming the
intervention data and test the transformed data for autocorrelation. The steps
for accomplishing this are shown in the next table:
Command Purpose
>Getcsv() Open jennyab.csv data to transform the
baseline.
>attach(ssd) Attaches data.
>diffchart(cry,pcry, “A”) Runs first difference transformation on
baseline data.
Select Y when prompted in the Console. This saves the transformed data in the place
of your choosing.
Enter “tjennya” as a file name and save it in a location Saves transformed baseline data.
you will remember.
>Getcsv() Open transformed data tjennya.
>attach(ssd) Attaches data.
>ABrf2(diff, phase, “A”) Runs autocorrelation on transformed data.

The transformation of the baseline data shows that the rf2 = –0.617; however,
that autocorrelation that exists is not significant (sig of rf2 = .109).
We now merge the two transformed datasets and use this to examine
whether there are significant differences between baseline and intervention.
To combine the two datasets, enter the commands in the following table:
Command Purpose
>Append() Open transformed baseline data first: tjennya. Then open the transformed intervention
file tjennyb. Now save the combined file as tjennyab.
>Getcsv() Open the tjennyab.csv file.
>attach(ssd) Attaches data.

NOTE: When you use the Append() function, the behavior variables remain
named diff, and the phase variables remain named phase.
Because there seems to be no significant problem of autocorrelation in
either phase with the transformed data, we can proceed with the t test after
the baseline and intervention phase data are merged since we have already
determined that there is not a trend in either the baseline or intervention
phases.
In order to conduct the t test on the transformed dataset, enter the
following command in the Console:
>ABttest(diff, phase, “A”, “B”)
The results of the t test on the transformed data are presented in Figures 5.22
and 5.23 (see p. 119).
Figure 5.22 Console output results of t-test comparing differenced crying across phases
Figure 5.23 Visual comparison of mean differenced crying across phases

When looking at the test to examine the equality of variances, we see that
they are significantly different (p < .05), so we look at the t-test results for
unequal variances. When we look at these results, we note p = .4. As a
result, we cannot reject the null hypothesis that the difference in the means
between the baseline and intervention is greater than zero. Therefore, we
would have to conclude that, statistically, the intervention did not have an
impact on Jenny’s crying behavior, although it appears as if change is going
in the right direction.

Conclusion
In this chapter, we covered the use of a number of hyothesis tests to test for
Type I error between phases. SPC charts have the advantage of testing for
Type I error while producing graphs that are easily understood. We also
covered other statistical tests, including proportion/frequency, chi-square, the
CDC, the t test, and ANOVA. While these tests exist, it is important to
consider the conditions under which it is appropriate to use each of these,
and we recommend using the decision trees in Appendix C to help you select
the test of Type I error that is most suitable for your data.
 In single-subject research, we recommend using hypothesis tests in
conjunction with effect sizes to holistically assess your work with clients.
While statistical tests alone are sometimes used to examine change in all
types of evaluations, using a combination of both will likely lead to better
practical decision-making.
Finally, a method for transforming data and merging datasets was
discussed. Once new datasets are created using the Append() function, they
can be tested for Type I error using the techniques discussed in the chapter.

Chapter Exercise
Answer each of the following questions to determine whether there are
statistically significant differences between Brenda’s baseline and
intervention phase oppositional behavior:
1. Are there issues of autocorrelation in either the baseline or intervention phases? To make your
justification, provide both the rf2 and the significance of the rf2 for each phase.
2. Is there a significant trend that needs to be addressed in either phase? Justify your answer.
3. Conduct an appropriate hypothesis test with justification for why you are using a particular test.
Is there a significant difference in Brenda’s behavior after the introduction of the intervention?
4. What is the effect of the intervention? Choose an appropriate effect size with justification on
why you are utilizing that type of descriptive statistic. Interpret your findings.
5. Assume you are working with Brenda. Based on your findings, what practice decisions would
you make? Justify your rationale with the results of your evaluation.
6
Analyzing Group Data

Introduction
Macgowan (2008) pointed out the importance of measuring performance in
evidence-based group work (Macgowan, 2008, 2012). A set of functions
exists in SSD for R for analyzing measures of group behaviors.
This chapter discusses how to enter data on a group target behavior into
Excel or any other program that can export data to the .csv format. It also
explains how to use the SSD for R functions to analyze group outcomes.

Entering Group Data

Except for a few extra steps, entering data about groups is very similar to the
method discussed in Chapter 1. The steps to do this are presented next:
1. Open Excel or whatever program you are using to create your .csv file.
2. On the first row (labeled “1”), enter the names of your variables across the columns, beginning
with column A.

You will need to create several variables in Excel: (a) a group behavior
variable that corresponds to the group behavior that you are measuring; (b) a
phase variable that corresponds to the group behavior variable (the phase
variable will indicate in which phase, baseline or intervention, the
measurement occurred); (c) a time unit variable in which the behavior
occurred (this could represent the group session in which the measurement
was taken, e.g., “pweek”); and (d) a group member variable. This variable
will indicate which member of the group is being measured for each
occurrence.
 In order to do this systematically, we recommend giving the group
behavior variable a meaningful name; the phase variable name should be
similar to the behavior variable name. For example, if you are measuring the
degree of group mutual aid, your behavior variable could be named “muaid”
and your phase variable could be named “pmuaid.” In the example presented
in this chapter, we used pweek as our time unit variable to denote that the
groups were conducted weekly. Finally, we used “member” to represent a
group participant. Starting in row 2, begin entering your data for each
occurrence the behavior is measured.
IMPORTANT NOTE: When phases change, you will need to enter “NA”
into the row between a change in phases for each variable.
In Figure 6.1, note that we have entered baseline and intervention data for
a mutual aid group. Note that there are eight group members (labeled as
member 1 through 8 in column D) within each week (labeled pweek in
column B) displayed in Figure 6.1. The data are entered in order of the time
unit as this is the order in which data for this group were collected.
Figure 6.1 Example for entering group data

You can also track group members’ individual behaviors in the same
spreadsheet where you are tracking your group behavior variable. These
behaviors are typically not the same as the behavior that is measured for the
entire group, and specific behaviors measured can differ for each group
member. Note in Figure 6.1 the eight behavior variables m1, m2, . . . m8 and
the eight phase variables pm1, pm2, . . . pm8. The column labeled m1 is a
behavior variable for the first group member, and pm1 is a phase variable for
the first group member. All eight members have individual behaviors
recorded. The data were entered using the same steps discussed in Chapter 1.
 3. Once your data are entered into Excel, you will need to save it as a “CSV (Comma delimited)”
file or “CSV (Comma Separated Values)” in your ssddata directory. To do this, choose SAVE
AS and choose a name for your file. Do NOT click SAVE, but instead select one of the CSV
options from the drop-down menu for SAVE AS TYPE or FORMAT. After you finish this, you
should click SAVE and close Excel. You may receive several warnings, but you can accept all
of these by selecting CONTINUE.
4. Once you enter your data into Excel, you can import it into SSD for R. First, remember to load
the SSD for R package by typing require(SSDforR) in the Console and pressing the
<RETURN> key. You can now begin your analysis using the Getcsv() function. Once the file is
open, type attach(ssd).

Now type listnames() to obtain a list of variables in the file.

Analyzing Group Data

Many of the same issues discussed in previous chapters with regard to
understanding the baseline, using descriptive statistics, and comparing
phases are also relevant when analyzing group data.
An example file social skills group.csv is used throughout this chapter to
illustrate how to analyze group data. This dataset contains a group measure
of the degree of mutual aid, with higher scores indicating a higher degree of
providing/using mutual aid within the group. An individual behavior, the
degree of self-esteem experienced by the individuals in the group, is also
measured using a self-anchoring scale ranging from 1 (least) to 10 (highest).
To begin, use Getcsv() to open the file. Once the file is open, type
attach(ssd) <ENTER> followed by names(ssd) <ENTER>. The list of
variables in the order in which they were entered into Excel is shown in
Figure 6.2 (see p. 124) and will be displayed in the Console.
Figure 6.2 Variable list for social skills group

The variable muaid is the group behavior variable, pweek is the time unit
variable, pmuaid is the phase variable denoting whether the measurement
was made during the baseline or intervention, and member is a number
representing the group member being measured. The columns labeled m1
through m8 are the self-esteem measures for each group member. The phase
variable associated with m1 is “pm1,” m2 is associated with “pm2,” and so
on. As previously stated, these represent a target behavior (i.e., self-esteem)
and phase for each of the individual group members. There are 10 weeks of
baseline data and 10 weeks of intervention data.
The first step in analyzing your data is to describe each of the phases. The
typical conventions of single-subject research that use visual analysis and
descriptive statistics (e.g., mean, median, quantiles) are used to accomplish
this; however, because of the nature of group data, some of the functions
designed for individual measures do not adequately describe the data, and
functions especially designed for groups are used in their place.
Descriptive Statistics
It is appropriate to describe the behavior by a time unit variable that, in our
example, is weeks. Let’s begin by looking at the group behavior by week
more closely. A box plot can be created as follows on muaid by pweek by
using the following function:
>ABdescrip(muaid, pweek)

To make this a bit more descriptive, a line based on the median for the
baseline can be added using the following function:
>Gmedian(muaid,pmuaid,“A”)

The ABlineD(muaid) can then be used to place a vertical dashed line

between phases at the x-ordinate of 10.5 because the baseline phase changes
to the intervention after the 10th occurrence. The ABtext() function can then
be used to label the phases. Figure 6.3 (see p. 125) displays this annotated
box plot.

Figure 6.3 Boxplot of mutual aid group with baseline median displayed

The values for the mean, median, standard deviation, quantiles, and other
descriptive statistics by week for mutual aid appear in the Console and in
Figure 6.4 (see p. 125).
By examining Figures 6.3 and 6.4, we see data with considerable variation
in both phases, but with more in the baseline. The minimum value is 32, and
the maximum is 79, both of which occurred during the baseline. We also
note that the median value for each week during the intervention is above the
median for the baseline.

Figure 6.4 Descriptive statistics for mutual aid group

Each of the statistics shown in the Console is covered in detail in Chapter

3, and should be interpreted in the same manner with the understanding that
what is displayed in Figure 6.4 is weekly data and not phase data. For
example, the median (Md) is the value for which 50% of the observations
fall above and below. The median can be used to express the typical value in
a given phase. In this example, the median level of group mutual aid is
compared over a 10-week period. Similarly, the other measures of central
tendency and variation are calculated within each week. They all tend to
reflect a great deal of variation between weeks. For example, for Week 3 the
mean level of mutual aid is 52.00 with a standard deviation of 15.520
compared to Week 19 with a mean of 60.375 and a standard deviation of
5.423. Note that beginning in Week 7, the median degree of group mutual
aid begins to increase above the baseline median. This is important to note
because it occurs prior to the intervention commencing.

Comparing Phases
The ABdescrip() function can also be utilized to compare descriptive
statistics between phases. In our example, we accomplished this by entering
the following command in the Console:
>ABdescrip(muaid, pmuaid)

Figure 6.5 displays a box plot comparing the A (Weeks 1 through 10) and B
(Weeks 11 through 20) phases. The box plot displays a higher median during
the intervention phase. However, consider the results of Figure 6.3, which
shows that the degree of mutual aid had begun to increase during Week 7,
prior to the intervention being introduced during Week 11. This example
shows the usefulness of displaying group data by week, which provides
greater detail than overall phase data, and its use should always be
considered.
Figure 6.5 Boxplot comparing mutual aid between phases

Figure 6.6 shows the output for the phase data displayed on the Console.
From this, we observe that there was a mean increase in mutual aid from
baseline to intervention, and, as displayed by the standard deviations,
variation decreased during the intervention.
Figure 6.6 Descriptive statistics for mutual aid by phase

Autocorrelation and Trend

As discussed in Chapters 3 and 5, statistical tests rely on various
assumptions about the nature of the data being analyzed. Regardless of
whether individual or group data are measured, when these assumptions are
not met, incorrect conclusions about calculated values can be made, such as
deciding that observed differences between phases are significant when they,
in fact, are not (i.e., Type I error). On the other hand, we may erroneously
fail to detect a difference between phases when those differences do exist
(i.e., Type II error). As with individual data, it is important to consider the
presence of autocorrelation or trends in any phase before deciding on how to
compare phases.
For our group data example, enter the following command into the
Console to test for baseline lag-1 autocorrelation:
>GABrf2(muaid,pmuaid,pweek,“A”)

Note that the time interval variable pweek is included in this group
command. This is because there are multiple measures per week (i.e., the
eight members); therefore, the unit of analysis is the mean of members per
week (in consultation with Huitema, 2013).
Figure 6.7 presented in the graph window to the bottom right displays a
red dot representing the mean for each of the 10 baseline weeks. The results
of the baseline autocorrelation test are shown in Figure 6.8 (see p. 129).
Figure 6.7 Graphical output from GABrf2() function for mutual aid baseline example
Figure 6.8 Test for autocorrelation and trend for mutual aid in baseline

The rf2 is a measure of the degree of autocorrelation. Regardless of

direction, positive or negative, the higher the rf2 the more influential the
autocorrelation will be. The rf2 in this case is close to zero, with a value of
0.092. The “sig of rf2” indicates the chances of making a Type I error with
regard to the value of the rf2. Because this value is above 0.05 in our
example (sig of rf2 = .846), it is not statistically significant. As a result, we
can conclude these data are not autocorrelated.
Figure 6.7 displays a moderate trend in the data, with the blue regression
line displaying the slope; however, note that the individual data points are
not clustered tightly around the regression line. In addition to the graph,
additional output is produced in the Console under “regression,” as
displayed in Figure 6.8. These values show that the degree of change can be
quantified by the estimate for x1 of 1.1553, which is the slope of the
regression line. This can be interpreted as follows: For every one-unit
increase in time (in this case, a week), there is a 1.1553 increase in group
mutual aid. The column labeled “t value” is the calculated value for the
statistical significance of the slope and constant. The last column, labeled
“Pr(>|t|)” is the probability of making a Type I error (i.e., concluding that the
coefficient for the slope is greater than 0). Because the probability of the
coefficient is less than the commonly accepted threshold of .05, the slope is
considered statistically significant. This also demonstrates that even though a
trend exists in the data, the data are not autocorrelated, as explained
previously.
Similarly, the following command is needed to test for baseline lag-1
autocorrelation for the intervention phase, the results of which are displayed
in Figures 6.9 and 6.10 (see p. 131):
>GABrf2(muaid,pmuaid,pweek,“B”)

Figure 6.10 (see p. 131) presents the statistical results of the analysis. The rf2
for the intervention phase is 0.291. Because the sig of rf2 is above 0.05 in our
example (sig = .537), it is not statistically significant. As a result, we can
conclude the intervention data are not autocorrelated. Figure 6.9 displays a
lack of a trend in the data, and the coefficient of –0.1477 [Pr(>|t|) > 0.05]
indicates little change over time. Despite an insignificant p value, however,
we may still want to consider the fit of the data around the regression line
because, again, significance may be hard to achieve, particularly with small
sample sizes.
Figure 6.9 Graphical output from GABrf2() function for mutual aid intervention example
Figure 6.10 Test for autocorrelation and trend for mutual aid in intervention

If the data are autocorrelated or you are assuming autocorrelation, you

could attempt to transform the phase data in order to reduce the effect of
autocorrelation. If the data have a high degree of variation, using a moving
average could smooth them and reduce the impact of autocorrelation. The
moving average is just the plot of the mean of two adjacent observations:
([point 1 + point 2]/2). Alternatively, if the data are autocorrelated and have
a trend, attempting to transform the data using differencing is recommended.
For more information on using the moving average or differencing functions
in SSD for R, refer to Chapter 3.

Using Effect Size to Describe Group Change

Thus far we have described visual techniques combined with measures of
central tendency and variation to describe change in behavior across phases.
Effect size quantifies the amount of change between phases. Measuring
effect size has become a common method for assessing the degree of change
between phases as it gives a sense of practical, or clinical, significance
(Ferguson, 2009; J. G. Orme, 1991). When effect size is used to describe
data, it has advantages over tests of statistical significance in that the focus is
on the magnitude of change between phases rather than whether differences
are statistically significant (Kromrey & Foster-Johnson, 1996). An additional
advantage is that effect size is also relatively easy to interpret. SSD for R
calculates two forms of effect size that are appropriate for group data: ES
and d-index.
Entering the following command in the Console will produce the output
displayed in Figure 6.11 (see p. 133).
Figure 6.11 Effect size calculation for mutual aid group

>Effectsize(muaid,pmuaid,“A”,“B”)

Notice the request to choose one of the following: (s)ave, (a)ppend, or

(n)either results? (s/a or n); select n. The other two choices are used to store
the effect size results to a file for use in meta-analysis.
All effect sizes ES (0.51867), d-index (0. 59932), and Hedge’s g
(0.59647) indicate a small degree of change, as noted in the interpretation of
effect sizes toward the top of Figure 6.11. The percentages of change for the
ES and d-index are 19.8% and –22.55%, respectively.

Tests of Statistical Significance

SSD for R contains two tests of statistical significance that can be applied to
group data. There is a function that is designed specifically to do a t test
using group data, which differs from the t test described Chapter 5. The
proportion/frequency test, as described in Chapter 5, can also be used with
group data. Additionally, three statistical process control (SPC) charts, the
X-bar R chart (X̄-R chart) and two R charts are also appropriate for the
analysis of group data.

t-test
The t test compares the differences in the means between phases. The null
hypothesis is that the difference in means between the phases is zero (i.e.,
there is no change). If the t test indicates statistically significant differences
between the phases, that is, p ≤ .05, then we accept the alternative that the
mean differences are greater than zero. The t test, however, is nondirectional,
meaning that significant p values could indicate either improvement or
deterioration in the observed behavior. If higher values are desired, the
difference should be positive, and if lower values are desired, the difference
should be negative.
As mentioned in Chapter 5, there are some caveats for the use of the t test.
It should not be used if there is a trend in the data or if either phase has
problematic autocorrelation. If the data are autocorrelated, a transformation
of the data can be used to try to remove it. If the problem of autocorrelation
is resolved, the transformed data can then be analyzed using the t test.
The t test also assumes equal variation between the phases. If the
variances are unequal, then a t test for unequal variances should be used. The
group t-test function in SSD for R provides a test for the differences in
variances between phases, which can be interpreted to determine which form
of the t test should be used. Additionally, output for the group t-test function
includes results for both equal and unequal variances, and the appropriate
test results can then be analyzed.
Looking at our example of the mutual aid group, we recall that the
baseline data are not autocorrelated but do have a moderate trend. Although
there is a trend in the baseline, for illustrative purposes, enter the following
command in the Console to get the output that is displayed in Figure 6.12:
Figure 6.12 t-test comparing baseline and intervention for mutual aid group

>GABttest(muaid,pmuaid,pweek,“A”,“B”)
Note that the time interval variable pweek is included in the command. This
is because there are multiple measures per week (i.e., the eight group
members); therefore, the unit of analysis is the mean of the members for
each week.
There are three tests displayed in the Console panel to the left of the
graphical output: a t test for equal variances at the top of the Console, an F
test to compare the variances in the middle, and a t test for unequal variances
at the bottom. To begin, look at the p value for the F test to compare the
variances. Because the p value for equality variances in our example is less
than .05 (.006367), we reject the null hypothesis that the variances are equal.
Therefore, we use the t-test results for unequal variances to determine
whether there are statistically significant differences between the baseline
and intervention phases.
Looking at the results for unequal variances, located at the bottom of the
Console, we reject the null hypothesis because the p value is less than .05
(.0006419). Therefore, we accept the alternative that the mean difference
between the phases is greater than zero. We also observe that the mean went
up from 54.8125 in the baseline to 60.7125 in the intervention.

Proportion/Frequency
If you are using task completion data or any binary outcome, the
ABbinomial() function can be utilized to test for statistical significance
between phases. This command is discussed at length in Chapter 5. No
adjustments are needed to this function for group data.

Using an X̄-R Chart to Visualize Group Change

The X̄-R chart can be used with group data because there are multiple
observations per sample (J. G. Orme & Cox, 2001). In the case of our mutual
aid group example, the group data were collected on a weekly basis, which
resulted in 20 samples (i.e., 10 weeks of baseline plus 10 weeks of
intervention), with eight observations (i.e., group members) per sample.
Using the mean of these samples increases the accuracy of our measure,
which are the individual data points displayed in the X̄-R chart (Mohammed
& Worthington, 2012; Orme & Cox, 2001).
To create the X̄-R chart displayed in Figure 6.13 (see p. 136), enter the
SSD for R commands presented in the following table in the Console:
Figure 6.13 X-R chart for mutual aid grou

Command Purpose
>XRchart(muaid,pweek,2,“week”,“Level of >Mutual Creates graph using two standard
Aid”,“Group Mutual Aid”) deviations
>ABlines(muaid) Adds line between phases
>ABtext(“57.763”) Enters mean text
>ABtext(“64.605”) Enters Uband (upper band) text
>ABtext(“50.920”) Enters Lband (lower band) text
>SPClegend() Creates a legend

NOTE: Once you create a legend, you will no longer be able to alter the
graph. Therefore, we add this as a last step in creating SPC charts. Be sure
that your graph looks exactly as you would like before adding the legend.
In this example, the desired zone would be above the upper band (64.605)
since the group’s goal is to increase the level of mutual aid. Although 9 of 10
weeks during the intervention are above the mean, none of the intervention
observations is in the desired zone.
In terms of practice decisions, the group facilitator would probably want
to continue measuring this process because it seems that the intervention is
going in the desired direction. To increase confidence, continued tracking
would be recommended.
Using R Charts to Assess Variability of Group Data
As discussed in Chapter 4, R charts can be used to detect changes in
variation over time. As these SPC charts use samples, they are appropriate
for use with group data. In the example of our mutual aid group, it is most
appropriate to use the standard deviation form of the R chart, which is
illustrated in Figure 6.14 (see p. 137). To create this chart with two standard
deviations, the following command was entered into the Console:
>Rchartsd(muaid, pweek, 2, “week”, “standard deviation”, “Group Mutual Aid”)

The graph was further annotated by noting values for the mean, upper, and
lower bands using the ABtext() function and adding a legend in the same
way that the X̄-R chart was annotated, above, with the SPClegend()
function.

Figure 6.14 R-chart using standard deviations for mutual aid group

In looking at the output, we see that the data became more stable in the
second half of the baseline and throughout the intervention. Combined with
our previous analysis, this could help us conclude that, over time, the
group’s level of mutual aid was moving in the desired direction, and that, as
a whole, the overall level of mutual aid was beginning to become less
variable between group members.

Using P Charts to Assess Binary Outcomes in Group Data

If you recall, P charts can be used to assess binary outcomes between phases.
P charts work well with group data as they use a grouping variable (e.g.,
weeks). While our sample dataset does not contain any binary outcomes, this
SPC chart can be used with group data in the way that it was described in
Chapter 4.

Individual Change
As mentioned, the individual change of group members can also be
measured with SSD for R. In our hypothetical example, the group members
used a self-anchoring scale weekly to rate their self-esteem. The scale ranged
from a low of 1 to a high of 10. The measured score for each group member
is contained in the columns labeled m1 to m8. The corresponding phase
variables for these are contained in columns labeled pm1 to pm8.
Measured scores for these individuals can be presented in multiple line
graphs that are displayed simultaneously. This is helpful since all the
members’ progress can be visualized at once. An example of this type of
graph is presented in Figure 6.15.
Figure 6.15 Individual self-esteem scores for group members

The first step in creating the graph is to set up the graphic environment
using the plotnum() function, which provides parameters for the desired
number of rows and columns. We want to arrange our eight graphs as four
rows of two columns, so we enter the following command into the Console:
>plotnum(4,2)

Now, each graph can be added to the graphic environment using the
ABplotm() commands seen in the next table. Note that this function is
different from the ABplot() function:
NOTE: Once you create an individual graph, you will able to alter and
annotate the graph until the next graph is added. Be sure that your graph
looks exactly as you would like it before adding another. If the following
error occurs, “figure margins too large,” you will have to increase the size of
the Plots pane. If this does not solve the issue, reset the Plots pane by issuing
the following R command in the Console: dev.off(). If you do this, be aware
that any graphs currently in the pane will be removed.
Command Purpose
>ABplotm(m1,pm1,“weeks”,“self esteem”,“member 1”) Creates graph
>ABlines(m1) Adds line between phases
>ABplotm(m2,pm2,“weeks”,“self esteem”,“member 2”) Creates graph
>ABlines(m2) Adds line between phases
>ABplotm(m3,pm3,“weeks”,“self esteem”,“member 3”) Creates graph
>ABlines(m3) Adds line between phases
>ABplotm(m4,pm4,“weeks”,“self esteem”,“member 4”) Creates graph
>ABlines(m4) Adds line between phases
>ABplotm(m5,pm5,“weeks”,“self esteem”,“member 5”) Creates graph
>ABlines(m5) Adds line between phases
>ABplotm(m6,pm6,“weeks”,“self esteem”,“member 6”) Creates graph
>ABlines(m6) Adds line between phases
>ABplotm(m7,pm7,“weeks”,“self esteem”,“member 7”) Creates graph
>ABlines(m7) Adds line between phases
>ABplotm(m8,pm8,“weeks”,“self esteem”,“member”) Creates graph
>ABlines(m8) Adds line between phases

These individual data can be analyzed in the same manner we discussed in

previous chapters. For example, you can obtain descriptive statistics for each
member using the ABdescrip() command. If the necessary assumptions are
met, you can calculate effect size and tests of statistical significance and can
create any of the graphs discussed previously. All the assumptions discussed
in the previous chapters need to be considered, such as baseline stability,
autocorrelation, trend, and standard deviation.

Conclusion
In this chapter, we discussed how you could use SSD for R to assess group
functioning. There are functions to compare change over time or between
phases. Using the group social interaction example data, you learned the
functions necessary to calculate these statistics. These functions generate
supporting graphs, box plots, SPC charts, and individual line graphs. SSD for
R also provides methods for computing group autocorrelation, group effect
size, and group t tests.

Chapter Exercises
Assignment 6.1—Create a Group Spreadsheet and Import It to
RStudio
A social worker serving mental health clients is assigned a group of four
clients having difficulty adapting to transitional housing because of the
quality of their activities of living (ADL) skills, such as cleaning, laundry,
maintaining their residences, shopping for food, preparing meals, paying
bills, and taking public transportation. The worker decides to start a group
and provide an intervention to teach the clients ADL skills. She measures
their ADL skills at each session with a scale ranging from 1 (poor) to 10
(excellent).
For this assignment, you will create a spreadsheet with the group’s
baseline and intervention data. BE SURE TO STORE THIS
SPREADSHEET IN A PLACE YOU CAN ACCESS LATER, as you will
use it for future homework assignments. The data you need for this is in the
following table:
ADL Score Week of Service Phase (A or B) Member ID
4 1 A 1
4 1 A 2
6 1 A 3
3 1 A 4
2 2 A 1
3 2 A 2
4 2 A 3
2 2 A 4
7 3 A 1
3 3 A 2
2 3 A 3
3 3 A 4
3 4 A 1
4 4 A 2
5 4 A 3
3 4 A 4
2 5 A 1
2 5 A 2
4 5 A 3
6 5 A 4
NA NA NA NA
5 6 B 1
7 6 B 2
8 6 B 3
4 6 B 4
6 7 B 1
5 7 B 2
7 7 B 3
6 7 B 4
6 8 B 1
7 8 B 2
6 8 B 3
5 8 B 4
7 9 B 1
6 9 B 2
7 9 B 3
6 9 B 4
6 10 B 1
7 10 B 2
8 10 B 3
6 10 B 4
7 11 B 1
6 11 B 2
5 11 B 3
 ADL Score Week of Service Phase (A or B) Member ID
7 11 B 4
8 12 B 1
6 12 B 2
7 12 B 3
5 12 B 4
6 13 B 1
7 13 B 2
7 13 B 3
7 13 B 4
8 14 B 1
6 14 B 2
7 14 B 3
6 14 B 4
7 15 B 1
7 15 B 2
8 15 B 3
8 15 B 4

Assignment 6.2—Visual Analysis

1. Create an X̄-R chart, labeling it appropriately. In the main title, include YOUR initials. For
example, my main title could read, “Group Memory—CA.”
2. Annotate the line graph to put a vertical line between phases.
3. Label the baseline phase “A” and the intervention phase “B.”
4. Export the graph to a Word document and add your answer to the following question: Based on
the line graph alone, do you think that the group intervention with the ADL group is having the
desired effect? Why or why not?

Assignment 6.3—Hypothesis Testing and Effect Sizes

Answer each of the following questions to determine whether there are
significant differences between the group’s baseline and intervention phase
memory behavior:
1. Are there issues of autocorrelation in either the baseline or intervention phases? To make your
justification, provide both the rf2 and the significance of the rf2 for each phase.
2. Is there a significant trend that needs to be addressed in either phase? Justify your answer.
3. Conduct an appropriate hypothesis test with justification for why you are using a particular test.
Is there a significant difference in memory after the introduction of the intervention?
4. What is the effect of the intervention? Choose an appropriate effect size with justification for
why you are utilizing that type of descriptive statistic. Interpret your findings.
5. Assume you are responsible for this group. Based on your findings, what practice decisions
would you make? Justify your rationale with the results of your evaluation.
7
Meta-Analysis in Single-Subject Evaluation
Research

Introduction to Meta-Analysis
Throughout this book, we have discussed methods for evaluating the degree
to which change can be observed and interpreted with the introduction of
intervention(s) to help remediate some identified problem(s). The findings
from these evaluations should inform your work with individual client
systems, but sometimes, we might want to know more.
Oftentimes our work involves many client systems with similar presenting
problems, and we might use similar intervention techniques to help address
those problems. We also know that some client systems respond better to
intervention than others. So, we might wonder, given clients with a
particular type of presenting problem (e.g., depression), how much change
could we expect to see if we use a particular intervention (e.g., cognitive
behavioral therapy)? And, does that change more generally make meaningful
differences in the lives of our clients?
Meta-analytic techniques can be used to aggregate evaluation results
across studies. In the case of single-subject research designs, we could
combine findings from evaluations with 5, 10, or 20 clients to determine, on
average, how effective an intervention is. We can combine effect sizes
discussed in Chapter 4 to report the direction and magnitude of change
across all those clients (i.e., individual studies). This is a more complex and
sophisticated way of understanding differences across studies than reporting
those changes qualitatively or simply reporting the individual effect sizes for
each study.
Meta-analysis is, in essence, a study of studies. It is most frequently used
to understand average effects of interventions across research in which group
designs are used, and it is predicated on the assumption that individual
studies are independent of one another. Therefore, it is only appropriate to
use meta-analytic techniques with different clients. That is, do not conduct a
meta-analysis with the same client across different outcomes.
In this chapter, we talk more about why meta-analysis is important to
consider in single-subject research, and then we demonstrate how to do this
using SSD for R functions.

How Do I Conduct a Meta-Analysis?

As with other aspects of this text, we do not go into great detail on how to
conduct a thorough meta-analysis as many excellent books and resources are
available on this in great depth. Some of these resources are listed in
Appendix D because we find them particularly relevant to single-subject
case designs.
In short, however, meta-analysis is not unlike methods used in other
research. For example, you should first start with an answerable research
question. Instead of generating data by conducting original research, meta-
analysis is dependent on conducting a systematic search for studies that can
help answer the research question and often entails a specification of
inclusion and exclusion criteria. That is, the researcher will formalize a
rationale for why studies should be included or left out of a meta-analysis.
Then, data have to be extracted from studies included in the meta-
analysis. This could include either the raw data used to compute effect sizes
or the effect sizes themselves along with other variables that could be
moderators. From there, data can be analyzed and interpreted based on
output from these analyses.

Why Is Meta-Analysis Important in Single-Subject

Research?
Single-subject research designs are wonderful ways to consider change
across individual client systems, but they are not generally considered
particularly rigorous research designs in the scheme of published research.
This is because it is impossible to generalize findings of these studies to
larger populations because we only have a sample size of one in these
designs. Meta-analysis, then, enables researchers to combine these findings
across studies to begin to build generalizability to larger groups.
The What Works Clearinghouse Single-Case Design Standards
recommend combining a minimum of five single-subject design studies in
order to begin building a basis for causal inference. That documentation
provides additional information on how to conduct rigorous single-subject
research studies for publication, and we recommend close reading of that
resource for more information.

Conducting Meta-Analysis With SSD for R

SSD for R provides two options for conducting a meta-analysis. The first
uses traditional effect size calculations, while the second uses a non-overlap
effect size method. The specific method you use is dependent on both the
type of data you have and whether you want to include control variables in
your analysis. Since non-overlap methods have been traditionally used in
single-subject research designs, this method may be preferable if you do not
need to include a control variable in your evaluation. Additionally, the
method we present here, the mean NAP, can be used in cases where data are
more highly autocorrelated and/or significantly trending (Manolov et al.,
2011).

Mean Non-overlap of All Pairs

Using the average Non-overlap of All Pairs (NAP) across studies is perhaps
the most flexible method to use in the meta-analysis of single-subject
research studies. Since trending and serial dependence are not as problematic
with the NAP, this would be the preferred method to use if these were issues
in any of the studies to be included in your meta-analysis. Therefore, we
present this method first.
To begin, you will need to create a file that saves the NAP results for all
cases you want included in your analysis. This was described in detail in
Chapter 4.
As an example, we have created a file Student NAP Interventions. The file
contains information about special education students’ ability to remain on
task prior to and during an intervention, which included reconfiguring
classroom space and praising on-task behavior. Data were recorded for seven
students, and effect sizes, as noted by NAP values, were calculated and
saved in this file. Load these data by entering the following in the Console
and navigating to the file location:
>Getcsv()

Be sure to attach the file before proceeding.

To calculate the average effect size, enter the following in the Console to
get the results displayed in Figures 7.1 and 7.2 (see p. 146):
> meanNAP(napES, Label, “Effect Size for 7 Special Ed Students”)

Figure 7.1 Console output results of meanNAP() function

Figure 7.2 Visual output results of meanNAP() function

When entering this command, you will always use napES and Label as
the first two parameters you enter unless you manually change these in your
.csv file prior to loading it. Figure 7.1 demonstrates that the mean NAP for
all seven students is 0.79 (standard deviation [sd] = 0.19). Looking at the
key below the computed output provides information on the interpretation of
the calculated values. An average effect of 0.79 is understood to be a
moderately effective degree of change.
Figure 7.2 provides additional information. The output in this graph orders
studies (in this case students) from lowest effect size at the base of the y-axis
to highest at the top of y-axis. We can see here that Students 3 and 5 did not
really improve since their NAP scores fell below the 0.67 mark, which is the
threshold for minimal effectiveness. Their NAP scores were much less than
the remainder of students included in the study, all of whose effect sizes
were greater than 0.8. This is valuable information, and if we were the
teachers in this classroom, we might wonder in what ways Students 3 and 5
might differ from those who saw greater effects. This could provide us the
beginnings of insights on how we might want to change interventions based
on characteristics of more than one study participant.
Using Traditional Effect Size Statistics to Conduct Meta-
Analysis
In Chapter 4, we discussed traditional effect sizes such as ES, d-index, and
Hedge’s g. If these effect sizes can be used across studies to be included in a
meta-analysis, it may be desirable to do so. Reasons for this, according to
Shadish and colleagues (2014), include being able to integrate these studies
more easily into comparisons with between-subject designs and making use
of well-known meta-analytic tools, such as forest plots and publication bias
analyses. Additionally, tools provided in SSD for R allow for the
introduction of a moderating variable in this analysis.
To illustrate, consider a child welfare agency that provides services to
parents whose children are at risk for foster care placement due to
allegations of neglect. The agency has recently developed an in-home
intervention for which they want to receive a grant from the state. The state,
however, will only consider funding the program if it can be shown that
participation is effective at keeping families intact. The agency tracked 14
clients who participated in the program using observational measures of
parenting before and during the intervention. Effect sizes have been
computed for all 14 using the techniques described in Chapter 4, and they
have been placed in the file Parenting intervention meta.csv that comes with
this text.
Use the Getcsv() and attach(ssd) functions to load these data. If you click
on the spreadsheet icon, you will see that there have been two additional
columns added to what was developed by saving and appending the effect
size file you learned how to create previously. The first is a variable called
Support, which assess the level of social support available to the parent on a
scale ranging from 1 (with no social supports available to the parent) to a
high of 10 (the parent has many supports available to them). Additionally,
the variable Single denotes whether or not the parent has a live-in partner.
Zero indicates having a partner, and one indicates being single.
Enter the following in the Console to view the mean effect size across all
studies, the results of which are displayed in Figures 7.3 and 7.4 (see p. 148):
>meanES(ES, Label, “Mean Parenting Skills”)
Figure 7.3 Console output results of meanES() function

Figure 7.4 Visual output results of meanES() function

When entering this command, you will always use ES and Label as the
first two parameters you enter unless you manually change these in your .csv
file prior to loading it.
The output shown in these figures looks similar, although not identical, to
what you saw previously. The mean effect calculated is 1.33 (sd = 0.99) and,
according to the key provided underneath this output, demonstrates a
medium effect since this value is between 0.87 and 2.67. Additionally, we
see that, on average, parents saw a 40.77% improvement in parenting skill
scores after the introduction of the intervention.
Figure 7.4 shows the effect sizes from smallest to largest. Here, we see
that Clients 1, 4, and 13 showed (in decreasing order) the most
improvement, while Clients 12, 8, 11, 3, and 14 showed (in decreasing
order) the least. Evidence of overall improvement in parenting skills makes a
compelling case for potential funders who want some assurances that
services they are paying for will actually help clients keep their children at
home.
To continue our analysis, you will want to conduct a more robust
investigation by using the metareg() function as it weights each study based
on the variance of the ES within each study. This function assumes that the
variance between studies is greater than zero, and that studies included in the
meta-analysis are a sample drawn from a larger universe. Therefore, we
assume that differences observed between studies is not due to sampling
error, but from real differences in the population. The information gleaned
from this function will be more substantial and useful. Begin by entering the
following in the Console:
> metareg(ES, V)

The output for this function is displayed in Figures 7.5 (see p. 149) and
7.6 (see p. 150). Figure 7.5 shows that the summary ES for these 14 studies
is 0.94 and is displayed as the estimate. The interpretation of this estimate is
based on Cohen’s guidelines, with 0.2 indicating a small effect, 0.5
indicating a medium effect, and 0.8 indicating a large effect. A summary ES
of 0.94 indicates nearly a 1-SD improvement from baseline to intervention,
which is considered a large effect. The standard error for the ES, the z
statistic, as well as the upper and lower bounds of the 95% confidence
interval are the additional information provided on this line. The p that is
displayed is the p value for the summary statistic. In this case, p < .05, so it
is statistically significant.
Figure 7.5 Console output results of metareg() function

The output from the metareg() function also gives us information about
the heterogeneity, or the variability, of the studies included in this analysis.
The QEp value is the Q statistic that tells us whether the heterogeneity
between the studies is statistically significant.
The forest plot displayed in Figure 7.6 (see p. 150) illustrates this
heterogeneity, or variability, across these studies. Here, we can see that some
studies show a very small effect that is not statistically significant (Studies
14 and 3, for instance), while others show much larger effects that are
statistically significant but may have wide confidence intervals (Studies 1
and 9, for example). Notice that some studies’ ESs (such as Studies 3 and
14) are represented by larger squares, while others are represented by
smaller squares. The size of the squares denotes the precision of the ES, with
larger samples having more precision. The diamond at the bottom of the
forest plot denotes the summary effect (0.940), and the length of it represents
the 95% confidence interval, ranging from 0.439 to 1.442.

Figure 7.6 Visual output results of metareg() function

Since heterogeneity is significant in this case, we should look more

closely at the output to determine what might be contributing to the observed
variation. Heterogeneity may arise due to sampling error within individual
studies. Alternatively, there may be true variation in ESs between studies.
 Previously in our analysis, we used the Q statistic to determine that there
was heterogeneity in our meta-analysis, but we can use I2 to determine both
the extent and source of that variation. Values close to 0 indicate that all
heterogeneity is associated with within-study sampling error, while values of
100 indicate that all heterogeneity is associated with true between-study
differences. The estimate for I2 provides us with information about the
magnitude of the observed heterogeneity, with 25% representing low
heterogeneity, 50% indicating a moderate amount, and 75% representing a
high level of heterogeneity. The I2 for our example is 48.4951%, indicating a
moderate amount of heterogeneity.
The 95% confidence interval provided for I2 may help us understand the
source of the observed heterogeneity. In this example, the 95% confidence
interval is wide, ranging from 0 to 82.3374. This illustrates a wide degree of
uncertainty in the source of the variation, perhaps due to the small number of
studies included in the meta-analysis.
Since there is significant moderate heterogeneity for unknown reasons in
this example, it might be helpful to consider moderator variables. We can
consider the influence of social support since, theoretically, parents with
good social support often have an easier job parenting. Enter the following
into the Console to get the output displayed in Figures 7.7 (see p. 151) and
7.8 (see p. 151):
Figure 7.7 Console output results of metaregi() function for support as a moderator
Figure 7.8 Visual output results of metaregi() function for support as a moderator

>metaregi(ES, Support, V)

In the output shown in Figure 7.7, notice that there is a new line, mods,
that includes any moderator variables added to the model. The intercept in
this case, however, is not significant since p = .353 and is greater than the
.05 threshold. We therefore ignore this coefficient and discard the model and
conclude, in this case, that social support does not improve our
understanding of the ESs of the included studies.
Since heterogeneity has not changed or improved, we can think about the
variable Single since single parents often have a more difficult time
parenting. The output displayed in Figures 7.9 and 7.10 (see p. 153) is result
of the following being entered in the Console:
Figure 7.9 Console output results of metaregi() function for single as a moderator
Figure 7.10 Visual output results of metaregi() function for single as a moderator

>metaregi(ES, Single, V)

At first glance, we see that the intercept is significant, with p being

displayed as 0, so we can consider this model more closely. Recall that
Single is a dichotomous variable, so the interpretation of the output with this
moderator is slightly different from that of continuous variables. The value
of the intercept is 2.213, which is the ES when Single = 0 (i.e., being
partnered). The p value for the moderator is also significant, so we can
understand that as the slope. That is, when going from 0 to 1 (i.e., partnered
to single), there is a 1.848 reduction in parenting skills. Therefore, the effect
size for single parents is 2.213 –1.848 = 0.365. It appears as if improvement
in parenting skills is moderated by parents being single or partnered.
Also note that heterogeneity is no longer significant since QEp is no
longer significant, and I2 is estimated at 0. Therefore, the inclusion of Single
in the meta-analysis improves our understanding of the ES of the parenting
intervention.

Conclusion
Being able to combine single-subject research designs is particularly useful
in both practice and research settings. From a practice perspective, it
provides more information about what practitioners might be able to expect
from clients as a result of implementing specific interventions. In the
examples provided, we were able to identify clients who responded to
interventions better than others. We also found that being single negatively
impacted parenting skills across studies. In this way, we might be able to
begin identifying the types of clients who might respond more successfully
to specific ways of working than others. From a research perspective, being
able to conduct meta-analyses of single-subject research designs adds to the
knowledge base of what is known about particular interventions by
providing rigor that is not attainable by analyzing one study at a time.
In this chapter, we covered two methods for analyzing multiple single-
subject research studies. The meanNAP() function is an extension of the
nonparametric NAP method and can be used when data in studies trend or
may have problems of serial dependency. The metareg() and metaregi()
functions make use of traditional effect sizes that are used in group research
designs; however, they cannot be used in cases where serial dependency is
problematic or data in individual studies trend. The advantages to these
functions, though, are that they are more easily understood in the larger
research community, they can be used in larger meta-analyses to build
research evidence around the effectiveness of specific interventions, and
moderator variables can be included.
8
Using RMarkdown to Present Your Findings

Introduction
In this chapter you will learn how to utilize RMarkdown to present SSD for R
findings in a well-ordered and reproducible manner. RMarkdown is a plain
text formatting syntax that makes writing research reports simple. The
language provides a simple syntax that formats text such as headers, lists,
boldface, and so on. This language is popular, and you will find many apps
that are compatible with it. For example, combined with other packages, like
SSD for R, users can easily create tables and graphics to present their
research findings. Another important feature of this markdown language is
that it will make your findings reproducible in that all of your files are
connected. Thus, if there are changes to your data, rerunning the analysis is
simple. As Baumer and Udwin (2015) suggested, an RMarkdown document
links computation, output, and written analysis to enhance transparency,
clarity, and ease to reproduce the research. Furthermore, sharing data is only
a click away (Baumer & Udwin, 2015).
As mentioned, RMarkdown can save and execute SSD for R code and
create high-quality reports from a single file. Doing so provides greater
insight into your results and clearly displays what you did to analyze your
data. Furthermore, RMarkdown makes it possible for other
clinicians/researchers to replicate findings. RMarkdown is a straightforward
language that allows you to create documents with headings, text, images,
margins, and more to enhance the communication of the findings produced
by SSD for R. Documents created in RMarkdown can be saved in different
file formats, such as doc, PDF, or HTML.

Getting Started
If you are using a stand-alone version of RStudio, the installation of a LaTex
package is required to create professional-looking files in PDF format.
LaTeX is already installed on RStudio Cloud.
To create a Word file, the installation of MS Word is required. For a
Windows computer, MiKTeX is recommended and can be installed from the
following URL: http://miktex.org/download. For a Mac, MacTex 2013 + is
recommended and can be installed from the following URL:
https://tug.org/mactex/mactex-download.html. When you create a report
utilizing RMarkdown, it will automatically convert the report to a PDF or
Word file without your direct interaction with the LaTeX software you
installed. Once again, LaTeX is already installed on RStudio Cloud.
The next step is to open RStudio or RStudio Cloud and type the following
commands in the Console one at a time to install the necessary packages to
create a report in PDF or Word format:
>install.packages(“rmarkdown”)
>install.packages(“knitr”)
>install.packages(“kableExtra”)

Creating an RMarkdown Document

One of the primary reasons for developing an RMarkdown file is its ability
to be reproduced by other clinicians and researchers. As a result, interactive
commands such as ABlines() in SSD for R will not run. These commands
have been replaced and begin with the prefix RM; for example, RMlines() is
substituted for ABlines() to draw a vertical line between phases. The table
that follows displays a list of all interactive commands that have substitutes
for writing RMarkdown scripts. The best way to obtain the coordinates for
the commands that follow is to create the graph in RStudio and then apply
them to the commands in the chunk.
Interactive RMarkdown Syntax Example
Command Substitute
ABarrow() RMarrow() RMarrow(X1,Y1,X2,Y2) X1 y1 From >RMarrow
coordinates of arrow. X2 y2 To coordinates of (5.9,5,9.5,5.1)
arrow.
Gline() RMGline() RMGline(Y) From Y coordinate to draw >RMGline(5)
horizontal line
ABlines() RMlines() RMlines(behavior, X) Behavior variable to >RMlines (yell,15.5)
draw vertical line From X coordinate to draw
vertical line
ABstat RMstat() RMstat(behavior, phase,“statistic”,X) X start >RMstat (yell,pyell,“A”,
of horizontal line “mean”,.1)
ABtext() RMtext() RMtext(“text”,X,Y) From X and Y >RMtext
coordinates to draw text (“argument”,2.5,5)
Getcsv() ssd<-read() ssd <- read.csv(“file”) file data file name >ssd <-
read.csv(“Jennyab.csv”)

Once this is done, you are ready to work through an example. The
example will focus mostly on creating an MS Word document. The first step
is to create an RMarkdown file (.Rmd). In RStudio, select File/New File/R
Markdown, and the menu in Figure 8.1 will be displayed. Select Create
Empty Document and a blank untitled RMarkdown document will appear in
the top left pane.
Figure 8.1 Markdown menu

The first step is to create a YAML where you can configure your
RMarkdown file. This portion contains the header, author, title, and type of
format you desire. This portion of the file starts and ends with ---. Here is an
example:
---
title: “SSD for R”
author:
- Charles Auerbach
- Wendy Zeitlin
subtitle: An R Package for Analyzing Single-Subject Data
output:
word_document: default
 pdf_document: default
---

In this case, there are two authors, and each author’s name begins with a
hyphen (-) separated by a space. Also, note that under the output section,
both Word and PDF documents are entered, allowing for substituting
between document types.
Below the YAML is where different code chunks begin, which contain
various R functions from SSD for R. For example, the code chunk below will
open the jennyab.csv file, which can be downloaded from the authors’
website (https://www.ssdanalysis.com) or the publisher’s website. The data
file must be placed in the same directory as the .rmd file.
```{r openfile,include=FALSE}
knitr::opts_chunk$set(comment = NA)
require(SSDforR,warn.conflicts=F, quietly=T)
ssd <- read.csv(“Jennyab.csv”)
attach(ssd)
```

Each code chunk begins and ends with three ``` (grave accents). Inside the
{} curly brackets, the r indicates the chunk will include R code. Each chunk
requires a unique name, and in this example, the chunk was given the name
openfile. After the comma (,) options can be added for the chunk. Using the
option include = FALSE will suppress code from being displayed for the
chunk in the output document. Using Echo = FALSE will suppress only
code, and results = “hide” will suppress only results; however, we didn’t
include those options in this example as we want those displayed. The code
require(SSDforR,warn.conflicts = F, quietly = T) loads the SSD for R
package. You can use this syntax in all RMarkdown code that requires the
use of a package by replacing the package name, in this case SSD for R, with
another package name. The knitr::opts_chunk$set(comment = NA) will
improve the look of output by removing comment marks (#). The ssd <-
read.csv(“Jennyab.csv”) syntax places the jenny.csv file into the vector ssd
so it can be accessed using the syntax attach(ssd). This code can be altered
to analyze other datasets by simply changing the name of the file within the
quotation marks.
Let’s add a chunk of code to produce an ABplot() of Jenny’s yelling
behavior to the script. Be sure to add four line returns after the ending ```. To
create a page break in your document between the title page and the graph
output, a \pagebreak command is placed in the middle of the white space
between the chunks. Below is the code to create the graph using
RMarkdown. Once again, the best way to obtain the coordinates for the
commands that follow is to create the graph in RStudio and then apply them
to the commands in the chunk.
```{r ABplot,echo=F,results=“hide”}
ABplot(yell,pyell,“Days”,“Amount”,“Jenny”)
RMlines(yell,15.5)
RMtext(“A”,5,6)
RMtext(“B”,25,6)
RMtext(“argument”,2.5,5)
RMarrow(5.9,5,9.5,5.1)
RMstat(yell,pyell,“A”,“mean”,.1)
```

Once again, the chunk begins with a title and the following options: echo =
F and results=“hide”. The echo is set to “F”, and the result options are set
to “hide”, resulting in the prevention of messages and results from being
printed in the document. The SSD for RABplot() produced will appear in the
output document. The function that follows will annotate the graph. The best
way to obtain the coordinates for the commands that follow is to create the
graph in RStudio and then apply them to the commands in the chunk. The
RMlines() function will place a vertical line on the x ordinate 15.5; the
RMtext() will place an “A” at the x ordinate 5 and the y ordinate 6; the
second RMtext() function will place a “B” at the x ordinate 25 and the y
ordinate 6. The final RMtext() function will place the text “argument” at the
x ordinate 2.5 and y ordinate 5; the RMarrow() will draw an arrow from the
text “argument” to the 10th observation. Finally, the RMstat() function
places a mean line in the baseline.
To create the document, we can knit the .rmd file to create the MS Word
document to view the graph. When you knit a file, your RMarkdown
document is automatically saved. In order to do so, Word must be installed
on your computer. As displayed in Figure 8.2, in the top left pane click on
the down arrow next to knit and select MS Word.
Figure 8.2 Knit menu for RMarkdown

The output will be displayed in a Word file. If the file is read only, use
File/Save As to save the file under a new name. If you are running RStudio
Cloud, when the output is created, a message will appear to downloaded the
file to your local machine. You will then be able to modify the file. Figure
8.3 displays the result that will be produced in a Word document.
Figure 8.3 ABplot() of Jenny’s crying behavior

You can also add text to annotate your findings. For example, the
following can be included by pressing return four times after the ending ```.
**Introduction**
The School Base Support team recommended that Jenny receive Cognitive Behavioral
Therapy (CBT) at the Child Help Center to reduce or eliminate maladaptive and
inappropriate behaviors. The primary goals were the development of self-control and
problem-solving strategies. The acquisition and internalization of these skills provide the
means for the child to regulate her behavior.
**Visual Analysis**

For data analysis, visual analysis was used. Results indicated that the trend of changes in
the behavior scores for the client was descending in the interventions phase (B) and
indicated improvement. Figure 1 illustrates the results from the AB design.

\pagebreak

Note the double asterisks (*) around the subtitle to bold the subtitles
“Introduction” and “Visual Analysis.” You add a page break by placing the
\pagebreak command as a line after the final sentence of the section’s
explanation.
Each phase was tested for autocorrelation by inserting the chunks that
follow:
 ```{r autoA,echo=T}
ABrf2(yell,pyell,“A”)
```

```{r autoB,echo=T}
ABrf2(yell,pyell,“B”)
```

Notice that once again, the option echo was set to T to display the SSD for R
code in the document. To view the results of the autoA and autoB chunks,
you will need to knit the .rmd file. The results are shown in Figures 8.4 and
8.5 in the resulting output from the autocorrelation chunk for the baseline
autocorrelation. Figures 8.6 (see p. 162) and 8.7 (see p. 162) display the
results for the intervention autocorrelation.
Figure 8.4 Baseline ABrf2() results
Figure 8.5 Baseline ABrf2() Graph
Figure 8.6 Intervention ABrf2() results
Figure 8.7 Intervention ABrf2() graph

The next chunk will produce the results of a t test comparing the A and B
phases. Note the title of the chunk is ttest and the option echo = T is added
to include the syntax in the output. The syntax options(scipen = 999) is
added to remove scientific notation from the output. The SSD for R syntax
ABttest(yell,pyell,“A”,“B”) performs the t test.
```{r ttest,echo=T}
options(scipen = 999)
ABttest(yell,pyell,“A”,“B”)
title(main= “Figure 2: Mean Differences”)
```

Once again, knit the .rmd file as displayed in Figure 8.2, and changes
made to the file will be automatically saved. Figures 8.8 (see p. 163) and 8.9
(see p. 164) display the result of the t test. Note because echo = T was
entered as an option, the syntax is also included in the document. The title()
function is included to provide a title for the mean bar plot produced by the
ABttest() function. The title() function adds the title inside the quotations to
the bar graph produced with the ABttest() function. Notice that “t” in t-test
is between asterisks, which will italicize the characters between them.
Figure 8.8 Results from ABttest() function
Figure 8.9 Graph from ABttest() function

The final chunk will add the results of an effect size. Once again knit the
.rmd file as displayed in Figure 8.2, and changes made to the file will be
automatically saved. Figure 8.7 displays the result of the effect size.
```{r ES,echo=T}
Effectsize(yell,pyell,“A”,“B”)
```

Again, you can also add text to annotate your findings and conclusions. For
example, the following can be included by adding four white spaces after the
ending ```.
**Statistical Analysis**
Both phases were tested for autocorrelation. The findings indicate that the degree of
autocorrelation was small enough in each phase (rf2 = 0.267 and rf2 = 0.383, respectively)
to conduct a *t*-test for statistically significant differences between phases. A Welch
correction was used to compensate for unequal variances. The mean number of yelling
episodes per day decreased from 2.8 to .94 from baseline to intervention. Figure 2 shows
these differences, which are statistically significant (*t* = 5.26, *p* = <001). The Cohen’s
*d* of 1.95 indicates a moderate degree of change between phases.

**Conclusion**
The School Based support team is proposing removing the intervention to form an A-B-A
design, which is sometimes referred to as an experimental removal of intervention design.
The design is experimental because it consists of testing whether the introduction of an
intervention was likely to have caused changes in the target behavior as a result of the
child’s integration of problem-solving skills to regulate her behavior.

To view the results of ES chunk along with the additional narrative, knit the
.rmd file once again. The results are shown in Figure 8.10 (see p. 165). If the
MS Word document is read only, to edit the document, it will need to be
saved under a new file name.

Figure 8.10 Reults from Effetsize() function

Conclusion
In conclusion, this chapter provided an introduction to developing
RMarkdown documents to effectively present your SSD for R findings.
There are a number of helpful texts and websites on using RMarkdown in
Appendix D. These examples provide a more in-depth explanation on
formatting text, tables, and figures.

Chapter Exercise
Assignment 8.1—Create an RMarkdown Report on Evaluating
Your Practice With Brenda
For this assignment, you will use Brenda’s baseline and intervention data to
create a final report describing your practice evaluation. You will utilize
RMarkdown to present SSD for R findings of the Brenda case example in a
well-ordered and reproducible manner. You can modify the RMarkdown
example included with the text to create your document.
1. Create a YAML header.
2. Create a chunk load SSDforR, open the Brenda baseline/intervention file and attach it.
3. Create a chunk to recreate an ABplot() you created in Assignment 4.1.
4. Add an Introduction
5. Create a chunk to display your findings from Assignment 4.3.
6. Create a chunk to display the results on the effect size you utilized in Assignment 4.3.
7. Create a chunk to display the appropriate hypothesis test used in Assignment 5.3. Also add a
description of your statistical analysis.
8. Finally add a conclusion.
9
Building Support for Practice Research

Introduction
We recognize that practice research does not occur in a vacuum, and those
desiring to do this type of evaluation work in many different types of
settings. Some practitioners work in solo practices, while others may be
employed in large multisite, multiprogram organizations with thousands of
employees and clients at any given time. Therefore, the notion of engaging
in practice research and the ability to do so should be considered within the
context in which one works, as flexibility, resources, and work demands can
vary from organization to organization.
In this chapter, we discuss common issues that arise when organizations,
whether small or large, wish to engage in practice research, and we provide
some suggestions for addressing these.
It has been widely recognized that engaging in practice research is an
essential component of good and responsible practice in many disciplines
(Shaw, 2011). While we discussed the benefits of this type of research
previously, practice research findings can also be used at an administrative
level to improve existing programs and plan for future services. Engaging
in practice-based research can also be used to develop organizational
policies and procedures and aid in decision-making about the allocation of
resources (Preskill & Boyle, 2008).
We believe, then, that the importance of practice research cannot be
overstated. Despite this, many wishing to engage in practice research often
face challenges, including issues related to the practical ability to conduct
research, often referred to as research capacity.

Research Capacity and Research Culture

Building research capacity “attempts to enhance individual and collective
research expertise and hence capacities to contribute to improving the
quality and quantity of . . . research” (Christie & Menter, 2009, p. 5).
Another description of research capacity includes the ability and motivation
to routinely conduct evaluations of practice (Preskill & Boyle, 2008).
Factors that contribute to engagement in practice research include abilities
and desire. Therefore, our discussion on building support for practice
research in this chapter focuses on both capacity and organizational culture.

Barriers to Conducting Research

It is necessary to consider the barriers facing those wishing to engage in
practice research, as identifying and addressing these will help
organizations build research capacity. Barriers to conducting research
within organizations generally fall into two broad categories: organizational
challenges and individual factors. Organizational challenges include
administrative issues and the role of practitioners within the organization.
Individual factors contributing to challenges with conducting practice
research include skills and the nature of the profession to which the
practitioner belongs.

Organizational Challenges to Conducting Research

Frequently, there may be an overall desire to engage in practice research,
but there are challenges that must first be overcome. For example, most
human service organizations wishing to evaluate their practices are not
primarily research institutions; they are service oriented. In these
organizations, conducting research may not be a priority, there may be
negative attitudes about research in general, there may not be administrative
support for quality improvement, or there may be an overall lack of support
for research activities (Beddoe, 2011; Epstein & Blumenfield, 2012; Fouché
& Lunt, 2010; McCrystal & Wilson, 2009). One reason given for this lack
of focus on research activities is management being unsure of what to do
with research findings and the value these activities could bring to its core
mission (Beddoe, 2011).
In many cases, there may be desire among individual practitioners or
administrators to evaluate practice, but an absence of management support
to engage in these activities can lead to substantial roadblocks. For instance,
organizations focused on service delivery may not have the resources
readily available to conduct research. This could include difficulty
accessing professional development that supports practice research,
computers, technology support, and/or scholarly literature (Beddoe, 2011;
Carman & Fredericks, 2010; Lunt et al., 2008).
In addition to administrative issues, another type of organizational barrier
to conducting practice research revolves around the role of practitioners
within organizations. A primary concern for practitioners wishing to engage
in practice research is time. Many practitioners working in human services
have heavy caseloads with barely enough time to complete the requirements
of their daily work. Insufficient time to conduct practice research is the
reason most often cited for not doing so (Beddoe, 2011; Carman &
Fredericks, 2010; Edwards & Cable, 2009; Epstein & Blumenfield, 2012;
Fouché & Lunt, 2010; Lunt et al., 2008; McCrystal & Wilson, 2009).

Individual Barriers to Conducting Research

The other most commonly cited reasons for not engaging in practice
research are related to ability. In many cases, practitioners lack the
confidence, knowledge, or skills to engage in practice research (Beddoe,
2011; Carman & Fredericks, 2010; Edwards & Cable, 2009; Epstein &
Blumenfield, 2012; Fouché & Lunt, 2010; Lunt et al., 2008; McCrystal &
Wilson, 2009). Specifically, practitioners are often unsure of what to
measure or how to measure client problems or their progress (Carman &
Fredericks, 2010). They may have difficulty managing data and using
statistical software (Carman & Fredericks, 2010). Finally, many
practitioners are not sure how to utilize or effectively disseminate research
findings in general (Edwards & Cable, 2009).
A related systemic problem surrounding practitioner research skills has to
do with the nature of the profession to which practitioners belong. For
example, while social workers have deemed the development of research
skills important, there has traditionally been little focus on this within the
discipline itself (Edwards & Cable, 2009; Lunt et al., 2008; McCrystal &
Wilson, 2009; J. Orme & Powell, 2008). This has been exhibited by
insufficient attention paid to developing research skills in professional
programs (Epstein & Blumenfield, 2012; Lunt et al., 2008; McCrystal &
Wilson, 2009). This is mirrored in other human services professions as well.
 Additionally, there are few postgraduate educational opportunities across
professions to teach practitioners about research methods and building
research skills (J. Orme & Powell, 2008).
Despite these challenges, and as we referenced in our introductory
chapter, we believe that there will be an increasing demand for practice
research in the future. The trends toward utilizing evidence-based practices
and increasing accountability in healthcare and behavioral healthcare are
indicative of the need to build organizational environments conducive to
engaging in practice research within all human services organizations.

Developing Research Capacity and a Culture of Research

Within Organizations
Action items have been developed and implemented in order to make
research activities accessible to practitioners and others wishing to conduct
research in human services organizations. These have been aimed at
creating environments that encourage research among practitioners while
building the infrastructure necessary to actually engage in these activities.
Action items fall into two categories—those related to infrastructure and
demand and those related to skill building. While we can categorize these, it
should be noted that to implement all of these activities requires concerted
effort and administrative support.

Developing the Infrastructure and Demand for Research

Building the infrastructure and demand for practice research includes
activities related to developing overall infrastructure, supporting research
activities, valuing research as an important activity at the organizational
level, and producing useful research outputs. While this may initially seem
burdensome, it lays the foundation for longer term success; when
administration values and supports research, evidences suggests more
effective research capacity building within organizations (Arnold, 2006;
Beddoe, 2011; McCrystal & Wilson, 2009; J. Orme & Powell, 2008).
A number of suggestions have been made to assist with building research
infrastructure specifically. All of these require thoughtful planning, the
consideration of integrating practice research into the core activities of the
organization, and intentionally building research capacity (Christie &
Menter, 2009). One recommendation is to conduct an evaluation of the
organization’s existing research capacity. Baizerman, Compton, and Hueftle
(2002) have developed a checklist to assist in this activity by examining
existing processes, practices, and professional capabilities.
Based on this, organizations should consider developing or identifying
practice research models that are congruent with the aims of the
organization and resources available to them (Carman & Fredericks, 2010).
Arnold (2006) further supported this by suggesting that evaluation activities
should be considered in human services organizations when programs are
initially designed. To assist with this, organizations should think about
developing a holistic approach to building research capacity.
Volkov and King (2007) have developed a checklist to help organizations
develop research capacity. This checklist, which is referenced in Appendix
D, looks at the context, including organizational culture in which research
activities occur, mechanisms within the organization that are conducive to
building research capacity, and identification of resources necessary to
conduct practice research.
Most obviously, building infrastructure should also include providing
access to computers, databases, and software to analyze data (Edwards &
Cable, 2009; J. Orme & Powell, 2008; Rock et al., 1993).
Another factor to consider in developing organizational research
infrastructure involves human resources. For instance, some suggestions
have included identifying a champion within the organization who
understands both practice and research in order to help integrate practice
research into the core activities of the organization. These activities require
incremental steps that include being responsive to practitioner needs and
work demands (Edwards & Cable, 2009; J. Orme & Powell, 2008; Rock et
al., 1993). In fact, practitioners should be included in regular meetings and
decision-making at all stages of capacity building in order to develop an
integration plan that can pragmatically be absorbed into the culture of the
organization (Rock et al., 1993).
Additionally, in order to build infrastructure, organizations should
consider supporting doctoral studies and including research and
development requirements in job descriptions (J. Orme & Powell, 2008).
Not only do these human resource considerations help build staff
capabilities, but also they include the expectation that employees will be
involved in practice research (J. Orme & Powell, 2008). To build more
support for practice research, evaluation findings should be shared with
administrators and practitioners in a manner that is both meaningful and
useful (Arnold, 2006). Finally, conducting practice research regularly helps
incorporate these activities into the awareness and day-to-day work of
practitioners (Arnold, 2006).
Building a culture of research occurs within organizations, but it is also
influenced by the community stakeholder groups. For example, the notion
of valuing the process of research and its outputs in any one organization is
supported by others within the same sector, board members, funders, and
policymakers (Carman & Fredericks, 2010; Lunt et al., 2008; Shaw, 2011).
Therefore, stakeholder groups can work together to build demand for
practice research activities at the administrative level. This can result in
administrators encouraging practice research and working to increase the
visibility of research activities within organizations (McCrystal & Wilson,
2009).

Building Research Skills Within Organizations

In general, building research capacity can be enhanced by the development
of pragmatic research skills. This includes developing individual skills and
building collaborative relationships that are supportive of research
activities.
Practice research skills can be developed among workers. As previously
stated, two ways of doing this are by developing formal policy decisions
that include supporting doctoral studies and including research skills in job
descriptions (J. Orme & Powell, 2008; Volkov & King, 2007). The goal of
these should include developing research knowledge and skills in the
majority of practitioners (Arnold, 2006).
Organizations should also create and support learning opportunities for
employees. This could include providing research training for service users
and front-line practitioners (J. Orme & Powell, 2008). Other ways of
supporting employees’ learning about research methods and skills involves
encouraging employees to participate in research internships and to attend
online or in-person trainings, workshops, and seminars; and providing
access to reading material (Christie & Menter, 2009; Preskill & Boyle,
2008). Finally, organizations can encourage employees who have the
capacity to become practice research leaders to participate in train-the-
trainer learning opportunities in which participants are not only taught
relevant research knowledge, but also are provided the resources and skills
to pass these along to others (Edwards & Cable, 2009).
Another way to provide organizations with usable research skills is
through collaborative activities. This can involve something as simple as
contracting with experienced practice research evaluators to obtain
technical assistance and to regularly work with practitioners (Epstein &
Blumenfield, 2012; Preskill & Boyle, 2008). One possible venue for
accessing these skills is by actively working to create university-
organization partnerships not to only conduct evaluations, but also to help
develop practice environments that are conducive to ongoing practice
research within service organizations (McCrystal & Wilson, 2009; J. Orme
& Powell, 2008). A suggestion provided by Shaw (2005) goes one step
further and directs universities to consider establishing centers of practice
research. In this way, universities can become known and easily identified
resources that allow organizations to build research capacity without the
burden of having to build their own research collaboratives.
However collaboration is included in practice research, work should
begin with teamwork between researchers and consumers of research to
determine what outputs are needed (Berger, 2010).
Regardless of how research knowledge is brought into organizations,
numerous recommendations have been developed to encourage continuing
skill development and ongoing research activities within the organization
itself. Rock and colleagues (1993) suggest that including practitioners in
research activities and decision-making not only improves attitudes about
the research process but also breaks down resistance to this across the
organization.
Research collaboration within organizations can take many forms. One of
the most commonly written about is research mentorships. In these types of
collaborations, more experienced researchers teach practitioners about
research methodology and skills by including them in meaningful projects
and helping them take responsibility for portions of the research (Christie &
Menter, 2009; Edwards & Cable, 2009; Preskill & Boyle, 2008). In this
way, one-on-one support can be provided that allows practitioners to gain
research skills incrementally (Arnold, 2006). Some variations of research
mentorship involve the creation of evaluation teams where more
experienced teams provide assistance to less experienced teams through
various projects (Carman & Fredericks, 2010). Another way to engage in
mentorship is to pair academic researchers with practitioners for co-learning
projects. In these, practitioners gain research skills while academics learn
about the pragmatics of organizational life (Fouché & Lunt, 2010). Finally,
Carman and Fredericks (2010) described research mentoring to involve the
creation of learning networks that integrate peer learning and the sharing of
best practices. In this way, mentorship is a reciprocal relationship where
practitioners and researchers participate in collaborative, mutually
beneficial activities over time.

Conclusion
The focus of this book has been on single-subject research design,
methodology, and analysis; however, in order to make this, or any type of
research, accessible to organizations and practitioners, a clear plan for
conducting practice research should be developed.
This chapter was written to give you the opportunity to consider the
context of your organization along with what may be required to include
practice research into your organization’s core activities. This chapter was
not meant, however, to be an exhaustive discussion of building research
capacity; rather, it was designed to provide readers a broad understanding of
factors that should be considered when attempting to implement research in
a practice environment.
Common obstacles to conducting research in practice settings discussed
in this chapter include administrative factors, work demands placed on
practitioners, the availability of research knowledge and skills, and the
research tradition of some professions. To address these, recommendations
have been developed to remediate these barriers. These involve building
support and demand for practice research by increasing its value to
stakeholders, the development and/or accessibility to research skills, and
providing the infrastructure necessary to conduct practice research.
In this chapter we discussed the importance of including practitioners in
the process of building research capacity. It should be emphasized that, in
order to increase the chances of success, capacity building must be
collaborative. All activities should include representation from all employee
groups that will either participate in the research process or will be
consumers of research. Organizational change of any type needs to consider
the needs of all involved (Choi & Ruona, 2011).
 With the publication of this book, we hope that we have begun to address
some of the barriers identified in this.
 APPENDIX A

ENTERING AND EDITING DATA DIRECTLY IN

R

ENTERING DATA DIRECTLY INTO SSD FOR R

Throughout the text, we are assuming that you are using Microsoft Excel or
a similar program to enter your data; however, there may be situations in
which these programs may not be available to you. Fortunately, there is way
to enter data directly into R for use with the SSD for R package. Before
beginning it is necessary to open RStudio. In the Console, type the following
command:
>require(SSDforR).

As an example, let’s track two client behaviors for a hypothetical client:

amount of crying and level of self-esteem. An A-B-B1 design is being
utilized. The A and B phases each have six observations, while the B1 phase
has eight observations.
Entering data in R has to be done in a specific format as displayed below.
>cry<-c(3, 4, 2, 5, 3, 4, NA, 2, 2, 3, 2, 1, 2, NA, 2, 2, 1, 2, 1, 0, 0, 0)
pcry<-c(“A”, “A”, “A”, “A”, “A”, “A”, NA, “B”, “B”, “B”, “B”, “B”, “B”, NA, “B1”,
“B1”, “B1”, “B1”, “B1”, “B1”, “B1”, “B1”)
esteem<-c(3, 4, 2, 5, 3, 4, NA, 2, 2, 3, 2, 1, 2, NA, 2, 2, 1, 2, 1, 0, 0, 0)
pesteem<-c(“A”, “A”, “A”, “A”, “A”, “A”, NA, “B”, “B”, “B”, “B”, “B”, “B”, NA, “B1”,
“B1”, “B1”, “B1”, “B1”, “B1”, “B1”, “B1”)

Note that the behavior variable cry is followed by a “<” (less than sign) and
a “–” (dash) which is used synonymously in R to an equal sign (=). Also
note the lower case “c” before the open parenthesis and that a comma
follows each observation entry. Note that for the phase variables, pcry and
pesteem the letters A, B, and B1 are enclosed in quotes. Finally, notice that
the same number of elements are entered for the behavior variables and their
corresponding phase variables. Additionally, the NA, inserted as described in
Chapter 1 to denote a change of phases, is located in the same position in the
behavior variable and its corresponding phase variable. For example, cry is
made up of a total of 22 elements, as is the corresponding phase variable,
pcry. We also note that phases change for both the behavior and phase
variables after six occurrences of the baseline, and six occurrences of the
first intervention phase. After you enter the close parenthesis, press the
<ENTER> key.
After the two behavior and two phase variables are entered, the following
command can be utilized to create a “data frame” that can be saved as a
“.cvs” file for later use:
>ssd<-data.frame(cry,pcry,esteem,pesteem)

To view the data frame, shown in Figure A.1 (see p. 176), type the following
command:
>ssd

Note how the “NA”s line up for each behavior variable and its
corresponding phase variable. Also notice that each element is numbered on
the far left with values ranging from the first (“1”) to the last (“22”).
Now the data frame can be saved as a.csv file by using the following
function:
>Savecsv()
Figure A.1 The ssd data frame with four variables.

Remove the data frame from memory using the following command:
>rm(list = ls())
The data you just saved can be opened using the Getcsv() function and then
attach(ssd). It can also be opened and modified using Excel.

USING THE R DATA EDITOR

After opening a dataset using Getcsv() and using attach(ssd), it can be
modified using the following R command:
>fix(ssd)

When you do this, the “R Data Editor” will open, which allows you to make
changes to your data in a spreadsheet-type format, as shown in Figure A.2
(see p. 177). After changing any data in the editor, remember to save it using
the Savecsv() function.
Figure A.2 The ssd data displayed in R Data Editor

Figure A.3 Adding a variable

IMPORTANT NOTE: This is an example of the Mac R Data Editor. The

Windows version of the editor looks slightly different.
Use the arrow keys to move left and right or up and down the editor. A
column can be added by moving the right arrow key past the last column,
“pesteem.” As shown in Figure A.3 (see p. 177) below, the column is labeled
“var5.”
Now double click on “var5” and a menu with attributes will appear. Select
“Change name,” as shown in Figure A.4 (see p. 178). Now change the name
of the variable to comment.
Now enter “went home sick” in the comment column and third row. The
results are shown in Figure A.5 (see p. 178) below.
Now type ssd in the Console to view your changes, displayed in Figure
A.6 (see p. 178).
You can now use the Savecsv() function to save your changes.
Figure A.4 Renaming a variable
Figure A.5 The comment variable added to the data set
Figure A.6 The ssd with the added comment variable
 APPENDIX B

SSD for R Quick Functions Guide

SSD for R Quick Functions Guide

Basic Functions
Open/Import.csv file
Description: This function opens “.csv” file created in Excel.
Command: Getcsv()
Ex: Getcsv()

List all functions

Description: List all functions available in SSD for R.
Command: SSDforR()
Ex: SSDforR()

List variables
Description: Lists all variables in active data frame.
Command: listnames()
Ex: listnames()

Save a data file

Description: Save a data file edited in SSD for R as “.csv” file.
Command: Savecsv()
Ex: Savecsv()

Graphing Functions
Create a line graph
Description: This function builds a simple line chart for a given
behavior across all phases. A space separates each phase.
Command: ABplot(behavior, phaseX, ABxlab, ABylab,
ABmain)
behavior: behavior variable
 phaseX: phase variable
ABxlab: label for x-axis; written between quotation marks
ABylab: label for y-axis; written between quotation marks
ABmain: main label for graph; written between quotation marks
Ex: ABplot(yell, pyell, “school days”, “yelling incidents”,
“Jenny’s Yelling”)

Create multiple line graphs in a single window

Description: This function builds a simple line chart for a given
behavior across all phases. A space separates each phase. This
function needs to be invoked for each graph separately.
plotnum() must precede use of this function.
Command: ABplotm(behavior, phaseX, ABxlab, ABylab,
ABmain)
behavior: behavior variable
phaseX: phase variable
ABxlab: label for x-axis; written between quotation marks
ABylab: label for y-axis; written between quotation marks
ABmain: main label for graph; written between quotation
marks
Ex: ABplotm(yell, pyell, “school days”, “yelling incidents”,
“Jenny’s Yelling”)

Sets the graphic environment when creating multiple line graphs

Description: This function specifies the number of rows and
columns to display in the graphics window when multiple line
graphs are to be built.
Command: plotnum(nr, nc)
nr: number of rows desired
nc: number of columns desired
Ex: plotnum(4,2)

Draw a solid line between phases on a graph

Description: This function enables the user to draw solid vertical
lines between phases on a graph. Once the function is
invoked, the user is prompted to accept the line or not.
Command: ABlines(behavior)
 behavior: behavior variable
Ex: ABlines(yell)

Draw a dashed line between phases on a graph

Description: This function enables the user to draw dashed
vertical lines between phases on a graph. Once the function is
invoked, the user is prompted to accept the line or not.
Command: ABlineD(behavior)
behavior: behavior variable
Ex: ABlineD(yell)

Label a graph
Description: This function enables the user to write text on a
graph. Users have three options for doing this: text with
neither subscripts nor superscripts, text with superscripts, and
text with subscripts. Commands for each and examples to
label a graph are displayed below. After the command is
invoked, users will be prompted to place the cursor where the
text is to begin. After text is placed, users will be prompted to
choose whether or not to accept the graph with the text.
Command: (for text with neither subscripts nor superscripts)
ABtext()
Text to be entered on graph must appear between quotation
marks
Ex: ABtext(“baseline”)
Command: (for text with superscripts)
ABtext(expression(text^superscript))
text: this is the text that is to be displayed on the graph
superscript: this is the actual superscript that is to be displayed
Ex: ABtext(expression(X^2))
Command: (for text with subscripts)
ABtext(expression(text[subscript]))
text: this is the text that is to be displayed on the graph
subscript: this is the actual subscript that is to be displayed
Ex: ABtext(expression(B[1]))

Draw an arrow on a graph

 Description: This function enables users to draw an arrow on a
graph. For example, an arrow can be drawn from a text label
of a critical event to a point on the graph. Once the function is
invoked, the user is prompted to accept the arrow or not.
Command: ABarrow()
Ex: ABarrow()

Adding statistical lines (mean, median, or sd) to an ABplot

Description: This function enables users to draw a line in a
phase representing the mean, median, or standard deviation.
Command: ABstat(behavior, phaseX, v, statX)
behavior: behavior variable
phaseX: phase variable
v: phase letter entered between quotes (e.g., “A”, “B”)
statX: mean, median, or sd written between quotation marks
(e.g., “mean”)
Ex: ABstat(yell, pyell, “A”, “mean”)

Adding trimmed mean line to an ABplot

Description: This function enables a user to add a line
representing the trimmed mean to any phase of an ABplot.
Command: Trimline(behavior, phaseX, v)
behavior: behavior variable
phaseX: phase variable
v: phase letter entered between quotes (e.g., “A”, “B”)
Ex: Trimline(yell,pyell,“B”)

Adding goal line to an ABplot

Description: This function enables a user to add a line
representing the goal line (a level of function set as a goal for
a client) to an ABplot.
Command: Gline(y)
y: y ordinate
Ex: Gline(6)

Adding standard deviation (SD) bands line an ABplot

Description: This function enables a user to add a line
representing the SD bands (±1, ±2, or ±3) to any phase of an
 ABplot.
Command: SDAband(behavior, phaseX, v,bandX)
behavior: behavior variable
phaseX: phase variable
v: phase letter entered between quotes (e.g., “A”, “B”)
bandX: SDordinate 1, 2, or 3
Ex: SDAband(yell,pyell,“A”,2)

Adding interquartile range (IQR) line to an ABplot.

Description: This function enables a user to add a line
representing the interquartile range to any phase of an
ABplot.
Command: IQRline(behavior, phaseX, v)
behavior: behavior variable
phaseX: phase variable
v: phase letter entered between quotes (e.g., “A”, “B”)
Ex: IQRline(yell,pyell,“A”)

Create a ±1 standard deviation (sd) band graph for a given phase

Description: This function builds a ±1 standard deviation band
graph for a given behavior based on a phase of the user’s
choice. A space separates each phase.
Command: sd1bandgraph(behavior, phaseX, v1, ABxlab,
ABylab, ABmain)
behavior: behavior variable
phaseX: phase variable
v1: phase letter entered between quotes (e.g., “A”, “B”)
ABxlab: label for x-axis; written between quotation marks
ABylab: label for y-axis; written between quotation marks
ABmain: main label for graph; written between quotation
marks
Ex: sd1bandgraph(yell, pyell, “A”, “school days”, “yelling
incidents”, “SD graph: Jenny’s Yelling”)

Create a ±1 standard deviation (sd) band graph across all phases

Description: This function builds a ±1 standard deviation band
graph for a given behavior across all phases. A space
 separates each phase.
Command: SD1(behavior, phaseX, v1, ABxlab, ABylab,
ABmain)
behavior: behavior variable
phaseX: phase variable
v1: phase letter entered between quotes (e.g., “A”, “B”)
ABxlab: label for x-axis; written between quotation marks
ABylab: label for y-axis; written between quotation marks
ABmain: main label for graph; written between quotation
marks
Ex: SD1(yell, pyell, “A”, “school days”, “yelling incidents”,
“SD graph: Jenny’s Yelling”)

Places legend at bottom of one standard deviation (SD1) band graph

Description: This function enables the user to place a legend on
a graph.
NOTE: Once this legend is in place, the graph can no longer be
altered.
Command: SD1legend()
Ex: SD1legend()

Create a ±2 standard deviation (sd) band graph for a given phase

Description: This function builds a ±2 standard deviation band
graph for a given behavior based on a phase of the user’s
choice. A space separates each phase.
Command: sd2bandgraph(behavior, phaseX, v1,ABxlab,
ABylab, ABmain)
behavior: behavior variable
phaseX: phase variable
v1: phase letter entered between quotes (e.g., “A”, “B”)
ABxlab: label for x-axis; written between quotation marks
ABylab: label for y-axis; written between quotation marks
ABmain: main label for graph; written between quotation
marks
Ex: sd2bandgraph(yell, pyell, “A”, “school days”, “yelling
incidents”, “SD graph: Jenny’s Yelling”)
Create a ±2 standard deviation (sd) band graph across all phases
Description: This function builds a ±2 standard deviation band
graph for a given behavior across all phases. A space
separates each phase.
Command: SD2(behavior, phaseX, v1,ABxlab, ABylab,
ABmain)
behavior: behavior variable
phaseX: phase variable
v1: phase letter entered between quotes (e.g., “A”, “B”)
ABxlab: label for x-axis; written between quotation marks
ABylab: label for y-axis; written between quotation marks
ABmain: main label for graph; written between quotation
marks
Ex: SD2(yell, pyell, “A”, “school days”, “yelling incidents”,
“SD graph: Jenny’s Yelling”)

Places legend at bottom of two standard deviation (SD2) band graph

Description: This function enables the user to place a legend on
the graph.
NOTE: Once this legend is in place, the graph can no longer be
altered.
Command: SD2legend()
Ex: SD2legend()

Create an interquartile (iqr) band graph for a single phase

Description: This function builds an iqr band graph for a given
behavior based on a phase of the user’s choice. The graph will
only be drawn for the phase selected. Statistical output in the
Console shows the interquartile bands.
Command: IQRbandgraph(behavior, phaseX, v1,ABxlab,
ABylab, ABmain)
behavior: behavior variable
phaseX: phase variable
v1: phase letter entered between quotes (e.g., “A”, “B”)
ABxlab: label for x-axis; written between quotation marks
ABylab: label for y-axis; written between quotation marks
 ABmain: main label for graph; written between quotation
marks
Ex: IQRbandgraph(yell, pyell, “A”, “school days”, “yelling
incidents”, “SD graph: Jenny’s Yelling”)

Create a interquartile (iqr) band graph through all phases

Description: This function builds an iqr band graph for a given
behavior based on a phase of the user’s choice. The graph will
be drawn for all phases. A space separates phases. Statistical
output in the Console shows the interquartile bands.
Command: ABiqr(behavior, phaseX, v1,ABxlab, ABylab,
ABmain)
behavior: behavior variable
phaseX: phase variable
v1: phase label entered between quotes (“A”,“B”,“B1”)
ABxlab: label for x-axis written between quotation marks
ABylab: label for y-axis written between quotation marks
ABmain: main label for graph between quotation marks
Ex: ABiqr(cry, pcry, “A”, “school days”, “yelling incidents”,
“SD graph: Jenny’s Crying”)

Places legend at bottom of IQRbandgraph or ABiqr band graph

Description: This function enables the user to place a legend on
a graph. NOTE: Once this legend is in place, the graph can no
longer be altered.
Command: IQRlegend()
Ex: IQRlegend()

Create a time series line graph

Description: This function builds a time series chart for a given
behavior across all phases. A space separates each phase.
There are no connecting dots.
Command: ABtsplot(behavior, phaseX, ABxlab, ABylab,
ABmain)
behavior: behavior variable
phaseX: phase variable
ABxlab: label for x-axis; written between quotation marks
 ABylab: label for y-axis; written between quotation marks
ABmain: main label for graph; written between quotation
marks
Ex: ABtsplot(yell, pyell, “school days”, “yelling incidents”,
“Jenny’s Yelling”)

Basic Statistical Analysis

Compute descriptive statistics for any phase

Description: This function produces descriptive statistics for all
phases. Statistics produced are mean, 10% trimmed mean,
median, standard deviation (sd), coefficient of variation (CV),
range, interquartile range, and quantiles. Graphical output for
this function is a box plot of data in each phase.
Command: ABdescrip(behavior, phaseX)
behavior: behavior variable
phaseX: phase variable on which test is based
Ex: ABdescrip(yell, pyell)

Ordinary least squares (OLS) regression for a single phase

Description: Conducts OLS regression for any phase.
Coefficients and residuals are produced. Also a simple line
graph for the specified phase with a regression line is
displayed in the graph window.
Command: Aregres (behavior, phaseX, v1)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
Ex: Aregres(cry, pcry, “A”)

Robust regression for a single phase

Description: Conducts robust regression for any phase.
Coefficients and residuals are produced. Also a simple line
graph for the specified phase with a regression line is
displayed in the graph window.
Command: Arobust(behavior, phaseX, v1)
behavior: behavior variable
phaseX: phase variable on which test is based
 v1: phase letter entered between quotes (e.g., “A”, “B”)
Ex: Arobust(cry, pcry, “A”)

OLS regression to compare phases

Description: Conducts OLS regression comparing any two
phases. Coefficients and residuals are produced for each
phase. Also a graph with a regression line is displayed for
each phase in the graph window.
Command: ABregres (behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: ABregres(cry, pcry, “A”, “B”)

Robust regression to compare phases

Description: Conducts robust regression comparing any two
phases. Coefficients and residuals are produced for each
phase. Also a graph with a robust regression line is displayed
for each phase in the graph window.
Command: ABrobust(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: ABrobust(cry, pcry, “A”, “B”)

Effect Size Functions

Calculates most common effect size indices

Description: Displays the percentage change and calculated
values for both the ES and d-index for any two phases.
Information for interpreting calculated values appears in the
Console.
Command: Effectsize(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
 v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: Effectsize(cry, pcry, “A”, “B”)

Calculates g-index
Description: Calculates effect size based on scores in the desired
zone.
Command: Gindex(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: Gindex(cry, pcry, “A”, “B”)

Calculation of Improvement Rate Difference

Description: Calculates Improvement Rate Difference (IRD) and
displays a graph. An increase in the targeted behavior is
desired.
Command: IRDabove(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: IRDabove(cry, pcry,“A”, “B”)

Calculation of Percentage of All Non-Overlapping Data.

Description: This function evaluates the Percentage of All Non-
overlapping Data (PAND). A decrease in the targeted
behavior is desired.
Command: IRDbelow(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: IRDbelow(cry, pcry,“A”, “B”)

Calculation of Percentage of All Non-Overlapping Data.

Description: This function evaluates the Percentage of All Non-
overlapping Data (PAND). An increase in the targeted
 behavior is desired.
Command: PANDabove(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: PANDabove(cry, pcry,“A”, “B”)

Calculation of Percentage of All Non-overlapping Data.

Description: This function evaluates the Percentage of All Non-
overlapping Data (PAND). A decrease in the targeted
behavior is desired.
Command: PANDbelow(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: PANDbelow(cry, pcry,“A”, “B”)

Calculation of Percentage of All Non-overlapping Data.

Description: This function calculates the Percentage of All Non-
overlap of All Pairs (NAP). An increase in the targeted
behavior is desired.
Command: NAPabove(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: NAPabove(cry, pcry,“A”, “B”)

Calculation of Percentage of All Non-Overlapping Data

Description: This function calculates the Percentage of All Non-
overlapping Pairs (NAP). A decrease in the targeted behavior
is desired.
Command: NAPbelow(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
 v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: NAPbelow(cry, pcry,“A”, “B”)

Calculation of Percentage of Data Exceeding the Median above the

median
Description: The Percentage of Data Exceeding the Median
(PEM) procedure offers a method to assess effect size and
adjust for the influence of outliers in the baseline phase when
desired values are above the reference line.
Command: PEMabove(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: PEMabove(cry, pcry,“A”, “B”)

Calculation of Percentage of Data Exceeding the Median below the

median
Description: The Percentage of Data Exceeding the Median
(PEM) procedure offers a method to assess effect size and
adjust for the influence of outliers in the baseline phase when
desired values are below the reference line.
Command: PEMbelow(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: PEMbelow(cry, pcry,“A”, “B”)

Create a legend on a PEM graph

Description: Adds a legend to a PEM graph. The graph cannot
be modified in any way after the legend is added.
Command: PEMlegend()
Ex: PEMlegend()

Calculation of Percentage of Non-overlapping Data above the

reference line
 Description: The Percentage of Non-overlapping Data (PND)
procedure offers a method to assess effect size based on the
highest data point in the comparison phase.
Command: PNDabove(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: PNDabove(cry, pcry,“A”, “B”)

Calculation of Percentage of Non-overlapping Data below the

reference line
Description: The Percentage of Non-overlapping Data (PND)
procedure offers a method to assess effect size based on the
lowest data point in the comparison phase.
Command: PNDbelow(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: PNDbelow(cry, pcry,“A”, “B”)

Create a legend on a PND graph

Description: Adds a legend to a PND graph. The graph cannot
be modified in any way after the legend is added.
Command: PNDlegend()
Ex: PNDlegend()
Statistical Process Control (SPC) Charts
Create an X-bar range chart (X-R chart)
Description: This function builds an X-R chart using range. A
space separates each phase. Used with multiple observations
per sample. In the example below, there are seven
observations (days) per sample (weeks). The “admitweek” is
the grouping variable. If there were 8 weeks, the values for
this variable would range from 1 to 8 with 7 observations for
each. Values for the upper band (Uband), mean, and lower
band (Lband) appear as statistical output in the Console.
 Command: XRchart(behavior, groupX, bandX,ABxlab,
ABylab, ABmain)
behavior: behavior variable
groupX: grouping variable
bandX: number of SDs desired (i.e., 1, 2, 3)
ABxlab: label for x-axis; written between quotation marks
ABylab: label for y-axis; written between quotation marks
ABmain: main label for graph; written between quotation
marks
Ex: XRchart(admits, admitweek, 3, “weeks”,“mean % of
admits”,“Social Admits”)

Create an R chart using mean range

Description: This function builds an R chart using the mean
range. A space separates each phase. Used with multiple
observations per sample. In the example below, there are
seven observations (days) per sample (weeks). The
“admitweek” is the grouping variable. If there were 8 weeks,
the values for this variable would range from 1 to 8 with 7
observations for each. This version of the R chart is
recommended when the sample size is small (less than 10). It
uses the mean range of the samples to track variation. Similar
to the X‾-R chart, using the mean range improves confidence
in the measure. Values for the Uband, mean, and Lband
appear as statistical output in the Console.
Command: Rchart(behavior, groupX, bandX, ABxlab, ABylab,
ABmain)
behavior: behavior variable
groupX: grouping variable
bandX: number of SDs desired (i.e., 1, 2, 3)
ABxlab: label for x-axis; written between quotation marks
ABylab: label for y-axis; written between quotation marks
ABmain: main label for graph; written between quotation
marks
Ex: Rchart(admits, admitweek, 3,”weeks”,“mean range % of
admits”, “Social Admits”)
Create an R chart using standard deviation
Description: This function builds an R chart using standard
deviation. A space separates each phase. Used with multiple
observations per sample. In the example below, there are
seven observations (days) per sample (weeks). The
“admitweek” is the grouping variable. If there were 8 weeks,
the values for this variable would range from 1 to 8 with 7
observations for each. This version of the R chart can be used
with samples greater than 10. Values for the Uband, mean,
and Lband appear as statistical output in the Console.
Command: Rchartsd(behavior, groupX, bandX, ABxlab,
ABylab, ABmain)
behavior: behavior variable
groupX: grouping variable
bandX: number of SDs desired (i.e., 1, 2, 3)
ABxlab: label for x-axis; written between quotation marks
ABylab: label for y-axis; written between quotation marks
ABmain: main label for graph; written between quotation
marks
Ex: Rchartsd (admits, admitweek, 3, “weeks”, “mean range % of
admits”, “Social Admits”)

Create line on R charts

Description: This function enables the user to draw solid vertical
lines between phases on the SPC R chart using standard
deviation and R chart using mean range. The user clicks the
mouse on a upper and lower y-ordinate.
Command:SPCline()
Ex: SPCline()

Create a proportion chart (P chart)

Description: This function builds a p chart. A space separates
each phase. Used with multiple observations per sample. The
behavior variable must be binary. Values for the Uband, mean,
and Lband appear as statistical output in the Console.
Command: Pchart(behavior, groupX, bandX, ABxlab, ABylab,
ABmain)
 behavior: behavior variable
groupX: grouping variable
bandX: number of SDs desired (i.e., 1, 2, 3)
ABxlab: label for x-axis; written between quotation marks
ABylab: label for y-axis; written between quotation marks
ABmain: main label for graph; written between quotation
marks
Ex: Pchart(group, wgroup, 3, “weeks”, “proportion of
attendance”, “Jenny’s Group Attendance”)

Create an X-moving range chart (X-mR chart)

Description: This function builds an X-mR chart and is used
with individual data. A space separates each phase. The X-mR
chart can be use to detect changes within and between phases.
This chart should not be used when there is a trend in the data.
As with the previous charts, large unexpected change in the
undesired zone (values above upper band or below the lower
band) may indicate the need to modify the intervention.
Values for the Uband, mean, and Lband appear as statistical
output in the Console.
Command: Xmrchart(behavior, phaseX, v1,bandX,ABxlab,
ABylab,
ABmain)
behavior: behavior variable
phaseX: phase variable on which bands are based
v1: phase letter entered between quotes (e.g., “A”, “B”)
bandX: number of SDs desired (i.e., 1, 2, 3)
ABxlab: label for x-axis; written between quotation marks
ABylab: label for y-axis; written between quotation marks
ABmain: main label for graph; written between quotation
marks
Ex: Xmrchart(esteem, pesteem,“A”, 3,”weeks”,“mean self-
esteem”, “Jenny’s Self-esteem”)

Create a C chart
Description: This function builds a C chart and is used with
individual (i.e., ungrouped) data. A space separates each
 phase. For use when the outcome variable is a count (i.e., ratio
level) variable.
Command: Cchart(behavior, phaseX, v1,bandX,ABxlab,
ABylab, ABmain)
behavior: behavior variable
phaseX: phase variable on which bands are based
v1: phase letter entered between quotes (e.g., “A”, “B”)
bandX: number of SDs desired (i.e., 1, 2, 3)
ABxlab: label for x-axis; written between quotation marks
ABylab: label for y-axis; written between quotation marks
ABmain: main label for graph; written between quotation
marks
Ex: Cchart(yell, pyell, “A”, 3, “days”, “Count”, “Count of
Yelling”)

Places legend at bottom of any SPC band graph

Description: This function enables the user to place legend on a
graph. NOTE: Once this legend is in place, the graph can no
longer be altered.
Command: SPClegend()
Ex: SPClegend()
Tests and Functions Related to Autocorrelation
Test for lag-1 autocorrelation
Description: This function tests for lag-1 autocorrelation. This
should be used any time the sample size is less than six. Any
phase can be tested. Also produces a regression line graph.
Command: ABrf2(behavior, phaseX, v1)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
Ex: ABrf2(cry,pcry,“A”)

Test for autocorrelation for any lag

Description: This function tests for autocorrelation for any lag.
Should be used with samples greater than or equal to six. Also
produces a significance graph for lags. The Box-Ljung test of
 significance is performed for all lags up to and including the
specified one.
Command: ABautoacf (behavior, phaseX, v, lags)
behavior: behavior variable
phaseX: phase variable on which test is based
v: phase letter entered between quotes (e.g., “A”, “B”)
lags: number of lags to be tested
Ex: ABautoacf(cry, pcry,“A”, 3)

Test for partial autocorrelation for any lag

Description: This function tests for partial autocorrelation for
any lag. Should only be used with samples greater than or
equal to six. Also produces significance graph for lags.
Command: ABautopacf (behavior, phaseX, v, lags)
behavior: behavior variable
phaseX: phase variable on which test is based
v: phase letter entered between quotes (e.g., “A”, “B”)
lags: number of lags to be tested
Ex: ABautopacf (cry, pcry,“A”, 3)

First difference transformation

Description: This function produces a first difference
transformation for any phase. The results can be saved for
later use by answering “y” to the prompt to save results. A
line graph is produced for the original and transformed data.
Statistical output includes displaying first differencing data in
the Console.
Command: diffchart(behavior, phaseX, v1)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
Ex: diffchart(cry, pcry,“A”)

Moving average transformation

Description: This function produces a moving average
transformation for any phase with every two scores being
averaged. The results can be saved for later use by answering
 “y” to the prompt to save results. A line graph is produced for
the original and transformed data. Statistical output includes
displaying moving average data in the Console.
Command: ABma (behavior, phaseX, v1)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
Ex: ABma (cry, pcry, “A”)

Combine two or more data files

Description: This function combines data files. This is useful
after data are created during transformations when using the
diffchart or ABma functions. Once files with different phases
are combined, you can use the saved file for significance
testing.
Command: Append()
Ex: Append()
Tests of Statistical Significance
Proportion-frequency/binomial test
Description: Binomial test comparing the number of
observations of a phase in a desired zone to another phase.
User needs to select the method for defining a desired zone
(e.g., below one sd)
Command: ABbinomial(phaseX, v1, v2, sucessA, sucessB)
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
successA: number of observations in desired zone for v1
successB: number of observations in desired zone for v1
Ex: ABbinomial(pyell,“A”,“B”, 1, 15)

Conservative dual-criteria test with desired zone above the lines

Description: Conservative dual-criteria test comparing the
frequency of observations above both the mean and OLS
regression line in any two phases.
Command: CDCabove(behavior, phaseX, v1, v2)
behavior: behavior variable
 phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: CDCabove(esteem, pesteem,“A”,“B”)

Conservative dual-criteria test with desired zone below the lines

Description: Conservative dual-criteria test comparing the
frequency of observations below both the mean and OLS
regression line in any two phases.
Command: CDCbelow(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: CDCbelow(esteem, pesteem,“A”,“B”)

Conservative dual-criteria test with desired zone above the lines

using robust regression
Description: Conservative dual-criteria test comparing the
frequency of observations above both the mean and robust
regression line in any two phases.
Command: RobustCDCabove(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: RobustCDCabove(esteem, pesteem,“A”,“B”)

Conservative dual-criteria test with desired zone below the lines

using robust regression
Description: Conservative dual-criteria test comparing the
frequency of observations below both the mean and robust
regression line in any two phases.
Command: RobustCDCbelow(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
 v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: RobustCDCbelow(esteem, pesteem,“A”,“B”)

Chi-square test with desired zone above the mean

Description: Chi-square test comparing the frequency of
observations above the mean in any two phases.
Command: meanabove(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: meanabove(esteem, pesteem,“A”,“B”)

Chi-square test with desired zone below the mean

Description: Chi-square test comparing the frequency of
observations below the mean in any two phases.
Command: meanbelow(behavior, phaseX, v1,v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: meanbelow(yell,pyell,“A”,“B”)

Chi-square test with desired zone above the median

Description: Chi-square test comparing the frequency of
observations above the median in any two phases.
Command: medabove(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: medabove(esteem, pesteem,“A”,“B”)

Chi-square with desired zone below the median

Description: Chi-square test comparing the frequency of
observations below the median in any two phases.
Command: medbelow(behavior, phaseX,v1,v2)
behavior: behavior variable
 phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: medbelow(yell,pyell,“A”,“B”)

Chi-square with desired zone above the trimmed mean

Description: Chi-square test comparing the frequency of
observations above the trimmed mean in any two phases.
Command: trimabove(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: trimabove(esteem, pesteem,“A”,“B”)

Chi-square with desired zone below the trimmed mean

Description: Chi-square test comparing the frequency of
observations below the trimmed mean in any two phases.
Command: trimbelow(behavior, phaseX,v1,v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: trimbelow(yell,pyell,“A”,“B”)

Chi-square with desired zone above the OLS regression line

Description: Chi-square test comparing the frequency of
observations above the OLS regression line in any two
phases.
Command: regabove(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: regabove(esteem,pesteem,“A”,“B”)

Chi-square with desired zone below the OLS regression line

 Description: Chi-square test comparing the frequency of
observations below the OLS regression line in any two
phases.
Command: regbelow(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: regbelow(cry,pcry,“A”,“B”)

Chi-square with desired zone above the robust regression line

Description: Chi-square test comparing the frequency of
observations above the robust regression line in any two
phases.
Command: robregabove(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: robregabove(esteem, pesteem,“A”,“B”)

Chi-square with desired zone below the robust regression line

Description: Chi-square test comparing the frequency of
observations below the robust regression line in any two
phases.
Command: robregbelow(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: robregbelow(cry,pcry,“A”,“B”)

t test
Description: Computes Student t test between any two phases.
This test should only be used if there is not a trend in either
phase AND there is no problem with autocorrelation in either
 phase. Graphical output is a bar chart displaying the mean for
each phase.
Command: ABttest(behavior, phaseX, v1,v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: ABttest(yell, pyell,“A”,“B”)

Wilcoxon Rank Sum

Description: performs a two-sample Wilcoxon test between any
two phases.
Nonparametric test to compare means. This test should only be
used if there is not a trend in either phase AND there is no
problem with autocorrelation in either phase.
Command: ABWilcox(behavior, phaseX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
v1: phase letter entered between quotes (e.g., “A”, “B”)
v2: phase letter entered between quotes (e.g., “A”, “B”)
Ex: ABWilcox(yell, pyell,“A”,“B”)

One-way analysis of variance

Description: Computes one-way ANOVA and performs Tukey
multiple comparison test. This test should only be used if
there is not a trend in any phase AND there is no problem
with autocorrelation in any phase. Use ANOVA instead of a t
test when comparing more than two phases
Command: ABanova(behavior, phaseX)
behavior: behavior variable
phaseX: phase variable on which test is based
Ex: ABanova(esteem,pesteem)

Scientific notation
Description: Coverts scientific notation to five decimal places
Command: SN(value)
value: value to be translated
 Ex: SN(2.73e-16)

Group Data Functions

Lag-1 autocorrelation for group data
Description: This function tests for lag-1 autocorrelation for
group data. Any phase can be tested. Produces a regression
line graph
Command: GABrf2(behavior, phaseX, timeX, v1)
behavior: behavior variable
phaseX: phase variable on which test is based
timeX: time variable (e.g., week)
v1: letter of phase being tested (e.g., “A”)
Ex: GABrf2(attend, pattend, week, “A”)

t test for group data

Description: Computes t test for group data. A bar graph
showing the mean for each phase is displayed.
Command: GABttest(behavior, phaseX, timeX, v1, v2)
behavior: behavior variable
phaseX: phase variable on which test is based
timeX: time variable (e.g., week)
v1: letter of first phase (e.g., “A”)
v2: letter of second phase (e.g., “B”)
Ex: GABttest(attend, pattend, week, “A”, “B”)

Draws median line group data

Description: Places median line for a single phase in group box
plot.
Command: Gmedian(behavior, phaseX, v)
behavior: behavior variable
phaseX: phase variable on which test is based
v: letter of phase (e.g., “A”)
Ex: Gmedian(attend, pattend, “A”)

Functions Related to RMarkdown

Draw an arrow in an RMarkdown document
Description: This function allows a user to draw an arrow on a
RMarkdown graph. For example, an arrow can be drawn from
 a text label of a critical event to a data point on the graph.
Command: RMarrow(X1,Y1,X2,Y2)
Ex: RMarrow(5.9,5,9.5,5.1)

Draw a solid line between phases on a RMarkdown graph.

Description: This function enables the user to draw solid vertical
lines between phases on a graph created in RMarkdown.
Command: RMlines(behavior, X)
EX: RMlines(yell,15.5)

Adding goal line to an ABplot in an RMarkdown document.

Description: This function enables a user to add a line
representing the goal line (a level of function set as a goal for
a client) to an ABplot in an RMarkdown document.
Command: RMGline(y)
y: y-ordinate
Ex: RMGline(6)
Command: RMstat(behavior, phaseX, v, statX)
behavior: behavior variable
phaseX: phase variable
v: phase letter entered between quotes (e.g., “A”, “B”)
X: start of horizontal line
statX: mean, median, or sd written between quotation marks
(e.g., “mean”)
Ex: RMstat(yell, pyell, “A”, “mean”, 7)

Adding trimmed mean line to an ABplot.

Description: This function enables a user to add a line
representing the trimmed mean to any phase of an ABplot.

Label an RMarkdown graph

Description: This function enables the user to write text on a
graph. Users have three options for doing this: text with
neither subscripts nor superscripts, text with superscripts, and
text with subscripts. Commands for each and examples to
label a graph are displayed below. After the command is
invoked, users will be prompted to place the cursor where the
 text is to begin. After text is placed, users will be prompted to
choose whether or not to accept the graph with the text.
Command: (for text with neither subscripts nor superscripts)
RMtext()
Text to be entered on graph must appear between quotation
marks.
Ex: RMtext(“baseline”,5,6)
Command: (for text with superscripts)
ABtext(expression(text^superscript,x,y))
text: This is the text that is to be displayed on the graph.
superscript: This is the actual superscript that is to be
displayed.
x: x coordinate
y: y coordinate
Ex: ABtext(expression(X^2,5,2))
Command: (for text with subscripts)
ABtext(expression(text[subscript],x,y))
text: This is the text that is to be displayed on the graph.
subscript: This is the actual subscript that is to be displayed.
Ex: ABtext(expression(B[1],8,2))

Meta-analysis Functions
Calculation of mean effect size
Description: This function calculates a mean and SD for
Cohen’s D effect sizes. A file containing saved effect sizes
must be opened by Getcsv() and then attached.
Command: meanES(ES, lab, esmain)
ES: effect size variable
lab: label for effect size variable
esmain: main title for graph
Ex: meanES(ES,“cry”,“ES For Crying”)

Calculation of nonparametric mean effect size

Description: This function calculates a mean and SD for NAP
effect sizes. A file containing saved NAP effect sizes must be
opened by Getcsv() and then attached.
Command: meanNAP ES, lab, esmain)
 ES: effect size variable
lab: label for effect size variable
esmain: main title for graph
Ex: meanNAP(ES,“cry”,“ES For Crying”)

Meta Regression
Description: Meta regression for saved effect sizes. A file
containing saved effect sizes must be opened by Getcsv() and
then attached.
Command: metareg(ES, V)
ES: effect size variable
V: Variance of the effect size
Ex: metareg(ES,V)

Meta Regression with a moderator

Description: Meta regression with moderaor for saved effect
sizes. A file containing saved effect sizes must be opened by
Getcsv() and then attached.
Command: metaregi(ES, I, V)
ES: effect size variable
V: Variance of the effect size
Ex: metaregi(ES,I,V)
Interactive RMarkdown Syntax
Command Substitute
ABarrow() RMarrow() RMarrow(X1,Y1,X2,Y2) X1 y1 From coordinates of arrow. X2
y2 To coordinates of arrow.
Gline() RMGline() RMGline(Y) From Y coordinate to draw horizontal line
ABlines() RMlines() RMlines(behavior, X) Behavior variable to draw vertical line
From X coordinate to draw vertical line
ABstat RMstat() RMstat(behavior, phase,“statistic”,X) X start of horizontal line
ABtext() RMtext() RMtext(“text”,X,Y) From X and Y coordinates to draw text
Getcsv() ssd<-read() ssd <- read.csv(“file”) file data file name
 APPENDIX C

Decision Trees

SPC Chart Decision Tree

Chi-Square Decision Tree

Hypothesis Test Decision Tree
 APPENDIX D

Bibliography of Additional Resources

Texts Covering Single-Subject Research Methodology

Bloom, M., Fischer, J., & Orme, J. G. (2009). Evaluating practice: Guidelines for the accountable
professional (6th ed.). New York, NY: Pearson.
Franklin, R. D., Allison, D. B., & Gorman, B. S. (Eds.). (1997). Design and analysis of single-case
research. Mahwah, NJ: Erlbaum.
Gast, D. L., & Ledford, J. (2010). Single-subject research in behavioral sciences. New York, NY:
Routledge.
Golper, L. A. C. (2012). Outcomes in speech-language pathology: Contemporary theories, models,
and practices. New York, NY: Thieme Medical.
Janosky, J. E., Leininger, S. L., Hoerger, M. P., & Libkuman, T. M. (2009). Single subject designs in
biomedicine (2009 ed.). New York, NY: Springer.
Kazdin, A. E. (2011). Single-case research designs: Methods for clinical and applied settings (2nd
ed.). New York, NY: Oxford University Press.
Kennedy, C. H. (2005). Single-case designs for educational research. New York, NY: Pearson
Education.
Kratochwill, T. R., & Levin, J. R. (Eds.). (1992). Single-case research design and analysis: New
directions for psychology and education. Hillsdale, NJ: Erlbaum.
Kratochwill, T. R., Hitchcock, J., Horner, R. H., Levin, J. R., Odom, S. L., Rindskopf, D. M. &
Shadish, W. R. (2010). Single-case designs technical documentation. Retrieved from What Works
Clearinghouse website: http://ies.ed.gov/ncee/wwc/pdf/wwc_scd.pdf
Nugent, W. R. (2009). Analyzing single system design data. New York, NY: Oxford University Press.
Orme, J. G., & Combs-Orme, T. (2011). Outcome-informed evidence-based practice. New York, NY:
Pearson.
Ray, W. J. (2011). Methods toward a science of behavior and experience (10th ed.). Belmont, CA:
Cengage Learning.
Richards, S., & Taylor, R. L. (2014). Single subject research: Applications in educational and
clinical settings (2nd ed.). Belmont, CA: Wadsworth/Thomson Learning.
Riley-Tillman, T. C., & Burns, M. K. (2009). Evaluating educational interventions: Single-case
design for measuring response to intervention. New York, NY: Guilford Press.
Rubin, A., & Babbie, E. R. (2013). Research methods for social work (8th ed.). Belmont, CA:
Brooks/Cole.
Schweigert, W. A. (2011). Research methods in psychology: A handbook (3rd ed.). Long Grove, IL:
Waveland Press.
Scruggs, T. E., & Mastropieri, M. A. (Eds.). (2006). Applications of research methodology. San
Diego, CA: JAI Press.
Shadish, W. R., Hedges, L. V., & Pustejovsky, J. E. (2014). Analysis and meta-analysis of single-case
designs with a standardized mean difference statistic: A primer and applications. Journal of School
Psychology, 52(2), 123–147.
Skinner, C. H. (2013). Single-subject designs for school psychologists. New York, NY: Routledge.
Thomas, J., & Hersen, M. (2011). Understanding research in clinical and counseling psychology
(2nd ed.). New York, NY: Routledge.
Thyer, B. A. (2009). The handbook of social work research methods. Thousand Oaks, CA: Sage.
Thyer, B. A., & Myers, L. L. (2007). A social worker’s guide to evaluating practice outcomes.
Alexandria, VA: CSWE Press.
Vannest, K. J., Davis, J. L., and Parker, R. I. (2013). A new approach to single case research. New
York, NY: Routledge.
Wallen, N. E., & Fraenkel, J. R. (2013). Educational research: A guide to the process (2nd ed.). New
York, NY: Routledge.

Recommended Reading on SPC Charts

Mitra, A. (2008). Control charts for attributes. In Fundamentals of quality control and improvement
(3rd ed., pp. 369–414). Hoboken, NJ: Wiley.
Orme, J. G., & Cox, M. E. (2001). Analyzing single-subject design data using statistical process
control charts. Social Work Research, 25(2), 115–127.

Recommended Reading Rmarkdown

Baumer, B., & Udwin, D. (2015). R markdown. Wiley Interdisciplinary Reviews: Computational
Statistics, 7(3), 167–177.
Wickham, H., & Grolemund, G. (2016). R for data science: Import, tidy, transform, visualize, and
model data. O’Reilly Media.
Xie, Y., Allaire, J. J., & Grolemund, G. (2018). R Markdown: The definitive guide. CRC Press.
Xie, Y., Dervieux, C., & Riederer, E. (2020). R Markdown cookbook. CRC Press.

Helpful Websites
Campbell Collaboration— http://www.campbellcollaboration.org. Produces freely available
systematic reviews in a number of broad disciplines, including crime and justice, education,
international development, and social welfare. This site contains many free resources, including
training on methodology related to the production of systematic reviews.
Gitlab. https://evoldyn.gitlab.io/evomics-2018/ref-sheets/rmarkdown-cheatsheet-2.0.pdf. Cheat sheet
for RMarkdown.
Outcome Informed Evidence-Based Practice. http://ormebook.com/. This site accompanies Orme and
Combs-Orme’s (2011) text listed in the Texts Covering Single-Subject Research Methodology. In
addition to information relevant to each chapter in the text, this site contains a lot of pertinent
information related to single-subject research in general, including bibliographies, templates for
collecting data and creating graphs, and links to standardized scales.
R for Data Science. https://r4ds.had.co.nz. Contains an online free-to-use copy of the R for Data
Science book.
RMarkdown. The definitive guide provides an online free-to-use version of R Markdown.
https://bookdown.org/yihui/rmarkdown/
RStudio RMarkdown. https://rmarkdown.rstudio.com. Contains a host of recourses on using
RMarkdown, including cheat sheets and a reference guide.
Single-Case Research. http://www.singlecaseresearch.org/. Information on single case research
design and analysis through the posting of published papers, manuscripts in press, and white
papers. Free calculators are available for the purpose of analysis along with instructional videos on
a variety of analysis topics.
U.S. Department of Education’s Institute of Education Sciences. http://ies.ed.gov/. Contains a host of
resources related to education, including methodological papers and presentations, datasets, and
 access to the What Works Clearinghouse.

Additional Resource
Volkov, B. B., & King, J. A. (2007). A checklist for building organizational evaluation capacity.
Retrieved from
https://wmich.edu/sites/default/files/attachments/u350/2014/organiziationevalcapacity.pdf
References

Archer, B., Azios, J. H., Müller, N., & Macatangay, L. (2019). Effect sizes in single-case aphasia
studies: A comparative, autocorrelation-oriented analysis. Journal of Speech, Language, and
Hearing Research, 62(7), 2473–2482.
Arnold, M. E. (2006). Developing evaluation capacity in extension 4-H field faculty. A framework
for success. American Journal of Evaluation, 27(2), 257–269.
Auerbach, C., & Schudrich, W. Z. (2013). SSD for R: A comprehensive statistical package to analyze
single-system data. Research on Social Work Practice, 23(3), 346–353.
doi:10.1177/1049731513477213
Auerbach, C., & Zeitlin , W. (2021). SSD for R (version 1.5.2). Vienna, Austria: R Foundation for
Statistical Computing. Retrieved from http://www.R-project.org/
Baizerman, M., Compton, D. W., & Hueftle Stockdill, S. (2002). New directions for ECB. In D. W.
Compton, M. Baizerman, & S. Stockdill (Eds.), The art, craft, and science of evaluation capacity
building (Vol. 2002, pp. 109–120). San Francisco, CA: Jossey-Bass.
Baumer, B., & Udwin, D. (2015). R markdown. Wiley Interdisciplinary Reviews: Computational
Statistics, 7(3), 167–177.
Beck, A. T., Ward, C. H., Mendelson, M., Mock, J., & Erbaugh, J. (1961). An inventory for
measuring depression. Archives of General Psychiatry, 4, 561–571.
Beddoe, L. (2011). Investing in the future: Social workers talk about research. British Journal of
Social Work, 41(3), 557–575.
Benneyan, J. C., Lloyd, R. C., & Plsek, P. E. (2003). Statistical process control as a tool for research
and healthcare improvement. Quality and Safety in Health Care, 12(6), 458–464.
doi:10.1136/qhc.12.6.458
Berger, R. (2010). EBP practitioners in search of evidence. Journal of Social Work, 10(2), 175–191.
Bloom, M., Fischer, J., & Orme, J. G. (2009). Evaluating practice: Guidelines for the accountable
professional (6th ed.). New York, NY: Pearson.
Borckardt, J. J. (2008). User’s guide: Simulation modeling analysis: Time series analysis program for
short time series data streams: Version 8.3.3. Retrieved from
http://www.clinicalresearcher.org/SMA_Guide.pdf
Brossart, D. F., Parker, R. I., & Castillo, L. G. (2011). Robust regression for single-case data analysis:
How can it help? Behavior Research Methods, 43(3), 710–719. doi:10.3758/s13428-011-0079-7
Buck, J. A. (2011). The looming expansion and transformation of public substance abuse treatment
under the Affordable Care Act. Health Affairs, 30(8), 1402–1410.
Carman, J. G., & Fredericks, K. A. (2010). Evaluation Capacity and Nonprofit Organizations: Is the
glass half-empty or half-full? American Journal of Evaluation, 31(1), 84–104.
Chambless, D. L., Baker, M. J., Baucom, D. H., Beutler, L. E., Calhoun, K. S., Crits-Christoph, P., . .
. Haaga, D. A. (1998). Update on empirically validated therapies, II. Clinical Psychologist, 51(1),
3–16.
Choi, M., & Ruona, W. E. (2011). Individual readiness for organizational change and its implications
for human resource and organization development. Human Resource Development Review, 10(1),
46–73.
Christie, D., & Menter, I. (2009). Research capacity building in teacher education: Scottish
collaborative approaches. Journal of Education for Teaching, 35(4), 337–354.
Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ:
Erlbaum.
Council on Social Work Education (CSWE). (2015). Educational policy and accreditation standards.
Alexandria, VA: Author.
Edwards, J. R., & Cable, D. M. (2009). The value of value congruence. Journal of Applied
Psychology, 94(3), 654–677.
Epstein, I., & Blumenfield, S. (2012). Clinical data-mining in practice-based research: Social work
in hospital settings. Routledge.
Ferguson, C. J. (2009). An effect size primer: A guide for clinicians and researchers. Professional
Psychology: Research and Practice, 40(5), 532–538.
Fisher, W. W., Kelley, M. E., & Lomas, J. E. (2003). Visual aids and structured criteria for improving
visual inspection and interpretation of single-case designs. Journal of Applied Behavior Analysis,
36(3), 387–406.
Fouché, C., & Lunt, N. (2010). Nested mentoring relationships reflections on a practice project for
mentoring research capacity amongst social work practitioners. Journal of Social Work, 10(4),
391–406.
Free Software Foundation, Inc. (2012). RStudio. Boston, MA: Author.
Gast, D. L., & Ledford, J. (2010). Single-subject research in behavioral sciences. New York, NY:
Routledge.
Glass, G. V., McGaw, B., & Smith, M. L. (1981). Meta-analysis in social research. Thousand Oaks,
CA: Sage.
Higgins, J., & Thomas, J. (Eds.). (2020). Cochrane handbook for systematic reviews of interventions
(Version 6.1.0). Chichester, UK: Wiley. Retrieved from
https://training.cochrane.org/handbook/current
Huitema, B. E., & McKean, J. W. (1994). Two reduced-bias autocorrelation estimators: rF1 and rF2.
Perceptual and Motor Skills, 78(1), 323–330.
Janosky, J. E., Leininger, S. L., Hoerger, M. P., & Libkuman, T. M. (2009). Single subject designs in
biomedicine (2009 ed.). New York, NY: Springer.
Kazdin, A. E. (2011). Single-case research designs: Methods for clinical and applied settings (2nd
ed.). New York, NY: Oxford University Press.
Kratochwill, T. R., & Levin, J. R. (2014). Single-case intervention research: Methodological and
statistical advances. American Psychological Association.
Kratochwill, T. R., Hitchcock, J. H., Horner, R. H., Levin, J. R., Odom, S. L., Rindskopf, D. M., &
Shadish, W. R. (2013). Single-case intervention research design standards. Remedial and Special
Education, 34(1), 26–38. doi:10.1177/0741932512452794
Kratochwill, T. R., Hitchcock, J., Horner, R. H., Levin, J. R., Odom, S. L., Rindskopf, D. M., &
Shadish, W. R. (2010). Single-case designs technical documentation. Washington, DC: US
Department of Education, Institute of Education Sciences. Retrieved from
https://files.eric.ed.gov/fulltext/ED510743.pdf
Krishef, C. H. (1991). Fundamental approaches to single subject design and analysis. Malabar, FL:
Krieger.
Kromrey, J. D., & Foster-Johnson, L. (1996). Determining the efficacy of intervention: The use of
effect sizes for data analysis in single-subject research. Journal of Experimental Education, 65(1),
73–93.
Lenz, A. S. (2012). Calculating effect size in single-case research: A comparison of nonoverlap
methods. Measurement and Evaluation in Counseling and Development, 46(1), 64–73.
doi:10.1177/0748175612456401
Logan, L. R., Hickman, R. R., Harris, S. R., & Heriza, C. B. (2008). Single-subject research design:
Recommendations for levels of evidence and quality rating. Developmental Medicine & Child
Neurology, 50(2), 99–103.
Lunt, N., Fouché, C., & Yates, D. (2008). Growing research in practice (GRIP): An innovative
partnership model. Wellington, New Zealand: Families Commission.
Ma, H.-H. (2009). The effectiveness of intervention on the behavior of individuals with autism. A
meta-analysis using percentage of data points exceeding the median of baseline phase (PEM).
Behavior Modification, 33(3), 339–359.
Macgowan, M. J. (2008). A guide to evidence-based group work. Oxford University Press.
Macgowan, M. J. (2012). A standards-based inventory of foundation competencies in social work
with groups. Research on Social Work Practice, 22(5), 578–589. doi:10.1177/1049731512443288
Manolov, R., Solanas, A., Sierra, V., & Evans, J. J. (2011). Choosing among techniques for
quantifying single-case intervention effectiveness. Behavior Therapy, 42(3), 533–545.
Matyas, T., & Greenwood, K. (1990). Visual analysis of single-case time series: Effects of variability,
serial dependence, and magnitude of intervention effects. Journal of Applied Behavioral Analysis,
23(3), 341–351.
McCrystal, P., & Wilson, G. (2009). Research training and professional social work education:
Developing research-minded practice. Social Work Education, 28(8), 856–872.
Mechanic, D. (2012). Seizing opportunities under the Affordable Care Act for transforming the
mental and behavioral health system. Health Affairs, 31(2), 376–382.
Miller, B. (n.d.). Single-subject research design (SSRD). Vancouver, BC, Canada: University of
British Columbia School of Rehab Sciences.
Mitra, A. (2008). Control charts for attributes. In Fundamentals of quality control and improvement
(3rd ed., pp. 369–414). Hoboken, NJ: Wiley.
Mohammed, M. A., & Worthington, P. (2012). Why traditional statistical process control charts for
attribute data should be viewed alongside an XMR-chart. BMJ Quality & Safety, 22(3), 263–269.
doi:10.1136/bmjqs-2012-001324
Morgan, D. L. (2008). Single-case research methods for the behavioral and health sciences. Sage.
Nagler, E., Rindskopf, D. M., & Shadish, W. R. (2008). Analyzing data from small N designs using
multi-level models: A procedural handbook. New York: Graduate Center, CUNY.
Nathan, P. E., & Gorman, J. M. (Eds.). (2002). A guide to treatments that work (2nd ed.). New York,
NY: Oxford University Press.
National Association of Social Workers. (2017). Code of ethics. Washington, DC.
Nourbakhsh, M. R., & Ottenbacher, K. J. (1994). The statistical analysis of single-subject data: A
comparative examination. Physical Therapy, 74(8), 768–776.
Orme, J., & Powell, J. (2008). Building research capacity in social work: process and issues. British
Journal of Social Work, 38(5), 988–1008.
Orme, J. G. (1991). Statistical conclusion validity for single-system designs. Social Service Review,
65(3), 468–491. doi:10.2307/30012410
Orme, J. G., & Combs-Orme, T. (2011). Outcome-informed evidence-based practice. New York:
Pearson.
Orme, J. G., & Cox, M. E. (2001). Analyzing single-subject design data using statistical process
control charts. Social Work Research, 25(2), 115–127.
Parker, R. I. (2006). Increased reliability for single-case research results: Is the bootstrap the answer?
Behavior Therapy, 37(4), 326–338. doi:10.1016/j.beth.2006.01.007
Parker, R. I., Hagan-Burke, S., & Vannest, K. (2007). Percentage of all non-overlapping data
(PAND). An alternative to PND. Journal of Special Education, 40(4), 194–204.
doi:10.1177/00224669070400040101
Parker, R. I., Vannest, K. J., & Brown, L. (2009). The improvement rate difference for single-case
research. Exceptional Children, 75(2), 135–150.
Parker, R. I., Vannest, K. J., & Davis, J. L. (2011). Effect size in single-case research: A review of
nine nonoverlap techniques. Behavior Modification, 35(4), 303–322.
Polit, D. F., & Chaboyer, W. (2012). Statistical process control in nursing research. Research in
Nursing & Health, 35(1), 82–93. doi:10.1002/nur.20467
Portney, L. G., & Watkins, M. P. (2008). Foundations of clinical research: Applications to practice
(3rd ed.). Hoboken, NJ: Prentice Hall.
Preskill, H., & Boyle, S. (2008). A multidisciplinary model of evaluation capacity building.
American Journal of Evaluation, 29(4), 443–459.
Pustejovsky, J. E. (2019). Procedural sensitivities of effect sizes for single-case designs with directly
observed behavioral outcome measures. Psychological Methods, 24(2), 217.
R Core Team. (2013). R: A language and environment for statistical computing. Vienna, Austria: R
Foundation for Statistical Computing. Retrieved from
https://www.scirp.org/(S(351jmbntvnsjt1aadkposzje))/reference/ReferencesPapers.aspx?
ReferenceID=1787696
Rock, B. D., Auerbach, C., Kaminsky, P., & Goldstein, M. (1993). Integration of computer and social
work culture: A developmental model. In B. Glastonbury (Ed.), Human welfare and technology:
Papers from the Husita 3 Conference on IT and the quality of life and services. Maastricht,
Netherlands: Van Gorcum, Assen.
The R Project for Statistical Computing. (n.d.). What is R? Retrieved from http://www.r-
project.org/about.html
Schlosser, R., & Wendt, O. (2008). Systematic reviews and meta-analyses of single-subject
experimental designs (SSEDs). National Center for the Dissemination of Disability Research.
https://ktdrr.org/systematicregistry/lib_systematic_search.cgi?location=sr&sel_1=104
Schudrich, W. (2012). Implementing a modified version of Parent Management Training (PMT) with
an intellectually disabled client in a special education setting. Journal of Evidence-Based Social
Work, 9(5), 421–423.
Scruggs, T. E., & Mastropieri, M. A. (1998). Summarizing single-subject research issues and
applications. Behavior Modification, 22(3), 221–242. doi:10.1177/01454455980223001
Scruggs, T. E., & Mastropieri, M. A. (2013). PND at 25: Past, present, and future trends in
summarizing single-subject research. Remedial and Special Education, 34(1), 9–19.
doi:10.1177/0741932512440730
Shaw, I. (2005). Practitioner research: evidence or critique? British Journal of Social Work, 35(8),
1231–1248.
Shaw, I. (2011). Evaluating in practice (2nd ed.). Burlington, VT: Ashgate.
Smith, I. R., Garlick, B., Gardner, M. A., Brighouse, R. D., Foster, K. A., & Rivers, J. T. (2013). Use
of graphical statistical process control tools to monitor and improve outcomes in cardiac surgery.
Heart, Lung and Circulation, 22(2), 92–99. doi:10.1016/j.hlc.2012.08.060
Smith, J. D. (2012). Single-case experimental designs: A systematic review of published research and
current standards. Psychological Methods, 17, 510–550.
Stewart, K. K., Carr, J. E., Brandt, C. W., & McHenry, M. M. (2007). An evaluation of the
conservative dual-criterion method for teaching university students to visually inspect Ab-design
graphs. Journal of Applied Behavior Analysis, 40(4), 713–718. doi:10.1901/jaba.2007.713-718
Swoboda, C. M., Kratochwill, T. R., & Levin, J. R. (2010). Conservative dual-criterion method for
single-case research: A guide for visual analysis of AB, ABAB, and multiple-baseline designs.
Wisconsin Center for Education Research Working Paper No. 2010-13. Retrieved from
https://wcer.wisc.edu/docs/working-papers/Working_Paper_No_2010_13.pdf
Tasdemir, A. (2012). Effect of autocorrelation on the process control charts in monitoring of a coal
washing plant. Physicochemical Problems of Mineral Processing, 48(2), 495–512.
Tate, R. L., & Perdices, M. (2019). Single-case experimental designs for clinical research and
neurorehabilitation settings: Planning, conduct, analysis and reporting. Routledge.
Thombs, B. D., Ziegelstein, R. C., Beck, C. A., & Pilote, L. (2008). A general factor model for the
Beck Depression Inventory-II: Validation in a sample of patients hospitalized with acute
myocardial infarction. Journal of Psychosomatic Research, 65(2), 115–121.
Thyer, B. A., & Myers, L. L. (2011). The quest for evidence-based practice: A view from the United
States. Journal of Social Work, 11(1), 8–25.
Vannest, K. J., Davis, J. L., & Parker, R. I. (2013). A new approach to single case research. New
York: Routledge.
Verzani, J. (2004). Using R for introductory statistics. Chapman and Hall/CRC.
Volkov, B. B., & King, J. A. (2007). A checklist for building organizational evaluation capacity.
Retrieved from
https://wmich.edu/sites/default/files/attachments/u350/2014/organiziationevalcapacity.pdf
Weisberg, S., & Fox, J. (2010). An R companion to applied regression (2nd ed.). Thousand Oaks,
CA: Sage.
Wendt, O. (2009, May). Calculating effect sizes for single-subject experimental designs: An overview
and comparison. Presented at the Ninth Annual Campbell Collaboration Colloquium, Oslo,
Norway.
Wendt, O., & Rindskopf, D. (2020). Exploring new directions in statistical analysis of single-case
experimental designs.
Wheeler, D. J. (2004). Advanced topics in statistical process control: The power of Shewhart’s charts
(2nd ed.). Knoxville, TN: SPC Press.
Woodall, W. H. (2006). The use of control charts in health-care and public-health surveillance.
Journal of Quality Technology, 38(2), 89–104.
Index

Symbols
``` (grave accents), 158
{} (curly brackets), 158

ABanova() function, 193

ABarrow() function, 180, 196
ABautoacf () function, 189
ABautoacf() function, 48
ABautopacf () function, 189
ABbinomial() function, 102–105, 190
ABdescrip() function, 63, 83, 126, 139, 183
ABiqr() function, 183
ABlineD() function, 124, 180
ABlines() function, 59, 76–77, 156, 180, 196
ABma() function, 44, 86, 190
ABmain argument, 22
ABplot() function, 21, 61, 138, 158–159, 179
ABplotm() function, 138, 179
ABregres() function, 65–67, 70, 184
ABrf2() function, 20, 106, 116–117, 189
ABrobust() function, 184
ABstat() function, 35, 61, 180, 196
ABtext() function, 35, 60, 76–77, 83, 124, 137, 180, 195–196
ABtsplot() function, 183
ABttest() function, 112, 163, 193
ABWilcox() function, 193
ABxlab argument, 22
ABylab argument, 22
accountability, 92
Adjusted R-squared, 41
alternative hypothesis (H1), 90
American Psychological Association (APA), 4
analysis of variance (ANOVA), 91, 112–116, 193
annotations
adding text in RMarkdown, 160, 164
adding to line graphs, 55, 59–62
client-annotated records, 13–14
ANOVA (analysis of variance), 91, 112–116, 193
Append() function, 87, 116, 190
Aregres() function, 183
arguments, 22–23
Arobust() function, 184
arrows, drawing on graphs, 180
assignments
annotating line graphs, 55
baseline analysis, 55–56
creating and importing group spreadsheets, 140–141
creating RMarkdown reports, 165–166
data entry, 17
hypothesis testing and effect sizes, 142
opening and displaying SSD for R files, 30
testing for Type I errors, 120
visual analysis, 142
visual comparison of two phases, 88–89
working with two phases of data, 87–88
attach(ssd) command, 47, 50, 57, 82, 86, 123
autocorrelation
functions associated with, 29, 189–190
in group data, 128–131
handling high levels of, 31–32
impact on visual interpretation, 5
removing in non-overlap methods, 82, 84–87
significance testing when high, 116–119
SPC charts and, 92
Type I errors and, 42–49

baseline phase data

analyzing, 31–32
autocorrelation, 42–49
defined, 31
trending, 40–41
types of, 31
visual analysis, 32–34
visual analysis example: Anthony, 49–54
visual analysis example: Jenny, 34–40
basic statistics, functions associated with, 29
BDI (Beck Depression Inventory), 11–12
Beck Depression Inventory (BDI), 11–12
behavior argument, 22
behavior measurement, using Excel to record client data, 14–16
binomial tests, 101–105
boldface, formatting in RMarkdown, 160
box plots, 34, 57–59, 83, 124

Cchart() function, 188

C charts, 99–100, 188
CDC (conservative dual criteria), 91, 109–111
CDCabove() function, 109–110, 190
CDCbelow() function, 109, 190
change, using effect size to describe, 69–75
chi-square tests, 91, 105–109, 191–192, 198
client-annotated records, 13–14
client data. See data
code chunks, 158–164
Code of Ethics (NASW), 3–4
comma-separated value (.csv) files, 10, 14, 23, 179
community data, functions suitable for, 30
Comprehensive R Archive Network, 6
concurrent baselines, 31
conservative dual criteria (CDC), 91, 109–111
control charts, 91
Council on Social Work Education (CSWE), 3
critical incident recordings, 13–14
.csv (comma-separated value) files, 10, 14, 23, 179
CSWE (Council on Social Work Education), 3

data
baseline phase data, 31–56
combining files, 190
comparing baseline and intervention phases, 57–89
entering and editing data directly in R, 175–178
example data, 8
with high levels of autocorrelation, 31–32
saving data files, 179
using Excel to record client data, 14–16
what to measure, 10–14 See also group data
data frames, 175
data transformations
functions associated with, 29, 189–190
using differencing, 44–49, 116
decision trees, 197–199
descriptive statistics
comparing baseline and intervention phases, 57–62, 184
functions associated with, 29, 183
diffchart() function, 189
differencing, transformations using, 44–49, 116
d-index value, 69, 71, 72, 81
direct behavioral observations, 10–11
directional null hypothesis, 90
downloading, 18–20

Educational Policy and Accreditation Standards (CSWE), 3

effect size
functions associated with, 29, 184–187
non-overlap methods, 76–82
using in meta-analysis, 146–153
using to describe change, 69–75
using to describe group change, 131–132
Effectsize() function, 69–75, 184
equal variances, 135
ES value, 69, 71, 72, 81
evidence-based group work, 121
example data
 case study set up, 23–28
downloading, 8, 19
Excel
benefits of, 16
drawing line graphs with, 6
measuring target behaviors with, 10
using to record client data, 14–16

false negatives, 92
false positives, 90
figure margins too large error, 138
F tests, 112, 135
functions
accessing hyperlinks in RStudio, 21
associated with data transformations, 29, 189–190
associated with hypothesis testing, 30
basic statistical analysis, 29, 183–184
effect size functions, 29, 184–187
graphing functions, 29, 179–183
group data, 30, 193–194
help files, 20–21
listing, 20, 179
listing variables, 179
meta-analysis functions, 30, 195
opening and importing .csv files, 179
related to autocorrelation, 29, 189–190
related to RMarkdown, 30, 194–195
saving data files, 179
SPC charts, 29, 187–189
SSD for R functions, 21–23
suitable for system-wide data, 30
summary of, 28–30
tests of statistical significance, 190–193 See also individual functions

GABrf2() function, 193

GABttest() function, 194
Getcsv() function, 34, 47, 50, 57, 71, 82, 86, 123, 179, 196
getting help, 20–21
getting started
download and installation, 18–20
example data: case study of Jenny, 23–28
listing functions, 20
locating help, 20–21
role of SSD for R, 18
SSD for R functions, 21–23
summary of functions, 28–30
Gindex() function, 184
g-index value, 69, 72–74, 81, 184
Gline() function, 82–83, 104–105, 181, 196
Gmedian() function, 194
goal line graphs, 82–83, 181
Google Sheets, 14
graphing functions, 29, 179–183
group data
analyzing, 123–124
autocorrelation and trend, 128–131
comparing phases, 126–127
descriptive statistics, 124–126
entering, 121–123
functions for analyzing, 30, 193–195
individual change, 137–139
tests of statistical significance, 132–137
using effect size to describe group change, 131–132 See also data
group research designs, 3, 6

Hedges's g value, 69, 71, 72, 81

help files, 20–21
heterogeneity, 149–153
HLM, 6
hypothesis testing
decision tree, 199
functions associated with, 30
null and alternative hypotheses, 90
statistical versus practical significance, 91

Improvement Rate Difference (IRD), 7, 78–81, 185

individual change, of group members, 137–139
individualized rating scales, 12–13
installation, 18–20
interquartile range (IQR), 181
interrupted time series research. See single-subject research
intervention phase, comparing to baseline, 57–89
IQR (interquartile range), 181
IQRbandgraph() function, 182
IQRlegend() function, 183
IQRline() function, 181
IRD (Improvement Rate Difference), 7, 79–81, 185
IRDabove() function, 185
IRDbelow() function, 79, 185

labels. See text

LaTex, 155
line graphs, 6, 35, 50, 59–62, 179–183
listnames() function, 179
logs, 13–14

Mac R Data Editor, 176

MacTex, 155
Md (median), 57–59, 106, 180
mean, 61, 106, 180
meanabove() function, 108, 191
meanbelow() function, 191
meanES() function, 195
meanNAP() function, 145–146, 154, 195
mean non-overlap of all pairs, 145–146
measures of variation, 62–65
medabove() function, 108, 191
medbelow() function, 108, 191
median (Md), 57–59, 106, 180
meta-analysis
conducting with SSD for R, 144–145
functions for conducting, 30, 195
importance of, 144
mean non-overlap of all pairs, 145–146
overview of, 143
steps of, 144
storing effect size results for, 71
using traditional effect size statistics, 146–153
metareg() function, 148–149, 154, 195
metaregi() function, 150–154, 195
Microsoft Excel, 6
MiKTeX, 155
MS Word, 155

n = 1 research. See single-subject research

NAP (Non-overlap of All Pairs), 80–81, 145–146, 186
NAPabove() function, 185
NAPbelow() function, 186
NASW (National Association of Social Workers), 3–4
National Association of Social Workers (NASW), 3–4
National Institute of Child Health and Human Development, 4
nondirectional hypothesis, 90
non-overlap methods, 76–82, 145–146
Non-overlap of All Pairs (NAP), 80–81, 145–146, 186
nonparametric tests, 105
null hypothesis (H0), 90
Numbers, 14

OLS (ordinary least squares) regression, 65, 106, 183

one-way ANOVA (analysis of variance), 112–116, 193
ordinary least squares (OLS) regression, 65, 106, 183

page breaks, in RMarkdown, 160

PAND (Percentage of all Non-overlapping Data), 7, 78, 81–82, 185–186
PANDabove() function, 185
PANDbelow() function, 185
Patient Protection and Affordable Care Act (2010), 92
Pchart() function, 188
PEM (Percentage of Data Exceeding the Median), 77–78, 81–82, 186
PEMabove() function, 186
PEMbelow() function, 186
PEMlegend() function, 186
Percentage of all Non-overlapping Data (PAND), 7, 78, 81–82, 185–186
Percentage of Data Exceeding the Median (PEM), 77–78, 81–82, 186
Percentage of Non-overlapping Data (PND), 76–77, 81–82, 186–187
performance, measuring in group work, 121
phasex argument, 22
plotnum() function, 138, 179–180
PND (Percentage of Non-overlapping Data), 76–77, 81–82, 186–187
PNDabove() function, 76, 186
PNDbelow() function, 76, 186
PNDlegend(), 186
practical significance, 90–91
practice research
action items for supporting, 169
barriers to conducting, 168–169, 173
building research capacity and culture, 167
building research skills, 171–172
collaborative approach to, 172–173
developing infrastructure and demand, 170–171
importance of, 167
proportion charts, 96–98
proportion charts (p charts)
alternatives to, 103
assessing binary outcomes in group data, 137
creating, 97, 188
using with other SPC charts, 100–101
proportion/ frequency tests, 91, 101–105, 135
p value, 90, 107

QEp value, 149, 153

Q statistic, 149–150
quick functions guide, 179–196

R
benefits of, 18
download and installation, 18
entering and editing data directly in, 175–178
range, 62–63
Rchart() function, 187
R charts, 94–96, 136–137, 187–188
Rchartsd() function, 187
R Data Editor, 176
reconstructed baselines, 31
regabove() function, 108–109, 192
regbelow() function, 108, 192
research capacity, 167. See also practice research
research standards, 4–5
resources, 6–7, 201–202
RMarkdown
additional resources, 202
creating documents, 155–164
functions appropriate for, 30, 194–196
getting started, 155–156
 overview of, 155
RMarrow() function, 159, 194, 196
RMGline() function, 194, 196
RMlines() function, 156, 159, 194, 196
RMstat() function, 159, 196
RMtext() function, 159, 195–196
robregabove() function, 109, 192
robregbelow() function, 109, 192
RobustCDCabove() function, 111, 190
RobustCDCbelow() function, 111, 191
robust conservative dual criteria tests, 91
robust regression, 106, 111, 184
RStudio
accessing function hyperlinks, 21
download and installation, 18–20
entering and editing data directly in R, 175–178
importing group data, 140–141
LaTex installation, 155

sample data, downloading, 8, 19

Savecsv() function, 176, 179
scales
individualized rating scales, 12–13
self-anchoring scales, 12
standardized scales, 11–12
scientific notation, 193
SD (standard deviation), 64–65, 180–182
sd1bandgraph() function, 181
SD1() function, 182
SD1legend() function, 182
sd2bandgraph() function, 182
SD2() function, 182
SD2legend() function, 182
SDAband() function, 181
self-anchoring scales
analyzing group data, 123
designing, 12
measuring individual change, 137
self-monitoring, 13–14
serial dependence, 5, 31–32, 42
Shewhart charts, 91
significance testing
analysis of variance (ANOVA), 112–116
conservative dual criteria (CDC), 109–111
first step in, 90
proportion/frequency tests, 101–105, 135
purpose of, 90
statistical process control (SPC) charts, 91–101
statistical versus practical significance, 90–91
t tests, 111–112, 132–135
 using R charts to assess variability of group data, 136–137
using X̄ -R charts to visualize group change, 135–136
when autocorrelation is high, 116–119
sig value, 90
Simulation Modeling Analysis, 6
SingleCaseES, 6
single-case research. See single-subject research
single-subject research
benefits of SSD for R, 7–8
current state of, 3–4
definition of, 10
description and usage of, 1–2
future of, 5
history of, 2–3
tools for evaluating data, 6
visual verus statistical analysis, 5–6
Single-System Design Analysis, 18
SN() function, 193
SPC (statistical process control)
benefits of SSD for R, 7
functions associated with, 29
SPC chart decision tree, 197
SPC charts, 91–101, 187–189
SPClegend() function, 137, 189
SPCline() function, 188
spreadsheet packages, 6, 10
SPSS, 6
ssd<- read() function, 196
SSDforR() function, 179
SSD for R package
analyzing baseline phase data, 31–56
analyzing group data, 121–142
benefits of, 7–8
comparing baseline and intervention phases, 57–89
data import, 10–17
getting started, 18–30
meta-analysis in single-subject evaluation research, 143–154
quick functions guide, 179–196
statistical tests of Type I errors, 90–120
topics covered, 8–9
using RMarkdown to present findings, 155–166
standard deviation (SD), 64–65, 180–182
standardized scales, 11–12
standards, 4–5
statax argument, 22
statistical analysis
basic statistical analysis, 29, 183–184
versus visual analysis, 5–6
statistical process control (SPC)
benefits of SSD for R, 7
 functions associated with, 29
SPC chart decision tree, 197
SPC charts, 91–101, 187–189
statistical significance, 90–91, 132–137, 190–193
statistical versus visual analysis, 5–6
system-wide data, functions suitable for, 30

table() function, 103

tests of significance. See significance testing
text
adding in RMarkdown, 160, 164
adding to line graphs, 59–62, 180
textx argument, 23
time series research. See single-subject research
title() function, 163
tM (trimmed mean), 61–62, 106, 181
transformations. See data transformations
trends
assessing in baseline data, 40–41
measuring, 65–69
trimabove() function, 109, 192
trimbelow() function, 192
trimdbelow() function, 109
trimmed mean (tM), 61–62, 106, 181
t tests, 91, 111–112, 132–135, 193
Ttimline() function, 181
Two Sample t-test, 112
Type I errors, 42, 90. See also significance testing
Type II errors, 42, 92

unequal variances, 135

Ungraph, 6

v argument, 22
variables, listing, 179
variation, measures of, 62–65
visual analysis
autocorrelation, 42–49, 84–87
baseline example data: Anthony, 49–54
baseline example data: Jenny, 34–40
baseline phase data, 32–34
benefits of SSD for R, 5–6
descriptive statistics, 57–62
goal line graphs, 82–83
improvement difference, 78–80
measures of variation, 62–63
measuring trends, 65–68
non-overlap methods, 76–78
non-overlap of all pairs, 80–82
standard deviation, 64–65
versus statistical analysis, 5–6
 trending, 40–42
using effect size to describe change, 69–75

Welch Two Sample t-test, 112

What Works Clearinghouse, 4, 144
Wilcoxon Rank Sum, 193

X-bar range charts (X̄-R charts), 92–94, 135–136, 187

X-moving-range charts (X-mR charts), 98–99, 188
Xmrchart() function, 188
XRchart() function, 187