Kline R.B. Principles and Practice of Structural Equation Modeling
Подождите немного. Документ загружается.
x Preface and Acknowledgments
learning about SEM by responding to questions that test their concept knowledge. Some
exercises also involve the analysis of structural equation models with actual data sets
(i.e., learning by doing). All of these features also support self-study of SEM; that is, they
should help readers who wish to learn about SEM but are not participating in a formal
course or seminar.
3. Website support for this edition is even stronger than that of the second edition.
For example, readers can freely download for every detailed example in Part II all syntax,
data, and output files for each of three widely used SEM computer tools: EQS, LISREL,
and Mplus. This allows readers to reproduce the analysis on their own computer using
the corresponding computer tool. Even if the reader uses a different computer tool for
SEM, all of these files can be opened with a standard text editor, such as Windows Note-
pad. That is, the reader does not need to have EQS, LISREL, or Mplus installed on his
or her computer in order to view the contents of these files. And for readers who already
use one of three computer tools for SEM (e.g., LISREL), it can be educational to view the
results of the same analysis generated by a different computer tool (e.g., Mplus). Other
resources for readers may be found on the book’s website (described in Chapter 1), the
address of which is presented on page 3.
4. The chapter on hypothesis testing in SEM (Chapter 8) reflects some of the most
recent thinking in this area that is described by several different authors in a special
issue on SEM in the journal Personality and Individual Differences (Vernon & Eysenck,
2007). Briefly, there is a general consensus that (a) standard practices for evaluating
models in SEM have been lax and, consequently, (b) researchers need to take a more
rigorous, skeptical, and disciplined approach to hypothesis testing. How to do so is a
major theme of Chapter 8 and indeed of the whole book.
5. There is more coverage in this edition of two advanced topics in SEM: the estima-
tion of interactive effects of observed or latent variables and multilevel analysis (Chapter
12). Many developments have taken place recently in each of these areas, and more and
more researchers are estimating models in which these types of effects are represented.
Accordingly, the chapter on how to fool yourself with SEM (Chapter 13) is now expanded
to include the failure to consider these types of effects, among other more prosaic ways
to become irrational with SEM.
6. Several chapters feature topic boxes about concepts, techniques, or specialized
issues in the conduct of SEM. These boxes offer relatively short summaries of topics that
complement or elaborate on the presentation in the main text. More advanced topics are
covered in chapter appendices, which allows readers of various skill levels to get more
out of the book.
C. Deborah Laughton, Publisher, Methodology and Statistics, at The Guilford Press,
has a special knack for giving me exactly the type of feedback I need at precisely the
right moment in the writing process. She collected reviews of the second edition and
drafts for the third edition from a variety of scholars with differing backgrounds and
levels of experience, from those just learning about SEM to renowned professors whose
work is very widely known in their respective fields. C. Deborah sent these reviews
Preface and Acknowledgments xi
to me without identifying their authors, and the content of the reviews was extremely
helpful in the planning and writing of this edition. C. Deborah, thanks again for all
your work and support. The names of the reviewers were revealed to me only after the
writing was done, and their original comments were not associated with their names. A
big thanks to all the persons listed next (in alphabetical order) who put in a lot of time
and effort to communicate their thoughts about the book in various stages of its writing;
their comments and suggestions were invaluable:
Alan C. Acock, Department of Human Development, Oregon State University•
Noel A. Card, John and Doris Norton School of Family and Consumer Sciences, •
Division of Family Studies and Human Development, University of Arizona
David F. Gillespie, Department of Social Work, Washington University in St. Louis•
Debbie Hahs-Vaughn, College of Education, Department of Educational Research, •
Technology, and Leadership, University of Central Florida
Lance Holbert, Department of Communications, Ohio State University•
Jacob Marszalek, School of Education, Research and Psychology, University of •
Missouri–Kansas City
Richard A. Posthuma, College of Business Administration, University of Texas at •
El Paso
James Schreiber, School of Education, Department of Foundations and Leadership, •
Duquesne University
Greg Welch, School of Education, Department of Psychology and Research in Edu-•
cation, University of Kansas
Craig Wells, School of Education, Department of Educational Policy, Research, and •
Administration, University of Massachusetts at Amherst
Duan Zhang, Morgridge College of Education, Quantitative Research Methods, •
University of Denver
It was a pleasure to work with the Methodology in the Social Sciences Series Editor
at Guilford, Todd D. Little, in putting together the final version of this book. His com-
ments were very helpful, and it was a pleasure to meet Todd when he visited Concordia
University in Montréal in November 2009. Betty Pessagno served as the copyeditor for
the original manuscript, and her work and suggested changes improved the clarity of
the presentation. I also appreciate the efforts of the Guilford production editor, Wil-
liam Meyer, in preparing the final version of this book. I asked Lesley Hayduk of the
Department of Sociology at the University of Alberta to review a draft of Chapter 8 about
hypothesis testing in SEM. Les has long advocated for a more rigorous approach to test-
ing in SEM, and the rest of the field is catching up to this viewpoint. I was hoping that
Les’s comments would give the final version of Chapter 8 more backbone, and I was not
disappointed. Thanks, Les, for saying the kinds of things I needed to hear about this
crucial topic.
The most recent versions of computer tools for SEM were generously provided for
me by Multivariate Software (EQS), Muthén and Muthén (Mplus), and Scientific Soft-
xii Preface and Acknowledgments
ware International (LISREL). In particular, I wish to thank Linda Muthén and Peter
Bentler for their comments on earlier drafts of descriptions of, respectively, Mplus and
EQS. And once again, my heartfelt thanks to my wife, Joanna, and children, Julia and
Luke, for all their love and support while writing this book.
R
e x B. Kl i n e
Montréal
rex.kline@concordia.ca
xiii
Contents
Part I. ConCePts and tools
1 • Introduction 3
The Book’s Website 3
Pedagogical Approach 4
Getting Ready to Learn about SEM 5
Characteristics of SEM 7
Widespread Enthusiasm, but with a Cautionary Tale 13
Family History and a Reminder about Context 15
Extended Latent Variable Families 16
Plan of the Book 17
Summary 18
2 • Fundamental Concepts 19
Multiple Regression 19
Partial Correlation and Part Correlation 28
Other Bivariate Correlations 31
Logistic Regression 32
Statistical Tests 33
Bootstrapping 42
Summary 43
Recommended Readings 44
Exercises 45
3 • data Preparation 46
Forms of Input Data 46
Positive Definiteness 49
Data Screening 51
Selecting Good Measures and Reporting about Them 68
Summary 72
Recommended Readings 72
Exercises 73
4 • Computer tools 75
Ease of Use, Not Suspension of Judgment 75
Human–Computer Interaction 77
xiv Contents
Core SEM Programs and Book Website Resources 77
Other Computer Tools 86
Summary 87
Recommended Readings 87
Part II. Core teChnIques
5 • specification 91
Steps of SEM 91
Model Diagram Symbols 95
Specification Concepts 96
Path Analysis Models 103
CFA Models 112
Structural Regression Models 118
Exploratory SEM 121
Summary 121
Recommended Readings 122
Exercises 122
6 • Identification 124
General Requirements 124
Unique Estimates 130
Rule for Recursive Structural Models 132
Rules for Nonrecursive Structural Models 132
Rules for Standard CFA Models 137
Rules for Nonstandard CFA Models 138
Rules for SR Models 144
A Healthy Perspective on Identification 146
Empirical Underidentification 146
Managing Identification Problems 147
Summary 148
Recommended Readings 149
Exercises 149
aPPendIX 6.a. Evaluation of the Rank Condition 151
7 • estimation 154
Maximum Likelihood Estimation 154
Detailed Example 160
Brief Example with a Start Value Problem 172
Fitting Models to Correlation Matrices 175
Alternative Estimators 176
A Healthy Perspective on Estimation 182
Summary 182
Recommended Readings 183
Exercises 183
aPPendIX 7.a. Start Value Suggestions for Structural Models 185
aPPendIX 7.B. Effect Decomposition in Nonrecursive Models 186
and the Equilibrium Assumption
aPPendIX 7.C. Corrected Proportions of Explained Variance 187
for Nonrecursive Models
Contents xv
8 • hypothesis testing 189
Eyes on the Prize 189
State of Practice, State of Mind 190
A Healthy Perspective on Fit Statistics 191
Types of Fit Statistics and “Golden Rules” 193
Model Chi-Square 199
Approximate Fit Indexes 204
Visual Summaries of Fit 209
Recommended Approach to Model Fit Evaluation 209
Detailed Example 210
Testing Hierarchical Models 214
Comparing Nonhierarchical Models 219
Power Analysis 222
Equivalent and Near-Equivalent Models 225
Summary 228
Recommended Readings 228
Exercises 229
9 • Measurement Models and Confirmatory Factor analysis 230
Naming and Reification Fallacies 230
Estimation of CFA Models 231
Detailed Example 233
Respecification of Measurement Models 240
Special Topics and Tests 241
Items as Indicators and Other Methods for Analyzing Items 244
Estimated Factor Scores 245
Equivalent CFA Models 245
Hierarchical CFA Models 248
Models for Multitrait–Multimethod Data 250
Measurement Invariance and Multiple-Sample CFA 251
Summary 261
Recommended Readings 262
Exercises 262
aPPendIX 9.a. Start Value Suggestions for Measurement Models 263
aPPendIX 9.B. Constraint Interaction in Measurement Models 264
10 • structural regression Models 265
Analyzing SR Models 265
Estimation of SR Models 269
Detailed Example 270
Equivalent SR Models 276
Single Indicators in Partially Latent SR Models 276
Cause Indicators and Formative Measurement 280
Invariance Testing of SR Models 288
Reporting Results of SEM Analyses 289
Summary 293
Recommended Readings 293
Exercises 294
aPPendIX 10.a. Constraint Interaction in SR Models 295
xvi Contents
Part III. advanCed teChnIques, avoIdIng MIstakes
11 • Mean structures and latent growth Models 299
Logic of Mean Structures 299
Identification of Mean Structures 303
Estimation of Mean Structures 304
Latent Growth Models 304
Structured Means in Measurement Models 316
MIMIC Models as an Alternative to Multiple-Sample Analysis 322
Summary 325
Recommended Readings 326
12 • Interaction effects and Multilevel seM 327
Interaction Effects of Observed Variables 327
Interaction Effects in Path Models 331
Mediation and Moderation Together 333
Interactive Effects of Latent Variables 336
Estimation with the Kenny–Judd Method 337
Alternative Estimation Methods 340
Rationale of Multilevel Analysis 343
Basic Multilevel Techniques 345
Convergence of SEM and MLM 348
Multilevel SEM 350
Summary 354
Recommended Readings 354
13 • how to Fool Yourself with seM 356
Tripping at the Starting Line: Specification 356
Improper Care and Feeding: Data 359
Checking Critical Judgment at the Door: Analysis and Respecification 361
The Garden Path: Interpretation 363
Summary 366
Recommended Readings 366
suggested answers to exercises 367
references 387
author Index 405
subject Index 411
about the author 427
Part I
Concepts and Tools
3
1
Introduction
The book is intended to serve as a guide to the principles, assumptions, strengths,
limitations, and application of structural equation modeling (SEM) for researchers and
students who do not have extensive quantitative backgrounds. Accordingly, the pre-
sentation is conceptually rather than mathematically oriented, the use of formulas and
symbols is kept to a minimum, and many examples are offered of the application of
SEM to research problems in various disciplines, including psychology, education,
health sciences, marketing, and management. When you finish reading this book, I
hope that you will have acquired the skills to begin to use SEM in your own research in
an informed, disciplined way. The following adage attributed to poet Eugene F. Ware
is pertinent here: All glory comes from daring to begin. Let’s do just that.
the Book’s WeBsIte
This book has a website on the Internet; the address is www.guilford.com/kline
From the site, you can freely access or download the following resources:
Computer files for every example of SEM analyses in Chapters 7–12 for three widely •
used SEM computer tools—EQS, LISREL, and Mplus.
Links to related web pages, including sites with more information about computer •
data analysis in SEM.
A supplemental reading about the estimation of curvilinear effects of observed and •
latent variables in SEM.
The purpose of the website for this book is to support a learning-by-doing approach
to SEM. Specifically, the availability of both data summaries and syntax files means that
you can reproduce the analyses for most of the examples in this book using the corre-
sponding SEM computer tool. Even without access to a particular program, such as EQS,