This article was downloaded by: [Guru Nanak Dev University] On: 19 December 2012, At: 20:33 Publisher:

Psychology Press Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Structural Equation Modeling: A Multidisciplinary Journal

Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/hsem20

A Beginner's Guide to Structural Equation Modeling (3rd ed.)

Tiffany A. Whittaker
a a

The University of Texas at Austin Version of record first published: 17 Oct 2011.

To cite this article: Tiffany A. Whittaker (2011): A Beginner's Guide to Structural Equation Modeling (3rd ed.), Structural Equation Modeling: A Multidisciplinary Journal, 18:4, 694-701 To link to this article: http://dx.doi.org/10.1080/10705511.2011.607726

PLEASE SCROLL DOWN FOR ARTICLE Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material.

Structural Equation Modeling, 18:694701, 2011 Copyright Taylor & Francis Group, LLC ISSN: 1070-5511 print/1532-8007 online DOI: 10.1080/10705511.2011.607726

BOOK REVIEW

Downloaded by [Guru Nanak Dev University] at 20:33 19 December 2012

A Beginners Guide to Structural Equation Modeling (3rd ed.). Randall E. Schumacker and Richard G. Lomax. New York, NY: Routledge, 2010, 536 pages, $59.95 (softcover). Reviewed by Tiffany A. Whittaker The University of Texas at Austin As the popularity of structural equation modeling (SEM) grows, so does the number of introductory SEM textbooks that can be considered for adoption in a rst course (e.g., Bollen, 1989; Kaplan, 2009; Kline, 2010; Loehlin, 2004; Raykov & Marcoulides, 2006). This is the third edition of the Schumacker and Lomax textbook, and their coverage of SEM in this new editions reects their collective experience using the previous edition of the text in their own classes over the years as well as growth in the eld of SEM. This review is based exclusively on the contents in the third edition. A short introduction describing the purpose of the textbook and its new additions is provided rst, followed by summaries of the chapters in the text. These summaries might also include some commentary concerning certain features within a chapter. Subsequently, an overall evaluation of the text is presented, followed by a general conclusion discussing the appropriate audiences for the text. The purpose of the textbook is to introduce SEM using simple and easy-to-understand examples so that students and researchers can perform analyses in SEM independently and be more informed readers of SEM research. Schumacker and Lomax indicate that a number of new surprises are offered in the third edition. These surprises include the following: Step-by-step examples in LISREL 8.8 software (student version), which is freely available for download. More examples in more disciplines. Updated chapters with added material. An associated Web site where raw data sets used in the textbooks examples and exercises, links to Web sites and online resources, and answers to the entire set of the textbooks exercises for instructors are available. Extended treatment of advanced structural equation models such as multiple-group modeling, multilevel SEM, mixture modeling, second-order factor models, dynamic factor models, and Monte Carlo techniques.
Correspondence should be addressed to Tiffany A. Whittaker, The University of Texas at Austin, Educational Psychology, 1 University Station D5800, Austin, TX 78712, USA. E-mail: tiffany.whittaker@mail.utexas.edu

694

BOOK REVIEW

695

Downloaded by [Guru Nanak Dev University] at 20:33 19 December 2012

Extended treatment of power and sample size using software programs and reporting results of research. New references to journal articles to help further readers understanding of research published in SEM. Tips on troubleshooting and addressing problems encountered in SEM in Chapters 3 and 11. Additional examples of structural equation models in Chapters 13 through 16. New exercises (25%) with answers to odd-numbered questions provided in the text. More matrix model examples in Chapter 17. Updated references for fundamental topics in all the chapters.

CHAPTER SUMMARIES Chapter 1 presents a general introduction to SEM and mentions the topics and types of structural equation models that are covered in the remainder of the text. A history lesson on the developments in SEM is provided, as well as why SEM is popular among researchers. Available SEM software programs are mentioned, although the focus in the text is on LISREL software. As indicated by Schumacker and Lomax, one of the new additions to the third edition is the use of LISREL 8.8 (student version) when demonstrating the structural equation model examples in the text. Chapter 2 continues the focus on LISREL software by briey explaining the LISRELPRELIS program with examples using LISREL 8.8 (student version) to demonstrate data entry. Data considerations and the importance of data screening prior to conducting statistical analyses are briey discussed, such as scales of measurement, restriction of range, missing data handling options in LISREL-PRELIS with an imputation example provided, outlier detection methods in LISREL-PRELIS, linearity, nonnormality detection, and solutions to correct for nonnormality (e.g., data transformations, weighted least squares estimation, asymptotically distribution free estimation). In Chapter 3, types of correlations, ranging from the Pearson correlation to the polychoric correlation, are quickly reviewed with respect to scales of measurement. Topics related to the factors that impact correlation coefcients, including measurement scale, range of scores, nonlinearity, missing data, outliers, attenuation, Heywood cases, and sample size are discussed. The remainder of the chapter discusses semipartial and partial correlations (with formulas included), the relation between correlation and covariance, standardized versus unstandardized variables, and causation versus inference. Chapter 4 introduces the fundamentals of SEM, which adhere to the following ve-step sequence: (1) model specication, (2) model identication, (3) model estimation, (4) model testing, and (5) model modication. Model specication is presented in the context of correct specication of the model in that the model includes relevant variables or excludes irrelevant variables. The model identication subsection denes models that are underidentied, justidentied, and overidentied, as well as conditions, such as the order and rank conditions, for establishing whether a model is identied. Model estimation techniques (ULS, OLS, GLS, ML, WLS) with respect to data considerations (continuous or categorical variables and nonnormality) are introduced in approximately three pages. The discussion of model estimation seems a bit limited with respect to the depth of coverage, as well as the comparability between estimation

696

WHITTAKER

methods in LISREL and other SEM software. For instance, certain concepts (e.g., implied covariance matrix, asymptotic covariance matrix used as input in LISREL, SatorraBentler scaled chi-square) are introduced but are not elaborated upon. Although a full discussion of some of these concepts might be beyond the scope and purpose of the text, examples of implementing these different estimation methods in LISREL, which are not provided here, would be valued. Diagonally weighted least squares (DWLS) estimation, which is available in LISREL, is not mentioned in the chapter. Further, describing the similarities between estimation methods in LISREL and other SEM software [e.g., DWLS with the appropriate asymptotic covariance matrix provided and weighted least squares with mean-adjusted chi-square (WLSM) or weighted least squares with mean- and variance-adjusted chi-square (WLSMV) in Mplus] would facilitate better understanding when reading SEM research in which different estimation procedures within different SEM software platforms are implemented. Model testing in Chapter 4 is presented with respect to model t and parameter estimate signicance. Model modication introduces the concepts associated with a specication search using parameter statistical signicance, modication indices, expected parameter change estimates, and residual matrices. The model modication subsection concludes with a nicely suggested eight-step process that should be considered during the model respecication process. These ve fundamentals are subsequently discussed when certain types of models are presented in later chapters, such as regression models, path models, conrmatory factor analysis models, structural equation models, multiple-indicator multiple-cause (MIMIC) models, and mixture models. In Chapter 5, model t indices (with formulas and sample model t output from a two-factor conrmatory factor model example using LISREL-SIMPLIS) and model t issues (model t criteria, model comparisons, model parsimony, and parameter t examination) are comprehensively discussed. Another part of Chapter 5 is expansively devoted to power and sample size in which post-hoc power examples in SAS, SPSS, and G*Power 3 (Faul, Erdfelder, Lang, & Buchner, 2007); sample size determination using the Critical N statistic (Hoetler, 1983), SAS, and G*Power 3 in a sensitivity analysis; and power analysis using the root mean square error of approximation (RMSEA) as suggested by MacCallum, Browne, and Sugawara (1996) are presented. Model comparisons using likelihood ratios for nested models, power to detect a difference in the RMSEA as suggested by MacCallum, Browne, and Cai (2006) using a SAS program, and power for testing the signicance of a parameter estimate using SAS are also presented. An overview of other methods that are available to compute power (e.g., Muthn & Muthn, 2002) is provided, in addition to references that indicate how other factors (e.g., sample size and missing data) might impact power. The chapter ends with a brief description of two-step SEM and four-step SEM with suggested recommendations. Chapter 6 briey presents multiple regression models in which standardized and unstandardized partial regression coefcients, R2 , and adjusted R2 are reviewed. Useful illustrations of calculating effect sizes, building condence intervals around an R2 value, adding and dropping an independent variable from the model, and a calculation of an F test to determine whether a statistically signicant difference exists between two regression models are provided. An example of a multiple regression analysis is presented to help with concepts such as model specication, identication, estimation, model testing, and model modication, which were previously discussed in Chapter 4. The example is run in LISREL-SIMPLIS with sample output and interpretation. Adjustments for measurement error and the additivity of multiple

Downloaded by [Guru Nanak Dev University] at 20:33 19 December 2012

BOOK REVIEW

697

regression models are also discussed. Chapter 7 extends multiple regression models by discussing path models. A path model example is used to help illustrate model specication, model identication, model estimation, model testing, and model modication concepts. Path diagram symbols are also presented and discussed. The LISREL-SIMPLIS program associated with the path model example is given in the Appendix of Chapter 7. In Chapter 8, conrmatory factor analysis (CFA) is presented. The chapter begins with a brief discussion of the limitations of using path models with respect to measurement error and the distinction between exploratory factor analysis and CFA. A two-factor correlated model is introduced that helps to further illustrate SEM diagram symbols as presented in Chapter 7 and is used to briey describe modeling options in CFA, such as correlated factors and correlated errors. This example is subsequently used when discussing model specication, identication, estimation, model testing, and model modication with the LISREL-SIMPLIS program provided in the Chapter 8 Appendix. Although the two options for setting the measurement scale for latent factors (i.e., setting a factor variance to 1.0 or setting one indicator variables factor loading to 1.0 for a given factor) were quickly presented in the model identication subsection in Chapter 4 (SEM Basics), elaboration on this issue would be a nice addition to this chapter. Chapters 9 and 10 both describe developing structural equation models. In Chapter 9 (Developing Structural Equation Models: Part I), observed variables are again distinguished from latent variables and latent independent (exogenous) and latent dependent (endogenous) variables are introduced. The benets of using more than one indicator per factor are presented using an informative illustration of how this relates to reliability and error variance. CFA models are recapped and used to introduce structural models. Variance and covariance estimates in a structural model example are briey discussed, followed by a short discussion of the two-step and four-step modeling approaches. In Chapter 10 (Developing Structural Equation Models: Part II), another structural equation model is used to illustrate model specication, identication, estimation, model testing, and model modication concepts. Again, the LISRELSIMPLIS program associated with the structural model example is provided in the Appendix of Chapter 10. Chapter 11 provides useful guidelines and recommendations for reporting research in SEM. Helpful checklists are provided for information to include in the research document as well as for data preparation, model specication, model identication, model estimation, model testing, model modication, and model validation. Model validation is well discussed in Chapter 12. Validation of models using multiple samples is introduced and illustrations using LISRELSIMPLIS are provided. Cross-validation of models is presented using cross-validation indices such as the expected cross-validation index (ECVI; Browne & Cudeck, 1989) and the crossvalidation index (CVI; Cudeck & Browne, 1983) with LISREL-SIMPLIS demonstrations. Model validation using bootstrapping is also introduced using PRELIS. Chapter 13 begins to introduce additional types of structural equation models. Schumacker and Lomax indicate that their intention in Chapter 13, as well as in subsequent chapters, is to simply introduce additional types of models and refer readers to additional readings for each topic area. Multiple-sample, multiple-group, and structured means models are illustrated in Chapter 13. An example of each type of model is presented with LISREL-SIMPLIS sample program syntax and excerpts from the respective output. A discussion of invariance (e.g., metric, strong, and strict invariance), including testing procedures and implications (e.g., releasing constraints), as well as assumptions associated with structured means models are important

Downloaded by [Guru Nanak Dev University] at 20:33 19 December 2012

698

WHITTAKER

topics that were not included in this chapter. Further, the structured means model diagram displayed in the chapter does not incorporate the mean structure with the use of the constant pseudo-variable (typically represented as a triangle containing a constant value of 1.0), which would more effectively illustrate the concept of integrating and estimating means in a model. Chapter 14 continues briey introducing additional types of structural equation models such as second-order, dynamic, and multitrait-multimethod models. Again, an example of each type of model is used with LISREL-SIMPLIS sample syntax and excerpts of relevant output are provided. Chapter 15 covers MIMIC, so-called mixture, and multilevel models. A MIMIC model is presented in which three observed variables, treated as continuous, predict a latent factor with three indicator variables. The MIMIC model is analyzed using LISREL-SIMPLIS. A presentation of a MIMIC model in which a dummy-coded variable (group code variable) predicts a latent factor would have been welcomed to help interpret latent mean differences between groups as well as to present an alternative to structured means modeling, which was briey demonstrated in Chapter 13. The subsection introducing what the authors term mixture models begins by indicating that mixture models in SEM involve the analysis of observed variables that are categorical and continuous (p. 298). An example using continuous and ordinal indicator variables is demonstrated using PRELIS (with a polyserial correlation matrix as input) and steps are provided to obtain the SatorraBentler scaled chi-square statistic in PRELIS. The so-called mixture model presented in the chapter should not be confused with a latent class model or a factor mixture model in which parameters are estimated for different populations in the data set with the lack of manifest grouping information; nonetheless, references for latent class and factor mixture models in this context are provided at the end of the chapter. A multilevel model with repeated measures is presented with step-by-step instructions using LISREL to analyze the example model. Results from the analyses including the intraclass correlation and deviance statistic are discussed in brief. Chapter 16 introduces interaction models, latent growth models, and Monte Carlo methods. Interaction models are illustrated by (1) using multiple-group modeling in the context of categorical observed variables using LISREL-SIMPLIS; (2) creating latent variable scores and latent variable interaction scores in PRELIS, which are subsequently used in a path analysis model example in LISREL-SIMPLIS; and (3) using a two-stage least squares approach in PRELIS and SIMPLIS. The difculties inherent to modeling latent variable interactions (e.g., model specication, multicollinearity, distributions of variables) are briey discussed. A latent growth model example is succinctly presented using LISREL-SIMPLIS. Further discussion of slope and factor parameterization options, a discussion concerning identication, the incorporation of latent variable predictor variables, and elaboration on interpreting important parameters in a latent growth model would be welcomed. The chapter ends by providing examples of various Monte Carlo methods that can be used to generate population data, including a step-by-step data simulation example using PRELIS; examples in SPSS, SAS, and LISREL to generate population data from a specied covariance matrix; and an example using a covariance matrix from a specied model in SPSS, which is then used in LISRELSIMPLIS. Supplementary references concerning Monte Carlo studies are included at the end of the chapter. A valuable presentation of matrix notation in SEM is covered in Chapter 17. SEM model examples from different chapters are used to help illustrate free, xed, and constrained parameters

Downloaded by [Guru Nanak Dev University] at 20:33 19 December 2012

BOOK REVIEW

699

in the various matrices used for different types of models and the matrix command language in LISREL. More specically, the structural equation models from Chapters 9 and 10, the path model from Chapter 7, the structured means model from Chapter 13, and a modied interaction model from Chapter 16 are used in these illustrations. Multiple-sample model matrix notation is also briey provided.

EVALUATION I would like to preface the evaluation of this textbook with the acknowledgment that it is difcult for authors to know what topics to include, in what order the topics should be presented, and how thoroughly the included topics should be discussed in any written document. Every reader will not typically be fully satised with all of the contents of a textbook covering such a wide-ranging topic as SEM. In addition, tackling the writing of a textbook is praiseworthy in itself. On that note, Schumacker and Lomax have included many nice features in their textbook. For example, the inclusion of chapters on data entry and editing, guidelines and recommendations on reporting SEM research, model validation, matrix notation in SEM, and power makes this textbook appealing given that these topics are often overlooked in introductory SEM textbooks. The added examples in additional disciplines throughout the text will help students and SEM researchers from different disciplines understand the wide-reaching applicability of SEM and better conceptualize the use of SEM in their respective elds. The textbooks associated Web site will allow access to the raw data sets to practice using LISREL and replicate the textbooks examples and will provide access to online resources to students in classes for which the instructor has adopted the textbook. Further, the exercises provided at the end of each chapter will allow additional practice not only running the models presented in the text, but also with concepts covered in the chapter. Again, a new feature of the textbook is the use of LISREL 8.8 (student version) when presenting model examples and topics. Because the student version is freely available and SEM users will either solely use LISREL or might want to learn LISREL in addition to other SEM software programs, many might view the focus on LISREL as valuable. On the other hand, some might view the focus on LISREL software in the text as restricting a more extensive discussion of certain topics (e.g., theoretical background and uses pertaining to certain models, comparability of estimation methods across software programs). In a similar vein, the limited discussion of the more advanced techniques in SEM in Chapters 13 through 16 along with step-by-step instructions in LISREL, although commendable, might provide inexperienced SEM users with the knowledge of how to analyze such models (e.g., multilevel models in SEM), but not necessarily the requisite skills needed when analyzing or interpreting such models. As even Schumacker and Lomax point out, SEM software programs are now easier to use and contain features similar to other Windows-based software packages. However, such ease of use necessitates statistical training in SEM modeling and software via courses, workshops, or textbooks to avoid mistakes and errors in analyzing sophisticated theoretical models (p. 7). Although Schumacker and Lomax do indicate that they are simply introducing more advanced models and refer readers to additional resources concerning advanced SEM topics, key references covering some of the advanced topics seem

Downloaded by [Guru Nanak Dev University] at 20:33 19 December 2012

700

WHITTAKER

Downloaded by [Guru Nanak Dev University] at 20:33 19 December 2012

to be missing from the reference lists at the end of these chapters. In addition, it might better serve readers in a rst course to have more elaboration concerning the purposes of certain models, the appropriate procedures when analyzing certain models, the interpretation of important parameters in certain models, and the assumptions associated with certain models. These advanced chapters could then extend the coverage of topics, including multiple-group modeling, structured means modeling, MIMIC modeling, and latent growth modeling. With a few minor exceptions, the order of topics presented in the text generally facilitates the learning of SEM topics. Because Chapter 4 (SEM Basics) and Chapter 5 (Model Fit) introduce concepts not yet seen in any context (e.g., model identication, model estimation, model testing, chi-square test statistic, calculation of t indices), they might be better placed after readers have accessed Chapter 6 (Regression Models) and Chapter 7 (Path Models). Similarly, information concerning model estimation and model t, given their connection, might be better placed in the same chapter. Chapter 11 (Reporting SEM Research) and Chapter 12 (Model Validation) might be more informative if they followed Chapter 16 (Interaction, Latent Growth, and Monte Carlo Methods) to help reinforce Schumacker and Lomaxs message that model t is a subjective approach that requires substantive theory because there is no single best model (p. 219) and that theoretical models need to be replicated, cross-validated, and/or bootstrapped to determine the stability of the parameter estimates (p. 219). Further, the presentation of Monte Carlo methods in Chapter 16 might be more accessible to students if presented in a chapter subsequent to Chapter 17 (Matrix Approach to Structural Equation Modeling).

CONCLUSION Schumacker and Lomax indicated that the main goal in this third edition is for students and researchers to be able to conduct their own SEM model analyses, as well as be able to understand and critique published SEM research (p. xviii). The applied nature and nontechnical introduction of SEM with several examples are denite strengths of their textbook. The main weakness of the text, however, would be the limited discussion of certain topics (e.g., model estimation, invariance testing, model assumptions) and certain models (e.g., multiple-group, structured means, MIMIC, latent growth). Accordingly, this text would be appropriate to use in an introductory SEM course in which LISREL software is used. Experienced SEM users who are interested in learning LISREL or in acquiring a useful reference text would also appreciate this textbook. For those who are interested in learning the technical underpinnings of SEM, other texts (e.g., Bollen, 1989) might offer more of this type of exposure.

REFERENCES
Bollen, K. A. (1989). Structural equations with latent variables. New York, NY: Wiley. Browne, M. W., & Cudeck, R. (1989). Single sample cross-validation indices for covariance structures. Multivariate Behavioral Research, 24, 445455. Cudeck, R., & Browne, M. W. (1983). Cross-validation of covariance structures. Multivariate Behavioral Research, 18, 147167.

BOOK REVIEW

701

Downloaded by [Guru Nanak Dev University] at 20:33 19 December 2012

Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G*Power 3: A exible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods, 39, 175191. Hoetler, J. W. (1983). The analysis of covariance structures: Goodness-of-t indices. Sociological Methods and Research, 11, 325344. Kaplan, D. W. (2009). Structural equation modeling: Foundations and extensions (2nd ed.). Thousand Oaks, CA: Sage. Kline, R. B. (2010). Principles and practice of structural equation modeling (3rd ed.). New York, NY: Guilford. Loehlin, J. C. (2004). Latent variable models (4th ed.). Mahwah, NJ: Lawrence Erlbaum Associates, Inc. MacCallum, R. C., Browne, M. W., & Cai, L. (2006). Testing differences between nested covariance structure models: Power analysis and null hypotheses. Psychological Methods, 11, 1935. MacCallum, R. C., Browne, M. W., & Sugawara, H. M. (1996). Power analysis and determination of sample size for covariance structure modeling. Psychological Methods, 1, 130149. Muthn, B. O., & Muthn, L. K. (2002). How to use a Monte Carlo study to decide on sample size and determine power. Structural Equation Modeling, 9, 599620. Raykov, T., & Marcoulides, G. A. (2006). A rst course in structural equation modeling (2nd ed.). Mahwah, NJ: Lawrence Erlbaum Associates, Inc.