Structural equations with latent variables /

A graduate-level book intended primarily for social scientists, applied statisticians, market researchers, and other analysts. Structural equation system (also known as the LISREL model) provides a unified treatment of path analysis, classical econometrics, confirmatory factor analysis, and recursive and nonrecursive models. It also emphasizes the role played by substantive expertise in various stages of the modeling process. An interesting feature of this well-written and useful book is the inclusion of numerous empirical examples emphasizing the applications of the techniques presented. To be able to read it comfortably, one needs to have a reasonable knowledge of linear algebra, regression analysis, and asymptotic distribution theory. After providing a brief history of the general structural equation models, Bollen discusses various methodological tools in the form of model notation, covariances and covariance algebra, and path analysis; then he deals with the problem of causality, and with regression/econometric models for observed variables. The topic of random measurements error in observed variable models is discussed in Chapter 5, whereas the relationship between the error-free variable and the observed variable is examined in Chapter 6. After a presentation of confirmatory factor analysis in Chapter 7, the last two chapters are set aside for the general structural equation model with latent variables. Each chapter is followed by a summary. There is a reasonable list of references and a subject index, but no author index. Bollen includes some new material on measurement reliability and validity, model identification, and overall fit indices. Two appendixes, one on matrix algebra and another on asymptotic distribution theory, are also included. -D. V. Chopra, Wichita State University