ABSTRACT

It is well known that structural equation modeling (SEM) has become one of the most popular methods in multivariate analysis, especially in the social and behavioral sciences. In a SEM model with latent variables, the relationships among observed (manifest) variables is formulated through unobserved (latent) constructs. Because measurement errors are explicitly accounted for, coefficients in key parts of a model are uninfluenced by errors of measurement, implying greater theoretical meaningfulness and cross-population stability to the parameters than might be achieved with methods such as regression or analysis of variance, which do not correct for unreliability. This stability is a key goal of theory testing with SEM, where a substantive theory or hypothesized causal relationship among the latent constructs, facilitated by path diagrams, can be tested through SEM. With the help of popular software such as LISREL (Jöreskog & Sörbom, 1993) and EQS (Bentler, 2001), applications as well as new technical developments in SEM have increased dramatically in the past decade (e.g., Austin & Calderόn, 1996; Austin & Wolfle, 1991; Bollen, 1989; Tremblay & Gardner, 1996). There exists a vast amount of recent introductory (Byrne, 1994; Dunn, Everitt, & Pickles, 1993; Kline, 1998; Mueller, 1996; Schumacker & Lomax, 1996) and overview material (Bentler & Dudgeon, 1996; Browne & Arminger, 1995; Hoyle, 1995; Marcoulides & Schumacker, 1996).