ABSTRACT

The wide use of structural equation models with unmeasured variables, and the widespread adoption of the computer programs developed by Jöreskog ([5]; Jöreskog and Sörbom [7]; Jöreskog and van Thillo [8]) to estimate and assess such models have introduced a new vocabulary into the analysis of data by sociologists. A decade ago, a conversation about “specification,” “identification,” “goodness of fit,” “residuals,” and “correlated error” would have been incomprehensible to most sociologists. Methodological and pedagogical papers, especially the review by Long [11] and numerous publications reporting on the analysis of substantively interesting “multiple indicator models” have now made such terms a part of the recognition vocabulary of almost all quantitative sociologists. Furthermore, a substantial number of sociologists now have a reasonably sophisticated comprehension of the reasoning underlying structural equation models with unmeasured variables, of the available techniques for estimating the parameters of such models, and of some of the indices for assessing the fit between model and data.