ABSTRACT
Structural equation modeling (SEM), also known as path analysis with latent variables, is a general system for the analysis of dependencies/independencies in a set of measured variables and/or common factors of subsets of variables. It is rapidly gaining in popularity as a way to investigate “causal” relations in nonexperimental data. This is primarily because of the ease with which an investigator can draw a path diagram to express a conjecture about cause–effect relations, and “confirm” it with increasingly user-friendly commercial software. Given the usual caveats about drawing causal inferences from “mere” correlations, and for other reasons that appear later in this chapter, we might characterize it as a dangerously conjectural technique for asking essential research questions which otherwise are impossible to consider. Properly understood, and carefully used, it does have the advantage over less systematic modes of analysis intended to support causal theories that it makes its own assumptions and limitations explicit. Superficially applied, it is an easy way to give an impressive and plausible but possibly totally incorrect account of a set of measurements in the social and behavioral sciences. Applications of the second kind are possibly in the majority.
