Chapter 5: Evaluating Between-Group Differences in Latent Variable Means

Researchers often seek to compare groups hypothesized to differ on one or more outcome variables. In the simplest form, groups may be compared on a single, directly observable variable that is believed to be measured reliably. On the other hand, research questions frequently involve group differences on multiple outcomes that are thought to represent one or more constructs. Although multivariate analysis of variance (MANOVA) persists as a commonly applied method for analyzing mean differences on a set of outcomes, methodologists have argued that researchers should use structural equation modeling (SEM) methods for comparing groups on latent variable means if multiple dependent variables represent each of the underlying constructs of interest (e.g., Aiken, Stein, & Bentler, 1994; Cole, Maxwell, Arvey, & Salas, 1993; Green & Thompson, 2003a; Hancock, 1997). In this chapter, we discuss the utility of SEM for evaluating between-group differences on latent variable means and illustrate SEM approaches for doing so. Specifically, we address the following points: