UC Santa Barbara

Moderation analyses with a latent class variable allows researchers to study relations among exogenous (predictor or covariate) and endogenous (distal outcome) variables across the latent classes. Extending the latent class model to include auxiliary variables (both a predictor and an outcome) creates a context where the latent class variable serves as a moderating variable, referred to as a mixture regression. This thesis provides a pedagogical introduction on how to specify and interpret the moderation model with a latent class variable with the (ML) three-step manual approach (Asparouhov & Muthén, 2014) using data from the Longitudinal Survey of American Life (LSAL). Specifically, the latent class variable (science attitude classes) is hypothesized to moderate the relation between a predictor (science ability) and outcome (interest in science issues), controlling for the demographic variables. This thesis will outline the steps of the ML three-step approach, appropriate tables and visualizations used, and accompanying Mplus syntax.