Session: Advances in Regression Mixture Modeling (Society for Prevention Research 22nd Annual Meeting)

The search for differential effects, or individual differences in the relationship between a predictor and an outcome, has received increased attention in prevention science. There is growing interest in understanding why individuals exposed to similar conditions of risk or the same intervention experience variability in outcomes.  Regression mixture models are an exploratory approach for answering these type of research questions aimed at identifying heterogeneity in contextual or treatment effects. Regression mixture modeling falls in the broad class of statistical models called finite mixture models, which utilize empirically-derived and qualitatively-distinct latent classes to explain population heterogeneity in an unknown distribution. Unlike the use of statistical interactions, which test whether the effects of a predictor on an outcome vary as a function of a pre-specified, third variable, regression mixtures explore the data for evidence of groups of respondents who differ in the effects of the predictor on the outcome in magnitude and/or direction. The method offers promise for deepening our understanding of the more nuanced ways that risk and protective factors impact individual’s health. However, for prevention scientists to effectively use this relatively new methodology we need to better understand the conditions under which the methods work, how to correctly specify them, and to continue developing regression mixtures for use with common study designs. In this symposium, we present three papers that extend regression mixture models to allow for the examination of cross-level heterogeneity in effects and test the capabilities of these models under conditions similar to those used by prevention scientists.  The first paper examines the effects of an assumption that is often implicitly made in the estimation of regression mixtures: that the means of the predictor do not vary across classes. The second paper examines issues related to the inclusion of covariates in the regression mixture model, focusing on how and when to include predictors of latent classes. The final paper extends regression mixtures to the multilevel contexts for use in testing cross level interactions. This extension is new and critical for prevention scientists who want to examine heterogeneity in contextual risk and protective factors (aspects of communities and schools).