Session: Applying Latent Class Models in Prevention Science: Practical Solutions to Everyday Problems (Society for Prevention Research 27th Annual Meeting)

Latent class and latent profile analysis (LCA/LPA) are statistical tools that prevention scientists are turning to with increasing frequency to explain population heterogeneity by identifying underlying subgroups of individuals. The subgroups (i.e., classes) are comprised of individuals who are similar in their responses to a set of observed variables; subgroup membership is inferred from responses to the observed variables. Prevention scientists may be interested in using LCA/LPA to identify subgroups and their prevalences and to link subgroup membership to risk factors and adverse outcomes. Recent methodological advances with LCA/LPA have made it easier than ever for prevention scientists to incorporate latent classes into their theories and analyses.

Often referred to as “person-centered” methods, LCA/LPA are concerned with within-individual patterns of behaviors (or other characteristics) across multiple dimensions. This is in contrast to “variable-centered” methods, such as regression analysis, that are concerned with between-individual relations among variables. This symposium focuses on resolving practical problems faced when applying LCA/LPA to prevention-related data in real life, and it provides solid understanding of the presented solutions by demonstrating a step-by-step approach to each solution.

We start by considering missing data. Although missing data on indicator variables is typically handled via maximum likelihood estimation, multiple imputation is often desirable for missing data on other variables included in analyses (e.g., predictors). Our first speaker describes, demonstrates, and evaluates an approach for using multiple imputation with LCA with predictors. Next, we consider nested data. A common data structure in prevention research, multiple individuals may be nested within families or classrooms, or multiple observations over time may be nested within individuals. Our second speaker uses multilevel LPA (MLPA) to understand both family structures and family dynamics in intensive longitudinal data. Finally, we consider how to model long-term trends in latent class prevalences and in effects of predictors on latent classes. Our third speaker presents an approach to combining LCA with time-varying effect modeling (LCA-TVEM) while examining epidemiological trends in alcohol use from adolescence through adulthood. Our discussant is an expert in substance use among adolescents and adults, and in risk and protection in these populations, as well as in the implementation and evaluation of intervention programs. She will bring our three talks together by highlighting the contributions that can be made by person-centered methods like LCA/LPA, as well as the implications of our three studies and opportunities for prevention scientists.