Session: State-of-the-Art Approaches to Latent Class Analysis with Distal Outcomes: Recommendations for Today’s Prevention Scientists (Society for Prevention Research 23rd Annual Meeting)

Latent class analysis (LCA) is a statistical tool that prevention scientists are turning to with increasing frequency to explain population heterogeneity by identifying underlying subgroups of individuals. The subgroups (classes) are comprised of individuals who are similar in their responses to a set of observed variables; class membership is inferred from responses to the observed variables. In many empirical studies, prevention scientists are interested in understanding which characteristics predict latent class membership. For example, do adolescents’ friendship goals (i.e., a risk factor) predict substance use patterns (i.e., a latent class variable)? The mathematical model for predicting class membership from a covariate is well-understood. Questions related to associations between a latent class predictor and distal outcome (LCA with distal outcomes), however, present a more difficult methodological problem. Historically, classify-analyze strategies have been used to solve this problem, where individuals are assigned to classes using some rule based on posterior probabilities from the LCA, and then an outcome analysis is performed treating class membership as known (e.g., regressing the outcome on a set of dummy coded predictors for class assignment). This approach, however, is known to cause substantial attenuation in effect estimates. Solving this problem is a “hot topic” in the methodological literature right now. This symposium summarizes the three competing state-of-the-art approaches to LCA with distal outcomes in order to guide researchers in their own work. 

The talks in this symposium provide a solid understanding of these new approaches in the context of research on the etiology of substance use and related negative outcomes. The first speaker will explain the three approaches, summarize what we know about their performances, and discuss software options. This discussion will focus on the simplest case: a single latent class predictor and a single observed outcome. The second speaker will introduce an extension to add control variables and moderators so that more nuanced questions can be addressed. The third speaker will take the discussion one step further by integrating inverse propensity weights as a causal inference technique and LCA with distal outcomes. This discussion will be structured around a step-by-step primer to conducting such an analysis. The discussant is a world-renowned expert in risk and protection in adolescents and young adults, and in the adaptation, implementation, and evaluation of evidence-based prevention programs; she will bring the three talks together by highlighting the implications of these approaches and opportunities for prevention scientists.