Causal Mediation Analysis Using Categorical Latent Variables

Understanding causal pathways is essential for advancing scientific knowledge and addressing complex questions. In causal mediation analysis, the total effect of an intervention is decomposed into the natural direct effect (NDE) and the natural indirect effect (NIE). Although several methods exist, applications involving latent categorical variables from latent class analysis (LCA) remain limited. This study extends existing approaches by incorporating latent mediators and/or outcomes into mediation analysis. We evaluate the performance of these methods through simulation studies, considering both regression-based and marginal structural models. Procedures for estimation and inference are detailed for each scenario. Finally, the proposed methods are applied to data from an intervention promoting health behaviors among high school students in Brazil, illustrating their practical relevance and broad applicability across different fields.