Detecting Latent Classes Through Mediation in Regression Mixture Models

The current study aims to investigate mediation in regression mixture models. There has been little research that has examined the combination of mediation and regression mixture models to determine if there are latent subgroups that vary in their levels of mediation. This investigation aims to address this gap by simulating varying conditions of sample size, number of latent classes, mixing proportions, class intercept separation, direct effects, and class separation on mediating effects. Information criteria (i.e., AIC, BIC, aBIC) and likelihood ratio tests (i.e., LMR, VLMR, and BLRT) were evaluated for model selection. The results suggest that the BIC and BLRT perform best at identifying the correct number of latent classes. The class enumeration indices improved in accuracy as sample size, class intercept separation, and separation on the mediating effect increased. The current investigation identifies conditions where class enumeration is most accurate with mediation in regression mixture models.