A Bayesian Approach to a Multiple-Group Latent Class-Profile Analysis: The Timing of Drinking Onset and Subsequent Drinking Behaviors Among U.S. Adolescents

摘要

This article presents a multiple-group latent class-profile analysis (LCPA) by taking a Bayesian approach in which a Markov chain Monte Carlo simulation is employed to achieve more robust estimates for latent growth patterns. This article describes and addresses a label-switching problem that involves the LCPA likelihood function, which has multiple equivalent modes because it is invariant to permutations of class and profile labels. Our solution involves a dynamic data-dependent prior that can break the symmetry of the posterior distribution via preclassification of one or more individuals into latent subgroups. The article demonstrates this LCPA approach in an estimation of the effect of early-onset drinking on subsequent drinking behaviors among adolescents. The data are from a subsample of 4,773 adolescents (12-14 years old in 1997) studied in the National Longitudinal Survey of Youth 1997. The estimation results provide support for the view that patterns of sequential latent growth depend on the timing of drinking onset.