Bayesian Zero-Inflated Latent Class Model for Longitudinal Data

• Activity Number: 492
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Si Yang* and Gavino Puggioni
• Companies: University of Rhode Island and University of Rhode Island
• Keywords: Bayesian ; Finite Mixture Models ; Growth Mixture Models ; Latent Class Models ; Zero-inflated Poisson

Abstract

Using longitudinal data to model behavioral change patterns is a major interest in the field of developmental psychology and behavioral science. As heterogeneity often exist in the population, researchers are more and more interested in a topological representation of the growth trajectories. This paper focus on developing latent class models for longitudinal data, in addition to having zero-inflated count response variables. The goal is to model discrete longitudinal patterns of change on counts of rare event and to identify subject-specific covariates associated with latent class probabilities. Two kinds of discrete latent information exist in this type of models: first, a latent categorical variable classifies subgroups with distinct developmental trajectories and then a latent binary variable identifies whether an observation is from a zero-inflation process or a regular count process. A joint estimation of the latent variables and regression parameters is proposed in the Bayesian framework. The methods are illustrated through a simulation study and are applied to data on cigarette smoking from the National Longitudinal Study of Adolescent Health.