Robust Latent Class Analysis for Longitudinal Data

• Activity Number: 605
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Mental Health Statistics Section
• Author(s): Kari Hart*+ and John J. Hanfelt
• Companies: Ursinus College and Emory University
• Keywords: Artificial Likelihood ; Finite Mixture Models ; Generalized estimating equation ; Latent class analysis

Abstract

Latent class analysis is a likelihood-based approach for elucidating the structure underlying population heterogeneity. Existing methods for latent class analysis primarily focus on applications to cross-sectional data, while likelihood-based extensions for longitudinal data tend to be computationally intensive and sensitive to modeling assumptions. We present a novel and robust artificial-likelihood-based approach that accommodates high-dimensional longitudinal data. When the number of latent classes is assumed fixed and known, we consider a finite mixture of latent-class-specific generalized estimating equations, where the relative frequencies of class membership may be influenced by a set of covariates. Parameter estimation is performed via a modified expectation maximization algorithm that utilizes artificial-likelihood and requires specification of only the first two joint moments of the data. Simulation studies are presented to demonstrate our method's ability to accurately detect the presence of underlying latent classes when the number of latent classes is known and the sample size is sufficiently large. An application to mild cognitive impairment is also discussed.