Statistics and Its Interface

Volume 6 (2013)

Number 1

A Bayesian analysis of generalized latent curve mixture models

Pages: 27 – 44



Edward H. Ip (Department of Biostatistical Sciences, Wake Forest University Health Sciences, North Carolina, U.S.A.)

Jun-Hao Pan (Department of Psychology, Sun Yat-Sen University, Guangzhou, China)

Xin-Yuan Song (Department of Statistics, The Chinese University of Hong Kong, Hong Kong, China)


Latent curve models for longitudinal data have received increasing attention in medical, educational, psychological, and behavioral sciences. In these applied areas of research, heterogeneous longitudinal data are common. This paper proposes the use of generalized latent curve models for analyzing heterogenous longitudinal data. The basic model features a mixture of trajectories. It also employs a multinomial logit model for assessing the influence of fixed covariates and explanatory latent variables on the class membership probability within the mixture model. This broad class of models also handles non-normal data from the exponential family distributions. A Bayesian approach is implemented for data analysis. We report a simulation study that proves the satisfactory performance of the proposed approach. Furthermore, we analyzed a real data set extracted from the National Longitudinal Survey of Youth to illustrate the practical value of the proposed model and methodology.


latent curve mixture models, heterogeneous longitudinal data, Markov chain, Monte Carlo method, modified deviance information criterion

Full Text (PDF format)