Semiparametric Random Effects Selection in Generalized Linear Mixed Models

• Activity Number: 70
• Type: Contributed
• Date/Time: Sunday, August 3, 2014 : 4:00 PM to 5:50 PM
• Sponsor: Biometrics Section
• Author(s): Yong Shan*+ and Xiaoyan Lin and Bo Cai
• Companies: University of South Carolina and University of South Carolina and University of South Carolina
• Keywords: Random effects ; Generalized linear mixed models ; Semiparametric

Abstract

Random effects selection in Generalized linear mixed models (GLMM) is challenging, especially when the random effects are assumed nonparametrically distributed. In this paper, we develop a unified Bayesian approach for the random effect selection for the GLMMs. Specifically, we assume the random effects arise from a Dirichlet process (DP) prior with normal base measure. A special Cholesky-type decomposition is applied to the base covariance in the DP prior, and then zero-inflated mixture priors are assigned to the components of the decomposition to achieve the random effect selection. For non-Gaussian data, a Laplace approximation to the likelihood function is relied to apply the proposed MCMC algorithm. The performance of our proposed approach is investigated by using simulated data and real life data.