A0225
Title: Bayesian approach for structure selection in mixed-effects model with categorical responses
Authors: Chi-Hsiang Chu - National University of Kaohsiung (Taiwan) [presenting]
Abstract: A Bayesian structure selection approach is proposed for mixed-effects models with a single categorical response variable. This type of data structure frequently arises in medical studies, where mixed-effects models are commonly employed for analyzing categorical responses. Accurate and efficient selection of significant explanatory variables is crucial in such analyses. To handle the categorical response, a multinomial probit model is adopted by introducing appropriate latent variables. A row sparsity assumption is imposed on the coefficient matrix, as each row represents the effect of one explanatory variable. To facilitate variable selection, an indicator-based Bayesian variable selection approach is utilized, and a corresponding MCMC algorithm is developed to generate posterior samples for inference. The performance of the proposed method is demonstrated through simulation studies and an application to a medical dataset.
