A0752
Title: Bayesian cumulative logit random effects models for longitudinal ordinal data
Authors: Jiyeong Kim - Sungkyunkwan University (Korea, South) [presenting]
Keunbaik Lee - Sungkyunkwan University (Korea, South)
Abstract: In analysis of longitudinal categorical data, generalized linear mixed models (GLMMs) are typically used. The random effects covariance matrix in the GLMMs explains both subject-specific and serial correlation of repeated outcomes. However, estimation of the covariance matrix is not easy because of high dimensionality and positive-definiteness of the estimated one. In addition, the structure of the covariance matrix can be heteroscedastic. To solve these constraints, Cholesky-type decompositions of the covariance matrix have been proposed (modified Cholesky, moving-average Cholesky and autoregressive moving-average Cholesky decomposition). We analyze longitudinal ordinal data using one of GLMMs, cumulative logit random effects models (CLREMs) with autoregressive moving-average random effects covariance matrix. In addition, various Cholesky-type decompositions are compared to our methods. The method are illustrated by a lung cancer data set.
