ANALYSIS OF CLUSTERED ORDINAL DATA WITH SUBCLUSTERS
VIA A BAYES HIERARCHICAL MODEL USING GIBBS SAMPLER
Ming Tan
Department of Biostatistics and Epidemiology,
The Cleveland Clinic Foundation
Correlated ordinal data often arise in classifying disease severity
and mental or physical well-being. Although various mixed-effects
models for continuous and binary outcomes are available, developments of
random-effects models for correlated ordinal data are limited.
In particular, methods to analyze data with multi-level of nesting (or
clusters with subclusters) are rare and often the maximum likelihood
estimation is computationally prohibitive. As an extension to the
mixed-effects model, we propose a Bayes hierarchical model
to analyze repeated ordinal data with computation performed using Gibbs
sampler. The method is particularly useful for analyzing datasets with
small or moderate sample sizes and complex structures and is illustrated
with medical studies.