Abstract
A Bayesian analysis of count data using a hierarchical model with latent variables has been proposed. The model allows for fixed covariates as well as direct modeling of latent variables as a function of covariates which are considered as primary importance. A Gibbs sampling algorithm with adaptive rejection sampling method is suggested to find posterior densities of parameters and latent variables. The proposed method is applied to epileptic seizure data arising from a study of progabide as an adjuvant therapy for seizure episodes.