Bayesian Approach For Clustered Count Data With Excessive Zeros Using Compound Poisson Random Effec

Count data with excessive zeros are often encountered in a wide range of research areas. Zero inflated models, such as zero inflated Poisson (ZIP), zero inflated negative binomial (ZINB), and hurdle models, are widely used to handle zero inflated and overdispersed count data. Our work is motivated by model developed by Ma et al. (2009). They have proposed a multilevel random effects zero inflated Poisson model for the clustered count data with excess zeros. The main attraction of their model is they have used compound Poisson distribution to model cluster level random effects to accommodate the zero and nonzero count responses. The objective of this report is to incorporate the Bayesian approach to the ZIP model proposed by Ma et al. (2009). The regression parameters and the random effects parameters are estimated with Bayesian analysis using a Markov chain Monte Carlo (MCMC) algorithm. We use two zero inflated count datasets-hospital utilization data collected by the general hospital of the City of St. John's, and psychiatric outpatient service use data collected by ACCESS, to illustrate our proposed method.