Hierarchical Bayesian Models for Small Area Estimation under Overdispersed Count Data

Bayesian analysis was applied to small area models with overdispersed response variables. The benefits of implementing this strategy by Markov Chain Monte Carlo methods make inference straightforward and computationally feasible. In this paper, we apply the strategy into area-level modeling to predict the under-five mortality rate at the district level in Java Island, the most populated region in Indonesia. The result shows that the zero-inflated negative binomial model yields the reduced relative standard error and relative mean squared error when compared to district estimates, the zero-inflated generalized Poisson and Poisson models.