Abstract:
|
Bivariate zero-inflated poisson (BZIP) regression models have been used in several applications to model bivariate count data with excess zeros. Bayesian treatments of BZIP models have focused on diffuse prior structures for model parameters. These parameters depend on covariates through canonical link, generalized linear models. A common, relatively noninformative Bayesian prior approach is to place diffuse priors on regression coefficients and subsequently induce priors on the model parameters of interest. However, such an approach may be problematic in some cases as it can result in identifiability issues with the estimation of some parameters. We present an example to illustrate this. We offer an informative prior approach for BZIP model parameters. In particular, we propose a prior structure that uses expert opinion to elicit informative priors. Finally, we present an example in the medical context to illustrate our methods.
|