A0783
Title: Accounting for model uncertainty in Bayesian Poisson regression models
Authors: Gregor Zens - Bocconi University (Italy) [presenting]
Mark Steel - University of Warwick (United Kingdom)
Abstract: Variable selection in Poisson regression models is a standard task for applied researchers in various fields. While frequentist penalized likelihood methods are well established, Bayesian frameworks have received considerably less attention. We develop a novel, exact and computationally feasible hierarchical framework for model averaging and variable selection in Poisson regression models. Posterior simulation is based on automatic and efficient reversible jump Markov chain Monte Carlo algorithms. A simulation study demonstrates the strengths of the framework relative to a number of competitor models, and real data applications further illustrate the approach.
