Xiaomei Li

Victoria University of Wellington

Computational model selection for the Poisson regression model

The task of statistical model selection is to choose a parsimonious model from a collection of models, which gives the best approximation to the observed data. The ultimate objective for this project is to illustrate the non-Bayesian and Bayesian approaches in computational model selection for the Poisson regression model. We developed three commonly used model selection methods by using R and WinBUGS. They are developed based on hypothesis testing, deviance, Bayesian approach, and information criterion. Also, we compare the results which we obtain from the different approaches.