Bayesian Statistics Seminar
North Carolina State University
presents
Dr. Brian Reich
NC State University
"A Review of Bayesian Variable Selection Methods"
ABSTRACT
Selecting a subset of predictors for a regression model is a crucial aspect of many statistical analyses. The Bayesian approach to variable selection has several attractive features, including formalization of results using decision theory, straight-forward quantification of variable importance and model uncertainty, and the flexibility to handle missing data and non-Gaussian distributions. In this talk, we will review several popular approaches to Bayesian variable selection including Bayes factor methods and stochastic variable selection. Connections between Bayesian models and penalized regression approaches such as the LASSO will be highlighted and the application of Bayesian methods will be illustrated using the R and WinBUGS computing packages.
Slides and related codes are available from Speaker's website: Slides and WinBUGS code
Tuesday, September, 19, 2006
4:00 - 5:00 pm
208 Patterson Hall