Bayesian Variable Selection for Logistic Regression

• Activity Number: 263
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Yiqing Tian* and Howard Bondell and Alyson Wilson
• Companies: North Carolina State University and North Carolina State University and North Carolina State University
• Keywords: variable selection ; logistic regression ; bayesian ; penalized method

Abstract

We consider methods for Bayesian variable selection for logistic regression. There are several options available to select relevant predictors, which include penalized methods, screening, and forward selection. For Bayesian methods, a challenging problem is choosing an appropriate prior distribution, which can be particularly difficult for high-dimensional data where complete separation may occur. We consider an extension of Bondell and Reich (2013) and propose a hierarchical prior that we use to construct joint credible regions and perform variable selection. We discuss the development of the prior in detail and compare our results to those of other methods.