Alternative Strategies for Variable Selection in Logistic Regression Models with Missing Covariates

• Activity Number: 227
• Type: Contributed
• Date/Time: Tuesday, August 9, 2005 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Gang Liu*+ and Xiaowei Yang and Thomas R. Belin
• Companies: University of California, Los Angeles and BayesSoft, Inc. and University of California, Los Angeles
• Keywords: MCMC ; Imputation ; SIAS & ITS ; Model Selection ; GLM ; Missing Data

Abstract

A crucial problem in fitting a multiple regression model is the selection of predictors to include. Logistic regression model is a popular modeling approach to finding the relationship between a binary response variable and a subset of potential explanatory variables or predictors. In practice, it is common for there to be numerous variables measured and many of them have missing values. The proportion of incomplete cases sometimes is so high that traditional data analysis methods would provide biased estimators or invalid inferences if incomplete cases were excluded from consideration. In this paper, we develop two strategies for variable selection in logistic regression models with missing values on explanatory variables. One approach, which we call "impute, then select" (ITS) involves initially performing multiple imputation and then applying Bayesian variable selection to the multiply imputed datasets. The second strategy is to conduct Bayesian variable selection and missing data imputation simultaneously within one Markov Chain Monte Carlo (MCMC) process, which we call "simultaneously impute and select" (SIAS).