Model Averaging for Causal Inference

• Activity Number: 330
• Type: Invited
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: SSC
• Author(s): Matthew Cefalu*
• Companies: RAND Corporation
• Keywords: Confounder selection ; Variable selection ; Causal inference ; Model averaging

Abstract

Data-driven variable selection methods are useful tools for causal inference. This talk will highlight the use of Bayesian model averaging for confounder selection and discuss several related approaches. A key feature of Bayesian model averaging is that it allows researchers to incorporate prior information into the analysis. This prior can include conditions needed to be satisfied for a covariate to be considered a confounder and any prior information on which covariates are more likely to be confounders. All uncertainty in the confounder selection process is fully accounted for in the final effect estimates.