Penalized Regression Approaches to Variable Selection in the Potential Outcomes Framework

• Activity Number: 114
• Type: Topic Contributed
• Date/Time: Monday, August 5, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Health Policy Statistics Section
• Author(s): Debashis Ghosh*+
• Companies: Penn State University
• Keywords: Adaptive lasso ; Average causal effect ; Imputation ; Model misspecification ; Prediction ; SCAD penalty

Abstract

With all the interest in assessing causal effects in observational studies, an important question becomes as to what variables to include for modelling. We adopt the potential outcomes framework and observe that there is an inherent predictive nature to the notion of counterfactuals. Because of this, the usual penalized regression approaches are not directly applicable to the problem, and we instead focus on using predictive criteria. Using these functions in conjunction with standard L1 type penalties will lead to adaptations of existing LASSO algorithms. Some theoretical results will be given, and simulation studies and real data will be used to demonstrate the methodology.