Model-Free Selection of Covariates for Estimating Average Causal Effects

• Activity Number: 670
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Jenny Häggström*+ and Emma Persson and Xavier Luna de Luna and Ingeborg Waernbaum
• Companies: Umeå University and Umeå University and Umeå University and Umeå University
• Keywords: Unconfoundedness ; Covariate selection ; Average causal effect ; Dimension reduction ; Kernel-based smoothing

Abstract

In observational studies the choice of covariates to control for in order for unconfoundedness to hold should primarily be based on subject matter theories, although the latter typically give only partial guidance. Including unnecessary covariates is, however, suboptimal when the effect of a binary treatment is estimated nonparametrically. De Luna, Waernbaum and Richardson (Biometrika, 2011), embracing the Neyman-Rubin model, characterize subsets from the original reservoir of covariates that are minimal in the sense that the treatment ceases to be unconfounded given any proper subset of these minimal sets and propose data-driven algorithms for the selection of minimal sets of covariates. We have implemented these algorithms in a R-package for model free selection of covariates in situations where the parameter of interest is an average causal effect. It allows the user to choose between two model free dimension reduction techniques: marginal co-ordinate hypothesis tests (Cook, Annals of Statistics, 2004) and kernel-based smoothing (Li, Racine and Wooldridge, Journal of Business & Economic Statistics, 2009). The performance of the methods are evaluated in a simulation study.