Efficient, Stable, and Doubly Robust Estimation with Inverse Weighting

• Activity Number: 9
• Type: Invited
• Date/Time: Sunday, August 2, 2009 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Consulting
• Author(s): Zhiqiang Tan*+
• Companies: Rutgers University
• Keywords: missing data ; causal inference ; double robustness ; inverse weighting ; propensity score ; nonparametric likelihood

Abstract

Consider the problem of estimating the mean of an outcome variable in the presence of missing data under the assumption of ignorability given a vector of covariates. A closely related problem is to estimate the population average treatment effect in causal inference. A doubly robust approach makes use of both a outcome regression model and a propensity score model and derive an estimator that remains consistent if either of the two models is correctly specified. We extend the nonparametric likelihood approach of Tan and propose a new doubly robust estimator. The estimator involves two computationally simple steps through maximization of concave functions. In addition to double robustness, the estimator has advantageous properties in efficiency and stability. We compare the estimator with existing estimators theoretically and in a simulation study.