Doubly Robust Causal Inference Using Doubly-Matched Estimators

• Activity Number: 460 - Causal Methods for Discovery, Confirmation and Mechanistic Evaluation
• Type: Contributed
• Date/Time: Thursday, August 6, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Ruifeng Chen* and Karen Messer
• Companies: University of California, San Diego and University of California, San Diego
• Keywords: Doubly Robust; Average Causal Effect; Propensity Score and Prognostic Score; Double Matching

Abstract

Methods for doubly robust estimation (DR) using matching are needed. The well-known standard DR estimator combines inverse probability weighted (IPW) estimators with model imputed estimators to estimate the average causal effect (ACE). In this paper, we review and propose doubly-matched estimators to estimate the ACE. Double matching is based on both the propensity score (PS) and the prognostic score (PGS), and doubly-matched estimators are unbiased when either the PS or the PGS is correctly calculated. In addition, we propose an augmented doubly-matched estimator which involves matching, IPW and imputation to estimate the ACE. Outstanding questions include whether the proposed estimators are consistent and efficient are proved, and investigated by simulation. We choose three different simulation scenarios, with either categorical or continuous covariates. We compare the proposed estimators with other known methods for estimating ACE in both large and small sample size cases. The proposed estimators are also used in a real study using the PATH database to estimate the effect size of the effectiveness of e-cigarettes use on long-term cigarette cessation.