Estimating Generalized Propensity Scores with Survey and Nonresponse Weighted Data

• Activity Number: 502 - Propensity Score Methods to Conduct Observational Studies Using Complex Survey Data
• Type: Topic Contributed
• Date/Time: Wednesday, July 31, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Survey Research Methods Section
• Author(s): Beth Ann Griffin* and Michael Robbins and Brian G. Vegetabile and Daniel F. McCaffrey
• Companies: RAND Corporation and RAND Corporation and RAND Corporation and Educational Testing Service
• Keywords: generalized propensity score; survey weights; sampling weights; complex surveys; survey

Abstract

Prior work in causal inference has shown that using survey sampling weights in the propensity score (PS) estimation stage and the outcome model stage for binary treatments results in a more robust estimator of the effect of the binary treatment being analyzed. However, to date, extending this work to continuous treatments and exposures has not been explored nor has consideration been given for how to handle nonresponse or attrition weights in the PS model. Nonetheless, generalized propensity score analyses (GPSA) are commonly utilized for estimating continuous treatment effects on outcomes using observational datasets with survey or attrition weighted data. In practice, survey designs and attrition weights are not properly being accounted for in estimation of the causal effects of interest. Here, we extend prior work and show that using survey sampling or attrition weights in the GPS estimation stage and the outcome model stage for continuous treatments also results in a more robust estimator than one which does not.