A1705
Title: Propensity score weighting with complex survey data: Best practice
Authors: Guangyu Tong - Yale University (United States) [presenting]
Yukang Zeng - Yale University (United States)
Fan Li - Yale University (United States)
Abstract: A unified framework is introduced for integrating survey weights into propensity score weighting methods for causal inference with complex survey data. We incorporate survey weights into both propensity score estimation and outcome modeling under the balancing weights framework, establishing the asymptotic normality of survey-weighted estimators and extending the minimum variance property of overlap weights to survey settings. Furthermore, we develop three augmented estimators: moment, clever covariate, and weighted regression, each specifically adapted for survey weights to enhance robustness and efficiency. We also provide robust empirical closed-form sandwich variance estimators to ensure reliable variance estimation in survey settings. Through multistage sampling simulations designed to mimic real-world data collection scenarios, we demonstrate that balancing weights based on survey-weighted propensity scores can achieve effective population balance and provide consistent estimates for key population-level causal estimands, including PATE, PATT, and PATO. Our findings indicate that the augmented estimators effectively reduce bias and enhance efficiency, with the weighted regression estimator showing particular robustness across varying levels of covariate overlap and in the presence of model misspecification.
