Primary Submission Category: Dynamic Treatment Regimes
High-Dimensional Doubly Robust Inverse Probability Weighting for Dynamic Treatment Effects
Authors: Feiyang Yi, Jelena Bradic,
Presenting Author: Feiyang Yi*
Inverse probability weighting is a common approach for estimating treatment effects. However, this method can yield potentially biased estimation when propensity score models are incorrectly specified. When confounders change over time and their dimensionality is much larger than the sample size, correctly specifying the propensity score models becomes more challenging, which can introduce substantial bias. This paper proposes a robust inverse probability weighting estimator for dynamic treatment effects that allows high-dimensional, time-varying confounding. We prove that the proposed estimator achieves root-N consistency and asymptotic normality under a sequential model double robustness condition, where at least one of the nuisance models is correctly specified at each treatment stage. Simulation studies illustrate the advantage of the proposed estimator compared with other inverse propensity methods.