A1896
Title: Stable causal estimation with optimal transport
Authors: Yixin Wang - University of Michigan (United States) [presenting]
Abstract: In causal inference, the overlap (a.k.a. positivity) assumption is frequently violated in practice. Limited overlap causes severe instability for widely used estimators such as inverse propensity weighting (IPW), whose variance depends on density ratios that can explode as propensity scores approach zero or one. To address this challenge, the Stable Transport Estimator (STE), an optimal Transport-based approach for estimating average causal effects under limited overlap, is proposed. Unlike IPW, the variance of STE scales with the Wasserstein distance between treated and control covariate distributions, a quantity that remains finite under bounded moments, even when density ratios explode. By replacing reweighting with a geometry-aware Transport map, STE constructs Stable counterfactual comparisons without requiring density estimation or propensity score modeling. STE further extends to settings with no overlap at all: when covariate supports are disjoint, a linear interpolation Transport map enables principled extrapolation with explicit bias-variance trade-offs. Consistency, asymptotic normality, and sharp bias-variance characterizations are established under conditions substantially weaker than those required by standard methods. Across synthetic and real datasets, STE demonstrates improved robustness, reduced bias, and markedly more Stable variance compared to existing estimators.
