Optimal transport weights for causal inference

Imbalance in covariate distributions leads to biased estimates of causal effects. Weighting methods attempt to correct this imbalance but rely on specifying models for the treatment assignment mechanism, which is unknown in observational studies. This leaves researchers to choose the proper weighting method and the appropriate covariate functions for these models without knowing the correct combination to achieve distributional balance. In response to these difficulties, we propose a nonparametric generalization of several other weighting schemes found in the literature: Causal Optimal Transport. This new method directly targets distributional balance by minimizing optimal transport distances between treatment and control groups or, more generally, between any source and target population. Our approach is semiparametrically efficient and model-free but can also incorporate moments or any other important functions of covariates that a researcher desires to balance. Moreover, our method can provide nonparametric estimate the conditional mean outcome function and we give rates for the convergence of this estimator. Moreover, we show how this method can provide nonparametric imputations of the missing potential outcomes and give rates of convergence for this estimator. We find that Causal Optimal Transport outperforms competitor methods when both the propensity score and outcome models are misspecified, indicating it is a robust alternative to common weighting methods. Finally, we demonstrate the utility of our method in an external control trial examining the effect of misoprostol versus oxytocin for the treatment of post-partum hemorrhage.