Nonparametric Estimation of General Treatment Effects using Stabilized Weights

Abstract

Estimating the causal effect of a treatment or policy from observational studies is a challenging task due to confounding issues. We propose a nonparametric approach to identify and estimate general treatment effects using a weighted conditional expectation under unconfoundedness assumption. The weights are estimated through a generalised empirical likelihood method subject to an expanding set of moment equations. Our proposed estimator achieves semiparametric efficiency bounds for discrete treatments and is more efficient than the estimator constructed from the true weights for continuous treatments. We also derive an asymptotic influence function to facilitate statistical inference. Our framework includes a variety of treatment effects, such as average, quantile, and asymmetric least squares treatment effects. Additionally, it can be applied to data with measurement errors and functional components.