Primary Submission Category: Machine Learning and Causal Inference
Efficient estimation of modified treatment policy effects based on the generalized propensity score
Authors: Nima Hejazi, David Benkeser, Ivan Diaz, Mark van der Laan,
Presenting Author: Nima Hejazi*
Continuous treatments have posed a significant challenge for causal inference, both in the formulation and identification of scientifically relevant effects and in their estimation. Traditionally, focus has been placed on techniques applicable to binary or categorical treatments with few levels, which allow for application of propensity score-based methodology with relative ease. Efforts to evaluate causal effects of continuous treatments introduced the generalized propensity score, yet estimators of this nuisance parameter often rely upon restrictive, parametric assumptions that sharply limit the robustness and efficiency of inverse probability weighted (IPW) estimators. We formulate a nonparametric generalized propensity score estimator with favorable semiparametric rate-convergence properties and use it to construct nonparametric IPW estimators of a class of causal effect estimands tailored to continuous treatments. We outline several non-restrictive selection procedures for applying a sieve estimation framework to undersmooth the generalized propensity score estimator to obtain asymptotically efficient IPW estimators. We demonstrate that these IPW estimators are capable of achieving the nonparametric efficiency bound (comparable to so-called doubly robust efficient estimators) in a setting with continuous treatments, investigate their higher-order efficiency properties, and apply them to evaluate immune correlates of protection in a vaccine efficacy trial.