A0231
Title: Efficient estimation of modified treatment policy effects based on the generalized propensity score
Authors: Nima Hejazi - Harvard T.H. Chan School of Public Health (United States) [presenting]
Ivan Diaz - NYU Langone Health (United States)
David Benkeser - Emory University (United States)
Mark van der Laan - University of California at Berkeley (United States)
Abstract: Continuous treatments have posed a significant challenge for causal inference, both in formulating and identifying scientifically meaningful effects and in their robust estimation. Traditionally, the focus has been placed on techniques applicable to binary or categorical treatments with few levels, allowing for applying propensity score-based methodology with relative ease. Efforts to accommodate continuous treatments introduced the generalized propensity score. Yet, estimators of this nuisance parameter commonly rely upon parametric regression strategies that sharply limit the robustness and efficiency of inverse probability weighted (IPW) estimators of causal effect parameters. A flexible generalized propensity score estimator with rate-convergence properties desirable in semiparametric theory is formulated. With this estimator, nonparametric IPW estimators of a class of causal effect estimands tailored to continuous treatments are constructed. To obtain asymptotic efficiency for the proposed estimators, several non-restrictive selection procedures are outlined for applying a sieve estimation framework to under smooth generalized propensity score estimators. In numerical experiments, these novel, nonparametric IPW estimators are demonstrated capable of achieving the nonparametric efficiency bound (comparable to so-called double robust estimators) in a setting with continuous treatments, and their higher-order efficiency properties are investigated.
