Primal-dual Covariate Balance and Minimal Double Robustness via Entropy Balancing
Year: 2015
Type: preprint
Abstract: Covariate balance is a conventional key diagnostic for methods used estimating causal effects from observational studies. We show covariate balance can also be directly used in the estimation of mean causal effects by studying a recently proposed entropy maximization method called Entropy Balancing (EB). In order to avoid the tautology of repeatedly checking covariate balance, EB exactly matches the covariate moments for the different experimental groups in its optimization problem. We find that the primal and dual problems of EB correspond, respectively, to the implicit linear model of outcome and propensity score logistic regression, with features being the balanced moments. Consequently, we prove EB enjoys some desirable statistical properties: it is doubly robust with respect to the linear models and reaches the asymptotic semiparametric variance bound when both models are correct. Our theoretical results and simulations suggest EB is a very appealing alternative to the conventional propensity score weighting estimators.
Source: arXiv (Cornell University)
Author: Qingyuan Zhao
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2215646160