A1518
Title: Cross-balancing for data-informed design and efficient analysis of observational studies
Authors: Ying Jin - University of Pennsylvania (United States)
Jose Zubizarreta - Harvard University (United States) [presenting]
Abstract: Causal inference starts with a simple idea: compare groups that differ by treatment, not much else. Traditionally, comparable groups are constructed using only observed covariates; however, incorporating outcome data into the study design while preserving valid inference remains a long-standing challenge. The general problem of covariate adjustment, effect estimation, and statistical inference is studied when balancing features are constructed or selected with the aid of outcome information. Cross-balancing is proposed, a method that uses sample splitting to separate the error in feature construction from the error in weight estimation. This framework addresses two cases: one where features are learned functions and one where they are selected from a potentially high-dimensional dictionary. In both cases, mild and general conditions are established under which cross-balancing yields consistent, asymptotically normal, and efficient estimators. In the learned-function case, cross-balancing achieves finite-sample bias reduction relative to plug-in estimators and is multiply robust when learned features converge at slow rates. In the variable-selection case, cross-balancing requires only a product condition on the approximation quality of the selected variables. Cross-balancing is illustrated in simulations and an observational study, demonstrating that careful use of outcome information can substantially improve estimation and inference while maintaining interpretability.
