Cross-Balancing for Data-Informed Design and Efficient Analysis of Observational Studies

Ying Jin, José R. Zubizarreta · 2025

Read Cross-Balancing for Data-Informed Design and Efficient Analysis of Observational Studies by Ying Jin, José R. Zubizarreta (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-15896.txt.

Introduction

Ying Jin and José Zubizarreta propose cross-balancing, a method for causal inference in observational studies. Causal work compares groups that differ by treatment but little else. Groups are often built from background traits alone. Using outcome data to help design that comparison could raise accuracy, yet it risks breaking the statistics. That long-standing problem is what the authors address.

Sample splitting separates error in building balancing features from error in estimating weights. The frame covers features learned as functions and features picked from a large dictionary. Under mild conditions the estimators are consistent, near normal, and efficient. Tests in simulation and an observational case show gains, with less bias and robustness when learned features converge slowly.

Catalog record

CREATORYing Jin, José R. Zubizarreta
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE203 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-15896

Browse the full library · About the Tower of Babel