Data Decomposition beyond Splitting for Causal Estimation

In modern causal inference, the way we split and utilize data shapes both the efficiency and uncertainty quantification of treatment effect estimates. This manuscript explores emerging data manipulation strategies that go beyond conventional sample splitting. Building on a recent line of work, we introduce data decomposition methods tailored for causal estimation and examine how they can improve the performance of doubly robust estimators. Empirically, we show that these approaches lead to more precise and robust treatment effect estimates.