A New Central Limit Theorem for Augmented IPW Estimator in High Dimensions

Watch a Harvard CMSA lecture where Pragya Sur presents groundbreaking research on a new Central Limit Theorem for the Augmented Inverse Probability Weighting (AIPW) estimator in high-dimensional causal inference. Explore how this 51-minute talk challenges existing assumptions in estimating average treatment effects (ATE) by introducing a novel theorem that operates without traditional sparsity conditions. Learn about the cross-fit version of the AIPW estimator, its behavior under well-specified outcome regression and propensity score models, and discover key findings including substantial variance inflation and non-negligible asymptotic covariance between cross-fitting estimators. Delve into the technical aspects combining approximate message passing theory, deterministic equivalents theory, and leave-one-out approaches, while understanding their practical implications across various disciplines. Follow along as the presentation moves from foundational concepts in causal inference to advanced theoretical frameworks, concluding with empirical validations and discussions on assumption robustness.