Covariate-Balancing Weights for Causal Inference and Generalization
Weighting, including inverse weighting by propensity score, is a very common strategy to account for confounding in causal inference. To construct robust and stable weights, covariate-balancing constraints are incorporated into an optimization framework in many recent works.
This talk will start with a review for the idea of covariate-balancing weights for causal inference. Then we will talk about extension of such framework to construct weights for average treatment effect (ATE) generalization to a target population when individual-level data from a source sample and summary-level covariates data from a target sample are available. Numerical results and real data example will be shown for the ATE generalization setting.
