Overlap Weighting
This chapter allows analysts to flexibly specify a target population first and then estimate the corresponding treatment effect. It focuses on a special case of balancing weights, the overlap weight, which possesses desirable theoretical and empirical properties, and scientifically meaningful interpretation. Causal effects are contrasts of potential outcomes of the same units. Balancing weights include several widely used propensity score weighting schemes as special cases. An important special case of balancing weight is propensity score trimming, which focus on a target population with adequate covariate overlap. Matching and inverse probability weighting (IPW) with trimming methods target at similar populations with adequate overlap between treatments, but in a less transparent fashion. Survival, or more generally time-to-event outcomes are common in comparative effectiveness research and require unique handling because they are usually incompletely observed due to right-censoring. Propensity score weighting for covariate adjustment in randomized experiments bypasses the outcome model.
