|
Abstract:
|
Weighting methods are often used to generalize estimates of causal effects from a study sample to a target population of interest. However, most of the existing weighting methods are not directly applicable to target populations with complex structures, e.g., a single target individual. Moreover, such weights often have a multiplicative structure that may lead to increased variability in the final weights, yielding unstable causal effect estimates. We propose a new balancing approach to weighting for generalization and personalization problems, where the weights for each treatment group are constructed in ‘one go’ by minimizing a form of the dispersion of the weights, while directly constraining the covariate imbalances in the weighted group, relative to the target. The balancing approach does not require individual-level data from the target population, and can flexibly target different populations of interest, including a target individual. We discuss asymptotic properties of our approach, namely, consistency and asymptotic Normality of the corresponding average treatment effect estimators. We demonstrate the performance of our approach using a simulation study and a case study.
|
