Alternative Ansätze für Confounder-Adjustierung durch Gewichtung auf der Grundlage des PropensityScores

Abstract

Propensity score methods (PSMs) such as matching and weighting have become a mainstay for the estimation of treatment effects from observational data. PSMs require the choice of a treatment effect of interest given a specific target population or subpopulation. The most commonly used treatment effects in observational studies are the average treatment effect (ATE) and the average treatment effect among the treated population (ATT). While the traditional matching approach usually discards observations that cannot be matched and is therefore mainly used for estimation of the ATT, the weighting approach utilizes all observations in most cases and can be used for estimation of both, the ATE and ATT. One concern when using the weighting approach is the occurrence of large weights which may disproportionately influence the results and yield estimates with high variance. Recently several newer approaches, including matching weights and overlap weights have been proposed to overcome these limitations. However, this may complicate the interpretation of the results as the population to be inferred depends on the PS estimate itself.