Jackson
Propensity Score Stratification: Can We Do Better? (303720)
*Roland Albert Matsouaka, Duke University
Keywords: propensity methods, stratification, weighting, Mantel-Haenzsel weights, optimal weights, average treatment effect
Propensity score methods (PSMs) have gained immense popularity in the last three decades. They are commonly used to control for confounding and reduce bias in the assessment of causal treatment effects. Of the four major PSMs used in the literature, the propensity score stratification (PSS) method, despite its many practical advantages, often yields results that are less reliable and for which there are limited explanations as to why. For this presentation, we take a closer look at the different weights commonly used to aggregate stratum-specific treatment effects into the overall average treatment effect (ATE). We demonstrate that these weights rely on a set of underlying assumptions that are not always true and rarely verified in practice. This explains in part some the discrepancies in data analysis results.
As an alternative, we introduce and implement efficient weights for the ATE that are congruent to the PSS framework. We assess their performance through simulation studies, considering several data-generating scenarios. Finally, we illustrate the proposed method using the publicly-available datasets.