Optum Labs, Cambridge, MA, USA,
Appl Health Econ Health Policy. 2014 Feb;12(1):7-18. doi: 10.1007/s40258-013-0075-4.
This paper examines the use of propensity score matching in economic analyses of observational data. Several excellent papers have previously reviewed practical aspects of propensity score estimation and other aspects of the propensity score literature. The purpose of this paper is to compare the conceptual foundation of propensity score models with alternative estimators of treatment effects. References are provided to empirical comparisons among methods that have appeared in the literature. These comparisons are available for a subset of the methods considered in this paper. However, in some cases, no pairwise comparisons of particular methods are yet available, and there are no examples of comparisons across all of the methods surveyed here. Irrespective of the availability of empirical comparisons, the goal of this paper is to provide some intuition about the relative merits of alternative estimators in health economic evaluations where nonlinearity, sample size, availability of pre/post data, heterogeneity, and missing variables can have important implications for choice of methodology. Also considered is the potential combination of propensity score matching with alternative methods such as differences-in-differences and decomposition methods that have not yet appeared in the empirical literature.
本文考察了倾向得分匹配在观察性数据分析中的应用。先前已有几篇优秀的论文对倾向得分估计的实际方面以及倾向得分文献的其他方面进行了回顾。本文的目的是比较倾向得分模型的概念基础与治疗效果的替代估计方法。本文提供了文献中出现的方法之间的实证比较的参考文献。这些比较适用于本文考虑的方法的一个子集。然而，在某些情况下，特定方法之间的成对比较尚不可用，并且没有跨越本文调查的所有方法的比较示例。无论是否有实证比较，本文的目的是为在健康经济评估中替代估计方法的相对优点提供一些直观的理解，其中非线性、样本量、前后数据的可用性、异质性和缺失变量可能对方法选择有重要影响。本文还考虑了倾向得分匹配与尚未出现在实证文献中的差异分析和分解方法等替代方法的结合。
