Zhang Y, Gao Q, Wang T
Department of Health Statistics, School of Public Health, Shanxi Medical University, Taiyuan 030001, China.
Zhonghua Liu Xing Bing Xue Za Zhi. 2022 Apr 10;43(4):572-577. doi: 10.3760/cma.j.cn112338-20210827-00685.
Among kinds of methods for causal inference in observational studies, the propensity score (PS) method to control measured confounding is more widely used. PS method usually consists of two critical steps: first, estimating the propensity score, followed by calculating the causal parameters of interest by regression, weighting, matching, and stratification. Unlike the traditional dichotomous treatment, the generalized propensity scoring estimator used for continuous treatment has been proposed in recent years. Many methods have been developed to estimate the generalized propensity score or even estimate the balancing weight directly. This paper introduces the existing estimators from both the model-based and balance-based perspectives.
在观察性研究中的各种因果推断方法中，用于控制可测量混杂因素的倾向得分（PS）方法应用更为广泛。PS方法通常包括两个关键步骤：首先，估计倾向得分，然后通过回归、加权、匹配和分层计算感兴趣的因果参数。与传统的二分治疗不同，近年来已提出用于连续治疗的广义倾向得分估计器。已经开发了许多方法来估计广义倾向得分，甚至直接估计平衡权重。本文从基于模型和基于平衡的角度介绍了现有的估计器。
