响应倾向得分匹配法及处理效应估计  

Response Propensity Score Matching Method and the Estimation of Treatment Effect

作　　者：孙玲莉 杨贵军[2] SUN Ling-li;YANG Gui-jun(School of Statistics,Tianjin University of Finance and Economics,Tianjin 300222;China Center of Economics and Statistics Research,Tianjin University of Finance and Economics,Tianjin 300222,China)

机构地区：[1]天津财经大学统计学院,天津300222 [2]天津财经大学中国经济统计研究中心,天津300222

出　　处：《统计学报》2021年第1期16-25,共10页Journal of Statistics

基　　金：国家社会科学基金项目(20CTJ009、20ATJ008);天津市哲学社会科学规划项目(TJTJ19-001)。

摘　　要：在倾向得分匹配法的基础上引入响应倾向得分匹配法,将控制组的结果变量观测值按照升幂顺序排列取秩,建立观测值秩累计概率与协变量之间的响应倾向得分模型,依据样本响应倾向得分选择反事实匹配组,进行处理效应估计。模拟研究显示,该方法使用效果较好,在估计流动人口受教育程度对家庭消费的处理效应时,相比于倾向得分匹配法,响应倾向得分匹配法的处理效应估计偏差和均方误差更小。基于天津市流动人口动态监测调查数据的实证分析显示,在消除样本自选择偏误之后,流动人口受教育程度对家庭消费的处理效应要明显小于未消除样本自选择偏误的处理效应,响应倾向得分匹配法可以弥补倾向得分匹配法的不足,提高处理效应估计的可信度。Based on propensity score matching method, the paper introduced response propensity score matching method, and ranked the observations of control group’s outcome viariables following the ascending order, then constructed the response propensity score model between cumulative probability of rank and covariances. According to the score, the paper formed the counterfactuals matching group, and further estimated the treatment effect. Simulation study showed that, response propensity score matching method had less bias and mean squared error of treatment effect estimators compared with the use of propensity score matching method, while it was applied for estimating the teatment effect of the education level of floating population on household consumption. According to the dynamic monitoring survey data of floating population in Tianjin, the empirical analysis showed that, after eliminating sample selfselection bias, this treatment effect was significantly weaker than the effect with self-selection bias. That is, response propensity score matching method could make up for the shortcomings of original response propensity score matching method and improve the reliability of treatment effect estimators.

关 键 词：样本自选择偏误 处理效应 反事实 响应倾向得分匹配法 倾向得分匹配法 

分 类 号：F224.0[经济管理—国民经济]