摘要
倾向得分匹配-双重差分模型(PSM⁃DID)是政策评估及因果推断中最为流行的方法之一.但是在实际应用中,该方法面临着控制变量在处理组样本和控制组样本之间非平衡性的挑战.传统基于均值差异t检验的平衡性检验容易产生片面和误导性的结论,使得后续因果推断产生偏误.为克服上述问题,本文对传统的平衡性检验提出以下改进:一是推荐更全面的多维度的平衡性测度指标,便于在匹配后更严谨地比较处理组和控制组的平衡性;二是提出了适用于非平衡样本的新估计方法:倾向得分匹配-逆概率加权-双重差分(PSM⁃IPW⁃DID),该方法结合了倾向得分匹配(PSM)克服样本自选择内生性及对非平衡样本稳健的优势和逆概率加权(inverse probability weighting,IPW)利用全样本信息的长处,在不进一步删除样本的情况下得到一种更稳健的双重差分估计方法.数据模拟和应用实例显示,本文提出的新方法能更全面、客观地评价宏观、微观政策的作用,得到更为可信的因果推断.
The propensity score matching⁃differenced⁃differences model(PSM⁃DID)is one of the most popular methods for policy evaluation and causal inference.However,it faces the challenge of unbalanced control vari⁃ables between the treatment group samples and the control group samples.The traditional balance test based on the mean difference t test is prone to one⁃sided and imprecise conclusions,which makes subsequent causal inferences biased and incredible.In order to overcome the above problems,this paper proposes the following improvements on the traditional balance test:First,it recommends a more comprehensive and multi⁃dimen⁃sional balance measurement index,which facilitates(after applying propensity score matching)the balance comparison between the treated group and the control group in a more rigorous method.Second,a new estimation method for unbalanced samples is proposed:Propensity score matching⁃inverse probability weighted⁃differences(PSM⁃IPW⁃DID).This method combines the merits of propensity score matching(PSM)in overcoming sample self⁃selection endogeneity and robustness to unbalanced samples,and the advantage of inverse proba⁃bility weighting(IPW)in taking advantage of the full⁃sample information,to obtain a more robust difference⁃in⁃differences estimation method without further trimming samples.Furthermore,numerical simulation and real data applications show that the proposed new method can evaluate the effects of various macro and micro policies comprehensively and objectively,and yield credible causal inferences.
作者
蔡俊
杨岚
周亚虹
CAI Jun;YANG Lan;ZHOU Ya-hong(School of Management,Huazhong University of Science and Technology,Wuhan 430074,China;School of Statistics,Southwest University of Finance and Economics,Chengdu 611130,China;School of Economics&Dishui Lake Advanced Finance Institute,Shanghai University of Finance and Eco-nomics,Shanghai 200433,China)
出处
《管理科学学报》
CSSCI
CSCD
北大核心
2024年第2期30-48,共19页
Journal of Management Sciences in China
基金
国家自然科学基金资助重点项目(71833004)
国家自然科学基金资助专项项目(72342034)
国家自然科学基金资助项目(72173083)
国家自然科学基金资助青年项目(72303074)
四川省哲学社会科学基金资助青年项目(SCJJ23ND428).
关键词
倾向得分匹配-双重差分
平衡性
逆概率加权
双重稳健性
propensity score matching⁃difference in differences
covariate balance
inverse probability weighting
double robustness
