带有变量选择的协变量平衡倾向得分的估计:基于GMM-LASSO方法  

Covariate balancing in propensity score estimation with variable selection:Based on GMM-LASSO approach

作　　者：范菊逸 詹铭峰 蔡宗武[3] 方颖 林明[2,4] FAN Juyi;ZHAN Mingfeng;CAI Zongwu;FANG Ying;LIN Ming(School of Economics,Xiamen University,Xiamen 361005,China;The Wang Yanan Institute for Studies in Economics,Xiamen University,Xiamen 361005,China;Department of Economics,University of Kansas,KS 66045,USA;Fujian Key Laboratory of Statistical Sciences,Xiamen University,Xiamen 361005,China)

机构地区：[1]厦门大学经济学院,厦门361005 [2]厦门大学王亚南经济研究院,厦门361005 [3]堪萨斯大学经济系,堪萨斯66045 [4]福建省统计科学重点实验室(厦门大学),厦门361005

出　　处：《系统工程理论与实践》2021年第10期2631-2639,共9页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(71631004,72033008);国家杰出青年科学基金(71625001)。

摘　　要：本文基于Imai和Ratkovic (2014)协变量平衡倾向得分的估计方法,提出了带协变量平衡的GMM-LASSO估计方法.该方法既利用了协变量平衡的性质,同时又解决了如何基于数据来选取协变量的问题.理论上,文章证明了该估计方法是相合的.模拟显示,在满足一定的稀疏性的条件下,该方法可以显著地降低平均处理效应估计的绝对误差的中位数.最后,该方法被应用于研究2000年代初期意大利托斯卡纳地区的劳务派遣机制是否有助于工人寻找一份稳定的工作.Based on the covariate balancing propensity score method introduced by Imai and Ratkovic(2014),this paper proposes a new method to combine the GMM-LASSO type estimation method with covariate balancing approach.The proposed method not only utilizes the property of covariate balancing,but also solves the problem of how to select covariates based on data.Also,it is shown that the proposed estimator is consistent and simulations show that under the condition of sparsity,the proposed method indeed can significantly reduce the median of the absolute errors of the average treatment effect.Finally,the proposed method is applied to the data from the Tuscany region of Italy in the early 2000 s to study whether temporary work agency mechanism helps workers to find a stable job in the future.

关 键 词：协变量平衡 GMM-LASSO LOGISTIC回归 倾向得分 处理效应 

分 类 号：F224[经济管理—国民经济] O212[理学—概率论与数理统计]