协变量存在缺失的因果效应稳健估计  

Robust Estimation of Causal Effect With Missing Covariates

作　　者：韩锋 Han Feng(Teachers'College,Beijing Union University,Beijing 100011,China)

机构地区：[1]北京联合大学师范学院,北京100011

出　　处：《统计与决策》2020年第14期37-39,共3页Statistics & Decision

基　　金：国家社会科学基金后期资助项目(18FTJ003)。

摘　　要：文章借助逆概加权方法作用于影响函数,给出协变量存在缺失时倾向评分的估计方法。同时,针对现实研究中倾向评分取值可能接近0或1的情况,应用倾向评分扩展加权给出因果效应的估计。特别是推广Hirano(2003)的方法,导出其因果效应的渐近方差。模拟结果表明,提出的方法估计出的方差与样本方差接近,并且得到的因果效应优于传统方法,具有较小的Bias与MSE。This paper proposes the estimation method of propensity score by applying the inverse probability weighting method to the influence function when the covariates are missing.At the same time,in view of the fact that the value of propensity score may be close to 0 or 1 in actual research,the paper uses the extended weighting of propensity score to estimate the causal effect,particularly extending the method of Hirano(2003)to derive the asymptotic variance of its causal effect.The simulation results show that the variance estimated by the proposed method is close to the sample variance,and that the causal effect obtained by the proposed method is superior to the traditional method,with less bias and MSE.

关 键 词：协变量缺失数据 因果效应 渐近方差 随机缺失机制 

分 类 号：O213[理学—概率论与数理统计]