因果推断中平均处理效应的估计研究

Research on Estimation of the Average Treatment Effect in Causal Inference

医学、经济等领域的研究常关注某种处理或政策的有效性,并借助于估计处理或政策的平均处理效应及其方差,进行因果推断。Montes-Rojas(2009)已导出简单逆概加权方法得到的平均处理效应估计量的渐近方差公式。但当个体倾向评分取值接近于0或者1时,利用此公式可能会导致估计量及其方差异常大,从而估计量不稳健。本文类比Montes-Rojas(2009)导出渐近方差公式的方法,给出了扩展简单逆概加权方法下平均处理效应估计量渐近方差的计算公式。通过模拟研究,本文进一步验证了所导出的渐近方差公式的正确性及可行性。

The study in the medical,economic and other areas often focus on the effectiveness of a kind of treatment or of policy,and with the help of estimation the average treatment effect and its variance of the treatment or the policy we can draw causal inference.Montes-Rojas(2009) has been derived the asymptotic variance formula of the average treatment effect estimator by simple inverse weighted method.But when the propensity score for individual close to 0 or 1,using this formula could lead to estimator and its variance abnormal,thus the estimator was not robust.This paper imitated the methods of Montes-Rojas(2009) to derive asymptotic variance formula,and deduced the asymptotic variance formula of the average treatment effect estimator by extended simple inverse weighted method.Through the simulation research,we verified the asymptotic variance formula was correct.