适用于二元和连续处理变量的平均因果效应广义定义
统计方法学
2021-12-17 v2
摘要
因果推断的主要任务之一是估计定义明确的因果参数。其中一个主要的因果参数是平均因果效应(ACE)——目标总体中个体层面因果效应的期望值。对于二元处理变量,个体层面因果效应被定义为潜在结果之间的对比。然而,对于连续结果变量,在有限样本中存在许多这样的对比,从而阻碍了将其用作因果关系的有效总结。在此,我们提出了 ACE 的一个广义版本,其中个体层面因果效应被定义为个体层面因果剂量-响应函数在个体所接受的处理值处的导数(对处理变量求导)。这一定义等价于二元处理变量的传统定义,但也涵盖了连续处理变量。我们证明了该量可以在常规因果假设下进行估计,并通过模拟研究阐述了理论思想。
引用
@article{arxiv.2112.04580,
title = {A generalized definition of the average causal effect for both binary and continuous treatments},
author = {Fernando Pires Hartwig},
journal= {arXiv preprint arXiv:2112.04580},
year = {2021}
}
备注
The reason for withdrawal is that the proposed definition already exists in a literature that the author was unfamiliar with. More specifically, it exists in the econometrics literature under the name "average derivative effect". Therefore, although this work has been conducted independently of such previous work, it does not contribute anything new to the literature