Gauss因果模型中因果效应识别方法的比较

数学物理学报

Gauss因果模型中因果效应识别方法的比较

赵慧^1,2; 郑忠国³

(1. 华中师范大学数学与统计学学院武汉 430079; 2. 中国科学院数学与系统科学研究院北京 100190; 3. 北京大学数学科学学院北京 100871)

收稿日期:2006-03-28 修回日期:2008-01-20 出版日期:2008-10-25 发布日期:2008-10-25
通讯作者: 赵慧
基金资助:
国家自然科学基金(10571070)、中国博士后科学基金资助

Comparing Identifiability Criteria for Causal Effects in Gaussian Causal Models

Zhao Hui^1,2; Zheng Zhongguo³

(1. Department of Mathematics and Statistics, Huazhong Normal University, Wuhan 430079; 2. Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing 100190; 3. School of Mathematical Sciences, Peking University, Beijing 100871$)

Received:2006-03-28 Revised:2008-01-20 Online:2008-10-25 Published:2008-10-25
Contact: Zhao Hui

摘要/Abstract

摘要： 一个Gauss因果模型中常常存在不只一种识别因果效应的方法, 不同的方法对应的估计可能不同. 该文对Pearl等人提出的前门准则, 后门准则,工具变量准则等识别方法的估计精度进行了分析比较, 并给出了相应的模拟结果, 为实践中选择更优的识别准则提供了依据.

关键词: Gauss因果模型, 因果效应, 前门准则, 后门准则, 工具变量

Abstract: In a Gaussian causal model, there may exist several criteria for the identification of the same causal effect, but the corresponding estimators may be different from each other. This paper consideres three popular identifiability criteria, front-door, back-door and instrumental variables criteria for causal effects, compares their estimate efficiencies and gives some simulations.
These results can offer guidances for choosing better identifiability criterion in practice.

Key words: Gaussian causal models, Causal effects, Front-door criterion, Back-door criterion, Instrumental variables

中图分类号:

赵慧; 郑忠国. Gauss因果模型中因果效应识别方法的比较[J]. 数学物理学报, 2008, 28(5): 808-817.

Zhao Hui; Zheng Zhongguo. Comparing Identifiability Criteria for Causal Effects in Gaussian Causal Models[J]. Acta mathematica scientia,Series A, 2008, 28(5): 808-817.