因果图的无混杂与可压缩条件研究
Conditions for Non-Confounding and Collapsibility within Causal Diagrams
王婷 1孙毅1
作者信息
- 1. 新疆大学数学与系统科学学院,乌鲁木齐,830039
- 折叠
摘要
在观察性研究中,混杂现象往往会导致因果作用的估计出现偏差,从而影响因果关系推断结论的准确性.为了评估出真实的因果作用,本文利用因果图结构,研究关于因果作用无混杂与可压缩的问题.本文引入了线性排序集与条件化稳定的概念并研究了其性质.在此基础上,结合c-可去点,可反转边的定义与性质,提出无混杂与可压缩的若干充分条件.
Abstract
In observational studies,the phenomenon of confounding bias often leads to errors in the evaluation of causal effects,which in turn affects the accuracy of conclusions in causal inference.This paper details the properties of these two concepts of non-confounding and collapsibility in evaluating the true causal effects and proposes several conditions for non-confounding and collapsibility with knowledge of the constructed causal diagrams.In order to give characterizations for these conditions,we introduce the concepts of linearly ordered set and stability under conditioning and studies on certain properties.Based on the above arguments,we finally present sufficient conditions for non-confounding and collapsibility,respectively.
关键词
因果推断/无混杂/可压缩/因果图Key words
causal inference/non-confounding/collapsibility/causal diagrams引用本文复制引用
基金项目
国家自然科学基金项目(11726629)
国家自然科学基金项目(11726630)
国家自然科学基金项目(11701491)
东北师范大学应用统计教育部重点实验室开放课题(130028906)
出版年
2024
NETL
NSTL
万方数据