浅谈真实世界研究中的因果推断

本文旨在通过介绍因果推断中常见的悖论、主要统计模型和相应的假设,浅谈真实世界研究中的因果推断.本文通过对Yule-Simpson悖论、Lord悖论、替代指标悖论进行简要的介绍,结合潜在结果模型和概率图模型的因果推断定义和主要假设分析偏倚的常见来源.良好的对照是识别因果效应的基础,完全随机是满足因果推断两个模型重要假设的...

Brief introduction to causal inference in real world research

In this paper,we introduce the common paradoxes,main statistical models and corresponding hypotheses in causal inference,and we briefly discuss causal inference in real world research.By introducing the Yule-Simpson paradox,the Lord paradox,and the Surrogate paradox,combining the causal inference definitions of the Potential Outcome Model and Pearl Causal Model and their main hypothesis,we pointed out common sources of bias.A good control group is the basis for identifying causal effects,and complete random allocation is an ideal condition for satisfying the two important assumptions of causal inference models.When we design a real-world research,it is necessary to evaluate whether the assumptions of the causal inference model are satisfied,to reduce bias and avoid paradoxes,we should evaluate real-world study from the perspective of totality of evidence approach.