图形的逻辑力量:因果图的概念及其应用

The Logical Power of Graphs:The Concept of Causal Graphs and Their Applications

本文旨在系统性地引介因果图方法,一种社会科学领域新近发展起来的探究因果推断的非参数估计工具。首先对因果图的基本概念和构型进行介绍,讨论变量之间不同“通路”对应的开启和阻断规则及因果推断中的三种偏差来源,即混淆偏差、过度控制偏差、内生性选择偏差。在此基础上,本文将因果图框架与现有定量社会科学研究中基于回归模型的因果推断方法思路进行融合,结合实际案例使用因果图阐释包括遗漏变量、样本选择、自选择及联立性在内的四种内生性问题,并对多元回归与匹配、代理变量、实验、工具变量、面板模型等因果推断方法的运行机制进行了图形化。最后,本文使用因果图厘清一些关于因果推断的不准确理解。

Causal inference is a core problem in empirical research in the social sciences,but understanding the context of causal inference relies on algebraic derivation,a fact that hinders the prevalence of causal knowledge among sociologists.Causal graphs derived from computer science can intuitively present casual paths and control strategies in a graphical way,thereby providing people with a non-parametric toolkit for understanding causal problems.This paper aims to provide a comprehensive introduction to the causal graph method and integrate it with the existing framework of causal inference based on regression models.This article first introduces the conceptual rules and the three basic configurations of chain,fork,and inverted fork that make up causal graphs.Then,it discusses the opening and blocking of causal pathways between variables and the three sources of bias that can mislead the identification of true causal relationships,namely confounding bias,over-control bias and endogenous selection bias.The article further introduces the D-separation rule used to determine which variables in a causal inference should be controlled.On this basis,various empirical examples are brought in to interpret four endogenous problems of omitting variable bias,sample selection bias,self-selection bias,and simultaneity bias through causal graphs.Graphic expressions and implementing conditions of several causal inference methods are also identified,including multiple regression and matching,proxy,experiments,instrumental variable,and panel models.In addition,this article attempts to clarify two common misconceptions in causal inference:conditioning on a post-treatment variable does not necessarily lead to bias and conditioning on a pre-treatment variable may cause deviation.Finally,it is suggested that the application of the causal graph method can help standardise causality research and facilitate the teaching and dissemination of causal inference knowledge.