社会科学中的因果分析--潜在结果模型、因果网络模型与ABM

因果问题是社会科学的核心问题。在因果研究中采取反事实的因果定义已在各学科中达成共识。在这一前提下,当前社会科学的因果分析主要在三大方法论框架下展开:潜在结果模型、因果网络模型和ABM。本文对这三大方法论框架的核心思想及其主要模型和方法进行了回顾,首先按照数据类型和混淆变量的可观测性对倾向值匹配、加权、工具变量、断点回归、双重差分法等传统的模型进行了梳理,并介绍了与机器学习结合的最新进展;接着对因果网络模型的来源和其主要方法——贝叶斯因果图——进行了介绍,并简单阐述了贝叶斯因果图对于揭示横向因果机制的作用;最后对ABM的模型原理进行了介绍,分析了其在识别因果关系上的限制及其用于分析纵向因果机制的前提条件、优势与局限。本文希望通过系统性的梳理来为社会科学研究者了解因果分析方法体系及其前沿进展提供参考。

Causality is a core issue in social sciences research.The counterfactual definition of causality has reached a consensus in different research fields.Therefore,current causal studies of social sciences are mostly carried out under three frameworks:the Potential Outcomes Models,Structural Causal Models(SCM),and Agent-based Modeling(ABM).This paper reviewed the core concepts of these three methodological frameworks and their main models and methods.We started from reviewing the classical statistical models under the potential outcome framework such as PSM,IPTW,IV,RDD,DID,etc.,we classified them by the type of data and the observability of the confounders and then introduced their up to date progress in machine learning.Then we turned to the SCM,after a brief introduction of is history,we reviewed the Casual Bayesian Network,which is one of the most important models under this framework.Further,we introduced ABM and assessed its limitations when doing causal inference and its pre-conditions,strengths and weaknesses when revealing vertical causal mechanism.By systematically reviewing the three framework above,we hope to provide a clear clue for social scientists and help them to keep up with the state of the art methods of causal inference by this.