社会学定量分析中的内生性问题 测估社会互动的因果效应研究综述

The Endogeneity Problem in Quantitative Analysis:A Review of Estimating Causal Effects of Social Interaction

因果关系是社会学分析的核心议题。然而,基于调查数据的定量分析都会受到内生性问题的困扰,从而往往难以作出恰当的因果性推断。大量现有的社会学定量分析虽以解释为己任,但实质上要么仅仅停留在描述统计相关阶段,要么得出错误的因果性结论。本文以社会互动研究为例,详细讨论了遗漏偏误、自选择偏误、样本选择偏误和联立性偏误等四种主要的内生性问题的来源,并介绍了克服内生性问题的一系列模型识别方法。最后还基于中国综合调查(CGSS2003)问卷,提出了通过提高调查数据信息量以利克服内生性问题的可能性。

Cause-effect relationships are the core area in sociological analysis. However, sociological analysis based on survey data is confronted by the endogeneity problem which plagues causal inferences. Many existing studies aiming at providing explanations for social phenomena either merely describes the statistical associations among variables or arrives at problematic causal conclusions. Focusing on the social interaction studies, this paper addresses the major sources of potential endogeneity biases, namely, the omitted variable bias, self-selection bias, sample-selection bias and the simultaneity bias. Useful model identification strategies for correcting these problems are reviewed. Based on CGSS2003, this paper also discusses how to partially correct for the endogeneity problem through augmenting the volume of survey data.