摘要
从统计推断到因果推断的关键在于有效解决内生性问题。受限于数据可获得性,管理学领域的大量研究问题仍需以非实验数据开展,因而内生性问题尤为明显。本文在梳理内生性起源及定义的基础上,直观展示了实证研究中出现内生性问题的原因,同时结合管理学经典研究案例探讨了内生性的不同来源及对应的处理策略,最后进一步探讨了近年来应用广泛的动态面板模型和空间计量模型处理内生性问题的思路,并就后续管理研究中的内生性问题处理提出了相关研究展望和可行性建议。
The key from statistical inference to causal inference is to effectively solve the endogenous problem.Limited by the availability of data,the endogeneity problem is particularly obvious in the field of management where need to be carried out with non-experimental data.On the basis of sorting out the origin and definition of endogeneity,this paper first intuitively shows the econometric theories of endogeneity problems in empirical research.The essential cause of endogeneity is that the independent variable is correlated with the error term.In empirical studies,several sources can contribute to this correlation,namely,omitted variables,two-way causality or simultaneous equations,measurement errors,and selection biases caused by non-random experiments.Secondly,based on these different sources,we provide potential coping strategies of endogeneity in combination with the classic research cases of management.For the endogeneity caused by omitted variables,researchers can try to control these omitted variables by adding fixed effects or using proxy variables.Random assignment experiments and instrumental variable approaches can also be used,which are also commonly used to address endogeneity caused by simultaneous equations.For the endogeneity caused by measurement errors,a direct method is to find systematic bias-free measurements or try to interpret these measurement errors.For the endogeneity caused by selection biases,an additional method is to use Heckman’s two-stage selection model.And for self-selection biases,common methods to estimate average treatment effects can be used,such as,differences-in differences and regression discontinuity design.Then,we further discuss the effective solutions in dealing with more complex dynamic panel data modelsand spatial econometric models.The endogeneity of dynamic panel data models arises when controlling for individual fixed effects.Difference GMM and system GMM are commonly used methods.The endogeneity of spatial econometric models is from the simultaneity of the dependent variable and the key independent variable—the spatial autoregressive term,where spatial weighted exogenous variables are usually used as instrumental variable.If the spatial weighting matrix also endogenous,figuring out the sources of its’endogeneity and then applying the control function approach is a suggested method.Finally,this paper puts forward relevant research prospects and feasible suggestions on how to deal with endogeneity problems in subsequent management research.First,it is necessary for researchers to understand the importance of the issue of endogeneity,the key to which lies in identifying and recognizing the reasons for the existence of endogeneity.Secondly,researchers need to target and deal with the most obvious and serious type of endogeneity problem in the study,as endogeneity is very common.Third,researchers need to realize that the current academic approaches to endogeneity have their own conditions of applicability and need to be continuously improved.
作者
郭俊聪
郑维伟
瞿茜
Juncong Guo;Weiwei Zheng;Xi Qu(Antai College of Economics and Management,Shanghai Jiao Tong University)
出处
《管理学季刊》
2024年第1期20-32,124,共14页
Quarterly Journal of Management
基金
国家自然科学基金优秀青年项目“空间计量经济学建模与分析”(72222007)
国家自然科学基金青年项目“家庭消费与投资的同群效应——基于非线性空间计量模型的分析”(72303151)的共同资助
关键词
管理研究
因果推断
内生性
management research
causal inference
endogenous
作者简介
郭俊聪,上海交通大学安泰经济与管理学院助理研究员,于上海交通大学获得经济学博士学位,在Journal of Development Economics、Regional Science and Urban Economics、Economic Letters等期刊发表论文多篇,目前主要研究兴趣为空间溢出效应和同伴效应;郑维伟,上海交通大学安泰经济与管理学院博士研究生,在《财经研究》《南开经济研究》《国际贸易问题》等期刊发表论文10余篇,目前主要研究兴趣为城市经济与企业创新;通讯作者:瞿茜,E-mail:xiqu@sjtu.edu.cn,上海交通大学安泰经济与管理学院教授,博士生导师,于美国俄亥俄州立大学获得经济学博士学位,在Journal of Political Economy、Journal of Econometrics、Journal of Development Economics等期刊发表论文20余篇,目前主要研究兴趣为空间计量模型与同群效应。
