因果关系计量方法的内在逻辑与实践改进

不同于相关关系计量,因果关系计量不可避免地会遇到反事实这一难题。因果关系计量的核心是构建反事实,通过巧妙设计获得受干预个体在不受干预下取值的估值。倾向得分匹配法、双重差分法和断点回归法正是基于实现对反事实的构建逻辑探索得出的三种主要的因果关系计量方法。虽然都是构建反事实,三种计量方法的思路设计和具体方法存在明显差异。为了更好地进行因果关系计量,实践中可从充分使用大数据和机器学习技术、加强定量研究与定性研究的结合等方面加以改进。

Different from correlation measurement,causality mea⁃surement inevitably meets the problem of counterfactuals.The core of causality measurement is to construct counterfacts,and to get the value of the interfered individual without interference through ingenious design.Propensity score matching method,difference-in-difference method and regression discontinuity design are three main causality measures based on the logical exploration of constructing counterfacts.Although they are all counterfacts,there are obvious differences in the design and concrete methods of the three measures.In order to measure causality better,we can make full use of big data and machine learning technology,and strengthen the combination of quantitative research and qualitative re⁃search.