调节效应的陷阱

Pitfalls of the Moderating Effect

实证研究中,常在线性回归中加入交叉项来对调节效应建模。然而基于交叉项系数的统计显著来宣称调节效应,存在可能的误导。调节效应不能通过纯粹的统计分析来证实。如果缺乏可信的理论,以数据挖掘的方式来搜索调节效应很容易掉入虚假调节效应的陷阱。本文提出一系列预检验来避免虚假的调节效应,并建议了研究调节效应的实证策略。

Statistical significance of the interaction term in regression is often considered an evidence to support the moderating effect. There exist some pitfalls that make this statistical claim on moderating effect invalid, however. Modeling the moderating effect should be more guided by theory rather than pure data mining. If one attempts to data mine the moderating effect, some pre-tests suggested in this paper can help to avoid these pitfalls.