类别变量的中介效应分析

在心理学和其他社科研究领域,研究者能熟练地进行连续变量的中介效应分析,但面对自变量、中介变量或(和)因变量为类别变量的中介效应分析,研究者往往束手无策。在阐述类别自变量中介分析方法的基础上,我们建议使用整体和相对中介相结合的类别自变量中介分析方法,并给出了分析流程。以二分因变量为例,讨论了中介变量或(和)因变量为类别变量的中介分析方法的发展过程(即尺度统一的过程),建议通过检验Za×Zb的显著性来判断中介效应的显著性。用二个实际例子演示如何进行类别变量的中介效应分析。最后展望了类别变量的中介效应分析研究的拓展方向。

Mediation Analysis of Categorical Variables

In the research of psychology and other social science disciplines, researchers often do not know how to analyze mediation effect when the independent, mediator or/and dependent variable are categorical, even if they can skillfully conduct mediation analysis with continuous variables. The conventional mediation analysis, transforming multi-categorical variable into dichotomous or continuous variable or using analysis of variance (ANOVA), might be lack of efficiency when the independent variable is multi-categorical. A procedure is proposed and recommended to use the method integrating relative mediation with Omnibus mediation to analyze mediation effect when the independent variable is multi-categorical. The first step is to implement Omnibus mediation analysis. If Omnibus mediation effect is not significantly different from zero, the k-1 relative mediation effects are zero, where k is the number of the categories. Otherwise, go to the second step. In the second step, relative mediation analysis is used to find if each relative mediation effect is significant. If there is no relative mediation effect is significantly different from zero, the mediation analysis is end. Otherwise, go to the third step. In the third step, the results with relative direct effects are reported. An example is given to illustrate how to conduct the proposed procedure by using SPSS software.Then, the evolution of the mediation analysis method with categorical mediator or dependent variable is discussed, and the scale unified process is the focus. In early years, the product of coefficients (ab) obtained from the logistic regression was used to analyze mediation effect when mediator or dependent variable is categorical. Later, abstd was adopted to analyze the categorical mediation effect. Recently, was used to analyze the mediation effect. We suggest that is preferred to analyze mediation effect when mediator or dependent variable is categorical. In addition, we emphasize that asymmetric interval is used to test the significance of . We used an example to illustrate how to conduct the proposed procedure by using SPSS software. Directions for future study on categorical mediation are discussed at the end of the paper. In fact, in addition to dummy coding, sequential coding and contrast coding are alternative to code multi-categorical independent variable, and the result of mediation effect tests with these three coding methods are equivalent, but the coding method will influence the implication of the relative indirect, direct, and total effects. Furthermore, the method integrating relative mediation with Omnibus mediation to analyze mediation effect of multi-categorical independent variable could be expended to more complicated mediation models, such as single-step or parallel multiple mediator models in which there is more than one mediator. Finally, it is a promising direction to analyze mediation effect with binary dependent variable in Structural Equation Modeling.