基于两水平被试内设计的中介效应分析方法

在心理学和其他社科研究领域，通常的中介效应分析都基于被试间设计，研究者对于如何分析基于被试内设计的中介效应往往并不清楚。本文阐述了两水平被试内设计的中介效应分析方法(依次检验法和路径分析法)，综合各方法优缺点给出一个分析流程，并用应用研究实例演示如何分析两水平被试内中介效应，最后展望了基于被试内设计的中介效应分析研究的拓展方向。

The Analyses of Mediation Effects Based on Two-Condition Within-Participant Design

Mediation effects analyses are frequently applied to the studies of psychology and other social science disciplines. Because mediation is helpful to explain how and why two variables are related, not only questionnaire survey researchers but also experimental researchers are interested in it. Given within-participant design is one of the most commonly used designs in experimental psychology study, how to make a mediation effects analysis based on the data from such a design become a noteworthy issue. Suppose that every subject is assigned to both experimental treatments (X), and measurement for mediator (M) and dependent variable (Y) is conducted under each condition, according to general data input format (such as in SPSS), there are no X variable to regress Y and M on, and both M and Y have two column values. So the mediation analysis method for this design does not follow the well-known and traditional approach used in the questionnaire survey research designs that are cross-sectional or “between-participant” in nature. The aim of this study is to clarify how to analyze mediation effects based on two-condition within-participant design．There are two approaches for conducting mediation analysis based on the within-participant design: test of joint significance and path-analytic method. The former is easy to use, and with relatively low type I error rate. In addition, each of its step directly corresponds to the criteria of Baron and Kenny (1986) described in the context of between-participant mediation analysis. However, this method ignores the estimates of the total effect of X on Y (coefficient “c”) and the effect of X on M (coefficient “a”), and as the result, can not obtain mediation effect size measure and model diagram. Moreover, test of joint significance is hardly applied to models with multiple mediators. Path-analytic method makes up for the aforementioned shortcomings of joint significance approach, but it suffers from a higher rate of type I error than test of joint significance approach. After comprehensively considering the advantages and disadvantages of the above two methods, and making some improvements for the test of joint significance approach about how to get coefficient “c” and “a”, a procedure is proposed and recommended to integrate both methods to analyze the mediation effect when data is based on a two-condition within-participant design. (1) Adopt the test of joint significance for both the paths of the mediation. If every path is statistically significant, the mediation is supported. Even so, we had better go on to Step 2 to get the confidence interval of ab.(2) Compute the confidence interval of the coefficients product of the both mediation paths using nonparametric Bootstrap method. The mediation is not supported and testing is over if the test of joint significance does not pass and the confidence interval of ab contain 0. Otherwise, we should go to Step 3. (3) Test the direct effect of the mediation model, a significant result indicates a partial mediation. Meanwhile, if indirect effect and direct effect have a same sign, the ratio of indirect effect to total effect is recommended to report as the mediation effect size measure.An instance is given to illustrate how to perform the proposed procedure by using the SPSS and Mplus software. The directions for future study on mediation analysis for within-participant design are discussed at the end of the paper, including alternative mediation analysis approaches and multiple mediation effects based on two-condition within-participant design, mediation for multiple-condition(>3) within-participant design, and mediation for other type of within-participant design.