ISSN 1671-3710
CN 11-4766/R

心理科学进展 ›› 2022, Vol. 30 ›› Issue (9): 2131-2142.doi: 10.3724/SP.J.1042.2022.02131

• 研究方法 • 上一篇    


王阳1, 温忠麟2(), 王惠惠3, 管芳4   

  1. 1广东金融学院公共管理学院, 广州 510521
    2华南师范大学心理学院/心理应用研究中心, 广州 510631
    3宁夏大学教育学院, 银川 750021
    4清华大学心理系, 北京 100083
  • 收稿日期:2021-10-01 出版日期:2022-09-15 发布日期:2022-07-21
  • 通讯作者: 温忠麟
  • 基金资助:

The second type of mediated moderation

WANG Yang1, WEN Zhonglin2(), WANG Huihui3, GUAN Fang4   

  1. 1School of Public Administration, Guangdong University of Finance, Guangzhou 510521, China
    2School of Psychology/Center for Studies of Psychological Application, South China Normal University, Guangzhou 510631, China
    3School of Education, Ningxia University, Yinchuan 750021, China
    4Department of Psychology, Tsinghua University, Beijing 100083, China
  • Received:2021-10-01 Online:2022-09-15 Published:2022-07-21
  • Contact: WEN Zhonglin


心理学研究常用有中介的调节模型揭示调节作用通过中介变量间接实现的现象。介绍了第二类有中介的调节模型(meMO-II)的概念及优势; 将meMO-II与其它中介调节混合模型进行了辨析; 给出了meMO-II的建模方法和分析流程, 并用一个实例演示; 介绍了基于潜变量的meMO-II的分析方法、meMO-II建模方法的新进展以及meMO-II的变式。研究有助于推动调节机制研究的发展。

关键词: 第二类有中介的调节, 调节机制, 两水平有中介的调节, 变量系统, 潜变量


Mediated moderation (meMO) accounts for the moderating effect of a moderator (W) on the relationship between an independent variable (X) and a dependent variable (Y) transmitted through a mediator (M). It has been widely applied in psychological studies. However, the traditional meMO (also known as the first type of meMO, or meMO-I) is difficult to interpret, and thus, researchers often misapply it. In this regard, the second type of meMO (meMO-II), which has been in the foreground in recent years, can better articulate the definition of meMO. meMO-II indicates that the moderating effect of W on XY is mediated by an intervening variable, M, which is affected by W initially. An important theoretical implication of meMO-II is that it reveals how W moderates the effect of X on Y, for which meMO-I has limited applicability.

We compare and contrast meMO-II with other models that combine moderation and mediation. Then, we demonstrate the analytical procedure behind meMO-II. We conducted this analysis in three steps. The first step was to test whether there was a total moderating effect of W on X and Y. With Y as the dependent variable, we performed a hierarchical regression analysis in which the independent variable X and moderator W are entered first into the regression, followed by the interaction term XW. If there is a total moderating effect, this will be confirmed by XW's regression coefficient being statistically significant and ΔR2 being high enough (e.g., ≥ 0.02 or even ≥ 0.03). The second step was to test whether there was an indirect moderating effect. If the regression coefficient a1of W from the regression of M to W, and the coefficient b2from interaction term XM from the regression of Y to X, W, M, XW and XM, are significant, respectively, then the indirect moderating effect is significant (causal steps approach). If at least one coefficient was insignificant, we would apply the bootstrap method with higher power to test the product coefficient a1b2 directly. If the 95% bootstrap confidence interval of a1b2did not contain 0, the indirect moderating effect was significant (product of coefficients approach). The third step was to conduct a simple slope analysis. In accordance with the relation between the moderators W and M, we fixed the two variables at a set level to investigate the difference in the effect of X on Y at different points. This method is known as the pick-a-point technique. The Johnson-Neyman (J-N) technique can also be adopted to identify the range of W values for which the main effect between the X and Y is significant. Furthermore, we applied the above procedure to an empirical study as an illustration.

Then, we present two analytical methods of meMO-II based on latent variables: latent moderated structural equations (LMS) and the factor score approach (FS). Several recent advances in modeling approach of meMO-II, including variable system (VS) and two-level mediated moderation (2meMO) are also introduced. As for VS, the product of path coefficients from XWXMY is employed as the index of indirect moderation by multiplying X to the regression of M to W. Thus the disadvantages of endogeneity, incoherent paths and confusion about meMO-II, and the second-stage moderated mediation model in meMO-II, can be mitigated, whereas 2meMO uses error term partitioning in a multi-level model as a reference, applying the random effect of X on Y to the meMO-II model. This saves the trouble of verifying the homoscedasticity assumption.

Additionally, this paper also presents several effect sizes concerning indirect moderation. These include the ratio of the indirect moderating effect divided by the total moderating effect, υ (i.e., the square of a1b2), and an effect size based on the variance decomposition of coefficients ϕMO_ind(i.e., the ratio of the variance of a1b2 to the total variance of the path coefficients from XY).

Finally, we present the variant models of meMO-II, including multiple indirect moderation, suppression effect in moderation, multi-level indirect moderation, indirect moderation of categorical variables, moderated indirect moderation and indirectly moderated mediation.

Key words: the second type of mediated moderation, moderating mechanisms, two-level mediated moderation, variable system, latent variables