ISSN 1671-3710
CN 11-4766/R
主办：中国科学院心理研究所
出版：科学出版社

心理科学进展 ›› 2014, Vol. 22 ›› Issue (5): 731-745.doi: 10.3724/SP.J.1042.2014.00731

• 主编特邀 •    下一篇

中介效应分析：方法和模型发展

温忠麟;叶宝娟   

  1. (1华南师范大学心理应用研究中心/心理学院, 广州 510631) (2香港考试及评核局, 香港) (3江西师范大学心理学院, 南昌 330022)
  • 收稿日期:2014-02-06 出版日期:2014-05-15 发布日期:2014-05-15
  • 通讯作者: 温忠麟
  • 基金资助:

    国家自然科学基金(31271116)、教育部人文社会科学重点研究基地项目(11JJD190005)、中国博士后科学基金项目(2013M540535)、教育部人文社会科学研究青年基金项目(13YJC190029)和江西省博士后科研择优资助项目(2013KY08)资助。

Analyses of Mediating Effects: The Development of Methods and Models

WEN Zhonglin;YE Baojuan   

  1. (1 Center for Studies of Psychological Application / School of Psychology, South China Normal University, Guangzhou 510631, China) (2 Hong Kong Examinations and Assessment Authority, Hong Kong, China) (3 School of Psychology, Jiangxi Normal University, Nanchang 330022, China)
  • Received:2014-02-06 Online:2014-05-15 Published:2014-05-15
  • Contact: WEN Zhonglin

PDF (PC)

49737

摘要：

在心理学和其他社科研究领域, 大量实证文章建立中介效应模型, 以分析自变量对因变量的影响过程和作用机制。检验中介效应最流行的方法是Baron和Kenny的逐步法, 但近年来不断受到批评和质疑, 有人甚至呼吁停止使用其中的依次检验, 改用目前普遍认为比较好的Bootstrap法直接检验系数乘积。本文对相关的议题做了辨析, 并讨论了中介分析中建立因果关系的方法。综合新近的研究成果, 总结出一个中介效应分析流程, 并分别给出显变量和潜变量Mplus程序。最后介绍了中介效应模型的发展。

关键词: 中介效应, 间接效应, 逐步法, Bootstrap法, 因果

Abstract:

Mediation models are frequently used in the research of psychology and other social science disciplines. Mediation indicates that the effect of an independent variable on a dependent variable is transmitted through a third variable, which is called mediator. In most applied research, Baron and Kenny’s (1986) causal steps approach has been used to test mediating effect. In recent years, however, many methodological researchers questioned the rationality of the causal steps approach, and some of them even attempted to stop its use. Firstly, we clarify the queries on the causal steps approach one by one. Secondly, we propose a new procedure to analyze mediating effects. The new procedure is better than any single method that constitutes the procedure in terms of Type I error rate and power. The proposed procedure can be conducted by using observed variables and/or latent variables. Mplus programs are supplied for the procedure with observed variables and/or latent variables. Finally, this article introduces the development of mediation models, such as mediation model of ordinal variables, multilevel mediation, multiple mediation, moderated mediation, and mediated moderation.

Key words: mediating effect, indirect effect, causal steps approach, Bootstrap method, causality