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 ›› 2011, Vol. 19 ›› Issue (5): 765-774.

• 研究方法 • 上一篇    

中介效应的三类区间估计方法

方杰;张敏强;李晓鹏   

  1. 华南师范大学心理应用研究中心, 广州 510631
  • 收稿日期:2010-12-03 修回日期:1900-01-01 出版日期:2011-05-15 发布日期:2011-05-15
  • 通讯作者: 张敏强

Estimating Confidence Intervals of Mediating Effects by Using the Distribution of the Product, Bootstrap and Markov chain Monte Carlo Methods

FANG Jie;ZHANG Min-Qiang;LI Xiao-Peng   

  1. Research Center of Psychological Application, South China Normal University, Guangzhou 510631, China
  • Received:2010-12-03 Revised:1900-01-01 Online:2011-05-15 Published:2011-05-15
  • Contact: ZHANG Min-Qiang

PDF (PC)

5001

被引次数

104

摘要： 由于中介效应ab的估计量通常不是正态分布, 因此需用不对称置信区间进行中介效应分析。详述了三类获得不对称置信区间的方法, 包括乘积分布法(M法和经验M法)、Bootstrap方法(偏差校正和未校正的非参数百分位Bootstrap方法、偏差校正和未校正的参数百分位残差Bootstrap方法)和马尔科夫链蒙特卡罗(MCMC)方法。比较了三类方法在单层(简单和多重)和多层中介效应分析中的表现, 发现三类方法的表现相近, 与乘积分布法相比, 偏差校正的百分位Bootstrap方法表现较好, 但有先验信息的MCMC方法能更有效降低均方误。最后对中介效应不对称置信区间研究的拓展方向做了展望。

关键词: 中介效应, 置信区间, 乘积分布法, Bootstrap方法, 马尔科夫链蒙特卡罗(MCMC)方法

Abstract: Because the estimators of mediating effects are generally not normally distributed, it would be better use asymmetric confidence intervals to analyze mediating effects. There are three approaches to obtain the asymmetric confidence intervals of mediating effects: 1) Based on the distribution of the product (including M method and Empirical-M method); 2) Bootstrap methods (nonparametric percentile Bootstrap method, Bias-corrected nonparametric percentile Bootstrap method, parametric percentile residual Bootstrap method and bias-corrected parametric percentile residual Bootstrap method); 3) Markov chain Monte Carlo (MCMC) methods. After introducing each of the methods in details, we compared them and found the following results: 1) the behaviors of the three approaches were approximate. 2) Compared with Distribute of the product methods, bias-corrected percentile Bootstrap method was better. 3) The mean square error (MSE) of the MCMC with prior information was smaller than Distribute of the product methods. Directions for further research on asymmetric confidence intervals of mediating effects were discussed.

Key words: mediating effect, confidence interval, distribute of the product method, Bootstrap method, Markov chain Monte Carlo (MCMC) method