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 (7): 1083-1090.

• 研究方法 • 上一篇    

缺失数据处理的期望-极大化算法与马尔可夫蒙特卡洛方法

沐守宽;周伟   

  1. 漳州师范学院教育系, 福建 漳州 363000
  • 收稿日期:2010-10-29 修回日期:1900-01-01 出版日期:2011-07-15 发布日期:2011-07-15
  • 通讯作者: 沐守宽;周伟

Handling Missing Data: Expectation-Maximization Algorithm and Markov Chain Monte Carlo Algorithm

MU Shou-Kuan;ZHOU Wei   

  1. Department of Education, Zhangzhou Normal University, Zhangzhou 363000, China
  • Received:2010-10-29 Revised:1900-01-01 Online:2011-07-15 Published:2011-07-15
  • Contact: MU Shou-Kuan;ZHOU Wei

PDF (PC)

4017

被引次数

28

摘要： 缺失数据普遍存在于心理学研究中, 影响着统计推断。极大似然估计(MLE)与基于贝叶斯的多重借补(MI)是处理缺失数据的两类重要方法。期望-极大化算法(EM)是寻求MLE的一种强有力的方法。马尔可夫蒙特卡洛方法(MCMC)可以相对简易地实现MI, 而且可以适用于复杂情况下的缺失数据处理。结合研究的需要讨论了实现这两类方法的适用软件。

关键词: 缺失数据, 期望-极大化算法, 马尔可夫蒙特卡洛方法, 极大似然估计, 多重借补

Abstract: Dataset with missing data is quite common in psychological research, which usually creates major problems in statistical inference. Maximum Likelihood Estimator and Multiple Imputation based on Bayesian Estimator are most important methods of handling missing data. Expectation-Maximization Algorithm, included in Maximum Likelihood Estimator is quite advantageous to flexible use and accurate results, while Markov Chain Monte Carlo Algorithm may achieve multiple imputation more easily and can be applied to handling missing data in complex situations. Finally, suitable statistical software is discussed in the field of psychological study.

Key words: missing data, Expectation-Maximization algorithm, Markov Chain Monte Carlo algorithm, Maximum Likelihood Estimator, multiple imputation