带有缺失数据的结构方程模型中的模型选择问题  

Model Selection of Structural Equation Models with Missing Data

作　　者：李云仙[1] 王学仁[2] 

机构地区：[1]云南财经大学金融学院保险系,云南昆明650221 [2]云南大学数学与统计学院统计系,云南昆明650091

出　　处：《数理统计与管理》2012年第6期1010-1021,共12页Journal of Applied Statistics and Management

基　　金：国家自然科学基会资助项目(10761011)

摘　　要：结构方程模型在社会学、教育学、医学、市场营销学和行为学中有很广泛的应用。在这些领域中,缺失数据比较常见,很多学者提出了带有缺失数据的结构方程模型,并对此模型进行过很多研究。在这一类模型的应用中,模型选择非常重要,本文将一个基于贝叶斯准则的统计量,称为L_v测度,应用到此类模型中进行模型选择。最后,本文通过一个模拟研究及实例分析来说明L_v测度的有效性及应用,并在实例分析中给出了根据贝叶斯因子进行模型选择的结果,以此来进一步说明该测度的有效性。Structural equation models are widely used in social, educational, medical, marketing, and behavioral sciences. In these fields, missing data are commonly encountered. To deal with this problem, structural equation models with missing data have been proposed. In the application of this kind of models, one of the most important issues is model selection. In this paper, we proposed an alternative Bayesian criterion-based method called the Lv measure for model selection of structural equation models with missing data. A simulation study and a real example are presented to demonstrate the efficiency and the application of the Lv measure, and results based on Bayes factor in the real example are also presented to illustrate the performance of the Lv measure for model selection.

关 键 词：缺失数据 贝叶斯方法 模型选择 MCMC算法 

分 类 号：O212[理学—概率论与数理统计] F832[理学—数学]