IRT框架下不可忽略缺失数据的Bayes估计  

Missing data modeling and Bayes estimation in IRT

作　　者：李斌[1] 李晓毅[1] 付志慧[1] 

机构地区：[1]沈阳师范大学数学与系统科学学院,沈阳110034

出　　处：《沈阳师范大学学报（自然科学版）》2015年第2期216-220,共5页Journal of Shenyang Normal University:Natural Science Edition

基　　金：国家自然科学基金青年基金资助项目(11201313);辽宁省科技厅博士启动基金资助项目(20131107);吉林省社会科学基金项目(2012B115)

摘　　要：项目反应理论主要研究被试在测验项目上的反应和成绩与潜在特质间的关系,能否有效的估计模型中的参数是项目反应理论能否得以应用的前提。数据的完整性对参数估计有一定的影响。而项目反应过程中,数据的缺失是常见的。缺失数据的机制影响了处理方法。因此,针对不可忽略缺失数据,利用潜变量建模法,采用等级评分模型拟合观测指标,Rasch模型拟合缺失指标。同时用Gibbs抽样法抽取参数,给出估计。通过模拟研究,验证了所用方法有效的减小了由于忽略缺失数据估计参数时产生的偏差。Item response theory mainly studys the reaction of the person in the test project and the relationship between the achievements and latent traits. The application of item response theory depends on whether the parameters of the model can be effectively estimated. The integrity of the data has a certain effect on the parameter estimation, while the missing of data in the item response process is very common. The mechanism of missing data affects the processing method. As a consequence, using the graded item response model to fit the observed data and Rasch model to the missing data and according to latent variable modeling method, is a solution to the non--ignorable missing data. Finally, the Gibbs sampling method is applied to extract the parameter and provide the estimation. By simulation study, it verifies the conclusion that the method effectively reduces the deviation owing to ignoring the missing data during parameter estimation.

关 键 词：项目反应理论 缺失数据 GIBBS抽样 

分 类 号：O212.8[理学—概率论与数理统计]