贝叶斯框架下一组英语测试数据的统计分析  

Statistical Analysis for a Group of English Testing Data from Bayesian Frame

作　　者：徐宝[1] 王宇廷[1] 马艺光 

机构地区：[1]吉林师范大学数学学院,吉林四平136000

出　　处：《吉林师范大学学报（自然科学版）》2017年第1期46-50,共5页Journal of Jilin Normal University:Natural Science Edition

基　　金：吉林省科技发展计划项目(20150101007);吉林省社会科学基金项目(2014B137)

摘　　要：在贝叶斯偏差信息(DIC)准则下,研究了一组大型英语测试的二级反应数据在传统的项目反应模型、随机效应题组反应模型以及带有题组判别参数的题组反应模型之间的模型选择问题以及题目参数和应试者能力参数的估计问题.首先,基于Gibbs抽样程序得到的各个模型参数的MCMC样本,计算出了各个模型的DIC函数值,然后经过比较,确定带有题组判别参数的题组反应模型是该组数据较优的拟合模型,最后基于带有题组判别参数的题组反应模型,研究了该项测试的题目参数和应试者的能力参数的后验期望估计以及这些参数的95%的等尾置信区间,所得结果在合理范围内.For a group of English testing dichotomous response data,the problem of model selection among the traditional item response model,the random effect testlet response model and the testlet response model with the testlet discrimination and the problem of item parameters and examinees＇ability parameters were studied under the Bayesian deviance information （DIC） criterion.First,the DIC function values of the three models were obtained by the MCMC sample value which generated from Gibbs sampling procedure.Then,through comparative analysis, the testlet response model with the testlet discrimination was chosen to be the optimal fitting model for this data.Finally,we obtained the Bayesian expected a posteriori estimation and 95% credibility interval for item parameters of these examinees. Generally , the computational results are reasonable.

关 键 词：模型选择 DIC准则 GIBBS抽样 

分 类 号：O212.8[理学—概率论与数理统计]