基于残差分析的0-1评分认知诊断模型项目拟合指标研究  

Assessing item fit for dichotomous cognitive diagnosis models based on residual analysis

作　　者：李姗姗[1] 张淑梅[2] 金蛟[2] 张雪晴[1] 

机构地区：[1]北京师范大学数学科学学院数学与复杂系统教育部重点实验室,北京100875 [2]北京师范大学统计学院,北京100875

出　　处：《北京师范大学学报（自然科学版）》2015年第6期559-563,共5页Journal of Beijing Normal University(Natural Science)

基　　金：国家自然科学基金资助项目(11171029)

摘　　要：认知诊断模型超过100种,在实际中选择合适的模型进行数据分析是十分重要的.本文针对0-1评分的认知诊断模型,提出2个项目拟合指标,即标准化残差和统计量z,为解决模型选择问题提供了一种途径.模拟研究得到统计量z大样本时渐近服从标准正态分布的猜想,以及在属性间紧密度较低时,2个指标对模型拟合不足识别效果较好.More than 100 cognitive diagnosis models exist,therefore choosing a suitable model to do statistical analysis is rather a challenge.The dichotomous cognitive diagnosis models were studied,two item fit indices standardized residual and statistic z were used to solve the problem of model selection.Simulation confirmed conjecture that asymptotic distribution of statistic z was standardized normal distribution in a large sample,and the accuracy of two indices was high when the compactness among the attributes was low.

关 键 词：认知诊断 0-1评分 模型评价 项目拟合 残差 

分 类 号：B841[哲学宗教—基础心理学]