几种常用非补偿型认知诊断模型的比较与选用：基于属性层级关系的考量

当前国际上开发了60多种认知诊断计量模型(Fu &; Li, 2007), 各种模型各具特点, 实际应用者应根据实际情况选用恰当的模型。本研究以属性层级关系为切入点, 采用Monte Carlo模拟的研究方法, 比较了属性层级关系正确及有误两种情况下, 当前国际上常用的五种认知诊断模型的性能, 以充分考察不同认知诊断模型对属性层级关系的依赖程度, 及属性层级关系的错误界定对各认知诊断模型诊断正确率的影响, 从而为实际运用者在认知诊断模型选用上提供借鉴和参考。

Comparison and Selection of Five Noncompensatory Cognitive Diagnosis Models Based on Attribute Hierarchy Structure

Attribute hierarchy structure (AHS), which was considered as the basis of cognitive diagnosis, could largely affect the classification accuracy. However, in practical work, it was very difficult to determine whether the specified AHS was rational or not. Thus, it is necessary to explore how the AHS will affect the classification accuracy. This paper investigated the effect of different AHSs on the accuracy of diagnosis. And two AHSs were under investigation, one is the correctly identified AHS and the other is the incorrectly identified AHS. The commonly used Monte Carlo simulation method was employed to generate the data. And five cognitive diagnostic models, Rule Space Model (RSM), Attribute Hierarchy Model (AHM), General Distance Decision (GDD) Model, DINA_HC model and DINA model, were used to fit the same data. The results indicated that: (1) When the AHS was correctly identified, the attribute match ratios (AMRs) under RSM and AHM were both relatively low, while the AMRs under GDD, DINA_HC and DINA models were all relatively high. Furthermore, the AMRs under DINA_HC and DINA models were larger than that of GDD model. (2) When the AHS was incorrectly identified, the AMRs under RSM, AHM and GDD models were all relatively smaller compared to the case in which AHS was correctly identified, which indicated that the AHS significantly affects the accuracy of diagnosis of the RSM, AHM and GDD models. On the other hand, the influence of AHS on the accuracy of the diagnosis of DINA_HC model was moderate. But the accuracy of the diagnosis of DINA model will not be influenced by the AHS because the AHS information was not used in the DINA model.