【作者】 高旭亮； 汪大勋； 涂冬波；

【Author】 Gao Xuliang;Wang Daxun;Tu Dongbo;School of Psychology, Jiangxi Normal University;

【机构】 江西师范大学心理学院；

【摘要】 认知诊断理论是认知心理学和现代测量学的产物,它的目的在于对个体内部知识结构进行诊断,并为后续的补救教学提供指导。认知诊断计算机化自适应测验(CD-CAT)不仅具有自适应测验的优点而且能够实现诊断的功能,近些年,CD-CAT逐渐成为国内外研究的热点领域。在CD-CAT题库建设过程中,选择合适的认知诊断模型(CDM)标定题库参数是一个很关键的步骤,当前CD-CAT题库建设均是基于单一认知诊断模型下构建的。实际应用中,一份测验中不同题目可能会拟合不同CDM(Ma, Iaconangelo, de la Torre,2016),如果所有题目采用同一个CDM来标定题目参数,则有可能造成题目与CDM之间失拟。GDINA是一个饱和认知诊断模型(Cognitive Diagnosis Models, CDM),Wald检验被用于在题目水平上检验GDINA是否可以被简化模型(如DINA,DINO,ACDM和RRUM)替代,并为测验的每一个题目选择一个最恰当的CDM(简称混合CDM),据此,本研究提出了一种基于混合模型的题库建设思路。通过Monte Carlo模拟研究验证基于混合模型的题库建设的效果,并与传统的基于饱和模型的题库,也即选用G-DINA来标定题库所有项目参数,以及另外一种基于简化模型(包括DINA、DINO、A-CDM和RRUM模型)的题库,也即分别用一个简化模型标定所有题目参数的题库建设进行比较。结果发现,本文提出的基于混合模型题库建设整体效果较好;且与传统的基于饱和模型和单一简化模型题库建设相比,新方法不仅有利于提高属性模式判准率,而且有助于降低题库的曝光率。更多还原

【Abstract】 Cognitive diagnosis or diagnostic classification models(CDM) hold great promise for such contexts because they can potentially support relatively fine-grained information about respondents that can be used for developing targeted interventions. CDM are psychometric models involving multiple discrete latent variables developed to provide specific information about the cognitive skills or attributes required to solve problems in a particular domain. The saturated G-DINA model may produce the best absolute fit indices because it is highly parameterized. However, its parameter estimates may not be stable, especially when sample size is not large. In that case, a simpler model might be preferred. A simpler model also provides more straightforward interpretation of the attribute–item relationship. Finally, appropriate reduced models can provide better classification rates than saturated models, especially when the sample size is small. In de la Torre and Lee(2013), the Wald test was used to compare the G-DINA model with DINA, DINO, and A-CDM, and demonstrated high power while controlling Type Ⅰ error. Compared with model selection at the test level, item-level model selection does not require blanket acceptance of a model for all the items, which may avoid suboptimal choices, and enable the researchers to investigate the best CDM for each of the items. Cognitive diagnostic computerized adaptive testing(CD-CAT) purports to obtain useful diagnostic information with great efficiency brought by CAT technology. The item bank is a very critical part of the CD-CAT. Traditional CD-CAT item bank is to use the same cognitive diagnostic model(CDM) calibration of all the parameters of items, however, a diagnostic test can not completely fit the same diagnostic model. Ma, Iaconangelo and de la Torre(2016) study shows that indicated that in many, if not all, test applications, no single reduced model can be expected to satisfactorily fit all the items. In regular CD-CAT, a item bank uses the same cognitive diagnostic model to calibrate the item parameters, however, all the items in item bank can not be fully fitted with the same CDM. In this study we selected an appropriate CDM for each item of item bank within a G-DINA framework and used a complex simulation study to compare the performance of Mixed model and five special models. The basic manipulated test design factors included generating models(G-DINA, Mixed, DINA, DINO, A-CDM, RRUM), item quality(high vs. medium), test length(5, 10, 15, 20), and item selection strategies(MPWKL vs. SHE).The results indicated that across all conditions, the true model and the CDM selected by the Wald test always provided the most accurate attribute classification, followed by the Other special models. And Mixed model had the lowest test overlap rates except that generating model was A-CDM.更多还原