英语阅读测试中不同认知诊断模型比较研究

A Comparative Study of Different Cognitive Diagnosis Models on English Reading Tests

选取认知诊断研究中常见的一般化模型G-DINA、连接型约束模型NC-RRUM和DINA、补偿型约束模型C-RUM和DINO,从横向加工机制和纵向层级关系两个方面开展对比研究,考察不同类型诊断模型在英语阅读测试方面的适切性。使用似然比检验方法对比各类模型在相对拟合指标与绝对拟合指标上的差异,使用模型分类的一致性指标和精准度指标考察诊断的信度和效度。结果表明:(1)G-DINA和NC-RRUM模型与阅读测试数据的拟合度较好,二者显著高于其他模型,其中,一般化G-DINA模型属性分类一致性较高,约束化NC-RRUM模型属性分类精准度最优;(2)诊断模型与测试数据的拟合优度随着属性层级结构的削弱而增加,结构关系最为松散的独立结构模型的数据拟合度最佳,表明阅读能力不具备严格的层级关系。该结果可为研究人员探究智能化阅读诊断提供依据,为英语教师在阅读诊断实践中的模型选择提供参考。

The present research compared the performance of five widely-used cognitive diagnosis models(CDMs)for assessing subjects'mastery and non-mastery on a set of fine-grained skills(i.e.,attributes)in EFL reading comprehension,including two conjunctive models(DINA and NC-RRUM),two compensatory models(DINO and C-RUM),and a generalized model(G-DINA).The comparison was conducted from two perspectives:the conjunctive/compensatory mechanism of English reading ability,and the hierarchical structure of reading attributes.The performance of the models was compared with a likelihood ratio test in terms of relative and absolute fit measures,classification consistency and accuracy.Results of the study demonstrate that G-DINA and NC-RRUM achieved the best fit to the reading assessment data among the five CDMs.While G-DINA produced the most consistent attribute classification patterns,NC-RRUM generated the most accurate patterns of attribute classification.In addition,the hierarchical relationship was not identified among the attributes in English reading comprehension,suggesting that the nine attributes defined in the present study were independent of each other.The findings of this study have both theoretical and practical implications regarding the exploration of automatic diagnosis of test takers'language abilities.This study could also provide guidance for language teachers in selecting models to diagnose students'English reading abilities in the classroom.