一种多级评分的广义认知诊断模型

A Generalized Cognitive Diagnosis Model Under a Particuliar Polytomous Situation

认知诊断是近些年教育测量研究中的热点,大多数的认知诊断模型仅适用于0～1评分的情况。本文提出一种有多个潜变量多个滑动参数的多级评分认知诊断模型——GP-DINA,只要由评分标准和知识状态能确定理想反应模式,就可以利用此方法进行认知诊断分析。在该方法中,我们给出项目滑动矩阵的概念,将被试的观测得分均看成由某个理想得分的滑动,并采用EM算法估计滑动矩阵。在模拟研究中,采用每掌握一个属性得1分的评分标准,结果表明线性型、收敛型、发散型、无结构型和独立型五种属性层级结构均有较高的判准率。

Cognitive diagnosis has been a hot topic recent years. Most cognitive diagnosis models can be applied in merely dichotomous situation. This article considers a particular polytomous situation with each attribute is one point. At the same time, an examinee can slip to any possible score from his expectative score of an item with a specific probability. Based on this probability, this article gives a defi- nition of slip matrix and hence proposes a generalized model adapting to this particular polytomous situation. EM algorithm has been used to estimate model parameters. To test the accuracy of the model, this article also makes simulations considering five different kinds of attribute strictures with 3000 examinees and 35 items. The standards of accuracy are marginal match ratio and pattern match ratio. The result shows that both of the match rations decrease as the attribute structure changes.