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.