项目反应理论与认知诊断模型的参数估计:模型整合视角

将极大似然期望最大化(maximum likelihood estimation via the expectation maximization,MLE-EM)算法拓展到了广义诊断模型估计中,并详细描述了如何使用期望最大化算法计算模型参数的极大似然估计值.从理论上明确指出,在认知诊断模型中存在的2类参数,即项目参数和结构参数,都是从观察数据中自由估计获得的.据此对项目反应理论和认知诊断模型中所用的边际极大似然估计期望最大化(marginal maximum likelihood estimation via the expectation maximization,MMLE-EM)算法理论进行了澄清,指出以往一些研究出现错误结论的原因.最后从模型整合的视角上为后续的研究提出了4条建议.

Model parameter estimation in item response theory and cognitive diagnosis: perspective of integration

This article extends maximum likelihood estimation by expectation maximization (MLE-EM) algorithm to estimate model parameters of general diagnostic model.We present details of EM algorithm for computing maximum likelihood estimates of model parameters.There are two types of model parameter in CDM in theory.Both item parameters and structural model parameters are freely estimated from observed data.These findings help to clarify previously established results of marginal maximum likelihood estimation by expectation maximization in IRT and CDM.Causes for misleading results in previous studies are identified.For model integration,four suggestions for future studies are provided.