在GLMM框架下统一GT与IRT  

Using GLMM to Unify GT and IRT

作　　者：薛明锋 陈平 刘拓 甄锋泉 Xue Mingfeng;Chen Ping;Liu Tuo;Zhen Fengquan(Collaborative Innovation Center of Assessment toward Basic Education Quality,Beijing Normal University,Beijing,100875;Tianjin Normal University,Tianjin,300387;School of Psychology,South China Normal University,Guangzhou,510631)

机构地区：[1]北京师范大学中国基础教育质量监测协同创新中心,100875 [2]天津师范大学,天津300387 [3]华南师范大学心理学院,广州510631

出　　处：《心理科学》2021年第2期449-456,共8页Journal of Psychological Science

基　　金：中国基础教育质量监测协同创新中心基础教育质量监测科研基金项目(2019-01-082-BZK01,2019-01-082-BZK02);北京师范大学中国基础教育质量监测协同创新中心自主课题(BJZK-2018A2-18018)的资助。

摘　　要：本文首次提出使用广义线性混合模型(generalized linear mixed model,GLMM)对概化理论(GT)和项目反应理论(IRT)进行统合,即在一次统计中就能同时获得GT和IRT所需要的估计结果。模拟研究结果显示:相比于传统的GT方差分量估计方法——期望均值平方(expected mean squares,EMS),GLMM可以获得更准确的方差分量、概化系数(G系数)和可靠性系数(Φ系数)估计值;相对于传统Rasch模型,GLMM可以获得更准确的题目难度参数估计值。实证研究展示GLMM在实际心理测量数据分析中的应用。It is important to examine the quality of psychometric tools(e.g.,ability tests and personality scales)before they are formally administered.Thus,generalizability theory(GT)and item response theory(IRT)are becoming more and more popular.Though some efforts have been made to combine GT and IRT,most of the research continues to employ GT and IRT separately.That is because previous models including generalizability in IRT modeling(GRIM)and hierarchical rater model(HRM),are complicated and lack analysis programs.Therefore,this paper proposes using generalized linear mixed model(GLMM)to unify GT and IRT.GLMM is an extension of linear mixed model.By exploiting various link functions,response variables are no longer limited to be continuous data in GLMM.Thus,it is promising to use GLMM to analyze discrete data such as dichotomous data.There are a lot of advantages to unify GT and IRT under the framework of GLMM.First of all,GLMM can estimate both the variance components that are key components in GT and the difficulty parameters that are necessary for IRT in one step.Secondly,GLMM is simpler than other models.Besides,GLMM can be performed in many programs such as lme4 package in R,HLM,and so on.Last but not least,compared with expected mean squares(EMS),the traditional method to estimating variance components in GT,GLMM can avoid violating the assumption of the interval scale,thereby improving the reliability of the analysis.To illustrate the feasibility and strengths of GLMM,a simulation study and an empirical study were conducted.In the simulation,σp2 andσi2 were respectively set to be 2×π2/2 and 1×π2/3.The reason why they were set to be multiples ofπ2/3 was that the default residual variance of binominal GLMM using logit as linking function wasπ2/3.Setting true parameters of variance components in this way provided us with a simple proportional relationship.Person effect and item effect were randomly drawn from N(0,σp2)and N(0,σp2)respectively,and the item effect was treated as easiness parameter.By employi

关 键 词：广义线性混合模型 概化理论 项目反应理论 

分 类 号：B841[哲学宗教—基础心理学]