RT混合模型下五种线性等值方法跨分布比较  

Cross-distribution Comparison of Five linear Equating Methods Based on Mixed Model of Item Response Theory

作　　者：张敏强[1] 黎光明[1] 王小婷[1] 黄春汝 王幸君 

机构地区：[1]华南师范大学心理学院心理应用研究中心,广州510631

出　　处：《心理与行为研究》2015年第6期794-798,共5页Studies of Psychology and Behavior

基　　金：全国教育科学"十二五"规划教育部重点课题(GFA111009)

摘　　要：基于3PLM和GRM的混合模型下,采用非等组锚题设计,运用模拟方法比较平均数-平均数法,平均数-标准差法,稳健的平均数-标准差法,Haebara法和Stocking-Lord法五种线性等值方法,考虑被试数量和目标组被试能力分布对等值精度的影响。结果发现:(1)当目标组和基准组被试能力分布相似或差异较小时,平均数-平均数法胜出次数最多,等值精度最高;(2)当目标组和基准组被试能力分布差异增大时,平均数-标准差法逐渐占优。Based on Simulating some data of non-equivalent group for 3PLM and GRM mixed model of item response the- ory, this paper compared five linear equating methods such as Mean-Mean （MM） , Mean-Sigma （MS） , Robust Mean and Sigma （RMS） , Haebara （HA） and Stocking-Lord （SL） . The paper aimed at exploring whether the number of subjects and the ability distribution of target subjects have a significantly effect on equivalent accuracy. The result showed that： 1） when the target group and reference group subjects ability distribution were similar or their difference was small, the MM method was the best and had a highest equivalent accuracy; 2） when the difference of ability of distribution for the target group and reference group subjects increased, the MS method had a strong advantage.

关 键 词：IRT 混合模型 等值方法 跨分布比较 

分 类 号：B841.2[哲学宗教—基础心理学]