多维数据IRT真分数等值和IRT观察分数等值研究  

IRT True Score Equating and Observed Score Equating for Multidimensional Data

作　　者：刘玥[1] 刘红云[2] 

机构地区：[1]四川省教育科学研究所,成都610225 [2]北京师范大学心理学院,北京100875

出　　处：《心理学探新》2015年第1期56-61,共6页Psychological Exploration

摘　　要：实际应用中测验往往具有多维结构,如果仍采用单维方法进行等值,会得到不准确的结果。研究基于随机等组设计下英语测验,使用MCMC方法估计题目参数,将单维IRT真分数等值和观察分数等值方法推广到多维。比较了四种等值方法:单维IRT真分数等值和观察分数等值,多维近似单维IRT真分数等值和观察分数等值。结果显示,当数据符合多维结构时:(1)基于多维测验的IRT真分数等值和观察分数等值方法优于单维IRT真分数等值和IRT观察分数等值方法;(2)多维IRT观察分数等值略优于多维IRT真分数等值,但是两者之间的差异较小。In practice,tests usually measure more than one trait. In these situations, if the multidimensional structure is neglected and traditional unidimensional item response theory （UIRT） equating methods are still used, the equated parameters may be inaccurate. The research data was selected from a large - scale English test,using random - group design. Item parameters were estimated using MCMC method. UIRT true score equating and observed score equating were extended to multidimensional structure. Four equating methods were conducted and compared：UIRT true score and observed score equating,multidimensional item response theory （MIRT） true score and observed score equating. The results demonstrated that ： （ 1 ） due to the multidimensional structure of our dat, MIRT equating methods performed better than UIRT methods ; （2） MIRT observed score equating was slightly better than MIRT true score equating, but the resuits of these methods didn＇ t show significant difference.

关 键 词：测验等值 多维IRT 真分数等值 观察分数等值 MCMC估计 

分 类 号：B841.2[哲学宗教—基础心理学]