IRT测验等值模型的选择——以广东佛山市中考数学实测数据为例  

How to Select Equating Models for College Entrance Examination Based on Item Response Theory

作　　者：黎光明[1] 张敏强[2] 

机构地区：[1]广州大学 [2]华南师范大学

出　　处：《中国考试》2012年第2期8-13,共6页journal of China Examinations

摘　　要：以广东省佛山市中考数学实测数据为例,说明IRT测验等值模型的选择过程,其基本步骤包括等值设计、参数估计、量表化和测验等值等。供选择的四种IRT测验等值模型分别为SL、SN、NR和GPCM模型。分析发现,等值误差由小到大依次是SL、SN、NR和GPCM,SL与GPCM、NR存在显著性差异(p<0.001),SN与GPCM、NR存在显著性差异(p<0.001),GPCM与NR存在显著性差异(p<0.001),但SL与SN不存在显著性差异(p>0.05)。结果表明:(1)从模型等值的效果看,SL和SN最优且相当,NR等值效果欠佳,GPCM等值变异最大,效果最差;(2)选择IRT的SL或SN等级反应模型进行此类测验等值较为合适。This research exemplified college entrance examination of Foshan city in Guangdong province and explored how to select the decent mode from SL, SN, NR and GPCM by such as equating design, data collection, prameter estimation, scaling, test equating, which could fit data by comparing the result of observed-score equating for four IRT models Methods. Compared merits and demerits of IRT models in terms of value of difference, we find： Rank of total error squared root were SL, SN, NR and GPCM. There was significant difference （p〈0.001）between SL and GPCM and NR and between SN and GPCM and NR and between GPCM and NR, but there wasn＇t significant difference （p〉0.05）between SL and SN. The research show： （1） SL and SN were the best from the equating effect, but NR wasn＇t good and GPCM was the worse; （2） It was wisdom and decent to select SL or SN to do test equating like this.

关 键 词：中考数学实测数据 项目反应理论(IRT) 测验等值 心理与教育测量 

分 类 号：G405[文化科学—教育学原理]