基于CTT的锚测验非等组设计中四种等值方法的比较研究

采用锚测验非等组设计的数据收集方案,对4种基于经典测量理论的等值方法进行了比较研究。研究数据取自TIMSS1999数据库,兼用等值标准误和交叉验证方法作为各等值方法比较的检验标准,利用CIPE程序对实验数据进行分析。研究结果表明,针对本研究所设置的等值情境,线性等值优于等百分位等值,其中Tucker线性方法比Levine观察分数线性方法更好一些,Braun-Holland线性方法不宜采用,频数估计等百分位方法等值误差较大,亦不足取。

Comparison of Four Equating Methods in Common-Item Non-Equivalent Group Design Based on The Classical Test Theory

Four equating methods in common-item non-equivalent group design based on the classical test theory were compared.The research data was selected from the TIMSS1999 database.Both the equating Standard Error and a cross-validation analysis were used as the evaluation criteria.And the CIPE program was used to compute the data.We came to a conclusion that,in our research situation,the linear equating procedures were superior to the equipercentile method.Of the linear equating methods,the Tucker linear method was better than the Levine Observed Score equating method,and the Braun-Holland linear method was not appropriate.The Frequency Estimation equipercentile method was with large equating error,and should not be used.