全国英语等级考试锚测验非等组设计中样本量对等值结果的影响

The Effects of the Sample Size on the Result of Test Equating on the Common-Item Nonequivalent Group Design for PETS

本研究采用锚测验非等组设计,探究了锚测验样本量的变化对等值结果的影响。数据来自全国英语等级考试(PETS),使用以Bigsteps为核心的自主改进软件,基于Rasch模型估计题目参数。为了探究等值结果对合格分数线的影响,本研究对比了不同样本量锚测验参数估计值与锚题参数给定值的差异,并对不同样本量锚测验得出的实考试卷等值结果与最大样本量锚测验得出的实考试卷等值结果进行了差异分析。结果表明,当锚测验样本量达到150时,等值结果比较稳定。这一结果表明,PETS设定的300人左右的锚测验样本量是合理的。

This study explores the effects of the sample size on the result of test equating on the common-item nonequivalent group design, using data from the Public English Test System （PETS）. This study uses a self- improved Rasch-based software tool called Bigsteps to estimate the item parameter. To explore the effects of test equating on the passing score, the researchers compare the parameter estimate based on the different-sample-size anchor test and the given value of the parameter, and analyze the difference between the equating results on the live test based on the different-sample-size anchor test and the largest- sample-size anchor test. It turns out that the equating result becomes stable when the increasing sample size reaches 150. Therefore, the sample size of 300, which PETS uses, is reasonable.