匹配变量纯化的测验偏差检验方法

CSIBTEST方法是基于参照组和目标组2个测验信度对真分数进行估计,再按交叉位置分数将匹配分数划分为2类子样本,并分别计算其卡方统计量,然后将这2个独立的卡方统计量相加得到自由度为2的检验统计量.鉴于测验信度具有群体依赖性,即不同群体的测验信度可能不尽相同,而CSIBTEST方法将参照组和目标组分别划分为2类子样本,有必要对子样本上的测验信度也进行估计,由此拓展了CSIBTEST.新方法先使用CSIBTEST获得交叉位置参数,相当于进行DIF预分析,再使用子样本上的信度估计用于真分数估计,以在对匹配变量进行纯化后获得检测统计量.模拟研究结果显示:相比SIBTEST和CSIBTEST,匹配变量纯化的测验偏差检验方法对存在DIF试题有着更高的统计检验力.

The Matching Score Purification for Differential Item Functioning Method

The CSIBTEST method estimates the true scores based on the test reliability of the reference group and the focus group,then separates the matching scores into two kinds of sub-samples according to the crossing location,and computes the Chi-squared statistics respectively,and then adds the two independent statistics to obtain the test statistics with a degree of freedom of 2.From the view of the group dependence of test reliability,that is,the test reliability of different groups may be different,and the CSIBTEST method separates the reference group and the focus group into two sub-samples respectively,it is necessary to estimate the test reliability on the sub-samples for the extension of the CSIBTEST.The new method first uses the CSIBTEST to obtain the cross location,and then applies the reliability estimation on the sub-samples for the true score estimation as matching score purification to obtain the test statistic.The simulation study shows that the new method with matching score purification has higher statistical test power for the bias item than the SIBTEST and CSIBTEST.