正则化方法对连续协变量项目功能差异的检测及其应用  

Regularization Method for Detecting Differential Items Functioning of Continuous Covariate and Its Application

作　　者：甘凯宇 刘红云[1] GAN Kaiyu;LIU Hongyun(Beijing Normal University,Beijing 100875,China)

机构地区：[1]北京师范大学,北京100875

出　　处：《中国考试》2023年第9期44-51,共8页journal of China Examinations

摘　　要：提出检测连续协变量条件下项目功能差异的正则化方法,并将其与Logistic回归方法进行比较。模拟数据分析结果表明:1)在所有条件下,正则化方法的一类错误率比Logistic回归方法低。在DIF项目比例为20%时,正则化方法的检测效果优于Logistic回归方法。2)正则化方法对0.3的DIF值不敏感,检验力低。3)两种方法的一类错误率随着样本量增加、DIF值增加而增加,检验力随着样本量增加、DIF值增加、DIF项目比例减小而增加。将正则化方法应用于PISA2012数学测验数据,进行连续协变量下的DIF检测及正则化方法的实际应用,结果也发现正则化方法相比于Logistic方法可以更好地控制一类错误率。In this study,a regularization method was proposed to detect differential items functioning under continuous covariate conditions and compared with logistic regression method.Simulation results show that:1)the regularization method has a lower TypeⅠerror than logistic regression method under all conditions,and when the proportion of DIF items is 20%,the regularization method has better detection effect than logistic regression method;2)the regularization method is insensitive to the DIF value of 0.3 and has low power;and 3)the TypeⅠerror rate of the two methods increased with the increase of sample size and DIF size.The power increased with the increase of sample size,DIF size and DIF item proportion.The regularization method was applied to detect DIF under continuous covariate in the math test of PISA2012.The practical application of regularization method was introduced.The results also showed that the regularization method can control Type I error rate better than the Logistic method.

关 键 词：项目功能差异 连续协变量 正则化 LOGISTIC回归 PISA2012 

分 类 号：G405[文化科学—教育学原理]