甄别多项选择题考试中答案抄袭的不同方法的比较  

Investigating the Robustness and Effectiveness of Different Methods of Detecting Answer Copying for Multiple Choice Exams

作　　者：刘景玉 肖立宏[2] 

机构地区：[1]美国Educational Testing Service [2]北京教育考试院命题处

出　　处：《考试研究》2008年第3期90-101,共12页Examinations Research

摘　　要：在美国,各个考试公司都会用不同的统计方法来检测考试中的舞弊现象。本文研究了两个检测舞弊的指数:基于经典考试理论的g2指数和基于项目反应理论的w指数。文章模拟了四种真实测试情形中常见的抄袭模式和几个可能影响指数的变量,研究结果表明,对于g2和w指数,在各种情形下,按照有偏差的估计参数以及真实参数计算出来的第一类错误率都是类似的,并且较低。因此,用有偏差的估计参数来计算g2和w指数不会增加将被抄袭者误认为抄袭者的可能性。而基于有偏差的估计参数的g2和w指数,只有在抄袭题目百分比较高且测试长度较长的情况下,才可能实现较低的第二类错误率。当抄袭题目百分比较低时,即便使用真实参数,g2和w指数都会造成较高的第二类错误率。All testing companies detect cheating on test by different statistical methods.This paper discusses two statistical methods:the g2 index,which is based on classical test theory;and the w index,which is based on item response theory.The purpose of this paper is to examine the robustness and effectiveness of the g2 and w indices when item parameter estimates and copiers' ability estimates are biased.The results show that,for both g2 and w,the type I error rates computed from estimated/biased and population/unbiased parameters under various situations were similar and low.In other words,using estimated parameters and abilities to calculate g2 and w will not increase the probability of identifying non-copiers as copiers.Low type II errors for indices of g2 and w based on estimated/biased parameters were obtained only when the percentage of copied items was high and the test length was long.When the percentage of copied items was low,both indices g2 and w produced high type II errors,even when the population/unbiased parameters were used.

关 键 词：经典考试理论 项目反应理论 名义回应模型 考试舞弊 

分 类 号：G424.74[文化科学—课程与教学论]