A Quantal Choice Model for the Detection of Copying on Multiple Choice Examinations
提出基于决策理论量子选择模型的统计方法，通过三步流程估计学生答题匹配概率，比较四种方法后证明probit模型在检测抄袭时更准确且能减少误判。
ABSTRACT It is difficult to devise a statistical test to detect one student copying from another. Many prior efforts falsely accuse students of cheating. A general methodology based on the quantal choice model of decision theory overcomes these problems. Three steps are involved: (1) for each item estimate the probability and the variance that a given respondent will select each response, (2) for pairs of respondents, these probabilities determine the expected number of matches, and (3) compare the critical value to the number of items matched. Methods differ based on the probability estimation technique. Four methods (simple frequencies, Frary, Tideman, and Watts modification [8], logit, and multinomial probit) are compared on theoretical and empirical grounds. Theory and results show that it is crucial to incorporate the variance of the probability estimates. The probit model has theoretical advantages over the other methods and produces more accurate results.