Changing Answers in Exams — for the Better or for the Worse?

During the year 2004, Utah State University offered three sessions of an introductory long–distance statistics course (Stat 2000) for students in its International Program in Hong Kong. These sessions were based on the electronic textbook CyberStats. The exams were given electronically and all student answers were stored in the CyberStats data base. All means that if a student changed the answer to question multiple times, all previous answers were still accessible and not only the final answer. In this paper, we investigate how those changes in answers affected students’ scores. 98.6% of the multiple–choice questions and 84.4% of text–based questions got answered at least once. Conditional on the fact that a question got answered at least once, text– based answers got changed at a rate of 10.8% that almost doubles the rate at which multiple–choice answers got changed (5.7%). Text–based answers got changed for the second time at a higher rate (1.3%) than multiple–choice answers (0.8%). Conditional on at least one change, the second response resulted in significantly more points for multiple–choice answers as well as for text–based answers. For text–based answers, this trend continued as the third response also resulted in significantly more points than the second response.