Probability points to fraud

Statisticians have devised a simple model for detecting cheating in multiple-choice exams.

The system scans through test results to identify copying.

George Wesolowsky, a management scientist at McMaster University, Canada, said his model combined aspects of techniques devised by other researchers as well as his own methodology.

Studies have shown that as many as one in 20 students copies in such exams. Wesolowsky's model analyses every possible pair of students and compares their answers. It looks at the probability of getting a question correct depending on how difficult it is and the student's overall performance.

The model was devised to reduce the chances of picking out coincidental similarities between the answers of candidates and hence is conservative in its estimation of the level of cheating. "You do not want to make false accusations," Wesolowsky said.

He said the model was best utilised to determine whether there was a problem with the way exams were set and administered.

"Cheating is preventable by measures such as randomised seating and multiple versions of exams," he concluded.