The algorithm essentially compares the number of shared incorrect answers between students sitting next to each other, with those between students not sitting next to each other. In other words, students could sit immediately next to each other (i.e. not spread out around the room). Levitt and Lin find:
Students who sit next to one another on the midterm have an additional 1.1 shared incorrect answers... students who sat next to each other have roughly twice as many shared incorrect answers as would be expected by chance.Essentially, they find that "upwards of ten percent of the students cheated on the midterm in a manner that is detectable using statistics". Then for the final exam, they changed things up and cheating fell to levels that were basically null (the analysis suggested that four students cheated in the final exam).
What did they do? A number of things changed:
- There were four teaching assistants invigilating the exam, rather than one (for a class of over 240 students);
- There were two different versions of the exam, which were randomly allocated to students; and
- Students were randomly allocated to seats.
One last point - it is not straightforward to attribute the relative decrease in cheating to the randomisation alone. This is because the professor sent a series of emails to students calling for confessions of cheating following the midterm and reminding students that "cheating is morally wrong". So, students were probably primed for enhanced attention to be paid to their exam behaviour and extra-vigilant not to cheat (or to be perceived as cheating). No doubt this was reinforced by the change in procedures noted above.
Having said that, having students seated immediately next to each other for a test is just stupid (not having 100% multiple choice exams is probably a good idea either). If you don't want your students to cheat on exams, don't make it easy for them to cheat on exams. Increasing the costs of cheating (the penalties remain the same, but the probability of being caught increases when you have more supervision) and lowering the benefits (randomising seating so it is harder to sit beside or behind your friends) reduces the incentives for cheating. Then you wouldn't need Levitt and Lin to find the cheaters.