Yesterday I found this NBER-working paper via this tweet:

And indeed it’s true, the study has two interesting findings:

• Shared wrong answers were a more powerful indicator than shared right ones -> so focus on this one. The algorithm being used is indeed quite simple.
• And the solution:

Students seated themselves voluntarily, with the expectation that the seats they chose would be the ones in which they would take the exam. These seating choices were recorded. Prior to the actual test, however, students were randomly reassigned to different seats. Thus, we are able to observe the patterns in correlations among students who wanted to sit together, but then were not allowed to.

And surprise: when students are not allowed to pick where they would sit, the amount of cheating disappeared…

Abstract of the working paper:

We develop a simple algorithm for detecting exam cheating between students who copy off one another’s exam. When this algorithm is applied to exams in a general science course at a top university, we find strong evidence of cheating by at least 10 percent of the students. Students studying together cannot explain our findings. Matching incorrect answers prove to be a stronger indicator of cheating than matching correct answers. When seating locations are randomly assigned, and monitoring is increased, cheating virtually disappears.