Love seeing papers that use data like this.
The most common mistakes seem to be ones easily caught by the compiler.
The 18 mistakes studied were all detected from either compiler errors or running a custom linter on student’s code. That fact doesn’t invalidate that conclusion – those 18 mistakes were not chosen for their easy detectability, but were rather derived from 20 mistakes that educators identified as common in interviews. However, it is possible, though unlikely, that instructors merely notice lintable errors more often than more subtle logical errors, and that one of those logical errors is more common than the ones studied.