The study set up is a bit unusual. Consider three ways to teach: A) Do what's best for the student's long term learning. ("deep learning".) B) Optimize for t

The study set up is a bit unusual. Consider three ways to teach: A) Do what's best for the student's long term learning. ("deep learning".) B) Optimize for the student doing well on a specific type of test (e.g. AP Calculus AB). ("teach to the test") C) Optimize for the student doing well on a specific set of questions known in advance. In the study, instructors had all three options. In most high stakes testing, instructors have only options A and B. The study seems to conflate B and C. But A and B are more similar, than are B and C. It would have been an interesting study if the instructors hadn't had access to the exams beforehand. (Although typical US university high stakes exams are designed by course instructors who can do (C), I don't see this as relevant, as instructors can more easily help their students do well by making the tests easier.)