[mathgradthrow]

>regardless of the incentive, it does nobody any favors to put them in a situation where they're going to fail.

There are a number of incorrect assumptions hidden in this statement. Foremost is the idea that academic failure is always regular failure. I know people who started in math amd pivoted after they didn't think they could handle being full fledged mathematicians, but who benefitted from the math they did learn enough that it was worth it to them.

Secondly, you can't know in advance who is going to fail, and people are not obviously better served by a strategy that plays it safe.

[antasvara]

Colleges have limited capacity, and individual departments have even less. Given this information, what criteria would you suggest be used when determining admission into a department if not "likelihood of academic success?"

[dgacmu]

This is the core thing. Our acceptance rate is extremely low - under 10% for computer science. There are a lot of things we want to accomplish in admissions; _one_ of them is a statistical bias towards being able to succeed in the program. We obviously can't predict super accurately for a single student, but we try to create scoring criteria that result in success. Dropping the SAT caused a statistically significant reduction in success. We look at a lot of things, but, much as Harvard observed, a lot of those other things have even more bias. If you rocked the AIME, for example, we know a lot about your math skills, but there's a lot of bias ($$$) in which schools make it possible to participate in that, and there are a lot of students who will do well despite not having done AMC/AIME. Etc.