[numinary1]

This discussion misses an important element, the skill of the interviewer. It is unsurprising that unskilled interviewers' assessments are poor predictors of future performance. It would be interesting to measure the accuracy of interviewers who have had years of experience interviewing, hiring, and managing people.

Here's how I think it works. Skilled interviewers are biased toward rejecting candidates based on any negative impression. Structured interviewing has the same effect. It's the precision versus recall tradeoff. For this use case only precision matters. Extremely low recall is fine.

Also, in the GPA prediction example, the interviewer is penalized for predicting a low GPA for a person who performed well. But in hiring, there is no penalty for failing to hire someone who would have performed adequately.

(Yes, I understand there is an implicit assumption in my argument that candidates are not in short supply, but that's usually true, certainly at Google)

[ScottBurson]

> For this use case only precision matters. Extremely low recall is fine.

Only if you have arbitrarily large amounts of time you can spend interviewing. Most of us don't.