[pandaman]

The more you try to decrease false positives, the more you get false negatives, this is something taught in any intermediate Statistics class. Different people may have different opinions on what rate of false positives is acceptable but only ignorant can claim that you can eliminate false negatives "for free", without getting hit with more false positives.

The example given in my stats textbook was going on an expensive cruise and, as you ride to the port, you realized that you might have left an iron on (that was an old book, people used irons to make their clothes smooth). If you turn back then you miss your ship and take a loss on your whole cruise. But if you don't and the iron is actually on then you lose your house. So, do you want to take the false negative (the iron is off and you lost few grand you paid for the cruise) or the false positive (the iron is on and lost few hundred grand you paid for the house but, as a consolation, you enjoyed a cruise)?

Apparently businesses do not suffer from their preference to false negatives and there are not many (or any) companies with an easy interview process, which are also attracting many applicants, so the avoidance of false positives does not seem irrational.

[Certhas]

This is nonsense. Yes, given a prediction method that produces a probability, there is a tradeoff between false negative and positive. (Your example doesn't exhibit this though). But of course you can decrease your rate of false negative without affecting your fake positives: If you change your prediction method!

This is literally the first image in the relevant Wikipedia page:

https://en.m.wikipedia.org/wiki/Receiver_operating_character...

At a given acceptable level of false positives, different predictors have different rates of false negatives.

[pandaman]

This is what we get for replacing education with googling... Are you suggesting replacing interviews with some other metrics?

[mrjet]

He is suggesting that better methods of interviewing could increase “area under the curve”, improving both fase negative and false positive rates at once.

[pandaman]

I understand that. But even if you come up with a classifier that gives better false positives and false negatives rates than the existing process, it's still going to be tuned towards minimizing false positives and leaving false negative to grow freely. My point was that only companies that do that became successful enough to attract a whole lot of applicants who fail interviews and complain about them on social media.

[rsynnott]

> that was an old book, people used irons to make their clothes smooth

I mean, er, while I don’t personally own an iron, myself, I’m fairly sure anyone who regularly wears shirts and things _still does_.