[fsk]

Idiot: We don't want to make any bad hires! Therefore, we reject a lot of good candidates!

FSK: If you pass on many good candidates, and you have a small chance of hiring any given bad candidate, then each good candidate you reject actually INCREASES your odds of making a bad hire.

Idiot: We made a bad hire once! Never again! Now we reject lots of good candidates to avoid that repeat disaster!

FSK: But, if you want to minimize your bad hire rate, you also have to minimize the number of good candidates you reject.

Idiot: NO! NO! NO! The way you make sure you hire no bad candidates is to be so strict that you reject lots of good ones! That's what everyone told me so it must be right!

In one ear and out the other. Why do people who don't understand statistics get to be managers? If you don't understand this statistics argument I made, you're unqualified to work in any sort of technical area.

[ThrustVectoring]

It's actually fairly reasonable. The manager doesn't want to avoid hiring bad candidates. The manager wants to avoid being blamed for hiring bad candidates. They want to be able to say "I looked really really hard for reasons why Bob was a bad hire, and didn't find any, so you can't blame me for Bob being a bad hire."

Whether or not this actually reduces the number of bad hires is beside the point. It's a classic principle-agent problem.

[mcv]

You're missing something vital here. They're not rejecting good candidates because they're good candidates, they're rejecting them because they're not sure enough that they're good candidates.

They fear they might be bad candidates. They'd rather reject too many than too few, so they make sure they reject everybody who they're not 100% is a good candidate. That means they may reject some good candidates that aren't easily identified as such, but it doesn't increase their chance of hiring a bad candidate; it decreases it.

[smsm42]

Strictly speaking, I think raising the standards also could lower the probability of a bad hire (depends on the model - is bad hire completely random? does it depend on some parameter that is controlled?) together with a probability of a good hire, so I'm not sure it is a robust argument that raising standards always raises chance of a bad hire. I'd be happy to see a more rigorous proof (I know it's complete waste of time but for some people it's fun).

[ricardobeat]

On a side note, this kind of passive-aggressive comment is not very welcome here on HN. Let's please keep discussion civil.

[nrivadeneira]

Basing all of your management decisions purely on statistics is a bad strategy. You're assuming that the weighting of good vs. bad candidate is the same, when they're not.

Hiring a good employee is not nearly as impactful as hiring a bad one. If you can have a strategy that filters out 100% of bad employees, but unfortunately also filters out 90% of good employees, this is preferable to filtering only 50% of good employees, but also only filtering 90% of bad employees. You may have more good employees with the latter strategy, but the bad employees can kill the team.

[001sky]

People have an irrational belief in their ability to tell the difference. This is how one gets such flawed thinking in the first place (regarding priors and probability distributions).

Firms are hierarchy-based entities. "Technical Merit" is about a third or fourth order away from what really drives the hiring decision. Its also very imperfect predictor of actual performance.

The issues is that companies use the term all the time. They want people to believe they were hired for their merit, but merit is almost always 'fit' and not technical in nature. The technical hoops are just a CYA for when the 'fit' doesn't work out (mis-judgement) and they need something to point to as to that is not arbitrary in nature to explain how other people were not hired instead.