> There have been studies that our typical interview process gives us 17% predictability of how they will perform, but if we do a contract-to-hire (of just one week) that improves to 80%.

The pool of candidates interested in full-time jobs is not the same as the pool of candidates interested in doing contract-to-hire positions.

Contract-to-hire selects for people with the ability to risk working for a company for a period of time without a high risk of near-term unemployment if it doesn't work out. The people willing to take those jobs are usually more qualified to begin with because they have more career options open to them if the contract-to-hire doesn't turn into a contract job.

So you're basically pre-selecting your candidates.

There are two different types of "contract-to-hire". This type, I'm talking about working for 1 week, most people think of "contract-to-hire" as a 6-month gig that _might_ turn into a position. This is more of a "trial week", but is still technically contract to hire. They are not two different types of people, but it is true that it is much harder for someone who has an existing position to take off 5 days -- but we did it in our company and made accommodations to make sure we fit. It also weeded out people who wouldn't fit.

So, you're agreeing that it works, then.

I think they were trying to convey the idea that not all very qualified candidates would even consider a contract-to-hire option. I know I wouldn't. That’s an unnecessary risk for me to take.

> improves to 80%.

Sounds interesting. If you happen to have links to any articles about that, it'd be interesting to read

I wish I kept the references, I didn't think they would be hard to refind but they have been. This is a reference I just found that's more recent (I found the others around 2016): https://www.qualified.io/blog/posts/truly-predictive-softwar...

Thanks! I've seen that article, it was posted at HN 1.5 years ago: https://news.ycombinator.com/item?id=22704116

Turns out I still had it open, half read, in one of my 100+ browser tabs :-)

Nice to hear that you recommend it. (Hmm I wonder if maybe some of the other things you've read were also based on Schmidt & Hunter, 1998?)

I have in mind to do as they recommend (incl work sample tests and structured interviews and work knowledge tests), when/if later on I'll look for people to help me with the software I'm creating.

I wonder b.t.w. if you know about any automatically generated GMA tests?

(Edit: I found some, I can post a link if you want.)

It'd be nice if there were ways to auto generate GMA tests. Then maybe it could be just fine if everyone was allowed to practice as how as they wanted -- if there was an unlimited supply of new questions, because they were auto generated.

Maybe something with generating random 3D shapes and applying rotations.

But then it seems to me it's necessary to measure how well begin good at such things, correlates with being able to learn and get good at software, and scrap any poorly correlated tests.

you get 80% chance of the idiots willing to work supposed contract to hire positions. Not sure that's anywhere near the same thing.

Some people prefer to work as contractors since that tends to pay a lot more, and they might have many many offers to choose among. And, paired with (not in the US) a well functioning social welfare system in case of really bad luck