News for hackers. Stuff that matters.
Re:Why Coding Tests Are A Bad Interview Technique by btilly on Monday November 02, 2009 @11:50AM (#917156)
I've been on both sides of this situation. I don't think there is a clear answer.

The fundamental problem is that interviews are a horrible way to select people. There is little correlation between doing well in an interview and being capable at your job. So in theory the more you make the interview be like the job, the more accurate an assessment it is.

For instance I have never seen an interview method produce better results when looking for a front end HTML person than giving them a picture of the page you want, giving them cutouts of all of the pieces they need, and asking them to write a web page that looks like that, will scale reasonably as you resize the browser, and with reasonably clean HTML. Seriously, I've seen the results that this test produces, and invariably it gives you someone that can actually produce web pages. Occasionally it lets someone with poor work ethic slip through, but that seems to be better than the alternative. Particularly if the interviewer does not have the skillset you are looking to hire.

And then there is the time issue. I have seen interviewing for a new person suck up top talent for weeks on end. You need more resources, and you are losing proven, trained resources to do something they probably don't like! Yet unless you can have someone with the skills you're hiring in the room, the interview process can't weed out people who don't have the right skills. And as Joel's fizzbuzz test demonstrates, most candidates don't have the skills you need.

How bad is it? Well Joel claims a significant portion of candidates can't write a program to count numbers, but with every multiple of 3 replaced by "fizz", every multiple of 5 by "buzz", and every multiple of 15 replaced by "fizzbuzz". I believe it. I have found that over 2/3 of candidates when given a double-loop to compute the intersection of 2 arrays can't figure out how to make it more efficient. And good luck if you ask people to do something unusual like a depth-first traversal of a tree!

But there is a limit. It is easy to produce a simple application that looks reasonable, but which takes a lot longer to put together than you think it will. If you can't, knowing how it is supposed to go, personally do the exercise in an hour or two on a freshly set up machine, the exercise is too hard. Personally I far prefer having a test with some simpler questions rather than a full blown mini-application.

But the opposite extreme of depending on interviews to filter out bad people is worse. Much worse.

There are no comments on this comment yet.