> There were quite a few admissions by hiring managers at surprisingly well known companies that the ability to pass algorithm challenges does not correlate with success on the job.
There are a lot of anecdotes like that, but the statistics I've seen has clear correlation between performance on algorithm interviews and performance on the job. I don't think there are any public studies done on this though so if you don't work at a company where you can view it internally you just have to trust that someone has gone through the numbers at these big data-driven companies.
Well, I can say from personal experience that there are interviews when I've aced the algorithm question(s), and other interviews when I haven't. Not hugely different in time interviews, either. So I can say from personal experience that it doesn't correlate all that well with my own performance, or it would be constant.
I expect, though, that the issue is that the degree of correlation measured will depend on the population you're screening with it. If you are wanting to screen out fraudulent or otherwise utterly non-programmer persons, it will show a good correlation. However, among those who are actually programmers, it will show much less correlation, and if you're in a condition of developer scarcity, the false negatives could be more costly than the false positives.
> Well, I can say from personal experience that there are interviews when I've aced the algorithm question(s), and other interviews when I haven't. Not hugely different in time interviews, either. So I can say from personal experience that it doesn't correlate all that well with my own performance, or it would be constant.
Do you even understand what correlation means? There is a significant random element to it, yes, but that doesn't mean that there is no correlation, and the random parts from the same individual can be mitigated by doing many interviews after which you have reasonable correlation. Most interview styles has close to zero correlation, so even if we don't reach anything remotely close to 1 it is still pretty good.
Edit: About the points that it is enough to weed out people using simple problems, that isn't true either. There was significant correlation between job performance and interview performance even among those who did well. The statistics showed no signs of it capping out either, so for all we know we could make them even harder, just that there wouldn't be that many left that could pass them then.
> There are a lot of anecdotes like that, but the statistics I've seen has clear correlation between performance on algorithm interviews and performance on the job.
I can only claim to have personal anecdotes, but the critiques I hear and have are things like asking you to write Bash on a whiteboard an focusing on minor syntax errors (Bash was not the primary language for the job and the candidate was obviously familiar with the language). Or asking about details from compiled languages when the job exclusively dealt an interpreted language. I'm not saying these interviews shouldn't touch on these topics. I do like to see how deep a candidates knowledge goes (sometimes it can mean they're over qualified and would be bored).
I'm not sure if it's interviewers showing off, because they've "always done it this way," or what, but I often see both job posting criteria and interviews focus too much on things well beyond what the job requires.
Of course, I'm biased and think my way is better. Outside of the minimum requirements for the job I never really care if a candidate has an answer to most questions. In fact, when they get it too quickly it seems rehearsed and I don't think it reflects their skill. For me, the technical questions are a starting point to see them talk through how they approach the problem.
There are a lot of anecdotes like that, but the statistics I've seen has clear correlation between performance on algorithm interviews and performance on the job. I don't think there are any public studies done on this though so if you don't work at a company where you can view it internally you just have to trust that someone has gone through the numbers at these big data-driven companies.