Hacker News Reader

Generally you don’t get all the resumes at once, this is literally the secretary problem. Unfortunately you need some objective standard of measurement across candidates that is static over at least months, and ideally takes a minimum of you and your candidates’ time. Though imperfect, real-time technical questions are a pretty good solution to this problem, especially since a lot of the best engineers prefer them to take-homes.

Also, anecdotally after interviewing hundreds of candidates, I’ve found resumes to have a relatively high false positive rate. Surprising numbers of engineers look great on paper but can’t write working code for semi-complex problems when tested, even accounting for interview anxiety.

> Surprising numbers of engineers look great on paper but can’t write working code when tested, even accounting for interview anxiety.

How have you accounted for interview anxiety?

It’s definitely not an exact science.

1. I try to make the parameters of the interview as friendly as possible — language of their choice, don’t have to worry about code style or computational complexity, or talk while they’re coding, just write working code that solves the problem. They can use google freely. I’m there as a resource to talk through their solution, answer questions, and help them get unstuck if needed.

2. The algorithm to solve the question doesn’t require a trick and there are many viable solutions so most candidates can verbally explain to me a working algorithm quickly, +- a couple edge cases.

3. Candidates often get visibly flustered or hung up on irrelevant details, or unsure what exactly I want for them, and I focus on noticing this. I try to guide them down a correct path if they’re close and tell them not to over-architect if they start doing that.

Despite all this, too manly candidates do not have a dependable metaprocess for writing working code - a common scenario is they write code, it gives the wrong result, they drop some print statements to debug, but can’t effectively trace the program to figure out why they’re getting the debug output and so they guess and try to fix the wrong thing, causing further problems. I know that some of this is nerves but it can’t account for all of it.

> I know that some of this is nerves but it can’t account for all of it.

Why not?

If you've never experienced severe situational anxiety, you may not know how debilitating it can be.

And then, maddeningly, perhaps just minutes after the interview is over, you return to your "normal" self, and start to think about what you would have done in that situation if you weren't nervous.

It's basically stage fright, which can be terrifying if you're not used to the stage.

I guess I'm generally going off of an interviewee's body language, which isn't entirely fair since I'm not a mind reader. And I'm sure I'm missing some cases of stage fright, which is why again I mentioned all the techniques above that I try to mitigate this, but I agree is a really unfortunate side-effect of the interview process that I don't have a fool-proof solution for.

Still, I do have some EQ. A lot of candidates exhibit visible anxiety and then get into a rhythm and calm down when they start coding. Or they visibly calm down once I start asking them questions about how they're thinking and coax them towards viable implementation. On the flip side, a lot of candidates actually exhibit overconfidence running in the wrong direction, which I try to lightly mitigate but can't prevent in all cases without giving an unfair advantage.

Are you sure? I mean the 10 years of success spent over various companies projects on resumes have a high false rate because after they took your coding test they didn't do as well as you thought they should? Did you hire any of them and did they out perform your initial assessment.

If you didn't hire them and compare them to new hires who have bad resumes but did well on your test how would you know?

You can't objectivity judge a candidate by your test results only. Otherwise your test is just randomly filtering people. Until you do that you don't know how effective your test is. You could be flipping a coin and getting the same results.

I’ve hired people who have not passed the coding screen (but did pass subsequent interviews) and they’ve been some of my best hires. But the type of failure really matters. Did they write generally correct code but ran out of time because the algorithm they came up with under time pressure was too complicated? Or did they fail to debug a list indexing off-by-one error, even after printing the list repeatedly? It’s sometimes a judgement call but I believe there’s a lot of signal in watching people write code in real-time

How do you "quickly go through" a stack of 100 resumes and choose a handful of top candidates in a way that is not subject to personal (conscious or unconscious) bias?

I remember an interview training exercise, which also covered resume screening. Each person got the same stack of resumes, and the instructor said, "You have 60 seconds to pick out the top 3 you'd like to bring in for an interview. Find them." At the end of the exercise we found that we all basically picked out people who were a lot like us. White people picked out "whiter-sounding" names, women picked more women, etc. Also, Ken Thompson's resume was one of them (with a different name) and nobody picked him.

I think the point was that human eyeballs and brains are terrible at objectively or fairly pre-filtering.

> How do you "quickly go through" a stack of 100 resumes

Well, by quickly I didn't mean only 60 seconds. :-)

> in a way that is not subject to personal (conscious or unconscious) bias?

In my opinion, a common conceit is that it's possible to design a process that is not subject to bias. It's not possible. There will be bias, no matter what process you choose. It's thought that "quiz show" style interviews are "objective", but if you look at the results, they're clearly not helping. They're not increasing diversity. The poor results speak for themselves. So then some people look at the results of this so-called "fair" process and conclude that minorities are just worse at coding, because that's what the tests say. To me, that approach and attitude is even worse than just acknowledging personal bias.

The only way to try to counteract unconscious bias is to be conscious of it. Admit it, and keep it in mind while you're going through your hiring process. Keep it in mind while you're filtering the résumés. Keep it in mind while you're interviewing. But don't try to pretend like you can eliminate it entirely from the process, because then you're just deluding yourself and everyone else.

Agree with this for most companies. When at an airport we were government and had to be 'fair' it was about ten solid days of work to hire for one position.

> Don't do "screening" interviews. Only do "hiring" interviews.

Good way to put it. Exactly.

This only makes sense if they want top talent. Usually they want cheap/easy to push around talent and for that you need a screener that filters out the best and whoever remains will have the strength to deal with bully managers or lack of requirements