“Arbitrary” is the wrong word for it.
First, it’s clear that the system being used isn’t a “4-point” system but a 3-point system. 2 is negative, 3 is neutral, 4 is positive. 1 appears to be never used. And what we see is not that variance goes up with mean performance but that variance is highest closest to 3. That’s what we’d expect. Likewise, the maximum standard deviation is in a fair (p = 0.5) coin.
I think that technical interviews can be solved through aggregation, up to a point. The drawback is that, after a certain amount of technical grilling, people can tire out. Also, if you mix technical and cultural questions too quickly, you’re putting someone through a lot of context switching. That said, I think that 5 clean technical reads can be aggregated into a reasonable signal. Of course, you need some way for positive deviations to cancel out negative ones. If your bar is 18 (out of 20) on that 2-3-4 system, then you’re saying that 4 positive votes cancel out 1 negative one.
On the other hand, aggregation kills you when it comes to cultural vetting, because if someone has to be liked by 5 people (especially in the modern era when many of us, myself obviously included, have complicated internet presences) then you’re biasing your process in favor of non-dislikability. Obviously, you shouldn’t hire people who are jerks or who are demeaning, but if you ask 5 people “Do you like this person?” (which is not the same as, “Did this person do something objectively deserving of dislike?”) and require 5 “Yes” votes, then you’re going to lose out on your best people. And of course, “no cultural fit” is just a passive-aggressive way of saying that one doesn’t like someone. Over time, that sort of process results in people hiring people who are like themselves, but 97 percent as competent, and of course 0.97^N goes to zero as N grows.
Just scanning the article; 1 is used, exactly 4 times = 1.3% of the interviews graphed. (try expanding the full data viz; you’ll see some points go that way)
You’re correct. It’s then interesting (although probably not indicative of anything, given the small number of relevant data points) that all of the people who received 1’s received at least two 3’s and therefore averaged above 2.0.
The really interesting point is the one with 3 4’s, 4 3’s, and a 1. I wonder if there’s a story, or if it was just stray bad luck.
The charts shown here are the canonical example of colors you should use if you want 10% of the male population to be unable to read them.
Why is that?
~10% of the male population is color blind. For the life of me I would not be able to distinguish the small people icons of the first chart. I know from the legend below the chart that there are probably three different colors but I’ll never be able to see which one is which.
For whatever help it may be, the color on that chart is purely decorative - there’s a gradient from magenta on the left to blue on the right, and the color of each person-icon along the gradient appears to be determined solely by its x position.
It’s a useful thing for graphic designers to keep in mind that decorative features can still be accessibility issues when people have no way of knowing they’re decorative!
Wow, I didn’t expect that at all. I would never have imagined they only were decorative.
It is very interesting indeed, thanks!
I’ve met color bind people, but didn’t know that the likelihood was so high or that it more common in men. Thanks for sharing.
The genes that produce the color receptors are on the X chromosome. You only need one working copy to have color vision, but men only have one of those.
Also, it varies by heritage. 5% of men is a good rule of thumb for nearly all populations, but it can range up to 10% in a few (mostly island) groups.