I wish we had the actual paper, because the abstract and summary are odd:
We studied the effect of such stereotypes in an experimental market, where subjects were hired to perform an arithmetic task that, on average, both genders perform equally well.
If this is actually what they did, then it's a bit shady. Most of the data shows that men and women are the same on average, but men have a higher variance. So if managers are hiring the best, you'll get more men. Of course, the paper is paywalled, so who knows?
It is nice to know that objective measurements can reduce bias. Score 1 for hiring by github. I'd be curious to see what happens with incentives taken into account - make the employees do math problems and the manager gets $1 for every correct answer.
In this experiment, the employers picked between two candidates, one male and one female. If you pick the male on the basis that men tend to have a higher variance of performance, you will end up with a significant underperformance as much as you end up with a significant overperformance. In any case, the task was not to achieve the highest absolute performance - it was to choose the better of two candidates.
In other words, although the variance explanation might give a rational reason for hiring men over women in the real world (although I am very skeptical) it certainly doesn't give a rational reason for hiring men over women in this experiment. Your criticism does not apply.
Actually this paper is very interesting. So they did in fact reward employers for employee performance.
What I found the most interesting:
If, instead, we were to impose a random choice on employers, their earnings would drop by
11.4%, because employers do gain some relevant information from the appearance of the candidates, and this information allows them to make better-than-random choices (as can be seen in Fig. 1, which shows that employers in this condition choose the higher-performing candidate 55% of the time).
If we remove the anti-women bias in expectations [in the case of appearance only], employers would earn only 0.1% more in compensation.
So it turns out that anti-female bias is almost negligible when basing decisions entirely on candidate appearance. That's really surprising to me. I wish they included a table describing how much alpha could be gained by eliminating anti-female bias in all cases.
Also, it looks like the best (and least biased) predictor was Past Performance. Surprise surprise. Maybe people will now stop complaining about employers who ask for a github (aka "Past Performance") instead of a resume ("aka "Cheap Talk")?
We studied the effect of such stereotypes in an experimental market, where subjects were hired to perform an arithmetic task that, on average, both genders perform equally well.
If this is actually what they did, then it's a bit shady. Most of the data shows that men and women are the same on average, but men have a higher variance. So if managers are hiring the best, you'll get more men. Of course, the paper is paywalled, so who knows?
It is nice to know that objective measurements can reduce bias. Score 1 for hiring by github. I'd be curious to see what happens with incentives taken into account - make the employees do math problems and the manager gets $1 for every correct answer.