Fermi Pipeline Problems

There was some Twitter chatter the other night about a new arxiv paper called The Gender Breakdown of the Applicant Pool for Tenure-Track Faculty Positions at a Sample of North American Research Astronomy Programs:

The demographics of the field of Astronomy, and the gender balance in particular, is an important active area of investigation. A piece of information missing from the discussion is the gender breakdown of the applicant pool for faculty positions. For a sample of 35 tenure-track faculty positions at 25 research universities advertised over the last few years in astronomy and astrophysics, I find that the ratio of female applicants to the total number of applicants is ~0.2, with little dispersion and with no strong dependence on the total number of applicants. Some discussion is provided in the context of the fraction of women at the graduate student, postdoctoral researcher, and assistant professor levels, but strong conclusions are not possible given the limitations of the study. Current and future faculty search committees will likely be interested to compare their numbers to this distribution to decide whether or not they could be doing more to attract an applicant pool that is representative of the community.

“Representative of the community” here meaning “closer to the distribution observed at lower levels,” citing a demographic study from the AAS (pdf here) that found about 35% of astronomy grad students are women, and 29% of post-docs, and 30% of assistant professors. So, for women to be only 20% of the applicant pool for faculty jobs seems like a significant anomaly.

Of course, anybody who has done any academic hiring knows that the instant you post a job, you get a bunch of ridiculously inappropriate applications from people with the wrong degree, irrelevant experience, no references, etc., who just apply to every open position. Those are almost exclusively male, so my first guess as to what might explain this is that those guys are skewing the demographics– that the distribution of appropriate and qualified applicants is close to the post-doc distribution, but once you add in the folks who have no business applying, things look a lot worse.

But, you know, we’re physicists here, so we don’t have to just guess– we can do a back-of-the-envelope estimate to check whether that’s a reasonable explanation. So, let’s say that we have a job pool of 100 qualified candidates, 30 women and 70 men; how many unqualified jokers do we need to add to drive the percentage down to the observed level of the real pool?

Well, this can be rephrased as “30 is 20% of what?”, which you can see immediately is 150. So we would need to dump 50 unqualified men into the applicant pool for this to completely explain the effect.

Is that reasonable? Enh. Coincidentally, we read right around 150 folders for our recently completed job search, and while I don’t have solid stats for this, I would put the “no business applying” fraction a bit lower than that– probably 20-30. It’s probably a significant piece of the puzzle, but not the whole answer.

Of course, the other interesting thing about these numbers is the demographic rebound– that is, the way the percentage drops from 30% of the postdoc pool to 20% of the faculty applicants, then back up to 30% for assistant professors. If you take this as referring only to applicants meeting the formal qualifications, that would suggest that women have a rather significant advantage in terms of getting hired. Doing the same back-of-the-envelope sort of thing, with an imaginary pool of 100 applicants with 20 of them women (to match the observed fraction in the applicant pool), applying for 10 assistant professor jobs, ending up with 30% of the jobs going to women would mean that 15% of the women applying got hired compared to only 10% of the men.

And as with most things, there’s a way to explain this that makes it seem reasonable, namely that the women who didn’t apply would’ve been marginal candidates, while the marginal men applied anyway, because men are arrogant/overconfident that way. So the women in the pool were, on average, more qualified than the men, and thus deserved to be hired at a much higher rate.

Which, you know, is a thing you can do. Of course, if you want that to be the entire explanation, it would suggest that the “real” pool of candidates should be about 66 rather than 100– the 20 women who did apply, plus 46 well-qualified men. Which means that 34 of the men who did apply (about 43%) weren’t really viable candidates.

That also seems kind of high, but a combination of the two things is fairly plausible– some women with the right formal qualifications don’t apply, but the women who do apply tend to be somewhat better. And then there are a bunch of applicants who don’t have the right formal qualifications, who skew heavily male and thus make the demographics look worse than they should.

Of course, for both of these effects, the answer to the question raised in the abstract, namely how to “attract an applicant pool that is representative of the community” is “get fewer unqualified men to apply.” Which is a little counterintuitive, maybe, and also probably impossible, unqualified men being famously stubborn.

So, anyway, just in case you were thinking “Gosh, it’s been a while since we had an Uncertain Principles post noodling around about demographics of academic hiring…,” there you go. Also, if you were thinking that, you might want to take a minute to re-examine your priorities, because, really…