Most recent update: Aug. 31, 2017. To date, all updates have introduced only tiny quantitative changes to the original data, no substantive changes. A recent update adds an additional analysis (finite population confidence intervals). A second recent update slightly corrects the analysis of R1 vs non-R1 institutions because I originally misclassified one institution.
Recently I decided to quantify the gender balance of recently-hired ecology faculty in North America. “Recent” being operationally defined as “hired in 2015-16, or in a few cases in 2014”. Data on the gender balance of faculty is widely available only at the level of very broadly-defined fields like “biology”. Current faculty gender balance mostly (not entirely) reflects the long-term legacy of past hiring and tenure practice rather than current hiring practice (Shaw and Stanton 2012; ht Shan Kothari via the comments). And nobody’s anecdotal experience informs them about the outcomes of more than a tiny fraction of all ecology searches in any given year. So this seemed like a topic on which many people would welcome some reasonably comprehensive data. Follow the link for more details on how I compiled the data. In that old post, I also conducted a poll asking readers what they expected me to find.
Here are the answers: what fraction of recently-hired North American ecologists are women, and what do ecologists think that fraction is?
Many of you are going to be pleasantly surprised…
I’ll follow a methods-results-discussion format.
I checked over
250 300 ecology positions advertised in 2015-16, where “ecology” was broadly defined to include fields like wildlife ecology, fisheries, conservation, etc. (see the link to the previous post for details). The definition of “ecology” obviously is somewhat fuzzy, but a much narrower definition would’ve shrunk the sample size substantially while eliminating from the dataset many recent hires who self-identify as ecologists. Checking a given position often involved checking every assistant professor in the department. As a consequence, I sometimes stumbled across another ecologist who was hired in 2014 or later, in which case I added that position to the list. I included the occasional person I stumbled across who was hired in 2014 (rather than 2015-16) as a convenient way to increase the sample size a bit. I included positions at all colleges and universities, not just R1 universities. I included only assistant professor positions, not senior-level hires. In the end, I compiled 193 positions for which I could easily determine with high confidence that the position had been filled in 2014 or later by an ecologist (a small number of positions turned out to have been filled by people who weren’t ecologists even under my broad definition). In every case the gender of the person hired was clear from their name and photograph. I recognize that using a gender binary (male/female) isn’t ideal, but it seemed like the most practical choice. The remaining positions are those for which I couldn’t tell who was hired. In some cases perhaps no one was.
Of those 193 recent hires, 105 (54%) were women.
This is a sample, not a census. Assuming, not too unreasonably, that it’s approximately a random sample, the 95% confidence interval ranges from 47-61% women (normal approximation to the binomial distribution, which is fine here because the sample size is large and the estimated proportion of women is close to 50%).
In the comments on the previous post, a couple of you asked if the numbers are different for research universities vs. others. The answer is it doesn’t appear so. 83 of the identifiable positions were at R1 universities or their Canadian rough equivalents. Of those, 47% were filled by women, meaning that women filled 60% of the non-R1 positions. The 95% c.i. are wide: 36-58% women at R1/R1-equivs. and 51-69% women at non-R1s.
Both 95% c.i. include 50% (EDIT Aug. 31, 2017: no longer true after latest update) and the two proportions don’t differ significantly from one another (z=-1.80, p=0.07, two-tailed test). Note that the distinction between R1 institutions and others is somewhat arbitrary. If you were to define “research university” differently (say, as R1 + R2 universities, or not counting Canadian universities), you might get a different answer. Note as well that breaking down the results very finely (say, into all the Carnegie categories) would result in some small sample sizes.
Aug. 13, 2017 update: Upon reflection, the confidence intervals in the previous paragraph are too wide, because they assume sampling from infinite populations. In fact, the population of positions is finite and not all that much larger than the sample. I checked over 300 positions, which is probably the large majority of asst. professor positions in ecology that were filled in N. America in the 2015-16 faculty job season. For instance, if you assume that the population comprised 350 positions, then as of this writing the estimated proportion of 54% women is surrounded by a finite population 95% c.i. of 49-59% (normal approximation). That’s a bit narrower on either end than the confidence interval in the previous paragraph.
I did not break down the results by subfield. Personally, I think the sample sizes are too small and the definitions of subfields too fuzzy and arbitrary to make that a useful exercise, even just for purposes of data exploration rather than hypothesis testing. Plus, I didn’t have any a priori hypotheses and didn’t want to get sucked into data dredging. But if you want to look into that, you can download the data and go to town.
I welcome corrections and additions to the spreadsheet; please email me (email@example.com). I haven’t made the spreadsheet publicly editable, because I want to ensure the accuracy of any edits. I update the post periodically to reflect corrections and additions.
We got 468 respondents, evenly split between men (50%) and women (49%), plus 2% who declined to indicate their gender (note: percentages may not add to 100% due to rounding). Respondents comprised an even mix of grad students (28%), postdocs (31%), and faculty (30%), plus 10% in other occupations such as government scientists. I don’t know the mix of fields they’re in, but based on past reader surveys and the fact that this is an ecology blog that was polling about hiring in ecology, it’s safe to assume the large majority are in ecology or an allied field. Obviously, this isn’t a rigorous random sample from any well-defined population. But I think it’s a sufficiently large and diverse sample of mostly-ecologists to be worth talking about.
The majority of respondents think that recent hiring in ecology is skewed pretty heavily towards men, with the modal guess being 25-39% women. Surprisingly to me, there was a second, much smaller mode of guesses centered on 52-54% women (which as an aside is about where I’d have guessed had I completed my own poll). Guesses ranged widely, though. Every option, from <25% women to >75% women, was chosen by at least three people. Here’s the histogram of all the guesses:
The majority of guesses were off substantially, in three senses. First, over half the guesses were off by at least 10 percentage points, which seems like a fair bit to me in an admittedly vague and subjective sense. Second, and more objectively, only 43% of respondents (those who guessed anything from “43-45% women” to “58-60% women”) provided a guess that overlapped the 95% confidence interval of the correct answer. Third, the poll asked respondents to indicate their subjective confidence in their predictions. Specifically, I asked respondents to indicate what size of miss would surprise them (“I’d be surprised if the true value was X or more percentage points outside my chosen range”, where the choices for X were 1, 3, 5, 7, 9 or “other (describe)”). By their own admission, over 60% of respondents will find the result surprising.
I used chi-squared tests to check for associations between variables. There was no significant association between respondents’ genders and their predictions (p=0.94). There was no significant association between respondents’ positions (faculty, grad student, postdoc, other) and their predictions (p=0.13). There was no significant association between respondents’ predictions, and their subjective confidence in their predictions (p=0.85).
- I think these data are very good news. I think they represent progress. You surely wouldn’t have gotten the same answer if you’d collected the same data a few decades ago (Shaw & Stanton 2012).
- I hope that this good news will be widely shared. Ecologists as a group substantially underestimate the proportion of women among recent ecology hires. I totally understand why, but that seems very unfortunate to me. For instance, it potentially is quite discouraging to women looking to go into academic ecology to believe or be told, incorrectly, that women only make up a small minority of recent hires in ecology (not that anyone should make career decisions solely or even mainly on the basis of data like these, obviously!)
- As I hope is obvious, aggregate data like these don’t tell you anything about whether any particular search was conducted fairly, or whether any particular applicant for any particular search was evaluated fairly. They certainly don’t show that nobody ever experiences any gender bias during hiring. And I certainly don’t think they’re an argument for complacency about gender bias. Presumably, recent hiring is gender-balanced at least in part because the issue is taken seriously by many individuals and institutions. We shouldn’t stop taking it seriously.
- As I hope is obvious, these data don’t tell you anything about what’s going on at any other career stage (tenure, promotion, etc.), or in any other context in which ecologists and their work are evaluated (e.g., grant and fellowship applications).
- As I hope is obvious, these data don’t tell you what individual ecologists experience in their day-to-day professional or personal lives, and they don’t substitute for or devalue discussions of those experiences. Conversely, I don’t think individual experiences substitute for aggregate data like these. I think both have important and complementary roles to play in discussion about bias and what to do about it.
- As I hope is obvious, these data tell you very little about your own personal chances of obtaining a faculty position in ecology, since your own personal chances depend on many other factors besides your gender.
- I lack other data that might aid further interpretation of these data. In particular, I don’t know the gender mix and qualifications of the applicants for any of the positions. And I don’t know the gender mix of all ecology postdocs. Just based on what I know anecdotally about the gender mix of ecology graduate students and postdocs, I’d guess that most faculty applicant pools in ecology are roughly gender-balanced. But I’m speculating, obviously, and I wouldn’t put much credence in my own speculations.
- These data are perhaps somewhat surprising in light of what’s known about the “leaky pipeline” at the level of broader fields like biology. Shaw and Stanton (2012) quantified the effect of “demographic inertia” on the gender balance at different career stages in broadly-defined scientific fields. Basically, what would you expect the gender balance of biologists at any given career stage (grad student, postdoc, asst. prof, tenured prof) to be in 2006 (the most recent year of data they examined), given the “initial” gender balance back in the 1970s and changes over time in the gender balance of biology undergraduates, assuming no gender bias in individual probabilities of transitioning from each career stage to the next? That’s the “demographic inertia” baseline (really “demographic inertia plus changing undergrad enrollments”), to which you can compare the observed data in order to discover and quantify gender-biased transition probabilities. They find that demographic inertia has a big effect on current gender balance, but at all career stages there are fewer women in biology than expected under demographic inertia. The difference is largest at the transitions from grad student to postdoc and postdoc to assistant professor. That is, they find that those transitions are the biggest “leaks” in the “leaky pipeline” in biology, with the leaks at other career transitions being small or non-existent (indeed, they find a modest “reverse leak” at the tenure stage). Reviewing the broader literature, Shaw and Stanton (2012) refer to the transition from grad school to faculty as the most difficult transition for women as compared to men. In light of that, it’s perhaps somewhat surprising that recent faculty hiring in ecology is gender-balanced. But as noted in the previous bullet, I lack the data to compare the gender mix of recent hires in ecology to the mix that might’ve been expected given the gender mix of postdocs in ecology. Further, ecology is just a small subset of all of biology. So any comparison between my data and the results of Shaw & Stanton (2012) necessarily is tentative and based in part on speculation. Finally, Shaw & Stanton’s estimates of demographic inertia and deviations from it only go up to the early oughts. Extrapolating the long-term trends in their data forward by a decade would shrink the size of the “leak” during the transitions from grad student to postdoc to assistant professor, and reduce or eliminate the tentative potential contrast between their data and mine.
- I’ve been thinking a bit about what (if anything) these data suggest for blinding the initial stage of job searches. On the one hand, if current hiring procedures are working well, at least at the aggregate level and at least as measured by this one metric, does that reduce the motivation for trying different procedures? On the other hand, there are other arguments for blinding based on other considerations. Interested to hear what others think; I have no firm opinion. (aside: let me repeat what I said in that old post: I think experiments with blinding the initial stage of job searches are worthwhile.)
- The bimodality in ecologists’ predictions surprises and puzzles me. I expected a single mode, while being unsure where it would be centered. Further, the bimodality isn’t associated with any obvious attribute of the respondents. It’s not that we had, say, a small mode of faculty predictions centered on the correct answer, and a larger mode of grad student and postdoc predictions centered on an incorrect answer. And if everyone was guessing based on their own anecdotal experiences of the job market, why would respondents’ own experiences be bimodally distributed, with the larger mode being centered well below the correct answer and the smaller mode being centered on the correct answer? Any ideas why the guesses were distributed as they were? (Aside: No criticism of anyone’s guess is intended or implied.)
- For me, the contrast between the poll results and the data illustrates the value of caution about extrapolating from one’s own anecdotal impressions. My own guess as to the proportion of women among recent faculty hires in ecology would’ve been almost spot-on, and I was fairly confident in it. But in retrospect I think that was just lucky and I shouldn’t have been at all confident in my guess as to how the data would turn out. After all, I didn’t really have much to go on. Just knowledge of the outcomes of a few recent hires, plus various highly-indirect and quite-possibly-misleading lines of evidence. Knowing some people who’ve served on search committees and how seriously they take fairness. Walking around the ESA meeting and feeling like there are more women than there used to be. Knowing that the gender balance of faculty in most STEM fields has been slowly improving. Seeing lots of discussions on blogs and Twitter of the importance of taking gender bias seriously, giving me the vague impression that lots of people take the issue seriously these days…In retrospect, I really appreciate the several respondents who wrote in that they wouldn’t be surprised if their guesses were off by >15-20 percentage points; I wish I’d taken the same attitude.
- I said earlier that it’s not unreasonable to think of the data I compiled as an approximately random sample from the population of all recent North American hires in ecology. But there are ways in which my sample could be a bit biased. In particular, bigger and more research-intensive colleges and universities tend to have websites that are more informative about their faculty, and faculty employed by bigger and more research-intensive colleges and universities are more likely to have personal websites and/or Google Scholar profiles. For this reason, positions at larger and more research-intensive places were more likely to be included in my compilation. It was harder to identify who (if anyone) was hired at smaller and less research-intensive places. So if hiring at smaller and less research-intensive places skews a bit more female than hiring at larger and more research-intensive places (and as noted in the results, there’s a non-significant trend in that direction), then my compilation may somewhat underestimate the overall proportion of women among recently-hired ecologists in North America.
As always, very much looking forward to your comments.