EarlyToBed posted an interesting list breaking down the sex ratio for seminars at a number of Geosciences departments at Universities in the US today. This feeds into an ongoing dialogue about the representation of women in the sciences, and their lack of representation in higher profile sessions at meetings (see here for a great take by Jonathan Eisen, who includes a whole bunch of links at the end). When she tweeted the post I replied wondering how that relationship matches the departmental sex ratio. I posted that with the belief that higher ratios of women in a department would mean a higher ratio of seminars presented by women. To wit:
- H0: There is no relationship between the number of women in a department and the number of seminars presented by women in the departmental seminar series
- H1: Higher numbers of women in a department will mean more seminars presented by women.
Why do I believe that? I made a few assumptions to get to this idea. I assume that women in the geosciences likely have a higher number of women in their scientific social networks, and that departmental seminars often draw on members of the department to fill vacant spaces, so a department with more women should, presumably, have a higher chance of drawing women from the department to fill empty spaces.
So, what does the relationship look like? Not what I expected at all. . .
I took the list of seminars and their sex ratios posted by EarlyToBed and went to each department’s website, using the ‘People > Faculty’ page to assess the departmental sex ratio. For most departments this was fairly straightforward. I didn’t get a ratio for Lamont-Doherty since there is no single Faculty page, and faculty/researchers are represented on multiple sub-domain pages, meaning I could inadvertently double count individuals as I was going through the pages. I could have been more thorough, but this was supposed to be a simple exercise. A couple potential sources of noise:
- Some pages list lecturers, professors emeritus, and research associates, some don’t. If there is a sex bias in any of these groups this may affect the results.
- Some pages may not be completely updated. If new faculty are more likely to be included in the most recent seminars and those new faculty have a sex bias then this might also affect the results.
I fit the results using the bayesglm function for R with a binomial family (and a probit link).
So, it appears that there is in fact a negative trend, with departments having higher sex ratios showing lower sex ratios for the seminar series! Northwestern University in particular has the highest departmental ratio, but the lowest seminar sex ratio. The opposite end of this trend is anchored by UCLA, Arizona State, Princeton and Harvard, who all have < 15% women as faculty, but greater than (or equal to) 25% of women in their seminar series.
The model is significant with the bayesglm function, but not with the standard GLM in R, which is likely entirely due to Yale’s excellent behavior! If Yale is excluded both the bayesglm and glm fits are significant, if it is included then only the bayesglm fit is significant. The same cannot exactly be said for Northwestern University. Dropping Northwestern still produces a non-significant fit for the GLM, but there is still marginal support for the bayesglm fit. However, given the apparent sensitivity it would be nice to add a few more universities.
So let’s pretend for a minute that the results are not just a statistical anomaly due to dataset size. Why are universities with more women doing a worse job at including women in their seminar series? Can we invoke the Queen Bee hypothesis, meaning women are preferentially choosing men at a higher rate than their male counterparts? We know that both male and female faculty tend to rate female competence lower than male competence, even when application materials were identical. Is this bias also playing out in choices for seminar series?
A further complication is the fact that seminar series are not homogenous in terms of their aims and speaker pools. Speakers may be largely invited, or they may be made up of grad students and post-docs or faculty. Regardless, EarlyToBed’s results were really interesting, but they seem to get more interesting as we add the dimension of faculty sex ratios to the mix.
So how do we fix it? At this point a conscious effort needs to be made to explicitly include women across the board, there is evidence that this gender bias extends beyond seminars to all facets of academic pursuit, including hiring and grants. Even if we are not overtly sexist as a culture (although that is probably debatable), we still exhibit behavior and attitudes that result in the exclusion of women, in the absence of any sort of factual basis for exclusion, as the PNAS study seems to show. If we fail to make a conscious effort, then we fail to address the unconscious bias that we as a community continue to exhibit.
More interesting is why those universities that have higher numbers of women are inviting lower numbers of women to their seminars. Perhaps departments with higher proportions of males are making more conscious efforts at including women, and those with higher proportions of women are operating on the assumption that they are already creating diverse research environments, and so there is less of a conscious effort at inclusion. Either way, it would be nice to see how more data will inform the analysis.
If anyone has any thoughts or ideas, please let me know, either by emailing me or by leaving comments in the comments below. Also, if you think I’m doing it all wrong I’d appreciate a heads up!
For the interested, here is the raw data for analysis. The code is below:
library(arm) library(ggplot2) lectures <- read.csv('LectureRatio.csv') weights <- rowSums(lectures[,5:6], na.rm=TRUE) females <- as.matrix(lectures[,2:3]) model <- bayesglm(females ~ F.r, data=lectures, family=binomial(link='probit')) model.out = predict(model, type='response', se.fit = TRUE) mr <- data.frame(ymin = model.out[] - model.out[], ymax = model.out[] + model.out[]) qplot(data=lectures, x = F.r, y = S.r) + xlab('Faculty Sex Ratio') + ylab('Seminar Sex Ratio') + geom_ribbon(aes(ymin=mr$ymin, ymax=mr$ymax, alpha=0.3)) + geom_line(aes(x = F.r, y=model.out[])) + theme(legend.position='none')