Inside Higher Ed has a news squib about gender disparities in academic science, which points to a Nature story about a survey on job satisfaction (bad IHE, giving a false impression on the story!). The gender portion of the story is limited to a short section at the end of the article, and one graph:
The larger story is actually pretty positive, but I fear that as IHE did, too much attention will be focused on this one graph (which, by the way, is surprisingly badly done). There’s a fairly narrow point about the presentation of this that I think is worth making (even though it will likely be misinterpreted in an annoying way).
So, looking at this graph, it shows that, in North America at least, the salaries of men and women in science (not necessarily academia– the data seem to include both academic and industry positions) begin to diverge at 6-10 years from hiring. So, does this mean that a woman hired today can expect to be making significantly less than her male counterparts six years from now? Not necessarily. In fact, this is exactly the sort of graph you would expect to see if conditions were steadily improving for women in science.
I’m going to pull out just the North American data and present it in a slightly different way (data values estimated by looking at the graph on my computer screen, as Nature chose to give this the glossy newsmagazine treatment rather than presenting it like a real survey):
If you look at the 6-10 year data, you see a $5,000 (estimated) gap between men and women, and a $10,000 gap in the 11-15 year data. What does that really mean?
The important thing to remember here is that this kind of dataset is a lot like what you get in geology or astronomy: it’s a slice not only through salary space, but through time. That $5,000 gap reflects not only current conditions, but everything that has happened in the last 6-10 years.
You could get a $5,000 gap from a current and consistent bias toward men in promotion and pay increases. Or it could be the result of a significant gap in starting salaries for men and women 6-10 years ago, followed by consistent and equitable promotion and salary increases. Or some combination of the two. Or it might reflect something else entirely, like a greater tendency for women to take time off to have children in the first 6-10 years of their careers.
In fact, this sort of graph is exactly what you would expect to see if conditions for women have improved significantly over the last ten years. A large amount gender bias in the past giving way to a vastly smaller amount of bias in the present would produce exactly this effect: Men hired ten years ago would have benefited from salary bias back then, and ended up with significantly higher salaries, while men hired today would have little or no advantage.
How could you sort that out? Well, you could take these data, and compare them to a similar survey five year ago, or five years in the future. If the gap is indicative of some past bias that is significantly reduced, you would expect the gap in salaries to show up earlier in the older data, and later in future data. Or you could look at something like current salary as a fraction of starting salary, which would tell you if one group or the other has received disproportionate raises over the years. Those data are not presented here, though, so there’s no way to tell from this graph what the explanation is.
One thing is certain, though: This is not a reliable projection of the future salary trajectories of men and women in science. That will depend not only on current conditions (which appear to be reasonably equitable, salary-wise– if anything, women in North America would appear to have marginally higher salaries in the first five years, looking at the graph with the article), but also on future conditions, and the tricky thing about the future is that it’s damnably difficult to predict.
(This is very hand-wavy, but the actual article is pretty squishy as well. If they have their data available in any more quantitative format, I don’t see it.)
There’s a well-known problem with longitudinal data like this, where there are potentially influences from all of “age, period and cohort” — here, years since doctorate, the year of observation, and the year the doctorate was achieved. If you allow the average salary (or average salary difference by gender) to depend on all three in potentially arbitrary ways, then the problem becomes perfectly unidentifiable. The only way to get around this is to impose a priori restrictions, say by assuming that age, period and cohort effects are all additive. There is a good discussion in this paper by Hall, Mairesse and Turner, with application to measuring the productivity of physicists as a function of age.
Thank you for at least writing something about women in science!
Of course, this is a long way from an acknowledgement that women still face enormous hurdles in the physical sciences … and there is an enormous body of evidence for this, which a 1 minute google search could find. In that context, this graph is totally stupid, because in certain subjects the number of women barely makes a statistical sample.
I make less money than a girl? gee.