Monday, September 26, 2016

AMS/ASL Meetings, 1995-2016

Today I'm going to try to answer the same question for the AMS/ASL meetings that I answered on Saturday for the Annual Meetings: how has the representation of women among speakers changed over time? I'm considering the same factors:
  • Time span: The AMS/ASL meeting has only been an annual occurrence since 1995, so I'll take that as my starting year.
  • Types of talks: The only talks I know of at AMS/ASL meetings since 1995 are plenaries, so that's easy! Again, I'm including speakers who were invited to give a talk, accepted the invitation, and then couldn't attend since the goal is to study who is invited, not who is invited and not prevented from attending.
  • Numbers v. proportions: Once again, I'm going to argue for studying the proportion rather than the number. The 2 female speakers out of 10 in 2004 and the 2 female speakers out of 6 in 2009 don't demonstrate the same level of representation.
Here's my first stab at a model: a simple linear regression. Take a look at the scatterplot of the proportions of female speakers at a meeting versus the year of the meeting with the regression line added.
Note that I'm representing this one with a graph where the axis for proportions goes from 0 to 1 instead of 0 to 0.5 like I did last time: I want to remind everyone that the proportions never get close to their possible maximum! The equation of the regression line is
proportion = 0.004814(year)-9.501405.
I'm not going to mention any predictions one can make from this equation: the R2 values are dismally low. In fact, the adjusted R2 is negative (-0.002544). When I looked at the residuals, the first two flagged years were 1995 and 1996—the first two years I considered, in which 28.6% of the speakers were women (no other year had a percentage that high until 2007). The other two were 2012 and 2013. 2012 was the year with the highest-ever level of representation (42.9%), and there were no female speakers in 2013. These percentages occurring in consecutive years does not make a good linear model likely!

Looking at a LOWESS (locally weighted scatterplot smoothing) plot helps a lot in making sense of this data: representation was not horrible in the first two years, it tapered off to almost nothing for about a decade, and since then, there have been some better years and some very bad ones.

Summary: I'm not going to comment on the linear regression because that doesn't seem to be a reasonable model at all. Representation of women fell and then stayed extremely low for about a decade before beginning to increase again, but even the improvement since 2007 isn't stable: in six of those years, representation has been at least 28.6%, but it has dipped down to 0% in three rather regularly-spaced years.

Next up: the APA/ASL meetings!

2 comments:

  1. Well,well, interesting indeed. Mathematicians by and large, in my experience, don't take logicians seriously. Logicians don't take women seriously, so it's a double whammy.

    ReplyDelete