Saturday, September 24, 2016

ASL Annual Meetings, 1989-2016

The question I've been asked most since my first post is whether women have become better represented among ASL speakers over time. I'll try to answer it today for the Annual Meetings. Here are the factors I'm taking into account:
  • Time span: 1989 begins the era in which the Annual Meetings occur independently instead of in conjunction with either the APA or AMS, so I'm starting then.
  • Types of talks: In that time span, the Annual Meetings have had plenaries, tutorials, Gödel lectures, retiring Presidential addresses, symposia, and panels. I'm restricting this analysis to plenaries because those occur every year (unlike tutorials, symposia, and panels) and because there is more than one of them each year (unlike Gödel lectures and retiring Presidential addresses). I'm also leaving off speakers in special sessions for now: I'd like to analyze speakers invited by a "universal" program committee separately from speakers invited by other members of their subfield.

    There are a few instances of speakers who were unable to deliver their talks. Since the goal is to study who is invited, I'm including all speakers who accepted an invitation. 

  • Numbers v. proportions: It makes the most sense to me to consider proportions. In 1994, 2 women and 11 men spoke, and in 2015, 2 women and 5 men spoke. Although the number of female speakers is the same, women were certainly not represented equally well.
My first step was to construct a basic regression. Here's a scatterplot of the proportions of female speakers at a meeting versus the year of the meeting complete with regression line.
The equation of the regression line is
proportion = .009357(year) - 18.617671.
This predicts that -0.7% of the speakers in 1989 and that 24.6% in 2016 were female: low on both counts, but the model isn't an obvious bad fit for the data. So how good is it?

The R2 isn't too shabby but isn't great either: it's about 0.40. When I examined the residuals, they mostly supported a linear regression, but a few years were flagged as unusual: 2016, the only year in which we've had parity, 2009, the year with the second-highest proportion (33.33%), and 1999, the only year before 2009 in which the percentage of female speakers went over 20% (25%). In short, the unusual years are the years with a relatively high proportion of female speakers.

My next step was to apply a locally weighted scatterplot smoothing (LOWESS). Here's the result:

It seems like there are two different eras: 1989–1998, when the percentage of female speakers was often 0 and below 20% even when it was positive, and 1999–2016, when the percentage of female speakers was generally at least 10% and ranged up to 50%. I have no idea what brought about the change—Richard Zach informs me that the ASL statement on women in logic wasn't adopted until 2012.

Summary: The basic linear regression tells us that the proportion of female speakers is increasing by just under 1% each year. The LOWESS model suggests it's increasing faster than that now, but the trouble with LOWESS is that it lets you see trends more clearly but doesn't let you quantify them easily. It looks to me like there are a few very good years with relatively high proportions of women improving the predictions while the rest have much lower proportions (and very consistent lower proportions, too—look at those flat sections in the scatterplot!).

Next up: the AMS/ASL meetings! Let me know what other analyses you'd like to see.

No comments:

Post a Comment