Monday, May 4, 2015

Gender Representation at a Specialized Astronomy Conference


Today's guest blogger is Kyle Willett. Kyle is a postdoc in the School of Physics and Astronomy at the University of Minnesota. He works on galaxy morphology and their relation to active black holes, particularly as part of the Galaxy Zoo project.

Members of the astronomy community have recently been tracking participation at conferences as a function of gender. This is intended to address some basic questions about behavior at conferences, such as:
  • How equal are the allotments of talks among men and women?
  • Are men and women asking questions of speakers at the same rate?
  • Does it matter if the speaker/session chair is a man or a woman?
  • Are women/men more likely to ask the first question in a session? Does this affect the gender balance of remaining questions?
In the broader sense, this is intended to measure if our collective behavior at conferences — in choosing speakers, engaging with them, and interacting with the audience — is as inclusive as it should be. If not, then hopefully the community can come up with some ideas of best practices to follow.

Data-gathering efforts were started by James Davenport at the 223rd American Astronomical Society (AAS) meeting in 2014, and repeated at the 225th AAS meeting in 2015 and the National Astronomy Meeting (NAM) in the UK in 2014. The data collected focus on question/answer sessions following oral presentations, since it's public, quantifiable, and one of the main methods of professional interaction. They found that men ask disproportionally more questions than women, and that the gender of the session chair has a strong impact on the gender ratio of questioners. However, there aren't any data (yet) on whether smaller, more specialized meetings follow the same trends as AAS and NAM, which are much larger and thematically broader.

I tracked this for a recent conference on "Unveiling the AGN-Galaxy Evolution Connection", held in Puerto Varas, Chile from 9-13 March 2015. A one-person team (me) kept track of the gender of speakers, chairs, and questioners for all of the 72 talks given at the conference (hereafter referred to as AGN2015).

The full data and analysis are openly available as a Github repo and iPython notebook. My short version of the conclusions:
  • both the attendees and questioners at AGN2015 had significantly higher female/male ratios than the NAM or AAS meetings
  • the male/female ratio was similar for both attendees and speakers
  • both male and female speakers received about the same number of questions per talk
  • the gender of the first person to ask a question didn't affect subsequent questions
  • the gender of the session chair didn't affect subsequent questions
I'm not sure what my expectations were in looking at data for a smaller, specialized meeting vs. a large general meeting like the AAS or NAM. I might have guessed that a smaller, specialized meeting would more subject to the whims of an SOC (who could be more likely to be senior and thus might tend to invite their peers, which would presumably skew the gender ratio). On the other hand, postdocs and early-career researchers might travel more (fewer teaching responsibilities mid-semester) and be more invested in giving talks to establish themselves for permanent jobs, which would skew toward younger people and a higher female-male ratio. The demographics of the AGN/galaxy community, the relatively remote conference location in South America, and the timing of the meeting are all priors that should probably be taken into account.

The main negative was that the session chairs were overwhelmingly male (a possible reflection of a seniority bias). However, this is presumably simpler to correct for an SOC/LOC, is subject to small sample size variance, and doesn't appear to have affected the engagement with the audience, which is good. I also want to note for AGN2015, all the three end-of-day discussions were moderated by women, as was the conference summary. My unofficial impression is that this conference had a moderately younger-than-average age, especially among speakers. The targeted talks at this conference were unofficially designated as being reserved for "young people doing cool things" by the SOC. I'd love to see more of that in every conference, since I think it'd help with both scientific output and participant balance.

One extremely important point is that this is a sample size of N=1. We need similar data on other conferences, in different subfields, different parts of the world, and different conference structures. Unlike at the large meetings, data collection can pretty easily be done by one person, and I want to develop some standard plots and questions we can ask through open data sets (all of this data/analysis is available online). I hope other astronomers are willing to track this at future conferences; if so, please feel free to clone this dataset and analysis tools we can keep tracking this. I'm happy to help anyone who's interested in taking part.

So maybe smaller conferences are better at gender balance and engagement than larger meetings? Maybe — but we need lots more data to see if this effect is real, and to help guide what the astro community can do in the future.

Thanks to Ezequiel Triester and the conference organizers of AGN2015 for having me at the meeting, to Meg Urry for presenting some of these results in her conference summary, and to James Davenport, Jonathan Pritchard, and Karen Masters for their previous hard work at the AAS and NAM meetings. -KW

Data


What are the overall demographics of the conference?

I tracked gender participation for five aspects of the question/answer sessions:

  • speakers
  • chairs of the sessions
  • all attendees of the conference
  • people who asked questions of the speaker
  • people who asked the first question of the speaker for any given talk
Every one of these categories had more male than female participants. The gender ratio of the speakers closely matched that of the attendees as a whole (45% and 43%, respectively).

Women were slightly less likely to ask questions than men; women constituted 43% of the attendees, but only 35% of the people asking questions.

The gender of people who asked the first question in a given session is close to that of the overall gender of questioners (38% vs 35%).

Overall demographics for AGN2015 compared to other conferences

                       M         F
First question  0.614286  0.385714
Attendees       0.564246  0.435754
Questioners     0.641509  0.358491
Chairs          0.785714  0.214286
Speakers        0.549296  0.450704

Analysis of the demographics compared to other meetings/organizations:

The data from the 225th AAS meeting did not track the gender of either the session chair or the total number of attendees. The AGN2015 meeting had a higher proportion of women for both the speakers (45% to 38%) and the people asking questions (35% to 19%).

The NAM 2014 meeting tracked data in an identical format to AGN2015. The AGN2015 meeting had a higher percentage of women for speakers, attendees, and question askers. In particular, the gender distribution of people asking questions is twice as high for AGN2015 (35%) as for NAM (17%). This is driven by the large disparity in gender distribution between attendees and questioners at NAM. The gender of the session chairs is the only category in which NAM has a higher percentage of females than AGN2015. However, there were only 14 sessions at AGN2015, compared to 64 sessions at NAM. Some (although likely not all) of the high fraction of males at AGN2015 may be due to the smaller sample size.

Finally, I looked at the gender representation compared to the most recent data for members of the International Astronomical Union (IAU). The IAU is 85% male/15% female, a distribution significantly in excess of every demographic category measured for AGN2015. Membership in the IAU is likely not indicative of the astronomical community, especially since the requirements involve nominations by national professional organizations and thus skew toward more senior astronomers (which are more likely to be male).

How many questions per talk were there? Was this affected by the gender of the speaker?

There are 3.0 +- 1.4 total questions per talk
There are 2.8 questions per talk when the speaker is male
There are 3.3 questions per talk when the speaker is female
There is a 41.2 percent chance that the questions are drawn from
the same distribution for male and female speakers.
Engagement with the speakers was nicely high at the AGN2015 conference, with an average of 3 questions per talk being asked. This was higher than the averages for both the NAM (2.2±1.1) and AAS (2.8±1.5) meetings. The number of questions per talk was slightly higher when the speaker was female instead of male, with 3.3 compared to 2.8. However, a two-sided Kolmogorov-Smirnov test gives a 41% chance that the two samples are drawn from the same distribution, with no indication of a statistically significant difference.

A difference in the number of questions per talk depending on the speaker might be interpreted as either a positive or negative effect (preferentially ignoring speakers of certain genders, overaggressively questioning/harassing them, paying attention and engaging with them, etc). More analysis on this data would be pure speculation from me as to which might be responsible.

Did the gender of people of asking questions depend on the speaker's gender?

M         F
Male asks 1st question    0.694444  0.305556
Female asks 1st question  0.586538  0.413462
When women ask the first question in a session, women ask on average 40% of the total number of questions in such sessions. However, when men ask the first question in a session, women only ask 31% of the total questions in the session.

But this is clearly affected by the fact that the gender of the first question is fixed. To isolate that effect, let's look at the remaining questions in the session.
M         F
Male asks 1st question    0.666667  0.333333
Female asks 1st question  0.637931  0.362069
z = 0.354

When the first question is eliminated from the data, the fraction of questions asked by females are nearly identical for females and males (33% and 36%, respectively). A two-proportion z-test indicates the distributions cannot reject the null hypothesis at <1σ.
This result differs from the AAS meeting data, who found an extremely strong difference between the two. If the first question was asked by a male, then only 10% of the remaining questions were asked by a female. If the first question was asked by a female, 49% of the remaining questions were asked by a female.

Does the gender of the session chair affect the distribution of the questioners' gender?

M         F
Male chair    0.640000  0.360000
Female chair  0.648649  0.351351

When looking at the gender of the session chair (who might have selection biases when there are more questions than time permits), there is no difference in the gender distribution of questioners. In this case the data are nearly identical for male vs. female chairs, at 36% and 35% females respectively.

This null result differs from the NAM data, who saw a small but significant decrease in the fraction of females asking questions when the chair was male (16%) as opposed to a female chair (22%).