Skip to content

Computer Science Gender Gap

June 23, 2019


NY Times article on the paper

LinkedIn source

Lucy Lu Wang is the lead author of a paper released this Friday on gender parity in computer science. The paper is from the Allen Institute for Artificial Intelligence. The authors are Wang, Gabriel Stanovsky, Luca Weihs, and Oren Etzioni. We will call them WSWE for short.

Today we will discuss some of the issues this study raises.

The paper was highlighted by the New York Times in an article titled, “The Gender Gap in Computer Science Research Won’t Close for 100 Years.” The news article begins with an equally sobering statement of this conclusion:

Women will not reach parity with men in writing published computer science research in this century if current trends hold, according to a study released on Friday.

We are for gender-neutral opportunities and have always promoted this—see our earlier discussions here and here. We are for doing a better job in supporting women in computer science. The study by WSWE is an important paper that helps frame the problem. Quoting them:

The field has made more of an effort to reach a more balanced gender status. But the data seems to show that even with all the progress, we are still not making the change fast enough.

I suggest that you might wish to read the paper. Unfortunately there are many papers with similar conclusions—see this by Natalie Schluter, for example.

Main Points

Here are some points of the paper by WSWE:

{\bullet } There is a measurable gap. No one would, I believe, doubt this. But it is important to see that it is measurable.

{\bullet } The gap is shrinking, but slowly. Again this seems correct, but whether it is shrinking in all relevant measures of publication weight is still an issue.

{\bullet } The predictions. Perhaps it will not be closed for over a century.

{\bullet } Modern technology allows such a study. This is one aspect that we can all applaud. WSWE used automated tools that allowed this study to search millions of papers.

Issues

WSWE filtered a corpus of 2.87 million papers tagged as in computer science. The volume constrained their approaches to handling several basic issues.


{\bullet} How to tell the gender of an author? They use first names and try to detect gender from that alone. This is not easy. Not only can differently-gendered names in different nations or language groups have the same Romanized form, many names apply to both genders within those groups. The names Taylor and Kelley are perfect examples pf the latter.

WSWE used a statistical weighing method. So “Taylor,” for example, would be weighted as 55 percent female, 45 percent male. The weightings come from a large database called Gender API compiled from government and social agencies not directly related to computer science.


{\bullet} Another issue concerns the prediction part of their paper. They attempt to extrapolate and guess when there will be parity between female and male authorship.

As all predictions this is not easy. It is my main complaint with this and other papers on the gender-gap issue. They predict that parity will not be reached until 2167, in 168 years. An earlier study puts the parity point at 280 years away.

Open Problems

I believe that a major issue is hiring by computer science departments and other institutions. A major CS department just hired {N>1} assistant professors, of which {N} were male. This is a problem.

Should studies on the gender gap count all papers? Perhaps they should weight the papers by some citation indices. Are women writing more impactful papers? What percent of papers by gender have citations rate above X?—you get the idea.

Finally I wonder if parity is the right goal? How about aiming for more women papers than men? Why not?


[various formatting and word edits]

26 Comments leave one →
  1. June 24, 2019 12:01 am

    Does this weighing of first names make any sense? What happened to Bayes’ theorem?

    • June 28, 2019 1:39 pm

      It seems to make sense to me. They’re calculated the expected value, where the random variables reflect the distribution of gender across names.

      • June 29, 2019 2:30 pm

        If 10% of CScientists are women and 55% of Taylor are women, then the probability of a CScientist named Taylor to be a woman is (55*10)/(45*90+55*10), which is about 12%. So their model errs significantly with ambiguous first names. (Note that I’ve assumed that your first name doesn’t influence whether you become a CScientist or not.)

  2. June 24, 2019 1:00 am

    Regarding the right goal: I believe the better goal should be for every 15 year old to understand that computer science isn’t for one gender, etc.

    After that if people in aggregate make choices, so be it.

    • June 24, 2019 8:20 pm

      People in aggregate don’t make choices. Individuals do.

      • June 25, 2019 2:15 am

        I presume you know what I was trying to say?

      • June 25, 2019 10:08 am

        I think so. But it sounded a little like “if women as a group choose not to be in CS, so be it.” What matters to me are the many factors that could cause individuals, including women, to not feel welcome in CS. A lot happens between middle school, 15, college, and later that make people feel CS isn’t for them. So we can’t just do a good job of outreach to 15-year-olds and then assume our job is done.

      • rjlipton permalink*
        June 25, 2019 11:19 am

        Dear Cris Moore:

        Thanks for this and all your thoughtful comments. I have some theories why women sometimes feel they are not welcome. I have been on many search committees over the years. Often the most critical voices against a female candidate come from other female members of the committee. I can imagine why this happens, but it is not constructive. Not sure how to reduce this.

        Best

        Dick

      • June 26, 2019 9:48 am

        Hello Cris,

        You heard:

        “if women as a group choose not to be in CS, so be it.”

        This is not what I was trying to get across.

        When I say that “if people in aggregate make choices, so be it” what I mean is that if individuals all make their decisions to pursue whatever they want, and there are no discriminatory factors causing them stay away from CS, and if we then look at the proportion of, say, women in CS, and there are fewer women then men, then there is nothing wrong with that.

        I am not suggesting something as simple as outreach to 15 year olds. The work would be done, in theory, from the day they are born up to that point.

        Regards,
        J.P.

      • June 26, 2019 11:37 am

        Hi JP. I understand that you meant from childhood up to 15. But we can’t stop there. In our current world, there are many, many discriminatory factors beyond that point as well, both in school and in early to mid career.

        So while I agree with you that it would be wonderful if there were no such factors, and every individual were free to choose, we are very far from that world. If we ever got to that world, I would expect the demographics of CS to be pretty close to the demographics of the population, including gender, race, class, and so on.

        The fact that the fraction of women in CS used to be higher makes it clear that something has gotten worse recently, either in our institutions or in the surrounding culture. It would be good to understand this.

      • June 27, 2019 1:28 am

        Indeed there are various reasons leading to systematic bias against women. My experience from several decades of serving in various committees is different from Richard’s. Women members are usually more aware to systematic bias against women (although a welcome growing number of men are also aware of it). Sometimes men pays some special attention to cases were women committee members opposes women candidates.

  3. rjlipton permalink*
    June 24, 2019 6:31 am

    Dear J. P. McCarthy:

    Well said. Perhaps one day we will over haul education and place computer science to the front. Thanks for the comment.

    Best

    Dick

  4. June 24, 2019 9:02 am

    “Unfortunately there are many papers with similar conclusions” — is this unfortunate if you are looking for the paper, or unfortunate that so much effort is duplicated, or (most likely I guess) unfortunate that this situation exists?

    More seriously, anecdotal evidence suggests that things were much worse in academia in the past. As a result, a significant number of female academics concealed their gender by using initials, since they found that using their names meant that their papers were more likely to be rejected. However, this perhaps gives a source of bias in the analysis. (I confess I haven’t read the paper to see how this was dealt with.)

    • June 24, 2019 8:23 pm

      It depends how far back in the past you go. The number of women in computer science was significantly higher 15 years ago: http://www.aei.org/publication/chart-of-the-day-the-declining-female-share-of-computer-science-degrees-from-28-to-18/

      Personally, I blame the rise of social media and toxic brogrammer culture, but it’s not clear what the mechanism is. In any case, it’s clear that we’re losing a lot of talented young people.

    • June 24, 2019 8:27 pm

      It depends how far back in the past you go. The number of women in computer science was significantly higher 15 years ago. (I tried to past a link from aei.org but it seems to break the commenting system.)

      It’s unclear what the mechanism is — I personally blame the rise of social media and toxic brogrammer culture — but it’s clear that we’re losing a lot of talented young people.

    • June 28, 2019 5:41 am

      Cris raised a valuable point. It is indeed true that scientific blogs, MathOverflow, polymath projects, have lower participation of women and often women find these platforms uninviting and at times even hostile.

      • June 28, 2019 8:45 am

        I have to say that I find MathOverflow, polymath projects, etc. uninviting. I really don’t like the point scoring aspect of MO and StackExchange. I don’t think I have ever contributed to any of these, even GroupPubForum, which is by far the friendliest such list in my view. (I have replied to individuals who post on GroupPubForum if I have a piece of knowledge I think might be useful to them.)

        On a similar topic, I learned not so long ago about Impostor Syndrome, in a context which suggested that only women suffer from this. Absolutely not true. I think many mathematicians have it to some degree. After all, our results seem obvious to us (once they are proved), whereas other people’s worok (sometimes presented without much background) seems incomprehensible and therefore presumably deep.

        I am all for righting historic injustices. But for me, first and foremost, we are people!

      • June 28, 2019 10:51 am

        Yes, Impostor Syndrome is pretty universal. There are flavors of it that people from underrepresented groups have to deal with that I (for instance) don’t, but it is something I’ve struggled with as well. Maybe this is a good topic for another blog post.

  5. rjlipton permalink*
    June 24, 2019 11:08 am

    Dear Peter Cameron:

    The “unfortunate” was bad wording. But I mostly meant too bad these papers are needed.

    The initial comment is part of the discover issue that they have to solve: how to tell the gender from a name. Initials seems to be even harder. Of course there is G.H. Hardy.

    Best

    Dick

  6. June 28, 2019 1:57 pm

    Ruth Bader Ginsburg was spoke at the University of Chicago in 2017 and said “When I’m sometimes asked when will there be enough [women on the Supreme Court] and I say, ‘When there are nine,’ people are shocked. But there’d been nine men, and nobody’s ever raised a question about that.”

    Personally, I think there’s no good way to attempt to enforce gender parity, no do I think that’s a worthwhile moral goal. I think that in CS, like in all things, eyes should be set on eroding undesirable institutional, cultural, and political systems. The lack of women in computer science is a symptom of the problem, not the problem. Likewise, raising the proportion is expected to make progress on alleviating the problem not solve it.

    It’s easy (and common!) to conflate these things, but the bad guys here are discrimination, implicit bias, and systems of oppression. Not the precise number of women (or transgender people, or Latin@ people, or etc.) in the field.

    • rjlipton permalink*
      July 1, 2019 3:06 pm

      Dear Stella Biderman:

      Your quote from Justice Ginsburg is terrific. I agree with you other comments. The losers in any discrimination is us—“We have met the enemy and he is us”—as Pogo said. Any time some one does not elect to enter computer science and goes elsewhere, we all lose.

      Thanks again.

      Best

      Dick

  7. Sunnysideup permalink
    July 1, 2019 3:38 am

    This post is drenched in an extreme and dogmatic view of social justice. It seems to lay the blame for unequal representation of women in CS squarely at the feet of bias and an unwelcoming environment. Where is the proof of this? While no doubt these factors play some role, the extent of this role as compared to natural+cultural disparities in interests between men and women is very much an open question. With much more “broey” fields like medicine having much more success at attracting women, the burden of proof is on those who support the presumptions of this post. Equality of opportunity is desirable, but enforced equality of outcome — which this post implicitly supports — is tyranny.

    • rjlipton permalink*
      July 1, 2019 8:59 am

      Dear Sunnysideup:

      You comment is a bit harsh, but we do wish to hear all views. Essentially you are saying yes to equal opportunity. Good.

      I do not agree with your tone, however. The bottom line for me is that the more researchers we have the better. The more views we have the better. The more we have may help us solve some of the key problems that are still unsolved. Who knows who will be the one to make a breakthrough. This is why I have always supported diversity.

      Best

      Dick

    • July 1, 2019 10:35 am

      Sunnysideup, I think it’s important to address what people say rather than what you think they implicitly support. I don’t see anyone here calling to “enforce equality of outcome”.

      What I do see people calling for is a better understanding of the barriers that stand in the way of people getting into CS and staying there. Some of these barriers have to do mainly with the larger society (e.g. low-income rural people and first-generation students having a harder time getting into top institutions) and some are things we can do something about (how we present ourselves to the world, how we teach, what support we provide to people once they’re accepted to a program, how we award tenure, etc.). I’m sure you agree that if we can make CS more welcoming to more people, we should. We are not the Marines; our goal should not be to weed out almost everyone.

      • rjlipton permalink*
        July 1, 2019 2:50 pm

        Dear Cris Moore:

        Wonderful comment. I like the marine comment especially. Thanks again.

        Best

        Dick

  8. October 25, 2019 2:47 am

    This is a constant battle as a CS teacher & tutor. Reaching parity between genders in enrolment does feel like the current holy grail in our subject. Some days I do feel like it’s improving, and then I get another tutee ask for help because their teacher told them that girls don’t really do CS.

    One thing I think we should do less of as educators is to stop making things pink & sparkly to attract girls – just teach the subject like we do maths (for everyone) & expect them to be good at it!

Leave a Reply to Peter Cameron Cancel reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s