This paper made a bit of a splash when it was first published last year. To quote the MSNBC article:

The stereotype of computer scientists as geeks who memorize Star Trek lines and never leave the lab may be driving women away from the field, a new study suggests. And women can be turned off by just the physical environment, say, of a computer-science classroom or office that's strewn with objects considered "masculine geeky," such as video games and science-fiction stuff.

I decided, however, to hold off blogging about it until I actually had the chance to read it. I'm glad I did, as it actually contains a little more nuance than makes it into the popular press.

The paper actually has four individual studies in it, but the last three are really there to refine and shed light on the first one. If you were to watch the first study, here is what you see:

On some weekend day, a volunteer (an undergraduate who is not in CS and not a senior) enters the CS building of Stanford University to complete what they think will be a 'Career Development Center' survey. They are taken to a room, told to 'pay no attention to the stuff in the room' as it belongs to another group (and not the experimenter) and given a minute to themselves (while the experimenter goes to 'get the materials'). The volunteer then completes a six- minute word-stem completion task, a 18-question CS test 'to assess for stereotype threat' and then a questionnaire about their perceptions of the test. Then the volunteer fills out a questionnaire on their perceptions of computer science and how interested they are in it. When done with this, the volunteer is brought to the lobby for a debrief.

Most of this, of course, is smoke to distract the volunteer from the real purpose of the experiment: do the objects in the room affect the interest of women in computer science? Half the time, the room contained items judged 'stereotypical' of computer scientists (read: geek) and half the time, the room contained neutral objects. Did it matter? Seems to: women who saw the stereotypical objects were much less interested in computer science than the women who saw the 'non-stereotypical' objects.

Now, here's where the popular press will jump right to the 'Conclusions' section. But not only are there three other studies in this paper, I haven't even finished telling you about the first one! First of all, and just to poke at the validity of the experiment a bit, how did the experimenters choose the stereotypical and non-stereotypical objects? Could something fishy be going on there? And supposing that the experiment is valid, what does it mean? Why do these objects have this effect?

First, the objects: yes, they were chosen carefully. One group of volunteers was asked to (individually) list items found in the office/dorm of a 'stereotypical computer scientist' or 'stereotypical computer science major.' The experimenters kept the most common items, added similar but non- apparently-stereotypical objects, and asked a separate group to rate them on how much they associated the objects with computer scientists.

  • The most associated: a Star Trek poster, comics, video game boxes, soda cans, junk food, electronics, computer parts, software, technical books and magazines.
  • Unassociated: nature poster, art, water bottles, healthy snacks, coffee mugs, general-interest books and magazines.

Okay, so did the objects have an effect on women's interest in computer science? Oh my yes. A lot. If you look at the relevant diagram of the paper (Figure 1), you'll see that that it had a big effect. (It also had an effect on the men, too, but I can't tell if it is a statistically-significant one. I'll come back to both the statistics and the men below.) But all this only begs the next question: why?

The next three quarters of the paper are directly aimed at this question. One possibility, raised by a cited paper, is that the objects remind women that CS is a male-dominated field and that they might be the target of discrimination. Another possibility, which I came up with during my read of the paper, is that the stereotypical objects might convey unhealthiness. (Junk food and soda cans, versus water bottles?) But the hypothesis of the experimenters is that the objects influence the volunteers' senses of 'ambient belonging.'

Okay, what the heck is that? The core idea, I think, is that two things combine. First, people will infer things about a stranger or a group from the objects in that stranger's/group's environment. Thus, the objects act as what they call 'ambient identity cues.' Second, people want to feel like they belong. They want to be in situation where they 'fit,' and feel uneasy when they get the sense that they don't belong. Based on this, the authors define 'ambient belonging' as '...the feeling of fitting into an environment' and believe that their volunteers are gauging ambient belonging (into the field of CS) from the objects in the room.

Okay, but that only re-phrases the question. Why do the stereotypical objects tell women (as a population) that they will not belong? Because (the authors claim) the objects are gauged to be masculine. In other words, the objects tell women that CS = men, and they conclude that they don't belong.

The next three studies in the paper poke at this idea in various ways. Unlike the first study, the next three don't actually take place in the real world. Instead, the experimenters have the subjects read some text on a computer and answer questions. Basically, the subjects get to read about two companies / teams / etc., and rate their interest in each. The two groups are always exactly the same except for the objects mentioned in their descriptions-- one includes the stereotypical objects, the other uses the non-stereotypical objects. However, the three studies vary this basic idea in different ways:

  • In one study, both teams have a 50-50 gender ratio, and another study has both teams be all women.
  • One of the studies has the volunteer choose between the two potential employers, while another study has the volunteer rate their interest in each company individually.
  • Also, the three studies tried to measure this sense of masculinity and ambient belonging directly. One asked volunteers to rate the masculinity / femininity of each company as a whole, for example, while another asked volunteers to individually rate the masculinity / femininity of each object mentioned. Also, some of the studies asked volunteers to rate how much they identified with the teams, and so on.

In short, the next three studies confirmed that the objects mentioned will affect the interest of women (as a population) in a particular work- environment. And (using heavy-duty statistics I don't really understand) they confirmed a few things:

  1. Their 'ambient belonging' measure ('I belong here' / 'I don't belong here') better explained the reported interest of female volunteers than other explanations (such as the fear of future discrimination. Unfortunately, they didn't test my 'healthy' / 'unhealthy' hypothesis ;-)
  2. In women, the sense of ambient belonging was correlated with perceived masculinity and femininity of the objects.

(This might be a good time to re-iterate that I don't really understand the statistics, and may be a little off in my phrasing, here.)

So where is this going? The topic of 'women in computer science' is waaaay bigger than this one paper, and I still don't have anything new to say about it. The paper does point the way to one possible tactic we can use in the future: change the environments of undergraduate CS labs and classrooms. I'd be in favor of this on general grounds (man, some of the labs at my school were filthy!) but the paper suggests that this might at least reduce the entry-barrier for women. And since the 'CS nerd' stereotype is pretty damn exaggerated (as the paper notes) we might be able to make some progress just by not driving women away.

I have no evidence against this idea, but I have my doubts. It seems like a panacea. It's too easy, and I'm suspicious of things that are too easy. I can't believe that the massively disproportionate gender ratio we have in CS is due to a few pizza boxes and Star Trek posters. I'd be overjoyed if that were true, but I somehow suspect that there is something deeper going on here, and that we can redecorate all we want without seeing much effect. My studies continue. (But hey-- please, prove me wrong!)

Some other thoughts:

  • Man, psychology is HARD. The paper goes into great detail about all the little intricacies of their methodology, and the great pains they took to avoid biasing the volunteers / avoid confounding factors / screen for bad data / etc. It all sounds like a tremendous amount of work, and all in the hope of getting one statistically significant answer to one tiny yet/no question about one aspect of a very, very large and nebulous problem. I'm glad someone is doing this kind of work... and I'm glad it isn't me.
  • Again, I am completely unqualified to understand the statistics of this paper. I am aware that 'real' statistics is actually much less powerful than we see on TV, but I do wish I could understand this (discussing the first study):

Can environments stereotypically associated with computer science deter women's participation? In a 2 (gender) x 2 (environment: stereotypical, non-stereotypical) ANOVA on interest in computer science, we found no main effect on gender F(1, 35) < 1, ns, or environment F(1, 35) = 2.56, ns. However, there was significant interaction of gender and environment F(1,35) = 6.91, p < .05, η2p = .17. As predicted, in the stereotypical environment, women were less interested in computer science than were men (women: M = -.55, SD = 0.38; men: M = 0.22, SD = 0.85), F(1,35) = 4.58, p < .05, η2p = .12. However, in the non-stereotypical environment, there was no gender difference in interest in computer science (women: M = -.52, SD = 1.03; men: M = -0.04, SD = 0.81), F(1,35) = 2.50, ns.

In particular, can anyone tell me if this indicates now much of an effect the environment had? Can we use these statistics to estimate much of the gender disparity in CS is due to the effect being measured?

  • While the studies were designed to test the effects of these environments on women, there were a few interesting tidbits about the effects they had on men. For example: in one of the 'virtual' studies (read a description and answer questions) men also seemed to be turned off by the stereotypical objects (Study 3). However, this effect disappeared in the 'real world' (first) study, and men displayed the opposite behavior in one of the other virtual studies (Study 4). In that study, the stereotypical objects had no effect on men (when judged against a pre-study baseline) but found that men were actively turned off by the non-stereotypical environment. So, the paper didn't uncover a consistent effect on men, but then again, that wasn't its purpose.
  • Not being a psychologist, I can't quite figure out how to regard this theory of 'ambient belonging.' I agree that we can directly observe some things: the objects in the room, the reported interest in computer science, reported masculinity/femininity of the objects, etc. But in what sense are we observing ambient belonging? One answer is that we are observing a relationship between inputs (objects) and outputs (interest levels, self-reported senses of belonging, etc.) and concluding that there must be this intermediate thing called 'ambient belonging.' I'm okay with that-- I'm happy to believe in lots of intermediate things that cannot be directly observed. (Energy. Money. Facebook.) But is this paper claiming that ambient belonging is real in the sense that a joule of energy is real? That someday, we will some day be able to observe the 'ambient belonging' center in an fMRI? Or are they saying that 'ambient belonging' is a convenient simplification that will later be broken down into more concrete components?