“Why Women Don’t Code” — Can It Withstand the Scrutiny of Logic?

On June 19, 2018 a faculty member in my department, Stuart Reges, published an article arguing that women are underrepresented in academic and career Computer Science because they are generally less interested in CS than men are. Since then, there has been a furor of activity and discussion within the Allen School of Computer Science and Engineering and the broader public, with many insisting on the importance of having disagreements and dialogues even when they’re uncomfortable.

I agree that that having critical, uncomfortable discussions is a productive and crucial component of an institution premised on the free exchange of ideas. Such discussions demand a few things: first, a shared interest in developing the best ideas possible; second, an interest in well-developed, logically sound arguments; and third, a willingness to have one’s arguments subject to rigorous (and charitable) analysis.

In that spirit, what follows is a rigorous analysis of Reges’ argument. Given that Reges himself has expressed an interest in logically constructed arguments, I’ve worked through his piece and attempted to reconstruct it into premise-and-conclusion form. While I disagree strongly with his claims, I’ve done my best to analyze his argument as charitably as possible, and believe that in rigorously reconstructing it, its weaknesses come forward all the more clearly. For the sake of space, I have not reconstructed the entirety of his piece, but instead focus on the key elements of his argument: That men and women are fundamentally different (Argument A), that women are underrepresented in CS because they don’t want to be here (Argument B), and that we should accept that where we’re at in terms of representation is as good as it’s going to get (Argument C).

The first key argument to look at is Reges’ claim that men and women are fundamentally different. It can be reconstructed as such (exact quotes from the article in parentheticals):

Argument A:

• Premise 1: Men and women have different beliefs about success/failure (“…men believe in their successes and discount their failures while women believe in their failures and discount their successes.”)
• Premise 2: Men and women have different lifestyle priorities (“…men and women have different priorities” and gives the example of girls and women being more interested in “people, family, ‘balance in life,’ novels, and a good night’s sleep” (quoting Margolis and Fisher))
• Premise 3. Reges states: “How [are P1’ and P2’] possible if men and women don’t differ in fundamental ways?” Rephrased to match the logical structure of this article, P1’ and P2’ could be obtained if and only if men and women differ in fundamental ways.
• Conclusion 1. Men and women differ in fundamental ways.

This appears logically valid, but is it sound? That is, are all premises also true? Not quite. Both Premise 1 and Premise 2 are easily falsifiable — you only have to find one counterexample to prove either premise is false. We can easily find examples of women who believe in their successes, men who believe in their failures, women who don’t care about a good night’s sleep, men who care about family, and many other similar iterations. Let’s say we rephrase those each to be slightly less absolute, such that on average each claim is statistically viable. Is the argument logically sound then?

Well, still no — Reges states that Premises 1 and 2 are only possible if men and women differ in fundamental ways (Premise 3). That is, for Premise 3 to be true, the differences observed in general in Premises 1 and 2 could only be observed as a result of fundamental differences between men and women. Logically this is unsupported, and scientifically we have direct counter evidence that demonstrates the causal role social factors can (and often do) play in shaping statistically significant differences between identity groups [see, e.g., Cheryan, S., Ziegler, S., Montoya, A.M., & Jiang, L. (2017) and Ashcraft (2018)].

While we would agree that, on average, men and women differ in socially relevant ways, it is not logically or scientifically supported that these are “fundamental” rather than socially influenced differences (on average). Reges claims that one must concede that men and women differ in fundamental ways in order to support the aims of any diversity and inclusion effort. That’s not true. Labels like “man” and “woman” are convenient but imperfect ways to approximately measure whether our field is representative of people who have had a diverse variety of experiences in society. Importantly, there are many other experiences within society that are not well-represented in our community (CS in general, and also specifically the Allen School).

Maybe we’ll have more luck with the second main argument, i.e., that we should attribute the underrepresentation of women in CS to their freely choosing not to pursue it. That argument can be reconstructed as follows:

Argument B:

• Premise 1. Women disproportionately choose not to pursue degrees/careers in CS (“…women disproportionately chose to pursue other interests [than CS]”)
• Premise 2. One should never attribute to oppression that which is adequately explained by free choice (“…one should never attribute to oppression that which is adequately explained by free choice”)
• Conclusion 2. We should attribute the underrepresentation of women in CS to free choice (“…I believe that women are less likely than men to want to major in computer science and less likely to pursue a career as a software engineer and that this difference between men and women accounts for most of the gender gap”)

As stated, this is clearly not a logically valid argument, as the conclusion does not necessarily follow from the premises. However, if we add in several unstated premises, we can get closer:

Argument B (adjusted):

• Premise 1 [unstated]. Free choices are choices made in absence of “artificial barriers”
• Premise 2 [unstated]. Oppression is an “artificial barrier”
• Premise 3. Women disproportionately choose not to pursue degrees/careers in CS
• Premise 4 [unstated]. Women’s choices not to pursue degrees/careers in CS are generally free choices
• Premise 5. One should never attribute to oppression that which is adequately explained by free choice
• Premise 6 [unstated]. Women’s underrepresentation in CS is adequately explained by free choice
• Conclusion 2a [unstated]. We should not attribute women’s underrepresentation in CS to oppression.
• Conclusion 2b. We should attribute the underrepresentation of women in CS to free choice.

This is getting closer to being logically valid, but is it sound? In other words, are each of the premises true? Let’s give him 1, and we’ll even give him 2 for now. (However, it bears noting that Reges’ stated definition of oppression is nowhere in keeping with relevant literature on the topic, both in his over-individualization of the problem, and in his reducing oppression to something that arises from overt, intentional efforts to subdue a group of people. By contrast, significant literature on the topic insists on a definition that also recognizes factors that can be much subtler and may not involve individual perpetrators at all — see., e.g., Young, I. (2011). Justice and the politics of difference. Princeton, N.J.: Princeton University Press.)

Premise 3 is backed by statistical evidence, so seems fine. Premise 4 would be difficult to prove, and seems wrong, but for the sake of argument, let’s say it’s true. (However, it bears noting that we have plenty of evidence of it not being true that those choices are fully “free” and unconditioned by other factors — some of which Reges seems to agree with.)

Premise 5 is where things get really weird. It’s a strong claim and doesn’t come with much justification — Reges gives us more examples of various reasons women may choose other careers, but he does not give any argument that would help make the claim that it is untenable to ever attribute explanatory force to oppression when free choice is an adequate alternate explanation. (This is particularly puzzling since it would also lead to situations where a merely adequate free choice explanation would be preferred over even an excellent oppression explanation.) Further, for this premise to hold, one or both of the following must also hold:

A. Oppression and free choice are mutually exclusive explanatory forces.

B. The adequate explanatory force of free choice always takes precedence over any degree of explanatory force of oppression.

Perhaps our use of absolute language is objectionable, and perhaps one could arrive at a more convincing argument by using more tempered language. However, as the claim is so strongly and clearly stated in Reges’s article, it seems likely that any weakening of it would in turn fundamentally alter the spirit of the argument.

That being the case, it’s further unclear why we should accept either A or B as true without further justification. Why couldn’t something be explained by both free choice and oppression as coinciding factors? What if something looked like free choice because the oppressive forces were so covert? Why is free choice assumed to be primary — aren’t there cases where we might take oppression to be the more salient cause?

Turning to Premise 6, we can see that the truth of the claim hinges on what we take an adequate explanation to be, but with little argumentation to provide an answer. We’ve got Premise 4 as a starting point, but are we supposed to take that to be adequate evidence? What standard of adequacy is acceptable?

So clearly there are some serious problems with the second portion of Reges’ argument. While it seems plausible that some amount of representative difference might come down to free choice, we’ve been given no clear indication that it is the only or even the best explanation.

Let’s try the third segment of his argument to see if we can salvage it. The third key position Reges takes is that we’re not likely to make any further progress in increasing women’s representation in CS. His argument, with one added premise to get it closer to logical validity, is as follows:

Argument C (adjusted):

• Premise 1. If we don’t progress more in increasing women’s representation, it is likely we’ll move from positive stories about women succeeding to negative stories about men behaving badly, making women wonder if they should resent men and making men feel guilty for things other men have done (“I worry that lack of progress will make us more likely to switch from positive messages about women succeeding in tech to negative stories about men behaving badly in tech. Women will find themselves wondering if they should resent men and men will feel guilty for sins committed by other men. Women are not going to find this message appealing and men will find themselves feeling even more awkward around women than they would be otherwise.”)
• Premise 2. P1 will do more harm than good (“…I think [P1] will do more harm than good”)
• Premise 3 [unstated]. We should avoid doing more harm than good
• Premise 4. We aren’t likely to make further progress in women’s representation than 20% (“Our community must face the difficult truth than we aren’t likely to make further progress in attracting women to computer science.”)
• Conclusion 3. We should accept 20% women’s representation in tech (“…my honest view is that having 20 percent women in tech is probably the best we are likely to achieve”)

Once again, let’s go through each premise and check for truth content.

Premise 1 has a number of concerning twists and turns (twists and turns, it bears noting, that venture directly down a slippery slope). In talking of “switching” from narratives of women succeeding to narratives of men behaving badly, Reges is working off a number of unjustified assumptions, including: (a) our narratives are currently exclusively about women succeeding and not men behaving badly, (b) those two narratives can’t exist simultaneously. This premise also leaves one to wonder: what should we do about men (or anyone!) who behaves badly in tech if even talking about that bad behavior is out of the question?

Further, the causal relationship between the “switch” and Reges’ projected outcomes (i.e., women wondering whether they should be resentful, men feeling guilty for the sins of others, women not finding the message appealing, and men feeling more awkward around women) is not justified by the argument. It leads to a slew of questions that would need further explanation to hold water in the argument, including:

• Questions about causal role of the “switch:” Why does it follow from the narrative switch that:

— women will wonder whether they should be resentful?

— men will feel guilty? Will they perhaps instead feel like they should do something to help improve the situation? Will they feel aggressively defensive and write long articles about how silenced they feel?

— women will not find the message of men being bad appealing?

— men will feel more awkward around women (and do they feel some degree of awkward already)?

• Questions about what the premise assumes about the current state of things: Is it the case that many women aren’t already resentful in some way about the current status of CS? Is is the case that women don’t already find elements of CS unappealing? Is is the case that many men don’t already feel guilty about how other men conduct themselves?
• Questions about the implied negativity of projected outcomes: Why is it a bad thing for women to feel resentful? Why is it a bad thing for men to feel guilty about the sins of other men? Why is it a bad thing for women to not find this message appealing? Why is it a bad thing for men to feel awkward around women?

Premise 4 seems to be justified entirely by Reges’ arguing that representation rates of women have “stalled” in recent years despite concerted efforts to increase them. While we could concede that this simply is the case, it’s not logically clear why that would be a necessary concession. Maybe, for instance, things have stalled because the tactics we have been using have run their course. Maybe the primary causes for continued underrepresentation are the most difficult to combat. Further, none of these factors say anything about how likely we are to make progress, given that any potential progress would depend on what we actually do. Both Premise 4 and the conclusion are needlessly — and illogically — defeatist.