Abusing Science to Disguise Sexism

The sexist manifesto by James Damore, a now-former engineer at Google, has caused much conversation and distress over the last week. This manifesto attempted to claim that there are fewer women in engineering roles due to inherent biological differences between men and women. I’m not going to address all the issues with it, there are many, but I’ve noticed that defenses of Damore have focused on the science behind his arguments. They claim that since he used science, and science says there are gender differences, his arguments are valid and it was wrong to fire him. Here’s one from The Globe and Mail, and another from David Brooks of the New York Times. James Damore and his defenders aren’t just wrong, their conclusions are the exact opposite of what the research says.

The studies Damore references show there are some differences related to biology. However, those same studies show that these differences are basically nonexistent compared to environmental causes (society, culture, etc). It’s the same result we see with race, there is more variation within men and women than between them. I’m aiming here to help people understand the science of gender differences and see exactly why Damore’s arguments are nothing more than sexism disguised as science.

To establish some credibility, here’s a bit on my background. I received my PhD in Physics from UC-Berkeley in 2014. I wasn’t actually studying physics though, my research focused on how neurons in the prefrontal cortex of rats store short term memory. Before that, and during my undergrad years, I studied molecular dissociation dynamics which is a technical way to say I blew up molecules and watched how they fell apart. I taught myself Python and machine learning, then ended up teaching various data topics at Udacity where I’ve been for nearly two years. As a feminist, a scientist, and a man in tech, I feel I’m obligated to speak up.

The science of gender differences

In 2005, Janet Shibley Hyde collected 46 different meta-analyses on gender differences. Each of these meta-analyses themselves are a collection of studies on variables such as mathematics ability and body self-esteem. She found that for most variables studied, there is a practically insignificant difference between men and women compared to the variance within each group. That is, most psychological variables are determined almost completely by everything other than gender.

In the paper, gender differences are reported as effect sizes with Cohen’s d. This is a normalized measure, d is the difference in means of the two genders divided by the spread in each gender (technically the standard deviation). This way, you can compare the gender differences with the variance within the groups. Small effects will be dwarfed by random variances between people. For values of d less than 0.1, the difference is practically zero. Values between 0.1 and 0.35 are considered small, and values between 0.35 and 0.65 are medium. Below I’ve illustrated the reported gender difference in science ability. Each of these bell curves represents the distribution of science ability for men (in blue) and women (in orange).

You can see that even though there is an effect here, it’s rather small. Mostly there’s an extremely large overlap between men and women. The effect size for science ability is d=0.32, which is considered small to medium. We can also represent this effect using the Pearson correlation, r². This measure shows how much of some variable can be explained by gender. Science ability has an r² of 2.5%, than means 2.5% of any person’s science ability is explained by their gender while the other 97.5% is explained by everything else.

I also want to look at this in a way comparable to gender ratios we see when talking about gender gaps in tech. To do this, I calculated the probability that any given man has a greater ability than any given woman. For example, there is a 56% chance that any given man would be better at science than any given woman. In a pure meritocracy where people get jobs based only on ability, you’d see a gender gap of 56:44 (men to women) in scientific fields.

Here I’ve shown a sampling of variables from the paper with r² and gender ratio measures. I’m also showing the overlap between the distributions. These bars represent two standard deviations, roughly 95% of the populations.

You can see here that two measures of math ability show almost no difference between men and women, as well as leadership effectiveness. Aggression, stereotypically male, is much smaller than you’d assume at 6%. Note that there is a 40% probability that any given woman is more aggressive than any given man. It’s true that men show more aggression than women in general, but there are plenty of combative women. Damore claims that women are more neurotic and that is shown here. But again, this is a small effect at just 2.5% and it’s only a 44–56 split between men and women. Again, plenty of men more anxious than women out there.

Most of the variables covered in Hyde’s paper have near zero or small effects. Just over half the variables have r² values less than 1%. A bit over three-quarters have r² values lesson than 3%. What this means is for pretty much any psychological variable you want to measure, over 97% of what you see is explained by things other than gender. There is the same variance in ability between two women as there is between a woman and a man. Gender tells you practically nothing about how a person performs on math tests, as leaders, or as programmers.

Nature and Nurture

Damore and other science abusers claim that these gender differences are inherent. They say women are more neurotic than men and have worse reasoning skills due only to their biology. However, all these gender differences include effects from both biology and the environment these people exist in. It’s practically impossible to separate these. The environment changes how genes are expressed, genes inform behavior which changes the environment.

It’s well known by now that gender differences can be induced simply by telling people there is a difference. For example, women do worse on math exams when you tell them women are worse at math. This is called a stereotyped threat and it applies to people of color as well. From an excellent article by Gregory Walton and Steven Spencer, subtitled “Grades and Test Scores Systematically Underestimate the Intellectual Ability of Negatively Stereotyped Students”:

It arises from individuals’ awareness of widely known negative stereotypes and the possibility that they could be seen in light of them. When ethnic minority students perform in school, or when women perform in quantitative fields, they are often aware of stereotypes that impugn the ability of their ethnic or gender group. They may worry that a poor performance could lend credence to the stereotype. Hundreds of laboratory experiments demonstrate that this experience, termed stereotype threat, undermines intellectual performance.

In this study, Walton and Spencer analyzed results from multiple studies and found that stereotyped students performed worse on exams with effect sizes matching gender differences reported in Hyde’s study. Stereotyped threat is just one of many societal pressures that serve to push women and other underrepresented groups out of tech. It’s much more plausible that psychological differences we see between men and women are caused by societal effects than being inherent.

Comparing science to reality

Even if inherent psychological differences exist, Damore’s arguments still don’t make sense. In a Planet Money article in 2014, a now-classic figure was presented on the percent of women in college majors by field. Starting in the 1970s, women consistently became more and more represented in medical schools, law schools, the physical sciences, and computer science. This trend continued to today where women are almost at parity with men in all of these fields, except computer science.

If women are inherently less able to reason about complex problems, or less able to handle stress, why are these three other fields nearly at even representation? If it’s biological, why does it not apply to other fields, but only to computer science? And if there is a biological effect, apparently it only began to exist around 1985. The idea that there is some biological reason women aren’t in tech is completely absurd when you look at the representation of women in computer science over time.

Also, note that there is about a 80–20 male-female gap in computer science majors. This is far larger than anything seen in the gender differences data. Even the largest effects such as aggression are only 60–40, while math is around 53–47. This is more in line with what we see for law, medicine, and physical sciences. There is nothing in the gender differences literature that can explain such a huge disparity in computer science. Instead, women are being actively and passively excluded from the field.

Conclusion

James Damore claimed women make for worse programmers than men due to inherent personality differences. His defenders argue that these claims are worth discussing because of the science he references. However, if you look at the actual results, there is no evidence for inherent psychological differences related to software engineering, especially anything that could explain the massive gap in tech. This is an abuse of science to support sexist views. For another look at the science from someone in the field, see this great Quora response by Suzanne Sadedin.

I’m not going to speculate on Damore’s motivations here. I do understand his concerns about being able to discuss his opinions in the liberal environment of Google and other tech companies. We really all should be open to discussing what causes underrepresentation of certain groups. Hopefully we’d be able to change the minds of people like Damore who believe it comes from biology rather than a sexist society. It’s possible that with reasonable debate, Damore would see that biology can’t explain the gender gap in tech.

Damore says he discussed his views with others at Google and was apparently emboldened to share them on a larger scale. Writing the manifesto and sending it to the company, likely knowing it’d be shared outside of Google, is extremely hurtful to his female colleagues and most other women in tech. They already deal with frequent sexism and this is just another sign that they aren’t respected because of their gender.

Note: All the visualizations and other analyses I did for this post are available in this notebook.