The gender gap in STEM reflects natural differences in interests and abilities

The claim

Women are underrepresented in computer science, engineering, and physics not because of discrimination or structural barriers, but because of genuine differences in interests and, to some degree, abilities. Proponents cite surveys showing women more interested in people-oriented fields and men more interested in things, evolutionary psychology accounts of spatial reasoning differences, and argue that persistent gaps even after decades of diversity interventions prove the explanation must be biological. The 2017 “Google memo” by James Damore made this case explicitly, citing Simon Baron-Cohen’s work on systemizing vs. empathizing cognitive styles and arguing that Google’s diversity programs were misallocating resources by fighting an inevitable natural distribution.

The mechanism

The natural-differences argument proposes two linked mechanisms. First, a cognitive ability gap: men show higher average scores on some spatial rotation tasks, and the tail of mathematical achievement tests is slightly wider for males than females — meaning more men than women at the extreme right tail of certain assessments. Second, an interest gap: surveys consistently find men more drawn to objects, systems, and things; women more drawn to social roles, communication, and people. Together, these would predict fewer women in engineering and CS even in an unbiased world.

The first mechanism does not hold up to cross-national testing. If the ability distribution were driving the gap, we would expect similar male-to-female ratios in CS across countries with the same species-level biology. We do not find this. We find Malaysia with ~55% female CS graduates, Bulgaria and Romania historically above 50%, India with female enrollment in information technology programs well above the US ratio, and Israel with higher female representation in CS than the United States. Cognitive ability distributions do not vary this dramatically across national borders. Culture, economic structure, and framing do.

The interest-gap mechanism is more complicated but equally vulnerable. Interests are measured after socialization has already occurred. A girl who has been told since childhood that computers are “for boys,” who never saw a female programmer in any film or television show, whose parents bought a computer for her brother, who experienced stereotype threat in every math and science classroom — that girl’s reported interest at age 17 is not a clean read on her pre-social preferences. The question is not whether interest gaps are observed in adults (they are), but how much of the observed gap reflects innate preference structure versus socially installed expectations. The evidence strongly favors the latter as the dominant source.

The evidence

The 1984 peak and the home computer marketing inflection. The single most important US data point is the trajectory of women’s CS degrees over time. Women’s share of CS bachelor’s degrees rose steadily through the 1970s alongside other technical fields, reaching 37.1% in 1984 — higher than the current figure. It then fell sharply and continuously, reaching 18% by 2012 and recovering only slightly since. No other STEM field shows this pattern: women’s representation in biology, chemistry, and medicine continued rising through the same period. The divergence between CS and other STEM fields in the mid-1980s corresponds precisely with the explosion of home computer marketing — Apple, Atari, Commodore, IBM — which was targeted almost exclusively at boys. The home computer became a boy’s toy in American cultural framing. Boys arrived at university CS programs already experienced; girls did not. The introductory course assumed that experience, creating an immediate competence gap that was culturally produced, not biologically given. If the natural-differences account were correct, there is no mechanism by which it would produce a 19-percentage-point decline in female representation in a single field over 30 years while other STEM fields moved in the opposite direction.

Cross-national variation. Rohini Varma’s comparative research on women in computing in India and the United States documents that Indian women pursue CS at substantially higher rates than their American counterparts — and report higher computing self-efficacy. The explanation is not biological. In India, computing is understood as a clerical and intellectual profession suited for women; the cultural framing is different. In Malaysia, CS is associated with upward mobility and is gender-neutral in social coding. In the Soviet Union and its successor states, technical fields were explicitly promoted as prestigious for both sexes, and women filled technical and scientific roles at rates the United States has never approached. These differences map onto cultural and economic structures, not onto the cognitive architecture of human females.

Stereotype threat research. Claude Steele and colleagues developed stereotype threat theory to explain performance gaps that appear under conditions where a negative stereotype about a group’s ability is salient. Spencer, Steele, and Quinn (1999) tested this directly with math-identified women: when told before a math test that the test had shown gender differences in the past, women performed significantly below equivalent men; when told no gender differences had been found, the gap disappeared. The mechanism is cognitive — stereotype threat consumes working memory resources needed for difficult tasks. More than 300 subsequent studies have replicated and extended this finding. Effect sizes vary, and meta-analyses show some publication bias, but the core result — that activating gender identity in math contexts suppresses women’s performance — is among the more robust findings in social psychology. The “chilly climate” literature extends this: women in CS courses report more experiences of having their contributions ignored, their competence questioned, and their belonging in the field challenged. These are not individual responses to ability differences; they are structural features of environments that predict women’s departure from computing programs.

Ceci and Williams — the nuanced evidence. It is important to engage seriously with the strongest version of the individual-differences argument. Wendy Williams and Stephen Ceci (2011, PNAS) conducted a systematic review of hiring, grant funding, and publication patterns in academic STEM and found, to their surprise, that women who applied for faculty positions were not being disadvantaged in the hiring process — and in some fields were advantaged. Their finding is real and matters: the form of discrimination varies by career stage and field, and in the current academic hiring environment there is no simple story of explicit bias at the hiring gate. Ceci and Williams also acknowledge, however, that the pipeline problems are earlier: differential representation at the PhD applicant stage, which reflects earlier barriers. The absence of a hiring penalty for women who do apply to academic jobs does not resolve the question of why fewer women apply. Baron-Cohen’s systemizing-empathizing model (2003) has been challenged on the grounds that the tools used to measure these dimensions are culturally saturated and that effect sizes are much smaller than popular accounts suggest. Hyde’s meta-analytic “gender similarities hypothesis” (2005, American Psychologist) found that psychological gender differences on cognitive and personality measures are predominantly small to negligible, with a few exceptions (effect sizes below d = 0.35 characterize the vast majority of measures).

Harvey Mudd and Carnegie Mellon interventions. Harvey Mudd College in California raised its female CS enrollment from 10% to 40% within five years by making three structural changes: renaming the introductory course from “Introduction to Programming in Java” to “Creative Approaches to Problem Solving in Science and Engineering”; ensuring introductory sections were available for students with no prior programming experience; and sending female students to the Grace Hopper Celebration of Women in Computing early in their first year. No change to the cognitive demands of the curriculum was required. This is a near-controlled experiment in structural causation: holding constant the biology of the student population, curriculum redesign moved female enrollment by 30 percentage points.

The “gender-equality paradox” and its critics. A much-cited 2018 paper (Stoet & Geary, Psychological Science) claimed that in countries with higher gender equality, women are less likely to enter STEM — the “gender-equality paradox.” This finding has been extensively critiqued. Richardson et al. (2020, Science) identified coding errors in the original analysis and, using corrected data, found the claimed paradox substantially weakened or absent. The paradox has been used to argue that in “free” societies women reveal their true preferences by choosing away from STEM — but the empirical basis for this argument is fragile.

Who benefits

The claim that the gender gap in STEM is natural has specific economic and political beneficiaries. Technology companies facing legal exposure under Title IX, the Equal Pay Act, and state anti-discrimination statutes have a direct financial interest in maintaining that their underrepresentation of women reflects a supply problem (too few qualified women apply) rather than demand-side bias (hiring, retention, and promotion practices that discourage women). The Damore memo, circulated internally at Google, was written by an employee who argued that Google’s diversity initiatives were therefore misallocating resources — a position that, if accepted, would terminate programs that impose costs on the company. Funding for think tanks that produce natural-differences arguments — the American Enterprise Institute, the Manhattan Institute — flows substantially from technology industry and financial sector donors who have interests in resisting labor regulation. Simon Baron-Cohen’s Autism Research Centre at Cambridge receives corporate research funding. None of this constitutes a conflict of interest per se, but the direction of institutional interest is clear and worth noting.

The counter

The strongest version of the natural-differences argument is not the Damore memo version — it is Ceci and Williams’ careful academic work, which finds that at the point of academic hiring in most STEM fields, women are not currently being discriminated against, and that in some cases are advantaged. The residual interest gap — measured in adults, measured across cultures, measured even in relatively egalitarian Scandinavia — is real and does not disappear when you control for cultural exposure. Some portion of the observed interest difference may be genuine and pre-social. Baron-Cohen’s extreme systemizing–empathizing claim (large sex differences) is overstated, but Hyde’s gender-similarities hypothesis does not mean zero difference. The honest position is that some portion of the gap reflects constructed preferences, some portion may reflect innate preference variation, and the evidence cannot cleanly separate these — but the magnitude of variation across time and country rules out the strong natural claim that the gap is primarily biological. The structural account is not “discrimination is everywhere”; it is that the framing of computing as male, the historical accident of gendered home computer marketing, and the cumulative effects of stereotype-threat and climate effects in educational settings account for most of the observed gap.

References

Baron-Cohen, S. (2003). The essential difference: The truth about the male and female brain. Basic Books.

Ceci, S. J., & Williams, W. M. (2011). Understanding current causes of women’s underrepresentation in science. Proceedings of the National Academy of Sciences, 108(8), 3157–3162. https://doi.org/10.1073/pnas.1014871108

Hyde, J. S. (2005). The gender similarities hypothesis. American Psychologist, 60(6), 581–592. https://doi.org/10.1037/0003-066X.60.6.581

Richardson, S. S., Reiches, M. W., Bruch, J., Boulicault, M., Noll, N. E., & Shattuck-Heidorn, H. (2020). Scrutinizing the science of the gender-equality paradox. Science, 369(6509), 1291–1291. https://doi.org/10.1126/science.abd3996

Shetterly, M. L. (2016). Hidden figures: The American dream and the untold story of the Black women mathematicians who helped win the space race. William Morrow.

Spencer, S. J., Steele, C. M., & Quinn, D. M. (1999). Stereotype threat and women’s math performance. Journal of Experimental Social Psychology, 35(1), 4–28. https://doi.org/10.1006/jesp.1998.1373

Steele, C. M., & Aronson, J. (1995). Stereotype threat and the intellectual test performance of African Americans. Journal of Personality and Social Psychology, 69(5), 797–811. https://doi.org/10.1037/0022-3514.69.5.797

Varma, R. (2010). Computing self-efficacy and interest among women in India and the United States. Journal of Women and Minorities in Science and Engineering, 16(1), 1–20. https://doi.org/10.1615/JWomenMinorScienEng.v16.i1.10

Stoet, G., & Geary, D. C. (2018). The gender-equality paradox in science, technology, engineering, and mathematics education. Psychological Science, 29(4), 581–593. https://doi.org/10.1177/0956797617741719

American Association of University Women. (2015). Solving the equation: The variables for women’s success in engineering and computing. AAUW. https://www.aauw.org/resources/research/solvingtheequation/