The research team, led by Joseph R. Cimpian of New York University together with Zachary T. McDermott and Joseph R. King and colleagues, drew on two large longitudinal datasets that together follow American students across critical educational transitions. The first, the High School Longitudinal Study, tracked 6,340 students from ninth grade onward, capturing their trajectories as they moved through secondary school and into postsecondary education. The second, the Baccalaureate and Beyond Class of 2016, followed 12,950 students who earned bachelor’s degrees, providing detailed information about the majors they selected and the characteristics they brought to those choices. By using these nationally representative samples, the investigators were able to move beyond the convenience samples that dominate much of the literature on gender and academic sorting, offering a portrait of segregation that reflects the full diversity of American higher education.

The analytical approach centered on estimating relative representation across majors—a measure of how much more or less likely a given group is to be found in a particular field compared with a reference group. The fields examined spanned the full spectrum of gender composition, from overwhelmingly male-majority disciplines like physics, engineering and computer science at one extreme to female-majority health fields at the other. Rather than treating segregation as a single, monolithic quantity, the researchers disaggregated their estimates by achievement level, allowing them to test a hypothesis that has rarely been examined at scale: that the social dynamics of a field operate differently for students at different points of the achievement distribution.

The central finding is striking in its magnitude and consistency. Across both datasets, representation was most polarized among lower-achieving students, and the effect was particularly pronounced in male-majority fields. Among lower-achieving students, straight-cisgender males were substantially overrepresented relative to straight-cisgender females, with a relative ratio of 2.62 and a 95 percent confidence interval spanning 2.29 to 3.00. In practical terms, this means that a lower-achieving straight-cisgender male was roughly two and a half times more likely than his lower-achieving straight-cisgender female counterpart to be found in male-dominated majors—a gap that dwarfs the differences observed among higher-achieving students. The representation gap between men and women in these same fields was significantly smaller at higher achievement levels, a difference the researchers report as statistically robust at P < 0.001.

This achievement-dependent pattern carries profound implications for how scientists and policymakers understand the roots of gender segregation in science, technology, engineering and mathematics. A common assumption in the field has been that the underrepresentation of women in physics and engineering reflects a “leaky pipeline” in which talented women are progressively filtered out by stereotype threat, discrimination or attrition. The new findings complicate that narrative considerably. If segregation were driven primarily by the loss of high-achieving women, the disparity would be greatest at the top of the achievement distribution. Instead, the data show the opposite: the men and women who do enter male-majority fields at high achievement levels are relatively similar in composition, while the lower-achieving end of the distribution is where men cluster overwhelmingly and women largely disappear.

To explain this pattern, the researchers turned to the concept of field cultures and social norms that create in-group benefits for members of field-specific majority groups. The idea is that each academic field develops its own cultural environment—norms about who belongs, how competence is signaled, and what kinds of students feel comfortable and supported. For members of the majority group in a given field, these norms act as a kind of subsidy, easing their path into and through the discipline. Crucially, the researchers hypothesized that this subsidy matters most for students whose academic credentials are marginal. A lower-achieving student choosing a major is making a riskier bet than a high-achieving student; if the field’s culture provides a cushion of belonging for in-group members, that cushion may be decisive precisely for students who lack the achievement buffer to overcome a hostile or unwelcoming environment. High-achieving students, by contrast, may be able to succeed regardless of field culture, compressing the differences between groups at the top of the distribution.

One of the study’s most important contributions is its demonstration that these patterns cannot be explained away by the usual suspects. The researchers adjusted their estimates for prior achievement, coursework completion, expressed interests and stated aspirations, and the polarization among lower-achieving students persisted. This is a technically significant result because interests and aspirations are frequently cited as the primary explanation for gendered field choices—the argument being that women simply prefer people-oriented fields while men prefer thing-oriented ones. If measured interests and aspirations fully accounted for segregation, controlling for them would eliminate the disparities. The fact that substantial gaps remain after such adjustments suggests that something about the fields themselves—their cultures, their perceived climates, their social signaling—shapes who chooses them, particularly among students whose qualifications leave the least room for error.

The study also extended the analysis to LGBTQ+ identity, an axis of diversity that is almost entirely absent from large-scale studies of academic sorting. Here the picture was more complicated. In female-majority fields, the evidence for an analogous in-group benefit vis-à-vis LGBTQ+ identity was mixed across the statistical models the researchers employed. Unlike the stark and consistent pattern observed for gender in male-majority fields, the findings for sexual orientation and gender identity in female-dominated fields did not converge on a single interpretation, and the authors are careful not to overclaim. The asymmetry itself is informative, however: it suggests that the mechanisms producing in-group advantages may differ between male-majority and female-majority contexts, or that the datasets’ statistical power and measurement of LGBTQ+ identity limit the detectability of effects that do exist. Either way, the mixed results underscore how much remains unknown about the experiences of LGBTQ+ students across academic fields, and how thinly studied this population remains despite growing evidence that field climate matters for retention.

The methodological rigor of the work deserves emphasis. By replicating the analysis across two independent, nationally representative longitudinal datasets—one following students into college, the other following bachelor’s degree recipients—the researchers guarded against the possibility that their findings were artifacts of a particular sample, cohort or measurement strategy. The convergence of estimates across datasets with different designs strengthens the inference that achievement-dependent segregation is a real and general feature of American higher education rather than a statistical fluke. The use of confidence intervals and explicit comparisons across achievement levels, formalized in tests yielding P values below 0.001, reflects a quantitative standard that moves the conversation beyond anecdote and small-sample studies.

For the field of talent development, the implications are far-reaching. If the forces shaping field-level sorting operate most powerfully at the lower end of the achievement distribution, then interventions aimed at broadening participation may need to be redesigned accordingly. Efforts to diversify physics, engineering and computer science have historically focused on high-achieving students—scholarships for top performers, enrichment programs for advanced students, recruitment at elite institutions. But if the women and LGBTQ+ students who are missing from these fields are disproportionately those with more marginal academic credentials, then the bottleneck may lie elsewhere: in the culture of the fields themselves, which appears to welcome in-group members with weaker qualifications while filtering out out-group members with similar or better credentials. The finding that lower-achieving men can and do enter male-majority fields at high rates, while similarly situated women do not, is difficult to reconcile with purely meritocratic explanations of who ends up where.

The study also reframes how we should think about achievement itself in the context of field choice. Lower-achieving students constitute a large share of the college-going population, and their field decisions collectively shape the demographic composition of entire professions. If field cultures grant in-group benefits that are worth more to these students precisely because their academic position is more fragile, then the composition of male-majority fields is being determined not by a fair competition in which the best-prepared students of all backgrounds choose freely, but by an environment in which belonging functions as an unearned advantage. Conversely, the relative parity among high-achieving men and women in male-majority fields suggests that strong academic preparation can partially insulate students from unwelcoming cultures—an observation that is both hopeful and troubling, since it implies that cultural barriers remain but can be temporarily outspent by credentials.

The researchers are careful to note that their findings describe patterns of representation, not the individual psychological mechanisms that produce them. Future work will need to identify precisely which features of field cultures generate in-group benefits—whether it is classroom climate, peer networks, mentoring structures, grading practices or the images of who belongs that fields project to prospective students. But by demonstrating that segregation varies systematically with achievement, and that LGBTQ+ identity can be brought into the analytical frame at national scale, the study opens a new frontier in the science of educational inequality. The message for institutions hoping to broaden participation is clear: to change who enters a field, one may first have to change the culture that decides who feels entitled to be there, and that culture exerts its strongest influence on the students with the least margin to absorb its costs.