Women 1.5 Times More Likely to Leave STEM Pipeline after Calculus Compared to MenLack of Mathematical Confidence a Potential Culprit

Seventy-five percent. That's how much the number of women entering the STEM workforce would increase if women made it through Calculus I at the same rate as men. Not if we fixed hiring bias, not if we redesigned every lab culture — just this one course. So here's the question worth sitting with: is this a story about ability? Because the data from Ellis, Fosdick, and Rasmussen say no. And what they found instead is sharper and, in some ways, harder to fix. Calculus I sits at the narrow neck of an hourglass. Nearly every science, technology, engineering, and mathematics discipline in the United States requires it, which means a student's decision not to continue to Calculus II is effectively a decision to exit the STEM pipeline. Ellis and colleagues used data from a large national survey, part of the Characteristics of Successful Programs in College Calculus project, to examine exactly that decision. They surveyed students at the start and end of a Calculus I term, asking about their intention to take Calculus II, and followed up a year later to see who actually did. Students who maintained their intention to continue were labeled Persisters. Those whose intention fell by the end of the term were labeled Switchers. Across the full dataset, those categories captured over four thousand eight hundred students; the analytic sample with complete data came to two thousand two hundred sixty-six students, of whom seventeen point eight percent were Switchers. Two distinctions in the study design are worth holding onto. First, the researchers separated STEM-intending students — those committed to a STEM major at the start of the term — from STEM-interested students, who were still deciding but had indicated they planned to take Calculus II. Both groups could become Switchers, and the dynamics differed between them in revealing ways. Second, the statistical model the team used was a mixed-effects logistic regression, with institution included as a random effect. In plain language, the model compares Switchers and Persisters while holding constant what could be measured — academic preparedness, career intentions, quality of instruction — and while acknowledging that students at the same university aren't independent of each other. This is what makes the gender finding so difficult to dismiss. After controlling for all of that, a female student's odds of switching out of the calculus sequence were approximately one point five times those of a comparable male student. The ninety-five percent credible interval runs from one point fourteen to one point eighty-nine, which means this is not a borderline effect. To make it concrete: for a student with average test scores, a history of high school calculus, pursuing a science or technology career, and reporting average instruction, the male switching probability was estimated at eleven point seven percent. For an otherwise identical female student, it was sixteen point three percent. Change the profile — no high school calculus, pursuing a non-STEM career — and the male probability rises to thirty-one point three percent, the female to forty percent. The gap follows women across profiles. Preparedness, career goals, instructor quality — none of them close it. Three variables did predict persistence significantly. Higher standardized math test scores helped. Career intentions mattered — intending to be an engineer, for instance, was associated with greater persistence than a general science or math path. And higher instructor quality scores were associated with staying. Previous calculus experience and student-centered teaching practices didn't reach significance. But gender remained significant across all of it. When students who switched were asked why, one response stood out by gender. Among STEM-intending Switchers, thirty-five percent of women said they did not believe they understood the ideas of Calculus I well enough to continue — compared to fourteen percent of men. That's a statistically significant difference, with a p-value of zero point zero two six. Among STEM-interested Switchers, the figures were thirty-two percent of women versus twenty percent of men. Women, far more than men, are leaving with the explanation that they simply didn't understand the material. Here's where the paper makes its most important move. Ellis and colleagues don't just report that women say they don't understand. They go looking for whether that self-assessment reflects reality. And this is where the confidence finding arrives. To isolate high-ability students, the team defined mathematically capable as scoring at or above the national eighty-fifth percentile on standardized math tests. Among these students — the ones whose scores say they are well prepared — women began and finished Calculus I with lower self-reported mathematical confidence than men. All of these high-scoring students lost confidence over the semester, which is its own sobering finding about what introductory calculus does to people. But the gender gap didn't close during the course. It was there on day one and it was there at the end. Mathematical confidence in this study means students' own perception of their ability to understand and succeed in calculus. Mathematical ability is operationalized using actual standardized test scores. Two students at the eighty-fifth percentile, matched on preparation, matched on how their instructor taught — one of them walks in believing they can do this, and one of them doesn't. The one who doesn't is more likely to be the woman. And by the end of the term, that internal gap has persisted or widened, even as their actual performance profile remains comparable. The paper is careful about causality here. The link between lower confidence and higher switching rates is correlational in this dataset. Ellis, Fosdick, and Rasmussen cannot prove that the confidence deficit causes women to leave. But the pattern is consistent: women who leave cite lack of understanding more often than men, yet women with equivalent preparation show lower confidence, not lower ability. The paper names this directly: a lack of mathematical confidence, rather than a lack of mathematical ability, may be responsible for the higher departure rate of women. The study has real limits. Grade data were requested but only a small fraction of instructors submitted them, and variability in grading across institutions made grades unreliable as a measure of ability. The model accounts for institutional clustering through the random effect but can't fully separate every unmeasured characteristic of a specific department or professor. And because the confidence-departure link is correlational, the paper can't rule out some third factor driving both. The authors are upfront about all of this. But here's what makes the findings hard to set aside. The one point five times odds ratio holds after the most obvious alternative explanations are accounted for. The self-reported reason for leaving — not understanding the material — is given by women at more than twice the rate of men, yet prior research cited in the paper finds no gender difference in mathematical ability, and in some studies finds women outperform men. And the confidence gap appears specifically among the students who, by objective measure, should feel most prepared. The problem is not at the bottom of the ability distribution. It's happening in the middle and at the top. The workforce arithmetic that follows from this is direct. Ellis and colleagues estimate that if women persisted through Calculus I at the same rate as men, women would make up roughly thirty-seven percent of the STEM workforce instead of the current twenty-five percent. That seventy-five percent increase in the number of women entering STEM would also grow the incoming STEM workforce overall by about twenty percent. These numbers come from a single transition point — one course, one decision, the end of one semester. What that suggests about intervention is pointed. The study is not arguing for more remediation or earlier preparation programs, though those may help. The data point toward the course environment itself as the lever — specifically, environments that affect how students perceive their own understanding, separate from their actual understanding. Students can be objectively capable and still leave. The mechanism appears to operate through how they feel about their capability, not what their test scores say about it. That's what makes this more than a pipeline problem. It's a signal about what happens inside the room — in the specific texture of how a calculus course is taught, how students are made to feel about their place in it, and whether the gap between actual ability and felt ability is something an instructor, a department, or a curriculum can close. The research doesn't resolve that. But it tells you exactly where to look. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.