The Exam That Decides: What PISA 2025 Gets Right and Wrong About Girls and Mathematics

The latest PISA results show boys pulling ahead in mathematics and girls in reading, but new evidence suggests the mathematics gap is not a fixed fact about girls but a product of the environment in which they prepare for exams, writes Daniel Fernández Kranz.

Listen to this Article

The OECD published the results of PISA 2025 this month, and most of the attention has gone, understandably, to the fall in average scores and the role of AI. For example, Spain lost 23 points in reading and 16 in mathematics since 2022, roughly a school year of learning. But buried in the second chapter of the report is a finding that deserves as much attention, because it concerns not how much students learn but who gets to learn what.

Across OECD countries, 15-year-old girls now outscore boys in reading by 30 points, while boys outscore girls in mathematics by 13. In science the two groups are level. None of this is new. What is new is the direction of travel. Between 2022 and 2025 boys’ reading scores fell by 17 points on the OECD average and girls’ by 12, so the reading gap widened in girls’ favor by five points. In mathematics, girls fell by 11 points and boys by 7, so the mathematics gap widened in boys’ favor by four. Both subject gaps grew at once, in opposite directions. Whatever pushed scores down over these three years, it did not push boys and girls down equally.

The OECD report is also careful to show that averages hide the more important story. Boys’ scores are more spread out than girls’ in every subject. In mathematics, the gap at the bottom of the distribution is small and in most countries not statistically significant. At the top it is 24 points. One in ten boys reaches the highest proficiency levels in mathematics against one in sixteen girls, and there is not a single country among the 89 that took part where girls are more likely than boys to be top performers in mathematics. The reading picture is the mirror image, with boys over-represented among low performers: 37 percent of boys fail to reach basic reading proficiency against 25 percent of girls.

Why does the top of the mathematics distribution matter so much? Because that is where selection happens. Admission to engineering, medicine, economics and the quantitative degrees that lead to the best-paid careers is decided by performance on high-stakes exams, and the mathematics component of those exams is where girls lose ground. A well-known body of research, including work by Ghazala Azmat, Caterina Calsamiglia and Nagore Iriberri in Spain, shows that girls who do as well as or better than boys in ordinary classroom tests fall behind them when the same material is tested in an exam with real consequences. Boys, it seems, respond to stakes. Girls, it seems, respond to pressure.

The natural reading of these facts is that there is something about girls, whether confidence, risk aversion or an aversion to competition, that makes them underperform when it counts. My own recent work, with Caterina Calsamiglia of ICREA and the Barcelona Supercomputing Center, Yarine Fawaz of CEMFI and Junhee Lee of the Korean Ministry of Foreign Affairs, suggests a different reading. The gender gap in high-stakes mathematics is not a fixed feature of girls. It depends on the peer environment in which they prepare.

South Korea gave us an unusually clean way to test this. Under the country’s High School Equalization Policy, students in the big metropolitan districts were assigned to high schools by lottery, and the lottery did not distinguish between single-sex and coeducational schools. A girl in Seoul could end up in an all-girls school or a mixed one purely by chance. Those students then sat two national mathematics exams at very different stakes. In June of Grade 11 they took a national assessment with no consequences for their future. Seventeen months later, in November of Grade 12, they took the college entrance exam that decides admission to Korea’s selective universities, where only the top two to three percent of scorers reach the three elite institutions. We followed two cohorts, more than 370,000 exam records across 617 schools, and asked a simple question: does the gap between boys and girls in mathematics move between the low-stakes and the high-stakes exam, and does it move differently depending on whether the students are surrounded by the other sex?

In coeducational schools the answer is the one the literature would predict. The mathematics gap in favor of boys widens at the high-stakes exam. In single-sex schools the answer is the opposite. At the low-stakes exam, boys in all-boys schools are far ahead of girls in all-girls schools, more than five points on a scale with a standard deviation of twenty, a gap several times larger than in mixed schools. However, by the college entrance exam, the single-sex gap has shrunk to about two points. Girls in all-girls schools close three fifths of the distance to boys in all-boys schools precisely as the stakes rise, when everything we thought we knew said boys should pull further ahead.

Girls did not become more confident. They worked harder, and it paid off but it cost them too.

We spent a good part of the paper trying to see if this result would go away. It stuck. This narrowing of the gender gap from low- and high-stakes exam results is not explained by the share of female teachers, nor by the fact that most single-sex schools in Korea are private, nor by boys and girls choosing the science track at different rates in different kinds of schools. Within mixed schools, having more female peers or more female teachers does nothing of the sort either. What matters, our findings indicate, is not simply having more girls in the classroom but something about the all-girls school environment itself.  A complete absence of boys changes how girls approach the exam that decides their future.

A separate longitudinal survey of Korean adolescents lets us see what that something is. Compared with girls who were assigned to mixed schools, girls in all-girls schools steadily increase the time they devote to mathematics as the entrance exam approaches about two and a half extra hours a week by the final year, on top of a baseline of five. They study more on their own and they buy more private tutoring, and they do it for mathematics only, not for Korean or English. Boys show no such shift in either type of school. What girls do not do is change their minds about themselves. Their self-assessed ability in mathematics does not rise. Their self-esteem, their sense of efficacy and their educational and career aspirations do not move. Girls did not become more confident. They worked harder and paid for it: the same girls report markedly higher stress from grades and exams from the moment they enter high school, and by the final year their scores on a mental-health index, covering anxiety, disinterest and suicidal thoughts, are a third of a standard deviation worse than those of their peers in mixed schools.

The most parsimonious explanation is one that economists such as Roland Fryer of Harvard University, Leonardo Bursztyn of University of Chicago, and Alison Booth of Australian National University have been building for years. Visibly striving in a domain coded as male carries a social and well-being cost for adolescent girls, and that cost is higher when boys are watching. Remove the boys and the cost falls, so girls invest.

The point is not that all-girls schools are good for girls. It is more nuanced than that. Boys still gain more, on average, than girls do from single-sex schooling; the well-being cost we document is real; and Korea’s lottery is not a policy anyone is proposing to import. The point is that the mathematics gap at the moment of selection, the one PISA finds at the top of the distribution in every country, is not an inherent feature of girls’ mathematics performance. It is produced by the setting, and it responds to effort when the setting rewards rather than stigmatizes effort.

That has two implications for how we read PISA 2025. The first is that the widening mathematics gap is not a verdict on what girls can do. It is a measure of how far our classrooms still make it costly for a girl to be seen trying hard at mathematics. Closing that gap in coeducational schools, where almost all of our children are, means paying attention to the small social mechanics of the mathematics classroom: who is called on, who is praised for effort rather than talent, how competition is framed, and whether the students at the top are held up as models or as exceptions. The second implication is that the reading gap deserves the same seriousness. A third of boys in the OECD leave compulsory schooling unable to read at a basic level, and that share is rising. If the lesson from Korea is that gendered norms about what is worth striving for can be changed, it applies to boys and books as much as to girls and equations.

PISA will tell us again in three years whether the gaps have moved. Whether they move will depend less on the tests than on what we decide is normal for a 15-year-old to be seen working hard at.

 

© IE Insights.