Most education systems ask students to specialise somewhere between fourteen and sixteen. The choice feels small at the time. It is presented as a preference — what do you enjoy, what are you good at — and students select accordingly.
What is rarely made explicit is that a handful of those choices are not symmetric. Dropping physics closes doors that dropping history does not, because physics is a prerequisite for engineering, most physical sciences, and a large share of technical degrees, and history is a prerequisite for almost nothing.
The gender pattern in that specific choice is one of the most persistent in education statistics.
The size of it
In systems with early specialisation, girls are consistently underrepresented in advanced physics, and the imbalance has proved resistant to decades of intervention. Reported ratios vary by country and cohort, but a figure in the region of one girl for every three or four students in advanced physics classes is common in English-speaking systems.
Chemistry and biology look entirely different. Biology is often majority female. Chemistry is close to balanced in many systems. Mathematics sits somewhere in between and has improved substantially over three decades.
That variation matters. It rules out any explanation that treats "science" as a single thing girls avoid. Something specific is happening to physics.
What does not explain it
Attainment does not. Girls who take physics perform comparably to boys who take it, and in several systems slightly better on average — which is what you would expect from a more heavily selected group.
Prior mathematical performance does not. Studies that follow students with equivalent maths attainment at fourteen still find divergent take-up at sixteen.
Stated interest in science does not fully explain it either. Surveys of thirteen-year-olds find much smaller gaps in reported interest in physics topics than the eventual enrolment gap.
Something happens between reported interest at thirteen and enrolment at sixteen.
What the evidence points to
Three factors have the most support, and they compound.
Perceived difficulty relative to self-assessment. Physics is widely perceived by students as the hardest science. Where two students hold identical attainment but different self-ratings — the calibration pattern discussed elsewhere in this section — the student who rates herself lower will require a larger interest margin to choose the subject she believes is hardest. The self-assessment gap does not need to be large to change the decision at the margin.
Occupational image. Studies asking students what a physicist does, and what kind of person does it, find remarkably narrow and dated pictures. Where students associate a subject with a specific type of person, choosing it involves a claim about identity as well as ability. Biology's occupational image includes medicine and veterinary work, which are visible, prestigious and heavily female-associated at entry. Physics's image includes almost nothing comparable.
Departmental culture and teacher supply. Schools vary enormously in the proportion of their girls who continue with physics, and that variation is not explained by intake. Schools with specialist physics teachers, with physics taught as a separate subject earlier, and with active encouragement at the point of choice, produce substantially higher take-up. This is one of the few levers with reasonably direct evidence behind it.
The single-sex finding
Girls in single-sex schools take advanced physics at notably higher rates than girls in mixed schools. This finding is robust in raw numbers and is frequently used to argue for single-sex education.
It is also heavily confounded. Single-sex schools in most systems are disproportionately selective, fee-paying, or both, and their intakes differ on prior attainment and family background. Studies that control for those factors find the effect shrinks considerably, though several find a residual.
The more defensible reading is that something about the environment — plausibly the absence of a subject's gender-coding being visible in the room every day — has an effect, and that the effect is smaller than the raw comparison implies.
What the choice actually costs
The pipeline arithmetic is unforgiving. If a quarter of advanced physics students are girls, then even perfect retention thereafter produces a quarter of engineering graduates, and any subsequent attrition makes it worse.
Almost every intervention aimed at increasing women in engineering operates downstream of a decision that has already been made at fifteen. This is why university outreach, however well designed, cannot fix the numbers. The population it is recruiting from was determined years earlier.
There is a second cost that gets less attention. Physics prerequisites gate a set of careers that are unusually well paid, unusually stable, and unusually resistant to automation. Closing that door at fifteen is a financial decision that no fifteen-year-old is told she is making.
What has actually shifted numbers
Evaluations of interventions are patchy, but a few patterns recur.
Making the choice less final helps — systems that allow later entry to physical sciences show smaller gaps, because the decision is no longer irreversible.
Direct personal encouragement at the point of decision has among the largest measured effects of any low-cost intervention. A teacher telling a specific student that she should take the subject changes take-up at rates that assemblies and posters do not approach.
Making prerequisites explicit helps. Many students genuinely do not know what physics is required for. Telling them is close to free and moves numbers in the studies that have tried it.
None of this is dramatic. It is a set of small institutional changes made at the specific moment when an asymmetric door is closing, which is roughly the opposite of how the problem is usually addressed.