Gender differences in academic subject choice, particularly in physics, computing and engineering, are large and persistent across many countries.
The interesting finding is that the divergence in attitudes and self-concept appears considerably earlier than the point at which subjects are chosen.
The performance question
Worth addressing first because it's frequently assumed to be the explanation.
Average differences in mathematics performance between girls and boys are small in most contemporary datasets, vary considerably between countries, and have narrowed substantially over time.
The variation between countries is itself informative — if differences were primarily biological, you wouldn't expect the gap to differ so much by national context, or to correlate with measures of gender equality.
What differs more consistently than performance is confidence. Studies have repeatedly found that girls report lower confidence in mathematics and science at equivalent levels of measured ability.
The gender-equality paradox
A finding that complicates simple accounts. Some research has reported that countries with greater gender equality show larger gender differences in choice of scientific fields.
Proposed explanations include greater freedom to follow individual preference where economic pressure is lower, and differences in relative academic strengths within individuals influencing choice.
The finding has been contested on methodological grounds, including questions about the measures used and how the relationship was constructed.
It's worth knowing about because it's frequently cited, and worth treating as unsettled.
Where attitudes form
Research examining when children develop gendered beliefs about ability has found these emerging surprisingly early.
Studies have found that by around six, children begin associating high intellectual ability with one gender more than the other, and that this correlates with children's interest in activities described as being for very clever people.
Which suggests the relevant period is primary school rather than adolescence, and that interventions at the point of subject choice are addressing something formed years earlier.
The mechanisms
Adult expectations. Research on teacher and parent assessment has found differences in how the same performance is attributed — effort versus ability — with implications for how children explain their own results.
Attributing success to effort and failure to lack of ability produces a different response to difficulty than the reverse.
Field-specific ability beliefs. Fields whose practitioners believe success requires innate brilliance rather than sustained effort show lower representation of women, in analyses across academic disciplines.
The proposed mechanism is that where a field is understood to require a quality you've been taught you don't have, you don't enter it.
Stereotype threat. The finding that reminding people of a negative stereotype about their group can affect performance. Widely cited, and replication has been mixed, with effects appearing smaller and more context-dependent than early work suggested. Worth holding loosely.
Representation. Visible practitioners affect whether a field seems available. Evidence on role model interventions is mixed and generally positive.
Culture within the field. Particularly relevant in computing, where a documented shift in gender composition occurred from the 1980s, coinciding with the field's cultural association with a particular identity.
The historical point is significant: computing was not always male-dominated to the current degree, which indicates that the current pattern isn't inherent to the subject.
What interventions have evidence
Early exposure. Before attitudes consolidate, in primary school rather than secondary.
Framing fields as learnable. Emphasising effort and practice rather than talent, which addresses the brilliance-belief mechanism directly.
Contextualising the subject. Research has found that presenting computing and engineering in terms of applications and social purpose affects interest, particularly where the default framing is abstract or technical.
Changing the environment rather than the individual. Studies of classroom and departmental culture have found effects on participation from changes in how spaces are decorated, how work is organised and how competition is structured.
Structured group work. Where roles rotate rather than defaulting.
The retention problem
A dimension frequently neglected. Attracting people into a field is a different problem from keeping them.
Attrition data in several technical fields shows losses at multiple points — during study, at entry to the profession, and at mid-career.
Mid-career attrition is particularly notable and coincides with the period when caregiving demands are highest, which points at working conditions rather than at interest.
Which suggests that investment in encouraging girls into technical subjects, without addressing why women leave technical careers, addresses one end of a pipeline that leaks at the other.
What parents can do
Since attitudes appear to form early and at home as much as at school.
Watch how you talk about your own abilities. A parent saying they were never any good at maths transmits a great deal, particularly to a child of the same sex.
Attribute results to effort and strategy rather than to talent. Praise for being clever is associated with worse responses to subsequent difficulty than praise for effort.
Notice which activities and toys you offer, and whether the distribution is deliberate.
And take an interest in technical subjects yourself, visibly. Children calibrate what is for them substantially by what the adults around them treat as normal.