There is a great deal of activity in this area and remarkably little evaluation. A large proportion of STEM outreach programmes report participation figures and satisfaction scores, and never measure whether anyone subsequently studied engineering.
Among the programmes that have been evaluated with reasonable rigour, a set of common features separates the effective from the decorative.
Feature one: duration and repetition
Single-event interventions — a day, a visit, a talk — show small or null effects on subsequent enrolment in almost every evaluation that has looked.
Sustained programmes running over months or years show substantially better results. This is consistent with the self-efficacy literature: mastery experience requires time, and a day is not enough to accumulate any.
The implication is uncomfortable for funders, because one-day events are cheap, photogenic and produce large participation numbers. Sustained programmes are expensive, reach fewer people, and are the ones that work.
Feature two: producing something that functions
Programmes built around completing a real artefact — a working device, a piece of running software, a structure that holds — outperform programmes built around demonstrations and exposure.
The mechanism is straightforward. A demonstration shows the participant what engineering looks like when someone else does it. A completed project provides direct evidence that she can do it, which is the strongest available input to expectation of future success.
Evaluations of project-based programmes consistently report larger effects on stated intention to pursue the field than exposure-based programmes.
Feature three: the transition points
Timing matters more than intensity in the evaluations. Programmes positioned immediately before a consequential decision — subject selection at fourteen or fifteen, university application, the first year of a degree — outperform programmes positioned at ages where no decision is pending.
There is a great deal of activity aimed at eight to eleven-year-olds. It is popular, it is easier to organise, and the evidence that it changes eventual enrolment is thin. Enthusiasm generated at nine has to survive five years and a subject-choice decision made under entirely different pressures.
Feature four: cohorts rather than individuals
Programmes that recruit and retain students as a group show better retention than those working with isolated individuals.
Undergraduate cohort programmes — where students in a technical degree are placed in a shared cohort with common courses, study space and structured contact — have been evaluated at several institutions and have reported substantial improvements in retention.
The proposed mechanism is that being one of a group changes the interpretation of difficulty. A student struggling alone in a class where she is the only woman has an available explanation that is about her category. The same student struggling alongside eleven others in the same position does not.
Feature five: money, plainly
Scholarship programmes with academic support attached have among the best-evidenced results in this literature, and it is worth being honest that some of what looks like a retention effect is a straightforward affordability effect.
Engineering degrees are long, intense and hard to combine with paid work. Students who leave for financial reasons are not leaving because of culture, and no amount of mentoring addresses it.
What does not work well
Role model events in isolation. The vicarious-experience effect is real but is the second-strongest input to self-efficacy rather than the first, and a single talk delivers a small dose of it. Programmes where a role model has sustained contact perform better than those where she appears once.
Marketing campaigns. Evaluations of broad awareness campaigns have generally found effects on stated attitudes and little on behaviour. Attitudes towards engineering among girls are already reasonably positive in most surveys; the gap is not primarily one of approval.
Programmes that emphasise how difficult and exceptional the field is. This framing is common in recruitment material and it is counterproductive for exactly the population being targeted — students who already rate their own ability conservatively.
The evaluation problem itself
The honest state of this literature is that it is weaker than it should be. Many published evaluations lack control groups, rely on self-reported intention rather than behaviour, and have short follow-up periods.
Self-selection is the persistent difficulty: students who volunteer for a two-year engineering programme were more likely to study engineering anyway, and separating the programme effect from the selection effect requires designs that most programmes cannot afford.
This should induce caution about all of the above. The features listed here are the ones that recur across the better studies, not established causal facts.
The most defensible summary is that programmes producing sustained mastery experience in a cohort, timed to a decision, with the financial barrier removed, do better than events. That is a description of an expensive thing, which is probably why so much of the field consists of the cheap thing instead.