The standard image is of a pipeline with women leaking out along its length. It is a useful shorthand and a misleading one, because it implies the losses are distributed.
They are not. When you plot representation at each stage from early secondary school through to senior technical roles, the line is not a gentle slope. It has two cliffs and a lot of flat ground between them.
Cliff one: subject selection at fourteen to sixteen
The first and largest discontinuity happens at specialisation.
Up to the point where subjects become optional, participation is by definition equal — everyone takes science. At the moment choice is introduced, advanced physics and computing take-up drops sharply among girls while biology and chemistry do not.
This single transition accounts for a larger share of the eventual shortfall in engineering and computing than every subsequent stage combined. If a quarter of advanced physics students are girls, no downstream intervention can produce more than a quarter of engineering graduates.
The arithmetic here is the reason so much well-designed university outreach produces disappointing numbers. It is recruiting from a population that was fixed years earlier.
The flat stretch
Between subject choice and the first years of employment, the picture is considerably better than the pipeline metaphor suggests.
Girls who take advanced physics perform comparably to boys. Women who enter engineering degrees complete them at rates similar to or above men's in most institutions — degree completion is one of the few places where the raw numbers favour women. Entry into first technical roles is roughly proportional to graduation.
This matters because it locates the problem. Retention through education is not the primary failure. The commonly repeated claim that women drop out of STEM degrees at higher rates is not well supported in recent data from most systems.
Cliff two: the middle career years
The second discontinuity appears roughly five to twelve years into a technical career, and it is steep.
Studies tracking engineering and computing graduates find that women leave technical roles at higher rates than men in this window, and that a substantial share move into adjacent non-technical work — project management, sales engineering, technical marketing — rather than out of the workforce.
The stated reasons in exit surveys cluster more consistently than one might expect. Workplace climate and lack of advancement appear far more often than work-life conflict, which is the explanation usually assumed. Several large surveys have found that women leaving engineering cite the culture of their organisation and the absence of a route upward more frequently than family reasons.
This is a significant finding because it contradicts the dominant policy response, which has been overwhelmingly focused on flexibility and childcare. Those matter. They do not appear to be the main driver of this particular cliff.
Why the distinction between the cliffs matters
The two discontinuities require entirely different responses, and conflating them has produced a lot of misdirected effort.
Cliff one is about fourteen-year-olds, subject availability, teacher encouragement and what a physicist appears to be. It is addressed in schools, and mostly by people who are not in the technology industry.
Cliff two is about promotion criteria, allocation of technical work, credit attribution and management behaviour in specific firms. It is addressed inside organisations, and no amount of school outreach touches it.
An industry that funds coding camps for eleven-year-olds while losing its thirty-two-year-old engineers is addressing the wrong cliff, and it is doing so partly because the wrong cliff is more comfortable — it is somebody else's institution.
What "computing" hides
Computing deserves separate mention because its trajectory is unique among technical fields.
Women's share of computer science degrees in the United States rose through the 1970s and peaked in the mid-1980s at a level substantially above where it stands today. It then fell for two decades.
No other technical field shows this pattern. Medicine, law, biology and chemistry all rose steadily over the same period. Computing reversed.
This rules out explanations based on stable dispositions or on the intrinsic character of the work, because neither changed in 1984. Something about the field's social organisation, its entry expectations, or its cultural image did. That question is taken up separately in this section.
What the numbers do not tell you
Aggregate pipeline figures conceal enormous institutional variation. Individual universities differ by factors of two or more in the proportion of women in their engineering intakes. Individual firms differ similarly in technical retention.
That variation is the most useful information in the whole dataset, because it demonstrates that the outcomes are not fixed by the field. Some institutions are doing markedly better than others with comparable inputs, which means there is something to learn rather than merely something to lament.
The unglamorous conclusion is that the pipeline is not really a pipeline. It is a series of gates, most of them held open or shut by identifiable people making identifiable decisions, and two of those gates account for most of what happens.