"You cannot be what you cannot see" is one of the most repeated claims in this field. It is not wrong. It is considerably more conditional than the slogan allows, and the conditions determine whether a role model intervention helps, does nothing, or backfires.
The theoretical basis
Vicarious experience — observing a similar person succeed — is one of the established sources of self-efficacy. Its position on that list is second, behind direct mastery experience and ahead of verbal persuasion.
The word doing most of the work is similar. The theory has always specified that the effect depends on perceived similarity between observer and model, because the inference being drawn is "someone like me can do this."
A model who is not perceived as similar supports a different inference: "someone unlike me can do this," which conveys no information about the observer at all.
What the studies find
Exposure to female role models in technical fields produces measurable effects on stated interest, self-assessment and, in some designs, performance. Several field studies in university settings have found effects on persistence in technical courses.
The effects are generally modest. They are larger for participants who are earlier in their trajectory and less committed, which makes sense — someone already certain of her direction has little left to be influenced about.
And they are inconsistent. A number of well-conducted studies find no effect, and the variation between studies is large enough that any confident claim about the average effect should be treated sceptically.
The backfire condition
This is the finding that matters most for programme design, and it is routinely ignored.
Studies of upward social comparison find that exposure to an exceptionally successful model can lower rather than raise self-assessment when the gap between observer and model is perceived as unbridgeable.
The mechanism is comparison. A model whose achievement seems attainable provides an existence proof. A model whose achievement seems extraordinary provides evidence about what the field requires — and if the requirement appears to be extraordinariness, the reasonable inference for an ordinary competent student is that this field is not for her.
Related work on stereotypes of computing found that exposure to a woman who strongly matched the field's stereotype in other respects did not improve, and in some conditions worsened, women's stated interest.
Both findings point the same direction: the identity of the model is doing more work than her presence.
What this means for who to invite
The most-invited speakers are, predictably, the most eminent available. The evidence suggests this is often the wrong choice.
A woman five years into her career, from a comparable background, describing work that sounds difficult but ordinary, is likely to produce a stronger similarity effect than a distinguished professor describing a distinguished career.
The second-year student is a better model for the first-year student than the department head is. Peer and near-peer mentoring programmes have generally reported good results, and this is a plausible reason.
What the model should say
The content matters as much as the presence, and there is a consistent finding here.
Models who describe struggle, uncertainty and failure alongside success produce better outcomes than models who present a smooth narrative. A study in which scientists' stories of struggle were presented to students found improved performance among students who read about struggle, compared with those who read only about achievement.
This is the same mechanism as the ability-praise finding. A narrative of effortless success tells the observer that success requires being the kind of person for whom it is effortless.
The instruction that follows is specific and rarely given: tell them what went wrong, how long it took, and what you did not understand for the first two years.
Duration, again
Single exposures produce small and short-lived effects. Sustained relationships produce larger ones.
This is unsurprising and inconvenient, since sustained relationships are expensive and single talks are not. It also means that measuring the effect of a talk by the enthusiasm in the room afterwards is measuring the wrong thing at the wrong time.
The limit of the whole approach
Role model interventions address a perception. They do not address a structure.
If a student's obstacle is that her school does not offer advanced physics, no speaker resolves it. If a woman's obstacle is that her firm allocates the interesting technical work elsewhere, a mentor does not change the allocation.
The reason role model programmes are so popular relative to their evidence base is partly that they are cheap and visible, and partly that they locate the problem in the perceptions of the underrepresented group rather than in the arrangements of the institution.
The effect is real. It is a supplement to structural change and a poor substitute for it, and programmes that consist entirely of speakers should be honest with themselves about which of those they are.