The Office for Students (OfS) has published a second analysis of its 2025 sexual misconduct survey.
It slices the same 50k responses by subject, provider type, study location, parental education, disability type, and a long list of other variables the September headline release left untouched.
The temptation – especially inside providers who have their own cut of the data – will often be to respond defensively, and to mentally assess whether the PDFs that have been published will pass muster.
But beyond a descriptive prevalence picture, the analysis points at which combinations of student intake and educational design produce the heaviest concentrations of risk.
That ought to give providers a framework for thinking about their own risk profile – who they recruit, where their students actually spend their time, and which settings the provider has influence over even where it has less formal jurisdiction.
OfS’ descriptive analysis declines to draw conclusions from the patterns it surfaces. Providers and victims don’t have that luxury.
Subject patterns
Three subjects sit conspicuously above the 24.5 per cent sector average for sexual harassment – language and area studies at 42.4 per cent, veterinary sciences at 41.3 per cent, and medicine and dentistry at 40.3 per cent. Computing, at 11.1 per cent, sits at the bottom of the range.
For sexual assault and violence, against a 14.1 per cent sector average, veterinary sciences reports 29.0 per cent, language and area studies 25.0 per cent, and medicine and dentistry 23.3 per cent.
The same subjects show the highest figures across multiple measures of risk – lower confidence in seeking support among medicine, dentistry and veterinary students, characteristically different perpetrator-identity profiles, and patterns that hold for both female and male students even where overall prevalence differs sharply by sex.
The sex × subject cross-tabulation is more revealing than the headline subject figures suggest. Physical sciences runs at 44.8 per cent for women and 15.7 per cent for men – a near three-fold gap within a single subject. Architecture, building and planning is even sharper – 38.7 per cent for women, 8.1 per cent for men. In most subjects, what looks like a “subject effect” is partly a sex-composition effect, with female-dominated subjects pulling the overall figure up.
Veterinary science and medicine and dentistry are the exceptions. They are the only subjects where male prevalence – 25 per cent and 23 per cent respectively – crosses the lowest female rates anywhere in the dataset.
That’s unusual, and matters – it suggests the structural exposure in those subjects affects men too, in a way the compositional sex-effects in physical sciences or architecture don’t. It is the first hint, in the data itself, that something is happening in the medicine and vet experience that’s different from the high-prevalence subjects elsewhere.
Beyond the campus
This policy area having moved slowly from “this is all stranger danger” to “no, a lot of it will be other students” in the 2000s, one figure to watch is the share of harassment incidents allegedly involving “another student”. Sector-wide, the share is high – but this release shows the share collapses in some contexts.
In small Level 4 or 5 providers, 88.8 per cent of harassment incidents allegedly involve another student – almost everything is internal. The share drops to 50.6 per cent for harassment and 43.4 per cent for assault in medicine and dentistry, and to 49.2 per cent and 37.5 per cent respectively in veterinary medicine. For specialist creative providers – drama schools, conservatoires – the share is 46.6 per cent for harassment, with “Someone else” accounting for 41.6 per cent.
The OfS gloss is cautious – this “may indicate more incidents involving people outside the provider in these subjects”. The translation is likely that the perpetrator pool extends well beyond the institution’s disciplinary reach in exactly the contexts that folk like the British Medical Association (BMA), Surviving in Scrubs, the British Actors Network and the Medical Schools Council have all been warning about for several years.
Meanwhile, high-tariff providers report 34.9 per cent harassment and 20.7 per cent assault prevalence – the highest of any provider type. Medium tariff sits at 26.5 per cent and 15 per cent, specialist creative at 25.1 per cent and 15.7 per cent, while large Level 4 or 5 providers sit at 9.4 per cent and 5.1 per cent. This is a roughly four-fold gradient on the harassment measure, and four-fold again on assault.
It corresponds to a cluster of other findings pointing in the same direction. Students with at least one parent holding a higher education qualification report 31.5 per cent harassment compared with 23.1 per cent for those without. Students studying not local to home address report 29.2 per cent. The September release had already shown that students from the least deprived IMD quintile and those not eligible for free school meals reported elevated rates. The signal is consistent.
Student characteristics
The September report had only disability status as a binary. This release adds type, with students reporting a mental health condition showing 42.2 per cent harassment and 27.5 per cent assault, and those reporting multiple or other impairments showing 38.4 per cent harassment.
These students also report the worst experience of the formal reporting process and the lowest confidence in where to seek support – the inverse of what an effective system should produce.
Female lesbian, gay or bisexual students report 55.9 per cent harassment and 35.4 per cent sexual assault/violence, while male LGB students report 27.3 per cent and 18.2 per cent.
The pattern holds within both sexes – LGB students report higher prevalence than their heterosexual peers across the board, though the LGB-vs-heterosexual gap is somewhat smaller for men than for women. None of this fits either the placement-context dimension or the compositional-recruitment dimension. It is its own category of risk.
On reporting, the subject-level rates may cut against intuition. The highest formal reporting rates after harassment are among business and management students (21.9 per cent), creative arts and design (18 per cent), media, journalism and communications (17.1 per cent) and computing (16.6 per cent). The lowest are architecture, building and planning (5.2 per cent), geography, earth and environmental studies (7.9 per cent), physical sciences (8 per cent) and psychology (9.1 per cent).
Architecture is the most pointed entry on either list – mid-range prevalence at 27.6 per cent, but the lowest reporting rate in the dataset. The students who appear most at risk in a given subject are not, as a group, the students who appear most willing or able to use the formal reporting channel. That is a textbook E6 compliance signal, and one that institutional data – when it eventually publishes – should be expected to surface.
It does fit a wider theory about programmes that prepare students for professions where contacts, reputation and a need to not “rock the boat” have surfaced in previous deep dives.
The weighting question
The most common methodological concern about the survey is that an optional module on sexual misconduct, attached to the National Student Survey, will draw disproportionate response from students with experience of the topic. And it does remain the case that as a result, there’s a real risk that the method isn’t capturing the particular risks associated with postgraduate study, or those who leave the provider before the NSS email appears in Y3.
Response rates varied substantially by group – female respondents at 14.2 per cent against male at 9.3 per cent, disabled students at 15.8 per cent against non-disabled at 11.2 per cent, LGB students at 20.5 per cent against heterosexual at 11.1 per cent, and the least deprived IMD quintile at 15 per cent against the most deprived at 10.1 per cent. Groups with higher response rates were also groups with higher reported prevalence.
OfS modelled the issue. A logistic regression using 14 variables – including subject, student typology, provider region, disability type, sex, age, ethnicity, sexual orientation, IMD, free school meals, religion and several others – produced a response propensity for every individual in the eligible population, with weights assigned as the reciprocal. This corrected for the demographic component of self-selection.
The headline 24.5 per cent and 14.1 per cent figures are post-weighting, and the report notes that the weighting reduced prevalence relative to the raw data. So provider type and subject distribution have been brought into representativeness with the sector.
OfS is also clear that the model “explained only a little of the overall response propensity of individuals” because some drivers of response – attitude to surveys, interest in the topic, available time – aren’t captured by demographic variables. The weighting fixes the observable component of self-selection, but it can’t fix the within-cell motivational component.
Among similarly-characterised students, those who have experienced harassment may be more likely to complete the optional module than those who haven’t, and there’s no statistical correction for that without measuring what you’re trying to measure.
Nevertheless, the directional findings – medicine and dentistry, veterinary sciences, language and area studies, and specialist creative providers all sit at the top across prevalence, perpetrator-identity and support-confidence – survive the critique because that cross-measure consistency would be hard to produce by sampling noise alone. The precise gap between, say, 41.3 per cent and 40.3 per cent, not so much.
Familiar ground
The BMA’s report on medical students found 20 per cent of female medical students harassed or assaulted on clinical placement, with 69 per cent experiencing sexism on placements. The perpetrators identified included senior doctors, consultants, NHS staff and patients.
The Medical Schools Council acknowledged “fragmented reporting systems across different providers”, and only five medical schools have signed NHS England’s Sexual Safety Charter. This report’s perpetrator-identity findings for medicine and dentistry don’t just corroborate – they are the same pattern at national scale.
The running concern about specialist creative providers tells the same story. A few years ago the British Actors Network collected over 300 testimonies from drama school students, and allegations surfaced at the Guildford School of Acting, ALRA, East 15, the Royal Welsh College, the London School of Dramatic Art and elsewhere. Equity called for independent regulatory oversight.
Specialist creative is the only provider type in the May 2026 data where “another student” drops below 50 per cent as the harassment perpetrator, and “Someone else” accounts for over 40 per cent. Both numbers are unique to that provider type, and they are the structural pattern drama schools have been warned about previously.
Then there’s the Higher Education Policy Institute’s (HEPI) 2021 data on privately-educated young men – consistently lower confidence on consent and harassment, more accepting attitudes towards rigid gender norms, higher use of casual sex apps and pornography, and lower contraception use. A fair hypothesis was that universities recruiting heavily from the independent school sector would be carrying compositional risk into their student bodies.
The OfS data can’t test that directly, but it can be tested through proxies, and on at least five – provider tariff, parental higher education, free school meals eligibility, IMD quintile and subject choice – the data lines up. Students from privileged backgrounds, at high-tariff residential providers, studying high-private-school-recruitment subjects, report the highest prevalence. That is not the relationship most vulnerability indicators show.
The intersection that matters
Step back, then, from the individual findings and the underlying point is straightforward. Harassment and sexual misconduct are perpetrated by people, not by universities. What the data shows is that the perpetrator pool a provider faces is shaped by two things at once.
There’s a compositional aspect – who the institution recruits, what attitudes and behaviours arrive with them, and what institutional culture they encounter when they get here. Alongside that there’s a structural dimension – whether the educational design exposes those students to people who are not the institution’s staff or students, but over whose conduct the institution still has some influence.
Each dimension on its own produces some elevated risk. Where they compound, the prevalence peaks appear to turn up.
Medicine and Dentistry sits right in the middle of that venn, with a high independent-school intake meeting heavy clinical placement involvement with consultants, NHS staff and patients. Veterinary sciences sits there too – an even higher independent-school intake than medicine, and placements in practices, on farms, and in equine and zoo work.
Language and area studies has a strong private-school skew at high-tariff providers in modern languages and classics, alongside year abroad placements where the host institution and the support apparatus are someone else’s. Specialist creative providers sit in the intersection more than anywhere else – independent-school skew, intense industry exposure, freelance staff, alumni, and visiting practitioners all in play.
Subjects allied to medicine sit in only the second circle – heavy placement exposure, but a much more diverse intake. Their prevalence is real but lower. Computing sits in neither circle and produces the lowest prevalence in the sector. The framework predicts where the peaks are, and the peaks turn up there.
In other words, national analysis like this lets providers act on the risk profile more specifically than they could a year ago. Two operational questions follow.
One concerns who the institution recruits, and what is known about the populations it brings in. Prior education, attitudes formed before arrival, the cohorts where consent literacy is weakest – none of these are determinative, but they shape the perpetrator pool.
That has implications for induction, early-year teaching, residential and social settings, mentoring and pastoral capacity in the first weeks of arrival, and the way professional boundaries are framed from day one. It is not the same exercise at every provider, or in every subject – the new report effectively argues against a uniform approach.
The other concerns which settings the institution’s students are actually in – on its premises, in placements, year abroad, residencies, working with freelance practitioners, exposed to alumni and industry networks. Over which of those does the institution have influence, even where it does not have jurisdiction?
E6 already uses the language of “significant and credible difference”, which assumes influence rather than ownership. The May 2026 data identifies, with reasonable precision, where that influence needs to be most actively extended.
Neither question is straightforwardly answered by the standard compliance E6 toolkit – training modules, policies, reporting routes – applied uniformly across an institution. They are answered by an institution doing its own risk assessment of its own intake, its own subjects, its own placement structures and its own provider context, and configuring its response accordingly.
What’s next
OfS says it will run the survey again within the National Student Survey in 2027, giving the sector a tracking measure – with two data points to compare against the regulatory framework that came into force on 1 August 2025.
OfS also “intends to publish” institutional-level data from the 2025 and 2027 surveys together”. Former CEO Susan Lapworth’s 2022 framing – that high prevalence and low reporting at a particular institution would be cause for OfS to look more closely – becomes operational when that happens.
The phrase “intends to publish” is likely partly regulatory craft as well as transparency commitment – section 22 of the Freedom of Information Act lets public authorities decline requests for information they reasonably intend to publish in due course, which means provider-level prevalence data is now effectively sheltered from being turned into a league table between today and the 2027 release.
But publishing two data points together rather than one also creates a comparative frame by design. Providers will be expected to have something to show.
That aligns the publication moment quite tightly with the question of whether providers have done the risk-based analysis – and then acted on it – that this report invites.
