Overrepresented compared to whom?
Checking a common claim about higher education using only public census data
The claim
Official statistics report that women and most ethnic minority groups are a larger share of university students than of the general population. Women were 56.6% of first-degree undergraduates in 2021/22 (Advance HE). White pupils have the lowest higher-education entry rate of any broad ethnic group (gov.uk ethnicity facts and figures, with more in the House of Commons Library briefing). In the press this becomes one line: women and non-white people are overrepresented in higher education.
The question
“Overrepresented” only means something once you finish the sentence: overrepresented compared to whom? That comparison group is the benchmark, and the benchmark decides the answer. Students are young, and the young population is more ethnically diverse than the old. Compare students against the whole population and you can produce “overrepresentation” out of nothing but age. The honest comparison is against people of the same age.
How we check it
We use one free public source: the England and Wales 2021 census. It records both the population and whether each person holds a degree-level qualification (Level 4 or above, which covers a degree and higher). For residents aged 22 to 24, we compare each group’s share of the population with their share of the degree-level holders, region by region. If a group is a larger share of the holders than its share in the population, they are overrepresented, and the gap says by how much.
Two choices define the comparison group. Age: 22 to 24, because most people on a standard degree have finished by 22 and the band still reflects recent cohorts. Residence: about one in ten residents that age arrived in the UK within the previous five years, one in five in London. Include them and the count covers everyone now living here who holds a degree, wherever they earned it. Exclude them and it covers only people who were already here when the decision to go to university was made. The claim is about British higher education, so the main figures exclude recent arrivals. Where including them would change the answer, we say so.
Getting the data
On the ONS Create a custom dataset tool, choose Usual residents and set Area type to Regions. Two downloads are needed.
First, add Highest level of qualification, Age, Sex and Ethnic group. Set Age to the categories that include 22 to 24 and Ethnic group to 6 categories. This gives the sex panel and the unadjusted ethnicity figures.
Second, repeat with Length of residence in the UK in place of Sex. This gives the same table split by how long each resident has been in the country.
They have to be separate because ONS suppresses cells small enough to identify individuals. Asking for all five variables at once thins the table until half the regions are withheld. Five variables clear where six do not, so sex and residence are collected in turn rather than together.
This covers England and Wales. Scotland and Northern Ireland ran separate censuses and can be incorporated but it is unnecessary for our purposes.
What it shows
The hollow dot is each group’s share of the population, the filled dot their share of degree-level holders, and the gap between them is the over or under-representation.
Gender 22-24 y olds
Women are a larger share of degree-level holders than their share of the population in every region, by about 5 to 6 percentage points. Men are the exact mirror. For example, in London the share of young women who are degree-holders is 56.7%, while their share in the population is 52.1, a positive gap of 4.6%.
The sex figures come from the first download and so cover all residents. Applying the residence filter moves the women’s gap from +5.8 to +6.1 points, which changes nothing.
Ethnicity 22-24 y olds
Non-white residents are a larger share of degree-level holders than their share of the population everywhere except London, from 2.8 points in the North to 5.1 in the Midlands, against 4.6 for England and Wales overall. London is different from the rest of the country in the ethnicity composition of the population (see more on this in my next post). While non-white young people range from 11.6% in the South to 21.8 in the Midlands, their share in London is 48.9%. Relative to this high population proportion, the share of degree-holders falls 1.1% below.
Counting recent arrivals too adds between 0.6 and 1.2 points, most of it in the Asian group, and moves London from -1.1 to roughly level. No group changes sign.
And across age groups
The same comparison, now with the population share as the line and each age band placed by how far it sits above or below it. The number by each dot is that band’s share of degree holders.
Take 22 to 24 year olds. Non-white Londoners are 47.7% of Level 4 holders against 48.8% of the population, a gap of -1.1: under-representation. Hover over the two younger positive dots to see the exact gaps.
But almost nobody has finished a three-year degree by 21. Only 20.2% of London 20 to 21 year olds hold anything at Level 4, against 56.3% three years later, and most of that 20% is an HNC, HND, foundation degree or higher apprenticeship rather than a first degree. Non-white Londoners lead on those routes, not on degrees.
The gaps for women ease toward the line with age and turn negative in all regions for the over-50s.
What to keep in mind
The census measures who holds a degree, not who is enrolled today, and at 22 to 24 some are still studying.
What we learned
For women, the claim holds without qualification. Women aged 22 to 24 are a larger share of degree-level holders than their share of the resident population in every region, by roughly 5 to 6 points, and it persists in weaker form in older age bands.
For ethnicity, it holds in direction but not uniformly. Among residents already here five years before the census, non-white people aged 22 to 24 are overrepresented everywhere except London, where they are slightly underrepresented. The gaps run 2.8 to 5.1 points and narrow in older cohorts. Recent arrivals lift these figures, but by less than one might worry, and in London by enough to close the gap.
Reproducible from public data. The tables are built from the 2021 Census of England and Wales through the ONS Create a custom dataset tool, using the selections above.
Doing it yourself. You do not need to be a programmer. With a current AI model (as of mid-2026) as your guide, install WSL on Windows, then Miniconda, Python and a few libraries; the model gives you the exact commands and fixes anything that breaks. Download the table from ONS, hand it to the model, and ask for a script that draws any visualization you want. On a Mac or Linux machine most of this is already in place.