You’ll never believe where I found the secret LEO data!
David Kernohan is Deputy Editor of Wonkhe
Tags
Good news, everyone. I’ve tracked down that top secret data on the link between university courses and average graduate earnings.
It’s not a place that many people look for anything, but the government has hidden it on a little-known website called Discover Uni.
With persistence, and application, and an almost cavalier disregard for the rule book, you can get this site to show you graduate earnings (both an overall average, and a typical range) after 15 months, three years, and five years for any course currently on offer at any provider in the UK.
It’s not universal, of course. Some courses haven’t been running long enough to generate enough data to show – others have produced too little data to be able to publish anything meaningful. The data that is available is presented alongside sector information on similar courses, and wider details of graduate destinations and perceptions. Discover Uni also has employment rates and non-completion rates for pretty much every course.
Also: it’s not for the faint of heart, but the source data for this site is available in a machine-readable format for you to do your own analysis (pro-tip, you’re going to have to roll your own identifiers by combining the base UKPRN, provider course code, and mode code).
There’s also a bunch of LEO data available from the Department for Education. Again, it is well hidden (you have to search for “LEO data”) but I’m not afraid to spill the beans here. You are probably most interested in the graduate and postgraduate outcomes (this is the most recent release) by provider – and a few clicks will lead you to a spreadsheet. Really they could have made more of an effort to hide it.
It’s a release that’s kept me busy playing with dashboards over the years – so here’s the most recent one of these. The charts at the bottom are probably of the most immediate interest to the casual reader: you can see a ranked list of providers by subject area showing earnings one, three, five, or ten years after graduation.
And because we know a lot of things other than your choice of subject or course can affect earnings, you can also look only at graduates with a particular ethnicity, sex, POLAR4 quintile, or prior attainment marker. So we can keep to a sensible number of graduates in each group at each provider, the resolution of the subject area is a bit lower in the chart right at the bottom.
It’s not just me that mucks about with LEO data. Famously, the Institute for Fiscal Studies has published numerous reports that use LEO and a range of other techniques to think about graduate earnings, subject, and provider. As with all such attempts, past data is not a perfect indicator of future earnings success – but it makes for fascinating reading.
All LEO data shows a median, upper quartile, and lower quartile. That’s some fancy language for the middle of the range of earnings, the seventy-five percent line, and the twenty-five per cent line. A handy rule of thumb is that the bottom 25 per cent of graduate earners earn less than the lower quartile, the top 25 per cent of graduates earn more than the upper quartile, and the remaining 50 per cent are between the two – with the median marking the mid-point (as many graduates earn more than the median as earn less).
If you had anonymised individualised data, calculating quartiles and a median would be the first thing you would do – given the amount of things that can have an impact on graduate earnings looking at just one graduate is almost useless. By publishing these boundaries instead of anonymised data DfE makes LEO easier to use and understand.
There’s no curation of the data other than the fact that DfE doesn’t publish data that would pertain to really small groups of people. It wouldn’t be statistically meaningful to do so.
I hope this has been useful to the “looking for growth” team, and I wish you every success with your campaign to ensure that Barnet FC can once again play in Barnet.
Shame that all the data is misunderstood and interpreted as if it proves causation. So the more transparent the data, the more it is misleading.
It’s not at course level though.
Yes. It is.
Still just as misleading as it still doesn’t prove causation. Arguably it is more misleading as people are more likely to think that it does prove 100% causation if it is course level data. The extent of causation can only be attempted to be proved if we knew exactly what job the graduate was doing, and the extent to which, if at all, they are any better at that job because they spent an extra three years studying.
Extract from article : – “Wish you knew what you might earn as a graduate from a given course at a given university?” A more correct statement would be : – “Wish you knew what previous graduates who took any given course at a given university are now earning ” And this subtle change makes all the difference. And for clarity you might then add : – Limitation statement : ” The job that the graduate is doing may have nothing to do with their degree. Even when there is a connection then the graduate may have been able to… Read more »
This is all ground we’ve covered on Wonkhe many times before. Honestly if people believe that there is “secret” LEO data that is being suppressed I think the nuances of causality and correlation are maybe best served in a different article.
Ah, but it is not a ‘nuance’ (definition : ‘a subtle difference in or shade of meaning, expression, or sound’). It is fundamental to the whole interpretation and perceived usefulness of LEO data and thus highly relevant to the thrust of your article about whether the LEO data is hidden.
LFG’s point is that the raw data should be made publicly available so that right-wing think tanks can use it to estimate the level of public subsidy attached to each course/type of student/provider to inform their campaigning for these courses/types of student/providers to be excluded from public funding, with (working-class) young people “learning a trade” instead.
They are not interested in nuance or whether earnings differences are driven by course quality, it is about cutting public subsidy to higher education.
Seems a bit of an ask to expect the government to casually disregard the Data Protection Act and GDPR when they could just find a Goodwin-esque fellow traveller somewhere in a mid-ranking pre-92 to do any data torture that would help their cause.
I have no idea what the aims of LFG are , but perhaps it would be fairer to say non academic instead of working class in your argument above.
They’re Dominic Cummings’ representatives on earth aren’t they?
https://unherd.com/2025/10/dominic-cummings-new-nerd-army/
I think my summary of who right-wing think tanks and politicians believe should “learn a trade” instead of having a chance to study a degree is fair given who would be affected by such a policy, the ethnic mix of those affected is worth clocking too…
What datasets are not available that we should actually be campaigning to have released?
There are no further LEO datasets relating to course level data that need to be released if you ask me. It isn’t particularly useful data anyway as far as I am concerned as it is far more likely to be misinterpreted than enlighten public discourse and improve decision-making.
But I am curiously waiting for the new version of the annual LEO data release that will have both Graduate and non-graduate pay broken down by prior academic attainment, but this is already in the pipeline as the DfE have agreed to do this.
There’s been a big climb down from the ‘campaigners’ today. They’ve accepted that LEO is published on Discover Uni. On Twitter they note: “The Government only publishes heavily curated and disclosure-controlled data to the public, giving parents and students access to only a select preset of filtered breakdowns.” It’s curious, isn’t it, that the tax records of graduates are curated? They’d accused sector bodies of wanting to surprise the release LEO data, again they walk that back noting: “Some in the sector, like the University Alliance, have denounced the idea that LEO data should be used to ‘provide further indicators… Read more »