The proposal that the Office for Students put out for consultation was that the teaching excellence framework (TEF) will become a rolling, annual, event.
Every year would see an arbitrary number of providers undergo the rigours (and hefty administration load) of a teaching excellence framework submission – with results released to general joy/despair at some point in the autumn.
The bad news for fans of medal-based higher education provider assessments is that – pending the outcome of the recent ask-the-sector exercise and another one coming in the summer – we won’t get the first crop of awards until 2028. And even then it’ll just be England.
No need to wait
Happily, a little-noticed release of data just before Christmas means that I can run my own UK Alterni-TEF. Despite the absence of TEF for the next two years, OfS still releases the underlying data each year – ostensibly to facilitate the update of the inevitable dashboard (though this year, the revised dashboard is still to come).
To be clear, this exercise is only tangentially related to what will emerge from the Office for Student’s latest consultation. I’ve very much drawn from the full history of TEF, along similar lines to my previous releases.
This version of the TEF is (of course) purely data driven. Here’s how it works.
- I stole the “flags” concept from the original TEF – one standard deviation above the benchmark on each indicator is a single flag[+], two would be a double flag[++] (below the benchmark gives me a single[-] or double[- -] negative flag). I turned these into flag scores for each sub award: [++] is 2, [- -] is minus 2 and so on. This part of the process was made much more enjoyable by the OfS decision to stop publishing standard deviations – I had to calculate them myself from the supplied (at 95%) confidence intervals.
- If there’s no data for a split metric at a provider, even for just one flag, I threw it out of that competition. If you can’t find your subject at your provider, this is why.
- For the Student Outcomes sub-award (covering continuation, completion, and progression) three or more positive flags ( or the flag score equivalent of [+++] or above) gives you a gold, three or more negative flags or equivalent gives you a bronze. Otherwise you’re on silver (there’s no “Needs Improvement” in this game, and a happy absence of regulatory consequences)
- For the Student Experience sub-award, the flag score equivalent of seven or more positive flags lands you a gold, seven or more negative gets you a bronze.
- Overall, if you get at least one gold (for either Outcomes of Experience) you are gold overall, but if you get at least one bronze you are bronze overall. Otherwise (or if you get one bronze and one gold) you get a silver.
- There’s different awards for full-time, part-time, and apprenticeship provision. In the old days you’d get your “dominant mode”, here you get a choice (though as above, if there’s no data on even one indicator, you don’t get an award).
There are multiple overall awards, one for each split metric. To be frank, though I have included overall awards to put in your prospectus (please do not put these awards into your prospectus) the split metrics awards are much more useful given the way in which people in providers actually use TEF to drive internal quality enhancement.
Because that’s kind of the point of doing this. I’ve said this before, but every time I’ve shown plots like this to people in a higher education provider the response is something along the lines of “ah, I know why that is”. There’s always a story of a particular cohort or group that had a bad time, and this is followed by an explanation as to how things are being (or most often, have been) put right.
Doing the splits
In previous years I’ve just done subject TEF, but there’s no reason not to expand what is available to cover the full range of split metrics that turn up in the data. Coverage includes:
- Overall
- Subject area (a variant on CAH level 2)
- ABCs quintile (the association between characteristics OfS dataset)
- Deprivation quintile (using the relevant variant of IMD)
- Sex
- Ethnicity
- Disability indicator
- Age on course commencement (in three buckets)
- Graduate outcomes quintile
- Level of study (first degree, other undergraduate, UG with PG components)
- Partnership type (taught in house, or subcontracted out)
The data as released also purports to contain data on domicile (I couldn’t get this working with my rules above) and “year” which refers to a year of data in each metric. To avoid confusion I haven’t shown these.
In each case there is different data (and thus different awards) for full time, part time, and apprenticeship provision. It’s worth noting that where there is no data, even for a single indicator, I have not shown that institution as having provision referring to that split. So if you are standing in your department of mathematics wondering why I am suggesting it doesn’t exist the answer is more than likely that there is missing data for one of your indicators.
Here, then, is a version that lets you compare groups of students within a single institution.
And a version that lets you look at a particular student subgroup for all providers.
For each, if you mouse over an entry in the list at the top, it shows a breakdown by indicator (continuation, completion, progression for student outcomes; six cuts of National Student Survey question group data for student experience) at the bottom. This allows you both to see how the indicator compares against the benchmark, view flag scores (the colours) by indicator, and see how many data points are used in each indicator (the grey bar, showing the denominator).
More detail
The Office for Students did briefly consider the idea of a “quality risk register” before it was scrapped in the latest round of changes. In essence, it would have pointed out particular groups of students where an indicator was lower than what was considered normal. To be honest, I didn’t think it would work at a sector level as well as at the level of the individual institution – and I liked the idea of including NSS-derived measures alongside the outcomes (B3) indicators.
So here’s an alternative view of the same data that allows you to view the underlying TEF indicators for every group (split) we get data for. There’s a filter for split type if you are interested in the differing experience of students across different deprivation quintiles, ethnicities, subjects, or whatever else – but the default view lets you view all sub-groups: a quality risk register of your very own.
Here the size of the blobs show the number of students whose data is included in each group, while the colour shows the TEF flag as a quick way for you to grasp the significance and direction of each finding.
Understanding the student experience in 2026
Data like this is the starting point for a series of difficult institutional conversations, made all the harder by two kinds of financial pressure: that faced by students themselves (and affecting the way they are able to engage with higher education) and that faced by providers (a lack of resources, and often staff, to provide supportive interventions).
There’s no easy way of squaring this circle – if there was, everyone would be doing it. The answers (if there are answers) are likely to be very localised and very individual, so the wide range of splits from institutional data available here will help focus efforts where they are most needed. Larger providers will likely be able to supplement this data with near-realtime learner analytics – for smaller and less established providers releases like this may well be the best available tool for the job.
More than ever, the sector needs to be supplementing this evidence with ideas and innovations developed by peers. While a great deal of public funding has historically been put into efforts to support the sharing of practice, in recent years (and with a renewed emphasis on competition under the Office for Students in England) the networks and collaborations that used to facilitate this have begun to atrophy.
Given that this is 2026, there will be a lot of interest in using large language models and similar AI-related technologies to support learning – and it is possible that there are areas of practice where interventions like this might be at use. As with any new technology the hype can often run ahead of the evidence – I’d suggest that Wonkhe’s own Secret Life of Students event (17 March, London) would be a great place to learn from peers and explore what is becoming the state of the art, in ways that do not endanger the benefits of learning and living at a human scale.
Meanwhile it is ironic that an approach developed as an inarguable data-driven way of spotting “poor quality provision” has data outputs that are so useful in driving enhancement but are completely inadequate for their stated purpose. Whatever comes out of this latest round of changes to the regulation of quality in England we have to hope that data like this continues to be made available to support the reflection that all providers need to be doing.