The grade inflation mutant algorithm, 2026

The publication of the Workplace for College students’ annual data set on degree classifications and grade inflation was initially scheduled for October of final yr.

It was delayed till now to allow additional information checking – a pause that triggered many data-literate observers to invest that maybe the venerable and much-criticised OfS algorithm (which compares the classification of levels awarded to the proper yr that was 2010-11; controlling for age, entry {qualifications}, and topic of examine solely) could be in for an overhaul.

This algorithm has generated outcomes up to now that implies that greater than half of the classifications truly awarded to undergraduates have been “unexplained” – the present quantity is slightly below 40 per cent.

So, both 4 in ten levels awarded in UK larger training are problematic – or a really simplistic algorithm isn’t truly superb and wishes fixing.

Occam’s razor

So we thought OfS would take the additional weeks to rethink the algorithm. This has not occurred.

As a substitute, we get a extra nuanced tackle what’s seen on this assortment, which is price quoting in full:

The time period ‘unexplained’ on this context signifies that adjustments within the traits of the graduating cohort included in our modelling can not clarify statistically the adjustments in attainment over the interval.

We’re not in search of to know what different elements could be driving the noticed adjustments. We acknowledge that parts equivalent to enhancements in instructing high quality might account for them. Our modelling can not account for will increase in diploma awarding because of adjustments made in response to the pandemic. Neither can it account for entry necessities equivalent to efficiency in an audition or the submission of a portfolio, as entry {qualifications} are restricted to plain A-levels, BTECs and direct equivalents.

Equally, it can not account for adjustments in entry {qualifications} because of the teacher-assessed grading necessitated through the pandemic. Because of this, we additionally classify these adjustments as ‘unexplained’.

In studying this very welcome clarification it’s possible you’ll need to suppose again to November’s OfS intervention on these subjects. After investigating three suppliers (beginning in 2022) England’s regulator appeared to determine that the issue was diploma algorithms.

A level algorithm is the mechanism utilized by suppliers to calculate diploma classifications from a set of module marks achieved by a pupil throughout their undergraduate examine. This can be a notably British drawback – in most techniques globally a grade level common backed by a full transcript is way extra necessary than any classification provided.

Within the three investigations OfS performed it recognized two specific facet of diploma algorithms – awarding a pupil the very best end result from a number of algorithms, and discounting credit score with the bottom marks – that it was not sure have been appropriate with the necessities of registration situation B4 (which offers, partly with the “credibility” of levels awarded).

This was a brand new departure for a regulator that had beforehand been content material to make use of phrases like “unexplained” to solid suspicion on tutorial requirements extra typically. The truth that it discovered three suppliers liable to breaching B4 regardless of the absence of any present follow that may be in breach of B4 merely served as a sign that the sport has modified.

The toughest diploma

We get the standard information launch alongside the report. Right here’s a plot displaying the proportion level distinction between the precise grades awarded and the grades modelled by the algorithm (the so-called “unexplained” awards) – with the variety of graduates proven by the skinny gray strains. Filters will let you look simply at first-class honours or first and higher second levels, select the yr you have an interest in (the latest, 2023-24, is the default), and to decide on a minimal variety of graduates at a supplier for show (the default is 500).

Mousing over one in all these marks reveals, within the chart on the backside – the precise (orange) awards plotted alongside the modelled (blue) awards.

(Full screen)

Prime of the charts for “unexplained” first and higher second awards we discover Goldsmiths, East London, and Bradford. Aside from Goldsmiths’ all recorded a slight drop within the precise (noticed) award of undergraduate levels with these classifications annually.

Like many suppliers on the high finish of this chart, these establishments take satisfaction in serving under-represented and non-traditional candidates to larger training – and they’re superb at what they do. Goldsmiths’ is a big arts-focused establishment, with admissions decided by portfolio in lots of instances. East London and Bradford are vocationally-focused suppliers with robust employer hyperlinks, serving an area non-traditional inhabitants.

East London and Bradford award a far decrease proportion of first-class and higher second levels than – for instance – Durham, Tub, or Bristol. In any significant, student-facing, interpretation of this phenomenon it’s “simpler” to get a very good diploma at a selective supplier like that than at yet another targeted on serving the entire neighborhood. The toughest college to get a very good diploma at is Buckinghamshire New College – lower than half of those that accomplished their course in 2023-24 achieved a primary or higher second.

It’s maybe simpler to see this phenomenon on a scatter plot displaying each noticed and modelled awards.

(Full screen)

There’s a neat break up by supplier kind – each Russell Group college awards greater than 80 per cent of graduates a primary or higher second, whereas solely a handful (Tub, Loughborough, Lancaster, Arts, Goldsmiths’, Northumbria) do. Is that truthful?

Equity

The query for anybody involved with tutorial requirements is whether or not these supplier stage differentials are truthful. The OfS algorithm – as famous above – makes use of age, prior attainment, and topic as examine as explicatory elements. It’s price coping with every in flip.

• OfS reckons that college students with lower than stellar A ranges are much less prone to get good levels than these with AAA or above– so suppliers who recruit other forms of learner shall be penalised by the algorithm irrespective of how good they’re at treating non-traditional learners.
• Age doesn’t fairly work the way you may anticipate – mature college students are very barely extra prone to get a primary or an higher second than the standard 18 yr outdated entry cohort.
• And humanities or social sciences topics are judged to be tougher to get a primary in than bodily sciences: so when you have (say) an enormous legislation college and never many chemists you’ll battle with the output of this algorithm.

(Full screen)

I’d love to point out you the usual errors and p-values that supply reassurance on the standard of this info right here, however I perceive from OfS that there was a problem with calculating them accurately: the figures have now been faraway from the annex. The staff are happy that the coefficients are correct for what that’s price, but when you find yourself being investigated because of this information I might be asking some questions right here.

OfS has arrived at these insights by way of evaluation of earlier years of information – and it is a legitimate factor for them to have finished. The failure to foretell a lot of what has truly occurred suggests to me that different assumptions needs to be added to the mannequin. It used to have incapacity, ethnicity, intercourse, and TUNDRA – these have been axed from the mannequin in 2023 ostensibly as a result of they didn’t have a lot explanatory worth.

There’s a commendable readability within the technical annex that any hole between the mannequin and actuality is due to “a results of unobserved results between tutorial years that haven’t been accounted for and haven’t been included as explanatory variables within the mannequin”. It’s good to see language like that up high too, as a counterbalance to the moderately accusatory time period “unexplained”.

What of it?

We wrote in regards to the three investigations which have so far come about because of this information after we obtained the experiences printed final yr. What was notable from these judgements was that OfS didn’t discover any present proof of grade inflation at any of the three suppliers concerned, although at two of the three they did discover a historic concern about diploma algorithm that had been in place previous to the existence of the OfS and was addressed speedily when it turned obvious that it was inflicting issues.

I’m going to stay my neck out and say that there are prone to be no suppliers which can be finishing up deliberate and systematic grade inflation as a matter of coverage. If OfS feels that there are issues suppliers are innocently doing that will end in grades being lower than dependable what it must do is present causal and statistical proof that that is the case – and it’ll discover this simpler if it really works with suppliers within the spirit of enhancement and steady enchancment moderately than taking part in to the headlines.