Staff to student ratios explain far less about the student experience than we assume

I put the two sets of numbers side by side for around 160 universities and 32 subject areas to find out.

There are four different questions inside “does the ratio matter”, and they give four different answers.

Question one – same subject, different universities

Take law. Every university that teaches it appears in the data, and their ratios run from about 15 students per staff member to about 28. Now do the same for psychology, for history, for engineering, and so on for all 32 subject areas. Within each subject, does the university with more crowded classes get worse reviews than the university with roomier ones?

Essentially not at all. Five extra students per staff member is worth about a sixth of a percentage point on the teaching score. That is nothing. If you are choosing between two law degrees, the staffing ratio tells you next to nothing about how the students there rate the teaching.

Worth saying that this is the blurriest of the four comparisons, because the subject labels on the two data sets don’t quite line up. So it is fairer to say the effect is too small to see here than that it is definitely zero. Blurriness can hide an effect. It can’t invent one.

Question two – same university, different subjects

Now flip it round. Stay inside one university and compare its own subjects against each other. Its law school might run at 22:1 while its physics department runs at 11:1. Do its crowded departments get worse reviews than its roomy ones?

A bit, yes. Five extra students per staff member is worth about three quarters of a percentage point. Still small, but about four times bigger than the first answer, and this time it is consistent enough to be real rather than noise.

Question three – universities compared with each other

Now forget subjects altogether and treat each university as one number, the way a league table does. Does the university with 20 students per staff member get worse reviews than the one with 12?

About the same as question two – five extra students costs roughly three quarters of a percentage point. It is a real relationship but a small one.

Question four – comparing subjects nationally

Finally, ignore individual universities altogether. Work out the national average ratio for each of the 32 subject areas and the national average teaching score, and plot 32 dots on a chart.

Now the relationship looks enormous. Veterinary science averages about nine students per staff member, and 95 per cent of its students rate the teaching positively. Computing averages about twenty, and gets 82 per cent.

On this chart, five extra students per staff member looks like it costs more than two and a half percentage points – fifteen times the effect we found in question one.

Yebbut

It is tempting to take the big, dramatic number and run with it. You shouldn’t though, for two reasons.

The first is that 32 dots on a chart is a very small amount of evidence, and averaging away the messiness always makes a relationship look tidier than it is. This is a well-known trap – patterns between groups are routinely far stronger than the same pattern inside them.

The second reason is more important. When you compare veterinary science with computing, the staffing ratio is not the only thing that changed. You’ve also changed who gets in, how hard it was to get in, whether teaching happens in a clinic or a lecture theatre, how much equipment is involved, and how sure the students were about their choice when they applied.

The ratio is not causing the difference so much as labelling it. It’s a stand-in for “what kind of subject is this”.

You can see that by looking for the exceptions. Law runs at about 20 students per staff member – as crowded as computing – and gets 88 per cent on teaching. Engineering runs at under 15, much roomier, and gets 85 per cent. If the ratio were really driving the scores, that couldn’t happen.

So the closer you get to comparing like with like, the smaller the effect gets, until it more or less vanishes. Ratios only look powerful when you are comparing things that differ in a hundred other ways at the same time.

The odd bit

Crowded courses don’t score worse on everything, even at the level where we can see an effect. They score worse on contact with staff, on library and equipment access, and on whether the course felt intellectually exciting. But they score better on assessment and feedback – and pretty much the strongest relationship anywhere in the exercise is that courses with more students per staff member have clearer marking criteria.

That sounds weird until you think about it. If you are marking 400 essays with a team of markers, you can’t busk it. You need a written grid, agreed standards, and a process. If you are marking twelve, you can rely on judgement and a conversation.

The students on the big course get a clearer, more predictable system. The students on the small course get more of your attention.

Oh – and universities with high staff to student ratios are hardly a random group. They tend to be the ones with lower entry grades and less money to spend per student. And thus those universities have a well-known survey signature of their own, quite separate from staffing.

Compare whole universities without allowing for that and some of the correlations look astonishing. Allow for entry grades, spending per student, and how many students stay on the course, and almost all of it evaporates.

A university’s staffing ratio tells you very little extra once you already know what kind of university it is.

Change over time

There is one more way to ask the question. Instead of comparing universities with each other, compare each university with its own past. Between 2021 and 2025 some improved their staffing ratio and some let it slip. Did the ones that let it slip do worse on the survey?

On the face of it, yes, and strikingly so. Universities whose staffing improved gained about 3.6 percentage points on teaching between the 2023 and 2026 surveys. Universities whose staffing got worse gained about 2.6. The same gap appeared on all seven parts of the survey without a single exception.

But the students who filled in the 2023 survey had started their degrees in 2020. Their whole university experience was shaped by the pandemic, and scores that year were awful almost everywhere. By 2026 that had washed through and scores recovered – unevenly, because the universities hit hardest had the most ground to make up.

That alone produces a strong pattern with nothing to do with staffing. Allow for how low a university started, and most of the apparent staffing effect goes.

Measuring the ratio students actually lived through

At this point the measurement itself needs fixing, and the fix is obvious once you say it. An undergraduate student finishing in 2026 didn’t experience one year’s staffing ratio. They experienced three – their first year, their second, and their final year. Pinning their survey answers to a single year’s figure is sloppy.

So I rebuilt it as the average ratio each group of students actually studied under, across all the years of their own degree, and four years rather than three for Scottish universities where degrees are longer.

That immediately exposed a hole in my earlier check. I’d tried to avoid the pandemic problem by comparing the 2025 and 2026 surveys instead, both taken by students with no pandemic in their degree, and found nothing. But those two groups overlap by two of their three years. Their experienced ratios correlate at 0.995 and differ by an average of less than a tenth of a point. There was almost no difference for that test to detect. Finding nothing there proves nothing.

Rebuilt properly, part of the effect comes back. The clearest relationship is with library and equipment resources, and it holds up even after allowing for how low a university started in 2023. Learning opportunities holds up more weakly. Teaching is on the edge. Assessment, academic support, organisation and student voice don’t survive at all.

Which is a more believable result than the one I started with. Squeeze the staffing and the thing that visibly gives way is access to stuff – the books, the equipment, the space, the person you needed to ask. Not the quality of the teaching itself.

Where the numbers come from

There’s three sources. The staffing ratios are from the Guardian University Guide. The survey results are from the Office for Students, which runs the National Student Survey. The things I allowed for along the way – entry grades, spending per student, and how many students stay on the course – are the Guardian’s figures too.

Plonking them together is the weak spot. The two sets of numbers carve the world into subjects in different ways. Staffing is counted by where a university’s academic staff and money sit, the survey is counted by what students are signed up to study, and those are not the same thing. The Guardian already does a rough translation between them to build its subject tables, which is Matt Hiely-Rayner’s approximation rather than a measured fact, and I then did a second rough translation on top of his to get the 32 subject areas used here. Two subjects came out so awkwardly that I left them out altogether.

You can see the first translation in the raw data if you look. Cardiff reports exactly the same staffing ratio for accounting, for business and for marketing, and the same figure again for history and for all four of its nursing subjects. Those are not coincidences, they are one budget line being spread across several subjects.

So the subject labels are approximate on both sides, and that blurs the two comparisons that depend on them – questions one and two, and question one most of all, because that is where the real differences are smallest to begin with. It doesn’t touch the rest. Question three compares whole universities on a single figure each, and so does everything later in the piece about change over time, so no subject labels are involved at all. The one finding that survives to the end, on library and equipment resources, is in that second group.

It is worth being clear about which way the blurring cuts. It can hide an effect that is really there. It cannot invent one that isn’t. So I wouldn’t defend the exact size of any of these numbers, but the shape of the four answers is not something it could have produced.

A couple of other warnings. First, the way staffing ratios are calculated changed partway through the period we have data for, and the newest edition also treats franchised provision differently. Any comparison that straddles that break is doing something slightly dishonest. Two of the years students lived through are missing from the published data altogether, so even the rebuilt measure rests on two years of each degree rather than three.

Second, even where a relationship does show up, none of it proves cause. A worsening ratio usually means student numbers grew faster than staff numbers, and that tends to happen at universities under financial pressure, which are doing plenty of other things to the student experience at the same time.

What it adds up to

The staffing ratio is not a hidden dial that explains student satisfaction. Two universities teaching the same subject at very different ratios get remarkably similar reviews, and the dramatic-looking gaps between subjects are mostly the ratio taking credit for everything else that makes those subjects different.

Three narrower things do survive. Within a single university, the departments running hot score slightly worse, particularly on getting hold of staff and getting at resources. Where the ratio a cohort actually lived through got worse, their verdict on library and equipment resources got worse with it. And crowded courses really do have clearer marking criteria, which is the one relationship in the whole exercise that stays the same size no matter how you slice it.

That is a thinner set of conclusions than we started with. Every time the measurement got more careful the headline got smaller and more specific. That’s usually the sign that you are getting closer to something true rather than further from a good headline.