How the numbers work

Salary data

Graduate earnings at 1, 5 and 10 years after graduation come from the DfE Longitudinal Educational Outcomes (LEO) dataset, published annually by the Department for Education. LEO links HMRC tax records to graduate records, so it reports what graduates actually earned rather than what they told a survey they earned.

We use median earnings (50th percentile) for first-degree graduates, all employees, both sexes, weighted across CAH3 subject groups within each CAH2 category. Between those three points we interpolate in a straight line. Beyond year 10 — where LEO stops — every field grows at a flat 2% a year. That last assumption is uniform across subjects and is one of the cruder parts of the model.

The salary scenario selectormoves both routes to the same percentile, each using its own published dispersion: LEO's lower and upper quartiles for the degree, and ASHE's lower and upper quartiles (full-time, ages 22–29) for the matching occupation. Full-time rather than all-employee because ASHE reports annual pay, so part-time-heavy occupations otherwise show spreads that reflect hours worked rather than pay rates. Graduate earnings are markedly more dispersed than non-graduate earnings, so the lower-quartile case is worse for the degree than for the alternative — that asymmetry is in the source data, not an assumption we added.

The no-degree path is an estimate, not a published statistic.ASHE reports pay by occupation, not by whether the postholder holds a degree, so no official series measures “what this person would have earned without the degree.” Each field's figures are our estimate, anchored on the nearest ASHE occupational group. This is the least certain input in the model and the one the verdict is most sensitive to — the answer is driven mainly by the gap between the two curves, so a judgement here moves it more than any salary figure above. Treat the break-even year as a directional signal, not a forecast.

Caveat for humanities and social science fields:LEO reports earnings for everyone who graduated with a given degree, regardless of the job they went into. For subjects with diverse destinations — History, Sociology, Languages, Psychology, Politics — the median reflects a population scattered across law, finance, teaching, the civil service and much else. The figure is accurate for the typical graduate of that subject; it is not a prediction of earnings in a “history career.”

Cost assumptions

Tuition defaults to £9,790 a year, the England fee cap for 2026/27. From 2026 the cap rises annually with RPIX, so this figure changes every year and the model does not project future rises — it applies the current cap to every year of the course.

Maintenance defaults to £8,000 a year, which is deliberately below the maximum. Maintenance loans are means-tested on household income, and the 2026/27 maxima are £9,118 living at home, £10,830 living away outside London and £14,135 in London. The slider spans that full range; move it to match your circumstances.

Interest accrued during study is approximated on the average balance over the course, rather than modelled term by term.

Student loan repayment model

We model Plan 2 (English students who started 2012–2023), Plan 4 (Scottish students) and Plan 5 (starting September 2023 onwards). Plan 1 and postgraduate loans are not modelled. Welsh and Northern Irish students should read Plan 2 and Plan 5 as approximations — thresholds and write-off periods differ.

Each plan uses a single flat interest ratefor the whole term. Plan 2's real rate slides from RPI to RPI+3% with income; we approximate that with one mid-range rate rather than recalculating it each year, so Plan 2 balances will be somewhat wrong for very high and very low earners. Plan 4 and Plan 5 track RPI, which we hold at 3.2%.

Repayments are calculated on gross earnings above the threshold, as SLC rules require. Income tax and National Insurance are modelled separately, using 2025/26 thresholds and rates, and appear only in the take-home column of the year-by-year table — they never affect the repayment figure.

What “worth it (vs. no degree) after” means

Two versions of the same person leave school at 18. One goes to university; the other starts work immediately in the no-degree route named on that page. We accumulate each one's earnings year by year — the graduate's salary minus student loan repayments, against the non-graduate's salary — and report the year the graduate's running total overtakes the other. “Never” means it does not happen within 40 years.

The non-graduate starts earning several years earlier, so the graduate begins well behind. That head start, not the size of the loan, is what usually decides the answer: raising tuition from £9,250 to £9,790changes no field's break-even year at all.

Three things this comparison does not do. It uses gross earnings on both sides, so it ignores that the higher-earning path pays more tax — which flatters the degree slightly. It does not discount future money, so £1 in year 30 counts the same as £1 today. And it holds the repayment threshold fixed in nominal terms while salaries grow, which overstates repayments in later years if thresholds are uprated, and matches reality if they are frozen, as Plan 2 thresholds were from 2021.

How the verdict is decided

The GREEN / YELLOW / RED verdict is a rule, not a judgement, and it combines the break-even year with automation exposure and how the debt behaves:

• GREEN — breaks even within 8 years and AI-automation risk is 5 or below.
• RED — never breaks even, or takes more than 15 years, or AI risk is 8 or above.
• YELLOW — everything else.

The verdict deliberately ignores the loan balance.An earlier version went RED when the balance at year 20 exceeded 1.5× what was borrowed. That measured the opposite of what it intended: a high balance means low earnings and slow repayment, so those graduates hand over less in total. It flagged fields repaying about £67,500 on average while passing fields repaying about £107,600 — Teaching read RED while repaying the least of any subject (£44,000), and Medicine read GREEN while repaying the most (£194,000). The rule was removed. What the loan actually costs over its life is now reported as its own figure on every page instead, because it is a fact worth stating plainly rather than something to encode in a colour.

Note that a GREEN verdict does not mean the loan gets repaid. Only 4 of 27 subjects clear it before write-off on Plan 5; for most people it behaves as a 9% surcharge for the full term. Whether a degree beats not going, and whether the debt ever clears, are separate questions and the page answers both.

AI-automation risk

Risk scores run 1 (highly resilient) to 10 (high exposure) and reflect occupational automation exposure at the task level, not the job level. A score of 8 for Marketing doesn't mean marketing jobs disappear — it means the tasks that dominate entry-level marketing roles (first-draft copy, basic analysis, reporting) are ones current AI tools already perform competently. Pages display the inverse as a resilience score out of 10: 7–10 low risk, 4–6 medium, 1–3 high.

These scores are our qualitative judgement, informed by the sources below rather than computed from them. No published dataset scores automation exposure by degree subject, so this is the most subjective number on the site.

Confidence ratings (High / Medium / Low) reflect how stable we expect a score to be over the next 3–5 years, not certainty in the current estimate. High means regulatory or physical constraints make it unlikely to shift; Low means the entry-level task mix could look materially different within a single degree cycle.

The no-degree route

Each degree page links the National Careers Service profile for the occupation it compares against, and Find an apprenticeship. Both are government services; we are not affiliated with either and earn nothing from these links. Which occupation stands in for “no degree” in a given field is our editorial choice, and a debatable one in places.

Limitations

• Medians hide distributions. Law and finance are highly bimodal — the median is not representative of either the City track or the high-street track. Use the scenario selector to see the quartiles.
• The no-degree comparison is an estimate. It is the single input the verdict is most sensitive to, and no official statistic measures it.
• Nominal, undiscounted, gross. No inflation adjustment on the comparison, no discounting of future money, no tax modelled on either earnings path.
• Flat post-10-year growth. Every field grows at 2% a year beyond year 10, regardless of how steep its trajectory was up to that point.
• No career interruptions. The model assumes continuous employment from graduation. Part-time work, career breaks and unemployment lower actual repayments and extend the loan lifecycle.
• One institution, one outcome. LEO medians pool every university. Where you study changes earnings substantially and the model does not capture it.
• Money is not the only reason to study. Everything here is financial. It says nothing about whether a subject is worth studying for its own sake, or about non-financial returns to a degree.

Data vintage and refresh

Sources are on different cycles: graduate earnings are LEO 2023–24, occupational pay and its quartiles are ASHE 2025, and fees, maintenance and thresholds are 2026/27. LEO publishes each autumn and we update after each release; the fee cap now rises annually with RPIX and needs checking every year. AI risk scores are reviewed annually.

If you spot an error, or have a source that contradicts a figure here, please email data@degreeindex.co.uk.