When people talk about “REF rankings,” they usually mean league tables built by newspapers, universities, or analytics firms from the REF 2021 results. The Research Excellence Framework (REF) itself does not publish a ranking. It publishes raw quality-profile data for every submission, and third parties turn that data into the tables you see reported. Understanding the difference matters if you’re a research administrator, department head, or academic trying to interpret what a REF ranking actually says about a subject area or institution.
This guide explains how REF 2021 results are structured by Unit of Assessment, how the underlying quality profiles work, and how the most common third-party ranking methods (GPA, research power, market share) are calculated from that data. It does not repeat specific institutional league-table positions, which change depending on methodology and go stale quickly — for those, consult the official results portal or a current league table directly.
The REF does not rank institutions. It publishes quality profiles.
REF 2021 covered 157 UK higher education institutions, which made submissions across 34 subject-based Units of Assessment (UoAs) — the REF’s subject categories, ranging from Clinical Medicine to Music, Drama, Dance, Performing Arts, Film and Screen Studies. The full results are published on the official REF 2021 results portal, which lets you filter by institution or by UoA and cross-filter both.
For each submission, the portal shows an overall quality profile: the proportion of the submission’s research activity judged to fall into each of four starred quality levels, expressed in 1% steps. That’s it — there is no single number, score, or rank position published by REF itself. A submission’s profile might read, for example, 35% at 4*, 45% at 3*, 18% at 2*, 2% at 1*, 0% unclassified. Two submissions with very different profile shapes can look similar or different depending on which single number you choose to summarize them with — which is exactly why third-party rankings diverge from each other (see below).
How a quality profile is built: three weighted elements
Each REF submission is assessed across three elements, each producing its own sub-profile, which are then combined into the overall profile using fixed weightings set by the four UK funding bodies:
- Outputs — 60%. Journal articles, books, and other research outputs, assessed on originality, significance, and rigour.
- Impact — 25%. Case studies describing the demonstrable effect of research beyond academia, assessed on reach and significance. See REF impact case study for how these are structured and submitted.
- Environment — 15%. A statement and supporting data describing the research environment — strategy, people, income, infrastructure — assessed on vitality and sustainability.
Because Outputs carries 60% of the weight, a UoA’s overall profile is driven mainly by how its submitted outputs were scored, with Impact and Environment able to move the overall figure but rarely dominating it.
What the star levels mean
Every piece of assessed work — an output, an impact case study, an environment statement — receives a star rating from panel-based peer review (supplemented by citation data in some panels), using definitions set out in the official REF guidance:
- 4* — quality that is world-leading in terms of originality, significance and rigour (for impact: outstanding in terms of its reach and significance).
- 3* — quality that is internationally excellent, but falls short of the highest standards of excellence.
- 2* — quality that is recognised internationally.
- 1* — quality that is recognised nationally.
- Unclassified — quality that falls below the standard of nationally recognised work, or does not meet the published definition of research for the purposes of the REF.
A UoA’s quality profile is simply the percentage of its assessed activity that landed in each of these five bands, across all submitting institutions combined for a national picture, or per-institution for an individual submission.
How third parties turn profiles into rankings
Because REF doesn’t publish a single score, every media and analytics outlet that produces a REF ranking has to choose its own method for collapsing a five-band profile into one comparable number. The methods differ, which is why the same institution can appear in different positions across different published tables for the same subject. Times Higher Education, which publishes one of the most widely cited REF 2021 tables, has been explicit about its methodology, and it illustrates the three approaches most commonly used across the sector:
- GPA (Grade Point Average). The most common method. Each star band is given a point value (4* = 4, 3* = 3, 2* = 2, 1* = 1, unclassified = 0), and the GPA is calculated as (4×%4* + 3×%3* + 2×%2* + 1×%1*) ÷ 100, producing a score between 0 and 4. This is a straightforward quality-only measure: it says nothing about the size of the submission.
- Research power. GPA multiplied by the total full-time-equivalent (FTE) staff submitted to that UoA, then typically rescaled so the top-scoring submission reads 1,000. This rewards submissions that combine high quality with scale, and is often used as a rough proxy for an institution’s overall research funding potential in a subject area — a large submission with a slightly lower GPA can outrank a small, extremely high-quality one on this measure.
- Market share / QR-weighted measures. Some tables weight 4* far more heavily than 3* (commonly around 4x) and exclude 2*, 1*, and unclassified activity entirely, in an attempt to mirror how the funding bodies’ own quality-related (QR) funding formulas work — since QR funding is concentrated heavily on 4* and 3* activity.
None of these methods is the single official REF ranking — they are analytical choices layered on top of the same published profile data, and a research office comparing tables from different publishers should check which method each one uses before treating the resulting positions as directly comparable.
A caveat on small submissions
Where a UoA submission from a given institution involves roughly fewer than 4 FTE staff, the results portal suppresses or anonymises the detailed profile to avoid identifying individuals, which means GPA and similar scores can’t be calculated for those submissions — they typically show as a dash in published tables rather than a rank position.
Panel-level patterns
REF’s 34 UoAs are grouped under four main panels (A: life sciences and medicine; B: physical sciences and engineering; C: social sciences; D: arts and humanities), each overseeing a set of related subject panels and applying broadly consistent standards within its remit. According to Times Higher Education’s analysis of the REF 2021 results, mean GPA varied somewhat by main panel — Panel B (physical sciences and engineering) averaged around 3.33, Panel A (life sciences and medicine) around 3.29, Panel D (arts and humanities) around 3.2, and Panel C (social sciences) around 3.14. Differences of this size are more useful for understanding assessment patterns across broad disciplinary groupings than for comparing individual institutions, since panel-level averages smooth over a lot of UoA-to-UoA and institution-to-institution variation.
Nationally, REF 2021 also saw a large increase in the volume of activity assessed relative to REF 2014 — driven mainly by a shift to a near-universal staff-submission requirement in 2021, versus the more selective submission practice institutions used in 2014. The official REF guidance explicitly cautions against directly comparing REF 2021 outcomes to REF 2014 outcomes for this reason: a change in who gets submitted changes the profile independently of any change in underlying quality.
Why REF results matter beyond the league tables
REF outcomes are not just a reputational exercise. In England, Research England’s quality-related (QR) research funding — the block grant funding it distributes to institutions largely on the basis of REF results — rose to roughly £2 billion a year following the publication of REF 2021 results, according to Times Higher Education’s reporting on the funding settlement. The other three UK funding bodies (the Scottish Funding Council, Medr in Wales, and Northern Ireland’s Department for the Economy) use REF results similarly within their own funding formulas. This is the practical reason REF results get so much institutional attention: a UoA’s quality profile feeds directly into how much unrestricted research funding its institution receives for that subject area going forward.
Research offices and departments typically use their REF results in several concrete ways beyond funding formulas: benchmarking a department’s profile against comparator institutions in the same UoA, citing strong Impact or Environment scores in funding bids and recruitment, and identifying which elements (Outputs, Impact, Environment) most affected their overall profile to inform strategy ahead of the next cycle.
Looking ahead: REF 2029
The next assessment cycle, REF 2029, is already underway in terms of rule-setting, and several structural decisions have been confirmed or are still being finalised. If you’re preparing a submission strategy rather than interpreting 2021 results, see our guides on REF 2029 output eligibility and REF 2029 output portability rules for what’s confirmed so far about how the next cycle will differ from 2021.
Frequently asked questions
Does REF itself publish a ranking or league table?
No. The official REF 2021 results portal publishes quality profiles — the percentage of assessed activity at each star level — for every submission, filterable by institution and Unit of Assessment. Ranked league tables are constructed separately by media outlets, universities, and analytics firms using their own methodologies applied to that published profile data.
What is the difference between GPA and research power?
GPA (Grade Point Average) measures quality alone, calculated from the percentage of activity at each star level regardless of submission size. Research power multiplies GPA by the total FTE staff submitted, so it also reflects scale — a large submission can score highly on research power even with a moderate GPA, while a small, very high-quality submission may rank lower on research power than on GPA alone.
Can REF 2021 results be directly compared to REF 2014 results?
The official REF guidance advises against direct comparison. REF 2021 required near-universal submission of eligible staff, while REF 2014 allowed institutions to submit more selectively. That difference in submission practice alone can shift a UoA’s profile, independent of any real change in research quality.
What are Units of Assessment?
Units of Assessment (UoAs) are REF’s 34 subject-based categories that institutions submit into — for example, Clinical Medicine, Computer Science and Informatics, or Music, Drama, Dance, Performing Arts, Film and Screen Studies. UoAs are grouped under four main panels covering broad disciplinary areas.
Why do small departmental submissions sometimes show no score in published tables?
Submissions with roughly fewer than 4 FTE staff have suppressed or anonymised profiles on the official results portal, to avoid results being individually identifiable. This means GPA and other scores can’t be reliably calculated for them, so they typically appear as a dash rather than a ranked position in third-party tables.
