UK Subject Rankings: How to Compare Courses Within a Discipline
How to compare courses inside a subject using official UK sources: what REF unit-of-assessment results do and do not say, what LEO subject breakdowns can add, and why unirank.uk itself publishes institution-level figures only.
unirank.uk does not publish course-level or department-level rankings. As recorded on 2026-10-02, its dataset contains 141 university records within a six-dimension framework, and 138 of those records have REF data; it contains no per-course, per-department or per-UoA score field (unirank.uk dataset, 2026-10-02). For subject comparison, its institution-level proxies are the number of UoAs an institution submitted to, its submitted research FTE, its overall REF GPA, its four-star share and its provider-level LEO outcomes. These can describe an institution-wide profile, but they cannot identify the performance of a specific course (unirank.uk methodology and dataset, last verified 2026-08-11 and compiled 2026-10-02). A defensible within-discipline comparison must therefore use the official discipline and subject breakdowns while keeping their different units and populations separate.
What the official subject-level unit is
In REF 2021, the official research comparison unit is the unit of assessment (UoA), not a course or a generic subject label. The REF 2021 guidance reports 157 UK HEIs submitting into 34 discipline-based UoAs, producing 1,878 submissions containing 76,132 full-time equivalent staff; 185,594 outputs and 6,781 impact case studies were submitted (REF 2021, “Guidance on REF 2021 results”). Each HEI’s submission in a particular UoA is called a submitting unit. The relevant discipline boundary is therefore the selected UoA, and an HEI is assessed separately only in the UoAs in which it submits.
The overall research outcome is a weighted profile. Outputs account for 60%, impact case studies for 25% and environment for 15% (REF 2021, “Guidance on REF 2021 results”). Outputs are assessed for originality, significance and rigour; impact case studies for reach and significance; and environment for vitality and sustainability (REF 2021, “Guidance on REF 2021 results”). Four stars is the highest band, described as world-leading quality in originality, significance and rigour (REF 2021, “Guidance on REF 2021 results”). The published overall result should therefore be read with those weights and component judgements, rather than as equally weighted observations.
REF’s result has an HEI–UoA level. unirank.uk then converts the REF 2021 overall GPA from a 0-4 scale to 0-100 and takes an FTE-weighted mean across every UoA in which the institution submitted (unirank.uk methodology, last verified 2026-08-11). That institutional Research Quality figure is not a subject score: it combines disciplines with very different output conventions and cannot show how the institution performed in the selected UoA.
How to read the REF 2021 results browser
The official REF 2021 Results browser requires a selection of higher education institution and unit of assessment, and its displayed results can be filtered (REF 2021 Results browser, 2021 results). Read each HEI–UoA result row as the selected HEI’s submission in the selected UoA—the “submitting unit” defined in the REF 2021 guidance. It is not the whole university, a department’s taught offering or an individual course. Its overall outcome reflects the weighted output, impact and environment profile; in the accompanying submissions data, FTE describes the research staff base of that submission, not its taught-student volume.
A real comparison should hold the UoA constant, compare the overall profiles for the HEIs in that UoA, and then inspect the component results. The official REF 2021 Results analysis hub supports this with submissions data; sector and main panel averages; quality by UK nation and region; output profiles; environment data; institutional average results; institutional shares of quality and volume; ranges within main panels; UOA summary data; and comparative data (REF 2021, “Results analysis”, 2021 results). These products allow a research submission to be read against its disciplinary and sector context without relabelling it as a course ranking.
Why outcomes must be compared within subject
LEO answers a different question. The Department for Education’s “LEO Graduate outcomes provider level data, tax year 2022/23” covers UK-domiciled first degree graduates from all types of higher education provider in Great Britain, with outcomes measured one, three and five years after graduation (Department for Education, LEO 2022/23). Those periods refer to the first, third and fifth full tax years after graduation. For tax year 2022/23, the respective graduating cohorts are academic years 2020/21, 2018/19 and 2016/17 (Department for Education, LEO 2022/23). The five-year figure consequently describes the 2016/17 cohort, not graduates observed in tax year 2022/23.
Within a subject, the spread of outcomes across providers changes the meaning of a provider’s result. The Department for Education identifies Philosophy and religious studies as having the greatest variation between providers in the proportion in sustained employment, further study or both, with proportions from 63.0% to 94.1% for the middle 90% of providers. Medicine and dentistry has the least variation, from 89.8% to 97.4% for the middle 90% of providers (Department for Education, LEO 2022/23). These are middle 90% provider ranges, not unrestricted minimum-to-maximum ranges. A provider’s percentage therefore needs to be read against the distribution for the same subject; the spread does not measure a specific course.
Across subjects, the dispersion itself differs markedly. For median graduate earnings, the Department for Education reports ranges of more than £32,000 for Computing, Law, Economics, and Business and management, between the lowest and highest median earnings of the middle 90% of providers. Excluding Celtic Studies, it reports ranges of less than or equal to £10,000 for Nursing and midwifery, Medicine and dentistry and Performing Arts (Department for Education, LEO 2022/23). These contrasts show that variation between providers is subject-dependent. The supported numerical comparison is therefore between each subject’s own provider distribution, not between raw percentages from different subject groupings as though those percentages shared the same provider distribution.
LEO subject breakdowns use HECIS subject groupings, not REF UoAs; the classifications should not be assumed to align (Department for Education, LEO 2022/23, and REF 2021 guidance). The coverage also excludes postgraduate and international-only routes (Department for Education, LEO 2022/23). It is therefore a broad population dataset, not a course table or an individual record.
Why these sources cannot rank a course
REF measures research, not teaching quality or graduate outcomes. It cannot serve as a proxy for the taught experience in a specific course: a world-leading research submission does not establish what that course will be like (REF 2021, “Guidance on REF 2021 results”, 2021 results). On unirank.uk, the UoA count measures the breadth of research coverage; submitted FTE measures the scale of the research base; the GPA and four-star share are institution-wide aggregates; and LEO outcomes are provider-level figures. None resolves the course-level questions raised by the title (unirank.uk methodology, last verified 2026-08-11).
The National Student Survey does not close that gap. The Office for Students provider-level dashboard displays NSS 2026 and NSS 2025 and allows visualisation by participating higher education provider (Office for Students, National Student Survey provider-level dashboard, NSS 2026 and NSS 2025). The survey concerns final-year undergraduates, but its published level is the provider, not the course or department. It therefore cannot be read as a rating of a particular course.
Source vintage must also remain visible. The unirank.uk methodology record, last verified 2026-08-11, combines NSS 2025, LEO 2022/23, REF 2021 and global rankings 2025/2026 editions; those are distinct editions and should not be presented as observations from a common year (unirank.uk methodology, 2026-08-11). The defensible endpoint is a transparent comparison within a named REF UoA and, separately, within a named LEO subject. A course-level or department-level conclusion would require evidence that the official datasets above do not contain.