How the provider index is built

What the index is for

One question: among the universities teaching law in England, which ones do more for the students they actually admit. Not which ones admit the best students, which is what a raw indicator mostly measures.

Edition 2026-1, built 12 October 2026, publishes 91 providers in 7 bands and carries 68 more with their figures and no score. The table is here. Each annual re-run is a dated edition with a change log, and the previous edition stays online.

That is why the scored metrics are, wherever the publisher provides one, the difference between a provider’s result and an intake-adjusted benchmark for that provider. The benchmark is the Office for Students’ own construction, not ours: it is what students with that provider’s intake characteristics achieve elsewhere. Ranking on the difference from it is a value-added measure, and it is the one thing no commercial table does.

The metrics and the default weights

One hundred points, four scored families. The default weights are an editorial judgement, which is why they are printed here rather than buried. Reweighting them yourself is in preparation, and nobody is charged for it until it exists.

Progression carries the most weight because it is the outcome a careers site exists to inform. It means being in managerial or professional employment, or in further study, fifteen months after completing. It is taken from the Graduate Outcomes survey and it is not an earnings measure. The published numeric threshold for it is 60.

The lower quartile of earnings is scored as well as the median because the downside matters more to a seventeen-year-old than an average does, and because the publisher releases the quartiles so the spread can be shown rather than hidden.

The default view is unweighted

The published default is every metric separately, with its own source, licence, cohort year, denominator and suppression status. The composite is a layer on top of that, and the weights above are only a starting point. If you care about earnings and not at all about selectivity, set competitiveness to zero and earnings to forty: the table will reorder and the method will still be the same method.

How the numbers are turned into points

1. Compare law with law. Each metric is standardised within the law-provider cohort only, never against a whole-university or all-subject distribution.
2. Clip the outliers. Standardised values are clipped at three standard deviations, so a single extreme provider cannot dominate a family.
3. Scale and weight. Each clipped metric is rescaled to a nought-to-one-hundred range and multiplied by its weight.
4. Use differences where the publisher gives one. The outcome and student-experience families are scored on the difference from the intake-adjusted benchmark. Earnings and the offer rate have no benchmark, so they are scored on the value.
5. Score on the weight that exists. Where a publisher withheld a figure, the provider is scored on the points it does have: the score is the points earned over the weight available, and every row prints how many of the hundred points its score actually stands on. A provider is never given a made-up value and never penalised for its publisher’s suppression rule, and a reader can see which scores are thin. A provider with fewer than 60 of the hundred points published is left out of the index altogether rather than scored on a fragment.

One row, worked all the way through

Anglia Ruskin University, UKPRN 10000291, is the row the regulator’s own file is usually checked against. Its law first-degree full-time progression indicator is 67.8% against an intake-adjusted benchmark of 67.8%, on a population of 170 qualifiers from 2018-19 to 2021-22, with a published numeric threshold of 60. The difference is -0.1 pts, and that difference — not the 67.8% — is what the index scores.

Across the 91 providers in the index with a published figure, that difference averages -0.13 percentage points with a standard deviation of 4.09, so Anglia Ruskin University sits 0.01 standard deviations from the cohort mean. Scaled onto nought to one hundred and multiplied by the sixteen points this metric carries, that earns 8.02 points of a possible 16. Repeat for every metric, divide by the 100 points available, and the provider scores 44.6 with a 95 per cent interval of 40.3 to 49.0, which puts it in Band 6. Its median law graduate earnings five years out are £33,300 from 95 graduates of the 2017/18 cohort.

Ties, and refusing to invent an order

The Office for Students publishes a standard error for every difference from benchmark, and confidence intervals at twelve levels from 75 to 99.7 per cent. Those standard errors propagate: a provider’s composite carries the root sum of squares of its scored metrics’ contributions, and the published score is never printed without its 95 per cent interval.

The bands are then built by reading the table in score order. The first provider opens a band; each provider after it joins that band while its interval still overlaps the interval of the provider that opened the band; the first provider whose interval does not overlap opens the next band. A band therefore means one thing precisely: not statistically distinguishable from the provider at the top of this band. Inside a band the rows are printed in score order so a name can be found, and that order is not a finding.

On the current files that produces 7 bands for 91 providers, the largest holding 30 of them. That is a feature. A table that separates two indistinguishable providers by a decimal point is telling you something the data does not support, and it is the single biggest difference between this method and the commercial ones.

Two things about those intervals that we would rather say than have found out. First, they are understated: the earnings and competitiveness families publish no standard error at all, so they contribute no uncertainty to the composite, and the interval is a floor on the uncertainty rather than a measurement of it. The true bands are wider than the ones printed, never narrower. Second, the February 2026 correction to the student outcomes file fixed a standard error for the difference from benchmark that had been too low — which means any significance test anyone ran on the earlier file was wrong in the direction of over-confidence. We use the corrected file, and that is why its version date is printed in the sources.

Missing data: named, never filled in

Nothing is ever imputed, modelled or estimated into a gap. A missing metric reads insufficient data, with the publisher’s own suppression code named on the face of it.

Earnings

The Department for Education suppresses provider medians as c, low or x. In the latest release that affects roughly twelve to fourteen per cent of law provider medians: 18 of 127 at one year after graduation, 15 of 124 at three years and 16 of 121 at five years.

Outcomes and student experience

The Office for Students marks suppressed cells [low] and [DPH]. Of the 430 law first-degree taught rows in the student outcomes file, 345 are unsuppressed, 80 are [low] and 5 are [DPH].

Scale

Rows are omitted entirely where a provider has fewer than two students in a combination, and [DPL] cells are not treated as zero. An absent row is not evidence of a small cohort.

Survey results

RR marks a provider that took part in the survey but did not reach the response-rate threshold — 50 per cent for the student survey, 30 per cent for the graduate survey. Those results are not published and are not reconstructed by us. Cambridge’s seven student-experience measures for law are all RR, which is why its score stands on 75 of the 100 points.

Admissions

UCAS publishes no suppression code of its own: where a count is small it simply leaves the offer-rate cell blank. Those blanks are named on the face of the figure as published blanks, not as UCAS codes, because that is what they are.

Three further exclusions are ours rather than a publisher’s, and are labelled as ours wherever they bite.

• A benchmark with no headroom. The Office for Students flags benchmarks of 95 per cent and above. A difference from a benchmark that high is not a measure of anything a provider did, so those cells are carried and shown but not scored.
• A provider the publisher split in two. The earnings release records The University of Law under two registered regions, South East for the earlier cohorts and London for the later ones, and its own total row is a per-region total rather than a provider-wide one. The two rows are different populations. We take the region row with the larger population, name the other on the face of the figure, and never add or average them.
• A provider UCAS and the regulator name differently. UCAS publishes its own institution code and no UKPRN anywhere, so the offer rate has to be matched to the rest by name. Matching is by name tokens, with a hand-checked list of the providers that do not match that way, and an ambiguous match stops the build rather than guessing. Where no match exists the offer rate is empty, not inferred.

A provider enters the index only with at least 100 full-time first-degree law entrants across the regulator’s four-cohort window, unsuppressed data in at least 3 of the 4 scored families, and at least 60 of the hundred default points published. On the current files that gives 91 English law providers of the 159 the four datasets name between them.

The other 68 are carried with every figure we hold for them and are not scored, each with the reason on its own row: 49 have no law row in the regulator’s outcome file at all, which is what being outside the English register looks like; 7 have a published law cohort below the threshold; 5 have a cohort size the regulator withheld, so we cannot test the threshold and will not assume it; 3 have too few families published; and 4 have too little of the method published to carry a score. That is a fact about the published statistics, not a judgement about the law school.

Coverage, stated rather than fudged

The full index is England only, because the Office for Students’ outcome and student-experience files cover English-registered providers only. A separate earnings and admissions view covers Scotland and Wales, where the earnings release holds 879 and 407 law rows respectively but no regulator indicators exist. Northern Irish providers appear on admissions alone, because the earnings release contains no Northern Ireland providers at all. No page here claims a UK-wide law ranking.

Vintage, and the thing every other table hides

Every earnings figure is stamped with its graduation cohort. In the current release, the five-years-after-graduation figure describes the 2017/18 graduates, the three-year figure the 2019/20 graduates and the one-year figure the 2021/22 graduates. A seventeen-year-old choosing a course in 2026 is being shown what happened to people who started university in 2014, and that sentence appears on the page beside the number rather than in a footnote.

The region-adjusted earnings columns in the same file are not used at all. At provider by law by year level, almost every adjusted value is suppressed, so no cost-of-living-adjusted law earnings comparison is possible from that source. If we ever publish one it will be our own estimate and will be labelled as one.

One population, one definition, one date

The Office for Students publishes three population types: taught, taught-or-registered, and partnership. The index uses the taught population throughout and never mixes them, because mixing them produces numbers that are silently wrong. The same discipline applies to UCAS, where an applicant-level column and an application-level column sit next to each other and summing across them double-counts, and where a cycle-level offer rate is never presented as an individual’s probability of an offer.

Every cell carries its publisher, publication, cohort year, denominator, rounding convention and suppression status. Earnings are rounded to £100 and graduate counts to 5 by the Department for Education; UCAS rounds its counts to 5.

When it changes, and when it does not

None of this data changes hourly. The earnings release is annual, in June, with the next edition expected in June 2027. The student outcomes file is annual with corrections, last updated 5 February 2026. The Teaching Excellence Framework file is periodic, last updated 24 February 2026. UCAS publishes after each cycle closes in late November, and the 2025 cycle is the latest set as at 12 October 2026 — both candidate addresses for the 2026 cycle returned a not-found error on that date, which is how we know rather than assume.

Each annual re-run is a dated, versioned edition with a published change log, and the previous edition stays online. The hourly refresh this site runs updates the copy of the figures at the edge; it does not change the statistics, and the pages carry both timestamps.

What the index deliberately leaves out, and why

• Entry tariff per law course. The only source for it that we could find is a dataset whose licence we could not read on the publisher’s own site, which refused every request we made. Until that licence is confirmed there is no tariff metric, and we will not build one from scraped prospectus pages.
• A research quality metric. The licence position on the national research assessment was not established, so it is not used. We will not assume an open licence because a dataset looks public.
• Teaching Excellence Framework ratings. The downloadable files are indicator data, not ratings, and we could not find a machine-readable list of the gold, silver and bronze awards on the regulator’s own pages. The index therefore contains no rating.
• Current-year national student survey files. Their download page carries no open licence statement at all, and the survey is run on behalf of four bodies through a contractor, so single ownership cannot be assumed. The first edition uses the framework file’s survey-derived measures instead. If the licence is confirmed in writing, the current-year data replaces them and the framework file becomes the back series.
• Any SQE provider pass rate. The regulator publishes none. Why that matters.
• Bar course providers. The Bar Standards Board’s provider-level report is the only provider outcome data a regulator publishes for English and Welsh professional legal training, and it carries no licence statement. Its site terms exclude the right to create derivative works, so a recomputed chart is outside them, not merely a copied table. It stays as attributed prose on the cost page until the Board answers a written request.
• Any competitor’s position, score or column. Named, described and criticised in our own words; never reproduced, re-ranked, cached or harvested.

What we have not been able to confirm

Three gaps are open, and they are listed here rather than left for someone else to find.

• The Office for Students attaches a condition to each data download that users must follow its guidelines for appropriate use, and earlier candidate addresses for that document returned a not-found error. Reading the two Get-the-data pages’ own markup resolves most of it: on both pages that phrase is itself the anchor text of a link to the Open Government Licence version 3.0 at the National Archives, and there is no separate guidelines document behind it. We do not treat that as conclusive — it may be a content-management error rather than a considered statement — but it is what the publisher’s pages actually do, and it points the same way as the licence. We have still written to ask, and if the answer changes anything in this method we will say so here with the date.
• The course-level dataset behind the government’s own course comparison site is published by a body whose website refused every request from our systems, so we have not read its licence on a primary page. Secondary sources describe it as openly licensed. That is not good enough to publish on, and we do not state its licence here.
• Scottish and Northern Irish qualification are separate regimes, not variants of the England and Wales route, and we have not yet done that research. Nothing on this site should be read as covering them.

How to disagree with this

1. Change the weights. The composite is reweightable by a subscriber, and the unweighted per-metric tables are the default, so you never have to accept our judgement to use the data.
2. Check the inputs. Every source above is linked and openly licensed, so you can download the same files and recompute. Subscribers on the plan that includes exports can take our tables as CSV, with the source, licence and refresh date attached to each.
3. Tell us it is wrong. Email enquiries@assizedata.co.uk. We answer in two working days. If a figure or a method here is wrong we will fix it and say that we have, with the date.

Assize Careers is not a regulator, professional body, university, law school or official careers service, and it is not part of or endorsed by one. Nothing on this site is legal advice.