What’s measured, and how far it agrees

Q1: region × industry output, against ONS’s own figures

In short:

  • Nothing else publishes output per job or per hour at this detail. ONS’s regional GVA reaches 47 industries at ITL3, but no published jobs or hours denominator matches it; dividing by BRES jobs is the only free route. Below ITL2 × 33 industries, though, the GVA is allocated rather than measured.
  • Whole places agree with ONS once the self-employed are counted. Our GVA per filled job is a median 0.96–0.98 of ONS’s across 169 ITL3s (2015–2023), with rank correlations of 0.98–0.99. The GVA is the same; the gap is in job counts.
  • Sectors within a region agree; regions within a sector agree less. Against ONS’s lapsed INDBYREG, industries’ ranking within a region matches at 0.96 (median), regions’ ranking within an industry at only 0.83, and as low as 0.53.
  • Regional GVA departs from the Annual Business Survey by more than the survey’s sampling error in 24% of region × industry cells (2012–22), where 5% would be expected.
  • GVA per job isn’t simply pay, at region level. The plan expected regional GVA per job to track pay where labour’s share of income is high. It doesn’t: the labour share doesn’t predict which industries track pay.
  • ONS’s own quality scores for the estimates that set ITL3 detail collapsed from 2022. The share of cells scored low rose from about a fifth to 47% in 2023 and 86% in 2024 (provisional).

This page covers the plan’s Q1: how regional GVA linked to BRES compares with other ways of measuring output, and whether anything else reaches this level of industry detail. It draws on R1’s survey of sources (the landscape note, sources checked 2026-11-03) and A1’s four consistency checks (the checks note, 2026-11-04).

1 What reaches region × industry

No other source publishes output per job or per hour at ITL3 × 47 industries, or ITL2 × 74. ONS publishes balanced regional GVA at 47 non-overlapping industries for ITL3, 74 for ITL2 and 82 for ITL1, but no matching jobs or hours denominator. ONS does the division internally (BRES jobs × APS hours) and publishes only whole-economy productivity below ITL1. The finest published alternatives are much coarser:

Source Owner Geography × industry Measure Years Status
Industry by region labour productivity (INDBYREG) ONS ITL1 × 16 industry groups GVA per hour and per job 1998–2019 Lapsed (July 2021)
Scottish Annual Business Statistics Scottish Government 32 council areas × 24 groups Approximate GVA per head, non-financial business economy 2008–2023 Current (2023 edition)
Regional database, labour productivity by activity OECD UK TL3 × 10 activities GVA per worker To 2024 Current
ARDECO European Commission UK NUTS3 × 10 sectors Hours and employment (GVA by sector in current prices) Hours to 2019 Current; UK only in the NUTS 2016 version
Annual Business Survey, regional results ONS ITL1 × about 80 divisions Approximate GVA, turnover, purchases (no jobs) 2008–2024 Current
Regional and subregional productivity ONS ITL1 to local authority GVA per hour and per filled job, whole economy only To 2023 (2024 due 24 November 2026) Official statistics in development

Internationally the picture is similar. Eurostat’s regional tables carry 10 industries and no longer any UK rows. Statistics Canada currently publishes labour productivity by province at 2-digit industry only. The US publishes GDP and jobs by county × industry, but no hours below state level.

Detail isn’t independence. ONS balances its income and production estimates at ITL2 × 33 industries. Below that, regional GVA is allocated rather than measured. ITL3 detail comes from the income approach, which spreads compensation of employees using ASHE pay × BRES jobs. So the BRES-linked figures reach further than anything else partly because they rest on allocation: they are GVA per job as allocated. The checks below test how far that matters.

2 GVA per job against ONS’s own figures

ITL3, whole economy. Ours: GVA (without imputed rent) per BRES employee job, and per filled job (employees plus the self-employed). ONS’s: subregional GVA per filled job (June 2025), with imputed rent taken out. Our GVA matches ONS’s to within half a per cent, so the ratio is really ONS’s job count over ours.

Table 1: Our GVA per job ÷ ONS subregional GVA per filled job, 169 ITL3s, 2015 to the latest year. Spread = interquartile range ÷ median across ITL3s.
Version Median ratio Spread Range, any year Rank correlation
Per employee job 1.08–1.12 0.058–0.074 0.94–1.38 0.93–0.95
Per filled job, hybrid (default) 0.96–0.98 0.019–0.035 0.81–1.07 0.98–0.99
Per filled job, national split 0.96–0.98 0.050–0.061 0.81–1.18 0.95–0.97
Per filled job, APS-led 0.95–0.97 0.018–0.033 0.80–1.06 0.98–0.99
Scatter of 169 ITL3 areas. Almost all lie in a narrow band around a median ratio of about 0.96, whatever ONS's level of GVA per job. Rural areas such as Wiltshire and North Yorkshire sit above the band; a few, such as Inverclyde, East Renfrewshire and Renfrewshire, sit below.
Figure 1: Our GVA per filled job (hybrid) ÷ ONS’s, by ITL3, against ONS’s level (log scale). The line is the median ratio; the six ITL3s furthest from it are named.
  • In 2023, 93% of ITL3s lie within 5% of the median ratio per filled job, against 67% per employee job.
  • We count more jobs than ONS in employment centres, fewer in rural areas. In rural ITL3s local BRES leaves out farm employees, which pushes our GVA per job above ONS’s. In city centres our count of employee jobs alone comes close to ONS’s filled jobs.

ITL1 × 16 industry groups, against INDBYREG (2015–19).

Table 2: Our GVA per job ÷ INDBYREG’s GVA per productivity job, 11 GB regions × 16 groups, 2015–2019.
Version Median ratio Rank correlation, all cells Within a region, across industries (median) Within an industry, across regions (median) … lowest
Per employee job 1.08–1.09 0.96–0.97 0.95–0.97 0.76–0.88 −0.05
Per filled job, hybrid 0.95–0.96 0.98–0.99 0.96–0.98 0.83–0.93 0.51
  • For comparing sectors within a region, our figures reproduce ONS’s ordering almost exactly. For comparing regions within a sector they agree less well: the comparison the research plan already flags as weakest.
  • ONS’s own revisions move some industries more than the denominator does. INDBYREG used regional GVA as published in 2021. Summed over GB regions, our 2025-vintage GVA for finance in 2019 is 1.29 times INDBYREG’s.

3 Regional GVA against the Annual Business Survey

The Annual Business Survey (ABS) is the one independent survey measure of output by region × industry. Its approximate GVA (aGVA) isn’t the same measure as regional GVA: it covers only the non-financial business economy, and has none of the national accounts’ adjustments. So A1 compared each region’s share of the UK industry total, using the standard errors ABS publishes for region × division (2024 quality-measures edition, 4 June 2026). Regional GVA has no published error, so “diverging” means outside ABS’s sampling error, not wrong.

Line chart, 2012 to 2024. In both releases about 17 to 27 per cent of cells lie outside ABS's interval in most years, far above the 5 per cent sampling error would give, rising sharply in each release's latest year: 41 per cent for 2023 in the April 2025 release.
Figure 2: Share of region × industry cells where regional GVA’s share of the UK industry lies outside ABS’s 95% interval, by year, for two GVA releases.
  • About five times what sampling error alone would give. Pooled over 2012–22, 23.6% of 7,868 cells lie outside ABS’s 95% interval.
  • The latest year of each release diverges most, as the figure shows.
  • By region: most often in Northern Ireland (45%) and Scotland (33%), least in the South East (14%).
  • The levels differ by industry, as expected, but in a telling pattern. In 2023, UK regional GVA ÷ aGVA is lowest for head offices and management consultancy (0.32); air transport (0.37); warehousing and transport support (0.41); architectural and engineering (0.51), and highest for libraries, archives, museums and other cultural (3.49); manufacture of motor vehicles (2.73); programming and broadcasting (2.32); sewerage (2.31). Some of the high ratios are coverage (households and sole traders aren’t on the business register). The low ones suggest a classification difference: ABS’s regional results classify each site by its own activity (§6).
Table 3: The largest divergences in 2022, by GVA at stake (the gap in shares × the UK industry’s GVA).
Region Industry Regional GVA’s share ABS’s share z GVA at stake
Scotland Human health activities (86) 9.2% 3.8% 32.7 £6.8bn
London Human health activities (86) 17.0% 21.6% −2.0 £5.8bn
London Wholesale trade (46) 20.7% 27.0% −7.8 £5.6bn
South East Human health activities (86) 12.4% 15.5% −2.2 £4.0bn
London Construction of buildings (41) 24.3% 16.5% 4.2 £3.7bn
London Computer programming and consultancy (62) 36.7% 41.6% −3.2 £3.0bn

So ABS’s sampling error isn’t a sufficient error band for regional GVA by industry. The divergences mix regional GVA’s own error with differences of definition and classification, so they don’t measure that error directly. Nothing comparable exists below ITL1.

4 Is it just pay? The circularity test

The worry. ONS spreads compensation of employees, about half of GVA, across regions using ASHE pay × BRES jobs. Divide by BRES jobs and that part of GVA per job is ASHE pay per job, by construction. So the higher an industry’s labour share, the more its regional pattern of GVA per job should just be its pay pattern. (RegionalEconomicTools’ jobs-and-hours note §7 calls the labour share “the severity dial”.)

The test. For each industry and year, across the 11 GB regions: the correlation between log GVA per employee job and log ASHE pay per job (mean gross annual pay, full-time and part-time weighted as the jobs are counted), with industries split by labour share from the Blue Book 2026 supply and use tables.

Scatter of about 70 industries. Correlations range from about minus 0.7 to 0.94 with no upward trend as the labour share rises from 20 to 87 per cent. Manufacturing industries mostly sit low; several services such as health, education, wholesale and retail sit high.
Figure 3: Each dot is an industry (ITL1 × mostly 2-digit SIC). Vertical: the correlation across the 11 GB regions between GVA per job and ASHE pay per job, median over 2015–2023. Horizontal: labour share of GVA, UK.
Table 4: Correlation across regions between GVA per job and pay per job, by labour-share band, ITL1 × mostly 2-digit SIC, 2015–2023.
Labour share Industries Median correlation Share of industry-years above 0.9 Median share of regional variation removed by dividing by pay
Below 45% 11 0.42 3% 5%
45–65% 28 0.56 6% 15%
Above 65% 32 0.51 4% 21%
  • The prediction fails at this level. Correlations don’t approach 1 in labour-intensive industries, and the labour share doesn’t sort industries by how much GVA per job tracks pay: the rank correlation between them across 71 industries is 0.06.
  • Where GVA per job does track pay (median correlation 0.78–0.94): auxiliary financial services, human health, legal and accounting, education, wholesale, accommodation and food, retail. Health and education are non-market services whose output is measured by their inputs, so this is circular by definition; the test finding them is a sign it works.
  • Where it doesn’t: most of manufacturing, social work, residential care, scientific research.
  • London carries part of it. Without London the median correlations by band fall from 0.42–0.56 to 0.28–0.37.
  • At ITL3, where the worry is sharpest, it can’t be tested with published data: ASHE publishes pay by industry only for regions, and regional pay with the local job mix barely varies within a region.

5 ONS’s own quality scores

With each release ONS publishes the quality scores it uses to balance the income and production estimates, for each ITL2 area × 33 industries × year, on a 0–1 scale. A score of 0.6 corresponds to a coefficient of variation of 20%; below 0.6 is “low” here. The income approach (ASHE × BRES) is the one that sets ITL3 detail.

Line chart 1998 to 2024. Both approaches have about 17 to 27 per cent of cells scored low until 2021. The income approach then jumps to 43 per cent in 2022, 47 per cent in 2023 and 86 per cent in 2024, while the production approach stays around 20 per cent.
Figure 4: Share of ITL2 × industry cells that ONS scores low (below 0.6), by approach and year.
  • This is a recent-years story. Scoring 2023’s BRES-linked cells with each year’s metrics, the share of jobs in cells low on the income approach is 2.7% with 2015’s scores, 4.2% with 2019’s and 61% with 2023’s.
  • The weakest cells, low on both approaches in 2023, hold 17% of jobs:
Table 5: ITL3 × category cells where both approaches score below 0.6 in 2023, by jobs. ITL3s take their ITL2’s scores, so these are upper bounds on ITL3 quality.
Category Jobs in weak cells ITL3 cells ITL2 areas
Human health activities 2.33m 150 38
Social work activities 0.81m 150 38
Residential care activities 0.67m 150 38
Other personal service activities 0.23m 134 35
CD-CG (19-23) 0.22m 110 28
Food and beverage service activities 0.20m 22 6
Activities of membership organisations 0.15m 134 35

Human health sits on the weakest estimates and its GVA per job tracks pay across regions, so for comparing places it adds little to pay, especially in the latest years.

6 Open questions

  • Why do our job counts exceed ONS’s in city centres? ONS’s local jobs may come from the household survey, which counts fewer jobs than employer surveys like BRES; not checked.
  • Do ONS’s subregional and INDBYREG jobs include the armed forces? BRES doesn’t. Public administration in the South West and South East, and rural ITL3s with large bases, are where our jobs fall furthest short of ONS’s.
  • A site-versus-business classification mismatch? ABS’s regional results classify each site by its own activity. If national-accounts industry totals follow each business’s main activity instead, GVA per BRES job would be understated in head offices, warehousing and wholesale, and overstated in their parents’ industries. The low GVA ÷ aGVA ratios fit; the next test is to compare ABS’s regional results (by site) with its national results (by business).
  • Why did the income approach’s quality scores fall from 2022? The file doesn’t say. ASHE and BRES are the inputs.
  • Why does circularity fail at ITL1? One reading is that below the 33 balancing industries, ONS takes ITL1 and ITL2 industry detail from the ABS-led production approach, which spreads a firm across its sites by employment. Not confirmed.
  • The ITL3 circularity test needs local pay by industry: ASHE Table 7 (all-industry pay by local authority) as a proxy, or secure ASHE microdata.
  • Should ONS’s per-cell quality scores travel with the build’s outputs, flagging 2022 onwards? Parked for now (decision 12).

Data and code. A1’s scripts: claude_research/productivity/R/a1_00_fetch.R (downloads), claude_research/productivity/R/a1_01_per_job.R (check 1), claude_research/productivity/R/a1_02_abs.R (check 2), claude_research/productivity/R/a1_03_circularity.R (check 3), claude_research/productivity/R/a1_04_quality.R (check 4), with helpers in claude_research/productivity/R/a1_functions.R; they read the hours build’s outputs (page 2). Sources: ONS regional GVA (balanced) by industry, v13 and the September 2026 release; ONS subregional productivity (June 2025); INDBYREG (July 2021); ABS regional results and quality measures (4 June 2026); ASHE Table 5; Blue Book 2026 supply and use tables (30 October 2026); ONS regional GVA quality metrics (23 September 2026). The landscape: the landscape note, sources checked 2026-11-03.