Productivity by place and industry
What the numbers say, and what they assume
These reports are the technical record of the productivity strand of MONIAC2, a project that builds small, open models of how the economy’s “plumbing” works so that the assumptions behind economic arguments are visible and arguable. The strand started from a simple measure, GVA per full-time job by place and industry, built from ONS regional GVA and BRES jobs in the RegionalEconomicTools web blocks, and asked five questions of it. The reports set out what we found, where questions remain, and the detail behind each answer.
The five questions
- What’s measured (page 1): how does regional GVA linked to BRES jobs compare with other measures of output, and does anything else reach this industry detail?
- Jobs or hours? (page 2): how do levels compare once we estimate the hours behind BRES full-time and part-time jobs?
- What GVA per hour is made of (page 3): how much do capital, pay and other factors matter, in very different sectors?
- Demand or supply? (page 5 and page 6): how can we tell whether output fell because demand fell or supply did?
- The 2D frame (page 4): Webber & Huaccha’s 2D model of GVA counts goods, which may break comparisons across sectors. How should it be conceived, and what does its “potential” mean?
What we found
Measures (page 1). Nothing else publishes output per job or per hour at ITL3 × 47 industries. Once the self-employed are counted, whole places agree closely with ONS (GVA per filled job a median 0.96–0.98 of ONS’s figure, rank correlation 0.98–0.99). But comparisons of regions within an industry are weaker than comparisons of industries within a region. ONS’s own quality scores for the estimates behind ITL3 detail collapsed from 2022: 47% of cells scored low in 2023, against about a fifth before 2020.
Jobs or hours (page 2). GVA per full-time job overstates GVA per full-time equivalent by 21% for the GB economy, and by up to 68% in part-time-heavy sectors. Adding the self-employed matters as much as counting hours: GVA per hour falls from £48.9 (employees only) to £43.7 (all workers), and by a third or more in construction, the arts and personal services. Against ONS the shape agrees, and the level gap (our hours 0.93 of ONS’s) is about what ONS’s change in how it measures hours predicts.
What GVA per hour is made of (page 3). Pay explains little of the gap between sectors: across 13 sectors it accounts for 10% of the spread in GVA per hour in 2023, the “excess” over a required return on capital for 67%. The excess is concentrated in capital-heavy and rent-earning sectors (electricity and gas, pharmaceuticals, mining, finance) and volatile year to year. Every assumption behind the split is a switch, and the page maps what each one moves.
The 2D frame and “potential” (page 4). Webber’s rectangle travels across sectors only in money: in pounds, the paper’s two equal-GVA firms sit on one isoGVA curve; in units they don’t. “Potential” can mean four things, giving four different demand gaps for the same industry. A statistical filter run in real time re-reads a long slump as lower potential: for UK GDP in 2011–13 it put output 0.5–1.5% above potential, where hindsight puts it below.
Demand or supply (page 5). In a pilot with rules fixed before the data were seen (15 industries, 3 downturns, 3 signals), the signals agree no more often than chance (Fleiss’ kappa 0.02). Where they agree, it’s demand. The price index chosen flips 50% of the price–quantity labels.
Rising costs (page 6). Rising costs can’t tell demand from supply on their own: in a simulation with a known answer, reading them as supply is right 67% of the time. Variation that only one cause can produce does better. In UK manufacturing, divisions spending more of their output on energy raised prices 5.9 log points more per percentage point of energy intensity between 2019 and mid-2023, and cut volumes 8.7 points more: the energy shock’s own footprint was supply-side.
Where questions remain
The owner’s decisions so far, and what’s still open, are in the research plan (§6). The largest open questions:
- Do the self-employed work longer weeks than other workers? It would close most of the remaining gap with ONS’s hours (page 2).
- Is warehousing’s and head offices’ low GVA per job a classification artefact? Jobs counted by each site’s activity, output by the parent business’s (page 1).
- Why did ONS’s quality scores for the income approach collapse from 2022? (page 1)
- Which meaning of “potential” should a model use, and should it adapt to demand? (page 4)
- How much of 2022–23’s common price rise was demand? The exposure design identifies only the energy-driven difference between industries (page 6).
How to read these reports
- Every number in the text is computed when the page is built, from the strand’s data files. After an ONS release, re-running the pipeline and rebuilding updates them; the prose is re-read after each rebuild.
- Two layers, kept apart. Money (current prices) for comparing sectors and places at a point in time; volume (chained volume) for growth, always with the deflator named.
- Denominators are named. GVA per full-time job, per FTE, per paid hour, per employee hour or per hour by all workers: each figure says which.
- Unverified claims are left out or set as open questions. Each page ends with its open questions and a “Data and code” note linking the R scripts that made its numbers.
- No code in the pages. The method is in the linked scripts in
claude_research/productivity/R/, and the full working is in the strand’s notes.
Data vintages
| Source | Vintage used |
|---|---|
| ONS regional GVA (balanced) by industry | April 2025 release as corrected 10 October 2025 (“v13”); September 2026 release for some checks |
| BRES (Nomis) | Release of 28 October 2025, years 2015–2024 |
| ASHE Table 5 | 2015 revised to 2026 provisional |
| Self-employment | ONS JOBS04 (15 September 2026); Annual Population Survey to December 2024 |
| ONS subregional productivity | June 2025 (2024 edition due 24 November 2026) |
| ONS output per hour by division | 18 August 2026 |
| Supply and use tables | Blue Book 2026 (30 October 2026) |
| Producer prices, GDP(O) low-level aggregates | 21 October 2026; 30 September 2026 |
Built with R and Quarto. Source: claude_research/productivity/report/. Rendered 4 November 2026.