On most medium-sized jobs somebody spends part of every week writing documents they know will not be read at the workface. Usually a safety advisor; sometimes an agent doing that job on top of their own. Another method statement for an activity the gang has done forty times, because the prequalification portal wants one and because being the person who did not write it is a worse place to stand. Repeat that across a generation and you have the documented half of construction safety.
The measured half is more stubborn. Great Britain's construction sector has spent a generation adding documented safety activity: construction phase plans, risk assessments, method statements, inductions, permits, prequalification packs, health and safety files. The Health and Safety Executive's construction bulletin for 2025 still has to say that the fatal-injury rate showed a long downward trend "with signs of flattening out in more recent years," and that the latest three-year window for self-reported non-fatal injury is not statistically significantly different from 2016/17–2018/19.
More files. Rates that have stopped moving, in the window HSE itself will call flat. That deserves care, not a punchline, not least because people are still dying, most often from a fall from a height.
What can be compared, and what cannot
The comparison is documented safety activity against injury and ill-health rates, not headcounts. Counts move with how many people are on site.
The window is the recent official construction series: fatal rates on the long RIDDOR run from 1981 to 2024/25p, and Labour Force Survey non-fatal injury and ill health averaged over 2022/23–2024/25, set against 2016/17–2018/19 as HSE does.
The left-hand axis is where honesty costs something. There is no accredited official statistic for pages of method statements, live procedures per worker, or hours of safety administration. Nobody collects it. If you have sat through the slide where a rising document curve crosses a flat incident curve, that curve was drawn by hand, and whoever drew it owed you the caveat.
The side that has numbers
Construction statistics in Great Britain, 2025 — data to March 2025, published 20 November 2025 — is accredited official statistics. Construction is section F of SIC 2007, about 6 per cent of the Great Britain workforce.
Fatal injuries. Thirty-five workers in 2024/25p, against a five-year average of forty (2020/21–2024/25p). Four members of the public, matching that five-year average. The fatal-injury rate is 1.92 per 100,000 workers — about 4.8 times the all-industry rate. Over the same five years, 53 per cent of those deaths were classified as falls from a height. HSE tells you to read the rate rather than the count. It fell for decades before arriving at that sentence about flattening. A later single year can print a lower death count; one year is not a trend. The figures here are the 2025 bulletin's.
Non-fatal injuries. The Labour Force Survey is HSE's preferred source. An estimated 50,000 construction workers sustained a non-fatal workplace injury, averaged over 2022/23–2024/25. About 2.5 per cent of workers in the sector — statistically higher than the 1.8 per cent all-industry figure. Twenty-nine per cent of those injuries led to more than seven days' absence. Before the pandemic the rate had been falling. The latest period is not statistically significantly different from 2016/17–2018/19.
RIDDOR, for the same latest year, recorded 3,726 non-fatal injuries to employees: 1,525 specified, 2,201 over-seven-day. That is not a second estimate of the same thing. HSE is explicit: non-fatal RIDDOR is substantially under-reported, especially for the self-employed, and reporting rates can vary by industry and over time. HSE's sources note puts employee reporting at around half of comparable Labour Force Survey cases. The 2013 RIDDOR change from over-three-day to over-seven-day injuries also breaks the count series. Quoting a RIDDOR employee total as the sector's injury rate skips the part of the bulletin explaining why.
Ill health. An estimated 79,000 workers with work-related ill health (new or long-standing) over the same three-year Labour Force Survey window. Fifty-three per cent musculoskeletal. The overall ill-health rate is not statistically different from all industries; musculoskeletal disorders are significantly higher — about 2.0 per cent against 1.2 per cent. The latest ill-health rate is not statistically significantly different from 2016/17–2018/19 either. About 2.2 million working days lost a year, most of them to illness, not injury. Economic cost for 2023/24: about £1.4 billion.
| Measure | Construction | Comparator | Period |
|---|---|---|---|
| Fatal injuries (workers) | 35 (2024/25p); five-year average 40 | Rate 1.92 per 100,000 — about 4.8× the all-industry rate | 2020/21–2024/25p |
| Falls from a height | 53% of construction deaths | — | Five-year window |
| Non-fatal injury (Labour Force Survey) | ~50,000 workers, ~2.5% of the sector workforce | 1.8% all-industry | 2022/23–2024/25 |
| Non-fatal injury (RIDDOR, same period) | 3,726 (1,525 specified, 2,201 over-seven-day) | Substantially under-reported vs LFS | 2024/25p |
| Work-related ill health | ~79,000 workers; 53% musculoskeletal | MSD rate 2.0% vs 1.2% all-industry | 2022/23–2024/25 |
| Economic cost | ~£1.4 billion; 2.2 million working days lost a year | — | 2023/24 |
| Documented H&S policy (ESENER) | UK 97% of establishments | EU-27 average 91% | 2019 |
| Fatal-injury rate, UK vs EU | 0.62 per 100,000 | EU-27: 1.29 | 2018 |
Long improvement, then recent rates HSE will not describe as a further significant fall.
The paperwork side has no series
No national document-volume series exists. The published record still has a shape.
The legal architecture expanded, then tried to slim itself. The Construction (Design and Management) Regulations of 1994, 2007 and 2015. The 2015 explanatory memorandum said the new regulations were meant to be "more streamlined and less bureaucratic" than 2007, after competence paperwork had become "bureaucratic, costly and repetitive" and the focus had drifted onto the process. HSE's Construction Information Sheet 80 still has to say that a construction phase plan "does not need to be complicated." Professor Löfstedt's 2011 review, Reclaiming health and safety for all, located the problem less in the regulations than in how they were applied: third parties promoting "unnecessary paperwork," employers advised to keep large numbers of records, "paperwork for every possible risk." The official diagnosis, more than a decade ago, was already that documented activity had outgrown its use: not a page count, but a regulator and a government reviewer describing the same inflation.
Surveys measure having documents, not how thick they are. HSE's Health and Safety Statistics in the United Kingdom compared with European countries, 2025 reprints the 2019 European Survey of Enterprises on New and Emerging Risks: around 97 per cent of UK establishments — five or more employees, all sectors — had a document explaining health and safety responsibilities and procedures, against an EU-27 average of 91 per cent. Around 90 per cent said they carry out regular workplace risk assessments, against 77 per cent. Managers in the UK sample were likelier than their European counterparts to have a documented policy and to do regular checks. That is prevalence in 2019, not growth from 2009, and it is not construction-only. It tells you the United Kingdom is a high-documentation jurisdiction. It does not tell you how many pages sat in a site cabin in 2024.
The European pairing cuts both ways. The same HSE comparison shows the UK among the lowest standardised fatal-injury rates in Europe — 0.62 per 100,000 workers in 2018, against 1.29 for the EU-27. High documented practice and a relatively low national fatality rate can sit together. They do not prove that more paper produced the low rate, and they do not prove the opposite. A within-sector recent flattening in construction is a different claim from "paperwork never worked."
The research literature has a name for the leftover. Rae, Provan, Weber and Dekker, writing in Policy and Practice in Health and Safety in 2018, call it safety clutter: "the accumulation of safety procedures, documents, roles and activities that are performed in the name of safety, but do not contribute to the safety of operational work." The mechanisms are duplication, generalisation and over-specification. The driver is asymmetry, which anybody who has tried to retire a template will recognise: easy to add a form, hard to withdraw one. Rae and Provan, in Safety Science the same year, separate "safety work" from "the safety of work." The former is activity performed to manage safety. The latter is whether people are hurt. Evidence of the first is what organisations can produce; evidence of the second is what they often cannot.
Robson and colleagues' systematic review of occupational health and safety management-system interventions, published in 2007, found generally positive but methodologically weak studies and concluded that the evidence was insufficient to recommend for or against those systems. A management system is not the same thing as a pile of method statements. It is the nearest experimental literature we have to "more documented safety activity, better outcomes," and it will not carry a causal claim.
Putting the two beside each other
On the outcome side, HSE's recent construction window is flattening: fatal rates after a long decline; non-fatal injury and ill health not significantly different from the late 2010s. On the documentation side, the United Kingdom already sat at the top of European surveys for written procedures and regular assessments; CDM 2015 was rewritten partly because the previous regime had become a paperwork process; and safety research has a name for activity that accumulates after the useful work is done.
Nobody has run the regression, because one variable has never been collected. More files can mean more care. They can also mean theatre — surface compliance, a closed induction, a permit that never reached the workface. The published record will not settle which. The leading killer in the fatal series is still a fall from a height — not a rare exotic risk that was waiting on a new template.
How you would prove this wrong
The claim has two limbs. Documentation volume grew. Incident rates did not move. Either limb can fail.
The first limb fails if a pre-defined, national measure of documented safety activity — hours of safety administration per 100,000 hours worked, a count of live procedures, or an equivalent agreed before looking at outcomes — is flat or falling through the same window. Opinion about "too much paper" is not that measure. ESENER prevalence is not that measure.
The second limb fails if, after the known breaks are handled — RIDDOR's 2013 over-seven-day change, pandemic years, employment mix, hours — HSE's preferred rate series shows a statistically significant further decline in construction through the window in which documented activity is said to have grown. Use the Labour Force Survey for non-fatal injury. Use the fatal rate, not the death count. A single-year fall in deaths does not falsify flattening, and a RIDDOR employee total set against a Labour Force Survey worker estimate does not falsify it either.
The stronger, causal version — that documentation growth caused the stall — is already beyond the evidence. It would take a pre-registered pairing of those two defined series finding a significant inverse relationship after those controls, or a high-quality intervention study in which specified documents reduced specified rates. Robson and colleagues did not find that body of evidence in 2007, and it has not been published as an official Great Britain construction statistic since. Both tests are answerable; neither has been answered, because nobody funds the left-hand axis.
What none of this licenses
It does not license abolishing risk assessment, the construction phase plan, or the health and safety file. CDM 2015 still requires them, and HSE still says they should be proportionate. It does not license treating the long decline in fatal rates since the 1980s as unrelated to better planning and law. And it does not license assuming that a software system, a mandate, or a model will move the next rate on its own.
What they will support together is narrower: the industry likes a visible product of safety work, and the official outcome series for construction has, in the recent window HSE will stand behind, stopped showing a further significant fall.
Where the effort should go instead
Start with the missing axis, because it is the cheapest. A principal contractor can count, on one job, the live procedures a supervisor is meant to hold in their head and the hours a week the team spends writing safety documents rather than supervising work. Run it for a year and the industry has the start of a series it never had.
Make withdrawal somebody's named duty: a new form should name the document it supersedes, signed off by somebody senior enough that deleting one does not feel like accepting liability. Clients can cut more than anyone by taking a recognised prequalification scheme instead of a bespoke pack. Löfstedt said as much in 2011; the packs got longer anyway.
Then spend where the fatal series points. Falls from a height were classified as 53 per cent of construction deaths in that five-year window. Edge protection, access plant, competent supervision and the float to use them properly cost real money at tender, and none of it photographs as well as a signed template. Clients and principal contractors price that trade-off; the advisor writing Thursday's method statement cannot. Ask at tender, not at the inquest.
Read next: The principal designer's information duty that nobody resources, Permit to work on paper: a system designed to fail quietly, CDM 2015, ten years on: what the regulations assume about information.