The Roofing Fatality Atlas

Two federal counts, two different populations

Almost every published claim about how dangerous roofing is rests on one of two BLS numbers without saying which. They are produced by the same programme: the Census of Fatal Occupational Injuries, which is a census rather than a survey and cross-checks death certificates, workers compensation records, OSHA reports and news accounts, but they classify the same deaths along different axes.

The occupation figure counts people whose own job was roofing, under the Standard Occupational Classification code 47-2181. In 2024 that was 104 fatal work injuries. It includes a roofer employed by a general contractor, a roofer employed by a school district, and a self-employed roofer working alone. It excludes the labourer, the driver and the office manager who worked at a roofing company and died on the job.

The industry figure counts everyone who died while working for an establishment classified in NAICS 23816, roofing contractors, whatever their own occupation. In 2024 that was 120 fatal work injuries. It includes the labourer, the driver and the office manager. It excludes the roofer employed by the general contractor next door.

The gap between them in 2024 was sixteen deaths, and the direction of the gap is not fixed by anything: in a year when more roofing-company drivers die in road crashes the industry number rises without the occupation number moving at all. They are not interchangeable, and a page that quotes one and labels it the other is wrong even when both numbers are real. Everything below states which basis it is using.

One more distinction matters and is routinely lost. A count of deaths answers "how many people died". A rate answers "how likely was it". Roofing has a small count and an enormous rate: 104 roofers died in 2024 against 950 truck drivers, and roofing is still by a wide margin the more dangerous work. The rate BLS publishes for roofers is 48.7 per 100,000 full-time equivalent workers, and the FTE denominator comes from the Current Population Survey, so it does include the self-employed.

Eight years of the industry count, and the hole in it

The published industry series runs 107, 106, 102, 123, 124, 134, 120. The dash is real. CFOI Table A-1 for reference year 2019 does not carry a row for roofing contractors, or for specialty trade contractors at all; we checked the current and archived HTML editions on 3 September 2026 and the string does not appear. We have not estimated the missing year and the chart draws it as a hole rather than joining the line across it.

Over the seven years that are published, 816 people died working for roofing contractors. 650 of those deaths, 79.7 per cent, were classified as falls, slips and trips. No other event category comes close. Exposure to harmful substances or environments is a distant second and includes both electrocution and heat.

HyreRoof analysis. BLS publishes an hours-based fatality rate for detailed industries in a spreadsheet we could not retrieve (see Limitations), so we computed our own denominator: deaths divided by QCEW annual average covered employment in NAICS 238160, times 100,000. On that basis the rate was 54.2 in 2017 and 47.9 in 2024, with a seven-year mean of 52.4. It moved between 47.5 and 56.0 and shows no trend in either direction.

This is an upper bound on a true per-worker rate, because the numerator includes self-employed and unpaid family workers while the QCEW denominator does not, and the roofing trade has a large self-employed population. The number to take from it is the flatness, not the level.

Set that flatness against the growth underneath it. QCEW counted 197,328 covered employees in roofing contracting in 2017 and 250,564 in 2024, a rise of 27.0 per cent, across an establishment count that went from 21,874 to 31,562. The industry got substantially bigger and substantially more fragmented, and the death rate per employee did not fall.

What the enforcement record looks like

OSHA publishes a public tool that returns every standard cited against a NAICS code in the most recently completed federal fiscal year. Run against 238160 for citations issued October 2024 through September 2025, it returns 10,667 citations across 4,281 inspections carrying $46,519,129 in current penalties. Run against NAICS 23, all of construction, it returns 26,558 citations across 10,794 inspections and $108,935,100.

HyreRoof analysis. Roofing contractors accounted for 39.7 per cent of the federal OSHA construction inspections that produced at least one citation, 40.2 per cent of the citations and 42.7 per cent of the penalty money, from 3.1 per cent of construction employment. That is an over-representation factor of 12.9 on inspections and 13.9 on penalties. Both the numerator and the denominator of the inspection share are drawn from the same federal jurisdiction, so the comparison inside construction holds even though the absolute inspection count omits the state-plan states entirely.

The concentration inside roofing is more striking than the concentration of roofing inside construction. Of the 4,281 roofing inspections that produced a citation, 3,647, 85.2 per cent, included a citation under 29 CFR 1926.501, the duty to have fall protection. Put the other way round: roofing contractors accounted for 54.9 per cent of every federal OSHA construction inspection in fiscal 2025 that cited 1926.501 at all. More than half of American construction’s fall-protection enforcement happens on roofs.

The next three standards on the list say the same thing in different registers. Ladders (1926.1053) appear in 1,485 inspections. Fall protection training (1926.503) appears in 1,304: a citation that means the employer could not show the crew had been taught the system it was supposed to be using. Fall protection systems criteria (1926.502), which is what you get cited under when the equipment exists but is wrong or wrongly rigged, appears in 314.

Money follows the same shape. HyreRoof analysis: the mean current penalty per cited roofing inspection is $10,866, against $10,092 across construction. Fall protection alone accounts for 58.0 per cent of every dollar penalised against roofing contractors in the year.

1,535 severe injuries, and the height they happened at

Since 2015 employers under federal OSHA jurisdiction have been required to report every inpatient hospitalisation, amputation and loss of an eye within 24 hours. OSHA publishes the resulting file with a short narrative and coded event, source, nature and body-part fields. HyreRoof analysis: of the 105,996 reports in the January 2015 to November 2025 release, 1,535 are in NAICS 238160, 1.4 per cent of all severe injuries reported in the US economy, and 8.1 per cent of those reported in construction. 1,514 involved a hospitalisation and 50 an amputation.

1,212 of the 1,535, 79.0 per cent, were falls. That figure was computed independently of the fatality data and lands within a percentage point of the 79.7 per cent falls share of the CFOI death series. Two different federal collections, different definitions, different jurisdictions, and the same answer: roofing is a falling problem with some other problems attached.

The height distribution is the most useful thing in this dataset and the most counterintuitive. Among the 626 reports carrying a specific height band, 493, 78.8 per cent, were falls of twenty feet or less, and 177 were from ten feet or less. Only 38 were from over thirty feet.

The modal severe roofing fall in the United States is from eleven to fifteen feet: a single-storey eave, a garage, a porch roof. This is the reason the OSHA fall-protection trigger height in construction is six feet and not something more dramatic, and it is the reason a homeowner’s instinct that a low roof is a safe roof is wrong.

The source field adds the second useful finding. 175 of the reports, 11.4 per cent, are coded as a fall through a surface or an existing opening rather than off an edge, and 132 name a skylight somewhere in the source description. A skylight in an existing roof is a hole with something brittle over it, and it does not read as a hazard until it fails. Ladders are the recorded source in 326 reports, 21.2 per cent, which is a reminder that a large share of roofing injury happens before anyone reaches the roof.

782 of the 1,535 reports, 50.9 per cent, are coded as fractures. 56 are exposure to environmental heat, a category that barely existed in the public conversation about roofing ten years ago.

The nonfatal numbers are falling. The fatal ones are not.

This is the part of the atlas we find hardest to explain away, and we are going to state the evidence rather than resolve it.

On the nonfatal side, roofing contracting looks like an industry that has transformed itself. Its total recordable case rate was 5.3 per 100 full-time workers in 2014 and 2.4 in 2024, a fall of 54.7 per cent. Its days-away-from-work rate went from 2.3 to 1.1. Over the same eleven years all private industry went from 3.2 to 2.3, a fall of 28.1 per cent, and the specialty trades went from 3.8 to 2.3. In 2014 roofing’s recordable rate was 1.66 times the private-industry average. In 2024 it was 1.04 times it. On this measure roofing is now an ordinary job.

On the fatal side nothing of the kind happened. The occupation rate is 48.7 per 100,000 FTE. Our derived industry rate has a seven-year mean of 52.4 and ends the series close to where it started. The count of severe injuries reported to OSHA under a mandatory 24-hour rule has no trend either: 128 in 2015, 118 in 2024, with the highest year in the middle of the series.

HyreRoof analysis. Put the SOII case count against the QCEW headcount and the divergence sharpens. SOII estimated 9,000 recordable cases in roofing contracting in 2015 and 5,200 in 2024, a fall of 42.2 per cent, while QCEW covered employment rose from 177,954 to 250,564, up 40.8 per cent. Cases per 10,000 covered employees therefore fell from about 506 to about 208, a drop of 59.0 per cent, in an industry whose fatality rate and mandatory severe-injury reports did not move.

This is editorial inference and we label it as such: the most parsimonious reading is that the two collections have different failure modes rather than that roofing became three times safer to be hurt in and no safer to be killed in. SOII is a survey of employer-kept OSHA logs among establishments that are sampled and respond; CFOI is a census that cross-checks death certificates; the severe injury file is a legal reporting duty with a 24-hour clock. A fatality and an amputation are very hard not to record.

A recordable sprain in a nine-person roofing company with no safety officer is not. As the industry fragmented (21,874 establishments in 2017 to 31,562 in 2024) the share of the work being done inside very small employers grew, and very small employers are where recordkeeping is thinnest.

We cannot prove that from these datasets, and we are not going to pretend otherwise. What we can say is that the nonfatal series and the fatal series cannot both be describing the same underlying change, and anyone citing the nonfatal improvement as evidence that roofing has become safe is citing the weaker of the two.