Harm channels
Harm reaches people through four routes, and each is counted differently because each is known differently. Two are observed at an incident. Two are attributed through a risk factor. The difference decides what can be said about them.
Three of the four channels hold data. Their totals appear side by side and are never added together, because a person can be in two of them at once — a miner killed in an armed attack is a conflict death and a workplace death — and nobody has yet measured how often that happens. Adding them would be the easiest way to publish a wrong number, and it is the one this page refuses.
What is in the ledger
Every figure is the publisher’s, republished with the credit its licence requires. Nothing here has been attributed to a product, a company or a material — that is a later step, and it is the step people will argue with.
| Channel | Deaths in the ledger | Countries | Years | State |
|---|---|---|---|---|
|
Conflict and political violence
C1 |
4,257,891 | 126 | 1989–2025 | ingested |
|
Work
C2 |
1,523,089 | 142 | 1970–2024 | ingested |
|
Deprivation
C3 |
— | — | — | no rows — evidence missing |
|
Exposure and pollution C4 |
— | — | — | no rows — evidence missing |
|
Use phase C5 |
— | — | — | future work |
There is no total row on this table, and its absence is deliberate. Until the overlaps below are measured, no honest total exists, and a number that cannot be defended is worse than a gap that can be explained.
Deaths only. Injuries are counted in the work channel, modelled in the conflict channel, and absent from the other two.
Why the columns are not added
A person can be harmed through two channels at the same time. A miner shot during an attack on the site he works at is recorded as a conflict death by the people who count conflict deaths, and as a workplace fatality by the people who count those. Both records are correct. Adding them counts him twice.
There are four such collisions between these channels. Each needs a measured fraction — how much of one channel’s harm also sits in another — and none has been measured yet. Until they are, this page shows four numbers and no fifth.
| Collision | What is counted twice | Overlap |
|---|---|---|
| C1 × C2 | Workers killed at work by organised violence — a miner shot during an armed attack on the site he is working. UCDP counts him as a conflict death because an organised actor killed him; a labour inspectorate would count him as an occupational fatality because he died at work. Both are right. | not measured |
| C1 × C3 | Siege famine and conflict cholera. A child who dies of cholera in a besieged city is an indirect conflict death and a WASH-attributable death at once, because the conflict is what destroyed the water system. | not measured |
| C2 × C4 | Occupational exposure counted twice — a worker whose lung disease is attributed both to workplace carcinogens (C2 disease) and to ambient particulates (C4). | not measured |
| C3 × C4 | Indoor against outdoor air. Household air pollution from cooking fuel (C3) and ambient PM2.5 (C4) act on the same lungs, and the published attributions for each are computed against different counterfactuals — WHO publishes a "joint effects" figure precisely because adding the two overstates. | not measured |
An unmeasured overlap is recorded as unmeasured, not as zero. The two are different claims: one says nobody has looked, the other says somebody looked and found nothing. Only the first is true here.
What is missing
Listed rather than estimated. Every one of these makes the published figures lower than the truth, and none of them makes them higher.
- Indirect conflict deaths — the famine and disease a war causes rather than its weapons — are in no channel at all. Published estimates put them at three to fifteen times direct deaths.
- Occupational disease is absent. Injuries at work are counted; the cancers and lung diseases that arrive decades after an exposure are not, and international estimates put those at roughly four times the injury toll.
- Undernutrition is absent, and it is the largest thing missing from the deprivation channel. No exposure–response relationship for it has been read and cited yet.
- Air pollution, indoor and outdoor, is absent from both channels it belongs to. The attributed deaths are published; the attributable fraction is not, and a death whose share of its cause cannot be stated does not enter this ledger.
- Occupational statistics cover the reporting system, not the workforce. Informal and artisanal work — where the most dangerous conditions concentrate — is largely outside them everywhere it matters most.
- Harm from USING a product — road deaths, tailpipe pollution, firearms, tobacco — is a fifth channel that is future work. Every figure here is production only.
Questions
Why not just add the channels up?
Because some people are in two of them. The four places that happens are listed above, and none has been measured, so any total would be an overstatement of unknown size. It is a short arithmetic step away and it will stay undone until the overlaps are evidenced.
Are these HumanToll’s figures?
No. They are the publishers’ — Uppsala, the ILO, Eurostat, the World Health Organization — republished with the credit their licences require and with the exact release and checksum they came from. HumanToll’s own figures attribute harm to products, and none is published yet.
Which channels are counted and which are attributed?
Conflict and work are counted: somebody observed the death and recorded it. Deprivation and exposure are attributed: nobody observes a death from unsafe water, so a share of the deaths from a cause is assigned to the exposure using a published relationship. The second kind is weaker and is always labelled as such.
Does a channel with no rows mean there is no harm in it?
The opposite. It means the harm is real and the evidence needed to count it has not been assembled — usually because the comprehensive source is licensed in a way this project cannot use. An empty channel is a statement about this project, not about the world.