Methodology

The research question

How much human harm — deaths and injuries — can be linked, through documented chains of cause and effect and a model of world trade, to making and using the goods and services people buy?

The question is on purpose not “what causes war?” and not “is this product harmful?”. Those are debates. This is a question of careful accounting, with a stated method, a range of uncertainty and a choice of scenarios — which is what makes it answerable, and what makes disagreeing with it useful.

What counts as harm

A death or injury is counted only when research has measured how much of that kind of harm is caused by something the economy produces, spreads or withholds — and that cause could be changed.

This is how “leave out natural deaths” is done: as a rule about data, not a matter of opinion. Deaths with no such known cause — heart disease of old age, cancers with no known outside cause, conditions people are born with, ageing itself — are left out automatically, because there is nothing to link them to. Nobody decides what is “natural”: if science has found no measured link between a cause and a kind of death, that death is not counted.

One consequence has to be stated wherever a figure appears: the model counts too little, never too much. Harm that cannot be linked to a product is left out. Every figure is a floor, never a total.

The kinds of harm

Four kinds are counted. The table below is read straight from the database, so it always matches what the model actually does.

War deaths (C1) are counted directly: a death in battle is recorded, not estimated. Work deaths and injuries (C2) are reported by each country’s labour statistics. Deaths from poverty (C3) and pollution (C4) are counted only through the share of deaths that research says a cause explains — for example, the share of diarrhoea deaths caused by unsafe water. That is what turns “deaths from disease” into something that can be counted.

A fifth kind — harm from USING a product rather than making it: road deaths, smoke from burning fuel, firearms, tobacco, alcohol — has been designed but not built. The database was prepared for it from the start, so adding it later is a matter of filling it in, not rebuilding. Until then every figure is labelled "production phase only" and is never presented as a lifetime total. For a car, the harm from using it is probably the larger part.

What is measured, and per what

Two measures lead and are shown equally prominently: lives lost and injuries. Three more are for specialists: years of life lost, years lived with a disability, and the two added together — the only fair way to add a death and an injury.

Every product result is given per item, per thousand and per million items, per kilogram and per €1,000 of spending — and turned around, as “one death for every N items”. Turning it around is not a trick: it is the same number in a form a person can picture. Money amounts are adjusted for inflation to one fixed year, so a result does not change just because prices or a currency did.

From harm to product, step by step

Harm is counted, checked so that nothing is counted twice, shared out among industries, divided among the raw materials those industries dig, grow or use, turned into an amount of harm per unit of output, and carried along the supply chain to finished products by a table of what every industry in every country buys from every other. The uncertainty is carried through every step.

The share of a war that goes to each industry is scored, never simply asserted. There are nine criteria: how much the fighting is over land with that industry on it, how much exports depend on it, how much the government’s income does, where the fighting happens, what the armed groups themselves say, who controls the resources, trade ties, where the weapons come from, and the historical record. Each is scored against a published guide, with a source for every score and the name of the person who scored it.

Inside every industry, harm is divided among its materials by what they are worth — tonnes times the price — so a gold mine counts for more than the same weight of iron ore, and a field of cocoa for more than the same weight of sugar cane. Crops are priced at what farmers in that country were paid; minerals at what the whole world paid for them in trade that year. Harm is only ever placed on what the country where it happened produces, never on a world average.

Harm that is linked to the arms industry is placed on the industry that made the weapons, not on the country where people were harmed: on the countries that supplied each side, in proportion to what each delivered, and on a side’s own industry, in proportion to what it makes itself. Weapons stay in use for years, so deliveries are counted as a stock that loses half its weight every ten years. Weapons whose maker the records do not show — most of what rebel groups use, and anything sold second-hand — are counted but placed on nobody. Without the arms records, this harm stays where it happened.

Arms transfers and military spending: Information from the Stockholm International Peace Research Institute (SIPRI), https://www.sipri.org/databases

Where harm is placed: in the country where the killing takes place

A death is counted where it happened, and the model keeps it there: a war’s harm is linked to the industries of the country — or countries — where the people died, and so to the products made there.

A war fought in Ukraine is charged to Ukraine’s oil and gas, coal, grain and steel — not to the Russian exports that pay for it — and Russian soldiers killed in Ukraine are counted in Ukraine. The only exception is weapons: they are a product with a maker, so their share goes to the country that made them.

This rule is wrong, and it is as wrong as the war itself. It puts the cost of a war on the people it is fought among, not on whoever decided it or pays for it. That is exactly what a war does: an invasion moves its cost onto the land, the work and the lives of the country it invades. The dead lie in that country’s fields, mines and towns because the war put them there. Moving them, in the accounts, back to the invader’s exports would describe what ought to be, not what is — and it would need a theory of who is responsible that the data cannot test and that every side of every war disputes. This project measures where harm falls and which economy it falls on. The injustice of that is the injustice of the war, shown rather than corrected.

• Products made from a war-torn economy carry its war: Ukrainian grain, Sudanese gold, Congolese copper.
• A country’s figures are harm suffered there — never harm caused by it.
• These figures do not say who is responsible for a war, and must not be read as if they did.
• Within a country, the data cannot say where. A war’s share for farming or mining is spread over all of that country’s farms or mines, wherever they are: the fighting in Amhara in 2023 reaches Ethiopia’s coffee, grown in the south and west; attacks in North Kivu reach the copper and cobalt mined in Katanga, more than a thousand kilometres away. Each product page lists the wars behind its figure.

Nothing is dropped, nothing is inflated

Every scenario keeps an "Other" share for harm the evidence cannot place. With Other included, the shares always add up to 100%, in every scenario, for every war and kind of harm. A scenario never deletes a share and never stretches the remaining shares to fill the gap.

Stretching the remaining shares sounds natural, and it is wrong in a way that can be measured. Suppose only 40% of a war’s link to the economy has evidence behind it, and the rest is thrown away. Stretching the 40% back up to 100% would turn an oil share of 30% into 75% — the model would make its strongest claim about oil exactly where its evidence is weakest. The "Other" share keeps every supported share at its true size and turns the gap into a published number.

Two things follow, and the software enforces both. "Other" is always shown, never hidden: under a cautious scenario a product page shows how much harm was passed along and how much was held back, because how much the model would not place is itself a finding. And a calculation that loses any harm between two steps stops with an error — it is a mistake, not a rounding difference.

See how each scenario uses this →

Counting nobody twice

There are three ways a person could be counted twice, and each is guarded against.

• Overlapping kinds of harm — a miner killed in an armed attack belongs both to war and to work; a child who dies of cholera in a city under siege belongs both to war and to poverty. The overlap is measured for each pair of kinds, each country and each year, from evidence, and removed before anything else happens. Records are adjusted, never deleted, so the adjustment can always be seen and undone.
• Overlapping sources — different datasets report the same events. For each event one source is the main one; the others are used only to check it and find gaps, never added on top. The database itself makes this impossible to break.
• Supply chains that loop — industries buy from each other in circles. The harm is applied only to what each industry buys directly, and the mathematics of the trade table takes care of the endless loops. Applying it a second time to total output would inflate every result.

How sure we are

Every figure has a grade, and the grade is worked out from how wide its range is, how many steps lie between the harm and the product, and how reliable the sources are — never typed in by hand. A result with no interval is not published at all: the software that shows figures cannot be used without a range (the low and high values), so a page cannot print a bare number even by mistake.

HIGH

Main source, measured directly, at most two steps from harm to product, and a range no wider than 40% either side of the central figure.

MEDIUM

Linked by the model, three or four steps, with a range of up to about three times between its low and high ends.

LOW

Little data, or estimates standing in for data, more than four steps, and a range that spans ten times or more.

INDICATIVE

Shown only as a range, never as a single figure.

Versions, and repeating the calculation

• Each version of the method is frozen when it is published and can no longer be changed or deleted — the database enforces this. Corrections become a new version; the old one stays readable and citable for good.
• When a data source publishes a revised version, a new calculation is made; the old one is never altered. Results for each year sit side by side rather than replacing each other.
• Every calculation records the software version, the data versions, the scenario, the method version, its settings, and a fingerprint of every input file. Anyone with that bundle gets exactly the same number.