Reading the numbers
How we count
Everything here comes from one idea: count deaths per unit of electricity, not per source in total. Here is what that means — in plain terms — and where it breaks down.
Four measures, one scale
A terawatt-hour, first
A terawatt-hour (TWh) is the unit of how much electricity — one billion kilowatt-hours, very roughly a year of power for 150,000 people in the EU. It is what we divide by, and it is what makes sources comparable.
Why we divide, with real numbers
Suppose coal is associated with roughly 250,000 deaths a year worldwide and solar with a few dozen. Coal looks catastrophically worse — but coal also generates far more electricity, so most of that gap is just scale. Divide each by the electricity produced and the scale cancels out:
- Coal: ~250,000 deaths ÷ ~10,000 TWh ≈ 25 deaths/TWh
- Solar: a few dozen deaths ÷ its far smaller output ≈ 0.02 deaths/TWh
On raw totals
Coal looks about 6,000× worse — but coal also makes far more electricity, so most of that gap is just size.
Per unit of electricity
Coal is about 1,000× worse. The gap shrinks — but the verdict holds. Normalizing changes the number, not the conclusion.
Now the comparison is fair: coal is on the order of a thousand times more deadly per unit, and that no longer depends on how much of each we build. Comparing raw totals would have flattered solar for a reason that has nothing to do with safety — it simply produces less.
How the number is built
For each source we take the deaths associated with it and the electricity it produced — both measured, not assumed.
Deaths ÷ terawatt-hours gives deaths per unit of electricity. This removes the effect of how much each source happens to produce, so the comparison is fair.
Accidents are counted events. Air-pollution and radiation deaths are modeled statistical estimates. We draw the two differently and never blur them together.
Mining, building, running, fuel, and waste all count — which is why even wind and solar are not exactly zero.
Every figure is a low–high band, not a single confident number. Where the science is unsettled, the band is wide.
Counted vs modeled — the key distinction
This is the hardest idea on the site, so here it is as a picture. On the left, deaths you could list by name. On the right, the same total spread thinly across a whole population.
A fall, a mine collapse, a dam failure. Identifiable people — you could list them.
A slightly raised risk across everyone downwind. Real deaths in the aggregate — but no single one is identifiable as coal’s.
That’s why the charts draw them differently — solid for counted, hatched for modeled — and never blur the two. For coal and gas almost the whole bar is hatched; for nuclear, most of the number is modeled cancer projections rather than the counted casualties of Chernobyl and Fukushima.
Where these numbers are weakest
Data vintage differs by source, the fossil rates come from a European baseline generalized worldwide, nuclear depends on the contested linear no-threshold model, and none of it models storage, reliability, or transmission. Country pages adjust for two of these — hydro's Banqiao tail and pollution controls — but the coefficients stay coarse. When in doubt, read the band, not the midpoint.
A few terms you'll see
A unit of “how much electricity.” One TWh is very roughly a year of electricity for 150,000 people in the EU.
The rate we compare. It answers “how many deaths per unit of electricity,” independent of how big a source is.
Counted deaths are recorded events. Modeled deaths are statistical estimates of pollution or radiation harm. Drawn as solid vs hatched.
A death assigned to an exposure by a model, rather than a single identified victim. Most of coal’s and nuclear’s numbers are these.
Ready to look? Compare the eight sources →