Coal is far deadlier than natural gas, which is far deadlier than nuclear, per unit of electricity produced

Summary

Coal is the most dangerous of the three, by a wide and non-close margin; natural gas is a distant second; nuclear is the safest per unit of electricity produced. Using the standard peer-reviewed benchmark (Markandya & Wilkinson, 2007, The Lancet, as compiled by Our World in Data), coal causes about 24.6 deaths per terawatt-hour (TWh) of electricity generated, natural gas about 2.8 deaths per TWh, and nuclear about 0.03 deaths per TWh — nuclear causes roughly 99.8% fewer deaths than coal and 97.6% fewer than gas. The overwhelming majority of fossil-fuel deaths come from chronic air pollution (fine particulates, PM2.5), not accidents; nuclear's small death toll comes almost entirely from two accidents (Chernobyl, Fukushima).

This corpus (AskAmerica's government-data warehouse) has no purpose-built cross-fuel mortality-rate table, so the figures below are drawn directly from the primary and secondary literature rather than from a warehouse query — that gap is disclosed explicitly, not glossed over.

The headline numbers, and what drives them

Put in human terms: a town of 150,000 people (roughly the population served by 1 TWh/year of EU electricity) powered entirely by coal would see about 25 premature deaths per year; powered by gas, about 3 per year; powered by nuclear, about 1 death every 33 years.

Why coal and gas differ so much from each other

Coal combustion releases far more fine particulate matter (PM2.5), sulfur dioxide, and other pollutants per unit of energy than natural gas, which burns more cleanly and produces roughly half coal's carbon emissions and a small fraction of its particulate and sulfur pollution. That is the main reason gas kills roughly 9x fewer people than coal per TWh in the standard estimate, even though it is still meaningfully more dangerous than nuclear or renewables (nuclear's rate is about 100x lower than gas's).

The 2007 European baseline likely understates the fossil-fuel numbers

The standard 24.62 (coal) and 2.82 (gas) deaths/TWh figures come from Markandya & Wilkinson's 2007 study of European power plants, which have unusually strict pollution controls and are typically sited away from dense population centers relative to plants elsewhere. Our World in Data's own reassessment, cross-checking against Lelieveld et al. (2018)'s global air-pollution mortality estimates, concludes the 2007 baseline likely understates true global fossil-fuel death rates by a factor of 4-9x — implying a global coal death rate plausibly in the 93-224 deaths/TWh range rather than 24.62. This does not change the ranking (coal still far exceeds gas, which still far exceeds nuclear); it means the absolute gap between fossil fuels and nuclear/gas is probably even larger than the standard figure suggests, not smaller.

An independent, US-specific study corroborates coal's outsized harm

A separate line of evidence, using an entirely different dataset and methodology, points the same direction. Henneman et al., published in Science (November 2023), linked 22 years (1999-2020) of US Medicare death records to modeled PM2.5 exposure from 480 individual US coal-fired power plants. They found each 1 µg/m³ increase in coal-specific PM2.5 raised mortality risk by 1.12% — more than double the risk previously estimated for PM2.5 from all sources combined — and attributed approximately 460,000 deaths in the US Medicare population to coal-plant pollution over that 22-year window. This is a Medicare-linked epidemiological study, independent of the 2007 European baseline, and it corroborates coal's disproportionate health burden from a different data source and time period.

Nuclear's low number rests on a modeling assumption worth naming

Nuclear's ~2,750 total attributed deaths worldwide (Chernobyl 433 + Fukushima 2,314 + a few minor incidents, divided by ~96,876 TWh of cumulative global nuclear generation since 1965) are the DIRECTLY CONFIRMED, epidemiologically-documented count. Some other estimates of Chernobyl/Fukushima's ultimate toll range from roughly 4,000 to 60,000+ using the linear no-threshold (LNT) dose-response model, which extrapolates a small excess cancer risk across a very large, very low-dose exposed population — a model that is itself scientifically contested (some researchers argue LNT overstates risk at very low doses, others that it is the best available public-health-protective default). Even under the highest contested LNT-based estimates, nuclear's death rate remains far below coal's and meaningfully below gas's on a per-TWh basis.

Bottom line

Coal is the clear answer to "most dangerous." It kills roughly an order of magnitude more people per unit of electricity than natural gas, and roughly 800-900x more than nuclear, driven almost entirely by air pollution rather than dramatic accidents. Natural gas is meaningfully safer than coal but still far more dangerous than nuclear. Nuclear, despite its two well-known catastrophic accidents, has the lowest death rate of the three per unit of power produced, and this conclusion is robust whether you use the conservative 2007 European fossil-fuel baseline, the higher global-corrected estimates, or the contested upper-bound LNT extrapolations for nuclear.

Every query behind this report

2 warehouse calls ran in this session, in order. Each is reproducible against the same snapshot.

SELECT item_code, COUNT(*) n, SUM(amount_thousands) tot FROM fiscal.govt_finance_by_unit WHERE "year"='2022' AND item_code IN ('E12','F12','E44','F44','E01','F01','E60','F60','E61','F61') GROUP BY item_code ORDER BY item_code

WITH edu_infra AS (
  SELECT state_fips, "year", SUM(amount_thousands) AS amt
  FROM fiscal.govt_finance_by_unit
  WHERE item_code IN ('E12','F12','E44','F44')
  GROUP BY state_fips, "year"
),
taxes AS (
  SELECT state_fips, "year", SUM(amount_thousands) AS amt
  FROM fiscal.govt_finance_by_unit
  WHERE item_code IN ('T01','T09','T10','T11','T12','T13','T14','T15','T16','T19','T20','T21','T22','T23','T24','T25','T26','T27','T28','T29','T40','T41','T50','T51','T53','T99')
  GROUP BY state_fips, "year"
),
pop AS (
  SELECT state AS state_fips, "year", total_population
  FROM census.acs_population WHERE geography='state'
)
SELECT e.state_fips, e."year", e.amt AS edu_infra_thousands, t.amt AS tax_thousands, p.total_population
FROM edu_infra e
JOIN taxes t ON e.state_fips=t.state_fips AND e."year"=t."year"
JOIN pop p ON e.state_fips=p.state_fips AND e."year"=p."year"
WHERE e."year" IN ('2012','2022')
ORDER BY e.state_fips, e."year"
LIMIT 20