Count Deaths per Unit of Power and Nuclear Beats Nearly Everything Else

Ask a room full of people to name the most dangerous way to make electricity and watch the hands go up for nuclear. Then ask them to name the source that kills the most people per unit of power, and watch the room go quiet, because the honest answer is coal, by a wide margin, and almost nobody has ever seen a documentary about it.

That mismatch is the subject of this post. Fear is not a unit of measurement. If we want to compare energy sources, we need a method that counts the same thing for each: how many people die for each unit of electricity produced. Do that, and nuclear power lands near the bottom of the death table, with wind, solar and hydro, and far below coal and oil. The arithmetic is simple. The politics is not, since a smokestack has a lobby and a statistic does not. Here is the argument in four numbered steps, and then the three accidents everybody remembers.

Defining the terms

A terawatt-hour (TWh) is a billion kilowatt-hours. A kilowatt-hour is the unit on your electric bill. A typical American household uses on the order of 10,000 kilowatt-hours a year, as a round figure, so one terawatt-hour is roughly the yearly electricity of 100,000 households. It is the right size of unit for comparing power plants and power sources.

A death rate per terawatt-hour is the number of people who die, over the full life of the source, for each terawatt-hour it produces. The deaths counted include accidents, such as mine collapses and dam failures, and the deaths from air pollution that the source causes in its normal operation.

Attributable deaths are deaths that scientists can trace to a cause with some confidence, as opposed to deaths that are projected statistically from a very small increase in risk spread across a very large population. The difference matters for nuclear, and I will come back to it.

The method

The standard way to make this comparison was set out in a well-known 2007 study by Anil Markandya and Paul Wilkinson, published in The Lancet. They tallied deaths from accidents and from the health effects of air pollution, for each source, per unit of electricity. Their figures put coal at about 25 deaths per terawatt-hour, oil at about 18 and natural gas at about 3. Later compilations that count Chernobyl and Fukushima put nuclear far lower, at a few hundredths of one death per terawatt-hour: Our World in Data, which assumes 433 deaths from Chernobyl and 2,314 from Fukushima and divides by the 96,876 terawatt-hours nuclear plants generated from 1965 to 2021, gets about 0.03. I use 0.07 below, a deliberately generous round figure for nuclear, so the example does not flatter it. The figures are meant to show the order of magnitude. Anyone who wants the latest numbers should look at the published studies and check the figures themselves.

Worked example. Take a hypothetical country that needs 100 terawatt-hours of electricity a year and consider two ways of supplying it. If all of it came from coal at about 25 deaths per terawatt-hour, the arithmetic is 100 times 25, which is 2,500 deaths a year. If all of it came from nuclear at the generous 0.07, the arithmetic is 100 times 0.07, which is 7 deaths a year. The ratio is 2,500 divided by 7, or about 357 to one. The inputs are round figures, and the exact multiple would change with better data. What does not change is that the gap is two orders of magnitude, not a rounding error.

That is the whole method. It is the kind of comparison we make for cars and airlines, where nobody counts the vivid crash and ignores the quiet one.

The argument, numbered

1. Deaths from energy are mostly quiet. The largest toll from fossil fuels is air pollution, the fine particles from burning coal and oil that settle into lungs over years. It does not make the news, because no single death is a headline.
2. Nuclear produces almost no air pollution in operation. It burns no fuel and releases no smoke. Its record on deaths is therefore dominated by a handful of famous accidents, and the accidents were few.
3. Measured per unit of power, nuclear is among the safest sources. The published studies put it in the same range as wind and solar, and far below coal, oil and, by a smaller margin, natural gas.
4. Therefore a policy that closes working reactors and replaces them with fossil plants raises the death toll. That is not a rhetorical flourish. It follows from the numbers, and it has happened.

The strongest case against

A fair opponent has real points, and they should be heard before they are answered.

The first is tail risk. A coal plant's harm is spread thin and predictable, while a nuclear accident, if it goes wrong, can be catastrophic in a way that averages hide. A rate per terawatt-hour is a statistical summary, and people reasonably fear the rare disaster more than the steady drip.

The second is waste. Spent fuel stays radioactive for a very long time, and the United States has not opened a permanent repository. That is a failure of policy, though not of physics.

The third is cost and delay. New reactors have been expensive and slow to build, and critics argue that the money would buy more clean power if spent on other sources.

The fourth is evacuation and displacement. Even where radiation does not kill, a nuclear accident can force people from their homes for years, with real human costs.

These points do not defeat the argument. They set its limits. The death-rate comparison answers one question, how many people die per unit of power, and answers it decisively. It does not answer whether a reactor is cheap, or where the waste goes. Those are real debates, and the case for nuclear has to meet them on the merits. The next section takes the tail-risk objection head on, because it is the one the three famous accidents seem to support.

Three Mile Island, 1979

On March 28, 1979, Unit 2 of the Three Mile Island plant, near Harrisburg, Pennsylvania, suffered a partial meltdown after a cooling malfunction and a series of operator misjudgments. It was the worst commercial reactor accident in American history, and it frightened the country.

What it killed is the interesting part. The Nuclear Regulatory Commission has concluded that the small release had no detectable health effects on plant workers or the public, putting the average dose to the roughly two million people in the area at about 1 millirem, against about 6 millirem for a chest X-ray. Some epidemiological studies have argued for a rise in local cancers, and proving cause in either direction is hard, but no deaths have been shown to result from the radiation. The reactor vessel held nearly all of the radioactive material from the damaged core. The accident did enormous damage to the plant and to the industry's reputation, and, on the official findings, almost none to the health of the neighbors.

The honest lesson is a double one. The accident showed that things can go wrong with a reactor, and the industry responded with training and procedures, a story told in Admiral Rickover's Navy Has Run Reactors at Sea for Seventy Years. It also showed that a design with multiple barriers can fail badly and still harm almost nobody.

Chernobyl, 1986

On April 26, 1986, a reactor at Chernobyl in the Soviet Union exploded during a safety test. It is the worst nuclear accident in history and the one that anchors the public's fear. The facts deserve to be set down plainly and without minimizing.

Twenty-eight people died of acute radiation syndrome within a few weeks. A large area was contaminated and some 350,000 people were evacuated or resettled in all. About 5,000 thyroid cancers have been attributed to the accident, with 15 deaths. Estimates of the eventual toll from radiation are statistical projections rather than counts, the real number is unlikely ever to be known precisely, and they have been argued about for decades, with some estimates higher and some lower.

The key point for an American reader is the design. The Chernobyl reactor was an RBMK, a Soviet design that used graphite to moderate the reaction and had a dangerous property under certain conditions: when the cooling water boiled away, the reaction could speed up rather than slow down. The original RBMK design also omitted a full containment building of the kind that surrounds American reactors, a thick, sealed structure meant to hold in the radiation if the reactor vessel fails. The explosion therefore sent radioactive material straight into the atmosphere. No American commercial reactor is an RBMK, and the safety features that failed or were absent at Chernobyl are exactly the ones American plants are required to have. Blaming a modern American reactor for Chernobyl is like blaming a modern airliner for a crash of a 1950s aircraft with no seat belts and a faulty wing.

None of that makes Chernobyl a small event. It makes it an event about a particular reactor, run in a particular system, one that punished candor and rewarded concealment, which is a lesson about governments and not only about physics.

Fukushima Daiichi, 2011

On March 11, 2011, a magnitude 9 earthquake struck off the coast of Japan and sent a tsunami onto the northeast coast. The earthquake and tsunami killed about 19,500 people, and none of those deaths came from the nuclear accident. It overwhelmed the sea wall at the Fukushima Daiichi plant, flooded the backup generators, and caused core meltdowns in three reactors.

The nuclear accident itself caused no deaths or cases of radiation sickness that have been reported. In 2018 Japan's government acknowledged a link between one former plant worker's lung cancer death and radiation exposure, a single case. What did cause deaths was the evacuation. More than 100,000 people were moved from the area, including hospital patients and elderly residents of care homes, and the stress, disruption and interrupted medical care killed a number of them. Japanese authorities counted 2,313 deaths among evacuees from Fukushima Prefecture classed as disaster-related, as of September 2020. The evacuation's toll is part of that figure, and the lesson is how fear can drive decisions that cost lives.

It would be wrong to conclude that nothing went wrong at Fukushima. Three reactors melted down, the site cost a fortune to stabilize, and a large area was contaminated. The plant's operator had not adequately protected the site against a tsunami of that size. But on the one measure this post is about, deaths per unit of power, the accident adds a small number to the nuclear column and nowhere near what the fossil alternatives add every year.

What happened when reactors were closed

The strongest practical evidence for the argument is what happened when countries acted on the fear.

After Fukushima, Germany accelerated the closure of its nuclear plants. A 2019 working paper by economists Stephen Jarvis, Olivier Deschenes and Akshaya Jha, published by the National Bureau of Economic Research, studied the result. They found that the lost nuclear output was replaced primarily by coal-fired production and net electricity imports, and they put the social cost at about $12 billion a year, with more than 70 percent of it coming from higher mortality risk from local air pollution. This is a single study, and readers who want the figures should read the paper, but its direction matches the death-rate arithmetic above. The wider fight over power prices is touched on in Twelve Questions About Rooftop Solar and Who Really Pays for Net Metering.

The lesson holds as a matter of arithmetic. If you remove a source that kills about 0.07 people per terawatt-hour and replace it with one that kills 3 or 25, you do not need a computer to see which way the toll moves.

Answering the tail-risk objection

The objection is serious, and the answer has four parts.

First, the actual tail has been measured by experience. Over many decades and many reactor-years of civilian operation, the world has had three major accidents. Their combined death toll, including the contested statistical projections for Chernobyl, is far smaller than the annual toll of fossil pollution.

Second, the worst accident involved a design American reactors do not use, and the other two accidents harmed few or no people through radiation.

Third, the safety record has been built by learning. The industry responded to Three Mile Island with training and standards, and the designs of newer reactors rely more on passive safety, which uses gravity and natural circulation in place of pumps and operators.

Fourth, a risk should be compared with its alternatives. The relevant choice is rarely between nuclear and no risk. It is between nuclear and the coal or gas plant that would otherwise supply the power, and that plant has its own steady, quiet toll.

What this does and does not prove

Honesty about limits is part of the argument. The death rates are estimates built on assumptions, and the studies differ in method; there are choices about which deaths to count and how to count pollution. The ranking of sources is robust, with coal and oil far worse than nuclear, but the exact figures are not.

The argument does not show that every proposed reactor is a good investment. Cost, financing, the pace of licensing and the problem of waste are separate questions, and conservatives can disagree about them. It does not argue for ignoring safety, since the numbers are low because of the safeguards that regulators and operators built.

What it does show is that a conversation about energy and safety that leaves out the denominator is not serious. The number of deaths per terawatt-hour belongs in every such conversation. It is also a reminder that real conservation of human life means counting the lives that are lost quietly, in smoggy cities and mining towns, as carefully as the ones lost in dramatic accidents. The Founders asked a related question in the language of their time: not what frightens us, but what actually harms us, and what remedy a free people can apply. For the wider story of how a clean source got stuck in permitting and fear, see We Raised the Empire State Building in About a Year, So Why Not a Power Line and Hundreds of Millions Still Cook Over Wood, and That Is a Moral Question.

The case in brief

1. The fair way to compare energy sources is deaths per terawatt-hour, which counts accidents and air pollution over each source's life.
2. In the published studies coal is near 25, oil near 18, gas near 3 and nuclear under a tenth of one death per terawatt-hour, so on 100 terawatt-hours that is about 2,500 deaths for coal against about 7 for nuclear.
3. Three Mile Island (1979) produced no detectable health effects on the public by the NRC's finding, Chernobyl (1986) involved a Soviet RBMK design without a full containment building, and Fukushima Daiichi (2011) is tied mostly to deaths from evacuation, not radiation.
4. When countries closed working reactors, fossil plants largely replaced them, and the added pollution cost lives.
5. Cost, waste and licensing are separate debates, but any serious conversation about energy and safety has to start with the death rate per unit of power.

What you can do: look up the death-rate figures in the 2007 Lancet study by Markandya and Wilkinson and an updated source, check them yourself, and ask your state public utility commission what plan exists to replace any reactor scheduled to close.