Optimize

The second externality

Electricity kills people — unevenly

Every source of power carries a death toll: mining and drilling accidents, and — far larger — the air pollution that shortens lives downwind. Per unit of energy, coal is more than 1,000× deadlier than solar — three orders of magnitude. The electricity market never charges for any of it. Optimize lets you put a price on it.

The headline result

Put a price on either harm — carbon or mortality — and the coal-heavy Midwest is transformed: coal collapses and grid deaths fall about 95%, from ~2,232 to a few dozen a year. It barely matters which harm you price.

2,232
No pricing
deaths / yr
52
Carbon $400
deaths / yr
110
Mortality $21.5M
deaths / yr

The subtler, more interesting part is the second act — where the two prices disagree. They part ways on gas, shown below.

The risk ladder

, by source. The scale is logarithmic — each step is a multiple, not an addition — because the sources span three orders of magnitude.

Deaths per TWh · log scalecounted (accidents)modeled (pollution/radiation)1 TWh ≈ powering 93k homes for a yearn/a= not among Optimize's optimized technologies (rides along with demand) — still carries a real death toll, counted in the Data Explorer's read of the actual grid, but never in an optimized scenario's mix

Counted deaths are recorded accidents — a mine-shaft collapse, a rig fire. Modeled deaths are air-pollution and radiation attributions from epidemiological models — a fatal heart attack months after chronic particulate exposure, which no coroner labels “coal.” They're shown distinctly (solid vs hatched) throughout.

The most surprising number: nuclear

Coal
24.6 /TWh
Nuclear
0.03 /TWh

hundreds of times safer than coal — accidents included

Nuclear's 0.03deaths/TWh counts every death from every accident — Chernobyl and Fukushima among them. The toll from electricity is dominated by coal's everyday air pollution, not by rare disasters.

A carbon price and a mortality price are the same object

Both take a harm the market doesn't charge for, attach a dollar figure to it, and let the optimizer respond. Optimize already prices carbon. Mortality is one more coefficient, one more slider, one more linear term in exactly the same cost function.

Carbon price

$ / ton CO₂

Multiplies each source's CO₂ intensity. Coal and gas get steadily more expensive as the price rises.

Mortality price

$ / death

Multiplies each source's death rate. Coal is hit hardest; gas much less, because gas is ~9× cleaner on deaths.

They don't fully agree. Coal exits under either. Gas is where they part ways — ~9× cleaner than coal on deaths but only ~1.7× cleaner on CO₂, so per MWh a mortality price barely touches gas while a carbon price leans on it hard. Over the full 25-year horizon that difference reaches the built mix: a carbon price drives gas far lower than a mortality price does (shown below). Move both sliders in the playground to explore.

What is a life worth?

To price mortality you need a dollar value per death — a . HHS's 2026 regulatory guidance publishes a range, not a single number (constant 2025 dollars):

$6.6M
Low
$14.1M
Central
$21.5M
High

That number comes from what people actually pay to avoid risk — wage premiums for dangerous jobs, spending on safety — scaled up to one statistical death. It is not the price of your life, or anyone's in particular.

Assigning a dollar value to a death is a moral choice you are making, not a technical parameter the model resolves. The VSL figures are HHS's published range, shown as published values — never as endorsement.

Outputs are framed as what would have to be true for this to be the cheapest grid — not as a recommendation of any particular number.

Every number here is a band

Deaths are reported as a central estimate with a high-side band, never a single figure. The band is real scientific uncertainty from two independent places: how deadly each source is (coal's air-pollution toll is modeled, not counted), and what a life is worth (the VSL is itself a low/central/high range). A priced-mortality cost inherits both — so any dollar figure that includes mortality is a band too. The honest read is the width, not the midpoint. How the bands are built →

Deaths are only part of the harm

The health toll of burning fossil fuels isn't only fatal. For every death there are far more non-fatal harms — asthma attacks, heart and lung hospitalizations, lost workdays — that this model doesn't price. Counting deaths alone therefore understatesthe burden. The control below makes that explicit: it doesn't change the model, it shows how much larger the monetized health cost would be if morbidity were added on top.

The model prices deaths only. Turn on an uplift to see morbidity — the illnesses that don't kill — added to the monetized health burden.

Illustrative. EPA's PM2.5 benefit-per-ton methodology monetizes morbidity as a share on top of mortality; the exact fraction depends on which endpoints are counted and is uncertain — which is why it's yours to set, and why the default is off.

What pricing mortality actually does

Take a coal-heavy region and set the mortality price to the central VSL, $14.1M. Here is what the optimizer now sees:

Coal

~$347/MWh

Coal's 24.6 deaths/TWh, priced at the central VSL, adds about $347 per MWh — many times what it costs to actually run the plant (a few tens of dollars). Coal can't compete, and the optimizer stops rebuilding it.

Gas

~$39/MWh

Gas's 2.8 deaths/TWh adds only about $39 per MWh — comparable to its running cost, not a knockout. Gas takes a hit but survives, and often fills in as coal leaves.

That's the disagreement in one comparison: per MWh, the same price is a death sentence for coal and a tax for gas. A carbon price, which sees coal and gas as far more alike, weighs them differently.

Coal exits under either price — but gas is where they diverge
No pricingcoal 80 · gas 88 TWh · 2,232 deaths
Carbon only ($400/ton)coal 0 · gas 13 TWh · 52 deaths
Mortality only ($21.5M)coal 0 · gas 32 TWh · 110 deaths
Bothcoal 0 · gas 9 TWh · 41 deaths
SolarWindNuclearGasCoalBattery
Final-year mix for MIDW over 25 years, from the pre-computed grid. Coal collapses and grid deaths fall ~95%+ under either price — that's the robust result. Now look at gas: the carbon price pushes it far lower than the mortality price does. The two prices genuinely disagree about gas — carbon-heavy but comparatively low-mortality — and over the full horizon that disagreement reaches the built mix, not just the marginal cost ($347 vs $39/MWh). Move both sliders in the playground.
Where these numbers come from

Optimize does not invent death rates. They are imported from Level, which draws on Our World in Data's analysis of the safest sources of energy. That analysis compiles the published epidemiological literature — air-pollution mortality from concentration-response studies, and accident death tolls from energy-accident databases. Optimize is not a second source of truth for these figures; every coefficient on this site links back to its Level source page, and Level to the original research.

What could make these numbers wrong

Stated plainly, because pretending otherwise would be the real error:

  • They're global averages. US coal with modern scrubbers is likely somewhat safer than the global fleet these figures describe. Trust the ranking more than any decimal.
  • Air-pollution death is modeled, not counted. The bulk of the toll comes from statistical attribution, which depends on exposure and dose-response assumptions that reasonable experts dispute — hence the wide upper bands.
  • Future deaths aren't discounted. A death in 2050 is weighted the same as one today; a different choice would change priced costs.
  • VSL is contested. The very idea, and the specific number, are debated on ethical and methodological grounds. That debate is the point — it's why the price is yours to set.

Where the deaths land

CO₂ is global; air-pollution deaths are local. Optimize reports production-baseddeaths, attributed to the region that generated the power. The model optimizes each region on its own and holds transmission between regions at its historical level, so it doesn't track where power ends up — a consumption-based view (deaths where the power is used) would need an explicit inter-regional flow model. Why, and what would change it →

Deaths are an accounting attribution to the generating region, not an atmospheric model — real pollution crosses regional boundaries, harming people on both sides of a border. The uncertainty band propagates into any priced-mortality cost, so a priced total is a band, not a single number.

Questions people ask

Isn't putting a dollar value on a life wrong?
You already do it — every time society declines to spend infinitely on safety, it has implicitly priced a life. A speed limit, a hospital budget, a workplace-safety rule all embed a number. This tool just makes the number explicitand lets you choose it. That choice is yours and it is moral, not technical — the model doesn't resolve it for you.
Why is coal's rate so uncertain — 24.6, up to 224?
Almost all of coal's toll is modeled air-pollution death, not counted accidents, and modeling it means choosing exposure and concentration-response assumptions. The central figure uses mainstream assumptions; the high bound uses more pessimistic ones. The width of that band is honest uncertainty, not sloppiness.
Why doesn't gas exit under a mortality price, the way it does under carbon?
Gas is about 9× cleaner than coal on deaths but only about 1.7× cleaner on CO₂. So a mortality price crushes coal while leaving gas relatively cheap — the optimizer keeps gas and lets it fill in. A carbon price treats coal and gas more alike, so it pushes past gas toward renewables. That disagreement about gas is the whole point of pricing both.
Where are Chernobyl and Fukushima in nuclear's 0.03?
Included. Even counting every death from every nuclear accident, nuclear comes out around 0.03 deaths/TWh — roughly 800× safer than coal. The toll from electricity is dominated by coal's ordinary, everyday air pollution, not by rare disasters. The number isn't hiding the accidents; the accidents are just small next to the smog.
Are these US numbers?
No — they're global averages. US coal with modern scrubbers is probably somewhat safer than the global fleet, so treat these as order-of-magnitude figures, not precise US values. Getting the ranking right (coal ≫ gas ≫ nuclear ≈ renewables) matters more here than any single decimal.
Why only deaths — what about illness?
Morbidity — asthma, hospitalizations, lost workdays — is real and far larger in count than death. It's also harder to value consistently, so it's deliberately out of scope for this first cut. Excluding it means these figures understate the full health burden.
Why does battery have no death rate?
A battery generates nothing of its own; it time-shifts electricity some other source already made, and that source's deaths were counted when it generated. Giving the battery its own rate would double-count.

The numbers here are imported from Level, the descriptive reference for these figures. Optimize turns them into a lever.

Price mortality in a custom run →