Whose plants, whose air — source state × receptor state

Attributed deaths per year, source state → receptor state

Attributed deaths per year, per pair

↘ the diagonal — a state's own air exactly zero (risk threshold) selection

Bin edges are quantiles of the — pairs outside those columns (25th, 50th, 75th, 90th, 97th), not equal steps — the values span five orders of magnitude, so equal steps would paint almost the whole matrix as one colour. The top bin is open-ended — the largest pairs (see the ranking at right) sit far above the last edge. Cyan marks selection only.

The zeros are a threshold artifact, not missing data

Every source reaches every receptor in this SR matrix, so there is no such thing as no-data here. The — all-zero columns (—) are states where the modeled baseline PM_2.5 sits below GEMM's 2.4 µg/m³ risk threshold for the whole population, so the remove-form calculation returns exactly zero however much ΔPM_2.5 arrives. That is not "no coal impact" — Wyoming receives among the highest ΔPM_2.5 in the country. A further — pairs outside those columns fall below the payload's emitted precision — drawn at the bottom of the ramp, because they are tiny, not absent.

What a cell is

A cell is a share, not a marginal effect. Retiring one state's fleet alone would avoid a smaller number than its row shows, because GEMM is non-linear in concentration and per-source marginal effects do not add up. Shares do add up exactly — which is why they are used, and why the row and column sums reconcile to the national payload to under a hundredth of a death.

No selection — the whole matrix

How to read this

Honesty ledger — assumptions & limits

• How a cell is attributed:

• Framing:

• Concentration–response function:

• Transport realism:

• What a row total covers:

• Non-US remainder, and the rest-of-world tail:

• GEMM risk threshold (the zero columns):

• y0 (baseline mortality rate) is per receptor state:

• All-ages population convention: National 25+-consistent correction factor ≈ — which is why the national figure is shown as a band.

• Present burden, not a repeal forecast: this page shows what the currently-operating US coal power plants' real 2025 emissions contribute today. The separately-labelled §111-repeal delta scenario lives on the national tile, not here.

• Plant names are withheld:

• VSL is illustrative: VSL = (2023 USD, EPA-anchored central value) — prefer a range around the central figure, never "cost of deaths" or "lives worth."

• Whole-cell population attribution:

• Climate TRACE input-layer caveats:

• No demographic or environmental-justice view — and why: there is zero demographic data on this box, so nothing here speaks to who bears the burden by race, income or vulnerability. Building that honestly would need US Census/ACS race-and-income tables, a social-vulnerability index, and a defensible population-weighted (dasymetric) transfer from this variable-resolution receptor grid down to tract level. It would also need its own caveats: Not attempted here rather than approximated.

More about this matrix

What a cell means, in full

Modeled counterfactual, not a body count. Every cell answers one question: if the coal plants in the row's state stopped emitting and the reduction were sustained, how many fewer steady-state annual deaths would we estimate in the column's state? Each cell is that source state's share of the receptor state's modeled burden by concentration — an attribution rule, not a marginal or leave-one-out estimate. No plant is named: state aggregation is the deliberate ceiling.

Rows: the — states hosting coal power plants, ordered by total attributed burden. Columns: all — receptor states — the first — repeat the row order, so in-state cells fall on the diagonal; the remaining — host no coal power plants and follow by burden.

How much of the burden crosses a state line

Share of the modeled burden that crosses a state line

—of attributed deaths land outside the source state

— states host the — currently-operating US coal power plants (in — source grid cells); their emissions are attributed across — receptor states — — source→receptor pairs, every one of them drawn. Only — of the burden stays in the state that hosts the plants.

Read as a lower bound on transport. The SR response is single-layer and ground-level, so it over-concentrates ΔPM_2.5 near the stack and thins the far field — the true out-of-state share is larger than the one drawn here, not smaller.

Stays in-state

—

the marked diagonal — — deaths/yr

Crosses a state line

—

— deaths/yr, off-diagonal

Widest exporter

—

— of its burden lands elsewhere

Largest cross-state pair

—

— deaths/yr

The national figure this matrix sits inside

—
25+-adjusted deaths/yr —
as-specified deaths/yr

Illustrative health damages —–— a year (VSL-valued, never a cost of deaths). The matrix's own universe is the US-matched receptor subtotal, — deaths/yr — a subset of that band, because a further —/yr (—) lands on non-US receptors and is off-matrix.

GEMM is an upper-bound curve. These counts use the GEMM concentration–response function (Burnett 2018), which runs ~30–70% above the MR-BRT central estimate — read the figures as a sensitivity ceiling, not a point forecast.

The biggest flows, and who imports the most