Where the burden falls — county count vs rate

Nine counties: deaths (count) vs deaths/100k (rate)

Two bars per county. Both figures are upper-bound modelled estimates of deaths that would be avoided, not recorded deaths.

0255075100 estimated values

How to read the two bars

Rows are fixed in count order, largest county-total first. Both bars are simple size comparisons against the Bay-wide largest — not a claim that one county causes deaths in another. The gap between a county's two bars is how far it travels between "how many people" and "how much risk per person": the wider the gap, the more the two lenses disagree about that county.

Same nine, ranked by rate

How to read the hollow markers

Bars are sorted by deaths/100k (risk per person), longest at top. The hollow marker on each bar shows where that county's count would fall on the same 0–100 scale (its count as a percentage of the highest county's count) — when the marker sits far from the bar's end, the county's rate rank and count rank disagree.

0255075100 rate · count rank

Which counties move most between the two lenses

What this estimate assumes and leaves out

Honesty ledger — assumptions & limits

• Population-only, not demographic. We have modeled population per cell — no race, income, or other demographic data. "Where the burden falls" describes where deaths and risk concentrate by headcount, not an environmental-justice or equity finding.

• Baseline death rate per 100,000 people per year ( per person per year, the y0 term in the health function) — real baseline mortality (IHME GBD 2023: California non-communicable disease + lower-respiratory infection, ages 25+, crude), applied uniformly across Bay cells.

• Deaths are a counterfactual estimate under full removal of modeled sector PM2.5, sustained.

• Hub sectors () pin mobile/diffuse emissions to one cell — spurious single-cell hotspots. Excluded from the PHYSICAL default; ALL is a labeled sensitivity.

• GEMM (Burnett 2018) is the upper-bound concentration-response sensitivity; the recommended MR-BRT function runs 30–70% lower — county ranking is not expected to change, magnitudes would shrink.

• Count and rate are normalized independently to their own Bay-wide maximum (each axis reads "percent of the largest county's value", not a share of anything anyone owns) — the difference between a county's two bars is a comparison of two different ranking lenses, not a single physical unit.

Data: mortality_cells.json → counties[] (9 rows), meta.totals. Case: . VSL/damages are on mortality-vsl.html; the ranking-stability argument is on mortality.html.