Archived release. You are reading this finding frozen at the data release below — the numbers here will never change; this is the permalink to cite. The always-latest version is at the latest page.
Surprising findings · a risk-scale contrast, stated in the open
Air toxics and the years between the two ends of the risk scale
Rank 21,366 US ZIP codes by EPA-modeled air-toxics cancer risk and compare the two ends: the tenth with the most modeled risk (2,133 ZIP codes) lives 1.36 fewer years than the 7,132 with the least. And it is not an income story: the higher-risk tenth is also the higher-income group — $85,553 against $78,107 — so income runs against this gap, not with it. Two records about the same places, described together; never a causal claim.
Air toxics: EPA AirToxScreen, 2020 (modeled)Life expectancy: CDC USALEEP, 2010–2015Income: Census ACS, 2020–202421,366 ZIP codes compared2,133 and 7,132 ZIP codes at the two ends
What a modeled air-toxics number actually is
EPA’s AirToxScreen models lifetime cancer risk from air toxics for every part of the country — counting industrial and mobile sources alongside natural ones (wildfire and vegetation chemistry) and pollutants formed in the air itself. “Cases per million” means: if a million people breathed this air for a lifetime, the model estimates that many extra cancer cases. It is a model of exposure, not a measurement of anyone’s health — and this page never claims a resident’s illness came from their air.
Least modeled air-toxics risk (7,132 ZIP codes)
modeled risk 19.1 per million on average · median household income $78,107 · life expectancy 79.69 y
Most modeled air-toxics risk — the top tenth (2,133 ZIP codes)
modeled risk 43.1 per million on average · median household income $85,553 · life expectancy 78.33 y — 1.36 fewer years
The two ends are cut at 20.0 and 37.3 modeled cases per million lifetime risk — on whole shared values only, so the 6,091 ZIP codes sharing the 20.0 boundary value are never split, which is why the low group is larger than a strict tenth. The income comparison is in the panels: the higher-risk tenth is also the higher-income group, so income runs against this gap, not with it.
The highest modeled air-toxics cancer risk among ZIP codes with 5,000+ residents
| ZIP code | Residents | Modeled cancer risk (per million) | Estimated life expectancy (95% range) | Median household income |
|---|---|---|---|---|
| 97038Molalla, OR | 16,440 | 219 | 77.6 ± 1.4 | $89,564 |
| 95005Ben Lomond, CA | 7,277 | 195 | 78.2 ± 1.3 | $104,375 |
| 97023Estacada, OR | 12,111 | 195 | 77.7 ± 1.8 | $76,762 |
| 95006Boulder Creek, CA | 7,561 | 190 | 80.5 ± 1.7 | $102,167 |
| 70084Reserve, LA | 5,629 | 157 | 73.6 ± 1.8 | $62,047 |
| 95018Felton, CA | 6,594 | 135 | 80.9 ± 2.2 | $129,464 |
| 90270Maywood, CA | 24,218 | 103 | 78.5 ± 1.5 | $61,165 |
| 97055Sandy, OR | 19,786 | 102 | 78.6 ± 1.3 | $112,323 |
| 50111Grimes, IA | 16,865 | 96 | 81.0 ± 1.3 | $120,769 |
| 97045Oregon City, OR | 57,526 | 93 | 78.4 ± 1.0 | $103,573 |
Modeled total risk counts every source — industrial corridors (70084, Reserve LA, is in the lower-Mississippi petrochemical corridor) and heavily forested areas (where natural and secondary formaldehyde chemistry dominates the model) appear side by side. That is why the finding is the end-versus-end contrast, never any single ZIP code’s rank.
ZIP codes with fewer than 1,000 residents, or missing any of the three measures (modeled risk, life expectancy, income), are not compared at all — the contrast uses only the 21,366 ZIP codes clearing both bars. 17,193 compared ZIP codes are not named in the table: they sit outside the two compared groups, below the population floor, or above the uncertainty ceiling. The full dataset — including the lowest-risk group’s named rows — is in the CSV.
Why we hold this finding to a stricter standard
The association here is about the same strength as one we know is confounded (radon exposure appears to “track longer lives” at similar strength in this data — a geography artifact, and obviously not causal). So this page leads with the end-versus-end contrast with the income comparison in the open, keeps every claim descriptive — two records about the same places — and publishes this caution beside the finding rather than in a footnote.
Cite this finding
The permalink /insights/findings/air-toxics-life-expectancy/2024-52826046 resolves to this exact release and never changes — it will keep serving these exact figures after the data updates.
How this was computed
- Sources and measurement windows. Air toxics: EPA AirToxScreen modeled lifetime cancer risk, 2020 release (a model of emissions and air chemistry — it does not read demographics). Life expectancy: CDC USALEEP, 2010–2015. Income: Census ACS, 2020–2024.
- Universe. 21,366 ZIP codes with at least 1,000 residents carrying all three measures.
- The comparison. Rank by modeled risk and compare the two ends, cutting only on whole shared values: 74.8% of modeled risk values are whole numbers, and 6,091 ZIP codes share the value at the low boundary — so ZIP codes with the same modeled risk always land on the same side of a cut, and the lowest-risk group (everything at or below 20.0 cases per million — 7,132 ZIP codes) is larger than a strict tenth. The top tenth is everything at 37.3 and above (2,133 ZIP codes). Both groups’ average life expectancy and income are then compared directly.
- The result. 79.69 vs 78.33 years — 1.36 fewer years at the high-risk end. The higher-risk tenth is also the higher-income group ($85,553 vs $78,107), so income runs against this gap, not with it. The association also holds within states, at modest strength (a correlation of -0.142 with income held constant). Named rows in the table additionally require at least 5,000 residents and a published standard error of 2.0 years or less (the ± figures shown are 95% ranges — 1.96 × that standard error).
- The caution, stated against ourselves. A same-strength association exists in this data for radon in the implausible direction — so association strength alone cannot carry this page, and we say so above.
- What this does and does not claim. A descriptive comparison of published estimates and modeled values across places — never a causal claim, and never a claim about any individual resident.
- Reproduce it. The full dataset is the CSV above; the method is published in full on the methodology page, and every threshold used here is stated in the steps above.