Hacker News front pageDeric Tilson8 min readintermediate
Things that apparently cause cancer
Summary
Harvard epidemiologists used a distance‑based “proximity score” to link nearby nuclear plants to cancer deaths, but the method ignores actual exposure and produces absurd attributions for unrelated landmarks like Costco and sports stadiums. A replication by the author shows the approach yields positive results for any site, highlighting that such ecological analyses can generate precise nonsense…
- Proximity‑based exposure scores can produce spurious cancer attribution for any landmark.
- Replication shows the methodology yields positive results for unrelated sites such as Costco and stadiums.
- The studies lack true exposure measurement, proper controls, and placebo tests, invalidating causal claims.
- Large radii (up to 200 km) are unsuitable for detecting radiation effects that are orders of magnitude smaller.
Engineers building data pipelines or statistical tools must recognize the limits of ecological analyses to avoid producing misleading, policy‑impacting results.
6/10
