Does Costco Cause Cancer?

[ergophobe]

Statistically, using the same methods as a recent study that showed elevated cancer risk for people who live near a nuclear power plant, researchers also demonstrated

- living near Costco raises cancer risk

- living near a 4-year university raises the risk of cancer 15X more than living near a nuclear plant

- living near a major league baseball field causes 19X as much cancer as living near a major league soccer pitch

 - Auto racing fans will be relieved to know that living near a NASCAR track carries a lower risk of cancer than living near Harvard. Using the original methods, Harvard alone accounts for almost as many deaths as all the nuclear power plants put together. Conclusion: Harvard must be stopped.

For reasons they don't understand, the authors of this paper get a positive correlation no matter what type of landmark they choose and even chose random points and applied the same analysis as the authors of the original paper and continued to get positive correlations. They theorize there is something fundamentally wrong with the proximity correlation approach, but they don't know yet what it is (they do note that the original authors used such big radii that they sometimes overlap, causing probabilities to compound)

https://www.breakthroughjournal.org/p/things-that-apparently-cause-cancer

https://marginalrevolution.com/marginalrevolution/2026/10/does-costco-cause-cancer.html

[Travoli]

>there is something fundamentally wrong with the proximity correlation approach

like not accounting for thousands of variables other than proximity that might be driving correlations?

[ergophobe]

like not accounting for thousands of variables other than proximity that might be driving correlations?

It's more than that. If that were the case, you should get spurious positive correlations AND spurious negative correlations (Wal-Mart saves lives!). But they don't. Even with random points, the second team only got positive correlations.

So the key thing is not that these are spurious correlations driven by other variables, but that using the method the original authors used, you will *always* get a positive correlation, so you can tell a story about the villain of your choosing. Because they started from the position that nuclear power plants must be dangerous, they did not check their basic statistical methods to see if they are predictive in *any* situation. And it turns out they are not.

It's like they did a Cheetos and Chocolate weightloss study that used one scale for before and one scale for after, showed remarkable results, and never checked that either scale was properly calibrated, because they knew ahead of time that Cheetos and chocolate would lead to weight loss.