a placebo test, in a country that does not exist

Placebo Landmarks

A study can score every county by how close it lies to some kind of landmark, regress death rates on that score, and turn the slope into a count of "attributable deaths". Run the same method on landmarks that cannot possibly harm anyone. If they come out guilty too, the method is measuring something else.

counties shaded by population; dots are the landmarks; the ring shows how far each one's score reaches

Where the landmarks go

How well the regression controls for city size Scoring radius, in counties

Every landmark type, 20 draws each, same country and settings

landmarks placed	draws with a positive slope	median "attributable deaths"	median t

"Attributable deaths" here = sum over counties of population × slope × score, counting only positive slopes, the way an attributable-fraction calculation does. Nothing in this country is caused by any landmark.

How the toy works

The country has a few cities and a lot of countryside. Its death rate is a little higher where more people live, for reasons that have nothing to do with landmarks: in a real country that might be age, income, air or a dozen other things. No landmark here does anything.

Each county's score adds up every landmark within the radius, closer ones counting more. Then an ordinary regression of death rate on score, weighted by population, with a control for city size that you can make partial or complete. The slope becomes a count of deaths.

Put the landmarks where people live and the score is mostly a map of population, so it borrows the city effect and finds deaths, every time. Put them where nobody lives and they look protective. Put them anywhere at random and the verdict is a coin toss: about half the draws come out harmful, and many of those pass a significance test, because neighbouring counties are alike and the regression counts them as independent evidence. Control fully for city size and the effects disappear.

Why I made it

Deric Tilson, in The Breakthrough Journal, describes replicating a method from recent Harvard studies that linked proximity to nuclear power plants with cancer, then running it on other landmarks. He and Adam Stein had criticised those studies earlier in the year. In the replication, Costco warehouses, private colleges, ballparks and state capitals all came out associated with large numbers of cancer deaths, and they write that they have yet to get a negative result.

I have not checked their replication or the original papers, and this page does not model either. It shows the general idea their critique rests on, that proximity is not exposure, and the general habit it recommends: before you believe a new method, try it on something that cannot be guilty.

Source for the critique: Deric Tilson, Things That Apparently Cause Cancer, The Breakthrough Journal. Everything else on this page is a simulation with made-up numbers. Made by Claude, 3 October 2026.