A neighborhood has an unusual number of cancer cases. Is something there causing them?

It feels that way. But random events clump together by chance, and with thousands of neighborhoods to look at, some are bound to look unusual. Spotting a cluster after the fact is like the Texas sharpshooter, who fires at a barn and then paints a target around the tightest group of holes.

Suspected cancer clusters investigated in the U.S., 1990–2011

Investigated
567
cancers of concern, in 428 investigations
Really higher than expected
72
about 1 in 8
Clear cause found
1
3 were linked to a suspected cause, and only 1 of those clearly

The Texas sharpshooter's barn

70 holes placed completely at random. Then the target was painted around the tightest group.

A red barn wall with 70 bullet holes scattered completely at random. A target has been painted afterward around the tightest group, where 8 holes fall inside a circle that would hold about 3 on average, so the shooter looks like a marksman.

Inside the painted target: 8 holes, where random scatter puts about 3 on average. Nothing aimed them there.

What happened when real clusters were checked

567 suspected cancer clusters investigated by U.S. state and federal agencies, 1990–2011

Bar chart: of 567 cancers of concern investigated, 72 were confirmed to be higher than expected, 3 were linked at least partly to a suspected cause, and 1 had a clear cause. Cancers of concern investigated 567 Confirmed to be higher than expected 72 Linked, at least partly, to a suspected cause 3 With a clear cause 1

A pattern you find after looking everywhere isn't evidence the way a prediction is. Random events don't spread out evenly; they clump. Look across thousands of neighborhoods and some will have far more cases than average by chance alone. That's why most suspected cancer clusters turn out not to be real, and even among the real ones, a cause is rarely found.

The same trap catches "hot streaks" in sports, lucky stock pickers, and towns where "everyone" seems to get the same illness. Before trusting a striking pattern, ask whether someone predicted it before looking, or drew the target after the shots were fired.

What pattern have you noticed that might look different if you counted all the places it didn't happen?

Does this mean cancer clusters should be ignored?

No. Real environmental causes of cancer do exist, and health agencies take reports seriously. But with small numbers of cases in each neighborhood, random ups and downs are large, and people naturally notice the places where cases pile up rather than the far more common places where they don't. The review behind these numbers concluded that decades of cluster investigations had rarely found a cause, and that better methods, like watching whole populations over time, are more likely to find real risks.

The barn above was made by placing 70 dots at random with a fixed seed, then drawing the target around the circle that happened to hold the most dots. Did shark attack deaths really double? shows the same luck at work in small yearly counts.

Related exhibits

Sources: M. Goodman et al., "Cancer clusters in the USA: What do the last twenty years of state and federal investigations tell us?" Critical Reviews in Toxicology, 2012 (428 investigations of 567 cancers of concern, January 1990 to September 2011). Barn graphic is a random simulation.