Among hospital patients, two unrelated diseases can look like they prevent each other
Studying only hospitalized patients, early researchers found that people with diabetes seemed less likely to also have gallbladder disease than expected, as if having one somehow protected against the other. No such link shows up in the general population. Why would looking only at hospital patients manufacture a negative correlation between two diseases that have nothing to do with each other?
Reveal the answer
Statistician Joseph Berkson identified the trap in 1946: if either disease alone is enough to get someone admitted to hospital, then among admitted patients, having one disease makes it less necessary to also have the other to explain their admission, manufacturing an artificial negative correlation. Berkson's paradox is now a standard warning about selection bias whenever a sample is conditioned on an outcome related to the very variables being studied.