← Back

Berkson's Paradox

The Paradox

Diseases A and B are completely independent in the population (like two separate coin flips). But if you only look at hospitalized people (those with at least one disease), A and B appear negatively correlated.

Why? If someone is in the hospital and does NOT have Disease A, they probably have Disease B (otherwise why are they there?). This selection bias creates a phantom negative correlation.

Correlation (r)

Population
--
Hospital Sample
--