Berkson's Paradox

Berkson's paradox is a statistical phenomenon in which two variables that are actually independent, or even positively correlated, in the general population can appear negatively correlated within a subgroup selected using a criterion related to both variables. It is named for the physician and statistician Joseph Berkson, who described it in 1946 in the context of hospital admission data, where he noted that studying only hospitalized patients could create an apparent negative association between two unrelated diseases simply because having either disease increased the chance of admission. The paradox is treated as a specific, statistically well-defined form of selection bias, and it remains a standard cautionary example in epidemiology and other fields that rely on data drawn from a non-random, already-filtered subgroup.