Berkson's Paradox

Berkson's Paradox is a statistical phenomenon where two variables that are unrelated — or even positively correlated — in the general population appear negatively correlated when you only observe a selected subset. The bias emerges from conditioning on a collider: when your sample is filtered by a criterion that both variables influence, a spurious inverse relationship appears. Classic example: among published actors, attractiveness and talent may appear negatively correlated — not because they're actually opposed, but because getting published requires enough of either, creating the illusion that having more of one means less of the other. The paradox is a warning about drawing conclusions from filtered pools.

Whenever you notice an apparent negative correlation between two desirable traits, ask: am I looking at a filtered sample? If your data comes from a pool that was selected based on a criterion influenced by both variables, the negative correlation may be an artefact of selection, not a real relationship. Test this by checking whether the correlation holds in the unfiltered population. If it doesn't, you've found Berkson's Paradox.

Berkson's Paradox creates false negative correlations in filtered samples. When both variables influence the selection criterion, an inverse relationship appears that doesn't exist in the unfiltered population. Always check whether your sample was pre-selected.