Causal Inference

Collider

A collider is a variable that is caused by two other variables — two arrows collide into it. Unlike a confounder, a collider creates no spurious association on its own; the danger appears only when you control for it or select on it.

Conditioning on a collider opens a fake path and can make two genuinely unrelated causes look related. This is a subtle and common mistake: adjusting for more variables is not always safer, because adjusting for a collider introduces bias rather than removing it. Knowing whether a variable is a confounder or a collider requires a causal diagram, not the data alone.

Among hospitalized patients, two unrelated diseases can appear negatively linked — selecting on admission (a collider) creates the illusion.