Collider Bias

“Filtering on a shared consequence can manufacture a relationship between its causes.”

01 / PLAINLY

What it means, plainly

A collider is a variable caused by two other variables in a diagram. Selecting or adjusting for that shared consequence can make its causes look related.

02 / CONTEXT

A little more

A collider receives arrows from two variables. Conditioning, adjusting, or selecting on it can open a previously closed path and create a spurious association.

03 / WHY IT MATTERS

Why it matters

It explains why controlling more variables is not always better and why a filtered sample can display an artificial relationship.

04 / EXAMPLE

A familiar example

Talent and connections affect admission to a selective school; among admits, conditioning on admission can make them look inversely related.

05 / LIMIT

What it does not mean

Not every third variable should be controlled; some must remain unadjusted to avoid opening bias.

06 / NOTICE

Notice it in your day

Draw talent → admission ← connections. Imagine looking only at admitted students and point to the variable used for filtering.

FINAL NOTE

The idea worth keeping

Before filtering, ask whether the variable is a consequence of two relevant processes.

QUESTIONS / 02

Questions people still have

What does a collider look like in a diagram?

It receives arrows from two variables, as in A → C ← B.

What can happen after selecting on a collider?

A spurious association can open between its causes.