Selection Bias

Selection bias arises when the way units enter your sample is related to both the treatment and the outcome, so the sample no longer represents the population you care about. Who is in the data is itself informative in a way that distorts the comparison.

It is everywhere: survey responders differ from non-responders, survivors differ from those who dropped out, and customers who churn never fill in the satisfaction form. Selection bias is often a collider problem in disguise — selecting on a common effect — and unlike sampling noise, collecting more data the same flawed way does not fix it.