Selection bias
Who ends up in your sample is tied to the very thing you're measuring.
What it means
Selection bias arises when the process determining which units are observed is associated with both the exposure and the outcome, distorting the estimated relationship. It can enter through how a sample is recruited, who agrees to participate, who is lost during a study, or which records survive. Because the distortion is built into the data-generating process, more data of the same kind does not fix it. In causal-graph terms, selection bias is often collider bias induced by conditioning on inclusion in the study.
Examples
Surveying gym members about exercise habits overstates the population's activity, because the unfit rarely joined the sampling frame.
Online product ratings cluster at five stars and one star because delighted and furious buyers bother to write; the indifferent majority never enters the sample.
A drug looks well tolerated in a trial where the patients who suffered side effects quietly dropped out before the final assessment, leaving only those who coped.
First described in Heckman (1979); causal formalization via Hernán, Hernández-Díaz & Robins (2004).