Selection Bias

Selection bias occurs when the procedure that determines who enters or stays in a study distorts the exposure–outcome relationship. It can enter at recruitment (who volunteers, who is reachable) or during follow-up (who drops out). The bias is consequential when selection depends jointly on the exposure and the outcome — formally, when the analysis conditions on a collider.

It matters acutely in population science because many data sources — clinic populations, registries, online surveys, surviving patients — are non-random samples of the public. Findings from a biased sample may not transport to the population the research aims to serve, undermining both internal and external validity.

A frequent pitfall is treating a convenience sample as if it were representative, or ignoring differential loss to follow-up in a cohort. Healthy-volunteer effects, self-selection, and survivor bias are all named species of selection bias. Remedies include population-based sampling, tracking and reporting dropout, and weighting methods such as inverse-probability-of-selection weights.

For example, recruiting cancer-screening study participants only from a wellness fair could oversample health-conscious people, biasing estimates of how screening relates to outcomes in the broader community.