Glossary // Glossary term Technical
Confounding
A distortion of an estimated association that arises when a third variable influences both the supposed cause and the outcome. For example, people who talk about their feelings more may also have more social support, so an association between talking and recovery could partly reflect support rather than talking itself. Random allocation reduces confounding by design; in observational data, statistical control can help only when the relevant variables are known, measured well, and are not colliders or mediators. How one could reason about recording a catharsis-like change of state: signals, features, inference and context, baselines, self-report, error costs and ethics, and why every step stays uncertain. Nothing on this site measures anyone. Every public claim on this site with its type, evidence level, confidence, sources and review date, and an appraisal of each reference. How claims are graded, how art is kept apart from evidence, and how to report a problem. The works the definition rests on, with the level of evidence each was appraised at. (2018). Thinking clearly about correlations and causation: Graphical causal models for observational data. doi:10.1177/2515245917745629 Supports statements that correlation does not establish causation, that statistical control depends on assumptions, and that controlling for colliders or mediators can bias estimates. Methods guidance. (2004). A structural approach to selection bias. doi:10.1097/01.ede.0000135174.63482.43 Supports the structural distinction between selection bias, which results from conditioning on a common effect, and confounding, which results from a common cause. Conceptual and methodological; examples come from epidemiology.Where the site uses it
Studies behind it