Confoundingin

Confounding is a bias that occurs in statistical analyses and observational research when the observed association between an exposure and an outcome is distorted by a third variable, the confounder, that is associated with both the exposure and the outcome. Confounders are not mediators; they influence both sides of the association, potentially creating a spurious link or masking a real one. For instance, older age may confound the relationship between physical activity and heart disease risk because age influences both likelihood of being active and disease risk.

Approaches to address confounding include design choices such as randomization, restriction, and matching, which reduce or eliminate confounding at the outset. In analysis, researchers adjust for measured confounders using multivariable regression, stratification, or propensity score methods, and may apply inverse probability weighting. Instrumental variable analysis can help when randomization is not feasible. Despite these methods, residual confounding from unmeasured or mismeasured variables can persist, and overadjustment or collider bias can occur if adjustments are inappropriate. The term confoundingin itself is not standard; discussions typically center on confounding and on strategies to mitigate it in observational and quasi-experimental studies.