What is confounding?

What makes a variable a confounder

A confounder must satisfy three classic conditions: it is associated with the exposure; it is an independent risk factor for the outcome (a cause of the outcome in its own right); and it is not an intermediate step on the causal pathway from exposure to outcome. A standard illustration is the association between carrying matches and lung disease: smoking is associated with carrying matches and independently causes lung disease, so it confounds the apparent match–disease link. Recognising the third condition matters, because adjusting for a variable that lies on the causal pathway can wrongly remove part of the real effect.

How confounding distorts findings

Confounding can mislead in several directions. It can produce a spurious association where none truly exists, exaggerate or shrink a genuine association, or even reverse its apparent direction. This makes it one of the central reasons that an observed association is not enough to establish causation. Distinguishing confounding from selection bias and from chance is part of the routine critical appraisal of any observational study, and it is a key consideration when interpreting measures such as relative risk and the odds ratio.

Controlling for confounding

Confounding can be tackled at the design stage and the analysis stage. In design, randomisation (in trials) tends to balance confounders — known and unknown — across groups; restriction limits the study to one level of the confounder; and matching aligns the groups on it.

In analysis, stratification examines the association within levels of the confounder, and multivariable adjustment (such as regression) statistically holds measured confounders constant. A key limitation is that these analytic methods can only address confounders that have been measured — residual and unmeasured confounding can remain, which is why randomised designs are valued where feasible.

Confounding and causal inference

Confounding is central to causal inference: assessing it is one of the considerations, alongside chance, bias and the strength and consistency of evidence, that epidemiologists weigh before interpreting an association as causal. Modern approaches use causal diagrams (directed acyclic graphs) to reason explicitly about which variables to adjust for and which to leave alone, helping avoid both under- and over-adjustment. Transparent reporting of how confounding was handled — which variables were considered and adjusted for — is required by guidelines such as STROBE and is essential to credible findings.