confounder

A confounder is an extraneous variable in a study that is associated with both the exposure being investigated and the outcome of interest, and is not part of the causal pathway linking the exposure to the outcome. Confounding can distort or obscure the true relationship between exposure and outcome, leading to erroneous conclusions about causality.

To qualify as a confounder, a variable must meet three criteria: it is associated with the exposure, it is independently associated with the outcome, and it is not a mediator on the causal path from exposure to outcome. For example, in observational studies of physical activity and heart disease risk, age can be a confounder because older individuals may be less physically active and also have a higher risk of heart disease. If age is not adjusted for, the estimated effect of activity on heart disease may be biased.

Control and adjustment for confounding can occur at design and analysis stages. Design-based methods include randomization in experiments, restriction to certain categories of the confounder, and matching participants on confounder values. Analysis-based methods include stratification by confounder, multivariable regression, and propensity score techniques. Directed acyclic graphs are often used to identify potential confounders and to specify appropriate adjustment sets.

Limitations remain: confounding can persist due to unmeasured or poorly measured variables (residual confounding). It is important to distinguish confounders from mediators, which lie on the causal path, and from colliders, which can introduce bias if conditioned on.