Confounding and Confounding Control

AbstractCurrently, the classical, counterfactual, and structural confounding definitions, are simultaneously used in teaching or practice context of epidemiology. Despite the resemblance, the three definitions are different in many aspects. A unified theory of confounding and confounding control has not been established. At the same time, the distinction between several confounding-related concepts is not clear-cut. Inspired by the structural causal model, this paper puts forwards an improved confounding definition to unify previous confounding definitions. The improved definition defines confounding as the association between exposure and the risk factor for the outcome, which cannot be interpreted by the causal effect of exposure on the risk factor. In so doing, confounding control methods is derived naturally, and several confusing confounding-related concepts is distinguished clearly.