Confounding: Essence, Correction, and Detection
The classical tradition in epidemiology has viewed confounding as a systematic distortion of a true effect by extraneous causal factors. The observed effect is imagined as a sum of the exposure effect and the effects of these confounders. A confounder is defined as a variable that is associated with exposure, is “causally” related to the outcome, and is not an intermediate variable between exposure and the outcome. The collapsibility perspective traditionally favored by many statisticians, on the other hand, does not address the issue of causality directly. Rather, confounding is said to be present if adjustment for additional relevant covariates would cause the estimated effect to change. Failure to observe such change, after adjusting for additional variables, is taken as an indication that confounding is absent. Identifying potential confounders is based on an intuitive determination of the causal factors that may be at work.
In 1981, an article with the title “Confounding: Essence and Detection” challenged the adequacy of the extraneous-variable and collapsibility formulations of confounding (Miettinen and Cook, 1981). Miettinen and Cook proposed that confounding is best understood as a matter of comparability between study groups. We will consider the essence and detection of confounding from a comparability perspective in light of the counterfactual framework. In addition, we have explicitly added consideration of ...