Questions: Confounding: Definition, Identification, and Causal Criteria

Question 1

A study finds that coffee drinkers have higher rates of lung cancer. A researcher proposes that smoking confounds this association. For smoking to be a true confounder, which combination of criteria must it meet?

• A. Smoking must cause coffee drinking and must be more common in the study population
• B. Smoking must be associated with coffee drinking, independently associated with lung cancer, and not lie on the causal pathway from coffee to cancer
• C. Smoking must be measured in the study and must be statistically significant in the data
• D. Smoking must be a stronger risk factor for lung cancer than coffee is

Answer: (B)

All three epidemiological criteria must be satisfied simultaneously: (1) the confounder is associated with the exposure (smokers tend to drink more coffee), (2) the confounder is independently associated with the outcome (smoking causes lung cancer regardless of coffee), and (3) the confounder is not on the causal pathway from exposure to outcome (smoking is not the mechanism by which coffee would cause cancer — it is a separate cause). Meeting only one or two criteria is insufficient.

Question 2

A variable that lies on the causal pathway between exposure and outcome (a mediator) should be adjusted for as a confounder to obtain the true exposure effect.

• T. True
• F. False

Answer: False

Adjusting for a mediator is a serious methodological error — it blocks the very causal mechanism you are trying to measure, leading to an underestimate of the exposure's effect (or no association at all). For example, if physical activity reduces heart disease partly by lowering blood pressure, and you adjust for blood pressure, you remove part of the protective effect of exercise. Mediators are on the causal pathway; confounders are not. The distinction is structural, not statistical.

Question 3

Why is confounding primarily a concern in observational studies rather than randomized controlled trials?

Model answer: Randomization distributes both measured and unmeasured potential confounders equally across comparison groups on average, eliminating systematic imbalances between exposed and unexposed groups. In observational studies, exposure is self-selected or naturally occurring, so confounders can differ systematically between groups — for example, smokers may differ from non-smokers in diet, exercise, occupation, and many other factors, all of which could independently affect the outcome.

This is the fundamental advantage of randomization: it creates balance on all variables, including those the researcher did not think to measure. Observational studies can control for measured confounders through design (matching, restriction) or analysis (stratification, regression), but unmeasured confounders always remain a potential threat to validity.