What is confounding, and how can stratification help control it?

• A. Confounding is a mediator in the causal pathway; stratification eliminates it.
• B. A third variable is associated with both exposure and outcome; stratify by confounder levels to compare within strata.
• C. Confounding means measurement error; stratification has no effect.
• D. Confounding is random variation; stratification controls by randomization.

Answer: (B)

Confounding happens when a third variable is related to both the exposure and the outcome, distorting the apparent relationship between them. For example, if older age is linked to both higher exposure to something and higher risk of a disease, age can make the exposure–disease link look stronger (or weaker) than it really is.

Stratification controls this by splitting the data into groups based on the confounder (like age categories) and comparing the exposure–outcome relationship within each group. By holding the confounder constant inside each stratum, you remove its mixing effect and get a clearer view of the true association. If the exposure–outcome relationship looks similar across all strata, it suggests the crude association was confounded by that variable. If it varies by strata, there may be effect modification.

Remember, stratification works when you can measure and categorize the confounder and have enough data in each stratum; it doesn’t address confounding from unmeasured variables.