Confounding Variables
Confounding: compare like with like
An association answers a descriptive question: how do observed groups differ? A causal question asks what would happen to the same target population under different exposures. Those questions coincide only with strong design or analysis assumptions.
A variable is a confounder when it precedes the exposure, is associated with exposure, and independently affects the outcome. For example, in a study of a revision workshop and grades, prior preparation can influence workshop attendance and grades.
It is not a mediator if it existed before the workshop, and it is not an outcome measured after grading.
Random assignment tends to balance baseline common causes. In observational work, restriction, matching, stratification, standardization, or an appropriately specified adjustment model can make measured groups more comparable. These methods need correctly measured pre-exposure variables and do not turn a precise association into a proven causal effect when an important common cause is absent.
Investigators study whether a campus reminder program increases influenza vaccination. The program sends reminders, and students who receive them are more likely to book an appointment. Which variable is best described as a mediator rather than a confounder?
- Booking an appointment after receiving a reminder
- Age measured before the program begins
- Prior attitudes toward vaccination measured before enrollment
- Whether vaccination was recorded at the end of the season
A clinic compares a wellness class with no class. Among patients younger than 50, 18 of 90 class participants and 24 of 120 nonparticipants reach a target. Among patients 50 or older, 12 of 60 class participants and 24 of 120 nonparticipants reach the target. What does this table support?
- The class effect is proven because 30 class participants reached the target.
- Age is irrelevant because the overall class-participant success rate is 20%.
- Within each age stratum, the observed success proportions are equal: 20% versus 20%.
- The class lowers success because there are more nonparticipants than participants.