What is another term for interaction in statistical terms?

• A. Confounding
• B. Effect modification
• C. Randomization
• D. Multicollinearity

Answer: (B)

Effect modification is when the association between an exposure and an outcome differs across levels of another variable. This is the sense in which interaction is discussed in statistics and epidemiology—the effect of the exposure changes depending on the level of a second variable. In practice, you test this by including an interaction term in a regression model (the product of the exposure and the modifier). A significant interaction means the exposure’s impact on the outcome isn’t the same across all groups—for example, smoking might have a stronger link to lung cancer in one gender than in the other, indicating the modifier changes the effect. This concept is distinct from confounding, which biases the observed relationship due to a third variable related to both exposure and outcome; from randomization, a design method to balance variables; and from multicollinearity, which complicates estimation when predictors are highly correlated.