Confounding and Effect Modulation

The relationship between a predictor or study variable and an outcome variable may vary according to the value of a third variable, often called a confounding variable or an effect modulator. This entry clarifies the distinction between confounding and effect modulation (also called moderation or mediation) through the use of path diagrams. The statistical tests for establishing these three relationships are somewhat different (main effects model only for establishing confounding variables; main effects model with interaction term for establishing moderating variables and Sobel-like tests based on a series of regression for establishing mediating variables), so they are

discussed separately and their interpretation clarified by example.