Confounding variable

A confounding variable changes along with the explanatory variable, so its effect and the explanatory variable's effect on the response cannot be told apart. This is why an observed association does not by itself prove causation. For example, if students who attend tutoring also study more, and tutored students score higher, study time confounds the effect of tutoring. Random assignment defends against confounding by spreading such variables evenly across the treatment groups on average.