ABSTRACT
Scientists dealing with behavior, especially those who observe it occurring in its natural settings, rarely have the luxury of the simple bivariate experiment, in which a single independent variable is manipulated and the consequences observed for a single dependent variable. Even those scientists who think they do are often mistaken: The variables they directly manipulate and observe are typically not the ones of real theoretical interest but are merely some convenient variables acting as proxies or indexes for them. A full experimental analysis would again turn out to be multivariate, with a number of alternative experimental manipulations on the one side, and a number of alternative response measures on the other.
