ABSTRACT
During the early development of experimental design and analysis techniques, it was recognized that some of the requirements (sometimes called assumptions) necessary to perform a proper analysis of data from a designed experiment could not be fulfilled. The principal requirement that was in jeopardy was that of independence of the error variation throughout the experiment. If the error is dependent upon, say, the order of running the experiment due to a learning effect or a chemical that becomes exhausted, then this systematic error could build up and superimpose itself on the factors we study and taint their actual effects! We would have confounded the error with the factors we have under study. One way to assure this independence and to prevent this type of confounding is to randomize the order of the experimental runs. By randomizing the runs, we are able to make any systematic but unknown variation appear as a random variation.
