Forcing a sequential experiment to be balanced

Abstract

Subjects arrive sequentially at an experimental site and must be assigned immediately to treatment or control groups. In order to avoid biasing the results of the experiment it is customary to make the assignments by independent flips of a fair coin, but in small-sized experiments this may result in a severe imbalance between the numbers of treatments and controls. This paper discusses a new method of assigning the subjects which tends to balance the experiment, but at the same time is not over vulnerable to various common forms of experimental bias.