Balancing allocation of subjects in biomedical research: a minimization strategy based on ranks

A new method for ensuring the comparability of groups of experimental subjects is proposed. All subjects are ranked on each measured pretreatment variable and differences between groups in the means and the standard deviations of these ranks are minimized by systematically exchanging subjects when those exchanges yield less imbalance between groups. This method was designed for small studies in which pretreatment data from all subjects are known before the start of the experiment. A series of computer simulations suggests that this procedure is effective in balancing groups on several variables although the amount of computer time required becomes excessive as the number of subjects is increased.