We consider the problem of randomizing a known number of subjects into two or more treatment groups when recruitment occurs over an extended time period, and thus the potential for confounding factors related to time is a concern. Our proposed design generates more balanced allocations than complete randomization, but can have higher entropy than randomized-block designs. These conclusions follow a careful analysis of the probability distribution induced on the allocations. Furthermore, analysis and simulation indicate this new design can exhibit power advantages over randomized blocking and complete randomization. Randomization tests have been applied in a series of cancer chemoprevention trials. The design may be useful in other experiments where experimental units are confined to some linear order, such as arrangement in space.
Department Technical Reports