Accounting for historical information in designing experiments: the Bayesian approach.

Two of the most important statistical problems in human and animal experimentation are the selection of an appropriate number of units to include in a given study and the allocation of these units to the various treatments. Properly addressing these issues allows the units to be used as efficiently as possible, which can contribute to addressing the overall issue of reducing the number of subjects in experimentation. To do so, reliable historical information is of particular importance. In the present paper, we describe the Bayesian approach to determining sample size and allocating units, with particular regard to how the use of historical data can be optimised. The paper focuses on two specific problems: the comparison of normal means and of binomial proportions.