On MSE-optimal crossover designs (Q1991683)

A crossover design is a repeated measurements design such that each experiment unit (e.g. the patient) receives different treatments during the different time periods, i.e. the patient crosses over from one treatment to another during the course of the trial. Often a treatment applied in a given period has in addition to its direct effect a so-called carryover effect, i.e. it may effect the measurement in the next period. Most papers on optimal crossover designs consider an estimate which is corrected for carryover effects. However, this corrected estimate has a large variance. In this paper the authors consider another way. They use the uncorrected estimate which has a smaller variance but which is biased if carryover effects cannot be neglected. They find crossover designs which are optimal in the sense that they minimize the mean squared error (MSE), i.e. the sum of variance and bias. Efficiency considerations show that often designs which are optimal for the corrected estimate are also efficient for the uncorrected estimate.