Crossover designs with correlated observations

Crossover designs are widely used in different medical investigations where a number of treatments have to be compared. Sequences of treatments are given to subjects, and in practice the observations within each subject are likely to be correlated. This paper is concerned with the construction of crossover designs for such cases. The design problem is nonlinear in the parameters, and design optimality depends on the parameters defining the correlation structure. When the correlation structure is known, local optimum designs are obtained. When the distribution of its parameters is known, optimum Bayesian crossover designs are constructed. The optimum sizes of the groups of subjects receiving the same sequence of treatments are also determined.