ABSTRACT
This chapter examines various approaches that are possible to the analysis of the 2 × 2 cross-over trial, and looks at some modifications that have been suggested to resolve the difficulty, such as the use of baseline measurements and “wash-out” periods. Probably the most widely used and most abused cross-over design in employ is the 2-treatment, 2-period, 2-sequence design. The treatment, period, and treatment-by-period interaction terms all appear in the within-subjects stratum. The residual sum of squares and mean squares are the same as in Parameterization 1, but the adjusted treatment mean squares differ, as does the interpretation of the effects. The chapter presents a numerical illustration of the various problems associated with the 2 × 2 design using the data set of H. I. Patel. The 2 × 2 cross-over design without baseline measurements leads to an overparameterized model that requires, if meaningful solutions are to be obtained, additional assumptions to be made to remove one of the parameters.
