ABSTRACT
Until comparatively recently, the methodology available for binary and categorical crossover data was rather limited, and amounted largely to a somewhat ad hoc collection of tests based on simple contingency tables. An exception to this was the Mainland-Gart test for direct treatment effect from a two-period two-treatment trial with binary data described in Section 2.13. Although introduced there in the context of a contingency table, Gart (1969) gave a derivation for this in terms of a logistic regression model and Altham (1971) considered the same model from a Bayesian viewpoint. More recently interest in the analysis of repeated and correlated categorical data has grown enormously and we now have available a much wider range of methodology; see, for example, Chapters 7 to 11 of Diggle et al. (2002). Much of this recent work is relevant to the analysis of data from cross-over trials and for this reason most of the techniques described in the earlier literature, and the first edition of this book, Jones and Kenward (1989), are now either redundant, or can be seen to be special cases of more general model-based approaches.
