MARGINAL MODELING OF CATEGORICAL-DATA FROM CROSSOVER EXPERIMENTS

Marginal models provide a useful framework for the analysis of crossover experiments when the response variable is categorical. In this paper we use the three-treatment, three-period crossover experiment with a binary outcome variable to demonstrate how marginal models can be used to perform a likelihood-based analysis of multiple-period crossover experiments. Other designs are discussed in less detail. Maximum likelihood estimation is performed using a constraint equation specification of the marginal model. Data from a crossover trial comparing treatments for primary dysmenorrhoea are used to demonstrate the utility of marginal models in analysing crossover data.