Multivariate AB-BA crossover trial

One way to analyze the AB-BA crossover trial with multivariate response is proposed. The multivariate model is given and the assumptions discussed. Two possibilities for the treatment effects hypothesis are considered. The statistical tests include the use of Hotelling's T2 statistic, and a transformation equivalent to that of Jones and Kenward for the univariate case. Data from a nutrition experiment in Mexico illustrate the method. The multiple comparisons are carried out using Bonferroni intervals and the validity of the assumptions is explored. The main conclusions include the finding that some of the assumptions are not a requirement for the multivariate analysis; however, the sample sizes are important.