Nonparametric testing under crossover design for ordered categorical response

This paper uses ridit introduced by Bross (1958) to develop a nonparametric procedure for testing the difference between two treatment effects with ordered categorical outcomes under two period crossover design in which subjects receive same or different treatment during the period of study. The performance of this procedure is compared numerically with that of a procedure obtained from Armitage (1955) in terms of type I error rate and power. The numerical study shows the ridit approach to tend to relatively more efficient in terms of power as the number of categories increases. The procedures are illustrated using the data from Matthews (1989) on crossover trial with two hypertensive agents.