Show that if a one-way analysis of variance is performed on the ranks of the observations instead of the observations themselves, it becomes equivalent to a test based on the H statistic.