implications of random allocation for the t test