When comparing the medians of two independent groups where the data violates the assumption of normality, which test is most appropriate?