A Note on Multiple Comparisons Using Rank Sums

In a one-way layout design, the rank-sum procedures proposed by Steel [6] and Dunn [1] for the problem of comparing all pairs of treatments are compared and found to have equivalent asymptotic properties. The asymptotic relative (Pitman) efficiency of these nonparametric procedures relative to the classical solutions is shown to be the same as the efficiency of the corresponding two-sample rank-sum tests relative to the two-sample Student's t-test.