THE NORMAL SCORES TEST FOR THE c ‐SAMPLE PROBLEM

Two forms of a c ‐sample normal scores test are presented as alternatives to the Kruskal–Wallis test (KW). The normal scores test statistics and multiple comparison procedures are illustrated for large samples requiring a correction for tied observations. Theoretical and empirical evidence is cited in support of using the normal scores test in preference to a rank test such as the KW. The normal scores test is asymptotically more efficient than the KW for samples from common non‐normal populations (e.g. uniform and exponential) and is less sensitive to non‐normality and unequal variances than is either the F ‐test or the KW test.