Assessing Magnitude of Effect from Rank-Order Correlation Coefficients

Abstract

While Spearman's rho (ri) and Kendall's tau (t) are equally powerful rank-order correlation coefficients under conditions of normality, they have quite different metrics. When applied to the same data set, t typically is smaller in absolute value, often no more than two-thirds the size of r,. Although these facts are duly stated in a number of works on nonparametric statistics, they are omitted in many other, less detailed presentations. Given increasing concern with the degree to which a statistical effect existsrather than just whether an effect is present at all—this difference in correlational metric appears to need emphasis.