A further note on the interpretation of ridit analysis.

Ridit analysis has been used by a variety of investigators (e.g., 1-5) and has appeared in a t least one textbook (6). Bross (7) introduced the technique of ridit analysis and Kantor et a1 (8) showed the relationship between the average ridit value and a quantile rank. The purpose of this article is to elaborate on the remark made by Bross (7) that "[ridit analysis is] closely related to distribution-free methods based on ranks, especially the Wilcoxon text." An expression for the covariance between two average ridits with a common relative identified distribution and a chi-square statistic for testing the homogeneity of a series of average ridits is also included. The relationship between the ridit and Wilcoxon procedures makes it possible to use much of the methodology developed for the Wilcoxon test in the context of a ridit analysis. Therefore, exploring this relationship strengthens the meaning and facilitates the interpretation of the ridit technique for the analysis of epidemiologic data. Ridit analysis, in some situations, is a more sensitive alternative to the chisquare statistic for comparing two independent samples when the variable under study can be classified into ordered categories. One primary reason for employing ridit analysis is that the procedure tends to suppress differences in distributional shapes (see reference 7 for more discussion). The Wilcoxon rank sum test is also often used to compare independent samples. Ridit analysis uses a probability relative to a reference distribution as a means of identifying differences between groups, whereas the Wilcoxon procedure uses the sum of a set of ranks as a measure of the