On the Use of Rank Tests and Estimates in the Linear Model

The current status, including small-sample behavior and ease of computation, of rank-based estimates and tests in the general linear model is reviewed. For the important special case of Wilcoxon scores, details of application of various procedures are discussed. The three different testing methods considered may each be motivated by connecting it to one of three forms of the usual least-squares F statistic. Possible algorithms for computation of rank-based estimates and tests are presented. Each procedure is applied to an example using data. Finally, the technical assumptions made to obtain large-sample properties of these procedures, including the general-scores case, are outlined and discussed. (Author)