Nonparametric methods in multivariate factorial designs

Abstract

A nonparametric approach to the analysis of multivariate data is presented that is based on seperate rankings for different variables and extends the results of Akritas ct al. (1997. J. Amer. Statist. Assoc. 89, 336-343.) to multivariate designs. Factorial designs are considered including degenerate distributions as well as discontinuous distributions. The asymptotic normality of a linear combination of dependent linear rank score statistics is shown under rather weak assumptions. Wald-type and ANOVA-type statistics are derived and their properties are compared in simulation studies. The theory is applied to a completely randomised two-way layout. (C) 2000 Elsevier Science B.V. All rights reserved. MSG: 62G05, 62G10, 62G20, 62H10, 62H12, 62H15.