This article explores different non-parametric (rank-based) tests for analyzing multivariate observations in balanced and unbalanced one-way lay outs. The performance of the tests is compared in terms of significance level and power for different simulated situations, with continuous and discrete observations. An approximation based on ANOVA-Type statistic performed best for positively correlated responses. For negatively correlated responses, an approximation based the Lawley-Hotelling type test performed best. The use of the tests based on the approximations is demonstrated with experimental data from a plant pathology experiment. (44 refs.)
展开▼