Nonparametric Methods for Multivariate Analysis Using Statistically Equivalent Blocks.

Abstract : Multivariate statistical procedures developed under normality assumptions are well advanced. Some of these procedures claim robustness properties, especially in a large sample situation, that may serve to broaden their range of application. Nonparametric methods for multivariate analysis have been pursued, but their more complete development awaits further research. This report considers nonparametric multivariate hypothesis testing in both one- and two-sample situations. Comparable univariate procedures do not extend readily to higher dimensions. The methods considered are based on the properties of statistically equivalent blocks. A new approach using a proximity-based cutting function for block construction is described. Statistically equivalent blocks, while holding the promise of important practical application, has received limited research attention.