Article ID: | iaor19971170 |
Country: | United States |
Volume: | 16 |
Issue: | 1/2 |
Start Page Number: | 5 |
End Page Number: | 48 |
Publication Date: | Jan 1996 |
Journal: | American Journal of Mathematical and Management Sciences |
Authors: | Romeu Jorge L. |
Keywords: | simulation |
A new methodology for assessing distributional assumptions of multivariate data, with graphical applications, is presented. The underlying procedure is based on transforming the multivariate sample into a set of uncorrelated samples and representing the order statistics of each transformed sample by linked vectors in a two dimensional space. The proposed method is described and its properties discussed. The multivariate normality tests are reviewed and a new classification scheme for them is proposed. The new test is then compared with a selection of the ‘best’ competing ones under an exhaustive Monte Carlo study. A selection of ‘best’ tests for several non normal alternatives, with advantages and disadvantages, is given. Graphical aspects of the new procedure are discussed.