A Multivariate Multisample Rank Test for Stochastic Simulation Validation
Final rept. 1 Oct 9193-31 May 1994
ARMY RESEARCH LAB ABERDEEN PROVING GROUND MD
Pagination or Media Count:
Nonparametric multivariate statistical methods address a broad class of problems in data anlaysis in which the assumption of normality is not feasible, and where the data occur naturally as n-tuples vectors rather than scalar values. This is the data structure that is most common in the engineering sciences and, coincidentally, the least tractable. A computer-intensive approach to the analysis of these data, usually referred to as randomization or permutation procedures, will be the specific focus of this work. Tests based on permutations of observations are nonparametric tests. This study considers a multivariate extension of the well-known Kruskal-Wallis rank sum test as a method for hypothesis testing, a technique commonly employed for simulation validation. The test statistic investigated is a nonparametric analogue of the classical Hotelling T2 used for the normal theory model. This undertaking is part of a broader based Army research program, the goal of which is to improve the ability of communications networks to deliver critical information on the battlefield when and where it is needed, despite a rapidly changing and often hostile environment. It will also support the ongoing effort to formalize the validation process for network simulation that, in turn, provides the groundwork for exploring alternatives and testing hypotheses throughout the research program. This formalization of the validation process can be readily transmitted to other organizations that rely on network simulations for their analyses.
- Statistics and Probability
- Computer Programming and Software
- Command, Control and Communications Systems