The analysis of small-sample multivariate data

Clinical pharmacology studies typically consist of few subjects per treatment group, but with many, possibly highly correlated, measurements taken per subject. Permutational methods for testing equality of the multivariate treatment means when the number of variables exceeds the number of independent subjects have been developed, but are highly computationally intensive. In this paper, a parametric test was derived, using Edgeworth expansions, for the case of two groups, and compared with competing test statistics proposed by Mercante and Johnson (7), Dempster (5), Chung and Fraser (3), Mantel and Valand (6). The proposed test compares favorably with the others in terms of type I error rate, as well as power, and has the advantages of not requiring computationally intensive resources and being easily extendable to more than two groups.