A general method for constructing a test of multivariate normality

Abstract

We present a general method of constructing a test of multivariate normality using any given test of univariate normality of complete or randomly incomplete data. A simulation study considers multivariate tests constructed using the univariate versions of the Shapiro-Wilk, Kolmogorov-Smirnov, Cramer-Von-Mises, and Anderson-Darling tests.