ABSTRACT
Four general strategies are defined here which can be used for assessing multivariate normality. As in the univariate case, normal probability plots are suggested both as a subjective test for multivariate normality and for assessing the direction(s) of departures from normality. The second strategy consists of objectively assessing marginal normality using univariate tests, using the necessary (but not sufficient) condition that a multivariate normal distribution is normal in each of its marginal distributions. The third strategy is the reduction of data to m' < m dimensions in such a fashion that the distribution of the reduced data has some known or simple structure. Finally, the fourth strategy is the direct assessment of multivariate normality from the multivariate observations.
