A graphical method for assessing multivariate normality (Q880901)

A new graphical tool is proposed for testing the hypothesis of normality/nonnormality of multivariate data. It is based on the fact that for i.i.d. normal samples the sample mean is independent from the sample variance. So a set of bootstrap subsamples from the original sample is taken and for some linear combination of observations entries sample means and variances are computed on each subsample. The results are depicted on a scatterplot. A dependence visually observed at the display corresponds to nonnormality. Applications to Iris data and simulated samples are presented. The authors claim that this method is robust w.r.t. heteroscedasticity of data (in contrast to standard Q-Q-plot procedure).