Figure 1 Plots of paired ordered chi-square percentile values and observed Mahalanobis distances for non-normally and normally dis- tributed data. The statistical power associated with using NORMUL remains to be empirically evaluated. Preliminary testing with 1,000 sets of multivariate non-normal data gener- ated with the use of the DATASIM code in Table 1 indi- cates a power of .92 for a Type I error rate of .05. Never- theless, further research on the power for varying types of non-normal distributions and on the comparative merits of this approach to other test statistics is clearly desirable.
Related Figures (1)
