3.11. Concluding Remarks

In closing this chapter, we emphasize that the inference in multivariate regression can be sensitive to the assumption of multivariate normality as well as the presence of unusual observations. We have briefly described some of the approaches to these problems here. It is a good practice to apply these checks before performing the formal multivariate analysis described earlier in this chapter. Some of the test statistics in a normal MANOVA are more sensitive to nonnormality than others. In general, if the data are not strictly normal, Pillai's test is the most robust (in terms of preserving the power of the test) of the MANOVA tests and therefore is especially recommended for such situations.