The mahalanobis function returns . If the data are multivariate normal with dimension m, then we expect to follow a distribution. We can check this graphically with a Q–Q plot as seen in Figure 11.2. We see that there are some outliers and that we can investigate the sensitivity of the PCA to these values by re-analyzing the data after removing these points. If you do this, you will find that it makes a substan- tive difference, especially to the second PC. An alternative to this outlier detection approach is to use robust PCA methods.