“…Linear discriminant analysis procedures that are robust (i.e., insensitive) to departures from the assumption of multivariate normality have been proposed (Todorov, Neykov, & Neytchev, 1994) by replacing the conventional least-squares estimators of means and covariances with robust estimators, such as M-estimators (Croux & Dehon, 2001), minimum covariance determinant (MCD) estimators (Hubert & Van Driessen, 2004;Rousseeuw, 1984), minimum volume ellipsoid (MVE) estimators (Rousseeuw, 1984), and trimmed estimators (Ahmed & Lachenbruch, 1977;Gnanadesikan & Kettenring, 1972;Srivastava & Mudholkar, 2001). However, their emphasis has been primarily on prediction and not on describing the variables that contribute to group separation in non-normal data.…”