Using a regression approach to discriminant analysis is often incorrect because it forces the use of a binary dependent variable which violates virtually any distributional assumption for a linear model. However, assuming a Laplace distribution in an LP framework leads to a theoretical foundation for MSD discriminant analysis.