The great discriminant wall of multivaria

A discriminant analysis can have utility by focusing attention on those manufacturing, aging, and environmental factors which influence the failure of a particular component. Appropriate corrective actions can then be initiated prior to the production of additional units and existing lower quality units may be fixed to improve their reliability. A useful measure of quality for both tested and untested units is the posterior probability of failure, which is based upon the discriminant function and the empirical priors. Some phases of discriminant analysis are formulated in terms of normality, but can be practically robust to that assumption. Graphical displays provide useful insight into this aspect of the analysis. The sensitivity of the discriminant line (hyperplane) to changes in the sample point configuration is also discussed.<>