THE SELECTION OF INDEPENDENT VARIABLES AND PRIOR PROBABILITIES AS A FACTOR INFLUENCING THE ACCURACY OF CLASSIFYING INDIVIDUALS TO EXISTING GROUPS.

This paper first examines some of the suggestions made by Lohnes and Gribbons (1970) for improving the hit rate of multiple discriminant classification analysis. Alternative to their ideas are advanced and the effect upon hit rate is determined. It is found that an improvement, over their best hit rate, of greater than 26% can be achieved by adopting the ideas proposed by this paper. These involve a more careful selection of the variables and the prior probabilities used in the discriminant function.