Classification based on dichotomous and continuous variables : Krzanowski's procedure

Abstract

This dissertation reports an evaluation of Krzanowski's location model approach to discriminant analysis. Three of Krzanowski's assumptions (all binary means equal within-population, all binary correlations equal within-population, all discriminant functions parallel within-category) are relaxed and the impact noted on theoretical and observed error rates for the two binary, two normal variable case. Krzanowski's procedure outperforms the LDF on theoretical error rate comparisons, and also on observed error rates but by a smaller margin.