Grade Correspondence Analysis and outlier detection

Abstract

This chapter and the next present the main procedures used in grade methodology. They are based on grade correspondence analysis (GCA) which is introduced in Section 9.2. The basic procedure of GCA solves the problem of permuting the rows and columns of a probability table in order to maximize the value of the grade correlation p*. This is done by alternately permuting the rows and columns according to the respective grade regression function until both these functions are nondecreasing and no further improvement of p* is possible. Such a table provides a local maximum of p*. GCA finds several local maxima starting from the original table after the random permutations of its rows and columns, and then selects the highest value of p* as the global maximum (or at least a very good approximation of the global maximum).