Alicja Ciok

A comparative study of exploratory methods applied to car switching data

816

Abstract

The car switching data concern car purchases 1979 in USA. They form 16x16 data table (x(i,j)) which shows how many owners of a car belonging to i-th segment in a given car classification bought a car belonging to the j-th segment (i,j = 1,...,16). The paper is devoted to compare various methods looking for latent trends or regularities. First (section 3) it is shown what can be done when comparing marginal distributions only (marginal frequency analysis-MFA). Section 4 presents the decomposition into directed components (DEDICOM) method which was previously applied to the car switching data by several authors. In section 5 the data are treated by means of the grade correspondence-cluster analysis, combined then in section 6 with the MFA, and the combination is next compared with the DEDICOM.