Chapter 3 - Bertin's Graphics and Multidimensional Data Analysis

Abstract

This chapter discusses the role of Bertin's graphics in the multidimensional data analysis. These graphics are straightforward and accurate methods for communicating the results of some multidimensional statistical methods such as the principal components analysis, correspondence analysis, and cluster analysis. Bertin's graphics provide a visual complement to the solutions of correspondence and cluster analyses. Data matrices are represented by a matrix of histograms that lie on the same scale where rows and columns are optimally permuted. This permutation is defined in terms of either progressive variation or sedation or by homogeneous groups distinct from one another or by block modeling. These permutation criteria—defined empirically by Bertin—are the criteria of the multivariate statistical methods; the diagonal sedation corresponds to the maximum correlation permutation of rows and columns in the correspondence analysis and the block criteria corresponds to the homogeneity of groups in the cluster analysis. The chapter discusses several concepts related to Bertin's rules of graphic syntax and presents a simple example of Bertin's graphics. A detailed discussion on the correspondence analysis, Bertin's graphics, and the cluster analysis is also presented in the chapter.