Correspondence Analysis Handbook

Correspondence Analysis Handbook

THE STATISTICAL SOFTWARE NEWSLEITER Book Reviews J.-P. Benz6cri (1992): Correspondence Anal. ysis Handbook. New York: Marcel Dekker, ISBN: 0-8247-84...

90KB Sizes 1 Downloads 156 Views

THE STATISTICAL SOFTWARE NEWSLEITER

Book Reviews J.-P. Benz6cri (1992): Correspondence Anal. ysis Handbook. New York: Marcel Dekker,

ISBN: 0-8247-8437-5, pp. 688, $150.00 In the 1960's a group of mathematicians around Benz6cri developed correspondence analysis as a means to explore large tables of data by interpreting them through Euclidean space. To the nonFrench-speaking, it became known as an application to contingency tables, mostly through the volume "rheory and Applications of Correspondence Analysis" by M.J. Greenacre. It took nearly ten more years until this concise handbook written by its originator was made available to non-French speakers. In the first part, the author gives a broad treatise on Euclidean space. Appealing to the reader's spatial intuition, he proceeds from two- and threedimensional to n-dimensionai space. Keywords of vector geometry which are central to correspondence analysis, such as inertia, barycentre, and principal axes, are thoroughly introduced. Notation is mostly given in coordinates, sums and scalar products rather than in vectors, matrices, and vector products. Because a probability-based approach to data is not used, hints to possible relations of correspondence analysis and inferential statistics are only rarely given. In the second part of the book the methods derived earlier are applied to a simple data set consisting of answers to a two-questions, twelve respondent poll. Each step is meticulously described. The third part of the book describes an application to a complex 65x18-table, pointing out the scope of correspondence analysis. It ends with a code listing of a FORTRAN program for the analysis. The rest of the book deals with other examples from different fields of research, discussing the interpretation of principal axes, distances, and patterns. In Correspondence Analysis, two major methodological problems arise. One lies in the combined plot of row and column variables and their interpretation. This book uses the combined plot of row and column coordinates, which is commonplace in correspondence analysis but hard to interpret, but it also discusses relationships to alternative, more

375

directly interpretable methods. The other problem is the simultaneous analysis of several categorical variables and their first-level interactions. The author states that there is no satisfactory way to treat multiple interactions simultaneously. In contrast, Greenacre chose the Burr matrix for multivariate Correspondence Analysis. This book is suitable for a practical researcher with a fair grasp of vector geometry. Whereas Greenacre's book is written for the statistician who wants to know how correspondence analysis fits into the analytical framework already known, "Correspondence Analysis Handbook" can be read and understood by somebody who has no idea of what a random variable is. It is concise and selfsufficient, yet demonstrates this fact a little too clearly by not including a bibliography. It also gives a good idea of what Benz~cri means by "statistics is not probability", though Greenacre's book is more easily accessible to the probabalistically spoilt mind. J. Fassbinder Dortmund, Germany

,~ff

~For Book Reviews section contact: Walter KrLrner, Universit~t Dortmund, Germany. Fax: +49.231.755-5284, E-mail: walterk@ amadeus.statistik.uni-dortmund.de

IASC-News Registration of the COMPSTAT label as trade mark On occasion of the forthcoming COMPSTAT meeting in Barcelona, notice is given on the document, that certifies IASC as owner of the trade mark COMPSTAT with the German patent office "Deutsches Patentamt". On behalf of the IASC, we are happy that the efforts to protect the label COMPSTAT based on an initiative of former president Norbert Victor was so successful. Figure 1 shows the first page of the certificate (Urkunde); Figure 2 part of the second page explaining the objectives of COMPSTAT: Running conferences and seminars, publication of books, journals and newspapers.