J. Opl Res. Soc. Vol. 45, No.2, pp. 237-240 Printed in Great Britain. All rights reserved

0160-5682/94 $9.00+0.00 Copyright © 1994 Operational Research Society Ltd

Book Selection Edited by JOHN M. WILSON BRIAN S.EVERITT and GRAHAM DuNN: Applied Multivariate Data Analysis RICHARD LAMMING: Beyond Partnership: Strategies for Innovation and Lean Supply OsMAN BALCI, RAMESH SHARDA and STAVROS A. ZENIOS (Editors): Computer Science and Operations Research: New Developments in their Interfaces HELGE RITTER, THOMAS MARTINETZ and KLAUs ScHULTEN: Neural Computation and Self-Organising Maps: An Introduction

237 238 239 240

Applied Multivariate Data Analysis BRIAN S.EvERITT and GRAHAM DuNN Edward Arnold, London, 1991. xii + 304 pp. £29.95 ISBN 0 340 54529 1

This is a book of considerable breadth, starting with simple ideas for representing multivariate data graphically and ending with sophisticated, computer intensive methods for modelling the covariance structure of a set of multivariate data. There are numerous examples in the text, mostly arising from the consultancy work of the authors and, therefore, of a psychiatric or medical nature. There are also practical exercises at the end of each chapter, some with solutions. The book is very well referenced-the list of references at the end of the book numbering almost 300. The book is divided into four sections: Approaches to analysing data, Exploring multivariate data, Regression models, and Latent variable models. Part I (Approaches to analysing data) provides an introduction to the authors' approach; in particular, it highlights their interest in building useful models rather than hypothesis testing and significance levels. Part I also provides an overview of the mathematical background necessary for understanding the remainder of the text. The book assumes a reasonably good mathematical background; matrices, maximum likelihood estimation and optimization (including the method of steepest descent) are all covered in 11 pages. Part II (Exploring multivariate data) covers principal components analysis, multidimensional scaling and cluster analysis as well as introducing a number of ways of displaying multivariate data such as weather-vane plots, Andrews plots and Chernoff faces. The chapter on multidimensional scaling goes beyond the standard two-dimensional Euclidean case to consider the three-dimensional case and to comment on how maximum likelihood techniques can be incorporated into multidimensional scaling. Similarly, the chapter on cluster analysis goes beyond the standard agglomerative hierarchical case to consider the more statistical approach provided by the use of mixture models. In the introduction to Part III (Regression models), the authors state that, in addition to most readers being familiar with multiple linear regression, 'readers should also be familiar with models for the analysis of variance (ANOVA) and the analysis of covariance (ANCOVA)'. As a result, the first chapter in Part III (Chapter 7) introduces the ideas of link functions, distributions (other than the Normal distribution) for the observations and deviance rather than the basic ideas of regression. Chapter 8 shows the essential equivalence between regression and analysis of variance before discussing the use of residuals in model checking. The breadth of the book is illustrated by the inclusion in this chapter of comments on PRESS (predicted residual sum of squares), iteratively reweighted least squares, Cook's D and leverage. Most of the remainder of Part III is concerned with extending the applicability of regression type models. Specifically, there are chapters on linear models for categorical data, 237

Operational Research Society is collaborating with JSTOR to digitize, preserve, and extend access to Journal of the Operational Research Society. ® www.jstor.org

Journal of the Operational Research Society Vol. 45, No. 2

models for rates and survival times and analysis of repeated measures. Part III finishes with a short chapter on discriminant analysis. Part IV (Latent variable models) includes a chapter on factor analysis, building on the earlier chapter on principal components analysis, before looking at more sophisticated covariance structure models. Although the authors quote Cliff1 as saying that these models are 'the most important and influential statistical revolution to have occurred in the social sciences', they also caution against the tendency to accept models as reasonable 'simply because it has been possible to fit them to the data' (p. 274). In any case, the chapter on covariance structure models is rather dense and serves more as an overview and an introduction to the literature than as an introduction that would enable the reader to use such models. Although the authors state that one of the special features of the book is 'interpretation of the output from statistical software', there is very little output from statistical software in the book. The package used in Part III (Regression models) is GLIM and the only real references to other packages such as SPSS and SAS are in a two-page appendix on statistical packages. In conclusion, this is an intermediate level text which requires some solid mathematical and statistical skills of the reader and which covers a very broad range of multivariate techniques. The book includes a large number of examples, although there are times when it seems that too many different examples have been used. Some of the examples seem to be passed over very quickly without being fully analysed and developed. However, this is only a small criticism of a very good book. DAVE CoATES

Reference 1. N. CuFF (1983) Some cautions concerning the application of causal modelling methods. Multivariate Behavioural Res. 18, 115-126.

Beyond Partnership: Strategies for Innovation and Lean Supply RicHARD LAMMING

Prentice Hall International, Hemet Hempstead, 1993. xvii + 299 pp. £19.95 ISBN 0 13 143785 2

The International Vehicle Program, a major research project investigating the automotive industry, was based at MIT between 1986 and 1990. At its conclusion, The Machine that Changed the World 1 was published, and its subsequent sales of over a million copies introduced 'lean' production concepts to a wide readership. Richard Lamming was involved in that research, and is here applying those concepts to automobile component supplier businesses. A detailed history of the relationship between manufacturers and their suppliers is given; the heart of the book is a distillation of this into a descriptive model that summarizes how these relationships have developed from traditional (prior to 1975), through stress (1972-85), resolved (1982-) and partnership (1990-) stages. The fifth stage, beyond partnership as the title implies, which is yet to be fully realized although there are already examples of it, is that of lean supply. This five-stage model is expressed in terms of nine factors: the nature of competition in the component supply market, the basis upon which sourcing decisions are made, the role played by information transfer, the attitude to capacity planning, delivery practices, how price changes are dealt with, the attitude towards product quality, the role of R&D and the level of pressure in the relationship. In terms of these nine factors, Lamming sees the lean supply future as one in which 'competition in lean production will be global and that part of the lean supply relationship must be a readiness on the part of the supplier to provide a local service to the assembler wherever it is required in the world', and there must be a highly developed working relationship between assembler and supplier in making sourcing decisions; a transparency in information exchange, including cost information; capacity planning jointly undertaken; JIT delivery; cost reduction arising from close co-operation; quality measured in 238