The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Product liability: The publishers cannot guarantee the accuracy of any information about the application of operative techniques and medications contained in this book. In every individual case the user must check such information by consulting the relevant literature. SPIN 10517148 21/3135 -5 4 3 2 1 0 -Printed on acid-free paper 1 Conference papers related to Internet problems were published in a separate volume in German: H.Chr. Hobohm and H.J. Watjen (Hrsg.): Wissen in elektronischen Netzwerken. Beitrage zur Strukturierung und zum Retrieval von Information im Internet. Bibliotheksund Informationssystem (81S) der Universitat Oldenburg, Oldenburg, 1995.VII results obtained from a correspondence analysis of their data matrix. The paper of S. Ohl is a nice illustration of various descriptive, graphical and multivariate data analysis methods: his data were obtained in the development of the Mercedes-Benz S-Class when car buyers were asked for their preferences in extras and options, and he reveals from these data various dependence and clustering structures. U. Streit gives an account of computerbased geographic information systems and discusses various aspects of the geometrical, topological and thematic modelling of spatial objects. He calls for a closer cooperation of statisticians, geo-scientists and information scientists in order to improve on these systems. M. Theus presents an analysis of spatio-temporal data using interactive statistical graphics. He describes the software package REGARD which uses exploratory and graphical techniques. As an alternative, e.g., to multidimensional scaling, U. Wille presents methods by which an ordinal objectsxdata matrix may be represented by linearly separated regions in the n-dimensional real vector space. She thereby generalizes Scott's axiomatization of represent ability.
Statistical models and methodsG. Arminger and D. Enache survey artificial neural networks considered as statistical nonlinear regression models with parameters (weights) estimated (adapted) by quasi-maximum likelihood methods and nonlinear leastsquares. Various distance or evaluation measures and several numerical algorithms are presented. G. Tutz proposes non-parametric smoothing methods for categorical response data, in particular discrete kernel regression and local likelihood approaches. Those approaches are investigated in detail for multi-categorical regression and the estimation of discrete hazard functions.C. Weihs and W. Seewald present the statistical expert system STAVEX which allows the interactive computer-based design of experiments in industry. Various optimization criteria as well as special features such as mixture factors (with 'Cox axes') and composite responses, are included. R.L. Wolpert proposes the 'conditional frequ...