Model-Oriented Data Analysis; Proceedings of an IIASA Workshop, Eisenach, GDR, March 9-13, 1987)

Fedorov, V. & Laeuter, H. (1987). Model-Oriented Data Analysis; Proceedings of an IIASA Workshop, Eisenach, GDR, March 9-13, 1987). Heidelberg: Springer-Verlag. ISBN 3-540-18596-8

Preview

Text XB-87-407.pdf Download (4MB) | Preview

Abstract

The main topics of this workshop were (1) optimal experimental design, (2) regression analysis, and (3) model testing and applications.

Under the topic "Optimal experimental design" new optimality criteria based on asymptotic properties of relevant statistics were discussed. The use of additional restrictions on the designs were also discussed, inadequate and nonlinear models were considered and Bayesian approaches to the design problem in the nonlinear case were a focal point of the special session. It was emphasized that experimental design is a field of much current interest.

During the sessions devoted to "Regression analysis" it became clear that there is an essential progress in statistics for nonlinear models. Here, besides the asymptotic behavior of several estimators the non-asymptotic properties of some interesting statistics were discussed. The distribution of the maximum-likelihood (ML) estimator in normal models and alternative estimators to the least-squares or ML estimators were discussed intensively.

Several approaches to "resampling" were considered in connection with linear, nonlinear and semiparametric models. Some new results were reported concerning simulated likelihoods which provide a powerful tool for statistics in several types of models. The advantages and problems of bootstrapping, jackknifing and related methods were considered in a number of papers.

Under the topic of "Model testing and applications" the papers covered a broad spectrum of problems. Methods for the detection of outliers and the consequences of transformations of data were discussed. Furthermore, robust regression methods, empirical Bayesian approaches and the stability of estimators were considered, together with numerical problems in data analysis and the use of computer packages.

Actions (login required)

View Item