ENBIS: European Network for Business and Industrial Statistics

Double Generalized Linear Model Estimates in the Case of Unreplicated Designs with Dispersion Effects

23 September 2008, 11:15 – 11:20

Abstract

Submitted by

Corinna Auer

Authors

Corinna Auer and Dr. Martina Erdbrügge

Affiliation

Technical University of Dortmund, Germany

Abstract

In many technical or industrial applications it is necessary to not only optimize the mean but also the variance of a process. Using replicated designs it is possible to estimate functional dependencies between mean and variance as well as possible dispersion effects within the same model framework. However, in the situation of unreplicated designs usually constant variances or only a functional relationship between mean and variance is assumed. Here a method based on double generalized linear models (dGLMs) is proposed, which can also take possible dispersion effects into account. To fit a dGLM an extension of the likelihood as for example the extended quasi-likelihood (EQL) or the pseudo-likelihood (PL) is needed. The resulting estimates are examined concerning desirable statistical properties like efficiency and normality.

Return to programme