ENBIS: European Network for Business and Industrial Statistics

Iterative identification of location and dispersion effects

23 September 2009, 11:15 – 11:35

Abstract

Submitted by

Corinna Auer

Authors

Auer, C., Kunert, J., Wilk, A., Homberg, W. and Hornjak, D.

Affiliation

Technische Universität Dortmund

Abstract

In many technical applications the assumption of constant variance as made in linear models is not correct. Frequently, functional dependencies between mean and variance are observed. And/or there exist dispersion effects, which directly have an effect on the variance. Having an unreplicated design, methods to identify active dispersion effects are always based on the location model identified before. Many authors point out difficulties of identifying the correct dispersion model in the presence of a false location model.
To reduce these problems we propose a graphical iterative procedure for identifying active location and dispersion effects. The proposed method has the advantage of also identifying an appropriate transformation reaching independency between mean and variance at the same time. This holds for replicated as well as unreplicated designs. The method will be demonstrated with some simulated data sets and a data set from a project in cooperation with mechanical engineers.

View paper

Return to programme