Abstract: Robust parameter design methods have been used successfully in industry for some time. Despite this, there has been some skepticism in the statistical literature about the feasibility of conducting industrial experiments to estimate both location and dispersion effects. It has been claimed that a large experimental run size is needed to estimate dispersion effects and that such experiments are not practical in industry where the emphasis is on studying many factors simultaneously using highly fractionated designs. We show in this paper that this misconception arises from the fact that the commonly used methods of analysis ignore the basic structure in parameter design studies and hence are unnecessarily inefficient. We consider different models and methods of analysis and quantify the gains to be made from exploiting the inherent structure in parameter design studies. The consequences of these conclusions for the planning of such studies are also discussed.
Key words and phrases: Design of experiments, dispersion effects, quality improvement, variation reduction.