A method to identify dispersion effects from unreplicated multilevel experiments

The use of experimental design to identify dispersion effects is an effective tool for improving quality. In this article the problem of identifying such effects from unreplicated experiments is studied. We focus on a method developed for two-level 2(k-p) experiments and discuss how the method can be generalized to a wider range of experimental designs. In particular, we show how dispersion effects can be identified from non-geometric Plackett-Burman designs and from experiments with more than two factor levels. In contrast to related methods our approach provides a test statistic with a well-known reference distribution. Three actual experiments are used to illustrate the method and to make comparisons to related methods. (C) 1997 by John Wiley & Sons, Ltd.