Abstract
Dispersion experiments examine the effects of control factors on dispersion introduced by error factors. Error factors can be controlled in the laboratory but cannot be controlled when a product is used. For dispersion experiments, Taguchi used direct product designs in which control-factor levels are set by one orthogonal array, error-factor levels are set by another orthogonal array, and, for each combination of control-factor and error-factor levels, an observation is taken. In contrast, compound designs, which include product designs as special cases, share several of the attractive properties of Taguchi's designs, but they are often smaller for a given strength, or stronger for a given size, where strength refers to strength of the orthogonal array. This article discusses several useful compound designs for dispersion experiments and several general methods for constructing such designs. In particular, compound designs can be built from existing tabulations of confounded factorial designs.