An algorithm for generating response surface split-plot designs

Abstract

In an industrial experiment, the presence of hard-to-change factors may force the experimenter to use a split-plot type experimental structure. The corresponding Response Surface model then is a linear mixed model. Different design methodologies and estimation techniques are available in the literature, for constructing a Response Surface split-plot design and for estimating the variance components of the model. We discuss the need for developing a new design construction methodology and present an algorithm for generating a D-optimal Response Surface split-plot design such that it has pre-specified numbers of degrees of freedom for estimating the variance components using the randomization based approach. An advantage of this approach is that it gives pure-error estimates of the variance components.