Journal of quality technology vol:33 issue:4 pages:436-450
Cost considerations and difficulties in performing completely randomized experiments often dictate the necessity of running response surface experiments in a split-plot format. The resulting compound symmetric error structure not only affects estimation and inference procedures but also has severe consequences for the optimality of the designs used. For this reason, the error structure has to be taken into account explicitly when constructing the design. In this paper, an exchange algorithm for constructing D-optimal split-plot designs is developed and the resulting designs are analyzed. Computational results show that substantial efficiency gains can be realized.