Back To Index Previous Article Next Article Full Text

Statistica Sinica 11(2001), 553-604



Shao-Wei Cheng and C. F. J. Wu

Academia Sinica and University of Michigan

Abstract: Standard practice in response surface methodology performs factor screening and response surface exploration sequentially, using different designs. A novel approach is proposed to achieve the two objectives on the same experiment, based on one design. Running a uni-stage experiment has the advantages of saving experimentation time and run size. The approach is based on a two-stage analysis that employs factor screening, projection and response surface exploration. Projection-efficiency criteria are defined to evaluate the performance of the projected designs. The projection-efficiency properties of the $3^{n-k}$ designs and three nonregular designs are studied, and comparisons with central composite designs are made. Nonregular designs appear to enjoy better projection properties. The strategy is illustrated with the analysis of a PVC insulation experiment.

Key words and phrases: Central composite design, $D$-optimality, fractional factorial design, orthogonal array, projection, response surface methodology.

Back To Index Previous Article Next Article Full Text