Robustness of Split-plot Central Composite Designs in the Presence of a Single Missing Observation.

Abstract

In this work, we present a Minimaxloss criterion, based on D-efficiency, and which accounts for the within-whole plot correlation among the observations, for constructing split-plot response surface designs that are robust to missing a single observation of the various design points. We then develop robust split-plot designs, which are relatively insensitive to a single missing observation. It was observed that the criterion is robust to changes in the degree of correlation among the observations but varies with changes in the number of whole-plot and subplot factors for a given k –factor design.