Prediction capability of industrial split-plot response surface designs when one observation is missed

Response surface designs under restricted randomization or Split-plot response surface designs are often used in agriculture experiments and in industrial experiments due to existence of one or more factors that can't change their levels easily some factors need to estimate more precisely. The motive of this paper is to prevail prediction capability of a particular class of split-plot response surface designs, known as Central Composite Designs by Vining, Kowalski and Montgomery (VKM CCDs) when one observation of any category is missed. Both numerical and graphical methods, based on scaled prediction variance (SPV) are applied. Robustness of above class of designs against one missing observation is investigated relative to G-efficiency and Minimax loss designs are proposed. The prediction capability is computed by graphical methods such as 3D Variance Dispersion Graph (3D-VDG), Fraction of Design Space (FDS) plots and contour plots for extraordinary efficiency standards are used to look at the impact of lacking observations.