ABSTRACT
Blocking becomes far more problematic when it involves a breakdown of a given mixture design into subgroups. If it is done without careful thought, model parameters can become highly correlated with the block effect. The Draper–Lewis blocking schemes use Latin squares as their basis. The blocked mixture experiment cited by Draper et al. and Lewis et al. in their 1993 and 1994 publications, respectively, was conducted by an English miller on four flours, each derived from a different variety of wheat. The mean square for the blocks exceeds the residual mean square by nearly fourfold, which is appreciable, that is, only by filtering out this variation over time was it possible for this design to reveal the formulation behavior exhibited in the trace plot. Orthogonally blocked mixture experiments are highly inefficient compared to optimal designs—Goos and Donev. The mixture side of the experiment studied pure blends, binary blends, and the centroid of three plasticizers randomized within groups.
