TABLE OF CONTENTS "Two is company, three is trumpery."
EDNA LYALL (1897, WAYFARING MEN)
In the previous chapter we introduced the central composite design (CCD) the first choice for practitioners of response surface methods (RSM) because of its flexibility. As illustrated by a case study, an experimenter can hedge on RSM by adding center points to a simpler, more economical two-level factorial. If curvature is not significant, it's time to move on: Why bother doing RSM if the surface is planar? On the other hand, if you detect a significant increase or decrease in response at the center point and it merits further attention, you can complete the CCD by adding the second block of axial points, including more center points to provide a link with block one.
In the previous chapters we detailed a study on a reaction process that revealed a significant curvature effect of about 6 grams above the expected level of 82 grams. The chemist evidently felt that this deviation was too important to ignore, so the second block of the central composite design was completed in order to construct a proper map of the nonlinear response. If the curvature had been overlooked despite being statistically significant, it would have made no difference in the end, because the same optimum emerges from the factorial model as that produced by the full CCD: low time, high temperature, and high rate. Thus we get into an issue of statistical significance versus practical importance. Highly controlled automated...
