Abstract
The properties of screening designs for constrained mixture experiments derived from classical two-level screening designs are investigated. A slightly modified version of the XVERT algorithm is discussed and illustrated for eight- and eleven-component examples. The technique can be implemented by hand and is shown for the eight- and eleven-component examples to yield designs with properties comparable to those of designs constructed from the set of all extreme vertices using optimal design software. Because the good properties of these screening designs appear to be due to the good properties of the classical two-level designs from which they are derived, it is concluded that designs for even larger numbers of components should also have good properties.
Additional information
Notes on contributors
Gregory F. Piepel
Dr. Piepel is a Senior Research Statistician in the Computational Sciences Department. He is a Senior Member of ASQC.