Selecting a Nonregular Fractional Factorial Design for Screening
Presenter: David J. Edwards, Virginia Commonwealth University

Co-authors: Robert W. Mee, University of Tennessee, Eric D. Schoen, University of Antwerp, Belgium and TNO, Zeist, Netherlands

Abstract:
This talk presents a comparison of criteria used to characterize nonregular two-level designs for screening purposes. We provide insights that will aid practitioners in understanding the relevance of various criteria to evaluate nonregular designs and in choosing a design with a given run size and number of factors. To articulate the relationships between criteria, we focus on seven-factor designs with 16-32 runs and 11-factor designs with 20-48 runs. Screening based on selected designs for each of the run sizes considered is studied with computer simulation using a forward selection procedure and the Dantzig Selector. Results are shown using both model dependent and model independent error estimates. The simulation offers a direct appraisal of the benefit of a few replicates or an increased run size.

2015 QPRC