Identifying the Best 16-Run Regular or Non-Regular Screening Design for 6 to 8 Factors Using Multiple Objectives

• Activity Number: 298
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Christine Anderson-Cook*+ and Lu Lu and Mark E. Johnson
• Companies: Los Alamos National Laboratory and University of South Florida and University of Central Florida
• Keywords: Pareto front optimization ; non-dominated designs ; desirability functions ; design generators ; supersaturated designs ; alias patterns

Abstract

When exploring first order models including two-factor interactions for six to eight factors using a sixteen run design, there are many possible design choices. Building on the Johnson and Jones (QREI, 2010) catalog of the non-isomorphic regular and non-regular design alternatives, we summarize which of these designs are most promising based on two common general criteria. While there are 27, 55 and 80 alternatives for the 6, 7 and 8 factor cases, respectively, the Pareto fronts based on the criteria E(s^2) and tr(AA^') suggest that only a handful should be considered further. The best design for a particular experiment depends on the relative emphasis given each of the two criteria. Numerical and graphical comparisons between the alternatives are provided to highlight the merits of the leading candidates. Possible generalizations to other criteria for comparing the designs are also described.