Optimal Design of Experiments for Multiple Objectives

• Activity Number: 165
• Type: Topic Contributed
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Quality and Productivity
• Author(s): Timothy Robinson*+ and Bradley Jones*+ and Roselinde Kessels*+ and Chris Nachtsheim*+
• Companies: University of Wyoming and SAS Institute Inc. and University of Antwerp and University of Minnesota
• Keywords: Design of Experiments ; Optimal Designs ; Response Surface ; Choice Experiments ; Robust Parameter Design

Abstract

Box and Draper (1959) suggest a suite of criteria to evaluate when creating a designed experiment. This list of considerations suggests balancing estimation of model parameters and good prediction of new observations for an assumed model with other criteria including robustness to model misspecification and difficulties in collecting the data. For the practitioner it is often difficult to determine which subset of the list to highlight. Then, once the practitioner succeeds in prioritizing the list for the problem at hand, it is a formidable task to implement a design construction procedure that optimally addresses the multiple competing objectives.

In this discussion, we present a number of the potential criteria to consider for different scenarios, as well as several different approaches to finding optimal designed experiments for multiple objectives. The panel will consider screening experiments, response surface designs, choice experiments and robust parameter designs through a series of examples, which outline the process for deciding which criteria to consider as well as how to implement computer algorithms for obtaining the optimal designs. We will compare the solutions w