Layered and Hybrid Designs for Model Robustness

• Activity Number: 111
• Type: Topic Contributed
• Date/Time: Monday, July 30, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Quality and Productivity Section
• Author(s): Greg Piepel*+
• Companies: Pacific Northwest National Laboratory
• Keywords: Model robust design ; Layered design ; Space-filling design ; Optimal Design

Abstract

Optimal experimental design methods (e.g., D-optimality) that require assuming a specific response model form are widely used. Design methods that are robust to the model form have been the subject of substantial research literature since 1959. Despite this, model-robust design methods are not widely discussed in textbooks and are not implemented in commercial software. Hence, these methods are seldom used by practitioners. This presentation discusses two model-robust design approaches that I have used for many years, which can be implemented using available software. Both approaches can be used for problems involving mixture variables, non-mixture variables, or both. The layered design approach involves using standard optimal design software to generate design points on two or more "layers" of the experimental region. The hybrid design approach consists of constructing a small D-optimal design and augmenting it with some space-filling points. These design approaches provide for model robustness through including points on the boundary as well as the interior of the experimental region. These two approaches to model-robust design are illustrated with examples.