Practical Assessment of Supersaturated Designs

• Activity Number: 157
• Type: Topic Contributed
• Date/Time: Monday, July 30, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): David Woods*+
• Companies: University of Southampton
• Keywords: Bayesian D-optimality ; Power ; Screening ; Shrinkage regression ; Variance Inflation Factors,

Abstract

Supersaturated designs are useful tools for industrial screening experiments that include many factors and limited resource. Such designs are often generated and assessed using criteria such as E(s^2)- or D-optimality, which encapsulate aspects of design performance based on pairwise column correlations. Such measures do not necessarily relate directly to the aim of the experiment, which is typically to identify a high proportion of the active, or important, factors whilst declaring few unimportant factors as active. Usually, it will also be necessary to consider linear dependencies between more than two factor columns.

This talk will discuss the assessment and application of supersaturated designs, motivated and demonstrated by examples from manufacturing in the pharmaceutical and chemical industries. Assessment is via simulation studies that vary many standard features of an experiment: the number of factors, the design, and the data generating process. From these results, some guidance on the effectiveness of supersaturated designs are established. Some new design selection criteria are motivated from the simulations, and demonstrated with respect to the industrial examples.