Fraction of Design Space Plots for Examining Mixture Design Robustness to Measurement Errors

• Activity Number: 28
• Type: Contributed
• Date/Time: Sunday, August 7, 2005 : 2:00 PM to 3:50 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Ayca Ozol-Godfrey*+ and Christine Anderson-Cook
• Companies: Wyeth and Los Alamos National Laboratory
• Keywords: mixture experiment ; measurement error ; FDS plot ; scaled prediction variance

Abstract

Mixture experiments are experiments where the components are the ingredients of a mixture. When conducting an experiment, measurement errors might appear due to inaccurate measurements of the individual factor amounts. For mixture experiments, these errors affect not only the mismeasured component, but other components because the overall volume (or weight) is changed. In choosing between competing mixture experiments, the practitioner's main interest might be the prediction properties, specifically the scaled prediction variance (SPV) properties of a mixture design. In this presentation, we use Fraction of Design Space (FDS) plots to examine design robustness to mixing measurement errors for SPV. Two types of measurement errors in the component amounts are considered: relative and absolute errors. To measure the impact on prediction performance, both types are randomly generated from normal and uniform distributions. Two standard mixture designs are compared for robustness to measurement errors using FDS plots under a variety of error structures.