Assessing Designs for Mixture Experiments for the Generalized Linear Models Case

• Activity Number: 331
• Type: Topic Contributed
• Date/Time: Wednesday, August 11, 2004 : 10:30 AM to 12:20 PM
• Sponsor: Section on Quality and Productivity
• Author(s): Ayca Ozol-Godfrey*+ and Christine Anderson-Cook and Edward L. Boone
• Companies: Virginia Polytechnic Institute and State University and Virginia Polytechnic Institute and State University and University of North Carolina, Wilmington
• Keywords: response surface ; mixture experiments ; FDS Plot ; non-normal data ; robustness to parameter misspecification

Abstract

A mixture experiment is a special type of a response surface experiment where the factors are the ingredients or the components of a mixture, and the response depends on the relative proportions of the components. Frequently, while examining these designs, normality for the response is assumed, and the linear model form is used. However, recently different types of non-normal data and the ways to examine them have been studied. We will study mixture experiments using Generalized Linear Models (GLM) for a non-normal response. The Fraction of Design Space (FDS) Plot, a new graphical design evaluation tool by Zahran, Anderson-Cook, and Myers (2003) will be reviewed, and it will be adapted to examine the mixture designs for the GLM case. A variety of types of robustness to initial parameter estimates will be evaluated using these plots. The methods will be illustrated with several examples.