Bayesian Factor Screening and Response Surface Methodology
by
Derek Bingham
Department of Statistics and Actuarial Science, Simon Fraser University
Coauthors: Hugh Chipman (University of Waterloo), Abhyuday Mandal (Georgia Tech)

The usual approach to response surface methodology is to first perform factor screening, followed by response surface exploration using different experiment plans. Recently, a new approach has been proposed (Cheng and Wu 2001) that aims to achieve both goals using one experimental design. The methodology uses a two-stage analysis approach which first attempts to identify the important factors and then projects the design into a more comprehensive model space for the important factors only. A design criterion based on the analysis approach was proposed. We demonstrate that this methodology can lead to spurious identification of effects. We instead proposed a new design and analysis methodology that aims to overcome these difficulties. The approach is Bayesian and attempts to more directly incorporate the standard assumptions of industrial experiments into the design and analysis. New designs are found and the performance compared to standard designs. Related computational issues are discussed.

Date received: August 26, 2003