Bayesian Model Selection Applied to Definitive Screening Designs

• Activity Number: 87
• Type: Contributed
• Date/Time: Sunday, July 31, 2016 : 4:00 PM to 5:50 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Patrick Gaffney*
• Companies: Eli Lilly and Company
• Keywords: Definitive Screening Designs ; rjags ; MCMC ; Bayes Factors ; Active factors ; DOE

Abstract

Definite Screening Designs (DSDs) have the ability to detect curvature in a relatively small number of runs and are gaining increased attention especially in the Biopharmaceutical Industry. Yet determining the number of active factors remains an issue for many practitioners. Some suggestions for performing model selection are given in Jones and Nachtsheim (2011). These include best-subsets regression based on the corrected Akaike's information criterion (AICc) and a recommended forward stepwise regression approach. Ramsey, Weese and Montgomery (2015) list other approaches and note that it is difficult to make blanket recommendations from the case studies they explored. This paper proposes an alternative approach to model selection for Definitive Screening Designs using a Bayesian framework and provides an assessment of its frequentist properties. Both rjags and R programming are used to draw samples from the posterior and to explore the effect of different priors - including the prior for the number of active factors. Bayes Factors are then used to provide inference for the number of active terms in the model.