A1173
Title: Optimal two-level designs under model uncertainty
Authors: Pi-Wen Tsai - National Taiwan Normal University (Taiwan) [presenting]
Steven Gilmour - KCL (United Kingdom)
Abstract: Two-level designs are widely used for screening experiments, with the goal of identifying a few active factors that have major effects. The model-robust $Q_B$ criterion is applied for the selection of optimal two-level designs without the requirement of level balance and pairwise orthogonality. A coordinate exchange algorithm is provided for the construction of $Q_B$-optimal designs for the first-order maximal model and second-order maximal model, and it is demonstrated that different designs are recommended based on different experimenters' prior beliefs. Additionally, the definition of $Q_B$-criterion is extended to regular and irregular block designs and the relationship between this new criterion and the aberration-type criteria for blocks is studied. Some trade-offs between orthogonality and confounding will lead to different choice of block designs. Some new classes of model-robust designs that respect experimenters' prior beliefs have been found.
