A0792
Title: Capturing column information of a design for models with uncertainty
Authors: Qi Zhou - Tianjin University of Finance And Economics (China) [presenting]
Abstract: Model-robust designs have been studied extensively in the literature. These designs are employed to achieve robustness over a set of possible models, which are usually assumed to have an equal probability of being the true model. However, they may not be appropriate for experiments where prior knowledge indicates that effects involving certain factors are more likely to be significant than others. In such cases, it is important to select designs that have superior estimation capacity and information capacity over a subset of the model space that contains models that are more likely to be important. This can be achieved by strategically assigning the factors to the columns of a fractional factorial design. The individual estimation capacity (iEC) and individual information capacity (iIC) are proposed, and these criteria are used to distinguish columns of a factorial design. For a given design, the maximum number of columns, g, is tabulated for which iEC=100\%. A new class of designs is proposed that maximizes g, and the trade-offs between the number of runs, the number of columns, and g are evaluated. The emphasis is on the model space that consists of models containing a subset of main effects and their associated two-factor interactions.
