A0913
Title: A complete catalog of D- and A-optimal designs with up to 20 runs
Authors: Mohammed Saif Ismail Hameed - KU Leuven (Belgium) [presenting]
Eric Schoen - KU Leuven (Belgium)
Jose Nunez Ares - KU Leuven (Belgium)
Peter Goos - Universiteit Antwerpen (Belgium)
Abstract: The literature on two-level D- and A-optimal designs for the main-effects model is very exhaustive for run sizes that are multiples of four. This is due to the fact that complete catalogs of D- and A-optimal designs exist for run sizes that are multiples of four. However, for run sizes that are not multiples of four, there are no such catalogs, and experimenters resort to heuristic optimization algorithms to create designs. This approach has multiple weaknesses. First, it requires computing time. Second, heuristic optimization algorithms often fail to return a truly optimal design. Third, even in the event the design produced is truly optimal for the main-effects model, it often exhibits substantial aliasing between the main effects and the two-factor interactions as well as among the two-factor interactions. The purpose is to explain how to enumerate complete catalogs of D- and A-optimal main-effects designs for run sizes that are not multiples of four and how to select the best of these designs in terms of aliasing between the main effects and the two-factor interactions and among the two-factor interactions. As a result, the use of heuristic optimization can be avoided for most optimal design problems where the run size is at most 20 in the event the primary interest of the experimenter is in a main-effects model and statistical software can provide a minimally aliased D- and A-optimal design instantaneously.
