Statistica Sinica 22 (2012), 1273-1285

MINIMAL DEPENDENT SETS FOR EVALUATING

SUPERSATURATED DESIGNS

Arden Miller and Boxin Tang

University of Auckland and Simon Fraser University

Abstract: This paper investigates minimal dependent sets used for evaluating supersaturated designs. Unlike the popular $E(s^2)$ criterion, the criteria based on minimal dependent sets directly capture the properties of designs in terms of estimation and identification of active factors. This paper provides a theoretical investigation into the number and structure of minimal dependent sets in a supersaturated design, and presents some construction results on supersaturated designs with large minimal dependent sets.

Key words and phrases: $E(s^2)$-optimality, estimation capacity, model discrimination, screening design.