2016-Isobel Loutit Lecture

Isobel Loutit Lecture 
Organizer and Chair: Hugh Chipman (Acadia)

CHRISTINE M. ANDERSON-COOK, Los Alamos National Laboratory
Realistic Optimal Designs in a Resource-Constrained World  [PDF]
Historically designed experiments were chosen from a collection of designs itemized in the statistics literature, with limited size and design region shape options. These designs had good general properties with some robustness across applications. Next came computer-generated designs optimal for a single criterion. These designs allowed greater size and region flexibility, but lacked overall robustness. An emerging area seeks to combine the best of both worlds: well-rounded performance with increased flexibility. Priorities of experimenters differ and economic pressures on experiments are increasing. Hence, identifying leading candidate designs based on user-specified design criteria allows for better designs. A flexible two-stage Pareto front approach is described, which objectively finds promising candidates using experimenter priorities. Diverse examples illustrate the approach.