While response surface experiments should ideally focus on predictive capability, the most common criterion for creating optimal response surface designs is D-optimality, which minimizes the variance of all factor effect estimates collectively. I-optimal designs minimize the average variance of prediction over the experimental region, making them more appropriate than D-optimality for response surface designs. This article discusses I-optimal design for split-plot response surface experiments. Several examples show these designs to have improved prediction over D-optimal split-plot designs and to produce very precise factor effect estimates.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados