CMStatistics 2015: Start Registration
View Submission - CMStatistics
B1274
Title: Supersaturated split-plot screening experiments Authors:  Emily S Matthews - University of Southampton (United Kingdom) [presenting]
Abstract: A key step in early industrial experimentation is screening to identify those factors that, when their levels are varied, have a substantive impact on the measured response. It is common for such experiments to investigate many factors in only a small number of runs, and increasingly supersaturated designs, with fewer runs than factors, are employed. The runs of the experiment are also often grouped by the levels of one or more hard-to-change, or whole-plot, factors, resulting in a split-plot structure and the need to account for correlation between observations from units in the same whole-plot. Motivated by an example from materials engineering, new approaches to both the design and analysis of supersaturated split-plot experiments are discussed. A linear mixed model is assumed to describe the response, and methods for optimal design, model selection, and variance-component estimation are developed and presented. The methodology is demonstrated on a range of examples, reflecting realistic industrial and scientific experiments, including the motivating engineering application.