S3RI Seminar - Optimal restricted-randomised experiments that allow for pure-error estimation of the variance components - Kalliopi Mylona, Kings College, London Seminar

Event details

Textbooks on response surface methodology generally stress the importance of lack-of-fit tests. For lack-of-fit tests to be possible, experiments should allow for pure-error estimation. Therefore, they should involve replicated treatments. Moreover, very often, the experimental runs cannot be performed under homogeneous circumstances, in which case the experiment is blocked. On other occasions, the experiment involves hard-to-change factors or multiple steps in a process. In that case, the experiment generally is a split-plot experiment. In this contribution, we present a novel approach to design blocked and split-plot experiments. The key feature of the approach is that it ensures that the two variance components resulting from the blocked or split-plot nature of the experiment can be estimated from pure error, in addition to a precise estimation of the treatment effects and the response surface model. Our novel approach involves a new Bayesian compound D-optimal design criterion which pays attention to both the variance components and the fixed treatment effects.