B0240
Title: A new Bayesian approach to control model misspecification in robust design
Authors: Irene Garcia Camacha Gutierrez - University of Castilla-La Mancha (Spain) [presenting]
Kalliopi Mylona - King's College London (United Kingdom)
Abstract: Robust design techniques are crucial in experiments where the behaviour of the response is poorly known prior to running the experiments. Although there are numerous alternatives to address this problem, the focus is on the approach introduced by a prior study, which used mean square error as a measure of design quality to control the bias introduced by model misspecification. This approach is quite general in the sense of covering a wide range of possible responses. Nevertheless, a strong assumption must be made by the experimenter under this framework: his/her degree of uncertainty about model adequacy. A Bayesian approach is proposed to deal with this assumption. A new optimality criterion is proposed, and numerical algorithms are provided to calculate this new class of optimal robust designs. Several examples illustrate the results.
