Assessing response's bias, quality of predictions, and robustness in multiresponse problems

Abstract

Optimization measures for evaluating compromise solutions in multiresponse problems formulated in the Response Surface Methodology framework are proposed. The measures take into account the desired properties of responses at optimal variable settings, namely, the bias, quality of predictions and robustness, which allow the analyst to achieve compromise solutions of interest and feasible in practice, namely in the case of a method that does not consider in the objective function the responses' variance level and correlation information is used. Two examples from the literature show the utility of the proposed measures. © 2011 Springer Science+Business Media B.V.