Efficiency of nearly optimal design in experiments with mixture.

 Paulo Leonardo Sirimarco GOMES[1]

Carlos Alberto Ribeiro DINIZ1



§    ABSTRACT: In this work it is considered optimal designs for second-order models in experiment with three blending component proportions in orthogonal blocking. The optimal designs have been used extensively in mixture experiments, mainly when it is necessary to consider some external conditions to the experiments such presence of process variables. However, optimal designs have a drawback in the sense that they do not accommodate additional constraints. A reparametrization proposed by Prescott (1998) simplify the expressions associated to some optimal design criteria so that the runs used in the experiments could include a minimum proportion of each mixture ingredient. In this article, using a modified Prescott measure of efficiency, the criteria D-, A- and Eoptimal are evaluated and compared with respect to the performance in blends consisting of all the available ingredients.

§    KEYWORDS: Experiments with mixture; nearly optimal design.

[1] Departamento de Estatstica, Universidade Federal de S˜ao Carlos, Caixa Postal 676, CEP: 13565-905, S˜ao Carlos, SP, E-mail: dcad@power.ufscar.br