Testes assintóticos para detectar consenso multivariado em painéis sensoriais

Abstract

The unidimensionality of a sensory panel is directly related to the panel consonance, i.e., a panel is considered unidimensional when assessors score in the same way a particular attribute. Due to the importance of the panel agreement to the reliability of sensory analysis panelists should be trained in order that they agree with each other regarding the characteristics of an attribute. In the literature several methods have been proposed to assess such agreement, although existing methods evaluate the marks for one attribute at a time, making the analysis slower. The objective of this study is to generalize the asymptotic test eigenvalues proposed by Ferreira (2008a), in order to infer about the multivariate consensus of sensory panels. From the generalization of the asymptotic test of eigenvalues it was possible to obtain four new test statistics. The evaluation of the tests was conducted via Monte Carlo simulation, in which were evaluated different scenarios resulting from the combination of the numbers of panelists (2, 5, 10 and 15), attributes (2, 5, 10 and 20), observations (10, 20, 30, 40, 50, 100 and 200) , degree of training of the sensory panel (0;1 ≥ p² ≥ 0;99) and the restriction n ≥ pq. Overall, analyzing the type I error and the power function of the tests, the test InvH2 was more efficient.