Testes Monte Carlo para autovalores na verificação de consenso multivariado de painéis sensoriais

Abstract

In sensory analysis, the concordance between the members of a panel is fundamental to verify the efficiency of the training received by it. To this purpose, some methods and tests are found in the literature, among which are the asymptotic tests of Ferreira (2017), which evaluate the consonance of a panel in a multivariate form. Because they are asymptotic tests, they require a large number of sample observations, which is not common in the sensory context. Therefore, the objective of this work was to propose Monte Carlo versions of the Ferreira tests (2017) to obtain better characteristics of power and type I error in any sample size, as well as to compare them with the asymptotic tests via computerized simulation. It was also an objective to adapt the panel performance verification protocol proposed by Tomic et al. (2009), inserting a multivariate consensus test step. The four Monte Carlo tests proposed were considered accurate and presented higher power in 95,24% of the analyzed scenarios, when compared with the asymptotic versions. The use of the InvH2mc test is recommended. The efficiency of this test could be proven by verifying the performance of a sensory panel when evaluating two brands of guarana soda flavor in both their traditional and zero sugar versions. It was found that the panel did not present a multivariate consensus, that is, the tasters did not score the attributes similarly. The insertion of the Monte Carlo test in the protocol allowed inferential analysis of the panel’s behavior, corroborating the exploratory phases.