Equações de estimações generalizadas para dados ordinais em análise sensorial de cafés especiais e critérios de seleção para matrizes de correlação de trabalho

Abstract

In this work two parts are presented. The first part considers the theoretical basis of this thesis. The second part is composed of two scientific articles. The first article refers to modeling in sensory analysis for multiple repeated responses in an experiment with specialty coffees. In the sensory analysis applied to specialty coffees, it was possible to construct a data set with repeated measurements at taster / genotype levels and over four crop seasons. This was due to the fact that different tasters for different cup tests carried out evaluations of the same genotype in two situations: throughout the crop seasons and during the execution of the tasting to assign the notes. In this sense, it was necessary to study the associations in two directions. The first one regarding the taster and the second direction associated with the grades to the effect of the harvest. It was concluded that the methodology proposed in this first article identified the sensory covariates that are similar throughout the harvests, as well as producing estimates of probability for the categorization of specialty coffees in the best grades classes, associated with tastings performed by harvest. The second article presents a selection criterion for labor correlation matrix, used in generalized estimation equations. This criterion, unlike the selection criteria presented in this paper, makes use of the limiting estimate of the association parameters as a measure for the choice of the work correlation matrix. For that, Monte Carlo simulation was performed with different scenarios, comparing its result with the other criteria. In addition, two applications are presented, one related to a set of literature data and the other refers to the set of data coming from a sensory analysis of specialty coffees. It was possible to conclude that the proposed criterion proved to be competitive to the other criteria.