Selection of robust estimator usedin analysis of sensory characteristics and identification of environments conducive to specialty coffee production

Abstract

We propose multivariate regression models to identify environments for specialty coffee production. Simulation studies were carried out, taking as criterion the relative statistical efficiency between robust and non robust estimators in the presence of outliers related to covariance matrix of data. After selecting the model, we considered the sensory variables Aroma, Flavor, Acidity, Body and Balance, and the following chemical variables: final contents of trigonelline, chlorogenic acids and caffeine in the analysis of 14 genotypes of Bourbon cultivar evaluated in experiments in different environments. We concluded that the regression model considering MCD method showed larger evidence for identifying environments for specialty coffee production. In the case of MVE method, the multiple model showed greater evidence for discriminating genotypes.