A method for panelists' consistency assessment in sensory evaluations based on the Cronbach's alpha coefficient

In sensory analysis attributes are measured on samples based on human judgment. The ability to detect differences is essential when selecting a panelist, as well as the repeatability in assessments and the agreement among panelists (or reproducibility), which is our definition of panel consistency. Our goal in this paper is to identify an efficient method to compare evaluation profiles from panelists measuring a given sensory attribute on different samples, assessing the panel's consistency. For that we investigate two methods available in the literature - consonance analysis (Dijksterhuis, 1995), and repeatibility and reproducibility analysis (Rossi, 2001) - and propose a new method, based on the internal consistency test and the calculation of the Cronbach's alpha coefficient (Cronbach, 1951). We tested our proposition using a dataset from a case study in which beef cubes in stew, used as combat ration by the American Army, are characterized by a sensory panel using the Spectrum protocol. Different product formulations based on military specifications yielded eight samples evaluated by nine panelists in quadruplicate. Twenty-four sensory attributes were assessed by the panelists. Results pointed to the Cronbach's alpha coefficient as best among the methods tested, which is justified threefold: (i) it allows identifying attributes better understood by panelists, (ii) it gives a ranking of panelists according to their consensus with the rest of the panel, and (iii) it is analytically simpler in comparison with other methods. (C) 2013 Elsevier Ltd. All rights reserved.