PERMANOVA testing and Poisson Log-Normal modelling unravel how two traditional cheeses are distinguished through sorting and verbalization tasks

• Published in: Food quality and preference Vol. 104; no. 1
• Main Authors: Grollemund, P.-M., Lenoir, L., Benoit, J., Chassard, Christophe, Bord, C.
• Format: Journal Article
• Language: English
• Published: Elsevier 05.10.2022
• Subjects: Food engineering; Life Sciences; Statistics; PDO cheeses; PLN model; Verbalization task; Free sorting; PERMANOVA
• ISSN: 0950-3293
• DOI: 10.1002/9781118445112.stat07841

This study uses two statistical methods (PLN modelling and PERMANOVA) to investigate the differences in the ways in which different panels perceive and describe two French uncooked PDO cheeses in a free sorting task and then in a verbalization task.Panelists studied 10 cheeses from two categories Salers and Cantal cheese made from Salers-breed cow's milk or cow's milk from other breeds. We selected three types of panel based on their technical expertise in relation to the supply chain (professional technical experts, wholesalers, and consumers) and verified their level of expertise using a knowledge questionnaire to test o their knowledge of cheese in general and Salers and Cantal cheeses in particular. Data from the sorting task was analyzed using the DISTATIS method, and data from the verbalization task was analyzed using a basic correspondence analysis. Furthermore, we employed an original approach mobilising PERMANOVA and a Poisson log-normal (PLN) model enabling us to investigate and quantify which exogeneous variables contribute most to explain verbalization data.The results unsurprisingly showed broadly overlapping cheese configurations between the three panels. However, none of the panels clearly separated Cantal from Salers cheeses. In the verbalization task, different types of panel used different sets of terms to describe the categories. The professional panel preferentially used descriptive terms related to flavor whereas the wholesaler and consumer panels both tended to use quantitative (intensity) and hedonic terms. PLN modelling revealed that the knowledge variable was one of the variables that best explains the use of different word categories between panels. PERMANOVA testing and PLN modelling emerged as novel approaches for identifying the key variables that explain the use of terms in the description task.