Analysis of sensory assessors in texture profiling of potatoes by multivariate modelling

The aim of the present study was to analyse the reliability and performance of the sensory assessors in texture profiling by a univariate (ANOVA) and a multivariate procedure using discriminant partial least squares regression (DPLSR). Assessor×product modelling is relevant in the data analysis of sensory data, since potato tubers are slightly heterogeneous, even when grown under controlled conditions and sorted with respect to size and density. Thus, different assessors may in fact get "different" products. Also the assessor level differences were removed by the modelling. Signal to noise plots for all attributes showed reflection, mealiness and firmness to be the most reliable attributes, and springiness and chewiness to be the least reliable attributes. Signal to noise plots for each attribute illustrated some assessors to be good in evaluation of some attributes and less good in others. The assessors were questioned about psychological states after each session in order to study if individual differences influenced the assessorsʹ performance. Some assessors reporting that they felt concentrated, were in fact found to be less reliable than others who felt stressed.