Decomposition of the level effect into overall and descriptor-specific components

Despite training, subjects in a descriptive panel can still differ in their use of the scale. Some subjects can score higher or lower than others (level effect) or spread more or less their scores on the scale (scaling effect). The scaling effect, as calculated in Brockhoff, Schlich, and Skovgaard (2015), was recently decomposed into an overall and a descriptor-specific component (Peltier, Visalli, &Schlich, 2015b). It was suggested that the overall component was related to psychological effect whereas the specific one was more related to a physiological one. This paper aims to extend this decomposition to the level effect. The overall level effect gives indications about the psychological component of the scoring level, which is obtained by averaging the level effects of all the descriptors, whereas the descriptor-specific component can reveal a subject's hyper-or hyposensitivity to a given descriptor and thus is related to the individual's physiological response. The relevance of this decomposition was demonstrated by a meta-analysis of 419 sensory profiling datasets. Finally, the summary table of performances MAM-CAP Table (Peltier, Brockhoff, Visalli, & Schlich, 2014) was improved in order to show both level and scaling effects (overall and specific) for facile monitoring of individual differences in the use of sensory scales by a panel.