A Multithreshold Model for Sensory Analysis

ABSTRACT:  A multithreshold Bayesian procedure involving a probit approach is presented to analyze data obtained from sensory panels. The procedure assumed that the analyzed trait follows an underlying Gaussian distribution and each panelist applied individual thresholding. The proposed method provides estimates of the effects that affect the underlying measure and also estimates the panelists' ability to discriminate between the categories. The procedure avoids the use of any form of standardization technique, and the results obtained are expressed as standard deviations of the underlying trait. As an example, the procedure was applied to previously analyzed data from an experiment involving rabbit selection, in which animals from the 7th generation were compared with those from the 21st generation. Seventh generation embryos were frozen, thawed, and implanted in 21st generation does. The control (C) and selected (S) groups were contemporary. Sensory analysis was carried on samples of the longissimus dorsi muscle. The parameters evaluated were: intensity of rabbit flavor (IRF), aniseed odor (AO), aniseed flavor (AF), liver flavor (LF), tenderness (T), juiciness (J), and fibrousness (F). There were substantial differences between the selected group and control group for IRF, AO, AF, and LF and between males and females for IRF, AF, T, J, and F, which confirmed previous results. The proposed procedure has a greater ability to detect these differences, and in addition provides information about the ability of panelists to discriminate between samples.