MODELING THE EVOLUTION OF THE PERFORMANCE OF A SENSORY PANEL: A MIXED-MODEL AND CONTROL CHART APPROACH

A sensory panel is often used to profile the same type of product with the same set of attributes for many years. We are interested in characterizing the evolution of the performance of such a panel (and its panelists) over time. This article presents a methodology based on a mixed‐model approach that takes into account the evolution of both panel and panelist in the same model. At the panel level, linear and quadratic evolutions of the performance are tested. At the panelist level, the method allows detection of whether some panelists perform better than others, and whether this difference remains the same or evolves over time. This mixed‐model approach is followed by a graphical representation using a control chart method to identify occasional outliers. Data used to illustrate this methodology are eight sensory profiling data sets collected on ready‐made frozen meals between 1997 and 2001 (every 6 months). The performance index chosen as an example in this study is the individual repeatability measured by standard deviation over replicates.     

MULTIVARIATE ANALYSIS OF CONVENTIONAL PROFILING DATA: A COMPARISON OF A BRITISH AND A NORWEGIAN TRAINED PANEL

Many studies have shown that conventional profiling provides reproducible and meaningful results. However, comparison of the technique as used in different countries appears to be nonexistent. In addition, data analysis is often approached differently, and this aspect is also addressed. This paper describes a study to compare the results obtained from profiling milk chocolate samples, using trained panels in Britain and Norway. Data were analyzed using principal component analysis, generalized Procrustes analysis and partial least squares regression. Results indicate that the underlying perceptual structure of the sample spaces obtained from both panels were similar, however, the emphasis on the underlying sensory dimensions differed. Moreover, it was possible to calibrate the two profiles, which has implications for marketing products for export, as well as providing a potential tool for panel monitoring and calibration across cultures.     

A STATISTICAL TEST OF CONSENSUS OBTAINED FROM GENERALIZED PROCRUSTES ANALYSIS OF SENSORY DATA

Matrix matching techniques such as Generalized Procrustes Analysis (GPA) always produce a matrix of maximal agreement which can then be used to graphically represent samples in “consensus plots”. The degree to which the consensus plots produced by GPA on sensory data (such as that obtained from free choice profiling) actually give a picture of true consensus among panelists, as opposed to being merely artifacts of the analysis, has not been examined. Using a Monte Carlo approach, a statistical test is defined for qualifying this consensus. Examples of the application of the test to sensory profiling data of fruit flavors are given.     

THE APPLICATION OF TWO FREE-CHOICE PROFILE METHODS TO INVESTIGATE THE SENSORY CHARACTERISTICS OF CHOCOLATE

The concept of free-choice profiling allows untrained assessors to participate in sensory profiling. However, it is sometimes the case that such individuals are unable to describe what they perceive when the samples are presented in the isolation of the sensory testing booths. This paper reports an experiment to compare the usual free-choice profile procedure, with a more structured approach based on Kelly's repertory grid method. Generalized Procrustes analysis was used to analyse the data. The sample configurations obtained from both approaches were very similar, as were the interpretations of the main perceptual dimensions.     

SELECTION AND SCREENING OF A DESCRIPTIVE PANEL FOR EWES MILK CHEESE SENSORY PROFILING

The aim of this work was to select a competent sensory panel considering its ability to describe ewes milk cheese sensory properties. The panelists evaluated the sensory characteristics of several ewes milk cheeses in terms of odor, flavor and texture. The performance of judges was assessed taking into account the following aspects: (1) individual discriminatory ability and reproducibility for each sensory attribute; (2) group discriminatory ability and reproducibility and agreement between judges for each attribute; (3) panel homogeneity; (4) panel consonance; and (5) overall panel discriminatory ability. The panel performance was checked with a set of statistical procedures. The results obtained by the selected group were better than the initial panel from the point of view of discriminant power, reproducibility and agreement. The suggested method can be easily applied in routine control of assessors performance and to discover deviant behavior or differences in individual perception of the samples.     

THE RELATIONSHIP BETWEEN CONSUMER PREFERENCES AND TRAINED PANEL RESPONSES

Eight samples of dried tomato soups were profiled by a trained panel, who also provided ratings of overall liking and liking for particular attributes on relative-to-ideal rating scales. A consumer panel tested a subset of four of the soups in home trials, assessing them for overall liking and for particular attributes on either hedonic or relative-to-ideal rating scales. The pattern of overall preferences differed between the trained and untrained panelists, demonstrating the inappropriateness of using trained panelists to provide measures of preference or acceptance. The conclusions regarding consumer preferences would differ depending on which rating scale was used; these differences disappeared when the relative-to-ideal ratings were converted into values of the same form as the hedonic ratings. The overall liking was best predicted by flavor rather than color or thickness. A principal components analysis (PCA) of the profile data compared well with a plot based on the trained panel preference data using MDPREF. The preference data were also fitted to the PCA dimensions using the PREMAP vector model, which gave a good fit for only six of the 15 trained panelists; the PREMAP ideal point model failed to show a better fit. In order to test these models adequately more than eight samples would need to be tested.     

MULTIDIMENSIONAL SORTING, SIMILARITY SCALING AND FREE-CHOICE PROFILING OF GRAPE JELLIES

The sensory data from three different methods, multidimensional sorting, similarity scaling, and free-choice profiling were compared using ten commercial grape jellies as a model system. Without any prior training, the overall similarities/dissimilarities between stimuli were judged using both multidimensional sorting and pair-wise scaling and the sensory attributes were rated using free-choice profiling by different panels. A two-dimensional stimulus configuration best represented the data from each of the three methods. The underlying dimensions of stimulus space were identified from free-choice profiling data and also explained some background variables. The texture, sweetness, sourness, grape flavor and color contributed to both dimensions. All three methods were very similar in describing the most important differences among stimuli as suggested by the highly significant correlation between their first dimensions. The Procrustes analysis coupled with permutation tests, as well as RV coefficient, indicated that similarity scaling and free-choice profiling reached maximum consensus, whereas multidimensional sorting shared slightly lesser consensus with the other two methods.     

GRAPES: A METHOD AND A SAS® PROGRAM FOR GRAPHICAL REPRESENTATIONS OF ASSESSOR PERFORMANCES

GRAPES computes individual and global analyses of variance for sensory profiling data, consisting of several sessions in which all the panelists gave scores to all the products for a number of attributes. The fitted model takes into account the session effect. GRAPES summarizes the results by means of graphical assessor scatterplots which allow to check and to compare panelist performances, such as the way of using scale, the reliability, the discrimination power and the agreement with the panel. In addition, GRAPES detects the outliers for each of these criterion. The usefulness of GRAPES for the panel leader will be demonstrated using texture and flavor profiling of 4 restructured steaks by 12 assessors for 15 attributes. The SAS® program GRAPES, available by e-mail from the author, requires the SAS/BASE®, SAS/STAT®, SAS/GRAPH® and SAS/IML® softwares.     

SENSORY EVALUATION BASED ON VERBAL JUDGMENTS

Studies of the repeatability and the homogeneity of expert panel scores in sensory profiling show that lasting and reliable evaluations of food products are difficult to obtain: strong inter- and intra-individual differences are commonly observed. Our hypothesis is that this variability is due to quantification methods that consist of asking panelists to furnish quantitative values (by attributing a numerical point to perceived intensity) and that using natural language in the form of verbal judgements in a hierarchical tree would allow improving the reliability of sensory evaluations. This hypothesis was tested by comparing a numerical value scale and a specific hierarchical semantic scale that subjects devised themselves prior to the experiment. The products (chocolate dessert creams) were evaluated using each of the two methods. The results show that the capacity to discriminate between products is better when verbal judgements are used. We discuss the advantages of hierarchical semantic scales for sensory profiling.