3 16
Abstracts
of Oral
Presentations
descriptive analysis. However, to our knowledge, no assumption has tested the predictive value of these tests on the future performance of the selected panelists. The aim of this study is to check if such links could be verified. Eighteen subjects were selected amongst about a hundred according to the procedure described by Issanchou et al. (submitted) in order to perform quantitative descriptive profile on Camembert cheeses. Seven sensory tasks were designed to test taste and smell sensitivities, odor memory and description ability of the candidates. Forty-five hours of training were conducted to learn panelists sensory procedures and to build a descriptive vocabulary of Camembert cheeses. Sensory measurements were then done in duplicates on sixteen samples of cheeses by using 19 flavor, 3 taste, 1 mouthfeel, 11 texture and 3 after-taste descriptors. The determination of panelists performances was based on repeatability and discrimination abilities, and on complexity of the individual sensory space of the 16 cheeses. Data analysis were conducted to expalin sensory profiling performances by selection scores. It revealed that subjects who obtained high scores on familiar odors recognition and on olfactory memory task were good discriminators on aroma perceptions. Thus, it appears that to be discriminating on flavor attributes, subjects must either have an initial olfactive culture or have a good ability to memorize olfactive stimuli. This study demonstrates the importance of the selection as significant effects were observed within a small range of selection scores.
33.SlMPLETOOLS FOR ASSESSING THE QUALITYOFASENSORY PANEL Tormod Naes, MATFORSK,
Oslovegen
1, 1430 As, Norway
Descriptive sensory analysis with a trained panel is an important way of measuring the quality of food products. The data from such analyses are usually submitted to simple ANOVA based on raw data, or by multivariate analysis based on the average profiles. However, sensory analysis is quite complicated and there are many pitfalls, so a quality check of the data before statistical analysis is often necessary. Many researchers in the area of sensory analysis have their own way of checking the quality of their panel and the data it produces, but better and more easily interpretable procedures are needed. Recently, a number of publications have appeared that use Procrustes analysis as a tool for panel evaluation, but Procrustes analysis is based on a specific model which may be unrealistic and it can be rather difficult to interpret for persons with little experience in multivariate statistics. The present paper will present some alternative tools which are quite simple and easy to interpret. In particular, we will discuss a newly developed method based on plotting the cumulative ranks of the assessors, the so-called eggshell
plot. In addition, we will discuss methods based on simple analysis of variance models. Both approaches provide simple graphs which are easy to interpret.
34.VALlDATlON OFSENSORYANALYSIS OF BEEFTENDERNESS Marit Redbotten, Per Lea and Kjell lvar Hildrum, MATFORSK, Norwegian Food Research Osloveien 1, N- 1430 As, Norway
Institute,
A critical point in sensory analysis is the degree of reproducibility of each single measurement. It is important that each assessor can discriminate between different samples and repeat the assessment when duplicate samples are served. This is equally important for all sensory attributes. This work will describe differences among panel members in ability to analyze different sensory attributes. In our experiment M. longissimus dorsi muscles from 40 animals of the Norwegian Red Cattle breed, seven cows and 33 young bulls were used to span a wide sample space in relation to texture. Eleven well trained sensory assessors used a QDA profile method for testing the samples. Intensities of the texture attributes, tenderness, hardness and juiciness were evaluated using a continuous non-structured scale. Individual differences in rating within the panel members were studied by using the statistical tool ‘Eggshell-plot’ (Hirst & Naes, 1994). The MSE-value (Mean Square Error) from the analysis of variance performed for each assessor separately, which measures the assessor’s variance over replicates, gives information on the panelist’s ability to repeat themself when judging identical samples blindly coded. This observation, in addition to the ability to discriminate between samples that are different (low p-value from the above mentioned analysis of variance), can be used as a measure of the panei members’ professional performance. Comparing with each panel members’ sensitivity-testing of sweet, sour, salt, and bitter taste (IS0 3972:1991 E), we found that the panelists with the highest sensitivity for tastes also are the best panelists in discriminating meat samples for texture attributes.
35. TWO-DIMENSIONAL COVARIANCE COMPONENTSAPPLIEDTOSENSORY DATA Per M. Brockhoff and Barbara GuggenbOhl, Department of Mathematics and Physics, Royal Veterinary and Agricultural University, Thorvaldsensvej 40, DK-1871 Frederiksberg C, Denmark
The two-dimensional generalization of the mixed linear model is presented as a way to analyze the correlation between sensory observations and chemical/physical mea-