Panel performance

This section details a way to proceed for the evaluation of the performance of a sensory panel in the TDS framework. It is proposed to use a dedicated testing protocol to better target the question of the panel performance, then to pre-process the data to fit with the performance context and finally to analyse the data according to a specific method to determine if the panel is performing well, highlight the good performing panellists and/or attributes and quickly identify the main points to be improved. 

Even if panel performance should be the first step before going for product evaluation, it is presented at the end of the data analysis part, since it relies on the product difference test concepts presented before. But contrary to product differences, the focus of Ihe panel performance is to assess the ability of the panel to give consistent product differences across subjects rather than evidencing product differences as such. For this reason, Lepage et al. (2014), proposed a simple a priori protocol to check panel performance prior to product evaluation. This is based on the evaluation of a limited number of samples (say four) representative of the product space to be evaluated. Because of the nature of a TDS evaluation (selection of attributes as dominant or not, i.e. 0/1 data), the subjects are supposed to test all products at least in triplicate. In addition, to ensure that subjects results are fully comparable and not biased by any differences in the product presentation order, the same presentation order is given to all subjects for all replicates. 

Rapid Sensory Profiling Techniques and Related Methods 

The four performance indexes are calculated at four different levels  

The calculation of each index is derived from the usual sum of square decomposition used in ANOVA, but applied to dominance rates. For instance, the discrimination index at panel level is simply the sum of square of the product effect (see the original paper for more details on the other indexes). However, the authors do not follow the F-test approach to test the significance for these indexes, since TDS data (or the residuals from any standard ANOVA model) are not normally distributed. They rather follow the permutation test approach proposed by Meyners and Pineau (2010) and Meyners (2011) and extend it to the scope of their indexes. The reader can refer to the original paper for more details about the testing procedure. 

---

A sandwich panel performs by improving loading characteristics such as in bending in the direction perpendicular to the plane of... 

CPT codes also indicate which laboratory procedures were offered. On the billing claim form, each laboratory test or laboratory panel performed is coded separately. An example of a CPT laboratory code is 83036, which corresponds to a hemoglobin A c determination (AMA, 2006). The CPT code 83036QW indicates a hemoglobin A c determination was performed using a CLIA-waived device. A complete list of CLIA-waived tests and their CPT codes is located on the CMS Web site (CMS, 2007b, 2007c). 

In three-layer flakeboards based on five softwood and hardwood species of 4, 10, and 18% moisture content, with press temperature of 177 °C, 6-min press time, and 3% isocyanate binder, the results showed that moisture content of wood was the most important variable at 18% moisture level, IB and bonding properties were lowest. Species of wood influenced strongly the bonding efficiency. In almost all cases the bending properties were the key characteristic of the panel performance. Southern pine produced the boards with lowest IB (81-116 psi), and red oak gave the highest IB values (98-213 psi) (J95). 

When the strategy is to confirm aU the negative results of the first test (see Table 15-7), the first test should be the more sensitive so as to minimize costs. As demonstrated in the two examples presented earlier, the decision rule used preferen-tiafty trades off sensitivity at the expense of specificity, or vice versa. Although independent tests have been used in these examples, the conclusions are the same for dependent tests. It should be remembered that it is the interpretive rule and the two tests that determine the overall panel performance and costs the order of testing does not affect performance but can dramatically affect costs. 

The vertical density profile provides a means of optimizing the press operation, as well as panel performance. While initially undertaken in the laboratory, the VDP can now be measured on the moving panel as it leaves the press, using an x-ray backscatter technique (Dueholm, 1996). This is an expensive tool, but the purchase is increasingly common as the value of real time VPD information is appreciated. 

The same approach is possible for the whole panel. The mean IFF value of the performance for the whole panel is called the panel performance factor. The mean value of the panel performance factor and the standard deviation give an indication of the quality of the whole panel related to 2-propanone concentrations. 

Figure 9 presents chamber data of only one set of hardwood plywood paneling performed at different loading and air change rates. 

The fact that the LMTO method can give only one set of z bands in any one calculation does not exclude the use of the method in cases where more than one principal quantum number gives rise to a band of z character. In such a situation one simply divides the energy range into panels, performs LMTO calculations with potential parameters appropriate to each panel, and pieces the individual bands together to form the complete band structure. That this in reality is the most efficient way of obtaining such bands may be seen... 

V Babrauskas, Sandwich panels performance in full scale and bench scale fire tests . Fire and Materials 1997 21 53. 

Then the task of the assessors is to mark across the line the point that represents the intensity of their perceptions. Repeated judgments from each panelist for each sample are collected and arithmetic means of the score for each attribute are obtained. Panelists do not discuss the data after each taste session, to reach a consensus, but data are statistically analyzed to determine differences between the products and to control for panelists and panel performance. 

In addition to this, one may want to assess the quality of FP measurements in terms of panel performance. In this regard, it should first be mentioned that panel performance cannot be understood as in conventional consensus-based methods, since panellists do not use the same attributes, and even if they do use the same words these words may not be assumed to convey the exact same sensory notion. Also, one should bear in mind that FP results are dependent on subjects descriptive skills, levels and fields of expertise. As already mentioned, should one wish to get the most comprehensive description of a product set, it is advisable to recruit experienced subjects with complementary fields of expertise. 

Figure 13.14 Data pre-processing for panel performance assessment.

Figure 13.16 Example of TDS perfonnance figure. Left part of the figure presents panel performance right part of the figure presents subject performance. Attrihutes are sorted according to n suhj ok subjects are sorted according to n att ok. Average eval time (s) average duration of an evaluation, n Att Sel/eval average number of attribute selection per evaluation. % max freq maximum frequency of selection of an attribute over time across products (extracted from TDS curves), n Att/subj ok number of attribute/subject with the diagnostic ok .

This example illustrates that such a panel performance tool can very directly help a panel leader to take corrective actions (i.e. retrain specific subjects on specific attributes). It is therefore a very useful tool at the end of the training phase in order to define if the panel can enter the main study or if it needs some further training. 

Lepage, M., Neville, T., Rytz, A., Schhch, P, Martin, N. and Pineau, N. (2014). Panel performance for Temporal Dominance of Sensations. Food Quality and Pr erence, 38, 24-29. 

Although the statistics are different, this is in line with panel performance routines performed on expert data. Such a procedure cannot be performed for methodologies in which the ideal product is estimated statistically. And, as for JAR scale, no similar routine exists, to our knowledge. 

Tomic, O., Luciano, G., Nilsen, A., Hyldig, G., Lorensen, K. and Naes, T. (2010). Analysing sensory panel performance in a proficiency test using the PanelCheck software. European Food Research and Technology, 230, 497-511. 

---