Using partial bootstrap to evaluate the uncertainty associated with TCATA product trajectories
Temporal Check-All-That-Apply (TCATA; Castura et al., 2016) is a temporal sensory method in which assessors track changes in the applicability of sensory attributes to describe a sample during an evaluation. Data provide information on the complex dynamic profile of products. TCATA curves can be used to show attribute citation proportions over time, or differences in citation proportions between pairs of products.
Exploratory data analysis techniques such as correspondence analysis or principal component analysis can uncover underlying structure in the data. Using such analyses, it is possible to show product trajectories, which summarize the different changes that occur in the products over time. Trajectories can be smoothed to avoid overfitting and reveal underlying patterns in the data. Now, resampling techniques are proposed to further investigate the quality of product trajectories. Following an approach similar to Husson et al. (2005) and Cadoret et al. (2009), we use the partial bootstrap to obtain virtual panels. Each virtual panel is identical in size to the original panel but comprised of assessors sampled with replacement. New product coordinates, based on data from these virtual panels, are projected into the sensory space. Resampled data from the virtual panels are shown at each time slice along the product trajectories. These sensory changes are visualized using an animated sequence, with the accumulation forming product contrails. These contrails reveal the variability associated with each product trajectory, and aid in interpretation by illustrating the uncertainty associated with estimates. The proposed approach is illustrated using empirical data from a trained red wine panel in a recent experiment conducted by Baker et al. (2015). Two red wine treatments varying in ethanol content, Low (“L”) and High (“H”), with 10 and 15.5% v/v, respectively, were produced from musts with different initial sugar contents (21 and 27 °Brix, respectively). Post-production, L was manipulated to develop a third wine (Adjusted, “A”) with the same ethanol concentration as a trained panel (n=13), using a consensus vocabulary and having practiced the TCATA methodology, used TCATA methodology to characterize the flavours, tastes, and mouthfeels that linger and evolve after swallowing red wine. The panel evaluated the 3 wine treatments in two sips and in quadruplicate. Investigation using the partial bootstrap technique helps to avoid potential over-interpretation and provides a greater level of confidence in the conclusions obtained from the product trajectories.
Castura, J. C., Baker, A. K., & Ross, C. F. (2016). Using partial bootstrap to evaluate the uncertainty associated with TCATA product trajectories. 14th Agrostat Symposium on Statistical Methods for the Food Industry. 21-24 March. Lausanne, Switzerland.