Principal component analysis of sensory panel results for a reference and multiple prototypes

Trained sensory assessors often evaluate samples by quantifying the intensities of sensory attributes. In some cases, samples are instances of in-market or prototype products. To explore results from the panel, it is conventional to obtain a products-by-attributes table of means, center and variance-standardize its columns, then conduct principal component analysis (PCA). The principal components that extract variance maximally from the products also extract variance maximally from all product paired comparisons. However, if there is a successful in-market reference product and multiple prototypes, then this PCA does not extract variance maximally from only the reference-prototype paired comparisons of primary interest. To investigate the relevant subset of paired comparisons, we create the input matrix for PCA using only the rows for paired comparisons of primary interest. Advantages of PCA of this modified input matrix are demonstrated visually and numerically using a case study involving cheddar cheeses. Data sets with related structures, including temporal sensory data sets and data sets in domains outside sensory evaluation can also be investigated using this approach.