Canonical variate analysis of sensory profiling data
Résumé
Principal Component Analysis (PCA) of product mean scores is generally used to obtain a product map from sensory profiling data. However, this approach does not take into account the variance of the product mean scores due to the individual panelist variability. Therefore, Canonical Variate Analysis (CVA) of the product effect in the two-way multivariate analysis of variance (MANOVA) should be considered as a natural alternative analysis to PCA. Indeed, it is the extension of the classical univariate approach used for the analysis of each descriptor separately. This analysis generates successive components maximizing product discrimination as measured by the usual Fisher statistics in analysis of variance. The outcome of CVA is a product map maximizing product separation, while gathering individual evaluations of the same product as close as possible. This article is an introduction to the rationale of CVA applied to sensory profiling data. In addition, it offers an R package (CVApack) and instructions for running it.