Confidence ellipse for the sensory profiles obtained by principal component analysis

Abstract Sensory profiles are classically summed up by a principal component analysis (PCA) performed on the table of means crossing products and descriptors. This paper proposes a way for constructing a confidence ellipse for each product in the PCA score space. It is thus possible to have an idea of the uncertainty of the position of each product. To do this, the resampling methods used take into account the special nature of sensory data. This methodology is applied to two datasets coming from panels having very different sizes. Some of the distances between products, apparently small in the PCA representation, appeared “significant” in one case and not in the other.