A quality measurement for sensory profile analysis The contribution of extended cross-validation and resampling techniques

Abstract

The sensory profile measures an interaction between products and subjects. Faced with this complex information, industry needs an operational tool for describing product sensory properties. Thus, the statistical analysis of profile data should yield a sharp and robust product picture. We set up quality criteria, based on extended cross-validation and re-sampling techniques, to quantify the relevance of factorial method results. We target PCA on mean product scores, and several MANOVA models. We used assessor cross-validation and assessor re-sampling to evaluate respectively the impact of each assessor and the impact of the panel make-up on the product results. PCA and simple MANOVA models are highly robust, whereas complex MANOVA models lead to unstable results.