Inferential non‐centred principal curve analysis of time‐intensity curves in sensory analysis: the methodology and its application to beer astringency evaluation

Abstract

Improving technologies and better understanding of sensory phenomena have lead sensory analysts to develop statistical methods to assess sensations that endure over time (e.g. the bitterness or astringency of a beer) dynamically. The data produced by this type of experiment is classically a time-intensity (TI) curve, and their analysis remains an active research topic. The classical approach, widely used in this context, starts by extracting some significant parameters from the initial curves (maximum intensity, area under the curve (AUC), etc.). Descriptive data analysis or statistical modelling is then applied to get information from these summary parameters. This paper presents a different method, called inferential non-centred principal curve analysis (INCPCA), for the analysis of TI curves. It combines multivariate analysis (to visualise the curves in a space of smaller dimensions) with statistical modelling (aimed at enhancing the significance of factor effects). Non-centred principal curves (NCPCs) are first extracted from the curves matrix. They decompose the TI curves into different interpretable components. Score plots are used to represent the projection of the initial curves in the space of the first principal curves and allow factors and judge effects to be visualised. Mixed modelling is then applied to test the significance of these effects using PCA scores as model responses. The classical and INCPCA methods are illustrated on a TI experiment exploring the relation between beer astringency and three factors of interest: pH, O 2 content and aging. Eight beers arranged in a 2 3 factorial design were tested in triplicate by eight trained judges.