Unified Multivariate Ordinal Model for analysis of sensory attributes

TL;DR

This paper introduces a unified multivariate ordinal model for sensory analysis, enabling comprehensive evaluation of food attributes with improved efficiency and accuracy over separate models, demonstrated through simulations and a real beverage study.

Contribution

The paper presents a novel unified multivariate model for sensory attribute analysis, integrating multiple responses into a single framework based on multivariate categorical data methods.

Findings

Unified model achieved over 80% concordance in simulations.

The model identified preferred prebiotic beverage formulations.

Unified analysis reduced computational time compared to separate models.

Abstract

Experiments involving sensory analysis of foods and beverages are beneficial for selecting healthy products and assessing the preferences of potential consumers. They are generally planned in incomplete blocks, and their attributes, such as aroma, colour, and flavour, are evaluated using a 9-point hedonic scale, characterising an ordinal variable response. Also, the generalised logit model with random effects for panellists is one of the appropriate models to relate the multivariate response to the covariates. This study aims to present a method for analysing sensory attributes through a unified multivariate model. Due to the nature of the variable, each separate model already corresponds to a multivariate analysis, so our proposal would incorporate a complete analysis with solely one model. This proposal is based on multivariate methods for categorical data and maximum likelihood theory. Our method was evaluated through a simulation study, in which we consider three distinct formulations with two attributes to represent various formulation selection scenarios via mixed discrete models. The simulated results demonstrated overall concordance rates exceeding 80\% for the unified model compared to the separate models. Moreover, as motivation is presented, a study of 13 prebiotic beverages based on cashew nut almonds added to grape juice, with 130 potential consumers. The attributes evaluated were overall impression, aroma, Body, sweetness and flavour, using a 9-point hedonic scale. The selected unified model considering all attributes was the non-proportional odds mixed-effect model. According to this model, the prebiotic beverage formulations most likely to be accepted were: 8\% sugar and 40\% grape juice ($F_4$), 6\% sugar and 44\% grape juice ($F_6$), and 9\% sugar and 30\% grape juice ($F_{13}$). The unified analysis and computational time showed the advantages of this proposal.