Sensometrics: Multivariate Analysis and Mapping of Sensory and Consumer data

Decision making in the industry worldwide is based more than ever on data, thus demanding for proper handling and modeling of data to a greater extent. In sensory studies, data are collected from trained assessors, based on hearing, sight, smell, taste and touch, to evaluate the characteristics of products and the dierences between them.

In consumer studies, the focus is on understanding the behavior and preferences on the products through a judging task given to a group of consumers. Relevant industries for sensory and consumer research span from food to audio, from fragrances to health care, from personal care to cars, and many others.

Mathematical and statistical methods are used to support market researches, product development and optimization, quality control and so on. The development and application of mathematics and statistics to problems from sensory and consumer science is called Sensometrics. Sensometricians are faced with the challenge of dealing with data coming from a perceptual process, as humans are used as measurement instruments.

The variation in the data can be modeled by using mixed models, where both xed and random components are taken into account. As a result, mixed eects models are generally preferred over xed eects models. Plenty of software tools for the specic sensory and consumer context have been developed. The multivariate nature of a typical dataset involving stimuli, attributes, assessors/consumers and replicates is evident.

Several scientists have raised concerns about the proper modeling of data in a multivariate setting and put some suggestions forward. However, there is a lack of comprehensive comparison of dierent multivariate approaches. Most of the analyses are still based on too simplistic models, where the nature of the data is not taken into account.

The intrinsic variation of the perpetual process needs to be properly handled in order to draw the right conclusions from the analysis of the data. Scope of this thesis is to bridge deterministic and probabilistic approaches to multivariate data analysis. Generic and novel multivariate methods are investigated and applied to sensory and consumer problems.

An algorithm and its implementation in the statistical software R is put forward and suggested as a suitable method for taking into account the variation inherent the perceptual process. This project brings closer together, and within the statistical framework, methods developed in specic scientic areas, namely chemometrics and psychometrics.

Concepts and methodologies are investigated in an attempt to unifying notation and terminologies and specically compared for sensory descriptive data using both a real dataset from the industry and a simulated example.