Linear mixed models in sensometrics

Today’s companies and researchers gather large amounts of data of different kind. In consumer studies the objective is the collection of the data to better understand consumer acceptance of products. In such studies a number of persons (generally not trained) are selected in order to score products in terms of preferences. In sensory studies the aim is the collection of the data to better describe products and differences of the products according to a number of sensory attributes. Here trained persons, so-called assessors, score the products in terms of different characteristics such as smell, taste, texture, sound - depending on the aim of a study. It is a common approach in both studies to consider persons coming from a larger population, which, from the statistical perspective, leads to the use of mixed effects models, where consumers/assessors enter as random effects (Lawless and Heymann, 1997).

Mixed effects models have been used extensively in analysis of both consumer and sensory studies. However frequently too simplistic models are considered, important effects are not accounted for and as a consequence important information is not gained or analysis leads to improper conclusions. The focus of this project is to propose a methodology for analyzing more complex models together with tools facilitating the methodology. This was accomplished by contributing to the mixed effects ANOVA modelling in general and specifically applied to sensory and consumer studies through a series of papers and software tools facilitating the developed methodologies. The primary advantage of the ANOVA approach is that it gives confidence intervals and significance tests for the various effects including the background variables used in the model and consequently a fast and reliable assessment and ranking of the importance of different factors.

There exists today very little easily available methodology and software which supports consumer studies with both sensory properties and background information related to health benefits, environment and user-friendliness. In close collaboration with the industrial partners an open-source software tool ConsumerCheck was developed in this project and now is available for everyone. will represent a major step forward when concerns this important problem in modern consumer driven product development. Standard statistical software packages can be used for some of the purposes, but for the specific problems considered here and for the typical users in industry, these programs are far from satisfactory. Therefore, the ConsumerCheck software represents a novel source of information for all quality-oriented industries. The effect is improved procedures for product development and hence improved quality of decision making in Danish as well as international food companies and other companies using the same methods.

The two open-source R packages lmerTest and SensMixed implement and support the methodological developments in the research papers as well as the ANOVA modelling part of the ConsumerCheck software. The SensMixed package is a package for semi-automated analysis of sensory and consumer studies within linear mixed effects framework. The lmerTest package supports tests for linear mixed effects models fitted with the lmer function of the lme4 package (Bates et al., 2013). While SensMixed is closely connected with sensometrics field, the lmerTest package has developed into a generic statistical package.
Reference manuals accompany these R packages.