Beta regression mixed model applied to sensory analysis

Abstract

Sensory analysis is an important area that the food industry can use to innovate and improve its products. This study involves a sample of individuals who can be trained or not to assess a product using a hedonic scale or notes, where the experimental design is a balanced incomplete block design. In this context, integrating sensory analysis with effective statistical methods, which consider the nature of the response variables, is essential to answer the aim of the experimental study. Some techniques are available to analyse sensory data, such as response surface models or categorical models. This article proposes using beta regression as an alternative to the proportional odds model, addressing some convergence problems, especially regarding the number of parameters. Moreover, the beta distribution is flexible for heteroscedasticity and asymmetry data. To this end, we conducted simulation studies that showed agreement rates in product selection using both models. Also, we presented a motivational study that was developed to select prebiotic drinks based on cashew nuts added to grape juice. In this application, the beta regression mixed model results corroborated with the selected formulations using the proportional mixed model.