Risks and pitfalls of sensory data analysis for shelf life prediction: Data simulation applied to the case of coffee

Shelf life determination by means of sensory analysis is thought to be of paramount importance even in case of a microbiologically stable food. Several approaches are found in literature, both in terms of data collection and data processing. Whatever method is used, the subjectivity in the choice of some parameters for data collection and analysis can deeply influence the final result. We put in evidence some typical pitfalls that the researcher should avoid when planning the test and analysing data. A comparison between the most utilized techniques in sensory data processing for shelf life prediction is reported, taking as a fil rouge the case of coffee. In particular, a non-linear regression, a logistic regression and a survival models were applied to simulated data frames of coffee. We evaluated the influence of the choice of acceptability limits, as well as the effect of data variability and we found out that they strongly influence predictions, as well as the panel and the batch of product do. We suggest that in case of microbiologically stable food, like coffee, shelf life is not univocal and it is a choice of the company or the researcher, rather than the result of the interaction between product and consumer.