A longitudinal data analysis approach for sensory shelf life estimation

Abstract Shelf life of a food product is a time threshold that determines when it no longer retains its quality. It can be estimated using data coming from sensory experiments where potential consumers taste aged portions of the product and state whether they will consume the product or not. The raw data obtained from the experiment consist of a binary sequence for each assessor. The standard approach to analyze this type of data is based on reliability theory and requires the coding of the raw data into censored intervals. This paper discusses how the aforementioned coding yields a loss of information and low coverage of confidence intervals. Furthermore, the paper introduces an alternative methodology to estimate shelf life based on longitudinal data theory. This methodology prevents loss of information, quantifies the effect of inconsistent responses, explicitly incorporates and diagnoses the correlation structure of the consumer’s responses, and provides an easier interpretation of the results for the final user. These features are shown using real data coming from sensory experiments.