Food sensory analysis

a comparison of tests for potential tasters selection

Abstract

Sensory analysis involves a product sensory attributes evaluation through human senses. The efficient panels formation is essential for quality control, therefore it’s necessary to perform selection and training steps. In selection, triangular tests are used to verify the discriminative ability of candidates, subjecting to training only those with acute sensory skills. Thus, it’s wanted to find the best test (in terms of lowest type I error rate and higher power) for tasters selection. This work aims to make a study of interval estimators for proportion, rewriting them as tests for tasters selection. Six tests (based on Normal and F distributions, Sequential test and Poisson approximation using χ² quantile) were compared. Through a Monte Carlo simulation study 1000 virtual tasters were generated and submitted to triangular tests over 5, 10, 15, ..., 100 trials. It was established a constant probability hit of these testers over time, set at 0,33, 0,35, 0,40, 0,45, ..., 1,00, evaluating the test performance in different test numbers (n) and proportion of correct answers (p). It was noted that sequential test presented lower type I error rate and higher power for a small number of trials (n < 20), while a Normal-based test obtained this behavior for n value equal to or greater than 20 (n ≥ 20). Therefore, both tests could be used to select tasters according to the stipulated number of tests.