Simulation of robust adaptive regression multi-level models for quality analysis of special coffees in cold storage

Abstract

Numerous factors contribute to specialty coffee quality, storage and cooling conditions. We may therefore assume that sensory evaluation results can be corrupted by measurement errors, especially when cuppers are not trained, leading to occurrence of observation outliers. Therefore, this study aimed to propose simulation scenarios considering parametric values of multilevel model fit with robust adaptive regressions to the presence of outliers in a real experiment with processed and unprocessed coffee beans stored at different times and temperatures. In this context, we considered computationally simulated scenarios in which sensory scoring errors can be made at L = 5 and 10 units. The proposed method was feasible for the sensory scoring of an experiment of coffee storage conditions and cooled environments. This is because it included robust characteristics of samples evaluated with up to 30% of outliers.