Use of Mandel’s bundle of lines model to improve agreement of a panel of tea tasters

Abstract

Sensory evaluation is of prime importance in the tea industry. The price of tea is predominantly determined by the tea taster’s judgement. Hence, statisticians and sensory analysts have significant interest in improving the reliability of sensory panel evaluations. This study suggests a statistical de-biasing approach to improve the agreement of panels of tea tasters by using the Mandel’s bundle of lines model. This paper analyses data from a panel of eight experts who evaluated 72 black tea samples collected from 13 different geographical regions of Sri Lanka over a one year (January 2010 ‒ December 2010) period at monthly intervals. The results showed that in only 20 instances out of 72, the intra-class correlation coefficients were greater than 0.5, but after bias correction, 64 instances out of 72 have intra-class correlations greater than 0.5. Therefore, the Mandel’s bundle of lines approach improves consensus among the experts by enforcing better statistical calibration.