Identifying Panel Inconsistency in Sensory Profiles using Multivariate Analysis of Variance (MANOVA) and follow–up Canonical Variate Analysis (CVA)

Inconsistency in sensory evaluation is a serious problem and it often leads to loss of large revenue. Screening tasters before the sensory evaluation is the only remedy to overcome this inconsistency. This paper aims at showing how the above requirement can be achieved by using Multivariate Analysis of Variance (MANOVA) and follow–up Canonical Variate Analysis (CVA). The approach was illustrated using sensory profiles of Sri Lankan tea. The principle behind the approach is that any discrepancy between assessors on several attributes is detected simultaneously using MANOVA, and the discrepancy interacts with the factors such as products or region is detected using CVA. Data used for the study consisted of sensory scores given by 8 tea tasters for 13 tea growing regions on 6 attributes; colour, brightness, strength, flavour, aroma, quality. Samples from four factories represented a region, and data were collected for a one–year period on a monthly basis. Data from each month were analyzed separately. The Wilk’s Lambda statistics of MANOVA revealed assessor effect as well as assessor × region interaction effect (P < 0.05) in every month indicating the inconsistency among assessors. The CVA for each region, specifically the 95% confidence regions of CV bi–plots, clearly identified clusters of assessors. Based on the location of these clusters in bi–plots, assessors who are suitable for different attributes were also identified. MANOVA followed by CVA, can effectively be used to identify discrepancies between assessors, discrepancy interacts with factors such as geographical region, and selecting consistent assessors depending on product or region and season.