Taking individual scaling differences into account by analyzing profile data with the Mixed Assessor Model

• Suggests a new and improved ANOVA of sensory profile data.
• The method is based on a simple covariate inclusion idea.
• The scaling handling leads to a novel post hoc confidence band construction.
• The importance of this is documented by a meta study using the SensoBase.
• The method makes a link between mixed modeling and assessor performance analysis.

Scale range differences between individual assessors will often constitute a non-trivial part of the assessor-by-product interaction in sensory profile data (Brockhoff, 2003, 1998; Brockhoff and Skovgaard, 1994). We suggest a new mixed model ANOVA analysis approach, the Mixed Assessor Model (MAM) that properly takes this into account by a simple inclusion of the product averages as a covariate in the modeling and allowing the covariate regression coefficients to depend on the assessor. This gives a more powerful analysis by removing the scaling difference from the error term and proper confidence limits are deduced that include scaling difference in the error term to the proper extent. A meta study of 8619 sensory attributes from 369 sensory profile data sets from SensoBase (www.sensobase.fr) is conducted. In 45.3% of all attributes scaling heterogeneity is present (P-value <0.05). For the 33.9% of the attributes having a product difference P-value in an intermediate range by the traditional approach, the new approach resulted in a clearly more significant result for 42.3% of these cases. Overall, the new approach claimed significant product difference (P-value <0.05) for 66.1% of the attributes compared to the 60.3% of traditional approach. Still, the new, and non-symmetrical, confidence limits are more often wider than narrower compared to the classical ones: in 72.6% of all cases.