STATISTICAL-ANALYSIS RELATING ANALYTICAL AND CONSUMER PANEL ASSESSMENTS OF KIWIFRUIT FLAVOR COMPOUNDS IN A MODEL JUICE BASE

We analyse the relationship between analytical sensory data and consum er data from a model juice base with three specific kiwifruit flavour compounds added, using statistical methods including exploratory graph ical methods, correlations, partial least squares, additivity and vari ance stabilising transformations (A AVAS), and multiple regression. Pa rtial least squares and correlation analysis both showed that the anal ytical panel was able to distinguish only one independent dimension ou t of the three possible, while the consumer panel data was two dimensi onal. The consumer panel was able to independently discriminate betwee n levels of two of the flavour compounds. Consumer 'kiwifruit aroma' w as not associated with any linear combination of analytical attributes , Nonlinear models, with quadratic and simple interaction terms, were fitted. These models were able to explain tin terms of concentrations of flavour compounds) 80-85% of the variance in consumer or analytical attributes, and define a relationship between consumer and sensory an alytical data. By comparison partial least squares could Produce only one dimension accounting for 15-20% of the variance, The models enable predictions of consumer response from analytical panel response but w ith less accuracy than that obtained by modelling the consumer respons e directly in terms of the chemical concentrations. It is recommended that further analytical attributes corresponding to the consumer 'kiwi fruit aroma' attribute be developed, This should result in better pred ictions. (C) 1998 Elsevier Science Ltd. All rights reserved.