Combining extrinsic and intrinsic information in consumer acceptance studies Elena Menichelli a,b,⇑ , Nina Veflen Olsen a , Christine Meyer c , Tormod Næs a a Nofima Mat AS, Osloveien 1, NO-1430 Ås, Norway b Department of Chemistry, Biotechnology and Food Science, The Norwegian University of Life Sciences, P.O. Box 5003, NO-1432 Ås, Norway c Unil, P.O. Box 290 Skøyen, NO-0213 Oslo, Norway article info Article history: Received 5 October 2010 Received in revised form 14 March 2011 Accepted 14 March 2011 Available online 17 March 2011 Keywords: Conjoint Preference mapping Factorial design ANOVA Fuzzy clustering abstract This paper proposes a methodology for combining extrinsic and intrinsic attributes in consumer testing of food products. The paper attempts to focalize on the main sensory drivers of liking or choice probabil- ity and their interaction with additional information, and to investigate effects related to the population as well as how consumers differ in their assessments. Two different data analysis approaches are consid- ered and compared on choice probability data from a consumer study of orange juice. One of the methods is based on mixed model ANOVA of individual differences, the other approach is based on fuzzy clustering related to regression residuals. The main results show that extrinsic consumer attributes are easily and efficiently related to the sensory properties of products, allowing for interactions. The methodology esti- mates population or segment means and gives an overview of individual differences in choice intent or liking. Ó 2011 Elsevier Ltd. All rights reserved. 1. Introduction Both intrinsic sensory attributes and extrinsic factors related to all other aspects of the product and its presentation are important for consumer choice probability or liking of food products. For in- stance, when buying yoghurt, the sensory properties of the yo- ghurt, information about sugar and fat content (Johansen, Næs, Øyaas, & Hersleth, 2010b) as well as the packaging are all impor- tant for the choice. In product development it is therefore useful to investigate consumers’ acceptance in light of all these aspects. Very often the two types of attributes are investigated in indepen- dent tests, but in some cases this may be insufficient. If for instance the difference in consumer choice probability between two prod- ucts depends on information about health benefits, this type of information is not possible to get without using a test where both aspects are involved. This facet is particularly important to take into account in research when the purpose is to understand pat- terns in human perception and liking or choice probability, but it may also be highly relevant in concrete industrial product develop- ment situations. In such cases it is therefore crucial to have tech- niques available that can be used to investigate both intrinsic and extrinsic attributes simultaneously. A number of studies have been conducted where consumers are given food samples together with additional information (Johansen et al., 2010b; Stefani, Romano, & Cavicchi, 2006; Urala & Lähteenmäki, 2006; Visschers & Siegrist, 2009), but most of these studies consider a number of fixed samples and end up with draw- ing conclusions more related to differences between the actual products than to the important sensory drivers of liking or choice probability (Enneking, Neumann, & Henneberg, 2007; Helgesen, Solheim, & Næs, 1997). In such cases, one typically uses standard factorial designs and Analysis of Variance (ANOVA) treating each product as a separate level of one of the experimental factors. This is an important methodology which can give a lot of insight, but its main drawback is that there is no or at least rather limited focus on the effects of the whole profile of sensory attributes of the products and how it influences consumer preferences. In other words, lim- ited information is obtained about what the main drivers of liking or choice probability are and also about how these interact with the extrinsic attributes. This type of insight may be of crucial value when optimising product properties. A methodology for solving this type of problem was proposed in Johansen et al. (2010b). This approach is based on first analysing a number of relevant samples for testing with the use of sensory analysis. This ‘‘large’’ number of samples may be obtained by for instance experimental design as done in Johansen et al. (2010b), but it may also be obtained by selection from a production process or from a store as long as the samples are relevant. This number, which may typically vary around 10, may, however, be too large for consumer testing in combination with other attributes. There- fore a selection strategy was proposed based on the scores plot from the Principal Component Analysis (PCA) of the sensory data. More specifically, the samples were selected to span a rectangular shape in the principal components plot, with the rectangular axes 0950-3293/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.foodqual.2011.03.007 ⇑ Corresponding author at: Nofima Mat AS, Osloveien 1, NO-1430 Ås, Norway. Tel.: +47 64970100; fax: +47 64970333. E-mail address: elena.menichelli@nofima.no (E. Menichelli). Food Quality and Preference 23 (2012) 148–159 Contents lists available at ScienceDirect Food Quality and Preference journal homepage: www.elsevier.com/locate/foodqual