Classification trees in consumer studies for combining both product attributes and consumer preferences with additional consumer characteristics Rosaria Romano a,⇑ , Cristina Davino b , Tormod Næs c,d a University of Calabria, Department of Economics, Statistics and Finance, 87036 Arcavacata di Rende, Cosenza, Italy b University of Macerata, Department of Political Sciences, Communications and International Relations, Piazza Oberdan 2, 62100 Macerata, Italy c Nofima Mat AS, Osloveien 1, NO-1430 Ås, Norway d University of Copenhagen, Department of Food Science, Rolighedsvej 30, DK-1958 Frederiksberg, Denmark article info Article history: Received 24 June 2013 Received in revised form 7 October 2013 Accepted 17 November 2013 Available online 23 November 2013 Keywords: Classification trees Acceptance pattern Questionnaire data Validation abstract The main objective of this paper is to describe and discuss the use of classification trees in consumer studies. Focus will be given to the use of the method in relating segments of consumers, based on their acceptance pattern, to additional consumer characteristics, including attitudes, habits and demographics variables. Advantages of the method in handling typical issues from consumer studies will be discussed. Primary interest will be given to the validation of the results, which will also be compared with results from alternative methods widely used in consumer studies. The approach will then be illustrated by using data from a conjoint study of apple juice. Ó 2013 Elsevier Ltd. All rights reserved. 1. Introduction In experimental consumer studies, one of the primary aims is to obtain information about consumer preference or purchase intent for a number of products. One of the most used methodologies in this field is conjoint analysis (Green & Srinivasan, 1978; Gustafsson, Hermann, & Huber, 2003), which studies the effect of a number of product characteristics on consumer acceptance. In conjoint stud- ies, product information is organized into a number of factors, each combination giving rise to a trial product to be presented to con- sumers. Consumers then express their degree of liking (or another hedonic characteristic) for each combination, or alternatively their ranking of the products or their choice (Louviere, 1988; Louviere, Hensher, & Swait, 2000). The data are generally analyzed using Analysis of Variance (ANOVA) (Næs, Brockhoff, & Tomic, 2010; Searle, 1971) or via rank order logistic modeling (Train, 1986; McCullagh & Nedler, 1989). Due to the intrinsic heterogeneity in consumer acceptance pat- terns, it is extremely important to investigate not only the drivers of liking at the population level, but also to explore individual differ- ences among consumers (Gustafsson et al., 2003; Næs et al., 2010). In addition, it is very important for the purpose of planning appro- priate marketing strategies to relate the individual differences in acceptance pattern to other consumer characteristics, including attitudes, habits and demographics. (Benton, Greenfield, & Morgan, 1998; Wedel & Kamakura, 1998) or to other external information such as sensory data (McEwan, 1996; Schlich & McEwan, 1992; Vig- neau & Qannari, 2002). To achieve this, many approaches have been proposed. One possible strategy is to segment acceptance values using some type of cluster analysis and then relate the obtained seg- ments to the additional consumer variables by tabulation; another option might include regression analysis (Næs, Kubberod, & Silvert- sen, 2001) or discriminant analysis (Ripley, 1996). Other important possibilities are based on ANOVA with the incorporation of effects for additional consumer characteristics (Næs et al., 2010), combina- tions of ANOVA modeling and multivariate analysis (Endrizzi, Menichelli, Johansen, Olsen, & Næs, 2011) and combining all data- sets into one single multivariate analysis using the L-PLSR method (Martens et al., 2005; Vinzi, Guinot, & Squillacciotti, 2007). An alter- native procedure allowing for a simultaneous analysis of the three blocks of information (product hedonic scores, product sensory descriptors and consumer attributes) has been recently proposed by Vigneau, Charles, & Chen (2014). The different blocks of informa- tion may also be related to each other in different ways by using some type of structural equations modeling (Guinot, Latreille, & Tenenhaus, 2001; Olsen, Menichelli, Sørheim, & Næs, 2012; Tenen- haus, Pagès, Ambroisine, & Guinot, 2005). 0950-3293/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.foodqual.2013.11.006 ⇑ Corresponding author. Address: University of Calabria, Department of Econom- ics, Statistics and Finance, Cubo 1/C, 87036 Arcavacata di Rende, Cosenza, Italy. Tel.: +39 0984 492448; fax: +39 0984 492421. E-mail addresses: rosaria.romano@unical.it (R. Romano), cdavino@unimc.it (C. Davino), tormod.naes@nofima.no (T. Næs). Food Quality and Preference 33 (2014) 27–36 Contents lists available at ScienceDirect Food Quality and Preference journal homepage: www.elsevier.com/locate/foodqual