Examining market segments and influencers of choice for wine using the Best-Worst choice method

Examining market segments and influencers of choice for wine using the Best-Worst choice method Dr Steve Goodman Adelaide Graduate School of Business The University of Adelaide South Australia 5006 Telephone: (61 8) 8303 4650 Facsimile: (61 8) 8223 4782 Email: steve.goodman@adelaide.edu.au Professor Larry Lockshin Professor of Wine Marketing School of Marketing, University of South Australia GPO Box 2471 Adelaide, South Australia 5001 Telephone: (61 8) 8302 0623 Facsimile: (61 8) 8302 0042 Email: larry.lockshin@unisa.edu.au Dr Eli Cohen Senior Lecturer Guilford Glazer School of Business and Management Department of Hotel and Tourism Management Ben-Gurion University of the Negev P.O. Box 653, Beer Sheva 84105, Israel Telephone: (972 8) 6472776 Facsimile: (972 8) 6472920 Email: elico@bgu.ac.ill Adjunct Senior Lecturer School of Marketing, University of South Australia GPO Box 2471 Adelaide, South Australia 5001 Email: eli.cohen@unisa.edu.au 1

Examining market segments and influencers of choice for wine using the Best-Worst choice method Abstract Wine marketers use market segmentation to target different products to different segments in order to increase sales, often with little evidence about what influences choice within or between segments. In this paper we provide initial results using a relatively new and very straightforward method for measuring consumer preferences: the best-worst scaling method. This paper shows how segmenting the consumers using factors such as gender, frequency of consumption, wine involvement and age produce segments with similar preferences for different varietal wines. Two country examples are used, Israel and Australia, to show the ability of the Best-Worst method to segment markets based on patterns of choice. Managerial implications of the importance of wine attributes that influence consumer drinks purchasing and wine style selection are discussed as well as suggestions for future research. Keywords: Best-Worst Scaling, segmentation, consumer choice of wine Examiner les déterminants du choix d un vin pour différents segments de marché en utilisant la méthode Best-Worst Résumé Dans cet article, nous présentons les premiers résultats d une étude mesurant les préférences du consommateur au moyen d une méthode relativement nouvelle et simple d application : la méthode best-worst. Cet article montre qu une segmentation du marché selon des critères tels que le genre, l âge, la fréquence de consommation, ou l implication dans le vin, met en lumière des préférences similaires, au sein de chacun de ces segments, quant aux différents vins proposés. Prenant l exemple de deux pays, l Israël et l Australie, nous montrons la capacité de la méthode best-worst à segmenter le marché selon l importance donnée aux différents attributs par les consommateurs. Les implications managériales de l importance qu ont les attributs à influencer l achat et la sélection des vins sont ensuite discutées ainsi que des suggestions pour de futurs travaux de recherche. Keywords: méthode best-worst, segmentation, choix du vin 2

Introduction Marketers in general and wine marketers in particular must constantly work to understand and forecast consumer product preferences. Wine is a unique product with a complex series of attributes, ranging from company brand, region or country of origin, grape variety, and price, to bottle shape, label design, and vintage date. Within each of these attributes there is typically more variation than in general packaged or fast moving consumer goods. For example, most supermarkets carrying wine would have at least 300 SKUs (stock keeping units) and some would have as many as 1000. There are 100,000s of wine brands in the global market along with several dozen main grape varieties and countries of origin. Prices can range from a few dollars per bottle to many hundreds of dollars, but even in a typical supermarket the price range for wine far exceeds the range for other product categories. Understanding consumer preferences for each of these attributes and the levels within them is a complex undertaking. There are many ways to measure consumer preferences. Most common are surveys with rankings or ratings and consumer panel data, which details individual purchases. Both of these methods have problems. Respondents to surveys do not use ratings or rankings the same way across respondents and the results are subject to a range of biases resulting in scores or ratings, which are too similar or too difficult to interpret (Cohen 2003; Cohen and Neira 2003; Finn and Louviere 1992). Consumer panel data provides powerful evidence of what consumers actually purchase, but is not suitable for testing new concepts or combinations of attributes. Consumer panel data shows what a consumer actually purchased, but may mask insight into their actual preferences; attributes or products that have bigger market share are more available for purchase and so are purchased more frequently. If five times more Chardonnay is available for sale than Sauvignon Blanc and so outsells it 5:1, does that mean consumers prefer Chardonnay, or are they just purchasing based on availability? From a strategic view, this is problematic as it gives a solid description of how things are, but is limited in providing cues for how things might be. Understanding what product features drive consumer choice is necessary for developing marketing and segmentation. Choice modelling provides a means to understand consumer preferences for product attributes and is much more predictive of actual marketplace choices than standard hedonic scaling (Lockshin and Hall 2003; Lockshin et al. 2006; Louviere et al. 2000). However, choice modelling confounds the scale and size of the utilities and therefore is not suitable for making comparisons among different data collections (Louviere et al. 2000). Finn and Louviere first published the Best-Worst method in 1992 and recently proved 3

the ability of the method to provide unbiased estimates across different data collections (Marley and Louviere 2005). Best-Worst scaling produces much less method variance than hedonic scaling and thus results in better separation among various alternatives. Goodman, Lockshin and Cohen introduced the method to the wine sector in 2005, which showed that it was easily applicable to measuring style preferences for wine in both Australia and Israel. This paper extends their work by utilizing the data from two countries and shows the ease of using this method to test for evidence of segmentation. We use two demographic and two psychographic measures to illustrate the method. The four variables were all significant in previous examinations of consumer choice for wine (Lockshin et al 2006; Perrouty et al 2005). Review of Literature Much of the literature on attribute importance in wine marketing is based on surveys, where consumers respond to questions on the importance of various intrinsic and extrinsic attributes. However, unless one alternative or attribute clearly dominates, it is difficult to identify the most important attribute or most preferred product. Treating the category ratings as equal interval scales may generate different conclusions than if they are treated as ordinal scales (e.g., relying upon median or 'top box' scores). Often the differences may be statistically significant, but it is difficult to assess whether a rating of 5.6 out of 7 is meaningfully different from 5.1 out of 7. What weighting scheme to apply to category ratings, or whether to rely on the alternative with the highest top box ratings, is a well-recognized problem in the case of purchase intention scales (see Morrison 1979, Jamieson and Bass 1989). Another issue is that each attribute is frequently measured with a single item rating scale newly developed just for the survey, so the reliability and validity of the scale is unknown. Attributes are usually not measured relative to other attributes or even products which must compete for the same (necessarily limited) consumer resources. Even if they are, respondents often are not allowed to indicate that they like many (if not all!) of them. Although some individuals truly might like nearly every attribute or combination, such responses don't provide adequate discrimination to help managers identify real priorities (Finn and Louviere 1992). As noted in the introduction, wine provides a complex set of products for the marketer to analyse. Hall and Lockshin (2000) found research on the following attributes used in wine buying behaviour, each with multiple articles: taste, type, alcohol content, age (of wine), color, price, brand, label/package, practical (usability for purpose), and region. Lockshin and Hall (2003) recently reviewed over 75 articles concerning consumer behaviour for wine. They 4

noted that many of the studies used simple surveys with rating scales to measure consumer preference for various wine attributes. Although there was much conflicting order in the rankings of the attributes for importance, previously having tasted the wine, the price, the origin, the grape variety, and the brand name of the wine were all mentioned frequently. The authors concluded that the best means to advance understanding of which attributes and combinations led consumers to purchase a particular wine was to use either choice-based experiments or analysis of actual consumer purchases. They also discussed the strengths and weaknesses of both approaches. Discrete choice modelling (Louviere and Woodworth 1983; Louviere, Hensher and Swait 2000) allows the measurement of utility (part worths) of attributes in various combinations, called product concepts. These part worths are calculated from the choices made and therefore, discrete choice is an indirect method of measuring utility or preference (Louviere and Islam 2004). The cost is similar to most other market research survey methods and it can yield useful information with relatively small sample sizes of 100-300 consumers, depending on the number of attributes and levels tested. It allows new attributes and combinations to be made and tested for preference. Since a survey is used, either on paper or online, other consumer characteristics can also be collected and used in the analysis. One of the problems of discrete choice when used for the wine industry, or any small sector, is that the design and analysis are complex and use sophisticated and often expensive computer programs. These are mostly provided by specialist researchers, which can also increase the cost. Another, and perhaps more serious, limitation to discrete choice models is the difficulty of interpreting the data including the inability to compare utilities across different experiments (Louviere, Hensher and Swait 2000). The Best-Worst (BW) approach, also known as Maximum Difference Scaling, was developed by Louviere and Woodworth (1990) and first published in 1992 (Finn and Louviere 1992). Recently, Cohen and Markowitz (2002) discussed the Maximum Difference Scaling (Max:Diffs) method and presented the advantages of the method. The Best-Worst approach assumes that there is some underlying subjective dimension, such as degree of importance or degree of interest and the researcher wishes to measure the location of some set of objects along this dimension (Auger, Devinney and Louviere 2004). The respondents are provided choice sets and choose the best/most important and the worst/least important from each set (an example of a choice set is presented in Appendix 1). There is no bias in the rating scale, since there is only one option to choose something that is most or least important (Cohen and Markowitz 2002). BW models the cognitive process by which respondents identify the two items with, respectively, the most and the least of a 5

characteristic, from designed sub-sets of three or more items. Technically, BW models the process of picking the two items that are the farthest apart on the underlying dimension of scaling interest (hence, maximum difference scaling ). BW produces an ordinal ranking of the items for each respondent, and an interval scale of the items based on sample or segment aggregate response (Louviere and Woodworth 1990). The method allows participants to gauge importance by multiple comparisons and they can dislike something as well as like something. The several studies cited here (Auger, Devinney and Louviere 2004; Cohen and Markowitz 2002; Finn and Louviere 1992) and our experience show that consumers find the task relatively easy and quick to complete. It does take more cognitive effort that standard hedonic scaling and because of that yields more realistic results. The major advantage to the researcher is the simplicity of the analysis, which yields a coefficient for each choice, whether it is a brand or attribute. The coefficients can be directly compared, which is not true for standard rating or ranking tasks or discrete choice analysis. The key issue for implementation is to design a series of choice sets that include all the items of interest and all possible comparisons an equal number of times for each respondent (Louviere and Woodworth 1990). Typically, experimental design software is used to create a balanced design. Auger, Devinney and Louviere (2004) and recently Marly and Louviere (2005) state that the total choices over all subsets of the implied pairs will be consistent with the multinomial logit model (MNL). An approximation of the model is achieved by calculating the differences of the total best and total worst counts for each item. Thus, as long as the experimental design is balanced, simply adding the number of times an item is chosen as worst and subtracting that from the total number of times it is chosen as best provides a scale that is about 95% as accurate as using multinomial logit to model the same data (Auger, Devinney and Louviere 2004). Method We collected data in Israel and Australia. The Australian data were collected among participants in several wine seminars in Adelaide and Perth during 2004. There were 145 valid responses from Australia. The other set was collected in Israel in 2005 where there were 130 valid questionnaires. In both studies, we presented 11 choices in 12 different choice sets, using the design developed by Finn and Louviere (1992), with further details discussed in Goodman, Lockshin and Cohen (2005). In the Australian data collection, respondents were presented with a BW selection set, consisting of 11 choices of 9 different wine varieties and 2 attributes of particular region and well known brand'. The Israel data consisted of 11 wine varietals. The items differed in each country due to the differing samples; wine aware participants Australia and less wine aware in Israel. However, the number of wineries in Israel 6

is growing and wine consumers are becoming more sophisticated and seeking better wines. In both studies respondents were asked to choose which varietal they most preferred and which one they least preferred from each choice set. The two sets of data differ fundamentally in the demographic breakdowns of respondents (Table 1), so true country comparisons are not possible, but the fact that segments can be determined even within these disparate groups shows the usefulness of the technique. For further research to better compare and map profiles it will be necessary to develop strict protocols involving quotas on income, gender, wine involvement, age and frequency if indeed these are the demographics of interest. The BW average score (level of importance) for each attribute (wine in our study) is calculated by subtracting the frequency of worst/least from the frequency of best/most of the same attribute, for each interviewee over all choice sets. The overall level of importance for an attribute (wine) for all respondents was determined by summing the BW scores of each wine for all respondents. Since the level of importance of each attribute depends on the number of respondents, the mean score was calculated by dividing the total Best Worst of an attribute by the number of subjects. The number of times each item was shown was equal in the two surveys. The level of importance (Best-Worst or BW score) of each attribute was used for the statistical analysis (general linear model) to test the significance of differences among the wine varieties. For the involvement segmentation, involvement was based on 3 questions describing the respondent's behavior concerning wine (Lockshin et al. 1997) as follows: 1). I have a strong interest in wine; 2). Wine is important to me in my lifestyle, and 3). Drinking wine gives me pleasure. Respondents were classified in two categories: "high involved" and "low involved". The sum of scores of 3 questions defined the classes where a score above 10 (the median value of the sum of scores) was classified as "high involved" and 10 or below was classified as "low involved". Results The results presented here are from initial studies undertaken to investigate and demonstrate the method and its application within the wine industry, especially to see if it would be possible to segment consumers based on the influences on choice using this approach. As discussed above, the Australian data is skewed towards highly involved wine consumers and generalizations at this stage are premature as a larger sampling of low involvement wine consumers needs to be included in the data set to give a more representative sample. Also, most of the Australian subjects (103 out of 145) were working full time and 103 subjects income was more than the average national income. In the Israeli 7

data 103 out of 130 subjects are working only part time and 99 subjects income were below the average national income. Table 1 shows the response by category in both markets and highlights the need in future collaboration to establish quotas for data collection. Table 1 - Response by Category Survey Location Israel Australia Gender Age Income Employment # of respondents 130 145 Male 72 85 Female 58 60 18-24 20 8 25-40 91 56 41-55 15 58 >55 4 23 Below average 99 23 About average 24 14 Above average 7 103 Full time working 19 103 Part-time working 103 19 Self-employed 7 15 Unemployed/ retired 1 7 Israeli Wine Varietal Preference Before applying segmentation to the data, it is necessary to examine each market as a whole to see the resultant rankings of what variety is preferred. In general, the Israeli people prefer Cabernet Sauvignon as the first choice followed by Merlot. However, there is no significant difference between these two varieties. Using analysis of variance and Tukey s post hoc test across all possible pairs of group means, we can see the statistically similar groups that emerge within the data (Table 2). The next variety preferred by Israeli consumers is Emerald Riesling (a very popular variety in Israel) followed by Chardonnay, Sauvignon Blanc and Muscat (usually slightly sweet). Shiraz is a relatively new variety in Israel and is not popular as a single variety wine in the market. Rosé, red and white house wines and sparkling wine are not popular in Israel and therefore they are less preferred. We note that the positive score means that for a given attribute best was chosen more frequent than worst and the negative score means the opposite, i.e., the distance between the coefficients provides the level of importance. General linear model and profile analysis throughout the segmentation approach was applied to see what different groups emerge as various segmentation schemes are trialled and compared back to the full data set. 8

Table 2 - Variety preference of wine, Israel (n=130) Variety Best-Worst Level of importance Comparison of means* frequency mean score Cabernet Sauvignon 269 2.1 X Merlot 223 1.7 X X Emerald Riesling 130 1.0 X X Chardonnay 107 0.8 X X X Sauvignon Blanc 73 0.6 X X Muscat 36 0.3 X X Shiraz 4 0.0 X Rosé -153-1.2 X Red house wine -174-1.3 X X White house wine -240-1.8 X X Sparkling wine -275-2.1 X * Tukey p<0.05 Gender The wine preferences profiles for males and females are presented in Figure 1. We can see that males prefer Cabernet Sauvignon and females prefer Merlot; males prefer Chardonnay and Emerald Riesling over Sauvignon Blanc, whilst females prefer Emerald Riesling over Sauvignon Blanc and Chardonnay but prefer Sauvignon Blanc to Chardonnay. However, there are no significant differences between males and females for most wine varieties, except for Shiraz. It seems that males prefer dry wines while females prefer white semi-dry wines (see Figure 1). Whilst this can provide a simple segmentation scheme we do not know if any of this ranking is of any statistical significance and must take a further step to do so. Figure 1: Relative importance of wine varieties for Israeli consumers gender segments Relative importance (marginal means) 4.000 2.000 0.000-2.000-4.000 Gender - Israel 1 2 3 4 5 6 7 8 9 10 11 Wine variety Wine varieties 1 White house wine 2 Sparkling wine 3 Rosé 4 Red house wine 5 Cabernet Sauvignon 6 Shiraz 7 Merlot 8 Chardonnay 9 Sauvignon Blanc 10 Emerald Riesling 11 Muscat Female Male 9

Consumption Frequency When segmenting by consumption frequency, there is very little difference in the ranking order (see Table 3). There are only several differences: high frequency drinkers (more than once a week) rank Merlot as the first choice and the low frequency drinkers rank the Cabernet Sauvignon as the first choice; lower frequency drinkers rank Muscat as the fifth most preferred compared with higher frequency rating it at number seven. There are no significant differences in preferring the wine varieties between the two segments except for Shiraz (p<0.001), where the high frequency wine drinkers prefer Shiraz wine. Since only a limited amount of Shiraz variety is in the Israeli market, the low frequency wine drinkers seem to drink the more familiar wines that available in the market. White house wine is less preferred by the high frequency wine drinkers (significant at p<0.05). Table 3 - Variety preference of wine for low and high frequency wine drinkers in Israel Low frequency wine drinkers (less than once week, n=87) High frequency wine drinkers (more than once a week, n=43) Level of Level of Variety importance Variety mean score importance mean score Cabernet Sauvignon 1.8 Merlot 2.7 Merlot 1.6 Cabernet Sauvignon 1.9 Emerald Riesling 1.1 Emerald Riesling 0.9 Chardonnay 0.9 Chardonnay 0.7 Muscat 0.6 Sauvignon Blanc 0.6 Sauvignon Blanc 0.6 Shiraz 0.4 Shiraz -0.2 Muscat -0.3 Rosé -1.3 Rosé -1.0 Red house wine -1.4 Red house wine -1.3 White house wine -1.5 Sparkling wine -2.0 Sparkling wine -2.2 White house wine -2.5 Involvement Involvement, one of the most used methods of separating wine drinkers preferences (Lockshin et al 1997; Lockshin et al 2001; Lockshin et al 2006; Quester and Smart 1998; Zaichowsky 1985) shows some differences in the Israeli wine market. We can see different patterns of preferred groups of varietals beginning to emerge (Table 4). Low involved wine drinkers have a top preference group that includes Cabernet Sauvignon, Merlot and Emerald Riesling, and the same grouping as the high frequency segment. High involvement consumers have Merlot as the sole most preferred wine varietal, significantly higher than the following varieties Cabernet Sauvignon and Chardonnay followed by Emerald Riesling, which the Israeli consumer has drunk for many years. High involved consumers tend to seek for new varieties and tastes and therefore they prefer Chardonnay as the first choice of white 10

wine. Rosé, red house wine, White house wine and sparkling wine are the least favourable for both segments, low and high involved consumers. Table 4 - Variety preference of wine for low and high wine involved consumers in Israel Low involved wine consumers (n=65) High involved wine consumers (n=65 Level of Level of Variety importance Variety mean score importance mean score Cabernet Sauvignon 1.9 Merlot 2.4 Merlot 1.7 Cabernet Sauvignon 1.6 Emerald Riesling 1.3 Chardonnay 0.9 Chardonnay 0.8 Emerald Riesling 0.7 Sauvignon Blanc 0.5 Sauvignon Blanc 0.6 Muscat 0.4 Shiraz 0.2 Shiraz -0.1 Muscat 0.2 Rosé -1.2 Rosé -1.1 Red house wine -1.5 Red house wine -1.2 White house wine -1.8 White house wine -1.9 Sparkling wine -1.8 Sparkling wine -2.4 As Israel is a developing wine market with a high number of lower knowledge consumers, a previous study (Goodman, Lockshin and Cohen 2005) shows that recommendation is the most important factor driving choice for most Israeli consumers. This is not surprising and the wine profiles for age groups, income categories and education look almost the same as for the whole sample of consumers in Israel. However, the Israeli wine consumers are becoming more sophisticated and seeking better wines. Australian Wine Varietal Preference The Australian choice set replaced 2 of the varietals with the choices of brand and region. Whilst this may not be included in replications, it was done so to enable an examination of fit with previous literature that shows the importance in the Australian setting of region and brand (see Hall and Lockshin 2000; Goodman, Lockshin and Cohen 2005 for more discussion). In line with market share, Shiraz was the most preferred, followed by Cabernet Sauvignon. Supporting other research (Hall and Lockshin 2000) was the fact that wine from a premium region was the third most important attribute when choosing varietal wine (Table 5). Interestingly, Sauvignon Blanc is preferred more than Chardonnay, which is in contrast to the market shares of these varieties, but in line with the growth rates of sales in Australia. 11

Table 5 Variety Preference of Wine - Australia (n=145) Variety Best-Worst frequency Level of importance Comparison of means* mean score Shiraz 255 1.8 X Cabernet Sauvignon 243 1.7 X X Wine from a premium Region 233 1.6 X X Cabernet/Merlot 178 1.2 X X X Wine from a well known brand 172 1.2 X X X Sauvignon Blanc 129 0.9 X X Chardonnay 71 0.5 X White sparkling wine -235-1.6 X Rosé -263-1.8 X Red house wine -297-2.0 X White house wine -486-3.4 X * Tukey p<0.05 Gender A quick examination of the Australian data shows that mapping by gender may show quite different preferences in several aspects as shown in the wine profiles for male and female (Figure 2). The most preferred attribute for male respondents is in line with market share (Shiraz and Cabernet Sauvignon) followed by wine from a premium region and wine from a well known brand, whilst for females in this sample wine from a premium region is the most important attribute when choosing wine, followed by Cabernet/Merlot and Sauvignon Blanc, which are in stark contrast to their market share. At first glance this appears to support females being a segment to target for niche markets as they appear to have preferences for attributes that are not big market shares. Statistical analysis shows that males have a specific preference for Shiraz and Cabernet Sauvignon, while females have a much broader range of grape varieties and wine attributes with no significant differences between them. This may partially be due to the relatively small sample size in this subgroup or it may indicate quite different choice processes between men and women. Statistical analysis shows significant differences (p<0.01) between male and female preferences for Shiraz, Cabernet Sauvignon, wine from a well known brand, rosé and white house wine. Males prefer Shiraz, Cabernet Sauvignon and wine from a well known brand while females prefer rosé and white house wine. Wine from a premium region is preferred by males compare to females although it is not statistically significant. Females prefer white sparkling wine compared to males p<0.05) and males choose red house wine more than females (p<0.05). 12

Figure 2: Relative importance of wine varieties for Australian consumers gender segments Australia - Gender Wine varieties Relative importance (marginal means) 4.00 2.00 0.00-2.00 1 2 3 4 5 6 7 8 9 10 11 1 White house wine 2 Sparkling wine 3 Rosé 4 Red house wine 5 Cabernet Sauvignon 6 Shiraz 7 Cabernet/Merlot 8 Chardonnay 9 Sauvignon Blanc 10 Wine from a well known brand 11 Wine from a premium Region -4.00 Wine variety Male Female Frequency of Consumption There are some key differences in ranking order between high and low frequency wine consumption groups (Table 6) in this initial data. High frequency wine drinkers are most influenced by premium region and prefer Cabernet Sauvignon, Shiraz, brand and Sauvignon Blanc over Cabernet/Merlot. The lower score for Cabernet/Merlot in this group is quite distinct from the ranking of the whole sample. The other key difference between the two rankings is that high frequency wine drinkers prefer Sauvignon Blanc to Chardonnay, against the market shares, whereas low frequency wine drinkers show almost the same preference for Chardonnay and Sauvignon Blanc (see Figure 3). Low frequency wine drinkers like Shiraz, as the preference is included in their top choice, whereas high frequency drinkers place this in their second tier. High frequency wine drinkers focus on wine from a premium region followed by top varieties, Shiraz and Cabernet Sauvignon and then on wine from a well known brand. The lower frequency drinkers have these in their preferences, but in a lower rank. This may be due to the smaller sample size of this group or it may indicate a true difference in how they choose wines. However, the only significant difference found between the two segments was for rosé wine, where the low frequency wine drinkers have higher preference compared to the high frequency wine drinkers (p<0.05, Figure 3). 13

Table 6 - Variety preference of wine for low and high frequency wine drinkers in Australia Low frequency wine drinkers (less than once week, n=43) High frequency wine drinkers (more than once a week, n=101) Level of Level of Variety importance Variety mean score importance mean score Shiraz 1.9 Wine from a premium Region 1.8 Cabernet/Merlot 1.6 Shiraz 1.7 Cabernet Sauvignon 1.6 Cabernet Sauvignon 1.7 Wine from a premium Region 1.1 Wine from a well known brand 1.4 Wine from a well known brand 0.7 Sauvignon Blanc 1.1 Sauvignon Blanc 0.4 Cabernet/Merlot 1.1 Chardonnay 0.4 Chardonnay 0.6 Rosé -1.0 White sparkling wine -1.6 White sparkling wine -1.7 Red house wine -2.2 Red house wine -1.9 Rosé -2.2 White house wine -3.0 White house wine -3.5 Figure 3: Relative importance of wine varieties for Australian consumers frequency of drinking wine (low frequency once a week or less, high frequency more than once a week) Relative im portance (m arginal m eans) 2.000 0.000-2.000-4.000 Australia - Frequency of drinking wine 1 2 3 4 5 6 7 8 9 10 11 wine variety Wine varieties 1 White house wine 2 Sparkling wine 3 Rosé 4 Red house wine 5 Cabernet Sauvignon 6 Shiraz 7 Cabernet/Merlot 8 Chardonnay 9 Sauvignon Blanc Wine from a well known 10 brand Wine from a premium 11 Region Low freq drink High freq drink Involvement The involvement segment gives fewer differences when comparing the rank order to the order of the sample. There are only minor differences in wine variety profiles (Figure 4); high involved consumers have Cabernet Sauvignon above Shiraz and low involved wine consumers have Shiraz almost the same level as wine from a premium region, both above Cabernet Sauvignon. Finally, low involved wine consumers prefer Sauvignon Blanc to Chardonnay, the opposite to high involved wine consumers. Both segments have Cabernet Sauvignon, Shiraz wine from a premium region, wine from a well known brand and 14

Sauvignon Blanc included in their top group of influencers. High involved consumers significantly (p<0.05) prefer Chardonnay compared to low involved consumers. Interestingly wine involvement shows less distinguishing features than the use of consumption frequency. Again, this may be due to the lopsided sample having a high proportion of high involvement buyers. Figure 4: Relative importance of wine varieties for Australian wine involved consumer Relative importance (marginal means) 4.00 2.00 0.00-2.00-4.00 Australia - wine involvement consumers 1 2 3 4 5 6 7 8 9 10 11 Wine variety LOW involved HIGH involved Wine varieties 1 White house wine 2 Sparkling wine 3 Rosé 4 Red house wine 5 Cabernet Sauvignon 6 Shiraz 7 Cabernet/Merlot 8 Chardonnay 9 Sauvignon Blanc Wine from a well known 10 brand Wine from a premium 11 Region Age The data collected in Australia enables a segmentation using the age variable. The wine variety profiles for the four age groups are presented in Figure 5. In general, rankings of each segment are somewhat similar to the overall sample with minor variations. However, although there are only a limited number of young consumers in this study (age 18-24, n=8), it is possible to see their preferences compared to the older age groups. Consumers of age 25-55 years prefer Shiraz as their first choice followed by Cabernet Sauvignon and wine from a premium region. Young consumers (18-24 years old) prefer Cabernet/Merlot as their first choice, significantly higher than Chardonnay, Sauvignon Blanc, wine from a well known brand, and wine from a premium region (see Figure 5). A possible explanation is that young wine consumers are less sophisticated and seeking for drinking pleasant wine provided by the blend of Cabernet/Merlot. Young consumers prefer rosé compared to the other age groups. A significant difference (p<0.05) was observed only for Shiraz. Consumers of age 25-55 years have significantly higher preference for Shiraz compared to young consumers (18-24) and adults of age over 55 years. 15

Figure 5: Relative importance of wine varieties for Australian consumers age groups Relative importance (marginal means) 4.000 2.000 0.000-2.000-4.000 Australia - Age groups 1 2 3 4 5 6 7 8 9 10 11 Wine variety 18-24 25-40 41-55 >55 Wine varieties 1 White house wine 2 Sparkling wine 3 Rosé 4 Red house wine 5 Cabernet Sauvignon 6 Shiraz 7 Cabernet/Merlot 8 Chardonnay 9 Sauvignon Blanc Wine from a well known 10 brand Wine from a premium 11 Region Limitations and Further Research This paper has presented the results from initial data collected to showcase the ease of making comparisons and segments using Best-Worst scaling. As such it is not in a position to make recommendations for managers. There are not sufficient data to make generalizations, nor are the two samples directly comparable. Future research will need to develop choice sets that can be used in identical designs across markets. It has shown the importance of conducting further research, with rigorous data collection protocols, and strict replication across markets in order to map the segments that emerge with different preferences to the market. As in so many industries talk has been focused on anecdotal observations, which are unverifiable as to true or false with regards to market segments and the resultant steps to successfully target them. A concerted effort in collecting sufficient quantity of data across a number of markets will enable the data to drive the identification of segments and enable researchers and practitioners to develop strategies to target them based on what we know empirically about the behaviour and demand of each segment. More data in each market would enable the analysis to be undertaken by multiple segmentation steps; looking at gender, age and income or looking at consumption frequency, income and age to see what segments emerge from the data that are empirically verifiable, are actually capable of being targeted, and offer some benefit to the research and managerial communities alike. The Australian data is more homogeneous than the Israeli sample; this has future research implications and may have managerial impact once that research has been conducted. It might be that homogeneity is indicative of the degree of development of the market. Data is 16

needed from Old World, the New World, established, emerging, and non-established wine markets in order to see if this holds true. This in turn may have implications for the type of marketing required, whether the message given is focused or broad, brand-based or categorybased. Although statistical analyses show differences between the segments examined, they only show what choice criteria are being used in the market by the different segments. As such it is suggested that for researchers and managers, a sequence be followed of examining rankings, and then the statistical differences between the segments. They are more likely to be a holistic picture, or map, than a prescriptive direction for target marketing. Further research would use Best-Worst scaling with whole product concepts, like in discrete choice analysis, to test how combinations of attributes (product concepts) are chosen by each segment. The use of Best- Worst for product concepts allows the calculation of comparable part worths within each product concept, something discrete choice cannot do. Conclusion This paper has sought to show the benefits from using consumer choice methodology, in particular the Best-Worse method, to understand consumer choice for wine. The approach offers much in terms of segmentation analysis and cross cultural research. To that end, there has not been lengthy discussion of various literatures. Previous research by others has discussed the theoretical underpinning of research into consumer attributes for wine choice and possible segmentation variables (see Lockshin and Hall 2003 for a review). This paper presented results from initial surveys into key segmentation variables to show how we intend to move research forward. It has provided a process for looking at the data to show how three steps can be used in a holistic manner in the analysis of BWchoice data. This will be the approach used to gather more data, across markets and begin to map the consumer influence models of the emergent segments. From the results presented here and the method outlined, there is much to be gained with researchers in various markets adopting the method and driving research in the area of wine marketing, from consumer understanding to distribution chain decision making. Furthermore, it is an exciting method that offers both academic research rigour and practitioner understanding and insight, as such can assist in bridging the divide between academia and practice without the cultural biases and problems of traditional rating and ranking methods, key in the global arena that is wine marketing. 17

References Auger, P., Devinney, T.M. and Louviere, J.J. (2004). Consumer s social beliefs, an international investigation using Best-Worst scaling methodology. Working paper, University of Melbourne, Melbourne Business School, Melbourne, Victoria, AU. Cohen, S. H. and Markowitz, P. (2002), Renewing Market Segmentation: Some New Tools to Correct Old Problems, ESOMAR (September). Finn, A. and Louviere, J.J. (1992). Determining the Appropriate Response to Evidence of Public Concerns: the Case of Food Safety, Journal of Public Policy and Marketing, 11(1), 12-25. Goodman, S. Lockshin, L. and Cohen, E. (2005). Best-Worse Scaling: A Simple Method to Determine Drinks and Wine Style Preference, 2nd international Wine Marketing Symposium Proceedings, (CD-ROM), Sonoma CA., July 8-9. Hall, J. and Lockshin, L. (2000). Using Means-End Chains for Analyzing Occasions Not Buyers, Australasian Marketing Journal, Vol. 8 (1), 45-54. Jamieson, L. F. and Bass, F.M. (1989). "Adjusting Stated Intention Measures to Predict Trial Purchase of New Products: A Comparison of Models and Methods," Journal of Marketing Research, 26 (August), 336-345. Lockshin, L. and Hall, J. (2003). Consumer Purchasing Behaviour for Wine: What We Know and Where We are Going, International Wine Marketing Colloquium, Adelaide, July, CD- ROM. Lockshin, L., Jarvis, W., D Hauteville, F. and Perrouty, J. (2006). Using simulations from discrete choice experiments to measure consumer sensitivity to brand, region, price, and awards in wine choice, Food Quality and Preference, 17, 166-178. Lockshin, L., Quester, P. and Spawton, T. (2001). Segmentation by Involvement or Nationality for Global Retailing: A Cross National Comparative Study of Wine Shopping Behaviours, Journal of Wine Research, 12 (3), 223-236. Lockshin, L., Spawton, A.L. and Macintosh, G. (1997). Using Product, Brand, and Purchasing Involvement for Retail Segmentation, Journal of Retailing and Consumer Services, 4 (3), 171-183 Louviere, J.J., Hensher, D.A. and Swait, J.D. (2000). Stated Choice Methods: Analysis and Application. Cambridge, UK: Cambridge University Press. Louviere, J.J. and Islam, T. (2004), A Comparison of Importance Weights/ Measures Derived from Choice-Based Conjoint, Constant Sum Scales and Best-Worst Scaling, CenSoC Working Paper No. 04-003, (September). Louviere J.J. and Woodworth G.G. (1983). "Design and Analysis of Simulated Consumer Choice Allocation Experiments: an Approach Based on Aggregate data," Journal of Marketing Research, Vol. 20, 350-367. Louviere, J.J. and Woodworth, G.G. (1990). Best-Worst Scaling: A Model for Largest Difference Judgments, working paper, Faculty of Business, University of Alberta. Marley, A.A.J. and Louviere, J.J. (2005). Some Probabilistic Models of Best, Worst, and Best-Worst Choices. Journal of Mathematical Psychology 49,464-480. 18

Morrison, D. G. 1979. "Purchase Intentions and Purchase Behavior," Journal of Marketing, 43 (Spring), 65-74. Quester, P., and Smart, J.G., (1998). The influence of consumption situation and product involvement over consumers' use of product attributes, Journal of Consumer Marketing, 15 (3), 220-238. Perrouty, J. P., d Hauteville, F. and Lockshin, L. (2005). The Influence of Wine Attributes on Region of Origin Equity: An Analysis of the Moderating Effect of Consumer s Perceived Expertise of Wine at the Visitor Centre [CD-ROM] presented at Second Annual International Wine Marketing Symposium, Sonoma State University, California, 8-9 Jul. Zaichkowsky, J. L (1985), Measuring the Involvement Construct, Journal of Consumer Research, 12 (December), 341-352. Appendix 1: Sample of a choice set for Best-Worst questionnaire In the following tables, please identify the MOST wine variety and the LEAST wine variety when you are choosing wine. Check ONLY ONE issue for each of the most and least columns, in each table. Each table will have one item ticked for the MOST preferred and one item for the LEAST preferred. Least/Worst Grape variety Most/Best Rosé X X Chardonnay Shiraz Merlot 19