Again, don't feel you have to persevere. But it's one of my earlier papers.
PRESENTED AT THE INAUGURAL ACADEMY OF WINE BUSINESS RESEARCH COLLOQUIUM, ADELAIDE, 2003
Keywords: Wine Marketing, Consumer Concentration, Pareto Share, Market Modelling
Cullen Habel, Cam Rungie, Professor Larry Lockshin, Tony Spawton
Wine Marketing Research Group
School of Marketing, University of South Australia
North Terrace, Adelaide SA 5000, Australia
PDF available at:
Abstract
This paper considers two performance issues for several types of alcohol – category penetration and consumer concentration. Consumer concentration is addressed using the performance measure of “Pareto Share”, which is defined as the percentage of category sales to the top 20% of its consumers. The beverage categories of beer, wine and spirits are first compared for their observed 1-week time period. The categories are then modelled, using the Negative Binomial Distribution in order to extrapolate market behaviour to longer time periods of observation – in this case a month and a year. Findings of this study are that the Pareto effect varies considerably across alcohol types and that the apparent Pareto effect increases as the sample time increases. The implications for managers are discussed and areas of further research highlighted.
Introduction
Categories have heavy and light users. In the first empirical study of this phenomenon, Twedt examined four categories of “Chicago Tribune” panel data in 1968, and asked the question “How important to marketing strategy is the ‘heavy user’?” He found the “heavy half” accounted for more than 80% of the category purchases for the two categories he considered. A question of growing importance to alcohol beverage marketers must surely be “How important are our heavy users?” We may sagely nod at the mention of a “Rule of 80:20” where the top 20% of customers account for 80% of our sales volume but do we really know anything about it? Nagging questions include:
¨ Is there 80:20 rule for wine and other alcohol beverage types, or is it simply a “rule of thumb”?
¨ What percentage of a category’s consumption is due to the consumption of the heavy user? (How does consumption data present this?)
Other questions also come to mind once we begin analysing Pareto type distributions:
¨ Does the “rule” apply to each alcohol beverage category in the same manner?
¨ Does the time period of sampling matter? Is the Pareto Effect for a week’s sample of consumer behaviour the same as a month’s or a year’s sample?
A search of the literature shows a few articles that have quantitatively addressed this question of buyer concentration ((Twedt 1968),(Schmittlein, Cooper and Morrison 1993), (Anschuetz 1997), (Rungie, Laurent and Habel 2002)) and none at all within the context of wine. Stanford (2000) has shown that wine consumers consumer multiple alcohols but does not differentiate the heavy/ light consumption concentrations.
The nature of this paper is not to describe the characteristics of heavy users. Even in the nascent field of wine consumption behaviour significant research has been conducted into that issue. Examples include (Keown and Casey 1995), (Blaylock and Blisard 1993), (Goldsmith and d'Hauteville 1998). These articles have all considered heavy users but in the sense of how heavy usage correlates to demographic, geographic and, interestingly, psychographic variables. This valuable in itself, but it differs from our research.
Our objective is more general. Many practitioners only have anecdotal evidence as to how many heavy users their categories have. Spirits marketers may think that a large amount of their volume is consumed by a relative few of their customers but have they ever demonstrated this? Wine marketers may like to think that their product has a broad appeal, so heavier users may account for less of the total consumption of wine, but how do they know?
Firstly, we aim remove some of the “hearsay” around this light/heavy user argument. Using real consumption data and modern modelling techniques, we answer the “what” question more than the “why”. What is the Pareto Effect for Wine as compared to that of Beer or Spirits? We acknowledge the existing concept of the “Pareto Effect” and define “Pareto Share” as the percentage of sales volume that is made to the top 20% of a category’s customers.
Secondly we aim to demonstrate that considering the customer base over a longer time period gives a dramatically different picture of customer concentration. Schmittlein, Cooper and Morrison (1993) first showed how buyer concentration (what we are calling “Pareto Share”) increases as we consider longer time periods. We aim to demonstrate this finding within the context of alcoholic beverages and compare the nature of that relationship for three alcohol categories.
The Negative Binomial Distribution (NBD) has been shown to often represent the category repeat purchase rate (Ehrenberg 1959). Thus the NBD will be the major modelling tool we use to consider the “what ifs” of the category consumer profile.
The paper evolves as follows: i/ We consider the reported weekly consumption data for Wine, Beer and Spirits for a population of 4800 survey respondents, fit the NBD and assess the fit of the model to the observed data. ii/ We compare the Penetration and Pareto Share across product categories. iii/ The time period of observation is extrapolated within the model to simulate the change in perspective of a senior manager from one week to one month or one year. iv/ Theoretical and managerial implications are discussed.
Pareto Share
A corollary to the unevenness in concentration that comes about from the presence of heavy and light users is the commonly quoted “Pareto Effect”. Pareto was a late 19th century Italian engineer turned economist that first developed a mathematical description for inequity in a country’s income distribution. (Persky 1992) His original empirical generalisation was later applied to areas such as statistical quality control and later to the social phenomenon of unevenness in customer concentration. (Weiner 2000)
To a marketing manager, Pareto means “80% of our sales are made to our top 20% of customers” (Buchanan 2002); (Sanders 1987). In reality, the proportion of sales to the top 20% of customers often seems closer to 60% and varies considerably, based on the time period of observation (Schmittlein, Cooper and Morrison 1993) and the market share of the brand (Rungie, Laurent and Habel 2002).
This percentage of sales to the top 20% of customers, or “Pareto Share” as we define it, appears to be a valuable tool to address the nature of product categories. It allows us to understand to what degree the heavy users of that category account for its turnover. “Pareto Share” may offer insight to brand managers as to whether to pursue increases in penetration of their brand (as per (Ehrenberg and Goodhardt 1990)) or increases in loyalty (Zeithaml, Rust and Lemon 2001) or purchase frequency (Peppers, Rogers and et al. 1999) (Baldinger and Rubinson 1997).
Most importantly, looking at Pareto Shares gives us the opportunity to compare categories – those with similar Pareto Shares are dealing with customer bases that are very similar in their behavioural loyalty to the category. That is, categories that may appear to be entirely dissimilar may have their customer bases behaving in the same sort of manner. Some early findings (Allsop 2002) show that - for example – “purchase of desserts” and “trips to the supermarket” are identical in how much volume is accounted for by the top 20% of their customers. This non-intuitive finding demonstrates the benefit of stepping back from the complexities and nuances of “what you see on the shelf” and using the tool of Pareto Share to clinically assess what your customer base is doing.
Results
Data Set
Survey data of approximately 4800 respondents was analysed. This was based on a quota sampling of the Australian population for the Australian Bureau of Statistics Population Monitor in 1995. Face to face interviews were taken in the house of the respondent. Whilst descriptive statistics for respondents were collected, it is the consumption statistics that will be analysed here. Respondents were asked to recall their consumption of alcohol products on a daily basis for the week prior to the interview.
Data has been recoded into weekly consumption (ml) of three alcohol types: Full Strength Beer, Wine (Comprising Red, White and Sparkling), and Spirits. The original data collection included a number of other liquor types that are not included in this analysis, namely liqueurs, fortified wines, light beers, extra light beers.
Observed Figures and Fit of the Model
We fitted the Negative Binomial Distribution to each of the categories. The NBD has a shape that will vary according to the two parameter values that are used in it. The act of fitting the distribution is a matter of determining which parameters create the shape of the NBD that most closely fits the observed data. The histograms of observed and theoretical consumption for each category were plotted on the same set of axes in order to a/ give a general picture of the category consumption behaviour and b/ indicate the fit of the model.
The first thing we see is that the theoretical figures fit the observed data quite closely. We also see that approximately 400 people consumed only one glass of wine in the week and that 200 people had only two glasses.
Aside from the poor fit of the model it is valuable to note that the numbers of light users of the category is less than that for wine and that the “tail” to the right hand side of the graph is fatter. We will later show that this higher number of heavier users gives a higher Pareto Share.
Spirits, again, exhibits the classic J-curve of the NBD under certain parameter conditions. Clearly the NBD is working quite hard to achieve a fit and describes the category consumption quite well.
It is important to consider the beer category, in particular, with care. It is plain to see that the fit is not as firm as the other two categories. This may be because we chose not to include light beer as part of the category or simply because of the limitations of the model in being able to fit all types of consumption/purchase behaviour.
Whilst looking at histograms of category consumption is valuable for considering the fit of the model and gaining a “picture” of the category, we need to look a little closer to gain insight as to the behaviour of the customer base.
The observed and theoretical Category Penetrations and Pareto Shares for the reported week are tabulated below.
26% of the sample consumed wine in the sampled week. This was significantly above beer (20%) and spirits (13%). This may infer that wine has a more universal appeal than beer or spirits in Australia.
The observed Pareto Share for wine was also below that of both beer and spirits. That is, the top 20% of wine consumers accounted for proportionally less of the total consumption than in the case of beer or spirits. Therefore the results indicate that as well as having a broader appeal across respondents, wine appears to be drawing its consumption less from heavy users than the other two forms of liquor.
In all three cases the NBD overestimated the Pareto Share of the category. This could well be due to the exclusion of the more marginal product types for each category (fortifieds for wine, light beer for beer, liqueurs for spirits). A predominance of light users of these marginal forms would explain this consistent overestimation.
Differing Perspectives
Another factor clouding the Pareto Effect is likely to be the perspective of the observer. A sales manager with a twenty-year history is unlikely to view his customer base from the same perspective as the sales representative who started a month ago. The sales manager is likely to recall customers who simply have not purchased recently. These “once a year” customers are more likely to enter the sales managers mental sampling time and change his perspective in two ways: i/ he will see higher penetration (the people who consume at all in the period), and ii/ the heavy users (the “once a week” buyers) become a smaller percentage of the total customer base.
This intuitive phenomenon can be modeled. The longer mental period that a sales manager observes can be replicated by extrapolating the parameters of the model. Thus while we have observed a week of market behaviour and captured it in the model, we can now look beyond the observed data. If the sales manager tends to consider a twelve-month period, we can do the same thing with the help of the model.
Extrapolation to longer time periods
Schmittlein, Cooper and Morrison (1993) demonstrated that the NBD could be extrapolated to estimate the market behaviour for longer time periods beyond the observed period. observation. We decided to assess the Category Penetration and Pareto share for our three categories based on a consumption period of one month, and one year.
Category penetration over longer time periods
Firstly the category penetrations (theoreticals only) for each of the time periods are tabulated below.

This straightforward result confirms our expectation that that sales manager will perceive a greater category penetration. It shows how the “once a year” customers will appear in the mental time frame (one year) of the sales manager and increase the overall penetration.
Whilst straightforward, the result is valuable in itself, in that it allows us to examine the penetration of wine on a long term basis. That is, in a year, close to 60% of people will drink some wine, compared to around 40% for beer and 30% for spirits.
Pareto Share Over Longer Time Periods
The Pareto Shares (theoreticals only) for the three categories, over the three time periods are plotted below.

In all three cases the Pareto Share increases as the time period of observation increases. That is, if we consider wine consumption for one year, the top 20% of consumers would account for 75% of the wine consumed. Interestingly, both Beer and Spirits’ Pareto Shares were about 80% for the one-year period which for these categories and that particular time period fits the terminology of “80:20 Rule”. The variability, however, across both categories and observation periods indicates this is more of a “Rule of Thumb” than a common occurrence.
Discussion
What is Pareto and why does it vary?
It seems that the “Pareto Effect” is not a clear 80:20 relationship but that it varies among categories. Whilst it is not safe to make any general rule, the Pareto Share is more likely to be around 55% to 65% and most importantly – it varies with the time period of observation.
This variation in Pareto Effect can be described using a simple example. Consider a woman aged 23, her mother and her grandmother. The 23 year old woman consumes wine about twice a week as part of her busy social life, the mother maybe once a month at family get – togethers, and the grandmother maybe once a year at special occasions.
For a sampling time of a week, the 23-year-old gets into the sample but neither of the lighter users do. The sample is then full of consumers with a purchase frequency similar to hers. When we observe for a month, the mother is captured in the sample. By that time the daughter has consumed 8 times compared to the mother’s one. Thus the concentration, or Pareto Share of the one month sample is greater than the one week. Once the sampling period is extended to a year, the Daughter, Mother and Grandmother are all in the sample, with volumes of 104, 12 and 1 respectively. By including more light users the heavier users become a smaller percentage of the customer base and the Pareto share increases.
Is wine different?
The higher penetration of wine in Australia as demonstrated in Tables 1 and 2 may be due to the multiplicity of functions that wine performs, such as a vehicle for learning, play, cultural assimilation, socialising (Groves, Charters and Reynolds 2000). The broadness of the appeal of wine across sexes (Keown and Casey 1995) is likely to be a contributing factor at least in Australia.
At the longer time periods of observation the category penetration rates all increase, and wine’s penetration increases its lead as we consider longer time periods. Wine consumption behaviour appears quite different to that of other forms of alcohol.
The lower observed Pareto Share for wine lends support to the notion that wine is the alcohol form that is consumed in moderation. The result may support the positioning of wine as an aid to health when consumed in moderation. (Spawton and Lockshin 2002)
The extrapolation to a longer time period shows an increase in Pareto Shares. The point of interest here is the variation in the rate of increase in Pareto Shares between categories. The Pareto Share increase for spirits is much greater in the transition from one month to one year than it is from one week to one month. (See figure 1) For wine the steps up are more even. The causes for this are worthy of further investigation, but the difference in the pattern is another indicator that wine is different to other forms of alcohol.
One less reason to panic
The increase of Pareto Share over longer time periods of observation is predictable, and theoretically well substantiated. Managers need to note this in their assessment of their customer concentrations. There is a case of a senior brand manager becoming alarmed when they considered their customer concentration, to find later that they had been considering 3 year data and comparing it to one year data. This paper has clearly demonstrated that this manager could have expected his concentration to appear different.
This brand manager’s alarm could have been soothed had they calculated the limit Pareto for their brand, and understood that concentration is dependent on the time period of observation.
A measure of the category’s health?
Pareto Share is a measure of how beholden a category (in this case three alcohol beverage categories) is to their heavier users. If the wine category is seen as competing with other categories for a share of the market’s consumption (Soft Drinks 2003) it may well be argued that other forms of choice modelling techniques may be used to describe the choice behaviour. One such model is the Dirichlet (Goodhardt, Ehrenberg and Chatfield 1984), which describes the multivariate purchase rate for brands competing within a category and involves a significant amount of the NBD theory contained in this paper. Under this particular model there are a number of parameters, but only one that relates directly to the strength of that brand. Each brand what is termed its “brand alpha” parameter, which in the model is the “weapon” it takes into battle with the other brands. Increases in a customer’s propensity to buy the brand in preference to others is directly represented as an increase in this “brand alpha”. It has been shown that an increase in the brand alpha is positively associated with an increase in Pareto Share of the brand. (Rungie, Laurent and Habel 2002)
As a result of its direct relationship to brand strength, Pareto share may well be the “signpost to the wine category’s health” that we are seeking.
Limitations and further research
We acknowledge the limitations of using face-to-face interviews, and relying upon respondent memory; this may have introduced a degree of collection bias. A cross sectional data set also has its limitations, as does the collection period of a single week which did not allow for observed/theoretical comparisons over differing time periods of observation.
Further research could include observations at other time periods with varying time periods of observation. As noted previously the exclusion of fortified wines, liqueurs and light beers may well have reduced the tightness of the fit of the NBD to our three categories.
This paper dealt with consumption behaviour – another interesting consumer characteristic is purchase behaviour. A similar analysis conducted on repeat purchase panel data is likely to yield interesting results.
At brand level there are Pareto patterns that vary with market share as demonstrated in FMCG. (Rungie, Laurent and Habel 2002) The variation of Pareto Share for wine brands could be modeled using similar techniques. Considering the crowding of the wine category this would constitute an analysis at the lower ends of the market share spectrum. We find no evidence of this having been done before.
Conclusion
We have demonstrated how customer concentration varies between three categories of alcoholic beverage and how that concentration varies with the time horizon. As such we consider this paper a valuable replication of Scmittlein Cooper and Morrison’s 1993 work in a different context. We have introduced the terminology of “Pareto Share” and defined it as the percentage of sales made to the top 20% of customers, and demonstrated how it may serve as a guide to a category (or brand’s) health. A deeper understanding of Pareto is likely to give the industry insight as to how to grow the category, and maybe give brand managers less cause to panic when they look at their panel data.
References
Allsop, J. (2002). Simple Quantitative Expression of Double Jeopardy. Marketing Science Centre. Adelaide, University of South Australia.
Anschuetz, N. (1997). "Profiting from the `80-20 rule of thumb'." Journal of Advertising Research 37(6): 51.
Baldinger, A. L. and Rubinson, J. (1997). "The jeopardy in double jeopardy." Journal of Advertising Research 37(6): 37.
Blaylock, J. R. and Blisard, W. N. (1993). "Wine Consumption by US Men." Applied Economics 24: 645-651.
Buchanan, M. (2002). "Wealth Happens." Harvard Business Review 80(4): 3.
Ehrenberg, A. S. C. (1959). "The Pattern of Consumer Purchases." Applied Statistics 8(No. 1): 26-41.
Ehrenberg, A. S. C. and Goodhardt, G. J. (1990). "Double Jeopardy Revisited." Journal of Marketing 54(3): 82.
Goldsmith, R. E. and d'Hauteville, F. (1998). "Heavy Wine Consumption: Empirical and Theoretical Perspectives." British Food Journal 100(4): 184-190.
Goodhardt, G. J., Ehrenberg, A. S. C. and Chatfield, C. (1984). "The Dirichlet: A Comprehensive Model of Buying Behaviour." Journal of the Royal Statistical Society 147(part 5): 621-655.
Groves, R., Charters, S. and Reynolds, C. (2000). "Imbibing, Inscribing, Integrating and Imparting: A Taxonomy of Wine Consumption Practices." Journal of Wine Research Dec2000, Vol. 11 Issue 3.
Keown, C. and Casey, M. (1995). "Purchasing Behaviour in the Northern Ireland Wine MArket." British Food Journal 97(1): 17-20.
Peppers, D., Rogers, M. and et al. (1999). "Is Your Company Ready for One-to-One Marketing?" 77(1): 151.
Persky, J. (1992). "Pareto's law." Journal of Economic Perspectives 6(2): 181.
Rungie, C. M., Laurent, G. and Habel, C. A. (2002). A New Model for the Pareto Effect (80:20) at Brand Level. Anzmac, Melbourne, Anzmac.
Sanders, R. (1987). "The Pareto Principle: Its Use and Abuse." Journal of Consumer Marketing 4(1): 47.
Schmittlein, D. C., Cooper, L. G. and Morrison, D. G. (1993). "Truth in Concentration in the Land of (80/20) Laws." Marketing Science 12(2): 167.
Soft Drinks, I. (2003). Soft Drinks Do What We Preach. Soft Drinks International.
Spawton, A. L. and Lockshin, L. (2002). Consumer Response to Societal Controls Designed to Modify Alcohol Usage and Abuse: Marketing Policy Implications. Bacchus to the Future Wine Conference.
Twedt, D. W. (1968). "How Important is Marketing Strategy to the 'Heavy User"?" Journal of Marketing 28: 71-72.
Weiner, J. (2000). "Writing Off 80% of Customers Is Economics Gone Berserk." American Banker 165(186): 11.
Zeithaml, V. A., Rust, R. T. and Lemon, K. N. (2001). "The Customer Pyramid: CREATING AND SERVING PROFITABLE CUSTOMERS." California Management Review 43(4): 118.