You are on page 1of 19

Journal of Business Research 57 (2004) 1307 – 1325

Understanding brand performance measures: using Dirichlet benchmarks
Andrew S.C. Ehrenberga, Mark D. Unclesb,*, Gerald J. Goodhardta

London South Bank University, London, UK School of Marketing, University of New South Wales, 3rd Floor, John Goodsell Building, Sydney NSW 2052, Australia Received 1 January 2002; received in revised form 1 August 2002; accepted 1 November 2002


Abstract Sales of a brand are determined by measures such as how many customers buy the brand, how often, and how much they also buy other brands. Scanner panel operators routinely report these ‘‘brand performance measures’’ (BPMs) to their clients. In this position paper, we consider how to understand, interpret, and use these measures. The measures are shown to follow well-established patterns. One is that big and small brands differ greatly in how many buyers they have, but usually far less in how loyal these buyers are. The Dirichlet model predicts these patterns. It also provides a broader framework for thinking about all competitive repeat-purchase markets—from soup to gasoline, prescription drugs to aviation fuel, where there are large and small brands, and light and heavy buyers, in contexts as diverse as the United States, United Kingdom, Japan, Germany, and Australasia. Numerous practical uses of the framework are illustrated: auditing the performance of established brands, predicting and evaluating the performance of new brands, checking the nature of unfamiliar markets, of partitioned markets, and of dynamic market situations more generally (where the Dirichlet provides theoretical benchmarks for price promotions, advertising, etc.). In addition, many implications for our understanding of consumers, brands, and the marketing mix logically follow from the Dirichlet framework. In repeat-purchase markets, there is often a lack of segmentation between brands and the typical consumer exhibits polygamous buying behavior (though there might be strong segmentation at the category level). An understanding of these applications and implications leads to consumer insights, imposes constraints on marketing action, and provides norms for evaluating brands and for assessing marketing initiatives. D 2003 Elsevier Inc. All rights reserved.
Keywords: Brand performance measures; Benchmarks; Repeat-purchase markets; Penetration; Loyalty; Segmentation; New brands; Marketing mix; Dirichlet model; Double jeopardy

1. Introduction 1.1. Overview Sales of Folgers, P&G’s leading brand of instant coffee in the United States, depended on it being bought three times a year on average by 11% of households. Roughly 12% of these customers were 100% loyal. Therefore, most customers bought other brands as well. In total, they did so about as often as they bought Folgers itself—a ‘‘share of category requirements’’ (SCRs) of only 50%. Such performance measures are routinely tabulated in marketing research reports and cited in marketing plans. But to interpret the figures, benchmarks and norms are needed. For example, is it that as many as 12% or only 12% are 100% loyal? The thesis of this paper is that there is a simple
* Corresponding author. Tel.: +61-2-9385-3385; fax: +61-2-9663-1985. E-mail address: (M.D. Uncles). 0148-2963/$ – see front matter D 2003 Elsevier Inc. All rights reserved. doi:10.1016/j.jbusres.2002.11.001

and general answer to such a question: most performance measures of most brands are just about normal. And in general, sales of a brand are largely determined by much the same predictable patterns of buyer behavior. Consumers are seen as mostly choosing from personal split-loyalty repertoires, typically buying one brand more often than another. Within such a framework of steady but divided loyalties, specific purchases then occur in a seemingly irregular or even ‘‘as-if-random’’ manner. These personal repertoires of brands differ from one consumer or household to the next. Yet such heterogeneous behavior aggregates to brand performance measures (BPMs) that follow much the same ‘‘lawlike’’ pattern from brand to brand. Key examples are that any loyalty-related measure is usually much alike for different brands, that a brand typically has many light buyers, and that few of its customers are 100% loyal over a sequence of purchases. Superimposed are regular relationships with market share, e.g., for purchase rates and for brand switching.

Ehrenberg et al. can they be of any interest to marketing practitioners who are always trying to change that steady state? The answers to both these questions appear to be a clear ‘‘yes. between competitive brands. etc. A new-brand case Many marketing analysts seem to be puzzled how a steady state model like the Dirichlet.. The model is defined for steady state and unpartitioned markets where market shares are stationary and there is no clustering of particular brands. Policy A was to position Brand X as a niche brand since 16% of consumers had rated it very high in placement tests. and potential critics. This probability is assumed to be steady for the time being but differing across heterogeneous consumers. The Dirichlet-type regularities and associated model still cause surprise. Both the observed and theoretical purchase rates for the different brands were very similar at about 3. sales promotions. (iii) Assessing and interpreting dynamic non-steady-state situations such as price promotions.’’ might nonetheless be of use in dynamic situations such as launching a new brand. motor cars.7 times a year by its buyers and hence by 3% of all consumers (since 8/2.2.’’ a U. This meant that the new Brand X also should be expected to stabilize at about 3. and sales trends. These patterns are in turn closely predictable from a single and parsimonious model. the Dirichlet mostly still holds and it provides useful benchmarks.’’ Given that steady state Dirichlet norms or benchmarks occur widely. But. with the theory being summarized in Appendix A. It would be competitively priced. / Journal of Business Research 57 (2004) 1307–1325 These ‘‘lawlike’’ patterns have been found in over 50 varied product and service categories from soap to soup.C. BPMs are all determined in the model just by the brand’s market share and only indirectly by the marketing mix or consumer-related factors acting through the market share. teachers/students.2. (Section 5). and their theoretical interpretation and practical use is at times controversial. Policy B was an extreme mass marketing positioning. it is possible to address many practical marketing issues (Section 4): (i) Auditing a brand’s performance to see whether or not it is in fact normal. (ii) Identifying market partitioning and other departures from the basic norms. satisfying the sales target of eight purchases per 100 households). Can there be ‘‘lawlike’’ patterns or even just a ‘‘near-steady state’’ in markets that are constantly subjected to competitive inputs. But the patterns are not always well-known or understood. To illustrate. X was therefore to be targeted at a small heavy-buying segment who really liked the brand.7 = 3). or relatively little. This shows the implications of knowing (or not knowing) the common Dirichlettype market pattern that brand loyalty varies little. with some 1% of the population buying it about eight times a year (thus. As a niche brand. which contains no explicit ‘‘decision variables. and the underlying theory—also has wider implications for our understanding of consumers. X was to be premium priced. this does not imply that all markets should be stationary and nonpartitioned (although many often are). prescription drugs. it would be bought about 2. the analysts concluded that if the new Brand X did reach its ultimate targeted share of 5% (or eight purchases per 100 households). advertising. Therefore. media usage. ‘‘X. relying on product and advertising pull rather than trade push.S. The management was then given the choice of two marketing policies.1308 A. technological innovations. as is discussed in Section 2. The new Brand X In considering a new instant coffee brand. Distribution was also to be up-market. the mathematical model. ‘‘try-something-different’’ advertising in mass media. Therefore. a realistic sales equation (where sales = percent buying  rate of buying) would be 3% buying 2. with awareness-building promotions. X would be an add-on or variety brand based on its special product formulation. practitioners. 1. with loyaltybuilding promotions and advertising in up-market media. The results in Table 1 showed that both policies A and B were totally out of step with the market. and in different countries and at different points in time. differentiation and positioning. or where there is some clustering. and environmental changes? Moreover. the Dirichlet (or ‘‘NBD-Dirichlet’’ in full). This target had been set by the management after having considered various marketing-mix expenditures and estimated revenues.1. even if near-steady state markets do exist. without expensive trade deals. The Dirichlet framework—the empirical patterns. It would reach the preset sales target of eight by being bought by some 8% of the population but only about once a year. we outline a simple new brand case based on an amalgam of practical experience. A and B. etc. brands. corporation had decided on a market share of 5% or eight purchases a year per 100 households. 1. and trade support bought by slotting allowances and eye-catching merchandising. However. our aim in this position paper is to bring together.S. The management then asked for a check against consumer panel data. . new brands. The empirical Dirichlet-type regularities have seldom been criticized. The Dirichlet model postulates that each consumer has a certain propensity—a probability in the model—to buy a given brand.7 times (a normal brand) compared with the hypothesized strategies of policy A where 1% buy eight times (a niche brand) or policy B where 8% buy once (an add-on brand). even where the market is not quite steady. and synthesize these findings for marketing researchers. Interpolating more precisely. update. This is outlined in Section 3. The model purports only to describe what markets are like when they are near steady and nonpartitioned.

Unilever.g. This forestalled that it would be one of the traditional ‘‘four out of five new brands that fail. .) (Bucklin and Gupta. Patterns of buyer behavior The basic finding about individual consumers’ varied buying behavior is that whichever way the data are aggregated.4 SCR 50 b Switching (annual) Percentage of Folgers buyers who also boughta MH 31 TC 24 Ne 21 Sa 12 HP 1 Ma 1 Br 1 Other 21 Observed a b For brand names see Table 1. GfK.2. She argued that in fact competitive brands are mostly much alike. years. The analysts countered that the close fit of the well-established theoretical norms (T) showed that the pattern was neither accidental nor unexpected.. Yet the marketing plan for X had claimed nothing of the kind. niche versus variety branding. IRI.2 3. etc.2.. A postmortem Later. Share of category requirements. instant coffee brand. But the number of customers buying Folgers increased greatly. From this.0 2.1.9 2. These patterns slowly came to be recognized over the years. First.6 2. Colgate. etc. a Outlier.e. As a result. yet some are able to achieve greatness.7 target.7 3.1. she questioned whether a me-too launch should have been ruled out as unattractive compared with launching a The variables analyzed are the standard BPMs used by large consumer panel operators (e. 2.8 0.1 2. Brand performance measures Data: Hallberg (1996)/MRCA.6 3. This struck her as an inappropriate ‘‘anything goes’’ approach to planning. 2. plans for Brand X were dropped.5a 2. O = observed.2 Percent buying Once 46 5+ 18 Category Purchases 6. We now step back from this introductory case history to review the main patterns of buyer behavior that have been observed for frequently bought products. are very different conceptually: Market share ð%Þ ¼ Total purchases of the brand Total purchases of the category Penetration ð%Þ ¼ The number buying the brand at least once The total number of potential customers Folgers market share (24% in terms of purchase occasions) was much the same in any length time period in a typically near-steady market. P&G. The NPD team now queried whether the similar average purchase frequencies of about three in Table 1 were not an unexpected coincidence.1 3.’’ 1. Sudman and Wansink.3 16 5 Percent buying O 11 10 9 6 5 1 0. Ehrenberg et al.C. / Journal of Business Research 57 (2004) 1307–1325 Table 1 Annual penetrations and purchase rates (leading brands of instant coffee) Instant coffee (USA 1992) Market share (%) 24 22 17 11 9 1 1 0. from about 1% in a week to 11% in a year. 2002).6 4. Second.2 8 3 Purchases (per buyer) O 3. TNSofres. she queried that two totally different policies. Kraft.8 2. with at least one big exception (for Maxim)..1. X would have to differ far more from the existing brands than they differed from each other.0 2.0 – T 3. management concluded that the new brand would not reach its 5% sales target via either the niche policy A or the mass market policy B. ACNielsen. they show regular patterns.A.1 3. They added that to break out of this pattern.) and their clients (e. followed by many replications across different products.S.3a 0. They are of three kinds. one with a distinctive built-in selling proposition).8 2. Nestle.S. T = theoretical Dirichlet predictions. 2. and countries to develop their generalizability.6 2.7 1309 Folgers Maxwell House Tasters Choice ´ Nescafe Sanka High Point Maxim Brim Other brands The new Brand X brand with a difference (i. had been put forward as alternatives for the same functional Brand X. senior management asked their new marketinginsights director to critique the earlier planning. Table 2 Common Brand Performance Measures (BPMs): Folgers USA 1992 Folgers Brand size Market share (%) 24 Percent buying in a: Week 1 Year 11 Loyalty-related measures (annual) Purchases (per buyer) 3. 1999.0 2.g. market share and market penetration. as illustrated in Table 2 for the leading U.2 8 – T 12 11 9 6 5 1 0. Brand-size-related measures The two main measures of size. They also felt that X’s allocated budget was not sufficient for an achievable me-too 3 Â 2.1 3.

On average.4 1.C. Compared with what: regularities from brand to brand To interpret these numbers for Folgers.9 0. Smaller brands therefore not only have far fewer buyers than the bigger brands but also show somewhat lower average purchase frequencies (i.4 Yearly 46 35 25 22 14 12 11 6 21 Loyalty-related measures (annual) Purchases (per buyer) Percent buying (5+ times) 24 21 16 13 19 11 11 13 16 Category Purchasesb 10 11 12 11 12 12 12 13 12 SCR (%) 39 33 26 24 26 22 22 24 27 100% loyal Percent buying 22 16 11 11 12 10 9 7 12 Purchases 4 4 4 4 3 3 3 3 4 Switching (annual) Percentage who also bought brand:a A 100 51 57 55 53 56 56 52 54 C 31 32 100 35 34 35 35 30 33 E 17 18 20 23 100 19 17 17 19 A B C D E F G H Average brand 3.4).e.g. These switching levels are themselves proportional to the brand’s penetrations—the so-called ‘‘duplication of purchase law.’’ as described in Appendix A. laundry detergents (2Â). any instant) 6. a A selection of brands.or switching-related measures either stay broadly the same from brand to brand or decrease far less (by factors of 2 or 3 at most). The markets in question were therefore nonpartitioned.5 1. e. toothpaste (2Â). 100% loyals are in fact light users of the whole category (on average about 4 such purchases versus 12 by all of the brand’s customers).. cheese.7 2. SCRs—The average SCR of a brand over the year is quite low.6 0.. Switching-related measures Table 2 also shows which other instant coffee brands Folgers’ customers bought once or more in the year. 3.. Loyalty-related measures Folgers customers averaged 3.. Which other brands are also bought—Much the same proportions of any particular brand’s customers also bought Brand A (all about 54%). about 27% annually (on average 3.2.g. Why are they what they are? Is an 11% annual penetration high or low? Is a SCR of only 50% as threatening as it seems? Does the relatively high purchase duplication of Folgers with Maxwell House mean that these two brands are highly competitive—much more so than with Maxim? 2. But half bought the brand only once. This atypical (upward) trend was called ‘‘natural monopoly’’ by McPhee in 1963. orange juice.1 2.  2. by almost 10-fold. They do not have many opportunities to be disloyal.5 1. To illustrate.8 3.2 3. giving the brand an SCR of 50% (i. i. from Brand A to Brand H. who explained it as statistical selection effect. household cleaners.2. But some 50% bought it just once in the year (as for Folgers in Table 2). Sources: AGB (UK). with even fewer for a small brand (DJ again).e. showing multibrand buying behavior.. GfK (Germany). we can compare them with those for other brands. This tendency to be ‘‘punished twice’’ just for being small was called ‘‘double jeopardy’’ (DJ) by McPhee (1963).9 3. The natural monopoly effect—How often customers of a brand bought the whole category increases slightly from 10 to 13 with decreasing market share. take-home beer. Any brand’s customers mostly buy other brands.2 Categories: Catsup. one in six of a brand’s annual customers bought it five or more times (with DJ again: smaller brands have even fewer heavier buyers). / Journal of Business Research 57 (2004) 1307–1325 2.S. 100% loyals—Only some 12% of a brand’s customers were 100% loyal in the year. It implies that large brands slightly ‘‘monopolize’’ light category buyers. These customers bought the category (i. cereals. Similarly for C (about 33%) or E (about 19%). Heterogeneity—Consumers are very heterogeneous.2/6.. lower repeat buying).2 2.1.7 3.6 2. Table 3 Performance measures brand by brand (averages for 8 leading brands in 12 product categories) Brands (by share) Brand size Market Share (%) 28 19 12 9 7 5 4 3 11 Percent buying Weekly 3. with no apparent clustering of some subsets of brands. The DJ effect is however small unless penetrations are very high. loyalty.2 purchases in the year. .e. 31% also bought Maxwell House. The question now is what these figures mean. paper towels (2Â). Table 3 compares a dozen performance measures across the top eight brands in 12 varied product categories internationally.3. e. and only 1% also bought Maxim.e.6 3. TCI (Japan).1310 A. In contrast. Several patterns are apparent in Table 3 and for each of the 12 categories:      Double jeopardy—Market shares and penetrations decrease greatly. Nielsen and IRI (USA).2/12).4 times on average. b Per brand buyer.2 0. Ehrenberg et al.7 0.1.

cleaners. appropriate analysis periods might be six months and a year. etc. Beyond that. -Private/own labels -Price bands  Time. cars. The other concerns brands. Ehrenberg et al. The fit for each separate measure can be summarized in two respects: (a) the virtual lack of systematic bias. loyalty-related measures are much lower in a longer period: nearly every buyer is 100% loyal in a week (when at most a single purchase of the whole category is usually made) but few are in the year. Similarly for duplicated buyer’s average rates of buying. The Dirichlet assumptions The length of the analysis period—A brand’s penetration being much higher in a year than in a week (as in Table 2) is less than pro rata due to some repeat-buying occurring. These notions are captured in the theoretical model by five specific distributional assumptions: two assumptions are about consumer heterogeneity for purchase frequency and for brand choice. 3. To use the model. whereas for gasoline—where average interpurchase times are shorter—a week and a quarter might be more appropriate. programs. Brands are characterized in the model by their purchase probabilities and hence their market shares. and subsequently by Bass et al. One concerns consumers. B2B -TV episodes. are likely to be ambiguous and misleading. In contrast. space. Consumers are thought of as highly experienced. Australasia. a consumer’s favorite brand typically may account for only some 60% of their annual category purchases. Japan. The Dirichlet model Given their wide generalizability.C. consumers can behave as if they have. forms.. and the final one is about the statistical independence of these two aspects.  Other measures—Similar regularities occur for other BPMs. which were shown in Table 3. / Journal of Business Research 57 (2004) 1307–1325 Table 4 Varied conditions for Dirichlet-type patterns Products and services -Food. and channels Brands and product variants -Large and small brands -Pack sizes. drink. with an average deviation. Studies where consumers are labeled as ‘‘loyal. To estimate the BPMs of any brand of particular interest its own market share has also to be inputted (see Table A1 of Appendix A). The individual assumptions are spelt out in Appendix A. flavors. for syrup.. financial services. The model is therefore very parsimonious. or differently marketed—this is all subsumed in the model’s steady state purchase probabilities. There is strong theoretical as well as empirical support for each of these assumptions. steady habitual personal purchase propensities—stochastic probabilities in the model—for when they buy and what brands they choose on different purchase occasions. it is perhaps unsurprising that these patterns of buyer behavior can all be described Two broad propositions underlie the model. Only four numerical inputs are needed. two about the probabilistic incidence of specific purchases of the product and of a brand. -Light and heavy buyers: demographic subgroups -Household or individual purchases -Different length analysis periods -Different points in time: 1950-2003 Market conditions -Near-steady markets -Dynamic markets (for loyalty-related measures) -Nonpartitioned markets -Partitioned submarkets 1311 by a single statistical model—the NBD-Dirichlet (‘‘Dirichlet’’ for short)—and that this model is simple. UK. Anyone with access to consumer panel or similar data can check on such patterns. 3. . They are therefore no longer easily influenced by a further purchase or seeing yet another advert (i. typically the penetrations and the average purchase frequencies of the category and of one specific brand. It is usual therefore to examine periods that equal and exceed average interpurchase times. for the time being.’’ without specifying the period of observation. aviation fuel.g. the model does not specify whether the brands are functionally differentiated or not. no additional ‘‘learning’’ occurs). the model’s critical feature is that it is able to describe the various observed brand performance patterns. as shown in Table 4. or mean absolute deviation (MAD). The model was developed by Chatfield and Goodhardt (1975) to account for the known empirical patterns.A. Crucially for the Dirichlet model.e.S. To us. and in that sense. Table 5 illustrates the Dirichlet model’s close theoretical predictions of the observed data. of less than 2 percentage points for the penetration-type percentages). along with procedures for calculating the theoretical BPMs. e. a consumer’s probability of buying a particular brand would not be affected by what other brands the consumer is also buying. Again. Germany. and people -United States. Conditions under which the patterns occur The generalizability of each of the above patterns is now widely established across a great variety of very different product and services categories. (1976) on a theoretical ‘‘utility’’ basis. it also helps explain and predict them. etc. quarter-by-quarter repeat buying is much the same for different brands (about 70% for the average of the 12 products in Table 3 but with DJ effects again). consisting of just a few well-based assumptions. formats. Hence. and personal care -Prescription drugs. 2.1. OTC medicines -Gasoline. and (b) the O and T values for the individual brands also differing little (e. it has to be calibrated for the chosen product category and brand. PCs -Retail chains..3.g.

But this occurs only occasionally (Ehrenberg et al. Namely. Moreover.7 2.1 2.0 A O 100 51 57 55 53 56 56 52 54 T 100 55 56 56 56 56 56 56 56 C O 31 32 100 35 34 35 35 30 33 T 31 31 100 31 32 32 29 30 31 E O 17 18 20 23 100 19 17 17 19 T 20 21 21 21 100 22 18 18 20 O = Observed as in Table 3.5 3.8 2. repeat rates over nonadjacent periods tend to erode somewhat by eight percentage points over a year (East and Hammond. Ehrenberg et al. 1988)... 1959).000+ households). Alternatively.S.1312 A.5 2. which then need to be explained (e. . there are slightly too many ‘‘medium’’ buyers in the model). 10.2 T 3. 1996).’’ This was found to reflect a shortfall of very frequent buyers.. Given the large consumer panels that are used nowadays (e. ‘‘differentiating’’) marketing value so far. 1993.1 4.’’ Follow-up work is needed.6 T 2.. some systematic discrepancies have been reported. for example:   Quarter-by-quarter repeat buying can be overpredicted (by 5 to 10 percentage points—Ehrenberg. Attempts to improve or elaborate the Dirichlet have so far not resulted in major gains in either predictive power or parsimony (see Appendix A). / Journal of Business Research 57 (2004) 1307–1325 Table 5 Annual observed and theoretical performance measures Brands (by share) Brand size Market share (%) Percent buying Loyalty-related measures (annual) Purchases (per buyer) Percent buying 5+ times Category purchases per buyer 100% loyal Switching (annual) Percentage who also bought brand: Percent buying O A B C D E F G H Average brand 28 19 12 9 7 5 4 3 11 46 35 25 22 14 12 11 6 21 T 46 36 25 21 16 11 10 7 21 O 3. not just a sampling error). The more limited ‘‘empirical Dirichlet’’ model can then be fitted instead. Fader and Schmittlein.g.2.8 1. and heavy buyers of a total category (or a subcategory) is sometimes a little ‘‘flatter’’ than predicted by the NBD part of the Dirichlet model (i.8 2. 1996. This is so far unexplained.e. 2.2 3. 1988).8 3. which was ‘‘explained away’’ as an artifact because few people buy a typical grocery product more than once a week or more than 13 times in a quarter (Chatfield. This is because the model is for the steady state and assumes that these effects will usually already have been subsumed in the brands’ market shares. In addition.7 3. Bhattacharya. But the discrepancy usually varies relatively little from brand to brand (except with market share) and has not been of diagnostic (i. Ehrenberg. how very similar the brands’ average purchasing rates are in Table 5 compared with the eightfold variation (800%) in their penetrations.9 1. Ehrenberg. Bhattacharya et al..6 3. 1995). Kahn et al. But the modeling goal here is not to predict the sales volume of individual brands—these are already known—but to elucidate market structure.2 2.  An early systematic discrepancy was the so-called ‘‘variance discrepancy.6 3.2. 1997.9 2. Deviations from the model Other variables such as marketing-mix inputs or consumer attributes do not have to be explicitly specified.. a 5% difference from an average buying rate of 3.1 2. The reasons are not yet understood.1 O 24 21 16 13 19 11 11 13 16 T 25 20 17 14 14 12 12 13 16 O 10 11 12 11 12 12 12 13 12 T 10 11 12 12 12 12 12 12 12 O 22 16 11 11 12 10 9 7 12 T 16 13 11 10 9 9 9 9 11 Purchasesa O 4. by a purchase or two for each of the 100 brands in Table 5). 1967. they will show up as discrepancies in the model predictions.g. Reibstein and Farris.4 2.  The annual purchase rates of 100% loyal buyers are consistently underpredicted (e. which in turn affect the brand’s other performance measures.0 as in Table 5—e.2 2. Such known discrepancies seldom curtail the model’s practical use.e. 1988. It may be due to large brands being more likely to be kept in stock and may not be as important as is sometimes thought..g. 1988). 1990. 3. a Consistently low—see discussion of systematic discrepancies in Section 3.. Nevertheless. such as penetration growth (Ehrenberg.C.9 2.8 1.8 or 3.7 2.0 1.g.9 3.9 3.4 3.9 1.7 2. T = theoretical Dirichlet predictions.9 2. For some market leaders.4 4..e. although at the cost of not being able to predict theoretical length-of-time period effects. medium.6 3.2—will of course matter in terms of sales.  The distribution of light. This is a real departure from a ‘‘steady state’’ and can be regarded as a ‘‘slightly leaky bucket. many of the reported discrepancies from the model will be ‘‘significant’’ (i.9 2. annual purchase frequencies are a unit or so higher than predicted..

An analysis excluding the promoted month would have shown this more explicitly (but the raw data were no longer available).  A remarkably high rate of new medical prescriptions occurred in 1986 for Squibb’s cardiovascular drug Capoten: 10 new prescriptions a year per prescribing doctor instead of the norm of about 5. for example. Extensions to new conditions The Dirichlet also provides benchmarks when analyzing data for another year.. 1995. In the outcome. This took place in all the analyses listed in Table 4. Stern and Ehrenberg. Wrigley and Dunn. Practical applications We now review various practical uses to which the Dirichlet framework has so far been put. 2001). Television—repeat viewing and switching for programs and channels (Barwise and Ehrenberg. the Dirichlet norms showed that all but one of the measures in the table was normal. aviation fuel contracts (Uncles and Ehrenberg. 1990b). ‘‘Impulse’’ purchases—made for immediate personal consumption (McDonald and Ehrenberg..S. 1995). 2000. and ready mix concrete (Pickford and Goodhardt. Kau et al. Durables and services—repurchase of PCs (Long and Ehrenberg. 1995. 1998. 2000 with Ehrenberg et al.. Moe and Fader. 1980).. as shown in Table 5.. Industrial purchasing and procurement—isolated cases so far. Examples are auditing the performance of established brands. one factor at a time. such as specific combinations of pack sizes. or category. Bound and Ehrenberg. 4.. It was therefore not a general characteristic of the brand but a one-off effect due perhaps to a fire in a warehouse (‘‘bad’’) or an exceptional promotion with an extralarge bonus of onceonly buyers (‘‘good’’ presumably). womenswear retailing (Brewis-Levie and Harris. Doctors’ medical prescriptions—where the doctor neither buys. Maxim’s very high annual purchase rate of 4. 1998) and to markets in ‘‘developing’’ countries (Bennett. private labels seemed to show normal repeat-buying loyalty (see Uncles and Ellis. it is simpler to check whether the Dirichlet patterns recur. Colombo et al. 1987).  For instant coffee in the UK. in press. 2003).  Brim also had a somewhat low average purchase frequency of 2. some for very different kinds of markets or market conditions. Thus. instant coffee.g. The prescribing rate fell back to normal when the inducement was withdrawn (Stern and Ehrenberg. and model specification (as for cars or PCs). but more work is needed.S. SKUs unquestionably differ functionally (e. 1989. 1998) and cars (Ehrenberg and Bound.       Geographical extensions—to Australasia and Japan (Wright et al. Ehrenberg et al. Goodhardt et al. as was tested in two ways: by grouping PLs together as a megabrand and by analyzing a PL within its own store chain. 1990a. 2000). 1994. 2000). Uncles and Hammond.1 (Table 1). contrast Leeflang and Wittink. and checking the nature of unfamiliar markets.. Brand performance audits A common use of the Dirichlet is to assess how existing brands are performing. as against the average of about 3—perhaps the households were running a bridge club!).2. 2000) and frequent flying by business travelers (Harris. flavors. such as chemical additives. say. In essence. and recently on-line electronic shopping (Danaher et al.. Kau et al. the market leader Nescafe ´ has over the years had a somewhat high buying rate. In practice. 1985).5 was found to be due to two ‘‘outliers’’ (two households making over 30 purchases each. 4.1. 1984. 2002). Uncles and Ehrenberg. 1995). large versus small pack sizes). / Journal of Business Research 57 (2004) 1307–1325 1313 4. Isolated deviations occur but often tend to have fairly simple explanations. and of dynamic market situations more generally. predicting and evaluating that of new brands. This was traced to doctors who were provided with a ‘‘free’’ PC if they prescribed Capoten often enough for medical monitoring purposes (this might seem a highly successful loyalty inducement but would soon become financially crippling). nor consumes the product (Stern.. paper and packaging (Easton. 1984.g. 1995). pays for. In another study. Store choice—repeat visits and switching between grocery outlets (Kau and Ehrenberg. SKUs—customer retention and switching between stockkeeping units.A. This was probably due to their inherent nonavailability in other chains. of partitioned markets. 2000). Early work shows . the Dirichlet consistently shows somewhat higher buying frequencies. 1999). 34 product categories were systematically assessed to see if BPMs for P&G brands were as expected (Uncles et al. Instead of unfocused data mining. country. This had not occurred the year before (the first thing to check). 1997). 1998. ´ even given Nescafe’s high 34% share (possibly this results from the occasional brand leader effect already mentioned)..C. which is bought on average as often as once a week and has distinctive solus-site retail distribution (Scriven and Ehrenberg. 1998. it was not clear in Table 2 whether any of the figures observed for Folgers were high or low (or ‘‘good’’ or ‘‘bad’’). as in the following five cases:   For private labels. instead of using all-in-one marketing mix ‘‘decision-orientated’’ modeling (e. Whereas brands are often similar (‘‘the commodity with a name’’). Unusually frequently bought categories—gasoline. the Dirichlet provides theoretical benchmarks for assessing price promotions or advertising. Examples are:    In Table 1 for U. 1994).

Another application is in evaluating a new brand some time after its launch. 1988). they were at the Dirichlet level as for any brand in that market (Ehrenberg and Goodhardt. The purchase duplications between pairs of brands were relatively high. Table 6 Market partitioning: unleaded versus leaded gasoline (percent of buyers of X who also bought Y. The figures are not zero mainly because of households with two cars. 1994) (for simplicity. 1998). Hardie et al. The general view for new brands is that loyalty grows slowly (e. TotW seemed doomed until their marketing advisors made Campbells aware of the DJ phenomenon. In contrast.. Each of these functional attributes may then attract a special ‘‘segmented’’ following. Typical brands) Britain (Quarter I. that is. this time for the UK gasoline market between unleaded and leaded variants (Scriven and Ehrenberg. but with differing ‘‘duplication coefficients’’ (i. just four brands are reported in the table). they show no special clustering of particular brands. one older and still requiring leaded gasoline. A more extreme exception was Unilever’s new toilet soap Shield some years ago. It follows that the previous isolated cases of near-instant loyalty were not High between the unleaded brands in the northwest quadrant (and also the leaded ones in the southeast quadrant) and decreasing from some 30% to 10% in line with the brands’ penetrations. The ‘‘law’’ usually continues to apply directly not only within each separate partition but also between the partitions. or middling but always decreased with decreasing penetrations:  Low at about 5% between two different brands of either leaded or unleaded in the southwest and northeast quadrants.. as implied in analyses by ‘‘depth of repeat’’).  Middling at about 15% for the same brand unleaded and leaded (in the southwest and northeast diagonals). have reported two isolated cases of near-instant loyalty occurring unexpectedly for two small new brands in Australia and the UK. Similar points emerged from an exhaustive study of car repurchasing in Europe.e. with consumer choice showing marked clustering for the more similar items (partitioning). Wright and Sharp (1999). with repeat buying and switching also very high immediately. Ehrenberg et al.g. This achieved a remarkable 20% share in a very traditional UK market almost immediately at launch. 1991). the ratios of the duplication levels and the penetrations—see Table A4 of Appendix A for the calculations). New brands Using the Dirichlet for initial new-brand planning has already been rehearsed in Section 1 (the case of Brand X) (see also Ehrenberg. with ‘‘luxury cars’’ not unexpectedly showing up as the most distinct of several clusters (Ehrenberg and Bound. and indeed as for an established 20% brand in that market (Wellan and Ehrenberg.3 prescriptions per prescribing doctor).. 2000). again decreasing with the penetrations. decaffeinated and regular. One instance is in Table 7 for doctors’ prescriptions for the then new antidepressant Prozac (2. Here is a case where having prior knowledge of the duplication patterns can be expected to work much better than a statistical multivariate ‘‘discovery’’ technique. Market partitioning Markets for directly substitutable brands are usually not partitioned. i..3. again because of 2+ car families buying both variants mainly at the same local gas station. Campbell’s newish premium priced ‘‘Tastes of the World’’ (TotW) soup in Britain for example showed low loyalty-related measures.. compared with its bigger competitors. a new empirical finding across 22 cases is that average purchase frequencies for successful new brands were in fact normal almost instantly.C. / Journal of Business Research 57 (2004) 1307–1325  that the Dirichlet-type patterns nonetheless still predominate (Singh et al. where a considerable number of techniques were applied and compared and where a focus on expected duplication patterns greatly assisted the analysis (Colombo et al. 2000). coffee by ground and instant. with the nonpartitioned Dirichlet-related ‘‘duplication of purchase law’’ as a well-established norm. This showed that TotW’s low loyalty was in fact normal for its size. that repeat-buying levels are predictably lower for smaller brands. Each of the three cases was regarded at the time as unusual and unexpected. Table 6 illustrates another instance of complex-yet-simple partitioning. However. Such clustering is brought out by relating the observed duplications of purchase between pairs of brands to these brands’ penetrations. that is.4..g. This allows for the penetration of each brand. 4. 1990) Buyers of Esso unleaded BP Mobil Gulf Buyers of leaded Esso BP Mobil Gulf Who also bought Unleaded Es BP Mo Gu Es 18 6 9 6 Leaded BP 6 15 7 4 Mo Gu 2 2 15 3 2 2 1 17 % 100 25 18 11 % 36 100 23 14 % 32 30 100 10 % 29 25 14 100 % % % % % 19 7 4 5 20 4 13 4 4 14 5 5 16 2 11 2 100 2 28 2 25 17 22 8 19 23 13 9 100 16 13 25 100 14 27 18 100 15 10 8 Penetrationa a Percent buying at least once in the quarter. But some subcategories exist.1314 A.e. low. for example. .. and Ehrenberg and Goodhardt (1968a) much earlier. TotW’s problem was that it had too few customers. 2000).S. A striking case of partitioning has been reported for the automobile market. 2000). 4. But no generalizable results of this have been reported (e.

1997). Japan. Ehrenberg et al. 4.. 2002). Consumers had not learnt to like their ‘‘new’’ brands during the forced switching (Charlton et al. Table 8 shows that the sales drop was hardly due to a loss of loyalty (neither light nor heavier QI buyers bought much less than expected in QII). . 2002). which shows that although penetration and loyalty both grow when share grows (in line with the Dirichlet/DJ cross-sectional results rehearsed here). and (iii) there are virtually no new buyers (Ehrenberg et al.  Stockouts—Temporary out-of-stocks need not lead to a longer-term loss of sales: CTA showed experimentally that before-to-after repeat buying was normal. In another long-run study. (ii) no before-to-after sales increase occurs.. It was mainly due to too few ‘‘new’’ buyers coming in only 25% QI nonbuyers buying in QII rather than the expected 34%. The steady-state norms can then be used to identify and interpret those measures that are still normal and the measures that have in fact changed. the giant pack of Unilever’s leading UK laundry detergent was promoted with extra product. Aquafresh. This gives predicted repeat-buying norms from QI to QII conditional on each consumer’s QI purchasing level (Goodhardt and Ehrenberg. But the CTA norms showed the opposite—the offer of extra product attracted recent nonbuyers and light buyers (Ehrenberg. The conclusion was that price promotions are mainly taken up by existing customers of the brand: promotions just bring forward purchase timing where the brand is already in the purchase repertoires of consumers.S. Nonetheless. Further applications Dirichlet norms have been used as benchmarks in exploring many other marketing issues.S. Repeat buying was tracked in a ‘‘conditional trend analysis’’ (CTA) as a diagnostic tool. More work into market dynamics is planned.. These various cases seem to suggest that penetration— the number of customers—is the main factor that changes when sales increase.5. Only the market shares and penetrations had changed.  The seasonal soup market—The CTA norms showed that the winter peak of soup sales was about half due to allyear-round buyers buying more and half due to peakseason-only buyers (Wellan and Ehrenberg. 1994b)..3 II 3 2.  Extra product promotions—In an isolated analysis.. This O = Observed measures. the loyalty levels were predictably Dirichlet both in 1992 and 1981 (Ehrenberg. 4. Some isolated cases published so far include the following:  segmentation would affect the timing of advertising and any attempts to spread peak demand.e. 1999).3 QII-VIII 14 2. 1997. 1995. Average new prescriptions a quarter per prescribing doctor. and the United States has shown that (i) before-to-after repeat buying is unaffected (i.A.3 1 1. Scriven.. 1994a).6. and Sensodyne (Ehrenberg et al.  Price-related promotions—A wider use of Dirichlet-type norms in analyzing 150 price promotions in Britain. 1997b. 1972). increased penetration is the key factor (Baldinger et al.2 VII 17 2. (iii) consumer loyalty programs (Sharp and Sharp. and (b) Folger’s brand share had doubled. Analyzing market dynamics The steady-state Dirichlet norms have also been used to diagnose dynamic situations where there are marked changes in market shares and/or in total market size. exceptional after all. 1990). Our general conclusion is that the well-grounded steady-state norms reflected by the NBDTable 8 Conditional trend analysis (CTA): powder detergent A 15% sales drop from QI to QII Buyers of powder detergent in QI Nonbuyers O (%) Percent buying powder in QII 25 T (%) 34 Once only O (%) 57 T (%) 63 Two plus O (%) 88 T (%) 90 A short-term trend—A few years ago. instant coffee market. 1996). and (iv) subscription markets (Sharp et al.S. / Journal of Business Research 57 (2004) 1307–1325 Table 7 A new brand: near-instant loyalty (doctors’ ‘‘new’’ prescriptions of Prozac) Prozac Quarters (I) Percent prescribing Average prescription rateb a b a 1315 Average III 8 2. Morrison. the total sales of U. had changed radically from 10 years before in two ways: (a) category sales had slumped by two thirds. Germany. no ‘‘learning’’). This finding of near-instant loyalty could not have become apparent if the close similarity of loyalty measures for established brands (as in Table 5) had not already been well established. 1999).. This is strongly supported by a large systematic study.. Empirical examples are: (i) cannibalization (Lomax et al. 1969). (ii) price sensitivities (Ehrenberg et al. as if there had been no out-of-stock. laundry detergents dropped by as much as 15% from one quarter to the next. as in Tables 1 and 2. 1967.9 VIII 21 2. Scriven and Ehrenberg. 1988).  Long-term trends—The 1992 U. Loyalty was unaffected by the market trends.0 Prozac was launched at some time during this quarter. the UK toothpaste market was similarly found to be Dirichlet both in 1967/1968 and 1990/1991 despite a major decline in sales of Gibbs and the introduction of successful new brands—Crest. T = theoretical NBD predictions. Management expected this to appeal to the brand’s heavy buyers.C.

customers of functionally distinct subcategories differ: cat food buyers have cats and dog food buyers have dogs. Horowitz and Louviere. in choice modeling (e. Here. Because consumers are generally highly experienced. Such brands can be near . 5. and even then. But such brand segmentation is not directly allowed for in the Dirichlet framework. making a further purchase of a brand would not affect a Dirichlet consumer’s probability of buying the brand (the model’s crucial ‘‘zero-order’’ assumption—see Appendix A). Indeed. In contrast.g. Indeed.g. and presweetened cereals are eaten more by children. as noted in Section 2. consumers might come to identify with their habitual brands (‘‘I use it. needs. the motherin-law coming. 1971. Gordon. there can be strong segmentation at the category or subcategory level.2. Precisely when a consumer buys the product—e.1. 1985. However.. Much of this literature is focused on techniques. This is the underlying supposition that leads to the Dirichlet model. the common questions are ‘‘milk or sugar?’’ or perhaps ‘‘decaffeinated or regular?’’ but not the ´ choice of brand ‘‘Maxwell House or Nescafe?’’ Nevertheless. 1994). 1995. Dall’Olmo Riley et al. which nonetheless successfully predicts most BPMs. Olshavsky and Granbois. motives. 1959. whether this week or next—and which particular repertoire brand is then chosen are both assumed in the model to be as-if-random.g... 5. 1996.g.. and customer types often differ (e.. 1995). and in the context of limited problem solving (e. Instead. Understanding consumers 5. and feelings—such as out-of-stock situations. this may have been due to some overlooked nonstationarity rather than real feedback (Shoemaker et al. These implications follow logically from the preceding patterns. 1982). therefore I like it’’—see Barnard and Ehrenberg. Kennedy and Ehrenberg. consumer habits. 2002). Ehrenberg. and the marketing mix. The notion that a consumer’s past purchasing behavior or current ‘‘state dependence’’ will invariably change the consumer’s brand choice propensities seems to us to put the cart before the horse. 1999)—only limited empirical support has been cited for this in steady frequent-purchase markets. / Journal of Business Research 57 (2004) 1307–1325 Dirichlet successfully provide benchmarks for an unusually varied range of situations. models. promotions. 2000).1316 A. This similarity of brand-users’ attitudes is consistent with the small role brands play when consumers actually consume the product: in offering a coffee.1. consumers are polygamous rather than either promiscuous or monogamous (except possibly in ‘‘subscription’’ markets). larger households are usually heavier buyers of the product as a whole. Brand segmentation Consumers are widely expected to fall into relatively homogeneous and recognizable subgroupings. users of leaded gasoline had older cars.2.. 1997. there is now much systematic empirical evidence that the user profiles of substitutable brands seldom differ (Collins. etc.1. and applications. the data show that buyers of Brand A feel about A much as buyers of Brand B feel about B (Barwise and Ehrenberg. specific choices tend to be governed by a variety of reasons.g. 5.S. This is not to say consumers literally toss mental pennies. Wedel and Kamakura.. needs. Although purchase feedback is at times hypothesized in the literature—especially in discussions of choice models (e.. For Dirichlet-type markets.. 1994). 2000). special displays. Foxall and Goldsmith. Bock and Uncles. McAlister and Pessemier. 1977). They usually have several steady partners—a repertoire—with one or two usually being favorites (Hammond. the varying behavior of consumers is sufficiently idiosyncratic and irregular to be successfully modeled mathematically as being quasi-random. Marketing implications The Dirichlet framework also has implications for our broader understanding of consumers. Typically. 1997. Ehrenberg et al.g.1.g. 1979). 1979. brands. as would tend to follow if nearly each of them uses several brands.. Nor do brand users generally differ in their attitudes to the brands they buy. and consumers’ own diverse habits. in qualitative studies (e. 5. the customers of similar brands are very similar. Engle et al. Franzen. Fournier and Yao.. Seetharaman et al.g. Any ‘‘learning’’ would already have occurred in the past. Polygamous consumers Over any sequence of purchases... moods. For example. But such subpatterns then tend to hold about equally for all the substitutable brands in that category or subcategory. 1997. East.C. Details are extensively discussed in standard consumer behavior texts and papers (e. consumers tend to buy several brands with largely steady habitual propensities. For a consumer to be ‘‘loyal’’ to a brand. not empirical results (e. Day et al. at least for the time being. Despite this lack of segmentation between brands.. the Dirichlet model does not presuppose any unique or explicit ‘‘commitment’’ to that brand. especially collectively. Understanding brands The Dirichlet model is a theory about choice between competitive entities such as brands. the vast segmentation literature is surprisingly lacking in explicit empirical cases where directly competitive brands do appeal to different kinds of people. 1994. the experienced consumers’ steady purchase propensities can be thought as the outcome of years of past experience. 1997). nor that the consumer be either a particularly heavy or exclusive buyer of it. In practice. 1988. 100% loyals over any extended sequence of purchases are rare and are light buyers anyway. In addition. Hammond et al. as we now outline. 2000). Thus.

In most instances.. This also appears to explain the quite limited ability of loyalty programs to increase brand loyalty dramatically (Dowling and Uncles. the different measures are all predictable from each other (i. the normal split loyalty patterns tend to persist. In this sense. Roquebert et al. and even normal hair. It seems that sustainable product differentiation of a major kind seldom occurs between competitive brands. 1997. consumers spend more time. 5. 1996. 1997. decaffeinated and regular coffee.g. shampoos for oily. 1997a. and in practice it usually does so. But in the Dirichlet. Nor is there evidence that brands that supposedly have high consumer-based equity do subsequently get bigger (Ehrenberg. 1990).S. they measure the same thing).. as SKU-related product variants: different pack sizes and flavors of soft drinks. cars having 2.C. / Journal of Business Research 57 (2004) 1307–1325 1317 identical in the theory itself—no differentiating attributes need to be specified except for the brand names and market shares. But both in theory and in practice. Aaker. While there clearly are big brands and smaller ones. the customer is deemed ‘‘loyal’’ but buys infrequently). 1996).1. any consumer generally shows loyalty to certain specific ones. Szymanski et al. as we have seen. 5. 1997. Moran. Foxall.. and some may appear almost meaningless (Carpenter et al. This is so even if the brand’s customers routinely buy other brands as well (‘‘split loyalty’’) and if the frequency of buying that brand is low (i. 1997.. but having a repertoire enables consumers to exercise some choice without having to reevaluate all brands on all their criteria or without somehow weighing up the expected ‘‘utilities’’ at each new purchase. This is reflected in the Dirichlet model where consumers are deemed ‘‘loyal’’ to the brand if they go on buying it... 1999. some functionally distinct submarkets or partitions have been deliberately created (e. The reason for consumers’ persistent polygamy may be some basic desire for ongoing variety. consumers in most countries will hardly know from which company the brands and variants come). or degrees of sweetness for mueslis. consumers then simply stay with some of them (e. Brand loyalty Although competitive brands tend to be similar..g. 1999). 2001). 1993a). 4. It could also be that having initially tried out many different brands.3. this does not seem to happen. dry. there is no evidence—from a consumer behavior viewpoint—that over and above this there are ‘‘strong’’ brands and ‘‘weak’’ ones (Ehrenberg. Goodhardt. 1999). They would just have had to buy more Folgers and less of the other similar brands—insofar as competing brands are perceived to be similar. Ehrenberg et al. direct competitors will then each launch product variants into these submarkets. unleaded and leaded gasoline). This is often so in practice since sales-effective product advantages and innovations are usually soon copied (e. 1994). which is usually small). 1994). But there is equally a lack of well-grounded reasons for consumers sticking with just one of several near lookalikes. Furthermore. but they may affect the . But competitive brands in a category tend to have much the same range of product variants. Theorists may feel that there is a lack of well-grounded reasons for consumers having steady repertoires of similar brands. Functional differentiation does however exist but usually within brands.. 2003). Our inference is this: staying with a brand repertoire requires less mental effort than making new choices. Feldwick.e. Loyalty is therefore a given: changes in sales do not—or hardly—arise from changes in loyalty but from changes in the sheer number of loyal customers. and distribution can have little if any impact on increasing loyalty. each measure tends to be similar for competing brands (apart from the market-share-based DJ phenomenon. car door handles. Instead. 1995). 3. or 5 doors. 1999. luxury cars. Many of these differences are not featured ‘‘on-pack’’ nor advertised. Uncles et al. 1976). and so on. In many product categories. the annual SCRs of a brand like Folgers was only about 50% and even lower in Table 3. such as different bottle tops. Sharp and Sharp.e. nearly all brands are umbrella brands and similar to each other as brands. The role of marketing-mix factors A major lesson of the Dirichlet-type findings is that brands often differ little in their loyalty-related measures but vary greatly in their penetrations. selling. Loyalty has over the years been measured in many different ways (for a recent review. As we have noted. such minor and nonpublicized differences may be noticed after consumers have started using the brand and may lead to longer-term brand preferences for otherwise similar brands. see Odin et al.. This means that varying marketing-mix inputs such as changes in price. Ehrenberg et al. and money choosing between functional specifications than between brands as such (Heath..A. Sometimes these product variants are given distinct brand names (possibly with the same house name such as Kellogg’s or Heinz for all of the variants. product formulation. This Dirichlet view of brand loyalty is out of kilter with the widely held view that ‘‘brand equity’’ is an idiosyncratic property of an individual brand (e. what we call ‘‘minor differences’’ between competing brands occur in practice. interest rates varying by withdrawal notice. Folgers’ customers could therefore have bought the brand up to twice as often without having had to drink more coffee. 1996. as is supposed in ‘‘rational economics’’ for example (see also Thaler. Charlton and Ehrenberg. However. effort.g. although not always—for Unilever and P&G brands.. Even early mover advantages tend not to last (Chiang and Robinson.. The nonpartitioned Dirichlet model can then apply directly.g. Nonetheless. An apparent paradox is that in Table 2.2. PCs with a great variety of technical specifications and software add-ons.

Tim Bock. mostly by reminding experienced consumers—‘‘Here I am. The outcome is therefore usually competitive equilibrium rather than actual big gains or losses. Appendix A. with profitable survival being preferable to the most likely alternative. how loyal they are. 5. consumers are seen to have. traditional loyalty-related measures for competitive brands are unaffected by any marketing-mix inputs. But driving at more or less steady speeds accounts for most of the distance covered. Hence. 1984). and in particular its market share and sales volume. consumers seem then to find it convenient and reassuring to have already developed their habitual split-loyalty choice propensities. we believe. In practice. the model consistently predicts fairly complex patterns and BPMs from very simple and limited inputs. / Journal of Business Research 57 (2004) 1307–1325 brand’s penetration. The observed outcomes of price-related promotions support this view. 1994a). Consumers just have to choose one of the available options of the right type: ‘‘Any reasonable brand will do’’ (Heath. 1974. namely. Jedidi et al. advertising does not.g. are marketing inputs such as advertising worth it if they do not soon lead to extra sales? Our view is that there is nonetheless plenty of scope for marketing (e. In this sense. it in no way implies or predicts that all markets should be near stationary and nonpartitioned. remember me’’—often in highly creative ways for impact and memorability (Ehrenberg. It also helps to explain the patterns. John Scriven. Ban Mittal.g. together with most markets being more or less steady most of the time. Robert East.1. the bulk of it occurs between competitive brands. Kathy Hammond. Two assumptions are about buying the product category: (i) A Gamma distribution for consumers’ differing average purchase rates Each consumer is assumed to buy the category at some steady long-run rate (e. In this. Acknowledgements We are greatly indebted to Helen Bloom for much valued advice and also to Byron Sharp.. once a year or 10 times a 6. it behooves the management to try and understand such steady markets and the factors that determine their sales. competition means running hard to stand still. These factors are how many customers a brand has. and (4) alternative models of buyer behavior. that under near-steady state conditions. Cam Rungie... 1987. This calls for maintenance of sales and retention of split-loyalty customers (both consumer and trade). But the blip stems almost entirely from past customers—these only need to change their quasi-random purchase timing of the promoted brand rather than more fundamentally switch to an unfamiliar brand (Ehrenberg et al. (2) how the model is calibrated. For example. 1999). Ehrenberg et al.1318 A. The NBD-Dirichlet assumptions In the NBD-Dirichlet model (or ‘‘Dirichlet’’ for short). . 1997. Marketing management issues The ubiquity of the predictable Dirichlet-type patterns of buying. and anonymous reviewers for helpful comments. Barnard and Ehrenberg. have to try to persuade consumers that Brand A really differs from the similar B. Dynamics matter: we have to start and stop. An analogy for the prime importance of the near-steady state in the Dirichlet view of markets lies in our use of automobiles. However. Large short-term sales blips (often 100% up or more) show that promotions have a dramatically ability to attract extra buyers. the Dirichlet model describes the widely observed patterns of near-steady state buying behavior that tend to occur. This involves five distributional assumptions (Goodhardt et al. A. 2002). Similarly. as noted earlier. steady personal purchase propensities—or stochastic probabilities—for when they buy the product and what brands they then choose. Hence..S. together with the Dirichlet type of constraints on all this. near-steady state markets generate the bulk of our sales and therefore they are likely to account for most of our revenue. for the time being. could add to existing doubts about the role of marketing. Any dramatic and lasting gain is a rare bonus. as we have already stressed. which competitive brands customers also buy. John Bound. it need not be surprising that promotional sales blips last only while ‘‘giving the product away’’ and do not affect repeat buying loyalty thereafter (Abraham and Lodish. such scope exists also for competitors. or even that it is better than B. While the Dirichlet model itself is explicitly defined for steady-state and nonpartitioned zero-order markets. Rachel Kennedy. between which there is usually little sustainable differentiation.C.4. 1999). Ehrenberg et al. Neil Barnard.’’ But since many markets or submarkets are like that most of the time. other than indirectly through changes in the brand’s market share.. The Dirichlet theory We discuss: (1) the theoretical assumptions of the Dirichlet.. For brand advertising. securing better distribution) since market shares differ greatly. The model merely describes what markets are like when they are more or less steady and nonpartitioned with no ‘‘purchase feedback. Conclusion To conclude. Advertising an established brand therefore must mainly publicize the brand. (3) how the theoretical performance measures are estimated. and how often they do so.

6. However. Schmittlein et al. consumers have ‘‘reasons’’ but behave as-if-random with their personal choice probabilities). The NBD has been used widely in repeat-buying studies where each brand is considered on its own. unless the mean is very high. Chatfield and Goodhardt. 0. a consumer does not always choose the brand with a utility of 0. The third and fourth Dirichlet assumptions are about brand choice: (iii) A multivariate Beta distribution of brand choice probabilities..g.g.g. Then any brand’s theoretical performance measures have to be estimated. Like the 80:20 rule of thumb.02 or 0.. Dirichlet..e. the Gamma invariably means there are many light buyers and few heavy ones. Luce. Morrison and Schmittlein.. In aggregate. This is in line with observed market shares being much the same for light. medium.. the parameters of the model must next be numerically calibrated. and 0. Thus. two penetrations and two purchase rates.02 or 0. This is the widely used zero-order ‘‘multinomial’’ distribution of brand choice (typically.C. about 0. say week-by-week or quarter-byquarter. the probability of buying Brand A on a particular occasion is independent (‘‘zero order’’) of the brands the consumer has previously bought. 1985. 1985). Rungie. Consumer heterogeneity in these individual average-buying rates is assumed to follow a smooth ‘‘Gamma’’ type of distribution (e. Strong theoretical backing for this assumption derives from the observed near independence of brand choice in line with the traditional theoretical notion of ‘‘independence from irrelevant alternatives’’ or IIA (Goodhardt et al. The zero-order assumption makes simple sense with the typically low weekly purchase probabilities: even for someone buying 10 times a year.. the NBD is used for the category distribution of purchases. An equivalent specification is used in choice models (e. if the deterministic part of the utilities are in the ratio of 0... . i. This reflects what the observed data tend to be like when tabulated. The Beta distributions of brand choice probabilities are taken to be the same irrespective of how often particular consumers buy. 1959. There are strong theoretical reasons for the Gamma relating to the observed near independence of buying the different brands (Goodhardt and Chatfield. Crucially. The deterministic part of the choice model might suggest that utility for Brand A is higher than the utility for other brands. 1986). a consumer is assumed to choose a brand as if randomly with their own fixed brand choice probabilities.. Ehrenberg et al. an individual’s purchase rate can change at times. and heavier category buyers.2. 1959). and 0. The model represents this as a gvariate discrete random variable with a joint frequency distribution given by the mixture of the Multinomial. (iv) Multinomial distributions for specific purchases On any one category purchase occasion. and 10% of the time in a random order. 30%. Ben-Akiva and Lerman. 1988. the three brands are bought 60%. Calibrating the model To calibrate the Dirichlet model for a given product or service category. Leeflang et al. / Journal of Business Research 57 (2004) 1307–1325 1319 year. 1988).S. perhaps with a change in lifestyle or in the marketing mix.258. The Poisson and Gamma assumptions combine to give the negative binomial distribution or ‘‘NBD’’ model (Ehrenberg. p.A. 1973.3. as has been widely tested for brand purchasing (Bass et al. 1988). Dunn et al. Having specified the model by these distributional assumptions. 30%.04 and hence near zero—no special assumption of ‘‘dead periods’’ between purchases has to be made. 1959. with weekly probabilities of about 0. Heterogeneity in consumers’ brand choice probabilities is assumed to follow a smooth Beta distribution of a multivariate ‘‘Dirichlet’’ type (after a French-named German mathematician).6. 1983. and 10% of the time. but the random error terms in the utilities can result in a consumer choosing a brand other than A. This we now describe in two stages. 1959. the Dirichlet model specifies how many purchases each household (or individual consumer) makes of each of the available (or specified) g brands in a chosen period of length T (not the T for the theoretical Dirichlet predictions).6. 2000. Luce. McFadden.20).1 will buy the three brands 60%. A. The fifth Dirichlet assumption is about the relationship between product category buying and brand choice: (v) The independence of purchase incidence and brand choice.20) and independently of when the previous purchase was made (a ‘‘zeroorder’’ process). In summary. (ii) Poisson distributions for individual purchases A given consumer’s specific purchases of the category are assumed to be spread irregularly (‘‘asif-randomly’’) over time with each consumer’s longrun probability (e. In the multibrand Dirichlet model. 1984. Ehrenberg. the chance of buying in two successive weeks is only 0. Poisson. 1973). four observed input measures are used. and Gamma distributions (e. Ehrenberg.g.3. 1984. as is shown in Table A1. Thus. 0.. 1999). a consumer with a three-brand portfolio and zero-order probabilities of 0.1 for a three-brand portfolio. This specifies ‘‘Poisson’’ distributions for each consumer.

3.1320 A. the penetration of one particular brand. 2000. 2. It is also possible to simulate individual Dirichlet-type consumers and then tabulate any required measure. i.  W. . Here: K þ n À 1 ½Sð1 À zx Þ þ n À 1Š A Cn ¼ CnÀ1 ðS þ n À 1Þ 1þA n where K. If the model fits the observed data exactly.g. Hewitt.. the average frequency of buying the category per category buyer (5. the average frequency of buying Folgers (3. A. but software is available (e.2).4.e. the inputs are as follows:  B.. Rungie et al. the unusual Maxim in Table 1 or brands with abnormal availability like a private label or regional brand).0 3.C. there are deviations. but as percent numerically). In practice.3 11 A brand’s market share (or equivalent) is needed to estimate its theoretical performance measures. such as a year. 2002). The algebra for the model’s theoretical estimate of a chosen performance measure is again complex. such as a particular brand’s penetration or its average purchase frequency. It requires first calculating SCn. households who bought the category of instant coffee at least once in the year. one can use computer software (contained in the same packages as for the model calibration) to give numerical estimates such as those in Tables 1 and 5 earlier.S. as one would with observed data (Goodhardt. how many consumers do not buy X but do buy the category any number of times. But their estimated performance measures would still be checked subsequently. (We note that penetrations are usually expressed as proportions algebraically. as is illustrated in Table A2.. and S are the parameters of the fitted Dirichlet model. any unusual brand can be left out of the calibration process (e. any brand could be used in this calibration process and would give the same results. In particular.  wF .’’ we said that far fewer people buy a small brand and buy it somewhat less often (this is like the wording commonly used to describe the normal distribution Table A2 The Dirichlet’s theoretical formulae for the penetration An illustration The Dirichlet theoretical formulae for the penetration of Brand X with market share zx is indirect. Simplifying approximations Alternatively. Thus.  bF . which are not covered by the available estimation software. for ‘‘DJ. 1989. the category penetration (i. It is then usual to fit the model for each of a number of leading brands and to average or smooth the results (in the computer program). The mathematics of this calibration process are fairly complex. and 5) Instant coffee (USA 1992) Market share Percent buying in year O Category Folgers Maxwell House Tasters Choice ´ Nescafe Sanka High Point Maxim Brim Average Brand a Purchases per buyer Percent buying 5+ times O T Category purchases per buyer O T 100% loyal Percentage who also bought: Folgers T O 5.0). A.g. simple verbal descriptions can be used to communicate what the theoretical estimators say—like those in Section 2 for the observed patterns in the data. Folgers say (11%). A method based on using the Juster scale to replace panel data has been described too (Wright et al. These purchase probabilities are aggregated in the model to give a theoretical estimate or ‘‘prediction’’ of any chosen performance measure. / Journal of Business Research 57 (2004) 1307–1325 Table A1 The inputs for calibrating the Dirichlet parameters (as required for Tables 1. 31%).. This can be useful for theoretical measures. 1990. each possible purchase made by any consumer is implicitly specified as a ‘‘stochastic’’ or probabilistic representation. Estimating the theoretical BPMs By calibrating the model. 2003). The selection of brands to use in such model calibration can be deliberate. For some chosen ‘‘base period’’ (which will be greater than the average interpurchase interval). Kearns. Uncles.e.2 T O T O T ´ Nescafe O T Maxim O T 31 24a 22 17 11 9 1 1 0. 1995). In practice.. Ehrenberg et al. A..

9/ 4. b/B) probably leads to further simplifications.6 4.g.1 3. These are much simpler than the exact Dirichlet expressions and therefore more insightful. 1966).8. and (ii) the ‘‘Duplication Law’’ relation bx/ySDbx (this relates the proportion bx/y of buyers of Brand Y who also buy Brand X in the analysis period.2 8 6 T 12 11 9 6 5 0 0. T = theoretical Dirichlet predictions. 1959.3a 0.1 3. there is some scope to improve goodness of fit. Ehrenberg. 1980. for bs of 10% or less as in Table A3.7 3. First. a semiparametric random effects model was found to capture heterogeneity in brand preferences across households. But the base B then varies with the length of the analysis period.0 3. 1988. simply proportional to bx.. Furthermore. the duplication coefficients D would tend to be close to 1 (implying independent brand choice.8 3. the smaller the b (as a proportion).0 2. A. 1993. where we say ‘‘5% lie beyond F 2 standard deviations’’ rather than spell out the pdf). Remarkably. The approximate formula predicts well. It appears that with these alternative models. but the predictive gains were slight and came at the expense of added complexity (Chintagunta et al. Reader and Uncles. or ‘‘duplication of purchase law. is Table A3 Exact and approximate estimates of the average purchase frequencies Instant coffee (USA 1992) Market share Percent buying O Folgers Maxwell House Tasters Choice ´ Nescafe Sanka High Point Maxim Brim Other Brands Average brand 23 22 17 11 9 1 0. More work is needed..1 2. 1996).5.5a 2. as illustrated in the final column of Table A3.0 T 3.1 3.9 Estimated duplications of purchase for Folgers Instant Coffee as in Table 2 = D Â Penetration.8 2. Jones and Zufryden.8 2. Only for brands with high penetration are the differences in average purchase frequencies quite large. Colombo and Morrison.0 Table A4 The duplication of purchase law: an illustration Instant coffee (USA. a D is estimated as (average duplication)/(average penetration) = 13.1. In a one-off study of saltine crackers.6 3. case-specific ‘‘mass points’’ have been used to model consumer heterogeneity rather than the Dirichlet’s smooth Gamma and Beta distributions (e. Thus. / Journal of Business Research 57 (2004) 1307–1325 1321 in statistics. Bass. 1992) Percentage who also bought: Average ´ Maxwell Nescafe Sanka Maxim Brim (8 brands) House 21 17 15 12 14 13 0 1 1 1 1 0 13.6 2. 1968b. Goodhardt. a constant for that set of brands and chosen time period (this relates Brand X’s average purchase frequency wx directly to its penetration bx). Relatively high penetrations can occur both for large brands.5 8 6 Average purchases (per buyer) O 3. 1988. The two most used simplifications are: (i) the ‘‘DJ’’ relation wx(1 À bx)Swo. Fader. but only by forsaking parsimony and generalizability.. consumer heterogeneity has been related to various possible causal sources such as sociodemographic and attitudinal factors (Allenby and Lenk.4. 1991).8 penetrationa % 28 Exact Dirichlet % 27 estimate Penetration % 10 6 5 0.0 2.9 A 3. the percentage of buyers of Brand Y who also buy X in the chosen analysis-period. Nor so far have these alternatives reported the same range of BPMs that can be derived routinely from the Dirichlet.C. the average purchase rates hardly differ (e. .9 2.9 = 2.8 2. This approximate relationship. 1996. Elaborating the Dirichlet model The Dirichlet and related stochastic approaches to modeling buyer behavior have been elaborated in two main ways. Ehrenberg. 1994.8 2.6 2. 1993. A. the ws are all about 3).2. Thus. 1989.. Expressing brand penetration in ‘‘relative’’ terms as the percentage of category buyers who buy the brand (i. Fader and Hardie. 1988). a very simple result). 1993. A = wo/ (1 À b). The duplication D is very simple to estimate as the average of the observed duplications for all pairs of brands (leaving out any extreme outlier) divided by their average penetration. to X’s overall penetration bx. Ehrenberg et al.2 3. A. and where D is defined below).1 3. For brands with fairly low penetration.S.0 3.g.g.5 Folgers buyers % 31 2.4. The ‘‘w(1Àb)=constant’’ approximation This says that w varies with 1/(1 À b). Bhattacharya et al.2 4. This is McPhee’s DJ notion that was noted earlier (but is not readily apparent from the exact but complex Dirichlet-type algebra such as in Table A2 where wx and bx/y would depend on all the observations for all the brands and not just on bx) (Ehrenberg. The ‘‘bx/y=Dbx’’ approximation This says that bx/y .1 3.e. several close quantitative approximations to the complex Dirichlet-type estimators are also available.2 3. 1970.3 0. Ehrenberg and Goodhardt. Fader and Lattin.9 0.9 2. the smaller is 1/(1 À b) and hence the smaller is w.0 2. the w(1 À b) formula copes with both cases.3 16 11 11 10 9 6 5 1 0. Brand X’s penetration in the whole population (e.9 12.. Russell O = Observed as in Table 1. a Outlier.0 2. Second.A.. and for small ones over long time-periods.2 0.’’ is illustrated in Table A4 for the duplication of purchase of Folgers with the other instant coffee brands.9 13.

Leeflang and Wittink... Barnard NR.. Wagner and Taudes. Guadagni and Little. Ehrenberg et al.37(1):21 – 30 [January/February]. London: Sage. Leeflang et al. Leeflang et al. Mark Sci 1987. 1998.1. Lodish LM. Consumer beliefs and brand usage. Many aim to predict changes in market shares from changing marketing-mix inputs (e. Lilien et al. Thaler. 2002. advertising or price).. In this.e..6. This uses the Duplication of Purchase Law formulation for pairs of purchases.g. 1985). and its underlying descriptive and benchmarking uses. 1962.. 1994). Erdem. which seem largely unresolved. Another purely structural model is the Hendry system (Butler. they are generally attempting to model nonstationary market conditions (e.. The future of research in marketing: marketing science. Ehrenberg ASC. J Am Stat Assoc 1994.g. Why brands grow. Echambadi R. Wrigley and Dunn. advertising or price changes).. Barnard NR. 1970) and has occasionally been mentioned ever since (e. The parameters of such econometric-type models are mostly respecified on each occasion. 1993b)..g. with no explicit ‘‘explanatory’’ variables specified (e. Ehrenberg ASC. Barwise PT.. A variety of factors have proven to be statistically significant in isolated studies. 1991..g.. But it makes very different assumptions from the Dirichlet and has different outcomes (e. It is seldom that a modeling theory and the facts clash so starkly.1322 A. Allenby GM. But they seldom touch on the issues of brand choice in highly competitive markets. 1994). 1993. Television and its audience. Such models usually require many—possibly hundreds—of parameters to be estimated and interpreted. In contrast. Massy et al. 1990. 1995. 1996). Kagel and Roth. (1976) did not refer to any empirical patterns. Ehrenberg and Goodhardt. Promoter: an automated promotion evaluation system. Lenk et al. ‘‘pairs of purchases’’ can however be used very constructively in the absence of continuous panel data (e. 1992).g. J Mark Res Soc 1985. Barwise PT. Ehrenberg ASC. 1974).S. Kannan and Yim.g.. Ehrenberg and Bound. A. More generally. Some of the models assume turbulent (i. Baldinger AL..g.. Ehrenberg ASC. 1998. Advertising: strongly persuasive or nudging? J Advert Res 1997. References Aaker DA. but systematic effects and generalizations have eluded analysts. as we have stressed. 2000). or do gains and losses operate quite differently from that? These are empirical questions for which it seems to us useful to know whether some answers generalize before theorizing extensively about them. Colombo et al. do any sales increases for a given brand follow the steady-state DJ pattern for large and small brands (implying mainly penetration growth rather than separate changes in buying rates or loyalty)? Do sales decreases simply show the reverse pattern from sales increases or are they more complex? Do increased sales for Brand X come from competitive brands pro rata to their size (as in the steady-state duplication of purchase law). 1996.g. 1983. Economists and psychologists such as Kahneman and Tversky have carried out very innovative research into broader issues of consumer choice. Specific results have been reported but with few generalizable findings and insights. 1994. New York: Free Press. 2000. 1996.98:1218 – 31. Modeling household purchase behavior with logistic nominal regression. Vilcassim and Jain.C. / Journal of Business Research 57 (2004) 1307–1325 and Kamakura. Other structural models The Dirichlet model is purely structural for steady-state markets.g. 1966) and the derivatives of it (e.. Lenk PJ. A. The alternative stochastic and/or econometric response models for buyer behavior in the literature are mostly dynamic. Moreover.g. Bennett. 2000.42(1):6 – 14 [January/February]. continually changing) choice probabilities for each consumer (e. For example. Kannan and Sanchez. Bronnenberg. differs from many other modeling approaches. Nearly all such modeling appears to be aimed at discovering and predicting effects that have not yet been observed. Other modeling approaches The Dirichlet model. Blair E. .. R&DI Report..30(1):1 – 6 [February]. An alternative structural model is the first-order Markov process. Even the early theoretical derivation of the Dirichlet model by Bass et al. In its simplest form. It is now known that (i) consumers are heterogeneous and (ii) that switching and repeat buying are independent of the brands as such but vary with market shares. Bass FM. These assumptions run counter to the Dirichlet-type empirical findings (and to much of the response modeling literature more generally). multicollinearity). Building strong brands. the Dirichlet-type approach first establishes what kinds of empirical changes have in fact taken place and then tries to model that knowledge (Ehrenberg. Indeed. Yim and Kannan.6(2):101 – 23 [Spring].27:81 – 93. the Markov approach also has no ‘‘causal’’ inputs but assumes (i) homogeneous consumers and (ii) fixed switching and repeat-purchase probabilities for each brand. 1999. J Advert Res 2002. these other approaches tend not to mention any of the ubiquitous empirical patterns reported in Section 2 or their theoretical modeling as described in Section 3 (see for example reviews by Bucklin and Gupta. South Bank University. 2000. Consumer attitudes and brand choice. 1999). 1991. Kuehn. with conceptual and at times computational complexities (e. 2001.g. 2000). This was popular in the 1960s and 1970s (e. Erdem and Keane. Abraham MM..6. covered by standard scanner panel tabulations and the Dirichlet approach. few major gains in predictive power have been either reported or even claimed for these more elaborate model specifications. 1986. Vilcassim and Jain. J Mark Res 1993. often along experimental and/or mathematical lines and frequently with roots in game theory (e. 2000. Hanssens et al.

The incidence of brand-switching. A brand-switching model with implications for marketing strategies. Barnard NR. Ehrenberg ASC. Fader PS. Buyer behavior under mini-test conditions. p. Croton-on-Hudson: Hendry Corporation. or not B. Colombo RA. Collins MA. R&DI Report. Kennedy R. p. Lerman SR.14(5):421 – 35. Int J Res Mark 2002.10:232 – 8. Turning data into knowledge. Kalwani MU. 1995. J Advert Res 1974. Goodhardt GJ. Miniard PW. Investigating heterogeneity in brand preferences in logit models for panel – 20. J Mark Res 1975. University of Derby. Ehrenberg ASC. Bennett D. Dall’Olmo Riley F.E? Proceedings of Advertising Research Federation 43rd Annual Conference. Omega 1980. Butler BF. Nakamoto K. Srivastava RK. Chatfield C. Do customer loyalty programs really work? Sloan Manage Rev 1997. Double jeopardy revisited. Ehrenberg ASC. Discrete choice analysis.7:77 – 84. Marketing research: state of the art perspectives (the Handbook of the American Marketing Association and the Professional Market Research Society).empgens. Goodhardt GJ. Ehrenberg ASC. Pras B. Bhattacharya CB. Chiang J. Ehrenberg ASC. April 7 – 9. The erosion of repeat purchase loyalty. Ehrenberg ASC. A comparison of online and offline consumer brand loyalty. or just right? Explaining deviations in loyalty from the Dirichlet norm. Goodhardt GJ. Goodhardt GJ. why aren’t you bigger? Admap 1993a. 1999.15(4):359 – 78. Chintagunta PK. Admap 1997. Market segmentation—the realities of buyer behavior. 28:417 – 28 [November]. London: Edward Arnold. 1997. Carrie D. Bound JA. Bloom H. Barwise PT. Carpenter GS. Hammond K. J Mark Res 1970.43:8 – 19 [Fall]. Danaher PJ. A taxonomy of differences between consumers for market segmentation. Press: 1988. Uncles MD.13:52 – 60. Int J Res Mark 1997. Mark Sci 1996. Gupta S.54:82 – 91. Bronnenberg BJ. Ehrenberg ASC. Barnard N. New York. Vilcassim NJ. The after-effects of pricerelated consumer promotions. B. Chatfield C. Ehrenberg ASC. Blackwell RD. Hammond K.C. J Mark Res 1976. Commercial use of UPC scanner data: industry and academic perspectives. Cranfield (UK): European Case Clearing House. Dowling GR. Jain DC. Lilien G. Some statistical models for buying behavior. Erdem T. Cambridge (MA): Marketing Science Institute. London: Prentice-Hall. Dordrecht (The Netherlands): Kluwer. Stochastic models of industrial buying behaviour. Fader PS. Ehrenberg et al. An empirical analysis of buying behavior in UK high street womenswear retailing using the Dirichlet model. Manage Sci 1976. Is your brand’s loyalty too much. Bennett D.19(3):215 – 24. Brand loyalty among petrol buyers. theory and applications.5:392 – 770. Ehrenberg ASC. Mark Sci [in press].31(3):339 – 50. Castleberry SB. Repeat-buying: facts.10(1):41 – 57. Orlando (FL): Dryden Press.220:304. Robinson WT. Davis R. / Journal of Business Research 57 (2004) 1307–1325 Bass FM. Appl Stat 1959.14 (5):437 – 50. Ehrenberg ASC. London University: Doctoral Thesis. Ehrenberg ASC. Can J Mark Res 2000. 2002. 1997. Research traditions in marketing. J Advert Res 1997a.8:89 – 99.14:171 – 83. Advertising and price. Barwise PT. 1997. J Mark Res Soc 1991. Brewis-Levie M. 1993b. Decision models or descriptive models.32(3):249 – 59. Differentiation or salience. Equilibrium stochastic choice and market penetration theories: derivation and comparisons. Wright GP. Reibstein D. Goodhardt GJ.17(2 – 3):147 – 58. Krannert School. Mark Lett 1993. Ehrenberg ASC. Available: http://www.14:25 – 34. Ehrenberg ASC.A. Ehrenberg ASC. Barnard NR. Purdue. Repeat-buying of a new brand. Bucklin R. Glazer R. J Mark Manage 2000. Ehrenberg ASC. Chatfield C. Uncles MD. Bass FM. The variability of attitudinal repeat-rates.22: 1051 – 63 [June]. Ben-Akiva M. Darius S. Nature 1968b. Ehrenberg ASC. Admap 1974. A consumer purchasing model with Erlang inter-purchase times. DeSarbo W. Ehrenberg ASC.8: 26 – 41. too little. editors. An investigation of the assumptions of the 1323 NBD model as applied to purchasing at individual stores. Diversity in analysing brandswitching tables: the car challenge. Ehrenberg ASC. An investigation into the order of the brand choice process.38(4):71 – 82 [Summer]. Ehrenberg ASC. Repetitive advertising and the consumer. Wilson IW. July 4. If you’re so strong. Barnard N. Decision-making under uncertainty: capturing dynamic brand choice processes in turbulent consumer goods markets. Charlton P. Academy of Marketing Doctoral Colloquium. Mark Sci 1989. Jeuland AP. Using two-purchase data. East R. The Hendry brand-switching coefficient.37(3):27 – 35 [May – June]. J Mark Res 1991. New brands: near-instant loyalty. Armed to the teeth: an exercise in brand management. The pattern of consumer purchases. Bock T. South Bank University.42(6):7 – 18 [July/August]. Meaningful brands from meaningless differentiation: the dependence on irrelevant attributes.10(1):9 – 16. A dynamic analysis of market structure based on panel data. 1966. Working paper. East R. Lodish LM. Goodhardt GJ.16(6):607 – 17 [Revised version] J Target Meas Anal Mark 2001. Keane MP. Goodhardt GJ. Morrison DG. J Advert Res 1997b. Day GS. 13 – 4 [October]. New brands and the existing market. Int J Res Mark 2000. Wrigley N. J Mark 1979. J Mark Res 1998.7(1):5 – 18. . Harris P.13:146 – 57. Erdem T. New York: Oxford Univ. Int J Res Mark 1997. Bhattacharya CB. 2000. J Am Stat Assoc 1973. Engle JF. 8:63 – 9. Ehrenberg ASC.37(6):7 – 14 [November/December]. J Advert Res 1994a. Givon M. Goodhardt GJ. Customer-oriented approaches to identifying product-markets. Easton G.33(4):285 – 99 [October]. In: Chakrapani C.2:200 – 5. Consumer behavior 8th edition. J Advert Res 2002. A model of multi-brand buying. Colombo RA. 2nd ed. Hendrodynamics: fundamental laws of consumer dynamics. Pymont B.4:99 – 112 [April]. Ehrenberg ASC. 23 – 46. Reader S. Ehrenberg ASC. 79 – 108. Theory or well-based results: which comes first? In: Laurent G. Mark Lett 1996. Goodhardt GJ. Br J Mark 1968a. Shocker AD. 1967. Bound JA. Barnard NR.3 (4):267 – 87 [Fall].S. Private label purchasing. J Mark Res Soc 1971.E. J Mark 1990. Mark Lett 1996. J Mark Res Soc 1972.34 (4):11 – 21 [July/ August]. Mark Sci 1996. Ehrenberg ASC. Mark Sci 1984.12(1):110 – 3. 2000.7(2):163 – 71 [March]. 1994b [reference M94-005:594-039-1/594-039-4/594-040-1/ 594-040-4]. Dunn R. The relationship between the marketing mix and share of category requirements. Ehrenberg ASC. Cambridge (MA): MIT Press. Integrating the Dirichlet-multinomial and multinomial logit models of brand choice. Int Rev Retail Distrib Consum Res 1999. editor. Sharp B. Advertising frequency decisions in a discrete Markov process under a budget constraint. Wright GP. Results concerning brand-choice. Chicago (IL): American Marketing Association. Charlton P.375: 17 – 9. Scriven JA. An experiment in brand-choice.19:23 – 36. Consumer behavior: advances and applications in marketing. Brand advertising as creative publicity. Uncles MD. Reprinted in the J Empir Generalisations Mark Sci 2000. Ehrenberg ASC. J R Stat Soc Ser C Appl Stat 1983. Do market pioneers maintain their innovative spark over time? Working paper.35:399 – 406 [August]. 1985. Scriven J.68:828 – 35. J Mark Res 1994.

18:190. Phillips RL. Kau KA.17: 323 – 34. Kau KA. Mark Sci 1995. J Mark Res 1984. J Mark Res 1980. The measurement of cannibalization. Horowitz J. Ehrenberg ASC. Manage Decis Econ 1999.6.6:145 – 59 [April].25:384 – 90 [November]. Zufryden F. Henley-upon-Thames (UK): NTC Publications. 1997. Mark Intell Plann 1996. Montgomery DB. Goodhardt GJ. 30:9 – 16 [October]. Nature 1973. Princeton: Princeton Univ Press. 2000. a speculation. Goodhardt GJ.17:653 – 64 [October]. McAlister L. DIPRED: a programme for predicting buyer behavior from the Dirichlet model. Chatfield C. Consumer decision making-fact or fiction? J Consum Res 1979. / Journal of Business Research 57 (2004) 1307–1325 Kannan PK. Accounting for heterogeneity and non-stationarity in a cross-section model of consumer purchase behavior. Rao AG. McPhee WN. Luce RD. The choice theory approach to market research. Kahn BE. Jedidi K. Englewood Cliffs (NJ): Prentice-Hall. 1970. 2000. South Bank University. Stochastic models of buyer behavior. London Business School. Fader PS.5:275 – 97. An empirical study of buying behaviour in an industrial market. Uncles MD.30:478 – 93 [November]. 1998. Schmittlein DC. 1994. Moe WW. Ehrenberg ASC. Personal buying is like family shopping. Fader PS. Wittink DR. editor. 1999. Market response models: econometric and time series analysis. R&D I Report. Roquebert JA. Fournier S. Collins M. Dirichlet no solver software. Kannan PK. A logit model of brand choice calibrated on scanner data. Mark Sci 1983. University of New South Wales. Kalwani MU.S. Glencoe (NY): Free Press. Farris PW. J R Stat Soc A 1984.1(2):10 – 7. J Mark Res 1988. An investigation of brand choice in repeat purchase markets: the case of business airline travel. Formal theories of mass behavior.17(2 – 3):105 – 26. London University. Int J Res Mark 2000. Learner]. Ehrenberg ASC. 45 – 57. Valette-Florence P. Building models for marketing decisions: past. Taking brand repertoires seriously. Gupta S. Kennedy R. J Mark Res 1996. Schmittlein DC. 1995. University of Derby.21: 399 – 409.12:39 – 54 [May]. Hanssens DM. Nonstationary conditional trend analysis: an application to scanner panel data. Admap 1999. 14(5):451 – 72. J Advert Res 1990. East R. Variety seeking behavior: an interdisciplinary review. 1996 [Supplementary tabulations with David E. R&DI Report. Morrison DG. The PC market’s retention and switching. Uncles MD. R&DI Report. Hardie BG. Lattin JM. Doctoral Thesis. J Bus Res 2001. Ehrenberg ASC. 1994. South Bank University. Reibstein DJ. Hardie B. Ehrenberg ASC. Naert P. J Brand Manage 1994. Advertising effectiveness. Strong and weak brands. Massey University. Handbook of experimental economics.244:316. Mela CF. Goodhardt GJ. Clemente M.1324 A. Conditional trend analysis: a breakdown by initial purchasing level. What is the role of consideration sets in choice modeling? Int J Res Mark 1995. Generalizing the NBD model for customer purchases: what are the implications and is it worth the effort? J Bus Econ Stat 1988. J Advert Res 1962. Schultz RL. Longitudinal data analysis: methods and applications.C. Kuehn A. Conditional trend analysis: a model that allows for nonusers. Conceptual and operational aspects of brand loyalty: an empirical investigation. Wittink DR. The substitutability of brands. Goodhardt GJ.34(3):14 – 7 [March]. 1959. Market share and distribution: a generalization. Mark Sci 1986. Consumer psychology for marketing. J Mark Res 1993. Version 2. Westfall P. J Mark Res 1993. Roth AE. 1990. New Zealand. Eur J Mark 1996.9:311 – 22 [December].4(1): 9 – 28.17:209 – 29 [June – July]. Louviere J. Chatfield C. J Mark Res 1969. Hallberg G. July. Mark Sci 1993. J Consum Res 1982. London: Routledge. Doctoral Thesis. Int J Res Mark 1997. The low-involvement processing theory. 2003. Moran WT.12: 304 – 17. University of South Australia. The television audience. Brand user profiles seldom differ. Lomax W.6:93 – 100 [September]. Pessemier E. Moorthy KS. Nature 1966. Yim CK. Pickford C. Proceedings of the Academy of Marketing Annual Conference. Ehrenberg et al. 2nd ed. Disaggregating repeat buying patterns. Amsterdam: October. Simulated Dirichlet customers: a software specification. Lenk PJ. Hammond K. Feldwick P. Odin N. Gamma distribution in consumer purchasing. Foxall GR.):G190 – 202 [Part 2 of 2]. Barnard NR. Morrison DG. Excess behavioral loyalty for high-share brands: deviations from the Dirichlet model for repeat-buying. Little JD.30(12):39 – 49. J Forecast 1998. Yao JL. Morrison DG.212:1616. School of Marketing. 1990. Goldsmith RE. The constant in duplicated television viewing. J Bus Res 2001. Wisniewski M. Consumer brand choice as a learning process. Markets versus management: what ‘drives’ profitability? Strateg Manage J 1996. Working paper. Heath R. Aldershot (UK): Gower. South Bank University. Reader S.18(1):1 – 22. The collection and analysis of consumer data. Uncovering patterns in cybershopping. Franzen G. Boston (MA): Kluwer. Leeflang PSH. Adding explanatory variables to a consumer purchase behavior model: an exploratory study. An investigation of the impact of promotions on across-submarket competition. 1963. Lilien GL. . Competitive brandchoice and store-choice among Japanese consumers. Goodhardt GJ. Fader PS. R&DI Report. Heterogeneity in brand choice. Individual choice behavior: a theoretical analysis. Kotler P.147:621 – 55. 1987.6:342 – 6 [August]. present and future. Int J Advert 1999. Goodhardt GJ.14(3. Port Sunlight (UK): Unilever Research Laboratory. South Bank University. Morrison DG.2(1): 25 – 30. The Dirichlet: a comprehensive model of buying behavior. Managing advertising and promotion for longrun profitability. Ehrenberg ASC. J Mark Res 1967.4:155 – 61. Tibrewala V. Guadagni PM. Rungie C.33:442 – 52 [November].40:1484 – 99 [November]. London: Pion. Brand presence and the perceptual frame.2(3):203 – 38 [Summer]. Goodhardt GJ. Goodhardt GJ. Ehrenberg ASC. Jones MJ. Fader PS. UK. Harris J. Hammond K.20: 241 – 57. 2000. Patterns of store-choice.6(1):27 – 39. Competitive market structures: a subset selection analysis. and some explanations. Kearns Z.14(7):20 – 8 [Reprinted in] J Prod Brand Manag 1997. p. Ehrenberg ASC. Dordrecht (The Netherlands): Kluwer. New York: Wiley. Brand loyalty for frequently-bought goods. 1995.53(2):75 – 84. Modeling consumer choice among SKUs. Long S. McDonald C. Ehrenberg ASC.30 (3):288 – 304 [August]. Calif Manage Rev 2001.53(3):137 – 49. Niching versus change-of-pace brands: using purchase frequencies and penetration rates to infer brand positionings. Olshavsky RW. Manage Sci 1994. Boston (MA): MIT Press. J Prod Brand Manag 1998. Hammond K. Fader PS. Market segmentation for competitive brands. An empirical comparison of new product trial forecasting models. Foxall GR. Marketing models. Mark Sci 1999. Goodhardt GJ. Hewitt I. Kagel JH. 1988. Reviving brand loyalty: a reconceptualization within the framework of consumer-brand relationships. Odin Y. 1992. 2002. Wedel M. Granbois DH. Massy WF. Parsons LJ. 2000.7(6):481 – 94. Gordon W. Sanchez S. Building models for marketing decisions. In: Uncles MD. Do we really need ‘brand equity’? J Brand Manag 1996. Leeflang PSH. Presentation to GVR using MRCA data. McFadden D.43:106 – 17 [Summer].

Customer loyalty and customer loyalty programs.17:319 – 31. Chintagunta PK. Market segmentation: conceptual and methodological foundations. The buying of packaged goods at US retail chains. Wrigley N. Dowling G. Purchase loyalty is polarised into either repertoire or subscription patterns.28:29 – 41 [February]. Int J Res Mark 1990a. Kamakura W. Environ Plann A 1984. 11 – 14 May.1185 – 91. Uncles MD. Proceedings of ANZMAC Conference. Int Rev Retail Distrib Consum Res 1995. 2nd ed. Troy LC. / Journal of Business Research 57 (2004) 1307–1325 Rungie C. Goodhardt GJ. J Mark Res Soc 1988. Proceedings of 28th EMAC Conference. Market statistics for the Dirichlet model: using the Juster scale to replace panel data. Ehrenberg ASC.30(1):35 – 44 [January]. J Consum Mark 2003.S. Singh J. How consumers respond to price. Prescriptions for branded and generic pharmaceuticals. Sharp B.36: 488 – 500. Hammond K. Proceedings of ANZMAC Conference. Ehrenberg ASC. Dunn R. Mark Res 1990. models and extensions.C. Kadane JB. Ehrenberg ASC. Wright M. Modeling purchase-timing and brand-switching behavior: incorporating explanatory variables and unobserved heterogeneity. Scriven JA. Hammond K. Driesener C. Wellan DM. A replication study of two brand-loyalty measures. Grocery store patronage. Uncles MD.19 (1):81 – 90. The structure of the UK petrol market. Multistore purchasing patterns and the Dirichlet model. Sharp B. Sharp B. Consumer behavioral loyalty: a segmentation model and analysis. Ehrenberg ASC.7(6):465 – 80. Taudes A. Berlin. Ehrenberg ASC. Ehrenberg ASC. Sudman S. Int J Res Mark 2002. Mark Sci 1995. Research note: brand performance measures from the parameters of the Dirichlet model. Shoemaker RW. Szymanski D. Uncles MD. Proceedings of the MRS Annual Conference. Gold Coast: Griffith University. Hammond K. 1999.2(3):177 – 83. J Mark Res 1999. Berlin. Loyalty programs and their impact on repeat-purchase loyalty patterns. 1995. Scriven JA. Sharp A. Uncles MD. Yim CK. 11 – 14 May. Wright M. Ehrenberg et al. Ehrenberg ASC. Jain DC. Why does the NBD model work? Robustness in representing product purchases. 2003.31:289 – 303 [May]. Sharp A. J Mark Res 1991. J Mark Res 1977. Reports prepared for Shell International Petroleum and BP Oil Europe. 1989. Environ Plann A 1985. Stock-keeping units: examining patterns of buyer behaviour. Thaler R. Kamakura WA. Are Australasian brands different? J Prod Brand Manag 1998. Uncles MD. Understanding brand competition using micro and macro scanner data. School of Marketing Working Paper. 1994. J Mark Res 1994. New brand effects in a Dirichlet market. Kannan PK. Davis RE.44:75 – 92. J Bus Res 1999. London: The Market Research Society. Loyalty programs and their impact on repeat-purchase loyalty patterns: a replication and extension. 2002. Morrison DG. Mark Sci 1985. Consumer panels. .66(3):278 – 96. Shoaf FR.23(3): 57 – 70. Wansink B. Australia. Ainslie A. Ellis K. Incorporating independent variables into the NBD and Dirichlet models. Chicago (IL): AMA. University of New South Wales. Goodhardt GJ. Sharp A. Wagner U. AMJ 2002. Wedel M. Proceedings of 28th EMAC Conference. Eur J Oper Res 1994. A multivariate polya model of brand choice and purchase incidence. Stochastic panel-data models of urban shopping behavior: 4.10(3):7 – 20. Hammond K. Industrial buying behavior: aviation fuel contracts.14:458 – 68. Uncles MD.14(5):473 – 86.4:255 – 66 [Summer]. Sharp A. 2000. The market performance of pharmaceutical brands. New York: Russell Sage Foundation. Wright M.16:759 – 78. J Mark 1995. Stern P.5(3):219 – 44.76:375 – 84. Mark Res Today. Sydney. 2000. J Brand Manage 1995. Wrigley N. Ehrenberg ASC. J Retail 1990b.5(3):287 – 302. Staelin R. Mark Sci 1986. Eur J Mark 1989. London: London Business School. Dunn R. Sharp A. Bharadwaj S. 1325 Uncles MD. Seetharaman PB. Sharp B. Investigating household state dependence effects across categories.20(4):294 – 316. Uncles MD. A case of seasonal segmentation. Goodhardt G. brand purchases and imperfectly recorded purchases. Ehrenberg ASC. University of South Australia. Vilcassim NJ. Sharp B. Schmittlein DC.285 – 92 [November]. BUYER: buyer behavior software. Patterns of buyer behavior: regularities.1:11 – 3. Int J Res Mark 1997. Relation of brand-choice to purchase frequency. Dordrecht (The Netherlands): Kluwer. Wellan DM. 2nd ed. Quasi rational economics. Order of entry and business performance: an empirical synthesis and reexamination. 1999. Scriven JA.59:17 – 33 [October]. Patterns of response to price changes. 1994. 1995. Wright M.7:57 – 68. Stochastic panel-data models of urban shopping behavior: 2. Bemmaor AC. Stern P. Ehrenberg ASC. 1999. The buying of own labels.14(3):G71 – 8 [Part 2 of 2]. A successful new brand: Shield. Russell GJ.A.