See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/237811503

A LOT CAN BE REVEALED BY A LITTLE DATA: TWO PURCHASE ANALYSIS OF

FAST FOOD BUYING

Article · December 2001

CITATIONS READS

4 212

2 authors, including:

Dag Bennett
London South Bank University
33 PUBLICATIONS   191 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Marketing Universitario View project

Luxury Goods buying behaviour View project

All content following this page was uploaded by Dag Bennett on 01 July 2014.

The user has requested enhancement of the downloaded file.

 A LOT CAN BE REVEALED BY A LITTLE DATA:
TWO PURCHASE ANALYSIS OF FAST FOOD BUYING

Dag Bennett and Andrew Ehrenberg

South Bank University

Abstract

Brand buying patterns such as portfolio loyalty and double jeopardy are well established in
many markets where panel data are available. This study uses a small set of survey data from
the Fast Food category to show that many loyalty related measures can be revealed through
analysis of just two purchases. The management implications are that if two purchase records
can be obtained the underlying market structures and purchasing patterns of data-poor markets
can be revealed.

The Patterns

This study is a preliminary exploration of whether well-established brand purchasing patterns

can be discerned using non-panel data. A previous study using pairs of purchases extracted
from panel data generated such patterns (Bennett et al 2000) but this is the first attempt to
apply the technique to survey data. This is important because panel data do not exist in all
markets (emerging markets), or may be unreliable or too expensive to gather (subscription
markets). In addition, there may be instances when a marketer would like a quick look at a
market to see whether a particular brand is performing as expected (Ehrenberg, Goodhardt &
Uncles 2001).

The benchmark patterns chosen for use in this study are those associated with the Dirichlet
model of buyer behaviour because “…the Dirichlet may be the best known example of an
empirical generalization in marketing…” (Uncles et al., 1995), and the model had been fitted
to data from the market in 2000, yielding the standard patterns (Sharp & Driesner, 2000).
However, no attempt has been made to fit the model to survey data. That would have violated
underlying assumptions of the model. Rather, this has been an attempt to generate the
patterns with small sets of survey data. In this it has been successful.

Brand loyalty and buying patterns have been extensively discussed among marketing scholars
(e.g. Uncles, et al 1995; Fader and Schmittlein, 1993) and the notion that brands’ loyalty
levels in established, competitive markets follow predictable norms has been repeatedly
confirmed and explained in over 50 product and service categories with the Dirichlet model.
These patterns have also been shown in subscription markets (Sharp & Wright, 2000),
European car markets (Ehrenberg & Bound, 1999) and the Australian car market (Bennett,
2001).

The established patterns relate to many different brand performance measures such as how
many people buy a brand, how often they do so, and which other brands they also buy. The
primary pattern is that all brand performance measures tend to vary together from one brand
to another. (Ehrenberg & Uncles 1995)

Another major pattern is that repeat-buying and brand-switching are dominated by how big
each brand is (its market share), and not by any other particular attributes or values of the
different brands. Nor do the patterns vary with consumers’ attributes (Kennedy & Ehrenberg
2001). Large share brands tend to score high on all measures and small ones score low.
Loyalty is therefore not specific to a brand. Instead, brands with the same market share tend
to have similar levels of loyalty. (Ehrenberg & Uncles 1995)

Finally, brand loyalty is usually divided. Most consumers buy more than one brand in a
category over any extended sequence of purchases, purchasing one less often than another yet
each brand more or less regularly, even if infrequently (Ehrenberg & Uncles 1995). Most
consumers therefore divide their loyalty over a portfolio of brands and switch back and forth
between brands within their portfolio only occasionally adding a new brand or dropping an
old one.

Quick Service Outlet Purchasing

The data in this analysis was gathered for a commercial market research project in South
Australia during February 2001. The sample contained 408 people who had bought meals
from Quick Service Outlets (QSOs) within the past 3 months. The data here came from three
questions about the last QSO used, the one before that, and the one before that.

Simple frequency analysis established brand shares (Table 1, last column), which were stable
(+/- 3%) from purchase to purchase. It also showed that the marketplace had a few larger
brands and many smaller ones. In this case, McDonalds (18%) and All Pizza (16%) were
largest, followed by Hungry Jacks (12%), etc. Several brands were amalgamated from
individual shops and small chains. Thus All Pizza includes a number of outlets and brands
which serve mainly pizza. The same holds for Chinese Take-Away, Local Deli, and Food
Court.

Table 1 is a duplication of purchase table showing the percentages of those whose first
purchase was the brand on the left who bought the brand listed on the top of the table at their
second purchase. For example, of those whose first purchase was McDonalds, 33% bought
McDonald’s the second time, 5% bought All Pizza and so on. Pairing two purchases shows
that about two thirds of QSO customers are not 100% loyal, even over just two purchases.
Table 1. Switching and Repeat percentages from First to Second purchase
2nd Brand purchased
st
1 Purchase McD Pizza H J LDel F&C F Sub KFC Oth C-TA Share
Ct
McDonalds 33 5 3 8 5 3 17 9 3 1 18.4
All Pizza 17 32 12 8 8 2 10 3 2 2 16.1
Hungry Jacks 13 11 32 6 6 8 4 4 6 2 12.5
Local Deli 7 9 9 34 7 11 5 5 2 0 11.7
Fish & Chips 9 9 6 3 41 0 9 3 3 12 8.6

Food Court 0 9 6 9 6 33 9 6 6 3 8.3

Subway 20 4 20 8 0 4 36 0 4 0 7.3
KFC 20 14 14 3 0 6 6 17 6 3 7.2
Other 23 9 5 9 9 0 5 5 32 5 7.2
Chinese TA 6 25 6 0 6 6 6 6 6 31 2.8

Ave Switch 11.0 9.6 8.1 5.4 5.6 4.0 7.0 4.1 3.7 2.7
The diagonal figures also show that repeat buying is similar for all brands, slightly higher for
some and lower for others, notably KFC. Overall, there is a small downward trend with
market share from 33% for McDonalds to 31% for Chinese Take-Away (the overall average
repurchase rate is 32.1%, excluding Don’t Knows). This is a classic Double Jeopardy pattern
found in many other product categories (Ehrenberg & Uncles 2000).

The same pattern was found, with some deviations, when the second and third purchases were
compared as can be seen in the second line of Table 2 which shows repurchase percentages
for each possible 2-purchase comparison, first and second, second and third and first and
third. The figures for consecutive purchases (top two lines) are similar, while repeat buying
between first and third purchases is lower by about 11 percent on average. This might be an
example of erosion (Hammond and East, 1996) in which external factors erode brand
repurchase rates over time. But given the short time frame in this study (86% bought once or
twice a month or more often) it is more likely due to variety seeking.

Table 2, Percentage of buyers who bought the same brand on both purchase occasions

Purchases McD All pizza HJ Deli F&C Fd Ct Sub KFC other C-TA Ave
Consecutive
1st & 2nd 33 32 32 34 41 33 36 17 32 31 32.1
nd
2 &3rd 36 32 31 50 37 52 24 13 29 19 32.3
Non-consecutive
1st & 3rd 28 21 26 25 26 30 24 6 22 6 21.4

Exceptions to patterns attract attention. Whether they should is another question

(Bhattacharya 1997). In this case, it is hard to tell. KFC for example, has repeat purchase
rates that are consistently lower than other brands. This may be due to exceptional marketing
activities during the survey period, or perhaps an unusual event in the marketplace such as a
shop closure. On the other hand, it may be due to sampling error which could have a large
effect on smaller brands.

Switching is in line with Penetration

Table 3 shows penetration for individual brands rising from purchase to purchase roughly in
line with market share. For example, 18% of buyers had bought McDonalds at the first
purchase, increasing to 29% by the second. On average, the number of people buying each
brand increases by about half from the first to the second purchase. Even so, by the second
purchase, large brands have a lot more customers than small brands. This is what
distinguishes a large brand from a small one, it has a lot more people buying it.

Table 3, Growth of Penetration from first to third purchases

Purchase McD Pizz HJ Del F&C FCt Sub KFC other C T-A Ave
1st 18 16 13 11 8 8 6 9 5 4 7.5
2nd 29 24 20 16 14 12 14 13 7 7 12.0

The penetration and switching levels are also in line with the percentage buying the brands at
all during the time period, i.e. both were higher for McDonald’s or Hungry Jacks than for
Chinese Take-away or KFC. This relationship of switching to penetration is an illustration of
the “Duplication of Purchase Law” which has been widely established for a large number of
categories. (Goodhardt 1966; Ehrenberg & Uncles 1995). For QSOs it reads:

The % of the previous customers of brand P

Who switch to the new brand N
≅ D x the penetration of N

In other words, more people switch to a big brand than a small one. The proportionality factor
“D” reflects the likelihood of switching from the previous brand P to a new brand N, relative
to how many people bought N at all during the time period.

D = % of customers of P who switch from P to N

% who buy N at all

Table 4 shows the average switching to figures for QSO brands (averaged over 2 purchases)
giving an average D-value for all brands of roughly .4. Thus 5% of McDonald’s customers
switched to Fish & Chips, which is about 40% of Fish & Chips’ overall penetration of 13%.
By and large, switching levels decrease in line with .4 times the brands’ penetrations from
high for McDonalds on the left to low for Chinese Take-Away on the right. (The correlation
is r = 0.93).

Table 4, Average Switching: Observed and Predicted

McD Pizz HJ Deli F&C FCt Sub KFC other CT-A Av.
Previous brand*
Av. Switch to 11.3 10.3 8.1 5.4 5.6 4.0 7.0 4.1 3.7 2.7 6.2
.4 x Penetration 11.3 10.3 9.4 7.4 6.3 5.6 5.2 5.1 4.0 2.9 6.8

• Average of 9 brands

There are however deviations: Only Subway’s switching to level is above its penetration norm
and this may be because Subway is perceived to be a lunch specialist while all other QSOs are
seen to offer 2 or 3 meals per day. It may also be related to an unusual level of marketing
activity by Subway during the study period.

Partitioning

Overall the average duplications in the top row in Table 4 were in line with brand penetrations
in the bottom row. This is essentially the standard Duplication of Purchase Law Pattern noted
above. Against this, the purchase duplications between pairs of brands reflect clusters or
submarkets within the QSO marketplace. Using provisional labels, the clusters are:
American-Style Chains (McDonalds, Hungry Jacks, Subway, Pizza Hut and KFC), Local
specialists (Local Deli and Food Court) and Take-Home (Fish & Chips and Chinese Take
Away). The clusters are derived from the differing likelihoods of switching between pairs of
QSOs, (assuming a high degree of substitutability), as expressed by their D-values.

Table 5 shows that there is slightly high duplication between American-Style chains, higher
for Local Specialists, and markedly higher for Take-Home food. In other words, there
appears to be a partition between different types of QSOs. The high D value shows that
customers substitute one take-home shop for another, even though Chinese is quite different
from Fish & Chips—the substitution seems based on buying and taking home, not the food
itself.

Table 5 Duplication-Coefficients D by submarket

Current Purchase
American-Style Local Specialists Take Home
Previous Purchase
American-Style Chains .5 .4 .3
Local Specialists .3 .6 .3
Take-Home .4 .2 1.0

Such simple clustering is not surprising to anyone already knowledgeable with the market.
No sophisticated statistical analysis techniques are needed to discern it (Ehrenberg et al,
2001). But the quantitative detail would be much less easy to ascertain without the
accumulated knowledge of the simple underlying patterns and theory.

Conclusions

Analysis of a very small number of purchase records reveals many brand-buying patterns such
as the dominance of market share in determining brand performance measures, double
jeopardy, and the duplication of purchase law. It is of course very helpful to know the
patterns to start with in determining the structure of a market.

Moreover, while this examination was performed on a small data set, it still managed to
generate results that are both in line with the expected patterns and accurate, especially for
large brands. The main limitations of this type of analysis are that not all brand performance
measures can be calculated (100% loyalty, Share of category requirements, etc.) and that
smaller brands are more susceptible to sampling error.

These results are encouraging and may hold the promise of a breakthrough towards a quick,
inexpensive and easy-to-use methodology for assessing brand performance. The research is
the first attempt for a methodical study of an under-researched area, and aims to make a
significant contribution towards the development of analysis techniques for use in markets
where data is difficult or expensive to acquire. In addition, the successful development and
application of the Two Purchase technique should provide marketers with a useful tool to
quickly and efficiently examine the brand buying behaviour of consumers.

The next steps in confirming the usefulness of the two-purchase analysis technique will be to
gather additional data sets in both well-known FMCG markets and less-well understood
markets such as those in developing countries and subscription markets.

References

Bennett, D R, Ehrenberg, ASC & Goodhardt, G (2000) “Two Purchase Analysis of Petrol
Purchasing”, ANZMAC 2000, Gold Coast, November 29-December 1
 Bennett, D R, (2001) “Two Purchase Analysis of Automotive Survey Data” unpublished
working paper at the Centre for Research in Marketing, South Bank University, London

Bhattacharya, C B, (1997) “Is your brand’s loyalty too much, too little, or just right ?:
Explaining deviations in loyalty from the Dirichlet norm”, International journal of Research
in Marketing, 14, 421-435

Dowling, G R & Uncles, M D, (1997) “Do Customer Loyalty Programs Really Work?”
Sloan Management Review 38 (4) 71-82

Ehrenberg, A S C & Bound, J, (1999) “Customer Retention and Switching in the Car Market”
Research Report 6, The R&D Initiative, South Bank University

Ehrenberg, A S C, Goodhardt G & Uncles, M D, (2001) “Using Benchmarks in

Understanding Buyer Behavior” Working paper at the Centre for Research in Marketing,
South Bank University, London,

Ehrenberg, A S C, & Long, S (1997) “Customer Retention and Brand Switching in the PC
Market”, unpublished working paper at the Centre for Research in Marketing, South Bank
University, London

Ehrenberg, A S C & Uncles, M D (2000), “Understanding Dirichlet-type Markets,” the R&D

Initiative Research Report 1, South Bank University, London

Ehrenberg, A S C & Uncles, M D (1995), “Dirichlet-type Markets: A Review,” unpublished

working paper at the Centre for Research in Marketing, South Bank University, London

Fader, P S & Schmittlein, D C, (1993) “Triple Jeopardy: Excess Behavioral Loyalty

Experienced by High-share Brands” Journal. marketing Research 30 (November) 478-493

Goodhardt, G (1966) “The Constant in Duplicated Television Viewing”, Nature, 212, 1616

Hammond, K. & East R, (1996) “Erosion of Repeat-Purchase Loyalty,” Marketing Letters,

Vol. 7:2 p163-171

Kennedy, R. & Ehrenberg, A S C, (2001) “There is No Brand Segmentation, ” Marketing

Research (in press)

Scriven, J and Bennett, D, (2001), “Brand Purchasing Patterns in the Petrol Market” Internal
paper available on request

Sharp, B, & Driesner, C, (2000) “The Dirichlet's Buyer Behaviour Assumptions Really Do
Matter”, ANZMAC 2000, Gold Coast, November 29-December 1

Sharp, B, & Wright, M, (2000) “There are Two Types of Repeat Purchase Markets”,
ANZMAC 2000, Gold Coast, November 29-December 1

Uncles, M D, Ehrenberg, A S C, Hammond, K (1995) “Patterns of buyer behavior:

Regularities, Models and Extensions ” Marketing Science 14 (3) G71-G78, Part 2 of 2

View publication stats