Post on 30-Jun-2020
transcript
RESEARCH P
GRADUATE SCH
STANFORD
Research P
EQUITYMAP: M and Prediction of Br
V. SriChan
Dae Ry
Apri
aper No. 1685
easurement, Analysis, and Equity and its Sources
nivasan Su Park un Chang
l 2001
APER SERIES
OOL OF BUSINESS
UNIVERSITY
EQUITYMAP:
Measurement, Analysis, and Prediction
of Brand Equity and its Sources
V. Srinivasan
Ernest C. Arbuckle Professor of Marketing and Management Science
Graduate School of Business, Stanford University, Stanford, CA 94305, U.S.A.
Phone: (650)723-8505, Fax: (650)725-6152
E-mail: seenu@stanford.edu
Chan Su Park
Associate Professor of Marketing
College of Business Administration, Korea University, Seoul, 136-701, Korea
Phone: +82(2)3290-1947, Fax: +82(2)922-1380
E-mail: chansu@mail.korea.ac.kr
Dae Ryun Chang
Professor of Marketing
College of Business and Economics, Yonsei University, Seoul, 120-749, Korea
Phone: +82(2)361-2516, Fax: +82(2)393-7272
E-mail: drchang@mail.yonsei.ac.kr
March, 2001
* The authors thank Mr. Ja Ik Koo at Samsung Electronics Co., Ltd. for his assistance,
Professor Allan Shocker, Doctoral Student Oded Netzer, and the participants in the marketing
seminars at Stanford and Duke Universities for their comments on an earlier version of this
paper. This research was supported in part by the Korea Research Foundation Grant (KRF-
2000-0000) and the Korea University Institute of Business Research and Education Grant
awarded to the second author.
EQUITYMAP:
Measurement, Analysis, and Prediction of Brand Equity and its Sources
Abstract
The authors propose EQUITYMAP, a new approach for measuring, analyzing, and predicting
a brand’s equity in a product market. It defines brand equity at the firm level as the
incremental profit per year obtained by the brand in comparison to a brand with the same
product and price but with minimal brand-building efforts. At the customer level, it
determines the difference between an individual customer’s overall choice probability for the
brand and his or her choice probability for the underlying product with merely its push-based
availability and awareness. The approach takes into account three sources of brand equity –
brand awareness, attribute perception biases, and non-attribute preference – and reveals how
much each of the three sources contributes to brand equity. In addition, the proposed method
incorporates the impact of brand equity on enhancing the brand’s availability. The method
provides what-if analysis capabilities to predict the likely impacts of alternative approaches to
enhance a brand’s equity. The survey-based results from applying the method to the Korean
digital cellular phone market show that the proposed approach has good face validity and
convergent validity, with brand awareness playing the largest role, followed by attribute-
based equity.
1
Since the late 1980s brand equity has been one of the most important marketing concepts in
both academia and practice. While several different definitions of brand equity have been offered
over the years, many of them are consistent with Farquhar’s (1989) definition of brand equity as
the value added by the brand to the product (see Keller (1998, p.43) for a summary of alternative
definitions of brand equity).
A key requirement for managing brand equity is the availability of good measures (Aaker and
Joachimsthaler 2000).1 In this paper, we propose EQUITYMAP, a new method for measuring,
analyzing, and predicting a brand’s equity in a product market that is consistent with the “added
value” notion of brand equity. We define brand equity at the firm level as the incremental profit
per year obtained by the brand in comparison to the same product (or service)2 and price but with
minimal brand-building efforts.
We conceptualize brand equity as arising from the following three sources: (i) increased
brand awareness, (ii) enhanced attribute perceptions, and (iii) favorable non-attribute preference.
In addition, we take into account the trade’s perspective on brands as articulated by Farquhar
(1989) by estimating the impact of the above three sources on the increased availability of the
brand. Stated differently, our measure of brand equity incorporates the impact of the increased
customer pull on brand availability.
Based on this conceptualization and using a multiattributed probabilistic choice model,
EQUITYMAP estimates brand equity at the individual customer level by determining the
incremental choice probability, i.e., the difference between an individual customer’s overall
choice probability for the brand and his or her choice probability for the same product and price
but with minimal brand-building efforts. Summing across customers (or a segment of
customers) the incremental choice probabilities multiplied by the corresponding category
1 Since previous researchers (e.g., Keller 1998 (chapters 8-9), Park and Srinivasan 1994, Srivastava and Shocker 1991) provide detailed review of extant measurement approaches (in addition to providing methods of their own), we do not provide a review of our own in this paper.
2
purchase quantities and the brand’s profit margin yields a measure of brand equity in profit
terms, which is more meaningful to managers than some of the previous measures that are more
in abstract terms.
In addition to providing a summary measure of a brand’s equity in profit terms,
EQUITYMAP can evaluate the relative contributions toward the incremental choice probability
from the three sources, thus helping brand managers better understand and evaluate the sources
of brand equity. In addition, EQUITYMAP permits a variety of what-if analyses evaluating the
bottom-line impact of alternative brand-building strategies. For example, brand managers can
answer questions such as what would be the impact on brand profitability of increasing brand
awareness by x% or improving attribute perception on attribute k by an amount y. With a
growing call for greater accountability in marketing activities and programs (Marketing Science
Institute 2000), the flexible what-if capability offered by the proposed approach could become a
valuable decision-support tool.
We believe that EQUITYMAP offers the following key contributions: (i) it provides a
conceptual framework for thinking about brand equity and relating it to its three sources, (ii) it
offers a method that simultaneously measures and provides an understanding of brand equity,
(iii) it determines the profitability implications of the current level of brand equity, (iv) it
provides a way of evaluating alternative strategies for enhancing brand equity, and (v) if
implemented as a tracking system, it provides a method for goal setting (e.g., levels of awareness
or attribute perceptions), monitoring, enhancing, and managing brand equity over time. The
approach is closest in spirit to Park and Srinivasan (1994). Compared to that earlier approach,
EQUITYMAP adds brand awareness as an important source of brand equity and incorporates
the effects of enhancing brand awareness and preference on the brand’s availability.
2 Hereafter we use the term “product” to include both products and services.
3
Furthermore, by adopting a probabilistic framework EQUITYMAP overcomes some of the
difficult scaling problems encountered in that earlier approach.
We begin by developing the conceptual basis and propose a method for measuring and
analyzing brand equity and its three sources. Then we provide an illustrative application of the
method to the Korean digital cellular phone market. We close with a discussion of the findings,
limitations, and directions for future research.
CONCEPTUAL FRAMEWORK
We define a brand’s equity in a product market as the incremental profit per year obtained by
the brand in comparison to a brand with the same product and price but with minimal brand-
building efforts (hereafter, base brand). The firm obtains the incremental profit because the
customer’s overall choice probability for the brand is greater than his or her choice probability
for the base brand. Note that our definition of brand equity is consistent with the commonly used
definition of brand equity (Farquhar 1989) as the added value endowed by the brand to the
product.
By the base brand, we do not necessarily mean an existing brand such as a weak national
brand or a store brand although such brands could provide useful information regarding
awareness levels, attribute perception biases, and non-attribute-based preference resulting from
minimal brand-building efforts. However, defining the base brand based solely on an existing
brand runs the risk of counting as brand equity the effect of push-based distribution. For
example, suppose that there are two firms, Firm A and Firm B, and neither of them makes any
brand-building efforts. Suppose further that Firm A’s brand in a new product market achieves
greater availability, awareness, and choice probability than Firm B’s identical new product
simply because Firm A has a stronger clout in terms of the distribution channel stemming from
4
its stronger salesforce. If we choose Firm B’s brand as the base brand for Firm A’s brand, we
would overstate the equity of Firm A’s brand. The base brand corresponding to Firm A’s brand
should have higher choice probability than Firm B’s brand. Let us call the levels of brand
availability and brand awareness created by push-based distribution push-based availability and
push-based awareness, respectively. Objective data for determining push-based availability and
awareness are usually not available. Therefore we allow for the use of industry experts to
estimate what would have been the likely result of push-based distribution by the particular firm
(in the particular product market) in terms of the resulting brand availability and brand awareness.
For instance, in the empirical application presented in a later section, we asked industry experts:
“In your best judgment, what would have been the levels of the brand’s availability (in terms of
the brand’s share of shelf space in a typical retail outlet of cellular phones) and its awareness had
the brand not conducted any brand-building activities and relied entirely on the current level of
push through the channel (by the salesforce)?” The precise operational definition of brand
availability (e.g., % ACV (all category volume), percentage of stores carrying the brand, percent
of shelf space) depends on the product market in question and can be determined after
discussions with industry experts.
The individual-customer-level brand equity is expressed as
(1) ,jijiij gpqe ∆=
where
ije = brand j’s equity ($/year) from customer i,
iq = customer i’s total purchase quantity in the product market (units/year),
ijp∆ = customer i’s incremental choice probability for brand j compared to the
base brand, and
jg = brand j’s contribution margin ($/unit).
For expositional convenience, we will often refer to the incremental choice probability ijp∆
as the customer-level brand equity, although strictly speaking, it is given in profit terms by
Equation 1.
Aggregating the individual measures of brand equity over N respondents in a representative
sample and scaling it to the overall market (or market segment) gives an aggregate-level (or
segment-level) brand equity measure ej:
(2) ∑=
∆=N
iijijj pqgQTe
1
)/( ,
where T denotes the total product category quantity per year for the entire market (in units) and
Q denotes the total quantity per year summed over the sample of respondentN
Thus (T/Q) “scales up” the sample of respondents to the population level. (
margin varies across customers, e.g., bank customers using tellers vs. ATMs,
enter into the summation as ijg .) Note that in order to minimize aggregation
otherwise result from nonlinear models, the choice probabilities are measure
level for each brand. Individual-level measurement also permits aggregati
bases for segmentation such as geography, demographics, psychograph
Although the measurement errors would be greater at the individual level,
diminish (as a percent of the mean) when aggregated to the segment or m
equation (2).
5
s (i.e., ∑=
=i
iqQ
1
).
If the contribution
the term jg would
biases that would
d at the individual
on using multiple
ics, and benefits.
the errors would
arket level as per
6
Equation 2 provides the equity of brand j in a particular product market, e.g., Sony’s equity
in the console television market in the U.S. It can be aggregated across geographic markets, e.g.,
different regions of the world. Furthermore, a narrower or broader product market definition
may be employed depending on the managerial objective, e.g., instead of console televisions, one
could have defined the product market more broadly as televisions.
The aggregate-level brand equity in Equation 2 can be interpreted as the incremental
contribution margin generated by the firm’s brand-building efforts which require considerable
investment. We view the investment as a fixed cost rather than a variable cost affecting gj. Thus
the firm can analyze the rate of return on its brand-building investment in the product market by
comparing ej with the amount of investment over the years.
The aggregate-level brand equity is measured in profit terms of $/year. In the stock market
brand valuation literature (Birkin 1994, Financial World (Meschi 1995), and Simon and Sullivan
1993), our measure of brand equity corresponds to yearly profit impact. To obtain brand
valuation we need to multiply brand equity as given by Equation 2 by a multiplicative factor to
incorporate (i) the number of years the firm is expected to reap the benefits of brand equity (with
appropriate discounting to express it in present value terms) and (ii) the extendibility of the brand
into categories beyond the product market definition used in the study. For instance, the
Interbrand approach (Birkin 1994) determines a multiplicative factor (maximum value = 20)
based on subjective ratings on factors such as leadership, stability, type of market, geographic
spread, trend, support and protection.
The brand equity definition of Equation 1 has a relationship to customer equity, defined as
the lifetime dollar value of the customer to the firm (Blattberg and Deighton 1996). In the
context of customer equity, there is no need to consider the increment over the base brand. Thus
by replacing pij in Equation 1 by the choice probability pij, we would obtain customer i’s dollar
value to the firm per year. To convert the per year number to lifetime value, we need a
7
multiplicative factor to incorporate the customer’s expected lifetime. In the customer equity
context, it would also be necessary to define the product category more broadly. For instance, in
the earlier Sony example, the product category may be defined more broadly as consumer
electronics rather than as console televisions.
Three Sources of Brand Equity
We assume the brand choice probability is determined by a probabilistic choice model such
as the multinomial logit model, as detailed later in the paper. Let p denote an individual
customer’s overall choice probability for the brand and p’ denote the customer’s choice
probability for the base brand. The probabilities p and p’ should have a subscript i to denote the
individual customer and the second subscript j to denote the brand. Consequently, pij summed
over j (all the brands in the market) would equal one for the ith customer. For expositional
convenience, however, we omit the subscripts whenever it is feasible.
The incremental choice probability, p – p’ (= p∆ ), is the result of the firm’s brand-building
efforts. To better understand what underlies p – p’, let us start by examining p’. We postulate
that the choice probability for a brand with minimal brand-building efforts (corresponding to the
right-hand-side square in Figure 1) is driven by the following three elements (corresponding to
the left-hand-side squares in Figure 1): underlying product (arrow a), push-based brand
availability (arrow b) and push-based brand awareness (arrow c).
Figure 1 about here
Next, we postulate that the incremental choice probability (= p– p’) arises from the
following three sources: increased brand awareness, enhanced attribute perceptions, and
favorable non-attribute preference. (The latter two are bundled together in Figure 1 under
“enhanced brand image”.)
Brand awareness can play a dominant role in brand choice if the customer has strong
awareness of some brands but not of other brands, in part because brands with little awareness
8
are unlikely to be considered for purchase. In addition to being a powerful driver of brand
purchase (arrow d in Figure 1)(Nedungadi 1990), a high level of brand awareness in a product
market will encourage the trade to stock the brand, leading to high brand availability and, in turn,
high brand choice probability (arrows f and h).3
If the customer is aware of the brand, the customer has perceptions (associations or image)
toward the brand. The literature on brand equity (e.g., Aaker 1991, 1996; Keller 1998) has
recognized brand associations as important bases underlying brand equity. Strong, favorable and
unique brand associations can enhance brand preference (Keller 1998) and favorably affect brand
choice probability (arrow e). In addition, increased brand preference can enhance the trade’s
interest in stocking the brand, thus increasing brand choice probability (arrows g and h).
Brand associations are quite diverse and there are several ways to classify them. For
example, Aaker (1996, p.79) postulates brand associations as consisting of twelve elements and
then organizes them around four categories: brand-as-product, brand-as-organization, brand-as-
person, and brand-as-symbol. He observes that many firms tend to focus solely on brand
associations related to product attributes in managing brands, and urges that firms adopt a broad
perspective on brand to include the other three dimensions of brand associations which are all
unrelated to product attributes. Keller (1998, p.93) classifies brand associations into three major
categories: attributes, benefits, and attitudes. Further, he distinguishes attributes into product-
related attributes and non-product-related attributes. Similarly, he distinguishes benefits into
functional benefits (corresponding to product-related attributes), symbolic benefits
(corresponding to non-product-related attributes), and experiential benefits (corresponding to
both product-related attributes as well as non-product-related attributes).
3 An increase in advertising expenditure increases brand awareness. The resulting increase in customer interest in the brand may encourage the trade to increase the brand’s distribution. The increased distribution could, in turn, increase brand awareness albeit to a much smaller extent. This, in turn, could increase distribution, …The net overall result of this dampened chain of events is captured in Figure 1 as the enhanced brand awareness and the corresponding enhanced brand availability. The detailed modeling of the dynamic effects is beyond the scope of this paper.
9
Some of the previously proposed measurement methods utilize similar but more
parsimonious categorizations of brand associations. Park and Srinivasan (1994) conceptualize
brand associations contribute to brand equity by creating an attribute-based component of brand
equity and a non-attribute-based component of brand equity. The attribute-based component of
brand equity is created by brand associations related to product attributes resulting in favorably
biased attribute perceptions. The non-attribute-based component of brand equity is created by
brand associations unrelated to product attributes such as user imagery, brand personality (Aaker
1997) (e.g., the rugged and masculine image conveyed by the Marlboro Man), and usage
situation imagery. Kamakura and Russell (1993) also develop a conceptual model that divides
brand equity into two components, although their empirical method does not break down the
brand equity estimate into the two components.
We conceptualize the incremental choice probability due to enhanced brand preference arises
from two sources, favorable attribute perceptions and non-attribute preference, which is
consistent with the previous measurement approaches. Park and Srinivasan (1994, p. 274)
provide further evidence supporting this conceptualization.
Related Measures
Ultimately what matters to the firm is the profitability of the brand as determined by p. This
is not only due to brand equity (p – p’), but also due to p’, the choice probability resulting from
the product itself and the push-based awareness and availability. Note that
(3) p = (p – p’) + (p’).
Multiplying both sides by jg iq and aggregating (see Equation 2), we obtain
10
(4) (T/Q) jg ∑i
iq pij = (T/Q) jg ∑i
iq ( pij – pij’) + (T/Q) jg ∑i
iq pij’.
In words,
(5) Total brand profitability = Brand equity + Base profitability.
The second term on the right-hand-side of Equation 5 reflects how successful the firm has been
in choosing its product attributes to suit customer preferences vis-à-vis its competitors and in
pushing the product through the distribution channels.
EQUITYMAP offers another useful way of decomposing brand equity. Note that the per-
unit contribution margin, jg , in Equation 2 can be divided into the following two terms:
(6) Contribution margin = Brand’s price – Brand’s unit variable cost
= (Brand’s price – Average market price)
+ (Average market price – Brand’s unit variable cost)
= (Price premium)
+ (Contribution margin obtained at average market price)
Hence, substituting Equation 6 into the contribution margin term in Equation 2 yields
(7) je = (Brand equity harvested through price premium) +
(Brand equity obtained at the contribution margin based on average market price).
While the first component would be larger for brands such as Lexus or Gucci, the second
component would be larger for brands such as Wal-Mart or McDonald’s.
An important issue in the field of brand equity measurement is finding an indication of brand
strength that is not simply the same as brand size (= market share which is � iq pij/Q) (Feldwick
1996). Working with choice probabilities in measuring brand equity yields an advantage of
providing a way of assessing brand strength separate from brand size. We define brand strength
jr as the weight
Thus
(8)
where ijδ = 1 if i
Since pij deno
is a customer of
expressed as
(9)
For frequently p
Russell and Srin
brand strength, t
11
ed average probability of purchase of brand j among the brand’s customers.
jr =
∑
∑
iiji
iijiji
q
pq
δ
δ
ndividual i is a customer of brand j and 0 otherwise.
tes the probability that customer i buys brand j, the extent to which individual i
brand j is also pij. By substituting ijδ = pij , a measure of brand strength can be
jr =
∑
∑
iiji
iijiji
pq
ppq =
∑
∑
iiji
iiji
pq
pq2
.
urchased products, rj is the aggregate repeat-purchase probability (Bucklin,
ivasan 1998), and thus intuitively a measure of brand strength. The greater the
he greater is the likely sustainability of the brand in the future. It is thus an
12
important input for determining the multiplicative factor to convert brand equity into stock
market brand valuation (see earlier discussion). As shown by Bucklin, Russell and Srinivasan
(1998), this measure is inversely related to the brand’s own-price elasticity, i.e., a stronger
brand’s sales are less affected by its own price changes.
MEASUREMENT AND ESTIMATION
The basic idea behind the measurement method is to determine p and p’ using a brand
preference-based probabilistic choice model (e.g., logit model based on brand preference) at the
individual customer level.
EQUITYMAP utilizes data from three sources: client firm, industry experts, and customer
survey. The client firm provides information on prices of brands, contribution margin of the
firm’s product, and availabilities of brands. In the empirical study, industry experts provided
objectively measured attribute ratings4 (e.g., Park and Srinivasan 1994), push-based availability,
and push-based awareness of brands.
The customer survey is conducted over a random sample of current users of the product
category, drawn from the geographic market(s) of interest to the researcher. The customer
survey measures the following: awareness of brands, recent purchase behavior, attribute
perceptions, attribute importances, non-attribute perceptions, overall brand preferences,
psychographics, and demographics.
In this section, we present basic building blocks of the proposed approach, i.e., models and
measures that are used to estimate the brand’s choice probability. In the next section, we
describe how to estimate the impact of the three sources of brand equity on the brand’s choice
4 For these one can also rely on sources other than industry experts, e.g., independent testing agencies (e.g., Consumers Union) or blind product tests. Depending on the nature of the attributes, one can rely on one source for a subset of product attributes and another source for a different subset of product attributes.
13
probability. For readers’ convenience, we present a summary of notation in Table 1.
Table 1 about here
Linking Overall Preference to Choice Probability
Let pij denote individual i’s probability of choosing brand j after taking into account brand
availability, brand awareness as well as overall brand preference. We can obtain the brand’s
choice probability using the following logit model:
(10) )(exp
)exp(
kik
iCk
jij
ij
Au
Aup
γα
γα
+
+=
∑∈
,
where
Ci = set of brands individual i is aware of,
iju = individual i’s overall preference for brand j∈Ci,
jA = availability factor for brand j∈Ci.
For expositional convenience, we will refer to those brands belonging to Ci as individual i’s
familiar brands. The parameters α and γ are to be estimated (see below). Since the choice set is
restricted to the set of familiar brands of the customer, if the customer is not aware of a brand, its
choice probability is set to zero.
To obtain the overall preference measure uij, we used (in our empirical application) a
constant-sum scaling method in which the preference scores sum up to 100 over familiar brands.
14
There are alternative ways of collecting overall preferences such as paired constant-sum scaling
for all pairs of familiar brands, followed by a log-linear regression (Silk and Urban 1978).
To obtain the availability factor Aj, one can use different methods depending on the product
market. For example, for customer packaged goods, one can use % ACV (All Commodities
Volume) reported by retail audit services, and for customer durable goods in which such retail
audit services may not be available, one may conduct survey of a sample of stores, or obtain
ratings from a group of industry experts.
In the empirical application, we estimated Equation 10 by using information on the last brand
purchased reported by the respondents. One can estimate the parameters α and γ by maximizing
the following likelihood function:
(11) ij
ijiCj
N
ipL
δ
∈=ΠΠ=
1,
where ijδ equals 1 if individual i last purchased brand j and equals 0 otherwise. An alternative
method of estimating Equation 11 in the context of frequently purchased consumer goods is to
obtain data from each respondent on the number of times each familiar brand was purchased
over the past (say) six months. The parameters α and γ can then be estimated by maximizing
Equation 11 with ijδ replaced by the purchase frequencies. One can potentially allow for
customer heterogeneity in the parameters α and γ using hierarchical Bayes methods.
Elicitation of Multiattribute Preference Structure
The overall preference measure uij captures brand preference arising from attribute
perceptions as well as non-attribute preference. To separate out the attribute-based source of
15
brand equity from the non-attribute-based one, we first need to obtain a multiattribute preference
measure.
Let sijp denote individual i’s subjectively perceived level of brand j on product attribute p
(p=1,2,…,q), and let m(s)ij denote individual i’s multiattribute preference for brand j based on
subjectively perceived attribute levels:
(12) ∑=
=q
pijpipij
sfsm1
)()( .
In the empirical application, we used the self-explicated approach to obtain the multiattribute
preference structure {fip} for each individual i in the sample. (For a detailed discussion on the
self-explicated preference structure measurement, see Green and Srinivasan 1990.) This
measurement method consists of two steps: (i) desirability ratings (say, 1 to 10) for the different
levels of each categorical attribute, and (ii) importance ratings for all attributes. The partworth
functions {fip} are obtained by multiplying the attribute importance ratings by the desirability
ratings for the levels of each attribute. (For continuous (i.e., non-categorical) attributes, we
assume {fip} is linear so that only the attribute importance rating is needed for such attributes.)
Each individual’s partworth function is then rescaled such that the partworth corresponding to
the least preferred level of each attribute equals zero, and the sum of partworths corresponding to
the most preferred levels of the attributes equals 100. (The partworth corresponding to the most
preferred level of an attribute also equals the attribute’s importance.)
An alternative method of obtaining the partworth functions {fip} is traditional conjoint
analysis. Green and Srinivasan (1990) provide a detailed discussion of the advantages and
disadvantages of alternative methods for measuring preference structures.
16
Elicitation of Attribute Perceptions
For some objective attributes (e.g., style of a cellular phone, i.e., bar or flip), the perceptual
bias induced by brand-building activities would be minimal. For such attributes, it is not
necessary to collect attribute perception ratings. For other attributes (e.g., reliability, durability,
quality of manufacture), each respondent provides attribute perception ratings sijp on an 1 to 10
scale for each attribute of each familiar brand. The measurement error in attribute perception
ratings can be reduced by using a multi-item approach to measurement: each attribute can be
worded in a few different ways, and the perception ratings sijp can be obtained as the average of
the ratings on the multiple items.
Linking Multiattribute Preference to Overall Preference
The multiattribute brand preference measure m(s)ij cannot be directly compared with uij
because the two are not on a comparable scale. Suppose the overall preference uij are rescaled (if
needed) so that ∑∈
=iCj
iju 100 . This allows us to relate uij to m(s)ij through the following logit-like
model:
(13) ))((exp
))(exp(
ikk
iCk
ijj
ij
sm
smu
βφ
βφ
+
+=
∑∈
*100,
where φj and β are parameters to be estimated. Since m(s)ij captures the multiattribute-based
preference, φj can be interpreted as capturing the effects of non-attribute preference resulting
from user imagery or brand personality (e.g., Aaker 1997) that go beyond product attributes.
Equation (13) can be estimated directly by nonlinear least squares. In the empirical application,
17
we simplified the estimation by converting equation (13) into a linear form in the parameters by
applying the following log-centering transformation (Nakanishi and Cooper 1982):
(14) ),)()(()(~log iijj
i
ij smsmu
u−+−= βφφ
where iu~ denotes the geometric mean of uij, φ denotes the average of the non-attribute constant
φj, and i
sm )( denotes the average of m(s)ij across the familiar brands for each individual i (j∈Ci).
(This regression was estimated for each set of familiar brands.) The non-attribute constant φj is
determined only up to an additive constant so that φ1 can be set equal to zero without loss of
generality. One can potentially allow for consumer heterogeneity in the parameters φj and β
using hierarchical Bayes methods.
Based on Equation 13, we can convert m(s)ij into a scale that is comparable to uij’s using the
following equation:
(15)
∑∈
=
iCkik
ij
ij
sm
smw
))(ˆexp(
))(ˆexp(
β
β*100,
where β̂ is the parameter value estimated in Equation 14. The transformed variable wij can be
interpreted as the constant-sum, attribute-based preference of individual i for brand j. (Note that
the sum of wij’s over the set of familiar brands of individual i equals 100 as uij’s do.) Letting nij
denote the non-attribute preference of individual i for brand j, we can separate out nij from uij
using the following equation:
18
(16) ijijij wun −= .
Equation 16 is close in spirit to Equation 3 in Park and Srinivasan (1994, p. 274). To understand
the sources of non-attribute-based preference qualitative market research methods may be
employed (Keller 1998, Park and Srinivasan 1994).
Linking Overall Preference to Brand Availability
In the conceptual model presented in Figure 1, we recognized the indirect effects of
enhancing a brand’s equity on its choice probability through improved pull-based brand
availability (refer to arrows f, g, and h in Figure 1). We now describe a method that allows us to
update the pull-based brand availability in response to a change in overall brand preference.
We start by postulating that a brand’s overall availability is determined by both push and pull
(refer to the lower left circle in Figure 1). Letting Aj denote brand j’s overall availability and Pj
denote brand j’s push-based availability, brand j’s pull-based availability is given by Aj - Pj.
From the trade’s perspective, brand j’s pull-based availability is driven by how much the
brand increases the overall attractiveness of the product category in the mind of customers. Let
us define the overall attractiveness of the product category to customer i as follows:
(17)
= ∑
∈ iCkikiC
uI )ˆexp(ln α ,
where α̂ is the parameter value estimated in Equation 10. Note that iC
I is the same as the
inclusive value (Maddala 1983) used in nested-logit model. Grover and Srinivasan (1992, p. 81)
explain its usefulness as a measure of the product-category attractiveness. To understand how
19
brand j affects the category attractiveness, suppose that brand j becomes unavailable. The
overall attractiveness of the category without brand j can be expressed as
(18)
= ∑
−∈−
}{}{
)ˆexp(lnjiCk
ikjiCuI α .
Let Vij denote the change in the overall attractiveness of the category due to the unavailability of
brand j:
(19) }{ jiCiCij
IIV −−= .
Note that Vij is zero if customer i is not aware of brand j because iC
I equals }{ jiC
I − in that case.
Averaging Vij across customers (weighted by category usage quantities) yields Vj that can be
interpreted as an average loss of category attractiveness due to the unavailability of brand j. In
other words, Vj represents how much the category attractiveness can be enhanced by adding
brand j to the product category. Thus it captures a key aspect of the trade’s decision to stock the
brand.
We relate brand j’s availability to push and pull as follows:
(20) jjj VPA θ+= ,
where θ is a parameter to be estimated by regression. As remarked earlier, the values for push-
based availabilities Pj are obtained from a group of industry experts. We use the average of
industry experts’ answers as the value for Pj.
Figure 2 shows a schematic representation of the proposed approach.
20
Figure 2 about here
MEASURING BRAND EQUITY AND ITS SOURCES
Having presented the basic building blocks of the proposed approach, we now discuss how to
determine the impact of three sources of brand equity – brand awareness, attribute perceptions,
and non-attribute preference – and the base brand’s choice probability (p’).
Determining the Impact of Awareness on Brand Equity
The impact of awareness-based source of brand equity is defined as awijij pp − where
awijp denotes customer i’s probability of choosing brand j had the customer’s level of awareness
for the brand equaled merely the push-based level of awareness. To illustrate, suppose that the
current awareness level of brand j and the push-based awareness level are 50% and 10%,
respectively. To compute awijp , we need to make 40%p (50% – 10% = 40% points) of the
respondents unaware of brand j. We can achieve this by running a simple simulation procedure
described below:
Adjust Awareness: Continuing with the previous numerical example, we start by randomly
drawing 40%p of the respondents from those who are currently aware of brand j and “make
them” unaware of the brand. In effect, 90% (= 50% + 40%) of the respondents are now unaware
of the brand. For these respondents, awijp is set to zero. For the remaining 10% of the
21
respondents who are aware of the brand, awijp will not necessarily be the same as pij because the
lower brand awareness in the market will decrease brand availability, thus lowering the brand
choice probability as shown below (refer to arrows f and h in Figure 1).
Adjust Availability: We can predict a change in brand availability due to a change in brand
awareness as follows: first, for each respondent, obtain awijV using Equation 19. For those 90%
who are not aware of brand j, awijV is zero. For the remaining 10% who are aware of brand j,
awijV equals ijV . Hence, aw
jV which is the average of awijV across all respondents will be lower
than j
V because of the 40%p of the respondents whose awijV was set to zero. Using Equation
(20), we obtain awjA as follows:
(21) )(ˆ aw
jjj
aw
jVVAA −−= θ ,
where θ̂ is the parameter value estimated in Equation 20.
For those 10% who are aware of the brand, we compute awijp by substituting aw
jA for jA in
Equation 10. (Recall that for the remaining 90%, awijp was set to zero in the previous step.)
Because of randomness in the simulation outcomes (i.e., which particular respondents were made
unaware of the brand), we need to repeat the above sampling procedure a number of times, say
30, and find the average (weighted by category purchase quantities) of awijij pp − across
respondents5. This is an estimate of how much brand awareness contributes to brand equity.
22
Determining the Impact of Attribute Perception Biases on Brand Equity
The impact of attribute-based source of brand equity is defined as aijij pp − where a
ijp
denotes customer i’s probability of choosing brand j had the customer’s subjectively held product
attribute perception biases for the brand equaled those for the base brand.
To describe a method to determine aijp , let us denote by a
ijsm )( respondent i’s multi-attribute
preference for brand j had the respondent’s product attribute perception bias for the brand
equaled that for the base brand. We define product attribute perception bias as jpijp
os − where jp
o
represents brand j’s objectively measured attribute value for attribute p. (Both sijp and ojp are
measured on the same measurement scale.) Note that since sijp = ojp + (sijp – ojp), m(s)ij can be
written as ∑=
−+=q
pjpijpjpipij
osofsm1
)]([)( .
For the purpose of determining the product attribute perception biases, we choose as the base
brand the particular brand in the product market having minimal brand-building efforts6. Using
subscript b to denote the base brand, the product attribute perception bias for the base brand
equals sibp – obp. We define aijsm )( as follows:
(22)
−+
−+=
∑
∑
=
=
otherwise, )],([
, brand base theof aware is pondent res if )],([)(
1
1
bpbpjp
q
pip
q
pbpibpjpip
a
ij
osof
biosof
sm
where sbp denotes the average of sibp among those respondents who are aware of the base brand b.
Note that the second part of the definition given in Equation 22 is necessary because some of the
5 Empirically we have found that 30 simulation iterations are sufficient to product an accurate result. 6 To the extent such a base brand used some brand-building efforts, our measure of attribute-based equity will be (conservatively) biased downwards.
23
respondents who are familiar with brand j may not be familiar with the base brand. (Recall that
we obtained attribute perception ratings for the respondent’s familiar brands only.) If the
respondent did not provide an attribute perception rating for the base brand, we utilize the
average sbp instead of sibp. By substituting aijsm )( for ijsm )( in Equation 15, we can obtain a
ijw .
Then we can obtain aiju by adding nij to
a
ijw .
We need to update the brand availability factor as well (i.e., arrow g in Figure 1) using the
procedure described in the previous subsections (i.e., Equations 17 to 21). Specifically, using the
updated overall preference aiju , obtain a
iCI , a
ijV and a
jV . Substituting a
jV for aw
jV in Equation 21
yields the updated brand availability factor .ajA Then we can compute a
ijp by substituting aiju
and ajA in Equation 10.
Determining the Impact of Non-Attribute Preference on Brand Equity
The impact of non-attribute-based source of brand equity is defined as nijij pp − where n
ijp
denotes customer i’s probability of choosing brand j had the customer’s non-attribute preference
for the brand (i.e., nij) equaled that for the base brand (i.e., nib). For the purpose of determining
nijp , we choose as the base brand the particular brand in the product market with the lowest level
of brand building efforts. To estimate nijp , we need to obtain a new brand preference measure n
iju
with nij replaced with nib. From Equation 16, we can obtain niju as follows:
(23)
+
+=
otherwise, ,
, brand base theof aware is respondent if ,
ijb
ijibnij wn
biwnu
24
where nb denotes the average of nib among those respondents who are aware of the base brand b.
Then we need to compute the updated brand availability factor njA that corresponds to n
iju
(refer to arrow g in Figure 1 and Equations 17 to 21). Finally, substituting niju and n
jA into
Equation 10 yields nijp .
Determining p’ (Base Brand’s Choice Probability)
The incremental choice probability ijij pp ′− is a key component in our definition of brand
equity (see Equation 2). Since determining the base brand’s choice probability requires a direct
application of the procedures explained in the previous subsections, we merely outline them here
using the previous numerical example (i.e., the awareness level of brand j and the push-based
awareness level being 50% and 10%, respectively). Table 2 summarizes how we operationalize
the base brand.
First, randomly draw 40%p of the respondents from those who are currently aware of brand j
and make them unaware of the brand. Then set Vij’ and pij’ to zero for those 90% of the
respondents who are not aware of the brand. Second, for the remaining 10% of the respondents
who are aware of brand j, find ijsm )( ′ using Equation 22 using the attribute perceptions biases of
the base brand. Then obtain wij’ by substituting ijsm )( ′ in Equation 15 and then obtain uij’ using
Equation 23 using the non-attribute preference of the base brand. And find Vij’ that corresponds
to uij’. Third, obtain Vj’ which is the average of Vij’ across all respondents and find Aj’ by
substituting Vj’ for aw
jV in Equation 21. Finally, for those 10% who are aware of brand j, obtain
pij’ by substituting uij’ and Aj’ in Equation 10. We need to repeat the above sampling procedure a
25
number of times, say 30, and use the average results.
Table 2 about here
What-if Analysis
EQUITYMAP allows brand managers to conduct a variety of what-if analyses to determine
the effects of a brand-building effort on any of the three sources on brand equity and profitability.
For example, brand managers can assess and compare the impact of increasing brand awareness
by x%, improving perceptions on a product attribute by an amount y, enhancing non-attribute-
based preference by an amount z, or any combination thereof, and decide on the best course of
action. The procedure for what-if analysis is a straightforward extension of the procedures for
computing the impact of the three sources of brand equity except that we use the what-if levels
of awareness, attribute perceptions, or non-attribute preference instead of those of the base brand.
For the sake of brevity, we describe a procedure for one scenario, which can be generalized
for analyzing many other scenarios. Suppose that brand j wants to increase its level of brand
awareness to 50% from a current level of 30% and improve its perceived level of product
attribute k up to the level of a leading brand, which we denote as brand l. We can analyze the
scenario using the following simulation procedure.
In the first stage, we increase the awareness level by randomly drawing 20%p (50% - 30% =
20% points) of the respondents from those who are currently unaware of brand j and “make
them” aware of the brand. For the remaining 50% of the respondents who are still unaware of
the brand, we set their Vij and pij to zero.
In the second stage, we update m(s)ij for those 50% of the respondents who are aware of
brand j (including those 20% who were made aware of the brand in the previous stage) by
26
substituting silk for sijk in Equation 12, holding perceived levels of other attributes fixed. Note
that those 20% who were made aware of brand j did not provide attribute perception ratings for
the brand at all. For these respondents, we utilize the average of the attribute perception ratings
provided by those 30% who were originally aware of the brand.
In the third stage, we update wij and uij for those 50% of the respondents who are aware of
the brand using Equations 15 and 23. For those 20% who were made aware of brand j, we utilize
the average of the nij from those 30% who were originally aware of the brand.
In the final stage, we obtain a new value for Vj which is the average of Vij across all
respondents using Equations 17 to 20 and then update Aj using Equation 21. Then for those 50%
who are aware of the brand, we update pij using Equation 10. We need to repeat the above
sampling procedure a number of times and use the average results.
AN ILLUSTRATIVE APPLICATION
We present results from applying EQUITYMAP to Korea’s digital cellular phone market.
Korea occupies a significant position in the global cellular phone market. Since it adopted a
digital cellular phone technology in 1996, its cellular phone market has been expanding at a
phenomenal rate. In just two years the number of subscribers in Korea has surpassed a 10
million mark in a country of 45 million people. Korea ranked fifth in the world (after the United
States, Japan, China, and Italy) in terms of the number of subscribers.
This product market offers an excellent setting for illustrating the proposed method for three
reasons. First, since digital cellular phones are high-tech products introduced recently, many of
the benefits offered by these products (e.g., signal reception capability, durability) are inherently
difficult for customers to evaluate. Thus brand equity is likely to be substantial. Second, major
competitors in this market seem to have strengths and weaknesses varying across the three
27
sources of brand equity proposed earlier, thus providing an ideal context to demonstrate the
diagnostic capability of the proposed method. Finally, the digital cellular phones are durable
goods, which represents a departure from previous brand equity measurement studies (e.g.,
Kamakura and Russell 1993, Park and Srinivasan 1994) that had applied their methods to
consumer packaged good categories.
Brands and Product Attributes
We considered four major players in the market (Samsung Anycall, LG Freeway, Motorola
MicroTac, and Qualcomm) who accounted for approximately 90% of the share in Korea’s digital
cellular phone market. Based on existing market research results provided by the client firm, we
chose seven product attributes. They are length, style (bar or flip), battery hours, signal reception
capability,7 durability, voice-activated dialing feature (present or absent), and price.
Data
As described earlier, EQUITYMAP utilizes data from three sources: client firm, customer
survey, and industry experts. In what follows, we describe the data collection procedures we
followed for the present study.
Client Firm: The client firm provided information on the objectively measured attribute levels of
competing models8 except for two attributes: signal reception capability and durability. For these
7 Signal reception capability means whether or not the subscriber can use his/her digital cellular phone in an area with a weak signal. 8 Since two of the brands included in the current study, Samsung Anycall and LG Freeway, were marketing four distinct models each, converting the objectively measured attribute levels into ratings was not straightforward. We calculated the weighted averages of the objectively measured attribute levels using each model’s relative sales volume as the weight.
28
two attributes, we obtained objective laboratory test results published by Customer Protection
Board in Korea which is an organization similar to Consumers Union in the United States. The
client firm also provided the data on availability (Aj) which measured each brand’s share of shelf
space in a typical retail outlet of cellular phones.
Customer Survey: Personal interviews of 281 users of digital cellular phones were conducted by
a commercial market research firm in four cities in Korea in December 1997. The customer
survey measured the following: awareness of brands, last brand purchased, attribute perception
ratings for familiar brands, attribute importance ratings, overall brand preferences, and
demographics. Since we have already described measurement scales used for attribute
perception, attribute importance, and overall brand preference ratings in the “MEASUREMENT
AND ESTIMATION” section, two other aspects of the survey warrant further explanation.
We obtained attribute perception ratings for two attributes only – signal reception capability
and durability – because all the other attributes were considered “objective” attributes in the
sense that customers’ subjective perceptions of these attributes would closely match their
objectively measured levels.
Each respondent provided attribute perception ratings for the set of brands she was aware of.
There are different ways of measuring brand awareness ranging from recognition to recall to top-
of-mind. Existing literature does not provide a clear guidance on which measure of awareness to
choose. Thus we measured brand awareness using the following two criteria. First, if the
respondent was not able to recognize a brand, she was considered unaware of the brand. Since
the purchase decision process for this high-tech product would be highly involving for most
consumers and there were only four major brands, the use of recognition rather than recall as a
measure of brand awareness seems appropriate. (In other contexts, recall may be more
appropriate.) Second, given the importance of brand knowledge in creating brand equity (Keller
29
1998, p.46), a customer may not be considered aware of the brand if she does not have a certain
level of knowledge about the brand despite the fact that she recognizes the brand. We
operationalized this second criterion by reclassifying a respondent as unaware of the brand if the
respondent was not able to provide attribute perception ratings for both “subjective” attributes,
i.e., signal reception capability and durability. For example, out of 89 respondents who were
able to recognize Qualcomm, three respondents could not provide attribute perception ratings for
Qualcomm’s signal reception capability and durability, thus reducing the number of respondents
who were aware of Qualcomm down to 86.
Industry Experts: A panel of six industry experts provided ratings on push-based availability and
push-based awareness of brands. The experts were drawn from a leading cellular phone
manufacturer and a leading telecommunication company in Korea. All of them had at least three
years of experience in marketing or sales of cellular phones. Bivariate inter-judge correlations of
expert ratings suggest that the ratings are fairly reliable: the average inter-judge correlation
coefficients for the push-based availability and awareness ratings were 0.93 and 0.61,
respectively. The lower inter-judge correlation coefficient for the push-based awareness resulted
from differing ratings for Motorola which once held a dominant position in the analog cellular
phone market (which was being phased out) but entered the digital cellular phone market much
later than competitors. The experts held differing views on how much of Motorola’s brand
awareness could spill over from the analog to the digital market given its belated entry.
Excluding ratings on Motorola, the average inter-judge correlation coefficient for push-based
awareness increased to 0.89. Table 3 shows the average ratings used in the subsequent analyses.
Table 3 about here
30
Estimation Results
We estimated Equations 10, 14, and 20 which serve as the basic building blocks of the
proposed approach. Table 4 presents parameter estimates. All the key parameters ( and in
Equation 10, in Equation 14, and in Equation 20) are statistically significant and have the
expected signs.
In estimating Equation 10, we included an “all other brands” term in the logit model to
estimate the effect of all other brands on the choice probability because there were other smaller
brands in this product market in addition to the four brands. (The respondent-reported market
shares accounted for by the four brands were 87.2%.) However, because the customer survey did
not collect an overall preference measure for these “all other brands”, the exp( uij+ Aj) term
in Equation 10 for “all other brands” was replaced by a parameter (cf. Park and Srinivasan
1994), that is, if the respondent’s brand fell into the “all other brands” category, we estimated the
choice probability for these “all other brands” pi,other by
(24)
)(exp,
kikCk
otheri
Au
p
i
γαδ
δ
++=
∑∈
.
Table 4 about here
Brand Equity
Table 5 presents each brand’s overall choice probability (p), base brand’s choice probability
(p’), and equity (p – p’). Note that although the choice probabilities are determined at the
31
individual level, we report the average probabilities in Table 5.
To obtain p’, we followed the approach described in the earlier section by setting brand
awareness and availability to their push-based levels for the brand in question. For the purpose
of determining the impact of attribute perception biases and non-attribute preference (see
Equations 22 – 23), we chose Qualcomm as the base brand because of its very little brand-
building efforts. We iterated the calculations 30 times and used the average as the estimate of p’.
The four brands show substantial differences with respect to brand equity. For example,
without its brand equity, Samsung Anycall would have achieved 25.6% of market share only.
However, its vigorous brand-building efforts brought in additional 28.2%p of market share,
leading to its 53.8% of market share. In contrast, Qualcomm’s brand equity contributes 2.0%p of
market share only.
Strictly speaking, brand equity should be given in profit terms as in Equation 2. We
converted the incremental choice probabilities in Table 5 into profit terms using the data
provided by the client firm. For example, our calculation shows that Samsung Anycall’s brand
equity contributes about 103 million dollars of contribution margin to its bottom line in 1997
alone. It compares favorably to the brand’s 30 million dollars of advertising expenditure since its
launch in 1994.
Table 5 also provides brand strength, a measure related to brand valuation (see Equation 9).
It indicates that brands with high market shares in this particular product market tend to
command higher brand strengths. However, it is interesting that the three smaller brands’
strengths are not very different from each other despite their widely differing market shares. We
note, in particular, that Qualcomm has a greater brand strength compared to Motorola despite its
smaller market share (p).
The measures of brand equity provided by EQUITYMAP were judged to have good face
validity by the managers involved with the study. We assessed convergent validity of the brand
32
equity measure by comparing it with the alternate measure proposed by Srinivasan (1979). Since
Srinivasan’s method does not provide individual-level estimates, convergent validity was
assessed only at the aggregate level. Specifically, we compared brand equity obtained by
EQUITYMAP with the “brand-specific effects” estimated from the TRANS procedure developed
in Srinivasan (1979). The two sets of brand equity estimates correlate at 0.97, indicating that
EQUITYMAP has very good convergent validity.
Table 5 about here
Sources of Brand Equity
One of the important advantages of the proposed method lies in its capability of offering
diagnostic information to brand managers by showing where the brand equity comes from.
Table 6 presents the impact of the three sources of brand equity. (As with p’, we iterated
calculations 30 times and used the averages.) For example, Samsung Anycall’s market share
would have been reduced by 23.4%p had its level of brand awareness equaled the push-based
awareness level. Similarly, its market share would have been reduced by 9.1%p and 2.4%p,
respectively, had its attribute perception biases and the non-attribute preference equaled those of
the base brand, Qualcomm.
The awareness-based equity determines the impact of the enhanced awareness, had it been the
only factor affecting the brand’s equity holding other sources of brand equity constant. In a
similar way, the attribute-based (non-attribute-based) equity measures the impact of enhanced
attribute (non-attribute) perceptions had it been the only factor. The total brand equity, however,
is the simultaneous result of all three factors. The three components summed together overstate
the total equity as can be seen by summing each row of Table 6.
33
The results in Table 6 show an interesting pattern across the four brands. Among the three
sources of brand equity, brand awareness contributes to brand equity by far the largest, followed
by attribute perceptions and non-attribute preference. This is consistent with existing conceptual
literature (e.g., Aaker 1991, 1996; Keller 1998) which postulates brand awareness as a
cornerstone of brand equity. In this durable product category attribute-based equity is larger than
non attribute-based equity, while Park and Srinivasan (1994) found that non attribute-based
equity was considerably larger than attribute-based equity for two consumer-packaged goods
(toothpastes and mouth washes). It is also interesting to note that LG Freeway almost entirely
depends on brand awareness for its brand equity and shows weaknesses with respect to attribute
perceptions and non-attribute preference vis-à-vis Samsung Anycall and Motorola MicroTac.
Table 6 about here
What-If Analysis
The proposed method provides what-if analysis capability since it measures brand equity and
its three sources in a unified framework based on choice probabilities. To illustrate the what-if
analysis capability, consider Qualcomm that has the lowest brand equity. Suppose that it
considers the following two strategies to build its brand equity: (1) enhancing its brand
awareness from its current level of 30.6% to 60% and increasing its brand availability from its
current level of 10% to 30%, or (2) enhancing its brand awareness to 50%, improving its
perceived signal reception capability up to the level of Samsung Anycall, and increasing its
brand availability to 30%. EQUITYMAP predicts the impact of each of the contemplated
strategies on its market share. After iterating the calculation 30 times, we obtained the average
as an estimate of new p. The results show that strategy (1) would increase Qualcomm’s market
34
share from its current 3.0% to 7.7% and strategy (2) would increase its market share to 6.8%p.
For simplicity, let us assume that these two strategies cost the same. Thus strategy (1) is
predicted to produce the greater impact on its market share and bottom line. This is consistent
with our finding that brand awareness is the most important source of brand equity in this
product market.
CONCLUSIONS
We propose EQUITYMAP, a new survey-based method for measuring, analyzing, and
predicting a brand’s equity in a product market. We define brand equity at the firm level as the
incremental profit per time period obtained by the brand in comparison to a brand with the same
product and price but with minimal brand-building efforts.
We conceptualize a brand’s equity as arising from the following three sources: brand
awareness, attribute perceptions, and non-attribute preference. In addition, the proposed
approach takes into account the impact of brand equity on the increased availability of the brand.
Based on this conceptualization, EQUITYMAP measures brand equity at the individual
customer level in terms of incremental choice probabilities, i.e., the difference between an
individual customer’s overall choice probability for the brand and his or her choice probability
for the underlying product with merely its push-based availability and awareness. Aggregating
the incremental choice probabilities across customers (or a segment of customers) and
multiplying it with the product’s margin yields a measure of brand equity in profit terms.
In addition to providing a summary measure of a brand’s equity in profit terms,
EQUITYMAP can evaluate the relative contributions toward the incremental choice probability
from the three sources, thus helping brand managers better understand and evaluate the sources
35
of brand equity. In addition, EQUITYMAP permits a variety of what-if analyses evaluating the
bottom-line impact of alternative brand-building strategies.
We believe that EQUITYMAP offers the following key contributions: (i) it provides a
conceptual framework for thinking about brand equity and relating it to its three sources, (ii) it
offers a method that simultaneously measures and provides an understanding of brand equity,
(iii) it determines the profitability implications of the current level of brand equity, (iv) it
provides a way of evaluating alternative strategies for enhancing brand equity, and (v) if
implemented as a tracking system, it provides a method for goal setting (e.g., levels of awareness
or attribute perceptions), monitoring, enhancing, and managing brand equity over time.
The major substantive findings from applying the proposed approach to Korea’s digital
cellular phone market are as follows. First, among the three sources of brand equity, brand
awareness contributes to brand equity by far the largest, followed by attribute perceptions and
only to a small extent, non-attribute preference. Second, the impacts of a brand’s equity on the
leading brand’s market share and profit are substantial. The methodological finding of this study
is that the proposed approach appears to have good face validity and convergent validity.
Limitations and Extensions
Any survey-based method, including that proposed here, involves measurement errors.
However, by aggregating the analyses over a large sample of consumers, their impact over the
overall equity measures should be small. For example, measurement errors are present in the
self-explicated approach to the customer preference structure measurement. However, good
validity results have been reported previously for the self-explicated method (e.g., Green and
Srinivasan 1990, p. 10). The choice model employed in calculating choice probabilities in the
illustrative application (see Equation 10) relies on the actual brand purchased to estimate its
36
parameter. Some errors exist in the estimate. For example, the model does not currently
incorporate price promotions because it uses a constant “regular” price across all the respondents.
This problem results from the inability of a customer survey to obtain accurate information about
the actual prices paid by the respondents and the store environment at the time of purchase.
Using both the survey and scanner panel data from the same respondents could solve the
problem, but doing so may not be possible in many situations. Future research should allow for
customer heterogeneity in the parameters of the probabilistic choice model.
We conceptualized non-attribute preference as a potentially important source of brand equity,
and previous research (e.g., Park and Srinivasan 1994; Aaker 1996) has found that brand
associations unrelated to product attributes can be a significant driver in differentiating a brand
from its competitors. However, more research needs to be done to uncover the underlying
dimensions (e.g., Aaker 1997) and measure the relative impact of each of the dimensions.
Another avenue for further research would be to relate the customer-based measure of brand
equity such as the one developed here to a stock market valuation measure of brand equity.
Although the proposed approach offers the incremental annual profits due to brand equity, it does
not provide the “multiplier” needed to convert the incremental annual profits to obtain a net
present value of the profit stream due to brand equity. The brand strength measure proposed in
Equation 9 could be a useful input in determining this multiplier.
37
Table 1
Summary of Notation
Symbols Definitions
eij Brand j’s equity ($) from customer i per time period (year).
pij Customer i’s probability of choosing brand j.
pij’ Customer i’s probability of choosing the base brand for brand j.
rj Brand strength (i.e., weighted average probability of purchasing of brand j
among its customers).
uij Customer i’s overall preference for brand j.
Aj Availability for brand j.
Ci Set of brands customer i is aware of.
sijp Customer i’s subjectively perceived level of brand j on product attribute p.
ojp Objectively measured attribute level of brand j on product attribute p.
m(s)ij Customer i’s multiattribute preference for brand j based on sijp.
fip Customer i’s partworth function on attribute p.
wij Constant-sum attribute-based preference of customer i for brand j.
nij Non-attribute preference of customer i for brand j.
icI Overall attractiveness of the product category to customer i.
}{ jicI − Overall attractiveness to customer i of the product category without brand j.
Vij Increment in overall attractiveness of the product category to customer i due
to brand j.
Pj Push-based availability factor for brand j. aw
ijp Customer i’s probability of choosing brand j had the customer’s level of
awareness for the brand equaled the push-based level of awareness. a
ijp Customer i’s probability of choosing brand j had the customer’s product
attribute perception biases for the brand equaled those for the base brand. n
ijp Customer i’s probability of choosing brand j had the customer’s non-
attribute preference for the brand equaled that for the base brand.
38
Table 2
Operational Definition of the Base Brand
Items Operationalizations
Objective product attributes and price (ojp) Objective product attributes and price of the brand (ojp)
Availability
Awareness
Attribute perception biases (sijp – ojp)
Non-attribute-based preference (nij)
Push-based availability of the brand
Push-based awareness of the brand
Attribute perception biases of the brand in the product
market with minimal brand building effort (sibp – obp)*
Non-attribute-based preference for the brand in the
product market with minimal brand building effort
(nib)*
* If respondent i is not familiar with the base brand, sibp is replaced by the average perception sbp, and nib is
replaced by its average nb.
39
Table 3
Availability and Awareness of Cellular Phones in Korea
Brands Availability (%) Awareness (%)
Total Push-based* Total Push-based*
Samsung Anycall 50.0 36.7 100.0 60.0
LG Freeway 30.0 16.7 100.0 25.8
Motorola MicroTac 10.0 8.3 100.0 43.3
Qualcomm 10.0 6.7 30.6 10.0
* Average ratings from industry experts to the following question: In your best judgment, what would have been the
levels of the brand’s availability (in terms of the brand’s share of shelf space in a typical retail outlet of cellular
phones) and its awareness had the brand not conducted any brand-building activities and relied entirely on the
current level of push through the channel (by the salesforce)?
40
Table 4
Parameter Estimatesa for Equations 10, 14, and 20
Equation 10 Parameters Estimates
b
N LL
0.036 (7.784) 0.028 (6.212) 4.649 (3.529) 281 -342.634 Equation 14
Group 1c Group 2d Parameters Estimates Estimates
Samsung
e
LG
Motorola
Qualcomm N Adj. R2
0.006 (10.806) 0.000 -0.058 (-1.112) 0.009 (0.179) -0.005 (-0.078) 344 0.332 Equation 20f
0.007 (17.606) 0.000 -0.084 (-2.989) 0.011 (0.355) n. a. 585 0.399
Parameters Estimates
N
20.972 (3.255) 4
a t-values in parentheses. b Parameter to capture “all other brands” (see explanation in text). c Respondents who were aware of all four brands (N=4 brands x 86 respondents). d Respondents who were aware of Samsung Anycall , LG Freeway and Motorola MicroTac only
(N=3 brands x (281–86) respondents).
e Samsung, LG, Motorola, and Qualcomm denote brand-specific dummy variables (with
Samsung set to zero for both groups).
f R2 is not meaningful for this model with no intercept term.
41
Table 5
Average Brand Equity and Related Measures
Brands
p*
p’ Equity
(p-p’)
Brand
Strength
Samsung Anycall 0.538 0.256 0.282 0.573
LG Freeway 0.184 0.044 0.140 0.244
Motorola MicroTac 0.117 0.035 0.082 0.188
Qualcomm 0.030 0.010 0.020 0.215
* Estimated p for “all other brands” is 0.131.
Table 6
Sources of Brand Equity
Brands
Awareness
(p – paw)
Attribute
Perceptions
(p – pa)
Non-Attribute
Preference
(p – pn)
Total b
(p – p’)
Samsung Anycall 0.234a
(67.0%)
0.091
(26.1%)
0.024
(6.9%)
0.282
LG Freeway 0.140
(90.3%)
0.015
(9.7%)
0.000
(0.0%)
0.140
Motorola MicroTac 0.070
(61.4%)
0.033
(28.9%)
0.011
(9.6%)
0.082
Qualcomm 0.020
(100%)
0.000c
(0.0%)
0.000c
(0.0%)
0.020
a Percentage ratio of the impact of Samsung Anycall’s brand awareness to its own brand equity
=0.234/(0.234+0.091+0.024)=67%.
b From Table 5. The sum of the three sources does not equal to the total because of overlap (see explanation
in text)
c Qualcomm served as the base brand when assessing the impact of attribute perception biases and
non-attribute preferences.
e
Preferenc
Sources of Brand Equity
42
Figure 1: A Conceptual Model of Brand Equity
d
Enhanced
Brand Image*
Push-based
Brand Awareness
Enhanc
A
A
Base Brand
Choice
Probability
Incremental
Choice Probability
Probability Product
Enhanced
Brand Awareness
h
f g
c
b
*Includes favorable att
e
Push-
based
vailability
ed (Pull-based)
vailability
ribute and non-attribute perceptions
Brand Availability
Brand Choice
a
e
Brand Preferenc
43
Figu
re 2
: EQ
UIT
YM
AP
– Sc
hem
atic
Rep
rese
ntat
ion
of th
e M
easu
rem
ent A
ppro
ach*
“obj
ecti
ve”
attr
ibut
e va
lues
o j
p
bran
d aw
aren
ess
mul
tiat
trib
ute
pref
eren
ce s
truc
ture
f ip
attr
ibut
e pe
rcep
tion
s s i
jp
attr
ibut
e pe
rcep
tion
bia
ses
s ijp
- o
jp
mul
tiat
trib
uted
pr
efer
ence
m
(s) ij
(12
) w
ij (1
3)-(
15)
non-
attr
ibut
e-ba
sed
pref
eren
ce
n ij
(16)
push
-bas
ed
avai
labi
lity
Pj
over
all
pref
eren
ce
u ij
pull-
base
d av
aila
bilit
y V
j (1
7) –
(20
)
avai
labi
lity
Aj
(20)
choi
ce p
roba
bilit
y p i
j (1
0)
* N
umbe
rs in
par
enth
eses
ref
er to
equ
atio
n nu
mbe
rs in
the
text
.
REFERENCES
Aaker, David (1991), Managing Brand Equity, New York: The Free Press.
_________ (1996), Building Strong Brands, New York: The Free Press.
_________ and Erich Joachimsthaler (2000), Brand Leadership, New York: The Free Press.
Aaker, Jennifer L. (1997), “Dimensions of Brand Personality,” Journal of Marketing Research,
34 (August), 347-356.
Birkin, Michael (1994), “Assessing Brand Value,” in Paul Stobart (ed.) Brand Power, New York:
New York University Press, 209-223.
Blattberg, Robert C. and John Deighton (1996), “Manage Marketing by the Customer Equity
Test,” Harvard Business Review, (July - August), 136-144.
Bucklin, Randolph E., Gary J. Russell, and V. Srinivasan (1998), “A Relationship between
Market Share Elasticities and Brand Switching Probabilities,” Journal of Marketing Research,
35 (February), 99-113.
Farquhar, Peter H. (1989), “Managing Brand Equity,” Marketing Research, 1 (September), 24-33.
Feldwick, Paul (1996), “What Is Brand Equity Anyway, and How Do You Measure It?” Journal
of the Market Research Society, 38 (April), 85-104.
Green, Paul E. and V. Srinivasan (1990), “Conjoint Analysis in Marketing Research: A Review of
New Developments,” Journal of Marketing, 54 (October), 3-19.
Grover, Rajiv and V. Srinivasan (1992), “Evaluating the Multiple Effects of Retail Promotions on
Brand Loyal and Brand Switching Segments,” Journal of Marketing Research, 29 (February),
76-89.
Kamakura, Wagner A. and Gary J. Russell (1993), “Measuring Brand Value with Scanner Data,”
International Journal of Research in Marketing, 10 (March), 9-22. 45
46
Keller, Kevin Lane (1998), Strategic Brand Management, New Jersey: Prentice-Hall.
Maddala, G. S. (1983), Limited Dependent and Qualitative Variables in Econometrics,
Cambridge: Cambridge University Press.
Marketing Science Institute (2000), Research Priorities 2000-2002, Cambridge, MA: Marketing
Science Institute.
Meschi, Robert L. (1995), “Value Added: Refinements in Our Brand Valuation Methodology,”
Financial World, 164 (August 1), 52-53.
Nakanishi, Masao and Lee G. Cooper (1982), “Simplified Estimation Procedures for MCI
Models,” Marketing Science, 1 (Summer), 314-322.
Nedungadi, Prakash (1990), “Recall and Consumer Consideration Sets: Influencing Choice
Without Altering Brand Evaluations,” Journal of Consumer Research, 17 (3), 263-276.
Park, Chan Su and V. Srinivasan (1994), “A Survey-Based Method for Measuring and
Understanding Brand Equity and Its Extendibility,” Journal of Marketing Research, 31 (May),
271-288.
Silk, Alvin J. and Glen L. Urban (1978), “Pre-Test-Market Evaluation of New Packaged Goods:
A Model and Measurement Methodology,” Journal of Marketing Research, 15 (May), 171-
191.
Simon, Carol J. and Mary W. Sullivan (1993), “The Measurement and Determinants of Brand
Equity: A Financial Approach,” Marketing Science, 12 (Winter), 28-52.
Srinivasan, V. (1979), “Network Models for Estimating Brand-Specific Effects in Multi-Attribute
Marketing Models,” Management Science, 25 (January), 11-21.
Srivastava, Rajendra K. and Allan D. Shocker (1991), “Brand Equity: A Perspective on Its
Meaning and Measurement,” Technical Working Paper, 91-124 (October), Cambridge, MA:
Marketing Science Institute.
47