Park 1994
Park 1994
Park 1994
SRINIVASAN*
The authors develop a new survey-based method for measuring and under-
standing a brand's equity in a product category and evaluating the equity
of the brand's extension into a different but related product category. It uses
a customer-based definition of brand equity as the added value endowed
by the brand to the product as perceived by a consumer. It measures brand
equity as the difference between an individual consumer's overall brand pref-
erence and his or her brand preference on the basis of objectively meas-
ured product attribute levels. To understand the sources of brand equity, the
approach divides brand equity into attribute-based and nonattribute-based
components. The method provides the market share premium and the price
premium attributable to brand equity. The survey-based results from apply-
ing the method to the toothpaste and mouthwash categories show that the
proposed approach has good reliability, convergent validity, and predic-
tive validity.
Brand equity is the added value endowed by the brand to rate of new product failures has led firms to acquire, li-
the product (Farquhar 1989). Although the idea of using a cense, and extend brand names to a degree unseen in previ-
name or a symbol to enhance a product's value has been ous decades.
known to marketers for a long time, brand equity has A review of the literature on brand equity shows that, at
gained renewed interest in recent years. Brand managers re- a conceptual level, some agreement exists as to what is
alize that after years of look-alike advertising and overcop- meant by brand equity. Several authors have provided defi-
ying with me-too brands, they now live in a world of prod- nitions of brand equity that are broadly consistent with Far-
uct parity. The ensuing price competition through short- quhar's (1989) definition of equity as the value added by
term price promotions reduces the profitability of brands, the brand to the product (e.g., Aaker 1991; Kamakura and
leading manufacturers to examine ways to enhance loyalty Russell 1993; Keller 1993; l..euthesser 1988; Simon and Sul-
toward their brands. In addition, faced with the increasing livan 1993; Srinivasan 1979; Srivastava and Shocker
power of retailers, manufacturers of consumer products re- 1991).
alize that having the strongest brands is vital to strengthen- Unlike the developments at a conceptual level, however,
ing their presence with retailers. Furthermore, the escalation the extant literature does not provide a satisfactory measure-
of new product development costs coupled with the high ment method for understanding the sources of brand equity
such as favorably biased attribute perceptions. An under-
* Chan Su Park is an Assistant Professor of Marketing, College of Busi- standing of where the equities of the firm's and competi-
ness and Economics, Hankuk University of Foreign Studies, Seoul, Korea. tors' brands come from is obviously essential for a brand
V. Srinivasan is Ernest C. Arbuckle Professor of Marketing and Manage- manager to enhance his or her brand's equity relative to
ment Science, Graduate School of Business, Stanford University. The au- those of competitive brands. Some of the previously pro-
thors thank MI NRfC Inc. for assistance in collecting the data used in this
study. Professors Kevin Keller and James Lattin and the editors and review-
posed measurement approaches take the firm's perspective
ers of this special issue of JMR provided helpful comments on earlier ver- and measure brand equity at the firm level. For example,
sions of this article. The authors are especially grateful to Professor Allan Simon and Sullivan (1993) estimate brand equity as the in-
Shocker for detailed comments on the article. This research was supported cremental cash flows that accrue to the firm due to its invest-
in part by the Doctoral Program, Graduate School of Business, Stanford
University, and a Business School Trust Faculty Fellowship awarded to the
ment in brands. Because their method relies on data aggre-
second author. gated to the firm level, their estimate of brand equity is not
very useful to brand managers managing an individual
brand in a multibrand firm operating in multiple product allowed to vary across brands. Note that the Swait and col-
categories. leagues definition of brand equity as total utility is different
A classical method for estimating brand equity is to in- from the commonly used definition of brand equity (Far-
clude brand name as a factor in the full-profile method of quhar 1989) as the added value endowed by the brand to
conjoint analysis performed at the individual level (Green the product. (Swait and colleagues also define an intangible
and Srinivasan 1978, 1990). By estimating brand equity at component of utility analogous to Farquhar's definition of
the individual rather than aggregate or segment level, brand brand equity; however, their use of perceived attribute val-
managers can aggregate the individual-level measures to ues in the utility function means that the component of
quantify both the mean and standard deviation of brand eq- brand equity arising from favorably biased attribute percep-
uity for any segment of interest. As explained subsequently, tions is inappropriately excluded from their intangible com-
starting with individual-level measures also avoids difficult ponent.) The Swait and colleagues approach also makes the
aggregation problems encountered in estimating the market unsupported assumption that attribute weights are homoge-
share premium and price premium attributable to brand eq- nous across consumers. More importantly, the Srinivasan
uity. A difficulty with the conjoint analysis method in the (1979), Kamakura and Russell (1993), and Swait and col-
context of brand equity measurement is that the conjoint leagues (1993) approaches have the important limitation
card-sort task can lead to unrealistic product profiles (i.e., that they do not break down the estimated equity into its
given that consumers have prior perceptions of brands on components that can be related to factors such as favorably
multiple attributes, orthogonal designs that combine brand biased perceptions. Thus, empirical results based on these
name with the attributes are likely to result in some unreal- methods will have somewhat limited managerial usefulness
istic profiles). Another important concern with conjoint anal- in terms of understanding the sources of brand equity and
ysis is that it does not provide an understanding of the suggesting directions for enhancing it.
sources of brand equity. We develop a new survey-based method for measuring a
Several authors have proposed useful methods for estimat- brand's equity at the individual consumer level and offer
ing a brand's equity (e.g., Kamakura and Russell 1993; Srini- brand managers an indication of the sources of brand eq-
vasan 1979; Swait et al. 1993). Srinivasan (1979) defines uity. Like the previous approaches (e.g., Kamakura and Rus-
brand equity (which he calls "brand-specific effect") as the sell 1993; Srinivasan 1979; Swait et al. 1993), the proposed
component of overall preference not explained by objec- measurement method rests on the multiattribute preference
tively measured attributes. He estimates brand equity by model.' However, unlike the previous approaches (except
comparing actual choice behavior with those implied by util- for the conjoint analysis method), it provides an individual-
ities obtained through conjoint analysis with product attrib- level measure of brand equity. The proposed approach uses
utes, but no brand names. His method avoids the problem a survey procedure to obtain each individual's overall
of unrealistic product profiles mentioned previously with brand preference and his or her multiattributed brand pref-
the conjoint method, but has a limitation of providing, at erence based on objectively measured attribute levels. After
best, segment-level estimates of brand equity. Kamakura scaling both preference measures to cents (dollars), the pro-
and Russell (1993) use the segmentwise logit model (Ka- posed approach subtracts the multiattributed brand prefer-
makura and Russell 1989) on single-source scanner panel ence based on the objectively measured attribute levels
data to estimate brand equity. They first estimate segment- from the overall brand preference to derive individual-level
level brand preferences by removing the effects of short- measures of brand equity.
term advertising and price promotions and then obtain seg-
Going one step further, the proposed method divides
ment-level brand equity estimates as residuals from a regres-
brand equity into attribute-based and nonattribute-based
sion equation relating segment-level price-adjusted brand
components, thus providing the brand manager with an in-
preferences to objectively measured product attributes. An
dication of different plausible bases of brand equity. The at-
attractive aspect of this method is that the researcher ob-
tribute-based component of brand equity captures the im-
tains brand equities from real consumer choices in the mar-
pact of brand-building activities on the consumer's attribute
ketplace rather than by relying on survey-based subjective
perceptions. In other words, the attribute-based equity incor-
methods. Like Srinivasan's approach, the Kamakura-
porates the difference between subjectively perceived and
Russell method has a limitation of offering, at best, seg-
objectively measured attribute levels. The non attribute-
ment-level estimates of brand equity. Furthermore, the Ka-
makura-Russell method of computing brand equity as resid- based component of brand equity captures brand associa-
uals in a regression equation tends to understate the actual tions unrelated to product attributes (e.g., the masculine
variation of equities across brands. For example, if there image conveyed by the Marlboro Man). Kamakura and Rus-
were as many attributes as the number of brands, all their sell (1993) also conceptualize brand equity as arising from
brand equities will be zero. two sources. However, their scanner data-based method
Swait and colleagues (1993) propose an approach to meas- does not permit the separation of overall equity (which they
uring brand equity, operationalized as the monetary equiva- call "intangible value") into its two components. This two-
lent of the total utility a consumer attaches to a brand. An ad- component decomposition of brand equity provides valua-
vantage of their approach is that it obtains brand equity
IBy attribute, we mean a product characteristic (e.g., flavor of a mouth-
from consumer choices to hypothetical choice sets rather wash such as cinnamon or mint) or benefit (e.g., breath freshening). For a
than by relying on survey-based subjective measures. An ad- useful discussion of the attribute-benefit linkage and its relation to brand eq-
ditional advantage of their method is that price sensitivity is uity, see Ratneshwar, Shocker, and Srivastava (1993).
Measuring and Understanding Brand Equity 273
ble information because the two components often relate to vasan 1979) based on objective product attributes:
different actions by brand managers. For example, although q
image-oriented advertising can create the nonattribute-
based component, attribute-specific advertising can create
(1) Uij=P~l fip(ojp)'
the attribute-based component of brand equity. The present where
study also provides a method for assessing the impact of a
brand's equity on its market share and profit margin in con- uij = preference of individual i for brandj G= 1, 2, ... , b),
trast with previous studies that did not explicitly address fip = function denoting individual i's partworth (i.e., utilities
for different levels) of attribute p,
those issues. The market share premium due to brand eq- 0jp = objectively measured level of brand j on the pth
uity tells brand managers how much of a brand's current attribute,
market share is attributable to the brand's equity, holding q = number of the product attributes, and
price fixed. The price premium due to brand equity pro- b = number of brands.?
vides the additional price the firm is able to charge cur- Because Equation 1 rests on objectively measured attrib-
rently for the brand, holding market share fixed. The mar- ute levels to predict brand preference, one can expect that
ket share and price premiums constitute meaningful sum- two brands that have approximately the same objective at-
mary measures of brand equity because they closely relate tribute levels should be preferred equally by consumers.
to brand profitability. However, two brands with almost identical physical charac-
We also develop a model that provides brand managers teristics often end up with very different outcomes in terms
with diagnostic information as to how a brand extension of consumer preferences. Allison and Uhl (1964) demon-
into a related product category builds on (or detracts from) strate a classic example of this phenomenon in their compar-
the equity of an established brand. The proposed model is ison of blind and brand taste tests of beer. Their results
based on linkages between the attribute-based (and the non- clearly show that regular beer drinkers could not identify
attribute-based) components of equities in the core and the their favorite brands in blind taste tests. Former Chrysler
extension product categories. Thus, when estimated using Corporation Chairman Lee Iacocca provides another real
data from existing brand extensions, the model offers in- life example. He said in the first commercial for the "Advan-
sights as to how the equity of an established brand can help tage: Chrysler" corporate campaign:
a brand extension succeed. In addition, brand managers can
Two cars comeoff the same assembly line,in the same
use the method to evaluate how the equity and consumer American plant. A Japanese nameplate goes on one
preferences for a new brand extension are determined in the [Mitsubishi Eclipse], an American nameplate goes on
extension product category. Brand managers who want to ex- the other [Plymouth Laseror Eagle Talon], andpeople
tend their established brands can use the model to select an preferthe Japanese version.
extension product category from a set of candidate catego-
ries. A potential limitation of the proposed method is that These examples suggest the existence of significant dis-
substantial measurement errors can arise because it is based crepancies between consumers' overall preferences for a
on a consumer survey. Consequently, we provide evidence brand (i.e., uij) and their multiattributed preferences based
indicating that the approach has good reliability, conver- on objectively measured attribute levels (i.e., }:;qp=lfi/ojp»'
gent validity, and predictive validity. The apparent discrepancies may result from different levels
We begin by developing the conceptual basis and pro- of brand-building activities (other than product and price)
pose a method for measuring brand equity, followed by the across different brands.' We operationalize brand equity as
method for evaluating the equity and consumer preferences the difference between an individual consumer's overall
for a brand extension. Then we apply the method to the brand preference and his or her multiattributed preference
toothpaste and mouthwash categories and assess reliability based on objectively measured attribute levels. Letting eij de-
and convergent and predictive validities. We close with a dis- note brand equity and u(O)ij denote consumer preferences
cussion of the findings, limitations, and directions for fur- based on objectively measured attribute levels, brand equity
ther research. is defined as
q
MEASURING AND UNDERSTANDING BRAND EQUITY (2) eij=uij-~l fip(0jp)=uij-u(0)ij'
We start with the conceptual framework underlying the
measurement method. We then propose a method for collect- Because overall preferences of individual i are measured
ing survey data, followed by the calculation of brand equity on an interval scale, we can, without loss of generality, ex-
and its two components. We close with the derivation of press them in mean-centered form so that individual i's
market share and price premiums resulting from brand mean overall preference ui = O. The proposed approach meas-
equity. ures equities of the b brands relative to each other so that
e
the mean equity i = O. Thus, a brand with a significant
Conceptual Framework
DriIe mathematical formulation uses the partworth model of preference
From the individual perspective, brand equity is the incre- not necessarily to imply that the attributes are categorical but to achieve gen-
mental preference endowed by the brand to the product as erality: The partworth model incorporates the vector and ideal-point mod-
els as special cases (Green and Srinivasan 1978).
perceived by an individual consumer. We start with the 3Undoubtedly, the discrepancies also may result from measurement er-
base case of a multiattribute model of a consumer's prefer- rors. To examine the extent of such errors, we assess reliability and valid-
ence for a brand in a product class (e.g., Shocker and Srini- ity of the brand equity measure.
274 JOURNAL OF MARKETING RESEARCH, MAY 1994
amount of brand building activities nevertheless could ob- charities also can create brand associations unrelated to prod-
tain a negative (relative) equity because other brands have uct attributes.
u e
done an even better job. Because both j and j are zero, it The distinction between the two components of brand eq-
follows from Equation 2 that the multiattributed prefer- uity holds both conceptual and managerial importance.
ences based on objectively measured attributes also should First, previous research on brand equity (e.g., Leuthesser
be expressed such that the mean u(o), =O. Furthermore, for 1988) develops a proposition that strong equity provides
the subtraction in Equation 2 to be meaningful, both Ujj and "meaning" to the product beyond actual product features.
u(o)j' must be expressed on a common scale (unit of meas- Furthermore, several authors (Hirschman and Holbrook
urement). As will be seen subsequently, this is accom- 1982; Zajonc and Marcus 1982) question the applicability
plished by expressing both Ujj and u(o)jj on a cent (dollar)- of the multiattribute preference model for some product cat-
metric scale so that equity e jj is also measured on the same egories (e.g., perfumes, fashion apparel) because product at-
cent-metric scale. tributes perceived by consumers account for only a small
The literature on brand equity (e.g., Aaker 1991; Keller portion of consumer preferences in these product
1993) has argued that brand associations underlie brand eq- categories.
uity. We postulate that brand associations contribute to Second, the two components relate differently to diverse
brand equity in two different ways. First, brand associations elements of the marketing mix. For example, though image-
related to product attributes create an attribute-based compo- oriented advertising and selective distribution (e.g., high-
nent of brand equity that is based on the differences be- end specialty stores) may contribute to the nonattribute-
tween subjectively perceived attribute levels and objec- based component, attribute-specific advertising may contrib-
tively measured attribute levels. In other words, various ute to the attribute-based component of brand equity. Thus,
brand building activities can bias an individual consumer's providing the two components' relative contributions to
perceived level of a product attribute away from its objec- brand equity helps the brand manager judge the suitability
of the marketing mix elements to the brand. In addition,
tive level. Second, brand associations create a non-tprod-
breaking down the attribute-based component into individ-
uct) attribute-based component of brand equity, which is a
ual attributes enables one to understand in detail the sources
part of a brand's overall preference unrelated to product at-
of the attribute-based component of brand equity.
tributes. For example, the masculine image conveyed by
Letting sijp denote individual i's subjectively perceived
the Marlboro Man has nothing to do with product attrib-
level of brand j on attribute p and u(s\j denote I,qp=/iP(SjP)'
utes, and yet it may playa significant role in accounting for
that is, individual i' s preference for brand j based on subjbc-
preferences for the brand. tively perceived attribute levels, the mathematical formula-
The marketing literature provides ample theoretical and tion of the two components is written as
empirical evidence supporting the two-component view of
brand equity. For example, Kamakura and Russell (1993) de- (3) eij = Ujj - U(O)ij
velop a conceptual model that divides brand equity into two
= [u jj - U(S)ij] + [u(s)ij - u(o)ij]
components, although their empirical method does not
break equity down into the two components. Research on
product ambiguity (e.g., Ha and Hoch 1989; Hoch and Ha
where n jj and a jj denote the nonattribute-based and attribute-
1986) shows that when a consumer sees ambiguous evi-
based components of brand equity, respectively. As re-
dence, prior hypotheses suggested by advertising bias con-
marked previously, the proposed approach measures the eq-
sumers' subjectively perceived attribute levels away from uities of the b brands relative to each other so that the mean
true levels. Research on product knowledge (e.g., Alba and
Hutchinson 1987) suggests that consumer knowledge may
e a
equities j, nj' and j are zero for each individual !.:. Corre-
spondingly, the mean preferences u j ' u(o)j' and urs), are
influence the attribute-based component of brand equity. equal to zero for each individual.
For example, more knowledgeable consumers might tem-
per halo effects (e.g., Beckwith and Lehmann 1975) with Method
more detailed evaluation of individual product attributes. The basic idea behind the measurement method is to obtain
And expertise may enable a consumer to avoid an errone- Ujj' u(s\j' and U(O)jt' on the cent-metric scale (i.e., preference
ous influence by an advertising message. measured in cents dollars]) so that subtracting one from an-
The nonattribute-based component can be created by other becomes appropriate. The proposed measurement
image-oriented (as opposed to attribute-specific) advertis- method requires two sets of data. One data set provides ob-
ing through its likable ad executions (Mitchell and Olson jectively measured attribute levels of different brands in a
1981) or the mere-exposure effects of advertising frequency given product category (i.e., 0jll)' Broadly, data on "objec-
(Ray 1973; Zajonc 1968). User imagery (i.e., what type of tively measured" attribute levels can come from laboratory
person uses the brand), usage consumption imagery (i.e., tests, blind consumer tests, or expert evaluations. Note that
what it feels like to use the brand [Hirschman and one can use one method for a subset of product attributes
Holbrook 1982]), and usage situation imagery (i.e., under and another method for a different subset of attributes, de-
what types of situations the brand is used) can endow a pending on the nature of the attributes.
brand with a personality such as masculinity and youthful- The second data set is survey-based and is collected from
ness, thus creating a nonattribute-based component. Endorse- a random sample of current users of the product category,
ments by celebrities and sponsorships of athletic events and drawn from the geographic market(s) of interest to the inves-
Measuring and Understanding Brand Equity 275
Table 1 brand (rather than with respect to the least preferred brand)
EXAMPLE OF CENT-METRIC PREFERENCE CALCULATIONS to avoid an extreme response.
We now apply a linear transformation to the ratings data
Uij
on the second column of Table 1 to express preferences on
Preference uij Actual (price- uij a cent-metric scale. Let ~Pi denote the answer to the price
Brands Ratings (cent-metric) Prices adjusted) (normalized) premium question given by respondent i, R. denote the over-
A 10 125 219 -94 24.5 all preference rating (1 to to) given by\espondent i to
B 9 112.5 219 -106.5 12 brand j, and mi denote the overall preference rating given
C 7 87.5 214 -126.5 -8
6 75 219 -144 -25.5
by respondent i to the second least preferred brand. Then
D
E 1 12.5 134 -121.5 -3 we obtain u ij by
IlP.
(4) u .. = __1_ *R ..
I) lO-ffi 1)'
i
tigator. The consumer survey consists of three parts: overall
preference rating (which provides the u r), preference struc- Continuing with the previous example, suppose that the re-
ture me.asurement (which provides fil:) and is needed for spondent gave 50 cents as the price premium for Brand A
computmg the u(o\j and u(s)ij)' and attnbute perception rat- over Brand D. Then, from the second column of Table 1
ing (which provides the srp)' To minimize any systematic U iA equals 50/(10-6)*10 = 125 (cents). Similarly, u equal~
iD
bias from a particular orddring of these three parts of data 50/(10-6)*6 = 75 (cents). (Note that the difference in pref-
collection, we recommend that the ordering be randomized. erences between Brands A and D is equal to 50 cents as per
In the following section, we describe how the informa- the original response.) The third column of Table 1 gives u.,
tion on overall preferences, preference structures, and percep- for each of the five brands. Note that so far we have ob~
tion ratings can be obtained. Tracking a brand's equity over tained the uij assuming that all the brands cost the same. Be-
time is an expensive proposition. Consequently, for each of cause that is generally not the case, we must subtract the ac-
the three types of information, we first propose cost-effec- tual price from the uij' leading to the fifth column. Finally,
tive telephone survey-based methods for collecting the because the u ij is determined only up to an additive con-
data. These are the methods adopted in the empirical study u
stant, we normalize them so that I,bj=l u ij = 0 (i.e., i = 0).
reported subsequently. However, in each case, we also sug- The
.
last column of Table 1 shows toe u.,y we will use to de-
gest more expensive alternative methods that have the poten- nve eij'
tial to reduce measurement error. An alternative way of collecting cent-metric overall pref-
erences is to ask the respondent to state the price premium
Elicitation of Overall Preferences he or she would be willing to pay for the preferred brand in
To obtain the overall preference rating, we recommend pro- every pair of brands (Pessemier et al. 1971). (Consequently,
ceeding in two steps. First, each respondent receives an iden- a ten-brand product category would require the elicitation
tical set of several brands available in the product cate- of price premiums for (10 X 9)/2 =45 paired comparisons.)
gory.' Holding prices equal, the respondent identifies the Construct a b X b matrix D in which d f provides the price
most preferred brand (which is then given ten points) and premium the respondent is willing to pay for column brand
the least preferred brand (which is then given one point), j over row brand i. (If i is preferred over j, d.. would be nega-
. N 1)
and then provides overall preference ratings for the remain- tive. ote that the diagonal elements d.. are zero and d.. =
• • 11 ' Jl
ing brands. (We do not use zero for the least preferred -d i) . Averagmg the numbers m each column provides the
brand because that tends to be interpreted by some respon- least-squares estimates of cent-metric preferences.
dents as the brand having no value.) For example, the sec-
Elicitation of Preference Structure
ond column of Table 1 shows overall preference ratings by
a respondent. We recommend a self-explicated approach to obtain the mul-
Second, the respondent provides the price premium he or tiattribute preference structure {fi } in Equation 2 for each
she would be willing to pay for the most preferred brand individual i in the sample. (For a 'detailed discussion of the
(i.e., Brand A from Table 1) over the second least preferred self-explicated preference structure measurement, see
brand (i.e., Brand D from Table 1). The question can be Green and Srinivasan 1990.) This measurement consists of
asked as follows: "Suppose that both Brand A and Brand two steps: desirability ratings (1 to 10) for the different lev-
D cost the same. Then from your previous responses you els of each categorical attribute, and importance ratings for
would prefer A to D. Now, suppose that the price of A is all attributes. For desirability ratings, we identify the most
higher than the price of D. How much more would you be and least preferred levels for each categorical attribute and
willing to pay for A than for D?" We suggest obtaining the anchor the scale values of the most preferred level at to and
price premium with respect to the second least preferred the least preferred level at 1. Then we elicit the relative de-
sirability of the remaining levels on the 1 to 10 scale. (For
"One could argue that we should collect data for only those brands in the price and other "continuous" attributes, we assume a vec-
respondent's consideration set. However, an important reason for a respon- tor model, that is, that preference is a linear function of the
dent to exclude a brand from his or her consideration set is unfavorable attribute level.) For importance ratings, we first identify the
brand perceptions. From a managerial perspective, it is important to under-
critical attribute, which is defined as the one for which the
stand the reduced equity that results from unfavorable brand perceptions.
Consequently, our approach collects data on all brands with which the re- improvement from the least preferred attribute level to the
Spondent is at least somewhat familiar. most preferred level is most valued, and set its importance
276 JOURNAL OF MARKETING RESEARCH, MAY 1994
rating at 10. Then we obtain the importance ratings (0 to brand equity, we suggest open-ended questions asking the re-
10) for the other attributes using the critical attribute as the spondent to tell the first few things that come to mind in de-
anchor (Srinivasan 1988). The self-explicated approach ob- scribing a brand. Although the open-ended questions are in-
tains the partworth functions by multiplying the attribute im- adequate to uncover fully various dimensions of brand asso-
portance ratings by the desirability ratings for the levels of ciations, examining what types of associations appear and
each attribute. what fraction of respondents share a particular association
Traditional conjoint analysis (Green and Srinivasan could offer useful insights. Depth interviews and projective
1978, 1990) is an alternative method of obtaining the part- techniques (Green, Tull, and Albaum 1988, p. 160-71) can
worth functions {fj }. (See Green and Srinivasan 1990 for a provide an even more comprehensive picture.
detailed discussionof the advantages and disadvantages of
Calculating Brand Equity and Its Two Components
alternative methods for measuring preference structures.)
The self-explicated approach to preference structure meas- Using the conjoint analysis framework (Green and Srini-
urement is considerably simpler than traditional conjoint vasan 1978), we can represent preference based on objec-
analysis methods such as the full-profile approach (card- tively measured attribute levels as an additive partworth util-
sort) and trade-off tables. Despite its simplicity (or perhaps ity function. Let price expressed in cents (to be consistent
because of it), the self-explicated approach produces a pre- with uij' which is in cents) be the qth attribute. Then
dictive ability (with respect to actual behavior of individu- q-l
als) comparable to those provided by the full-profile
method (Leigh, MacKay, and Summers 1984; Srinivasan
(5) U(O)jj=~1 fip(Ojp)-WjqOjq'
1988; Wright and Kriewall 1980). The self-explicated pref-
erence structure measurement is also especially well suited where W jq denotes the importance weight individual i at-
to telephone surveys, because it does not require a card-sort taches to the price attribute. (Recall that we assume prefer-
task or the completion of trade-off tables, as in more tradi- ence to be a linear function of price.) To obtain u(o)j' on a
tional conjoint surveys. cent-metric scale, we use the procedure proposed by 'Srini-
vasan (1979): Divide Equation 5 throughout by the weight,
Elicitation of Attribute Perceptions W jq, for price, thus making the new weight for price equal
Each respondent provides attribute perception ratings (i.e., to -1, so that u(o)j' now expresses preference in cents.
Sijp in Equation 3) on an 1 to 10 scale for each attribute of (Note that if price ofbrand j increases by I cent, then respon-
each brand.l To deal with the inter-respondent scale usage dent i's preference for that brand decreases by I cent, so
differences (response tendencies), each individual's attrib- that u(o)jj is measured in cents.) Because u(o\ is deter-
ute perception ratings for different brands are then normal- mined only up to an additive constant, we normalize it so
ized so that Ibj=ISjjp = 0 for each p.6 Hereafter, Sijp refers to that Ib'=lU(O)ij = 0 for each respondent (as was done for
normalized attribute perception ratings. The measurement ~ij)' Su6tracting u(o)jj from uij gives a measure of equity, e ij,
error in attribute perception ratings can be reduced by using III cents."
a multi-item approach to measurement: Each attribute can Dividing e ij into attribute-based (a i) and nonattribute-
be worded in a few different ways, and the perception rat- based (nij) components is straightforward (see Equation 3).
ing Sjjp can be obtained as the average of the ratings on the Replacing o'p with Sijp in Equation 5 gives U(S)ij' After ex-
multiple items. pressing U(Sjij in cents and normalizing it (i.e., Ibj=1U(S)ij =
0), we subtract u(s)ij from uij to obtain nW Likewise, subtract-
Additional Survey Data ing u(O)ij from u(s)ij yields ai,.
In addition to the three major data collection steps, the sur- Aggregating the individual measures of brand equity
vey should include questions to obtain information about over N respondents gives an aggregate-level (average)
each respondent's usage quantity per period (e.g., a week), brand equity measure. The aggregate-level brand equity e·
last brand purchased, and demographic and psychographic (measured in cents) is given by J
5For some objective attributes (e.g., number of doors in a car), the de- Market Share and Price Premiums due to Brand Equity
gree of the perceptual bias induced by a brand seems minimal. Thus, it is Because a brand's profitability is the primary concern of a
not necessary to collect attribute perception ratings for such attributes.
6Standardization (subtracting the mean and dividing by standard devia- brand manager, the market share and price premiums attrib-
tion) not only eliminates response tendencies in mean-ratings but also ac-
counts for response tendencies in variances. However, performing standard- 7Recall that ei , expresses relative brand equities so that ei = O. As a re-
ization becomes problematic if a large number of respondents have zero var- sult, if eij equals)0, it does not mean that the brand has no equity. It only
iance in the ratings on at least one attribute. means that that brand's equity equals the average over all brands.
Measuring and Understanding Brand Equity 277
utable to brand equity are useful summary measures of (i.e., brand equity measure for the store brand) for e ik while
brand equity that would help the brand manager assess its the equity measures for other brands remain unchanged.
managerial significance. Then, from our definition of brand equity, u(o)ik + e is would
The market share premium is the difference between the be an estimate of consumer i' s preference for brand k if
current market share and a benchmark share that the brand brand k had obtained the same equity as the store brand.
would have obtained at its current price had there been no Then the benchmark share, &kS' equals
equity. To estimate the benchmark share for brand k, we in-
clude in the survey a brand that is a priori presumed to have (9) m
A
- I~lPr'k
1= 1 S
q.1
ks- Q '
minimal brand equity. A store brand is an example because
typically retailers do not invest much in brand-building ac-
nvities." Then we estimate the benchmark share of brand k where Pr iks comes from the logit model in Equation 8 after
by the share that would have resulted if brand k had ob- replacing e ik with eis' Finally, the share premium equals
tained the same equity as the store brand. If a store brand is /I /I
(10) Amk = mk - ID ks •
not appropriate, a "weak" national brand or hypothetical
brand could be used. Similarly, we can obtain the share premium due to the at-
It should be emphasized that the inclusion of a store tribute-based or nonattribute-based component of brand
brand (or some other benchmark) in the analysis is needed equity.
only for the purpose of obtaining the market share and price A complementary measure to share premium is price pre-
premiums due to brand equity. If one is content merely mium. We define the price premium as the difference be-
with the equity measures e, n, and a, then there is no need tween the current price of brand k and the price that brand
to include a store brand. Recall that the equities for the k should have charged to achieve the current market share
brands are measured relative to each other so that e j , nij' and if brand k had obtained minimal equity such as that for the
a.,IJ average to zero over the brands for each consumer l. The store brand. In our approach, the price premium is deter-
use of a benchmark brand can be thought of as a way of mined as follows: First, estimate the current market share at
converting the relative equity measures into more "abso- the current price. Second, as was done in the share pre-
lute" measures. mium calculation, substitute for each individual e is for e ik
The first step in the share premium calculation for brand while the equity measures for other brands remain un-
k is to estimate the current market share (mk) as the average changed. Then reduce the price of brand k by one cent and
of each individual's choice probability weighted by the estimate the market share of brand k at the reduced price.
usage quantity. Letting &k denote the estimated market We repeat this calculation until the estimated market share
share for brand k, at a reduced price equals the current market share estimate
(i.e., &k)' Then we estimate the price premium as the differ-
(7) ence between the current price and the reduced price. 10
The share (or price) premium of the store brand is by def-
inition zero, and the share (or price) premiums for other
The probability Pr ik could come from a logit model that brands will, in general, be positive, though it can be nega-
links the overall preference measure to the brand choice tive. For example, a national brand may suffer from such a
probability": negative image that the brand actually reduces the per-
Pr = exp(l3 u ik) ceived value of the product.
(8) ik
Ir=l exp(l3 u ij ) II?J= 1 exp(r.t(u(o)
.....
..+e IJ
IJ
.. )) EVALUATING THE EQUITY OF A BRAND EXTENSION
We begin by presenting a conceptual model based on the
The next step is to estimate the benchmark share. A plausi- linkages between the attribute-based (and nonattribute-
ble way to do this is to substitute, for each individual, e is based) components of equities in the core product category
and a related extension product category. We then briefly de-
SOnemight wonder if a generic rather than store brand could be used to
obtain the benchmark share. We chose not to use a generic brand because,
scribe data requirements, followed by details on estimating
first, during the past decade, consumers who tried generics generally found the conceptual model and making predictions.
quality was not up to that of national brands and came to believe that ge-
nerics were strongly inferior, with the result that generic brands capture Conceptual Model
very little share in most product categories. Thus it would not be meaning-
ful to use as a benchmark a brand with such strong negative associations.
Previous research (e.g., Aaker and Keller 1990; Boush and
Second, a generic brand (with a generic name such as "Toothpaste") may Loken 1990; The University of Minnesota Consumer Behav-
be very difficult for the respondents to relate to during the survey. In con- ior Seminar 1987) shows that a direct, positive relationship
trast, a store brand (with a store name such as "Walgreen Toothpaste") between attitude toward an established brand and attitude to-
should be easier for a respondent to relate to. ward an extension exists when consumers believe that the ex-
!1Jsing the information on the last brand the respondents report having
purchased, one can estimate the parameter ~, which maximizes the follow- tension somehow fits with the established brand. Although
ing likelihood function (Silk and Urban 1978):
10Although average equity in Equation 6 is measured in cents, e k - e,
N b &
does not provide the price premium that brand k can extract. One reason is
L=TITI(Pr.) ij, that some respondents may view the brand favorably and thus have posi-
i=1 je l 1)
tive equity, yet not buy the brand because they prefer a competitor's brand
where 8ij equals 1 if individual i last purchased brand j and equals 0 even more. Consequently, the brand is unable to extract a price premium
otherwise. from such customers despite its positive equity.
278 JOURNAL OF MARKETING RESEARCH, MAY 1994
previous research conceptualizes fit at the product level- ( 12) s(2) .. = g(s(l).. 0<2). n(l)..)
I)P If' )P' 1)
suggesting that it is based on such factors as typicality + Q. (I) + + Q. (I) + 'Yp0(2)jp + IIpn (I)ij
- t-'pO + t-'p1s
- Q. Q. (\)
ij\ t-'p2S ij2 ... t-'pqs ijq
(Boush and Loken 1990) or substitutability, complementar-
ity, and manufacturing synergies (Aaker and Keller 1990)- + 8 ijp (p = 1,2,...r),
more recent research begins to conceptualize fit along two
where p denotes an attribute of the brand extension, q de-
dimensions, such as product similarity and brand concept
notes the total number of attributes for the core product cat-
consistency (Park, Milberg, and Lawson 1991) or product-
egory, r denotes the total number of attributes for the exten-
level and image-level fit (Bridges 1990). The conceptualiza-
sion product category, and s denotes random error. To illus-
tion of fit along two dimensions is supported by empirical
trate, suppose that the extension product is a mouthwash
findings that consumers positively evaluate extensions hav-
and the core product is a toothpaste. Further suppose that
ing low product-based similarities with an established
the toothpaste has four attributes or benefits (anti-plaque
brand if the extensions have brand concept consistency or
[A], breath freshening [B], cavity prevention [C], and teeth
image-level fit.
whitening [T] capabilities) and the nonattribute-based com-
These results suggest that in consumer evaluations of a
ponent of the established brand (i.e., n(l)ij) is mostly based
brand extension, associations with an established brand-
on sex appeal. Then the perceived breath freshening capabil-
be they related or unrelated to product attributes-are impor-
ity (B) for the new mouthwash brand extension can be writ-
tant. Drawing on the two-component conceptualization of
brand equity proposed previously, we conceptualize that ten as
both the attribute-based and nonattribute-based components (13) S(2\B = I3Bo + I3B1s(l\jB + I3B2s(l)ijA + I3B3s(l)ijC +
of the equity of an established brand can lead to consumer
expectations regarding the brand extension and thus influ- I3B4 S(l\T + 'YBo(2)jB + llBn(l)ij + 8 ijB.
ence brand equity and consumer preferences for the We expect that the estimate of I3 B I' which is the coefficient
extension. for the perceived breath freshening capability of the tooth-
To distinguish the extension from the core product cate- paste, would be positive and statistically significant. How-
gory, we denote the extension category by superscript (2) ever, it is also possible that consumer perceptions of other
and the core category by superscript (I). Because nij = uij - attributes of toothpaste might affect the extension's per-
U(S)ij from Equation 3, u(2\j can be written as ceived breath freshening capability. In addition, the sex ap-
peal image of the core brand (reflected in n(l\) may posi-
(I I) tively influence the perceived breath freshening capability
of the extension.
where r denotes the total number of attributes for the exten- A similar model can be proposed for n(2)ij' However, un-
sion product category. This formulation implies that a key like the previous equation for s(2).. , n(2).. has a simple rela-
J
to understanding consumer preferences for the brand exten- tionship to n(1\ because, by defi{iition: it is not based on
sion is to investigate the determinants of S(2)ijp and n(2)ij' product attributes:
What determines the subjectively perceived attribute lev- =h(n(l)..) ='no + 'n\n(l).. + e..,
(14) n(2) ..
els for a new brand extension (S(2)ijp)? Conceptually, there ap- I) I)"" I) I)
pear to be four different bases on which an individual con- where e denotes random error.
sumer can make an inference about the likely level of attrib- Note that the proposed model assumes only a one-way
ute p of a new brand extension. First, if attribute p is com- transfer of brand equity and does not incorporate reciprocal
mon and relevant to both product categories, then the percep- effects. However, this is not a serious limitation given the
tion about attribute p of the established brand may have findings of previous studies (e.g., Keller and Aaker 1992;
strong influence on the perception about attribute p of the Sullivan 1990), which show that the reciprocal effects tend
new extension. Second, if attribute p is unique to the brand to occur mostly when the extension is made within the
extension, then to the extent that consumers have learned same category as the core product (i.e., a line extension, not
ideas or "schemata" about linkages among product attrib- a brand extension as considered here).
utes, some attribute perceptions about the established brand Given this conceptual model, the method works as fol-
may influence the perception of attribute p in the extension. lows: With two product categories (one core and one exten-
In addition, the nonattribute-based component of equity of sion) and existing brand extensions from the core to the ex-
the established brand (n(l)ij) may affect the perception about tension category, Equations 12 and 14 can be estimated
attribute p of the extension, serving as a halo. Finally, objec- using the data from the existing extensions by the firm and
tive reality (0(2)jp) seems more likely to influence subjective its competitors. The estimated relationships reveal how the
perceptions of the extension's search and experience attrib- equities of established brands help brand extensions suc-
utes than for credence attributes.'! ceed. Then to evaluate a new brand extension, we propose
A mathematical formulation of the relationship gives inserting the values of its predictor variables into the esti-
11Search attributes are those whose true levels are learned by consumers and Karni 1973). Consequently, search and experience attributes in the ex-
prior to purchase. Experience attributes are those whose true levels are de- tension are less likely to be affected by perceptions of the established
termined only after consumption (Nelson 1970). Credence attributes are brand than credence attributes.
those whose true levels cannot be learned even after consumption (Darby
Measuring and Understanding Brand Equity 279
mated equations. To assess different product categories to method for three reasons. First, some of the primary bene-
which brand extensions can be made, brand managers can re- fits offered by these products (e.g., cavity prevention and
peat these steps for several candidate extension product cat- anti-plaque capabilities) are inherently hard for consumers
egories. In the following section, we suggest a procedure to to evaluate even after usage over a long period of time, thus
estimate the model and make predictions. making consumers susceptible to (correct or incorrect)
claims made by manufacturers. For example, contrary to
the belief held by many consumers, many dentists say that
Data Requirements brushing with water and no paste, flossing, and gargling
By applying the data collection method described previ- with fluoride rinse accomplish the same results as using a
ously to two different product categories, we obtain the data toothpaste (Wall Street Journal 1992). And the Food &
needed to estimate Equations 12 and 14, with the following Drug Administration has expressed concern over anti-
two caveats. First, respondents to the survey should be cur- plaque claims in mouthwashes (Drug Topics 1990). Thus
rent users of both product categories. Second, if there al- these categories are classic examples of "credence goods"
ready have been brand extensions from one product cate- (Darby and Kami 1973), in which brand equity is likely to
gory to the other, one should include as many existing be substantial. Second, there are some existing brand exten-
brand extensions as possible because that should give broad- sions (e.g., Colgate, Close-Up, and store brands) from the
based information to the brand manager and improve the pre- toothpaste to mouthwash category, thus providing a good
dictive power of the model. If there has been no brand ex- testing ground for a model that evaluates the equity and con-
tension, one might estimate Equations 12 and 14 using hy- sumer preferences for a new brand extension. After holding
pothetical brand extensions to approximate how subjective out the most recent brand extension (Close-Up), we can es-
perceptions can be formed. (Because a hypothetical exten- timate the model using the information on the existing
sion does not, by definition, have an objective reality, Equa- brand extensions and then test the predictive validity by com-
tion 12 has to be estimated after excluding the 0(2)jp term.) paring the predictions with the actual outcomes realized for
this brand extension. Third, these categories are ones with
which most consumers are familiar. Nearly 95% of adults
Estimating the Conceptual Model in the United States use toothpaste. About 62% of adults
The model shown in Equations 12 and 14 consists of r + 1 use mouthwash and the market is growing fast (Marketing
(r is the total number of attributes for the extension product & Media Decisions 1988). The high purchase incidence in
category) equations, because Equation 12 consists of r equa- these product categories made it possible to reduce con-
tions, one for each product attribute of the extension cate- sumer survey costs.
gory. Denoting the number of respondents by N and the
number of already existing brand extensions (or hypotheti-
cal extensions) by J, N * J is the total number of observa- Brands and Product Attributes
tions used for estimating each of the r + 1 equations. To reduce telephone interview time, we limited the sur-
Although one can separately estimate each of the r + I vey to four toothpaste brands (Crest, Colgate, Aqua-fresh,
equations by ordinary least squares (OLS), efficiency of the and Close-Up) and five mouthwash brands (Scope, Lister-
parameter estimates can be improved by jointly estimating ine, Close-Up, Plax, and Colgate). These brands account for
the r + 1 equations because the regression errors in Equa- 80% and 75% of the share, respectively, as reported by the
tion 12 for one attribute may be correlated with regression survey respondents. In addition, it was necessary to include
errors for another. Procedures for estimating a system of a store brand because it serves as a benchmark brand to de-
seemingly unrelated regression equations are available in sta- rive the market share and price premiums due to brand eq-
tistical packages such as TSP (Hall, Schnake, and Cummins uity. We included either Walgreen or Rite Aid (but not both
1988). for anyone respondentj'? in the survey, depending on
which chain had a larger number of outlets in the state in
Predicting the Equity of a Brand Extension which the respondent lived. By using both store brands, we
were able to cover larger areas in the United States (32
After estimating both Equations 12 and 14, we can predict states with 85% of the U.S. population) than by using one
S(2\p and n(2\ for the new brand extension by inserting the store brand alone.
values of the ~predictor variables into Equations 12 and 14. On the basis of previous studies involving the same prod-
Then inserting the predicted values for n(2)ij and S(2)ijp into uct categories (Bass and Wilkie 1973; Haley 1968; Ran-
Equation 11 gives the estimated preferences for the new gaswamy, Burke, and Oliva 1993; Shugan 1987; Wilson,
brand extension. Also inserting the values for n(2\ and S(2)ijp Mathews, and Harvey 1975), we chose five toothpaste and
into Equation 3 yields the brand equity estimates for die four mouthwash attributes (benefits) for this application.
new brand extension. The toothpaste attributes are anti-plaque capability, cavity
prevention capability, teeth whitening capability, breath
AN ILLUSTRAIWE APPLICKIION freshening capability, and unit price; the mouthwash attrib-
utes are anti-plaque capability, breath freshening capability,
We present results from applying the proposed method
to the toothpaste and mouthwash categories. These two prod- 12Walgreenand Rite Aid are, respectively, the largest and second largest
uct categories offer an excellent context for testing the (in terms of sales) drug store chains in the United States.
280 JOURNAL OF MARKETING RESEARCH, MAY 1994
19As in ASSESSOR (Silk and Urban 1978), we have used common para- dividual-specific Cl and 13 would require much additional data. It also
meters for the whole sample. Strictly speaking, the parameters Cl and 13 would be useful to obtain overall preference data on the "other brands"
should be allowed to vary across consumers. However, the estimation of in each respondent would consider in addition to the standard set of brands.
Measuring and Understanding Brand Equity 285
for the perceived breath freshening capability of a mouth- Dependant Adjusted Coefft·
wash brand extension (see Equation 13). Now we can esti- Variable R2 Predictors Est. t-stat p-value
mate Equations 14, 16, and 17 using information on two ex- nC2\j .13 Intercept -12.78 -3.23 .0013
isting brand extensions (Colgate and the store brand, thus j n(l)ij .34 5.56 .0001
= 1,2). sC2\jA .22 Intercept -.73 -2.97 .0051
However, severe multicollinearity appeared among the s(l)ijA .37 5.25 .0001
four attribute perceptions in Equation 16. To alleviate the f(1)ijA .33 3.25 .0012
o(2)'A .15 .15 .8808
problem, we reformulated Equation 16 as follows. First, we nCtlij .10 .56 .5754
ran three simple linear regressions relating s(l)ijB' s(l)f' and
s(2)ijB .36 Intercept -.22 -2.32 .0204
S(l)ijT (each in turn as the dependent variable) to s(l ijA (as SCl)ijB .39 6.24 .0001
the independent variable) and obtained the residuals. Sec- f<1\jB .29 3.26 .0013
ond, we obtained a single factor score variable by factor- o(2)'B .66 1.37 .1706
analyzing the three residuals and retaining the only factor nCtlij .28 1.82 .0688
with an eigenvalue greater than 1. Let fCl\jA denote the fac- ~e total number of observations without a missing item was 198 (104
tor score variable computed from the three residuals. Then for Colgate plus 94 for store brand). Two "outlier" observations were
we can modify Equation 16 as deleted resulting in a sample size of 196 for each regression.
286 JOURNAL OF MARKETING RESEARCH, MAY 1994
The proposed brand equity measure is only a relative level and the effects of brand equity on the financial market
measure. Although it tells how much more equity brand A value of a firm.
has vis-a-vis brand B (or a store brand), it does not concep-
tually provide the magnitude of brand A's equity in abso- REFERENCES
lute terms. However, this should not be a serious limitation Aaker, David A. (1991), Managing Brand Equity. New York:
for brand managers because they are most likely to use the Macmillan.
proposed method to investigate the competitive situation of - - - and Kevin L. Keller (1990), "Consumer Evaluations of
their brands against major competitors. In addition, we Brand Extensions," Journal of Marketing, 54 (January), 27-41.
Alba, Joseph W. and J. Wesley Hutchinson (1987), "Dimensions
would note that this limitation is inherent in any technique
of Consumer Expertise," Journal of Consumer Research, 13
for measuring preference that has, at best, interval-scale (March),411-54.
properties. Allison, Ralph I. and Kenneth P. Uhl (1964), "Brand Identifica-
The choice model relies on the last brand purchased to cal- tion and Perception," Journal of Marketing Research, 1 (Au-
ibrate its parameters. Some errors exist in these estimates be- gust), 80-85.
cause, though the last brand purchased could differ from Bass, Frank and William L. Wilkie (1973), "A Comparative Anal-
the "usual" brand because of, for example, price promo- ysis of Attitudinal Predictions of Brand Preference," Journal of
tions, the choice model did not allow for such possibility be- Marketing Research, 10 (August), 262--69.
cause of its using a constant "regular" price across all the Beckwith, Neil E. and Donald R. Lehmann (1975), "The Impor-
tance of Halo Effects in Multi-Attribute Attitude Models," Jour-
respondents. This problem results from the inability of a con-
nal of Marketing Research, 12 (August), 265-75.
sumer survey to elicit accurate information about the store Boush, David M. and Barbara Loken (1990), "A Process Tracing
environment in terms of prices and promotions of different Study of Brand Extension Evaluation," Journal ofMarketing Re-
brands. Using both the survey and scanner panel data from search, 28 (February) 16-28.
the same respondents could solve this problem, but doing Bridges, Sheryl A. (1990), "A Categorization Model of Brand Ex-
so has been impossible in many situations (Bucklin and tensions," doctoral dissertation, Stanford University.
Srinivasan 1991), though this may change in the future. Fu- Bucklin, Randolph E. and V. Srinivasan (1991), "Determining In-
ture researchers also should allow for consumer heterogene- terbrand Substitutability Through Survey Measurement of Con-
ity in the parameters a and 13 in the probabilistic choice mod- sumer Preference Structures," Journal of Marketing Research,
els (see Equations 8 and 15). 28 (February) 58-71.
Consumer Reports (1986), "Toothpastes" (March), 144-49.
The illustrative application reported here relies on expert
- - - (1989), "New Ways to Save Your Teeth?" (August),
evaluations as one way of obtaining "objectively" meas- 504-509.
ured attribute levels. Although we were able to demonstrate Darby, Michael R. and Edi Karni (1973), "Free Competition and
in this case that the expert ratings were consistent with Con- the Optimal Amount of Fraud," Journal of Law and Econom-
sumer Reports ratings and had lower variability than the con- ics, 16 (April), 67-88.
sumer ratings, we suspect that the expert judgments were Drug Topics (1990), "Oral Care Category Driven by Therapeu-
not free from the effects of various brand-building activi- tics," (September 17),74-75.
ties, thereby leading to downward biases in the attribute- Farquhar, Peter H. (1989), "Managing Brand Equity," Marketing
based component estimate of brand equity. Brand manag- Research, I (September), 24-33.
ers, however, do not have to use expert ratings. Because Green, Paul E. and V. Srinivasan (1978), "Conjoint Analysis in
Consumer Research: Issues and Outlook," Journal of Con-
firms routinely test competitors' as well as their own prod-
sumer Research, 5 (September), 103-23.
ucts, brand managers often should have access to laboratory ---and (1990), "Conjoint Analysis in Marketing
test data. Research: A Review of New Developments," Journal of Mar-
The predictive validity tests in the illustrative application keting, 54 (October), 3-19.
were not very powerful because both the existing brand ex- - - - , Donald S. Thll, and Gerald Albaum (1988), Research
tensions (i.e., Colgate and the store brand mouthwashes) for Marketing Decisions. Englewood Cliffs, NJ: Prentice-Hall,
and the new brand extension (i.e., Close-Up mouthwash) Inc. .
had small market shares in the extension product category. Ha, Young-Won and Stephen J. Hoch (1989), "Ambiguity, Pro-
cessing Strategy, and Advertising-Evidence Interactions," Jour-
Thus, further testing and refinement of the prediction model
nal of Consumer Research, 16 (December), 354-60.
would be a worthwhile avenue for further research. Haley, Russell I. (1968), "Benefit Segmentation: A Decision-
Relating the brand equity measure to its antecedents, Oriented Research Tool," Journal of Marketing, 32 (July), 30-
such as past advertising exposures and order of entry, 35.
would be important complements to the measurement Hall, Bronwyn H., Rebecca Schnake, and Clint Cummins (1988),
method developed here. Another avenue for further re- TSP User's Guide, Version 4.1. Palo Alto, CA: TSP
search that would assist brand managers involves develop- International.
ing a method that quantifies the importances of various Hirschman, Elizabeth C. and Morris B. Holbrook (1982), "He-
brand associations unrelated to product attributes (e.g., the donic Consumption: Emerging Concepts, Methods and Proposi-
tions," Journal of Marketing, 46 (Summer), 92-101.
importance of the masculine image of Marlboro Man) just
Hoch, Stephen J. and Yong-Won Ha (1986), "Consumer Learn-
as we measure the importance of product attributes in con- ing: Advertising and the Ambiguity of Product Experience,"
sumer preferences. Finally, because the share (or price) pre- Journal of Consumer Research, 13 (October), 221-33.
mium due to a brand's equity provides inputs into the esti- Kamakura, Wagner A. and Gary J. Russell (1989), "A Probabilis-
mation of the total value of a firm's brand equity, it would tic Choice Model for Market Segmentation and Elasticity Struc-
be fruitful to extend the proposed methodology to the firm ture," Journal of Marketing Research, 26 (November), 379-90.
288 JOURNAL OF MARKETING RESEARCH, MAY 1994
- - - and (1993), "Measuring Brand Value With Shocker, Allan D. and V. Srinivasan (1979), "Multiattribute Ap-
Scanner Data," International Journal of Research in Market- proaches for Product Concept Evaluation and Generation: A Crit-
ing, 10 (March), 9-22. ical Review," Journal of Marketing Research, 16 (May), 159-
Keller, Kevin L. (1993), "Conceptualizing, Measuring, and Man- 80.
aging Customer-Based Brand Equity," Journal of Marketing, Shugan, Steven M. (1987), "Estimating Brand Positioning Maps
57 (January), 1-22. Using Supermarket Scanning Data," Journal of Marketing Re-
- - - and David A. Aaker (1992), "The Effects of Sequential search, 26 (February), 1-18.
Introduction of Brand Extensions," Journal of Marketing Re- Silk, Alvin J. and Glen L. Urban (1978), "Pre-Test-Market Evalu-
search, 29 (February), 35-50. ation of New Packaged Goods: A Model and Measurement
Leigh, Thomas w., David B. MacKay, and John O. Summers
Method," Journal of Marketing Research, 15 (May), 171-91.
(1984), "Reliability and Validity of Conjoint Analysis and Self-
Explicated Weights: A Comparison," Journal of Marketing Re- Simon, Carol 1. and Mary W. Sullivan (1993), "The Measurement
search, 21 (November), 456-62. and Determinants of Brand Equity: A Financial Approach,"
Leuthesser, Lance (1988), "Defining, Measuring, and Managing Marketing Science, 12 (Winter), 28-52.
Brand Equity," Summary of Marketing Science Institute Confer- Srinivasan, V. (1979), "Network Models for Estimating Brand-
ence, Report No. 88-104. Cambridge, MA: Marketing Science Specific Effects in Multi-Attribute Marketing Models," Manage-
Institute. ment Science, 25 (January), 11-21.
Marketing & Media Decisions (1988), "Oral Fixations," (April), - - - (1988), "A Conjunctive-Compensatory Approach to the
111-28. Self-Explication of Multiattributed Preferences," Decision Sci-
Mitchell, Andrew A. and Olson, Jerry C. (1981), "Are Product At- ences, 19 (Spring), 295-305.
tribute Beliefs the Only Mediator of Advertising Effects on Srivastava, Rajendra K. and Allan D. Shocker (1991), "Brand Eq-
Brand Attitude?" Journal of Marketing Research, 18 (August), uity: A Perspective on its Meaning and Measurement," Work-
318-22. ing Paper No. 91-124. Cambridge, MA: Marketing Science
Nelson, Phillip (1970), "Information and Consumer Behavior,"
Institute.
Journal of Political Economy, 78,311-29.
Sullivan, Mary (1990), "Measuring Image Spillovers in Umbrella-
Nunnally, J. C. (1978), Psychometric Methods. New York:
Branded Products," Journal of Business, 63 (3), 309-29.
McGraw Hill.
Park, C. Whan, Sandra Milberg, and Robert Lawson (1991), "Eval- Swait, Joffre, Tulin Erdem, Jordan Louviere, and Chris Dubelaar
uation of Brand Extensions: The Role of Product Feature Simi- (1993), "The Equalization Price: A Measure of Consumer-
larity and Brand Concept Consistency," Journal of Consumer Perceived Brand Equity," International Journal of Research in
Research, 18 (September), 185-93. Marketing, 10 (March), 23-45.
Pessemier, Edgar, Philip Burger, Richard Teach, and Douglas The University of Minnesota Consumer Behavior Seminar (1987),
Tigert (1971), "Using Laboratory Brand Preference Scales to "Affect Generalization to Similar and Dissimilar Brand Exten-
Predict Consumer Brand Purchases," Management Science, 17 sions," Psychology & Marketing, 4 (3), 225-37.
(February), B-371-B-385. Wall Street Journal (1992), "Tooth-Brushers Take a Shine To Bak-
Rangaswamy Arvind, Raymond R. Burke, and Terence A. Oliva ing Soda" (March 2), B 1.
(1993), "Brand Equity and the Extendibility of Brand Names," Wilson, David T., H. Lee Mathews, and James W. Harvey (1975),
International Journal of Research in Marketing, 10 (March), "An Empirical Test of the Fishbein Behavioral Intentions
61-75. Model," Journal of Consumer Research, 1 (March), 39-48.
Ratneshwar, S., Allan D. Shocker, and Rajendra K. Srivastava
Wright, Peter and Mary Ann Kriewall (1980), "State of Mind Ef-
(1993), "On the Managerial Relevance of a Theory of Market
fects on the Accuracy with Which Utility Functions Predict Mar-
Behavior," working paper, School of Business Administration,
ketplace Choice," Journal of Marketing Research, 19 (August),
University of Florida.
Ray, Michael L. (1973) "Psychological Theories and Interpreta- 277-93.
tions of Learning," in Consumer Behavior: Theoretical Zajonc, Robert B. (1968), "Attitudinal Effects of Mere Expo-
Sources, S. Ward and T.S. Robertson, eds. Englewood Cliffs, sure," Journal of Personality and Social Psychology Mono-
NJ: Prentice-Hall, Inc., 45-117. graph, 9, 1-27.
SAS Institute, Inc. (1985), SAS User's Guide: Statistics, Version 5 - - - and Hazel Markus (1982), "Affective and Cognitive Fac-
Edition. Cary, NC: SAS Institute Inc. tors in Preferences," Journal of Consumer Research, 9, 123-
31.