Developing and validatinga multidimensional consumer-based brand ...

Journal of Business Research 52 (2001) 1±14
Developing and validating a multidimensional consumer-based brand
equity scale
Boonghee Yooa,*, Naveen Donthub,1
aDepartment of Marketing and Business Law, G. R. Herberger College of Business, St. Cloud State University, St. Cloud, MN 56301, USA
bDepartment of Marketing, J. Mack Robinson College of Business, Georgia State University, Atlanta, GA 30303, USA
Received 1 January 1997; received in revised form 1 July 1999; accepted 6 August 1999
Abstract
Little systematic research has been done to develop a scale to measure consumer-based brand equity. The authors report the results of a
multistep study to develop and validate a multidimensional consumer-based brand equity scale (MBE) drawn from Aaker's and Keller's
conceptualizations of brand equity. A total of 1530 American, Korean American, and Korean participants evaluated 12 brands from three
product categories (athletic shoes, film for cameras, and color television sets). Multistep psychometric tests demonstrate that the new brand
equity scale is reliable, valid, parsimonious, and generalizable across several cultures and product categories. The authors discuss theoretical
and practical implications of the study. D 2001 Elsevier Science Inc. All rights reserved.
Keywords: Multidimensional brand equity scale; Multistep psychometric tests; Brand equity
In the past decade, researchers have focused a tremen-
and marketing success (Ambler, 1997). Almost every
dous amount of attention on the brand equity construct,
marketing activity works, successfully or unsuccessfully,
which refers to the incremental utility or value added to a
to build, manage, and exploit brand equity (see Aaker,
product by its brand name. Using a consumer-based beha-
1991; Keller, 1993; Yoo, Donthu, and Lee, 2000). How-
vioral view of brand equity, we define brand equity as
ever, despite this considerable amount of interest, research
consumers' different response between a focal brand and an
that identifies and attempts to understand brand equity
unbranded product when both have the same level of
phenomena has been hampered because there has been no
marketing stimuli and product attributes. The difference in
agreement regarding what brand equity is and, more im-
consumer response may be attributed to the brand name and
portant, how it should be measured.
demonstrates the effects of the long-term marketing invested
Although several studies have examined brand equity,
into the brand.
their main focus was not on developing a brand equity
Researchers have found that a product's brand equity
measure. Currently, researchers use ad hoc measures such as
positively affects future profits and long-term cash flow
price premium (Aaker, 1991), conjoint analyzed value of the
(Srivastava and Shocker, 1991), a consumer's willingness
brand name (Rangaswamy et al., 1993; Cobb-Walgren et al.,
to pay premium prices (Keller, 1993), merger and acquisi-
1995), composite multiattribute weighted scores of the
tion decision making (Mahajan et al., 1994), stock prices
brand name (Park and Srinivasan, 1994), a collection of
(Simon and Sullivan, 1993; Lane and Jacobson, 1995),
consumer-based measures (Agarwal and Rao, 1996), and a
sustainable competitive advantage (Bharadwaj et al., 1993),
scanner data-based measure (Kamakura and Russell, 1993).
Other ad hoc measures include financial values of a brand,
such as future earnings (Aaker, 1991), incremental cash
flow (Simon and Sullivan, 1993), equalization price (Swait
* Corresponding author. Tel.: +1-320-255-3161; fax: +1-320-255-
et al., 1993), and momentum accounting-based value (Far-
4061.
quhar et al., 1991). However, these measures were devel-
E-mail addresses: byoo@stcloudstate.edu (B. Yoo), ndonthu@gsu.edu
(N. Donthu).
oped without rigorous psychometric tests, and they were not
1 Tel.: +1-404-651-1043; fax: +1-404-651-4198.
parsimonious enough to manage.
0148-2963/01/$ ± see front matter D 2001 Elsevier Science Inc. All rights reserved.
PII: S0148-2963(99)00098- 3

---

2
B. Yoo, N. Donthu / Journal of Business Research 52 (2001) 1±14
The purpose of this study is to develop a multidimen-
validated or replicated in other cultures (see Netemeyer et
sional measure of consumer-based brand equity and assess
al., 1991; Durvasula et al., 1993). We develop a brand
its psychometric properties. Here, consumer-based means
equity measure with an etic approach, in which a universal
measurement of cognitive and behavioral brand equity at
measurement structure across cultures is sought using multi-
the individual consumer level through a consumer survey.
ple cultures simultaneously (Geertz, 1973). The outcome
In contrast, for firm-based measures, researchers collect
measure that an etic approach produces is functionally,
financial market, accounting, and store-level scanner data
conceptually, linguistically, and metrically equivalent across
without contacting consumers; these then identify dollar-
cultures, which provides the basis for generating valid cross-
metric and financial brand equity at the firm or brand level.
cultural comparisons (Berry, 1980; Leung and Bond, 1989;
Unlike most previous studies, this study develops a mea-
Meredith, 1993; Rosenzweig, 1994).
sure of brand equity that is reliable, valid, and parsimo-
After surveying several cultures, we assess the univers-
nious. In addition, the measure's latent structure is assessed
ality of our brand equity measurement. Confirming the
for generalizability across multiple samples drawn from
universality enables cross-cultural benchmarking of brand
several cultures, specifically, Koreans, Korean Americans,
equity, because the measure is not bound to a particular
and Americans. In particular, our measure is developed
culture or country. Cross-cultural benchmarking is useful for
using the brand equity dimensions that Aaker (1991, also
tracking performance and developing strategies in the do-
see 1996) and Keller (1993) suggest, which have been
mestic and international markets, and in the cross-cultural
popularly accepted as valid and comprehensive. However,
study of brand equity.
key questions about Aaker's and Keller's measurements
We collect data from South Korea and the US because
remain unanswered, specifically, the structural validity of
these two countries show an adequate range of cultural
the measurement.
variation. According to Hofstede's (1991) work, Koreans
A consumer-based brand equity scale is beneficial in
and Americans are different in every major cultural dimen-
several ways. First, it offers a means to test brand equity
sion. In comparison with Americans, for example, Koreans
theories. According to Aaker (1991), brand equity provides
rate high on Confucian dynamism (long-term orientation),
value to customers by enhancing their interpretation and
low on individualism, and high on uncertainty avoidance.
processing of information, confidence in the purchase deci-
Therefore, South Korea is a good counterweight to the US,
sion, and satisfaction. Brand equity also provides value to
and Korean responses should reduce potential bias that
the firm by enhancing efficiency and effectiveness of
could result from developing and validating a measure using
marketing programs, prices and profits, brand extensions,
samples drawn only from a Western culture.
trade leverage, and competitive advantage. Similarly, Keller
In the following sections, we first examine the brand
(1993) proposes that enhancing brand equity results in the
equity construct and identify its relevant dimensions as
ability to command larger margins from consumers, elicits
proposed by Aaker (1991) and Keller (1993). Next, we
increased consumer information search, and improves mar-
discuss the procedures used to generate and select scale
keting communication effectiveness, licensing opportu-
items in the pilot and main studies. Then, we assess internal
nities, and consumers' responsiveness to brand extensions.
consistency, validity, and cross-cultural metric equivalence.
A brand equity measure would allow investigation of the
Finally, we discuss the implications of the final scale to both
role of brand equity in Aaker's (1991) and Keller's (1993)
practitioners and researchers.
models. Specifically, it may be used to measure the brand
equity of existing brands, then to examine the relationship
of brand equity to the resulting firm and consumer benefits.
1. The brand equity construct
Second, the measure would be more useful for testing
consumer-based brand equity theories than other previous
Brand equity has many definitions and forms, such as
measures have been. A consumer-based brand equity study
favorable impressions, attitudinal dispositions, and beha-
needs a measure that assesses an individual customer's
vioral predilections (Rangaswamy et al., 1993); brand
brand equity. However, some of the previous measures are
loyalty, brand awareness, perceived quality, brand associa-
designed to measure brand equity of aggregate products at
tions, and other proprietary brand assets (Aaker, 1991);
the industry or firm level (e.g., Simon and Sullivan, 1993;
brand knowledge such as brand awareness and brand
Mahajan et al., 1994). Others measure an individual custo-
associations (Keller, 1993); loyalty and image (Shocker
mer's brand equity (e.g., Rangaswamy et al., 1993; Swait et
and Weitz, 1988); the added value endowed by the brand
al., 1993; Park and Srinivasan, 1994; Cobb-Walgren et al.,
name (Farquhar et al., 1991); incremental utility (Kama-
1995), but the psychometric properties have not been
kura and Russell, 1993); the difference between overall
reported or fully analyzed. As a result, the previous mea-
brand preference and multiattributed preference based on
sures may not be appropriate to studies that examine
objectively measured attribute levels (Park and Srinivasan,
consumer-based brand equity phenomena.
1994); and overall quality and choice intention (Agarwal
Many scale development studies take an emic approach,
and Rao, 1996). One important consensus among the
in which a scale is first developed in one culture, then
definitions is that brand equity is the incremental value

---

B. Yoo, N. Donthu / Journal of Business Research 52 (2001) 1±14
3
of a product due to the brand name (Srivastava and
We designed perceived quality items to assess consu-
Shocker, 1991).
mers' subjective judgment about a brand's overall excel-
Collectively, brand equity consists of four dimensions:
lence (Zeithaml, 1988) because perceived quality represents
brand loyalty, brand awareness, perceived quality of brand,
overall quality rather than individual elements of quality
and brand associations, as proposed by Aaker (1991, 1996)
(Petroshius and Monroe, 1987; Aaker and Keller, 1990;
and Keller (1993). These dimensions may be used to
Boulding and Kirmani, 1993). We adopted seven of the
explore the findings of marketing and consumer behavior
items used by Dodds et al. (1991). We designed brand
research in relation to brand equity (see Barwise, 1993);
associations items to measure ``the strength of connection to
thus, we develop a brand equity measure that capitalizes on
a brand node as a function of both the amount or quantity of
these dimensions.
processing the information received at encoding and the
Aaker (1991, p. 39) defines brand loyalty as ``the
nature or quality of the processing of the information
attachment that a customer has to a brand.'' In our study,
received at encoding'' (Keller, 1993, p. 5). On the basis of
brand loyalty refers to the tendency to be loyal to a focal
this definition, we developed six items for the overall
brand, which is demonstrated by the intention to buy the
perceptual strength of brand associations.
brand as a primary choice (Oliver, 1997). In contrast, some
previous research has focused on the behavioral aspects of
2.1. Stimuli selection
brand loyalty (e.g., Guadagni and Little, 1983; Gupta,
1988). Brand awareness is ` the ability for a buyer to
We selected three product categories as stimuli: film for
recognize or recall that a brand is a member of a certain
cameras (a low-cost, fast-replacement cycle, short-term
product category'' (Aaker, 1991, p. 61). Thus, brand aware-
experience good), athletic shoes (a medium-cost, med-
ness consists of both brand recognition and recall (Rossiter
ium-replacement cycle, medium-term experience good),
and Percy, 1987; Keller, 1993). Perceived quality is ``the
and color television sets (a high-cost, slow-replacement
consumer's judgment about a product's overall excellence
cycle, longer-term experience good). An experience good
or superiority'' (Zeithaml, 1988, p. 3). It therefore is based
is a product that consumers must actually experience,
on consumers' or users' (i.e., not managers' or experts')
through consumption or purchase, to judge its quality
subjective evaluations of product quality. Aaker (1991, p.
(Nelson, 1974). Three criteria guided the selection of
109) defines brand associations as ` anything linked in
these product categories. First, the categories were differ-
memory to a brand'' and brand image as ` a set of [brand]
ent in price range, frequency of purchase, consumers'
associations, usually in some meaningful way.'' The asso-
product involvement, and consumption situation (e.g.,
ciations have a level of strength (Aaker, 1991; Aaker and
place, time, and interaction among users). This wide
Keller, 1990; Keller, 1993), and a link to a brand will be
variety of products was selected to enhance the assess-
stronger when it is based on many experiences or exposures
ment of the cross-product applicability of the scale.
than when it is based on few (Aaker, 1991).
Second, the participants in this study, college student
consumers, were familiar with the categories. Their ex-
perience with the products enabled them to provide
2. Item generation
reliable and valid responses to a questionnaire. For ex-
ample, in both Korea and the US, more than 90% of the
After careful consideration of the literature and on the
university student consumers purchase athletic shoes.
basis of the definitions previously established, we generated
Annual market reports by research firms in Korea and
a pool of 48 candidate scale items to reflect the dimensions
the US have shown that a significant number of both
of brand equity. To establish content validity, the items were
Korean and American students experience these product
evaluated for conformity to the theoretical definitions and
categories. Third, only consumer goods were selected
redundancy. After screening the items independently, then
because of their high brand equity. Consumer goods'
jointly, we retained 22 items for initial psychometric assess-
brand value variance is explicitly attributable to the brand
ment: five on brand loyalty, four on brand awareness, seven
equity variance (Simon and Sullivan, 1993).
on perceived quality, and six on brand associations.
We surveyed 12 brands: six athletic shoes (Adidas, Asics,
We designed five brand loyalty items to capture the
LA Gear, Nike, Puma, and Reebok), four films (Agfa, Fuji,
overall attitudinal loyalty to a specific brand rather than
Kodak, and Konica), and two color television sets (Samsung
directly measuring actual brand-loyal behavior (see Gua-
and Sony). We used two criteria in selecting the brands.
dagni and Little, 1983; Gupta, 1988). We adopted and
First, the brands had to be available in both Korea and the
modified Beatty and Kahle's (1988) brand loyalty items.
US. Second, the brands had to be markedly different.
Brand awareness may be measured as brand recognition or
According to the market reports mentioned previously, the
brand recall. We measured simple brand recognition rather
market share of the selected brands in the Korean and
than brand recall. Four items were based on previous
American markets ranged from 0.5% to 12.4% in athletic
research (Srull, 1984; Alba and Hutchinson, 1987; Rossiter
shoes, 0.5% to 41.2% in film, and 0.1% to 37.2% in
and Percy, 1987).
television sets. We selected brands that held the highest,

---

4
B. Yoo, N. Donthu / Journal of Business Research 52 (2001) 1±14
lowest, and mid-level market shares. In addition, we se-
for both Koreans and Americans, and one local brand,
lected brands with various countries of origin. The US is the
Converse or Prospecs, for Americans or Koreans, respec-
country of origin for LA Gear, Nike, Reebok, and Kodak;
tively. We prepared a different version of the questionnaire
Germany for Adidas, Puma, and Agfa; Japan for Asics, Fuji,
for each brand surveyed.
Konica, and Sony; and South Korea for Samsung.
3.3. Procedures and measures
3. Item purification: the pilot study
Different versions of the questionnaire were assigned
randomly to participants. The participants were told the
3.1. Subjects
purpose of the study was to provide managers with new
insights about effective ways to manage brands successfully.
The pilot sample was composed of 460 undergraduate
Approximately 60 participants were assigned to each of the
university students, 230 from South Korea and 230 from the
four different versions of the questionnaire. The participants
US. Elimination of incomplete data resulted in 414 useable
were asked to indicate any unclear or uncomfortable word-
surveys, 218 Koreans (117 men, 101 women) and 196
ing in the questions. The questionnaire consisted of items to
Americans (111 men, 85 women).
measure the dimensions of brand equity, as well as demo-
Researchers have suggested that the use of student
graphic questions. Brand equity items were evaluated with
subjects in measurement development research threatens
five-point Likert scales anchored at 1 = ``strongly disagree''
the external validity and generalizability of findings due to
and 5 = ``strongly agree.''
the non-representativeness and unique characteristics of the
population (e.g., Burnett and Dunne, 1986; Wells, 1993).
3.4. Analysis and results
However, in a cross-cultural study, well-matched (i.e.,
maximally homogeneous within and between cultures)
To select the items that would enter the main study, we
samples are more useful than representative samples be-
computed the reliability of the items of each construct. We
cause they allow more exact theoretical predictions and
dropped any items of a construct with low correlation to the
reduce the confounding effects of other factors (Hofstede,
composite variable, retaining only those that had a 0.70 or
1991). More important, students are accepted for theory
higher Cronbach's a coefficient for both samples (see
testing research in which the multivariate relationships
Nunnally and Bernstein, 1994). Next, to obtain reasonable
among constructs, not the univariate differences (i.e., mean
parsimony, when two items contributed similar Cronbach's
score comparisons) between samples, are being investi-
a coefficients, we dropped the weaker item. Also, we
gated (Calder et al., 1981). Students have been effective
reworded several items to enhance clarity, based on the
surrogates for non-students or adults in various empirical
participants' comments. With this process, we selected six
studies that have examined, for example, warranty and
items for perceived quality, three for brand loyalty, three for
consumers' risk perceptions (Shimp and Bearden, 1982);
brand awareness, and five for brand associations. We
country of origin, attitude toward advertising, and attitude±
reevaluated these items in the main study.
preference relationship (Yavas, 1994); product warnings
and safe behavior (Cox et al., 1997); price±quality percep-
tions (Lichtenstein and Burton, 1989); dissonance reduc-
4. Data collection and measures: the main study
tion behavior (Sheth, 1970); self-identity acculturation
(Owenbey and Horridge, 1998); and hiring recommenda-
In the main study, we administered the questionnaire to
tions (Olian et al., 1988).
Korean, Korean American, and American samples. Follow-
ing a process recommended by Anderson and Gerbing
3.2. Instrumentation
(1988), Nunnally and Bernstein (1994), and others, we
conducted a series of exploratory and confirmatory factor
We developed the questionnaire in English. Three other
analyses on the data. Our goal was to identify a final set of
marketing researchers and an English-language professor
items with acceptable discriminant and convergent validity,
examined its wording and the face validity of the questions.
internal consistency reliability, parsimony, and cross-cultur-
Next, two bilingual experts fluent in both English and
al metric equivalence. We analyzed data at an individual
Korean translated the questionnaire into Korean. The verbal
level, a multigroup level, and a pooled level.
equivalence between the Korean and English versions was
checked through back-translation with the help of two other
4.1. Subjects
bilinguists in Korea (Douglas and Craig, 1983). This process
was repeated until we agreed that the Korean questionnaire
Undergraduate students at major universities in South
was adequately translated with compatible meaning.
Korea and the US participated. A total of 650 Korean
In the pilot study, we used only four existing brands of
participants in South Korea and 350 Korean American and
athletic shoes as brand stimuli: Adidas, Nike, and Reebok
650 American participants in the US answered the question-

---

B. Yoo, N. Donthu / Journal of Business Research 52 (2001) 1±14
5
naire. After invalid responses, such as non-American citizens
items were measured as in the pilot study. Next, brand
in the American sample, were excluded, the self-adminis-
purchase intention, brand attitude, and product category
tered questionnaires yielded 1530 eligible responses: 633
involvement were measured as follows: Purchase intention
Koreans (435 men and 198 women), 320 Korean Americans
was measured as ` I would like to buy X'' and ` I intend to
(173 men and 147 women), and 577 Americans (272 men and
purchase X' (five-point scale). Attitude toward brand was
305 women).
measured with five-item scales of ``very bad/very good,''
We included Korean American participants to examine
` very nice/very awful,'' ` very attractive/very unattrac-
the validity of the sample selection. In the process of their
tive,'' ``very desirable/very undesirable,'' and ``extremely
acculturalization into the host (i.e., American) culture,
likable/extremely unlikable.'' Product category involve-
Korean Americans are theorized to show some transition
ment was measured with four five-point items: ` I am very
from Korean to American cultural and demographic
involved with product category X,'' ``I use (wear) product
characteristics (Wallendorf and Reilly, 1983; Deshpande
category X very often,'' ``I am a product category X
et al., 1986). In that sense, the culture of Korean Amer-
expert,'' and ` I am not interested in [the product cate-
icans is a hybrid of Korean and American culture. The
gory].'' The brand equity items were asked before the
analysis showed that Korean Americans (54.1%) had
items regarding brand attitude and purchase intention to
more male respondents than Americans did (47.1%) but
reduce the halo effect common to multiattribute attitude
fewer male respondents than Koreans did (68.7%). Korean
models, in which subjects distort their perceptions when
Americans (48.1%) had part-time or full-time jobs more
expressing their overall attitudes before they evaluate de-
often than Koreans did (19.3%) but less frequently than
tails that contribute to the attitudes (Beckwith and Leh-
Americans did (47.1%). In addition, household size
mann, 1975; Cooper, 1981).
among Korean Americans (4.0 members) was greater than
among Americans (3.2 members) but smaller than among
4.3.1. Product experiences
Koreans (4.2 members).
Data show that a significant number of the participants
had experienced the product categories. Those who had
4.2. Instrumentation and procedures
personally purchased athletic shoes, film, and color televi-
sion sets were 95%, 91%, and 83%, respectively, of
In addition to the brand equity items, the question-
Americans; 92%, 90%, and 64% of Korean Americans;
naire included items of other constructs: attitude toward
and 93%, 96%, and 16% of Koreans. Active users of
brand and purchase intention, product category involve-
athletic shoes, film, and color television sets were 93%,
ment and experience, and brand purchase experience. We
72%, and 92%, respectively, of Americans; 88%, 78%, and
developed 12 versions of the questionnaire for the 12
79% of Korean Americans; and 90%, 86%, and 95% of
brands surveyed in the three product categories. Across
Koreans. Participants who had purchased the athletic shoe
versions, we maintained the same format and order of
brands ranged from 36% (Asics) to 94% (Nike) of Amer-
question items.
icans, from 24% (LA Gear) to 88% (Nike) of Korean
American and Korean American participants answered
Americans, and from 26% (Puma) to 65% (Nike) among
the English questionnaire, whereas Korean participants
Koreans. For film, it ranged from 7% (Agfa and Konica) to
answered the Korean questionnaire for their language
96% (Kodak) of Americans, 9% (Agfa) to 95% (Kodak) of
convenience. Participants viewed one randomly assigned
Korean Americans, and 76% (Konica) to 94% (Kodak) of
version of the questionnaire and then rated their brand
Koreans. For television sets, it was 10% (Samsung) and
equity evaluations for the particular brand in the question-
36% (Sony) of Americans, 36% (Samsung) and 75%
naire. Almost the same number of responses for each
(Sony) of Korean Americans, and 38% (Samsung) and
version was obtained, and there was no significant differ-
2% (Sony) of Koreans. This rich product experience
ence in the number of responses among the 12 versions.
indicates that the participants might have developed strong
The number of responses per version ranged from 47 to 57
attitudes toward the brands (Fazio and Zanna, 1981; Smith
for Koreans, from 18 to 35 for Korean Americans and from
and Swinyard, 1983).
40 to 52 for Americans.
The four-item measure of product category involvement
showed acceptable reliability, 0.82, 0.85, and 0.64 for
4.3. Measures
Americans, Korean Americans, and Koreans, respectively.
Product category involvement was significantly different
Brand and product category experiences were measured
among the selected products: 2.41, 3.08, and 2.66 (F =
with yes or no items. The item measuring product category
32.78, p < 0.0001) for film, athletic shoes, and television
purchase experience was ` Have you ever bought any
sets, respectively, for Americans; 2.35, 2.94, and 2.69 (F =
brand of product category X?''; for brand purchase ex-
10.80, p < 0.0001) for Korean Americans; and 2.21, 2.71,
perience, ``Have you ever bought brand X?''; and for
and 2.76 (F = 46.33, p < 0.0001) for Koreans. This wide
usage and ownership, ``Do you currently use/own any
variety of involvement strengthens the generalizability of
brand of product category X?'' Subsequently, brand equity
the scale.

---

6
B. Yoo, N. Donthu / Journal of Business Research 52 (2001) 1±14
5. Analysis and results
squared correlation between awareness and associations is
greater than the average variance extracted for awareness
We conducted three levels of analyses to develop a brand
and associations (Fornell and Larker, 1981).
equity measure (see Durvasula et al., 1993). First, we
As we report in Part A of Table 1, the c2 difference test
performed an individual analysis to determine whether
between Models 1 and 2 was significant in each sample
common items and dimensions were found in each sample.
(c2degrees of freedom [d.f.] = 1 = 115.72 for Americans, 92.12 for
Second, we conducted a multigroup analysis to examine
Korean Americans, and 38.22 for Koreans). As shown in
factorial invariance of the items selected in the individual
Table 2, however, the correlation between awareness and
analysis (JoÈreskog and SoÈrbom, 1993; Bollen, 1989). The
associations was very high in Model 1 (0.88 for Americans,
factorial invariance becomes the basis of cross-cultural
0.80 for Korean Americans, and 0.90 for Koreans). Accord-
comparisons of the constructs (Steenkamp and Baumgart-
ingly, the squared correlation (0.77, 0.64, and 0.81 for
ner, 1998). Third, we conducted a pooled analysis to
Americans, Korean Americans, and Koreans, respectively)
identify culture-free universal dimensions of brand equity
was greater than the average variance extracted for aware-
in the pooled sample (Leung and Bond, 1989). Discovering
ness (0.82, 0.77, and 0.64) and associations (0.64, 0.58, and
the same dimensions in the individual, multigroup, and
0.54), as reported in Part B of Table 1. Therefore, despite
pooled analyses supports the universality of the dimensions
passing the c2 test, by failing the variance comparison test,
across samples. We conducted an O-factor analysis of the 17
discriminant validity between brand awareness and associa-
by 17 correlation matrix for the 17 brand equity items
tions was not proven in each sample.
correlated across the 12 brands (Rummel, 1970).
By combining these two dimensions, we created a new
measurement model, Model 3, composed of the three
5.1. Individual analysis
dimensions of brand loyalty, perceived quality, and brand
awareness/associations. To examine further whether com-
We explored whether a similar pattern of internal con-
bining awareness and associations is better than combin-
sistency and dimensionality among brand equity items could
ing any possible pair of the dimensions, we compared the
be found in each of the three samples. We selected items for
fit of Model 3 with the fit of all remaining three-dimen-
each construct until no higher reliability could be achieved
sional models (Models 4 through 8). Table 1 (Part A)
and obtained the same items across samples, five for
shows that the c2 fit index of the other models was worse
perceived quality and three each for brand loyalty, brand
than that of Model 3. To investigate whether three-
associations, and brand awareness. These fourteen selected
dimensionality is desirable, we also compared Model 3
items showed excellent reliability: perceived quality with
with the one-dimensional model, Model 9, in which all
0.92, 0.90, and 0.84 for Americans, Korean Americans, and
four dimensions are combined into one dimension. But the
Koreans, respectively; brand loyalty with 0.88, 0.86, and
c2 fit index of Model 9 was significantly poorer than any
0.87; brand associations with 0.83, 0.79, and 0.78; and
other previous model. Therefore, we used Model 3 in
brand awareness with 0.93, 0.91, and 0.84. However,
subsequent analyses.
exploratory factor analysis did not produce four distinct
Model 3 was also supported by other values of fit (see
factors among the selected brand equity items, mainly
Table 1). Its goodness-of-fit index (GFI) was 0.87, 0.87, and
because of the inseparability of brand awareness and brand
0.94 for Americans, Korean Americans, and Koreans,
associations. Only three factors (i.e., perceived quality,
respectively, and the adjusted goodness-of-fit index (AGFI)
brand loyalty, and brand awareness/associations) were con-
was 0.82, 0.82, and 0.92. Comparative goodness of fit
sistently found in each sample.
indexes in both comparative fit index (CFI) and incremental
To examine the dimensionality of brand equity, we
fit index (IFI) were 0.93, 0.91, and 0.95. The standardized
established three sets of measurement models, one-, three-,
root mean square residual (SRMR) was 0.061, 0.060, and
and four-dimensional, where the four-dimensional model
0.054. These fit indexes indicate an excellent level of fit of
comprises brand loyalty, perceived quality, brand aware-
the model (see Hu and Bentler, 1999). In addition, the
ness, and brand associations. We used confirmatory factor
loadings of the items to their corresponding dimensions
analysis and estimated the models through the LISREL 8
ranged from 0.58 to 0.87 for Americans, 0.50 to 0.93 for
maximum likelihood method (JoÈreskog and SoÈrbom, 1993).
Korean Americans, and 0.63 to 0.94 for Koreans. The
We then compared 2 four-dimensional models: Model 1, in
smallest t-value of the loadings was 9.31 across samples,
which the correlation between brand awareness and brand
which indicates high significance of the loadings. The
associations is specified to be free, and Model 2, in which
composite reliability estimates, which are internal consis-
the correlation is constrained to be equal with unity. Ac-
tency reliability measures as evidence of convergent validity
cording to the results of these methods, brand awareness and
(Fornell and Larker, 1981) were acceptable, ranging from
associations should be combined due to a lack of discrimi-
0.88 to 0.92 for Americans, 0.86 to 0.90 for Korean
nant validity when either (1) there is no significant c2
Americans, and 0.84 to 0.89 for Koreans. The average
difference between the two models (Bagozzi, 1980; Burnkr-
variance extracted for each dimension was greater than the
ant and Page, 1982; Anderson and Gerbing, 1988) or (2) the
squared correlation between the dimension and any other

---

B. Yoo, N. Donthu / Journal of Business Research 52 (2001) 1±14
7
Table 1
Dimensionality and internal consistency of the brand equity scale
d.f. = degrees of freedom, SRMR = standardized root mean square residual, GFI = goodness-of-fit index, AGFI = adjusted goodness-of-fit index, CFI =
comparative fit index, IFI = incremental fit index, and VE = variance explained for.
Part A: Dimensionality
Americans
Korean Americans
Koreans
c2 of the four-dimensional model
Model 1: fawareness:associations = free (d.f. = 71)
391.58
212.84
223.24
Model 2: fawareness:associations = 1 (d.f. = 72)
507.30
304.96
261.46
c2 of the three-dimensional model (d.f. = 74)
Model 3: Awareness and associations are combined
533.09
317.81
273.59
Model 4: Brand loyalty and perceived quality are combined
1143.35
549.37
1028.03
Model 5: Brand loyalty and associations are combined
1279.53
694.33
1089.00
Model 6: Perceived quality and associations are combined
1302.37
679.41
1153.37
Model 7: Brand loyalty and awareness are combined
1357.37
733.28
1129.49
Model 8: Perceived quality and awareness are combined
1999.40
946.51
1426.86
c2 of the one-dimensional model (d.f. = 77)
Model 9
2906.09
1533.59
2269.89
Fit statistics
Model 3: Three-dimensional model
Model 1: Four-dimensional model
Americans
Korean Americans
Koreans
Americans
Korean Americans
Koreans
c2
533.09
317.81
273.59
391.58
212.84
223.24
d.f.
74
74
74
71
71
71
SRMR
0.061
0.060
0.054
0.052
0.049
0.051
GFI
0.87
0.87
0.94
0.90
0.91
0.95
AGFI
0.82
0.82
0.92
0.85
0.87
0.93
CFI
0.93
0.91
0.95
0.95
0.95
0.96
IFI
0.93
0.92
0.95
0.95
0.95
0.96
Part B: Internal consistency
Americans
Korean Americans
Koreans
Composite a
Coefficient a
VE
Composite a
Coefficient a
VE
Composite a
Coefficient a
VE
Model 3: Three-dimensional model
Brand loyalty
0.88
0.88
0.71
0.86
0.86
0.68
0.87
0.87
0.69
Perceived quality
0.92
0.92
0.69
0.90
0.90
0.68
0.84
0.84
0.53
Awareness/associations
0.92
0.92
0.67
0.89
0.89
0.59
0.89
0.88
0.58
Model 1: Four-dimensional model
Brand loyalty
0.88
0.88
0.71
0.86
0.86
0.68
0.87
0.87
0.69
Perceived quality
0.92
0.92
0.69
0.90
0.90
0.65
0.84
0.84
0.53
Awareness
0.93
0.93
0.82
0.91
0.91
0.77
0.84
0.84
0.64
Associations
0.84
0.83
0.64
0.79
0.79
0.58
0.77
0.78
0.54
dimension and 0.50, which indicates the independence of
and SoÈrbom, 1993). In particular, we tested whether the
the dimensions (Fornell and Larker, 1981). In summary,
factor structure (i.e., item loadings to factors) was statisti-
brand loyalty, perceived quality, and brand awareness/asso-
cally invariant among the three samples by comparing an
ciations make three reliable and valid dimensions of brand
unconstrained and a constrained model. In the unconstrained
equity in each sample.
model, the factor structure is specified to vary across
cultures, whereas the factor structure is constrained to be
5.2. Multigroup analysis
the same across cultures in the constrained model. When the
c2 fit difference between these models is insignificant, the
We subsequently performed an invariance test for Model
factor structure is invariant across samples. As Table 3
3, which is useful in examining the equivalence of a
shows, the LISREL likelihood maximum method of estima-
factorial measurement or structure model across multiple
tion produced a c2 fit index of 1124.49, with 222 d.f. for the
samples (see Bollen, 1989; Durvasula et al., 1993; JoÈreskog
unconstrained model.

---

8
B. Yoo, N. Donthu / Journal of Business Research 52 (2001) 1±14
Table 2
Intercorrelations among dimensions
A = Americans, KA = Korean Americans, and K = Koreans.
Model 1: Four-dimensional model
Brand loyalty
Perceived quality
Awareness
Associations
A
KA
K
A
KA
K
A
KA
K
A
KA
K
Brand loyalty
1
1
1
Perceived quality
0.50
0.52
0.38
1
1
1
Awareness
0.30
0.15
0.28
0.45
0.38
0.18
1
1
1
Associations
0.38
0.29
0.30
0.52
0.37
0.29
0.88
0.80
0.90
1
1
1
Model 3: three-dimensional model
Brand loyalty
Perceived quality
Awareness/Associations
A
KA
K
A
KA
K
A
KA
K
Brand loyalty
1
1
1
Perceived quality
0.50
0.52
0.38
1
1
1
Awareness/associations
0.33
0.19
0.30
0.48
0.40
0.24
1
1
1
For the constrained model, the fit was c2d.f. = 250 =
strained model of c2d.f. = 222 = 1124.49. The test revealed
1215.63. Thus, the fit difference was not insignificant
that the metric inequivalence occurred because of three
(Dc2d.f. = 28 = 91.24, p < 0.0001). This result suggests that
perceived quality items (i.e., QL1, QL4, and QL5) and
the factor structure is not invariant across cultures; therefore,
one awareness item (i.e., AW1). By excluding these four
brand equity evaluations cannot be compared meaningfully
items and combining the remaining 10 invariant items, we
across samples due to lack of cross-cultural metric equiva-
generated an etic measure of brand equity. The 10 items
lence (Steenkamp and Baumgartner, 1998).
comprise three brand loyalty items, two perceived quality
To locate the source of inequality within the specified
items, and five brand awareness/associations items (see
matrix and discover an invariant measurement model across
Appendix A). The unconstrained model (c2d.f. = 96 =
cultures, we conducted a partial measurement invariance
424.65), in which the 10 factor loadings were specified to
test, as suggested by Byrne et al. (1989). We tested the
varyacross cultures, was compared with the constrained
invariance of each factor loading by declaring the loading
model (c2d.f. = 116 = 455.66), in which the 10 factor loadings
alone to be invariant across samples. To detect the invar-
were constrained to be invariant across cultures. The c2
iance, we conducted the c2 difference test with the uncon-
difference was not significant (Dc2d.f. = 20 = 31.01, p >
Table 3
Simultaneous tests of invariance
LO1, LO2, and LO3 = brand loyalty items; QL1, QL2, QL3, QL4, and QL5 = perceived quality items; AW1, AW2 and AW3 = brand awareness items; and
AS1, AS2, and AS3 = brand associations items.
Competing models
c2
d.f.
Dc2
Dd.f.
The unconstrained model: Model 3 with factor loadings variant
1124.49
222
±
±
The constrained model: Model 3 with factor loadings invariant
1215.63
250
91.24**
28
Model 3 with LO1 on brand loyalty factor invariant
1127.98
224
3.49
2
Model 3 with LO2 on brand loyalty factor invariant
1129.17
224
4.68
2
Model 3 with LO3 on brand loyalty factor invariant
1125.00
224
0.51
2
Model 3 with QL1 on perceived quality factor invariant
1130.99
224
6.50*
2
Model 3 with QL2 on perceived quality factor invariant
1127.81
224
3.32
2
Model 3 with QL3 on perceived quality factor invariant
1124.68
224
0.19
2
Model 3 with QL4 on perceived quality factor invariant
1144.63
224
20.14**
2
Model 3 with QL5 on perceived quality factor invariant
1142.24
224
17.75**
2
Model 3 with AW1 on awareness/associations factor invariant
1139.25
224
14.76**
2
Model 3 with AW2 on awareness/associations factor invariant
1128.73
224
4.24
2
Model 3 with AW3 on awareness/associations factor invariant
1127.32
224
2.83
2
Model 3 with AS1 on awareness/associations factor invariant
1125.12
224
0.63
2
Model 3 with AS2 on awareness/associations factor invariant
1127.26
224
2.77
2
Model 3 with AS3 on awareness/associations factor invariant
1128.32
224
3.83
2
* p < 0.05.
** p < 0.001.

---

B. Yoo, N. Donthu / Journal of Business Research 52 (2001) 1±14
9
Table 4
Maximum likelihood estimates and standard errors (SE)
LO = brand loyalty factor: QL = perceived quality factor; and AWAS = brand awareness/association factor. c2d.f. = 116 = 455.66, SRMR = 0.066, GFI = 0.91,
CFI = 0.96, and IFI = 0.96.
Across-culture equivalencies
Americans
Korean Americans
Koreans
Parameter
Estimate
SE
Estimate
SE
Estimate
SE
Estimate
SE
lLO1, LO
0.86
0.02
lLO2, LO
0.88
0.02
lLO3, LO
0.77
0.02
lQL2, QL
0.85
0.03
lQL3, QL
0.80
0.03
lAW2, AWAS
0.82
0.02
lAW3, AWAS
0.79
0.02
lAS1, AWAS
0.79
0.02
lAS2, AWAS
0.81
0.02
lAS3, AWAS
0.61
0.02
qLO1, LO1
0.26
0.02
0.38
0.04
0.26
0.02
qLO2, LO2
0.18
0.02
0.19
0.03
0.30
0.03
qLO3, LO3
0.43
0.03
0.40
0.04
0.39
0.03
qQL2, QL2
0.26
0.04
0.34
0.05
0.24
0.04
qQL3, QL3
0.37
0.04
0.41
0.05
0.32
0.04
qAW2, AW2
0.26
0.02
0.39
0.04
0.37
0.03
qAW3, AW3
0.32
0.02
0.40
0.04
0.42
0.03
qAS1, AS1
0.38
0.03
0.39
0.04
0.37
0.03
qAS2, AS2
0.25
0.02
0.34
0.04
0.44
0.03
qAS3, AS3
0.57
0.04
0.69
0.06
0.65
0.04
fLO, QL
0.49
0.04
0.37
0.04
0.54
0.05
fLO, AWAS
0.34
0.04
0.32
0.04
0.25
0.06
fQL, AWAS
0.49
0.04
0.25
0.04
0.41
0.06
0.05). Therefore, the hypothesis of an invariant pattern of
ables' mean scores is used as a data point per culture, the
factor loadings was tenable. Table 4 reports the maximum
relationships between the variables may appear. Second, a
likelihood estimates for the 10 items. All the factor loading,
within-culture standardization, in which any variable has a
error, and factor intercorrelation estimates were significant
zero mean and unity standard deviation within each cultural
at the 0.0001 level, and the fit indexes were SRMR = 0.066,
group, eliminates the patterning effect of culture, under which
GFI = 0.91, CFI = 0.96, and IFI = 0.96. The reliability was
a different culture shows a different relationship between
also satisfactory: 0.88, 0.86, and 0.87 for brand loyalty
variables. After these standardizations, the data are pooled
among Americans, Korean Americans, and Koreans, respec-
across cultures and analyzed, and the dimensions or measures
tively; 0.81, 0.77, and 0.84 for perceived quality; and 90,
obtained are etic measures.
0.86, and 0.86 for brand awareness/associations.
After we pooled data from the three samples using Leung
and Bond's procedure, we examined three kinds of mea-
5.3. Pooled analysis
surement models for the 10 items (i.e., four-, three-, and
one-dimensional), as we did in the individual analysis. First,
To check the cross-cultural validity of the 10 items further,
the four-dimensional model (c2d.f. = 29 = 219.82), in which
we conducted an individual-level multicultural factor analy-
the correlation between brand awareness and associations
sis, as suggested by Leung and Bond (1989). This technique
was unconstrained, was compared with the other four-
provides a meaningful way to pool cross-cultural data and
dimensional model (c2d.f. = 30 = 317.46), in which the
discover etic dimensions at the individual level while elim-
correlation between the two factors was constrained to be
inating response sets often found in cross-cultural data. Leung
unity. The c2 difference (Dc2d.f. = 1 = 97.64, p < 0.0001)
and Bond's procedure begins with a double-standardization
was significant. But high correlation (0.89) between brand
procedure. First, a within-subject procedure standardizes
awareness and associations suggested the inseparability of
scores across the variables for each subject. The mean of the
those two constructs. The squared correlation (0.79) be-
variables becomes zero, and the standard deviation becomes
tween the two factors was larger than the average variance
one for each individual in any culture. This removes the
extracted for either awareness (0.70) or associations (0.58).
cultural positioning effect, ``the relative location of the
Therefore, the four-dimensional model was not supported
responses made by the average individual from a particular
due to lack of discriminant validity between brand aware-
culture'' (Leung and Bond, 1989, p. 141). Under the cultural
ness and associations. Second, the three-dimensional model
positioning effect, even though certain variables show no
of brand loyalty, perceived quality, and brand awareness/
relationships in individual cultures, when a set of the vari-
associations showed a better fit (c2d.f. = 32 = 326.19) than

---

10
B. Yoo, N. Donthu / Journal of Business Research 52 (2001) 1±14
any other three-dimensional model. It was also better than
cient of brand equity for brand loyalty was 0.60 for Amer-
the one-dimensional model. Third, the model had excellent
icans, 0.56 for Korean Americans, and 0.69 for Koreans;
fit indexes in the pooled sample (SRMR = 0.042, GFI =
0.96, AGFI = 0.93, CFI = 0.96, and IFI = 0.96). The factor
correlations ranged from 0.35 to 0.48, and the composite
Table 5
reliability ranged from 0.82 to 0.88. In summary, the pooled
The MBE index (standard deviation) by brand
sample confirmed the identical factor structure for the model
Brand
as was found in the individual and multigroup analysis,
MBE
Brand
Perceived
awareness/
which is strong evidence of the universality of the items.
n indexa
loyalty
quality
associations
Americans
5.4. Measure of multidimensional brand equity (MBE) and
Camera films
92.36****,b 28.64**** 35.76**** 95.01****
Agfa
41 2.37 (0.44) 1.96 (0.73) 3.09 (0.25) 1.79 (0.83)
its index
Fuji
40 3.14 (0.37) 2.11 (0.77) 3.40 (0.57) 3.80 (0.67)
Kodak
52 3.79 (0.53) 3.27 (0.97) 3.97 (0.56) 4.06 (0.57)
We suggest this 10-item measure of MBE as a scale of
Konica
44 2.64 (0.40) 2.01 (0.72) 3.17 (0.37) 2.51 (0.87)
consumer-based brand equity. Because of the cross-cultural
invariance, the scores of the MBE and its dimensions can
Athletic shoes
27.42**** 15.28**** 16.63****
8.63****
Adidas
51 3.08 (0.49) 2.01 (0.79) 3.44 (0.53) 3.65 (0.83)
be compared cross-culturally, and the different scores may
Asics
46 2.97 (0.46) 2.01 (0.67) 3.34 (0.46) 3.43 (0.98)
be considered indicative of true cross-cultural differences
LA Gear
44 2.87 (0.40) 1.84 (0.89) 3.18 (0.58) 3.49 (0.75)
in the constructs.
Nike
49 3.82 (0.51) 3.01 (0.80) 4.09 (0.54) 4.26 (0.56)
Adding up the raw scores of the 10 items of the MBE
Puma
50 2.85 (0.55) 1.88 (0.79) 3.14 (0.71) 3.43 (0.88)
may not be an appropriate way to develop a MBE index,
Reebok
52 3.34 (0.52) 2.42 (0.85) 3.58 (0.70) 3.95 (0.71)
because they are not evenly distributed among the three
Television sets
62.13**** 24.19**** 25.97**** 37.67****
dimensions. More important, the three dimensions may
Samsung
48 2.67 (0.53) 1.94 (0.79) 3.21 (0.61) 2.65 (1.03)
contribute differently to brand equity. To develop the for-
Sony
52 3.46 (0.48) 2.76 (0.88) 3.82 (0.59) 3.67 (0.59)
mula for a single MBE index, the relationships between the
dimensions and brand equity should be considered. There-
Korean Americans
Camera films
23.26**** 16.23****
7.61***
19.82****
fore, we generated the higher-order three-dimensional mod-
Agfa
22 2.59 (0.52) 1.98 (0.85) 3.05 (0.62) 2.35 (1.12)
el that comprises the same dimensions and loading
Fuji
23 3.08 (0.33) 2.00 (0.75) 3.48 (0.57) 3.60 (0.62)
specifications as the MBE three-dimensional measurement
Kodak
21 3.65 (0.59) 3.37 (0.71) 3.69 (0.70) 3.93 (0.60)
model. The higher-order model is equivalent to the MBE
Konica
18 2.68 (0.31) 2.30 (0.66) 2.92 (0.43) 2.68 (0.58)
model because the intercorrelational paths of the MBE
Athletic shoes
19.03**** 12.82**** 11.78****
4.18**
model can be converted into the causal paths of the high-
Adidas
35 3.18 (0.57) 2.17 (0.86) 3.49 (0.65) 3.81 (0.82)
er-order model without adding any new path or deleting any
Asics
20 2.95 (0.41) 2.20 (0.71) 3.25 (0.38) 3.26 (1.19)
existing path (Stelzl, 1986). In the higher-order model, the
LA Gear
31 2.79 (0.38) 1.90 (0.73) 3.10 (0.51) 3.26 (0.71)
three dimensions are related to a higher-order factor, which
Nike
24 3.66 (0.46) 3.08 (0.70) 3.77 (0.44) 4.18 (0.64)
can be named ``higher-order brand equity.'' The fit indexes
Puma
32 2.49 (0.52) 1.52 (0.60) 2.66 (0.80) 3.40 (1.15)
Reebok
25 3.07 (0.44) 2.09 (0.81) 3.46 (0.64) 3.50 (0.81)
remain the same between these two different models be-
cause they are statistically equivalent.
Televisions sets
16.90***
9.80**
16.62***
3.56
All the causal paths of higher-order brand equity to the
Samsung
24 2.80 (0.34) 2.13 (0.67) 3.04 (0.41) 3.17 (0.56)
dimensions were significant at the 0.0001 level. The smal-
Sony
26 3.44 (0.68) 2.83 (0.90) 3.75 (0.75) 3.55 (0.83)
lest t-value of the paths was 10.88 for Americans, 5.44 for
Koreans
Korean Americans, and 7.41 for Koreans. The path coeffi-
Camera films
8.20**** 10.16****
1.79
2.71*
Agfa
55 2.52 (0.44) 2.04 (0.63) 2.95 (0.47) 2.75 (0.72)
Fuji
53 2.61 (0.48) 2.07 (0.70) 3.03 (0.56) 2.93 (0.75)
Kodak
53 2.87 (0.52) 2.63 (0.77) 3.11 (0.47) 2.96 (0.64)
Notes to Table 5:
a
Konica
49 2.43 (0.47) 1.98 (0.64) 2.90 (0.52) 2.58 (0.97)
The American MBE index = 0.296 (the mean of brand loyalty) +
0.414 (the mean of perceived quality) + 0.290 (the mean of brand
Athletic shoes
7.43****
2.69*
3.39**
9.52****
awareness/associations); the Korean American MBE index = 0.289 (the
Adidas
47 2.70 (0.48) 1.87 (0.64) 3.04 (0.56) 3.55 (0.92)
mean of brand loyalty) + 0.491 (the mean of perceived quality) + 0.220 (the
Asics
53 2.49 (0.52) 1.69 (0.65) 2.91 (0.71) 3.22 (0.73)
mean of brand awareness/associations); and the Korean MBE index = 0.411
LA Gear
50 2.32 (0.44) 1.67 (0.61) 2.94 (0.42) 2.58 (0.93)
(the mean of brand loyalty) + 0.315 (the mean of perceived quality) + 0.274
Nike
54 2.78 (0.54) 2.05 (0.81) 3.20 (0.64) 3.38 (0.73)
(the mean of brand awareness/associations).
b
Puma
53 2.35 (0.47) 1.69 (0.60) 2.82 (0.52) 2.81 (0.92)
ANOVA results for mean differences among brands by pro-
Reebok
57 2.64 (0.54) 1.86 (0.74) 3.15 (0.64) 3.22 (0.79)
duct category.
* p < 0.05.
Television sets
4.90*
7.50**
1.83
6.96**
** p < 0.01.
Samsung
55 2.78 (0.55) 2.41 (0.76) 3.21 (0.59) 2.83 (0.75)
*** p < 0.001.
Sony
52 2.54 (0.54) 2.00 (0.78) 3.37 (0.60) 2.41 (0.90)
**** p < 0.0001.

---