Professional Documents
Culture Documents
Measuring Brand Power Validating A Model For Optim
Measuring Brand Power Validating A Model For Optim
Assembling and validating There is a great deal of published research about brand equity; most of it
the macro-model deals either with attempting to value brand equity or with trying to
understand more about the structure and composition of the construct for
marketing purposes. In this paper a macro-model is first assembled by
combining and adapting several existing partial models, then the model is
validated by collecting and analyzing appropriate market data. The objective
is to reach an understanding of how best to optimize brand equity through the
most parsimonious use of marketing tools. Thus several suggestions made in
the literature are met: to improve our conceptualization of the brand equity
construct (Monroe and Krishnan, 1985; Shocker et al., 1994; Zeithaml,
1988); to establish the operationalisable indicators of brand equity from the
consumer perspective (Biel, 1993; Keller, 1993); to conduct empirical
research in order to develop valid benchmarks for the direct approach to
measuring customer-based equity (Keller, 1993; Park and Srinivasan, 1994);
and to conduct validation work to justify brand equity measurements and
methodology (Urban and Hauser, 1980).
In the work that follows, the ``brand power'' model is first briefly illustrated
and justified, then the empirical research undertaken is described in detail.
170 JOURNAL OF PRODUCT & BRAND MANAGEMENT, VOL. 8 NO. 3, 1999, pp. 170-184 # MCB UNIVERSITY PRESS, 1061-0421
has become clear in recent work on the topic, however, is that image cannot
be assessed by attribute measurement alone, but must include measurement
of consumers' perceptions of the value and benefits attainable by using the
brand ± an altogether more abstract, all-encompassing measurement is
required than typical attitudinal measurements (Aaker, 1991; Arnold, 1992;
Keller, 1993; Kirmani and Zeithaml, 1993; Park et al., 1986). So, the
antecedent factors of brand image should range in levels of abstraction from
the very concrete attribute data to the more ephemeral perceptions of benefits
and values. Zeithaml (1993) further suggests that these three levels of factors
should interact. These perceptions, that lead to the formulation of a
distinctive image in the consumers' minds, can be gleaned from a variety of
sources, both extrinsic and intrinsic to the specific product or service in
question; this is a typical assumption of the associative network theory from
which the basic idea is drawn (Arnold, 1992; Biel, 1993; Keller, 1993). The
image-level conceptualization can be combined with Keller's first,
awareness, stage to form an integrated model of brand power (Figure 1).
Benefits
Brand Image
Brand Equity
Weight derivation
Possible methods of There are several possible methods of deriving the weights. Conjoint
deriving the weights analysis was designed for this type of task, but measurement becomes
extremely tedious for designs incorporating large numbers of features.
Similarly, the card-sort task can lead to somewhat unrealistic product
profiles (i.e. given that consumers have prior perceptions of brands on
multiple attributes, orthogonal designs that combine brand names with
attributes are likely to result in unrealistic profiles), hence the technique is
not used here.
PREFMAP (Carroll and Chang, 1967) is also a possible method that also has
drawbacks in the present situation. The main problem with this method is
that although it can perform monotonic or metric regression of attributes
versus preferences for individuals when they rate many products (say, eight
or more), it fails when respondents rate only an evoked set of alternatives.
Urban and Hauser (1980) suggest that, because PREFMAP works on average
ratings and preferences in this circumstance and therefore sacrifices degrees
of freedom, it cannot be as accurate as an equivalent regression across
consumers. LINMAP is yet another alternative but, as this method uses non-
metric procedures, no statistical properties of the estimates or of the
goodness-of-fit statistic for the overall model are available (Shocker and
Srinivasan, 1979).
Preference regression is The most appropriate method appears to be, then, preference regression.
most appropriate method Other researchers have also concluded thus (Pessemier, 1975; Malhotra,
1993; Urban, 1975; Urban and Hauser, 1980). In particular, Urban and
Hauser recommend the use of a simple rank-ordering scale, which
overcomes the problems of respondents' scale allocation and rounding errors
experienced by constant-sum and dollar-metric scales. In addition, a global
assessment of satisfaction will also be gathered here. Using the calculated
elasticities, one can then predict changes due to modifications in the
independent variables and thus allow marketers to optimize their marketing
efforts to enhance their brand equity.
Model calibration
Let Bc be the set of all brands available to consumer c, and BPbc the
perception that consumer c receives from brands b ( Bc. Then the
probability of selecting brand b,Pbc, is:
Pbc = P(BPbc > BPjc) (j b)(Bc (3)
= P(Vbc + ebc > Vjc + ejc)
= P(Vbc ± Vjc > ejc ± ebc)
Advantages of the model The appeal of this model is that it explicitly models stochastic behavior at the
segment level in a manner which requires no metric assumptions about
rankings. A further advantage of a logit model is that elasticities can be
calculated in much the same manner as discussed above. For the model
postulated in Equation (5), the elasticity of consumer c's demand for
alternative s, P'BIPbc, from a 1 percent change in brand entity k, Ak, is:
Ecsk = (1-P'BIPsc)bkAk (6)
where Ecsk = elasticity of probability of purchase for c for alternative s of
brand entity k,
P'BIPsc = probability of BIP of consumer c's demand for alternative s,
bk = estimates of logit analysis, Ak, and
Ak = brand's perceived dimensions factor score.
where aXbc = the brand awareness of consumer `c' in category `b' and BIP =
Brand Image Power, from Equation (3).
Data generation
Generating list of attributes The first step is to find and operationalize the indicators of brand knowledge,
which are hypothesized in the brand power model illustrated above to be a
function of brand awareness, image and satisfaction. Focus groups, using
professional facilities and facilitators, were utilized for this purpose. Each
session was audio-recorded, and one of the authors and the brand manager
for the relevant product group observed from behind one-way glass. Two
groups of eight participants for each product were selected by gender, social
class and age ± to represent actual consumers in the target audience.
Each item from the verbatim comments generated on the brands by the focus
group participants was then ``trimmed'' (sentences synopsized into short
statements representing a single, specific facet of the product) then all these
items were sorted into attributes, benefits or values. This process was
undertaken to ensure that a complete list had been generated, and was
completed by one researcher and the relevant brand manager with no
difficulty or disagreement. Tissue is apparently judged by 11 attribute items
(e.g. ``not easily torn''), three benefit items (e.g. ``feels soft in use'') and six
value items (e.g. ``modern image''). Men's suits generated seven attribute
items (e.g. ``well-known brand''), four benefit items (e.g. ``comfortable'')
and seven value items (e.g. ``can be recommended with confidence''). Thus,
as expected, the structure of brand equity was found to be very dependent on
product category.
Items incorporated into The next stage in data generation was to incorporate the generated items into
questionnaire a questionnaire. Other than screening demographics, a preference ranking of
the brands was sought (after Urban and Hauser, 1980). Additionally, Likert-
type scales were used to give a validating measure of preference and an
overall measure of satisfaction.
Door-to-door interviews were conducted by professionals on a quota sample
of 810 respondents. The quotas were designed to represent the sample, based
No. Item F1 F2 F3
1 Easily available 0.809 0.148 0.029
2 Reliable 0.719 0.220 0.304
3 Traditional 0.714 0.291 0.289
4 Heavily advertised 0.692 0.241 0.010
5 Good brand name 0.684 0.226 0.196
6 Well-known brand 0.645 0.259 0.319
7 Feels soft in use 0.582 0.444 0.130
8 Feels soft to touch 0.494 0.283 0.358
9 Not easily torn 0.122 0.692 0.198
10 Satisfactory 0.164 0.686 0.058
11 Firm color 0.222 0.683 0.075
12 Rolled well 0.198 0.636 0.217
13 Properly cut 0.356 0.630 0.188
14 Highly absorbent 0.352 0.584 0.155
15 No dust produced 0.218 0.540 0.324
16 Natural pulp 0.420 0.514 0.249
17 Gives youthful feeling 0.134 0.187 0.754
18 Good for gift 0.166 0.097 0.728
19 Modern image 0.210 0.276 0.701
20 Neat image 0.395 0.396 0.500
Tissue Suit
Item Agree Disagree Item Agree Disagree
1 * 1 *
2 * 2 *
3 * 3 *
4 * 4 *
5 * 5 *
6 * 6 *
7 * ±
8 * 7 *
± 8 *
9 * 9 *
10 * 10 *
11 * 11 *
12 * 12 *
13 * ±
14 * 13 *
15 * 14 *
16 * 15 *
± 16 *
17 * 17 *
18 * ±
19 * 18 *
20 *
Table III. Inter-judge agreement on factors items
by an amount greater than twice the standard error. Table IV contains the
relevant data; the appropriate computation reveals that discriminant validity
has been achieved in both cases.
Convergent validity, on the other hand, is achieved by having a high level of
within-factor correlation. The simplest test here is to consider the Cronbach's
alphas already described. The average alpha is 0.852; the range of within-
factor alphas is from 0.665 to 0.899, which provides good evidence that
convergence has also been achieved.
Test of face validity A final test of (face) validity at this stage entails a more subjective approach.
Dividing the factor scores for each brand by the mean factor scores across all
brands created indexed factor scores. This way, a relative sense of how each
factor describes a brand is provided. The author and relevant brand manager
then compared the indexed scores to the real market situation to check that
reality was represented. There were no discrepancies noted.
At this stage it seemed appropriate to name the factors. As the items were
collected after categorization into values, benefits and attributes, it could be
assumed that the factors should represent this schema. In fact, this is not
necessarily the case, as there could be underlying reasons why some value,
for instance, may be more related to a benefit than some other value.
Nevertheless, the original classification does show to a point, as can be seen
from Tables I and II. The Tissue factors illustrate the above point well, as
Factor 2 clearly represents Tissue attributes and Factor 3 Tissue values,
whilst Factor 1 is a mixture of both. The authors made a value judgment and
called it Tissue attributes 2. The suit market is relatively straightforward,
with Factor 1 representing Suit benefits, Factor 2 Suit values. Factor 3 is also
clearly Suit attributes and so is the single-item (price) Factor 4.
Quantification
Satisfaction chosen as Two potential dependent variable constructs were captured, satisfaction and
dependent variable preference. As expected, the constructs do correlate (0.691 for Tissue and
construct 0.126 for Suits, p < 0.0001). To establish which construct has most bearing
on the scales, a regression analysis was run for both and the R2 values
compared. Satisfaction is clearly superior (R2 for satisfaction = Tissue,
0.573, Suits 0.548. R2 for preference = Tissue 0.252, Suits 0.201). Previous
studies have linked satisfaction to purchase, so satisfaction is consequently
used here. Normal preference regression can now be used, using the factor
scores as discussed in an earlier section, to analyze the relationship between
the factors and overall satisfaction. There should be no problem with
multicollinearity, as the scores used are derived from factor analysis.
However, adding factor variables one at a time to ensure that both the R2 and
adjusted R2 increase with each factor added provides a further check. The
preference regression models are shown in Table V.
The results of the brand power model suggest that over 50 percent of the
variance is explained; the beta coefficients are all significant and have good
explanatory power. It seems that the factor construct is well balanced and
exhibits criterion-related validity.
Calculating the importance From these regressions, the importance weights can be calculated by adding
weights all b coefficients and proportionalizing each variable (Urban and Hauser,
1980), The resultant values are displayed in Table VI, and provide an
additional opportunity to check validity. If the products are considered as
being functionally or symbolically-oriented, then the factor weights should
reflect this. Men's suits certainly fit into the symbolic category. Although
tissues are functionally weighted relative to suits, it is a surprise that values
should figure at all. Biel (1993) asserts, however, that in undifferentiated
markets values tend to come to the fore; it seems likely that this is what has
happened here.
Results
Measuring brand image power
Testing validity of brand The intermediate stage to calculating brand power is to measure brand image
image power power(BIP), using Equation (2) developed above. After calculating the
brand-specific means of the factor scores and inputting the data, the value of
BIP for each consumer is assessed and compared to their overall satisfaction.
To test the predictive ability of BIP, a parsimonious rule is applied, where, if
there is any difficulty in judging a consumer's preference correctly, it is
treated as a missing case. Thus suppose Consumer 2 had recently purchased
Brand 1 three times, and her calculated BIP was 63 for Brand 1, 45 for
Brand 2 and 44 for Brand 3. Yet as her overall satisfaction measure was the
same for both Brand 1 and Brand 2, the case is regarded as ``missing data''.
In this way the data set is slightly reduced, but all judged data sets are
``clean'' and not biased in support of the model. Urban and Hauser (1980)
suggest that a prediction for preference from 40-80 percent is good; by their
standards the results shown in Table VII are outstanding. Unfortunately the
purchase data for suits are not available, but the prediction of purchase for
tissues is also satisfactory.
Tissue Suit
Factor Weight Factor Weight
Attribute 1 39.2 Attribute 1 19.2
Attribute 2 31.4 Attribute 2 16.9
Value 29.4 Benefit 33.0
Value 30.8
Total 100.0 Total 100.0
Table VI. Weighted preference regression model
Elasticity
The practical value of the model can be further enhanced by considering the
elasticity attached to each factor for each brand, following the suggestion of
Roberts (1983) and Wells (1993). Thus, for example, for Brand 2 in the
Tissue market, the three factors have elasticities as follows: F1 = 0.00997, F2
= 0.0128, F3 = 0.0023. Factor 2 indicates where corporate resources should
be placed to enhance the brand equity and, consequently, sales.
Discussion
Soft attributes become The main point of this paper has been to show that brand equity can be
more important measured and, to a large extent, understood in terms of its antecedent factors.
There are several points about the taxonomy of brand equity discovered en
route that should be mentioned. It has been discussed in the literature, for
instance, that intrinsic and extrinsic cues are interrelated in the consumer's
mind and used during product evaluation. This becomes evident from an
inspection of the factor analyses. Similarly, Biel (1993) claims that as the
functional differences (hard attributes) between brands become more trivial,
then soft, or intangible, attributes will become more important. If the
``value'' dimension in the present context is equated to soft attributes, and
the ``attributes'' to hard, then there is plenty of supporting evidence here.
Again, Keller's idea regarding the three levels of attributes (attributes,
benefits and values) has been shown to be useful in discovering all the
relevant antecedents of brand equity.
Suits
Brand 1 16.91 0.09 1.522 29.5 n/a n/a
Brand 2 16.90 0.03 0.507 9.8 n/a n/a
Brand 3 12.67 0.001 0.013 0.2 n/a n/a
Brand 4 15.29 0.03 0.459 8.9 n/a n/a
Brand 5 15.83 0.001 0.016 0.3 n/a n/a
Brand 6 19.52 0.135 2.635 51.2 n/a n/a
Table VIII. Brand power and market share for each brand
References
Aaker, D.A. (1991), Managing Brand Equity: Capitalizing on the Value of a Brand Name,
The Free Press, New York, NY.
Aaker, D.A. and Biel, A.L. (1993), Brand Equity and Advertising: Advertising's Role in
Building Strong Brands, Lawrence Erlbaum Associates, Hilldale, NJ.
Arnold, D. (1992), The Handbook of Brand Management, The Economist Books, London.
Bagozzi, R.P. and Warshaw, P.R. (1990), ``Trying to consume'', Journal of Consumer
Research, Vol. 17, September, pp. 127-40.
Bearden, W.O., Netemyer, R.G. and Mobley, M.F. (1991), Handbook of Marketing Scales:
Multi-items Measures for Marketing and Consumer Behavior Research, Sage Publications,
Newbury Park, CA.
Beckwith, N.E. and Lehmann, D.R. (1975), ``The importance of halo effects in multi-attribute
attitude models'', Journal of Marketing Research, Vol. 12, August, pp. 265-75.
Biel, A.L. (1993), ``Converting image into equity'', in Aaker, D.A. and Biel, A.L. (Eds), Brand
Equity & Advertising's Role in Building Strong Brands, Lawrence Erlbaum Associates,
Hilldale, NJ, pp. 67-82.
Braun, M.A. and Srinivasan, V. (1975), ``Amount of information as a determinant of consumer
behavior toward new products'', Combined Proceedings, American Marketing Associa-
tion, pp. 373-8.
Carroll, J.D. and Chang, J.J. (1967), Relating Preference Data to Multidimensional Scaling
Solutions via a Generation of Coombs' Unfolding Model, Bell Telephone Laboratories,
Murray Hill, NJ.
de Chernatony, L. (1991), Strategic Brand Planning: The Way for Brands to Succeed in the
1990s, Working paper, City University Business School.
de Chernatony, L. and McWilliam, G. (1990), ``Appreciating brands as assets through using a
two-dimensional model'', International Journal of Advertising, Vol. 9 No. 2, pp. 111-19.
Farquhar, P.H. (1989), ``Managing brand equity'', Marketing Research, September, pp. 24-34.
Green, P.E. and Wind, Y. (1973), Multi-Attribute Decision in Marketing, The Dryden Press,
Hinsdale, IL.
Hauser, J.R. and Koppelman, F.S. (1979), ``Effective marketing research: an empirical
comparison of techniques to model consumers' perceptions and preferences'', in Shocker,
A.D. (Ed.), Analytic Approaches to Product and Marketing Planning, Marketing Science
Institute, Cambridge, MA.
Hauser, J.R. and Urban, G.I. (1979), ``A normative methodology for modeling consumer
response to innovation'', Operations Research, Vol. 25, July/August, pp. 579-619.
Johnson, E. and Meyer, R. (1984), ``Compensating choice models of non-compensating
processes: the effect of varying context'', Journal of Consumer Research, Vol. 11,
January, pp. 528-41.
Keller, K.L. (1993), ``Conceptualizing, measuring, and managing customer-based brand
equity'', Journal of Marketing, Vol. 57, January, pp. 1-22.
Kirmani, A. and Zeithaml, V.A. (1993), ``Advertising, perceived quality, and brand image'',
in Aaker, D.A. and Biel, A.L. (Eds), Brand Equity & Advertising: Advertising's Role in
Building Strong Brands, Lawrence Erlbaum Associates, Hilldale, NJ, pp. 143-62.