Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

A New Approach to

Market Segmentation
Although market segmentation is still considered a major topic
of interest by the marketing research community, it might be
useful to question some premises of traditional segmentation.

PAUL E. G R E E N

Paul E. Green is S. S. Kresge Professor of Marketing at The the complexity of segmentation methods is
Wharton School, University of Pennsylvania.
more apparent than real. Two basic ap-
proaches to segmentation--a priori and post
Few, if any, topics in marketing have com- h o c - a r e described, and it is shown that all
manded more attention than market segmen- current approaches boil down to one of these
tation. Hardly a journal issue appears that types, or possibly a hybrid of the two. A new
does not include some article on the subject. method is then proposed for examining rela-
Almost all marketing texts devote at least a tionships among alternative bases for segmen-
chapter to the topic, and whole books have 61
tation that may be under consideration b y the
been written on segmentation. researcher. The third part of the article
Techniques for carrying out segmentation describes a new approach to s e g m e n t a t i o n -
have also increased in variety and sophistica- componential segmentation. The characteris-
tion. Multiple regression and discriminant tics of this method are described nontechni-
analysis were the standard approaches a dec- cally, and the results of an empirical applica-
ade ago. N o w they frequently take a back seat tion are reported. 1
to such procedures as Automatic Interaction
Detection, factor analysis, cluster analysis,
perceptual and preference mapping and SEGMENTATION AND MARKETING
conjoint scaling. With the profusion of tech-
niques has come a profusion of segmentation Contemporary segmentation studies often col-
bases--perceptual-preference segments, needs- lect such a welter of different kinds of data
oriented segments, common-problems seg- that it is difficult to k n o w h o w the separate
ments, benefits-seeking segments, psycho- facets of the survey are to be integrated. For
graphic segments, and so on. One could easily
example, in segmenting the market for auto
get the impression that the topic has b e c o m e
too complex for mere mortals to compre-
hend. 1. Componential segmentation has been developed joint-
ly with J. Douglas Carroll, Bell Laboratories, and Frank J.
One of the purposes of this article is to Carmone, Drexel University. More details on the model
argue that such is not the case. I first discuss appear in Paul E. Green, J. Douglas Carroll and Frank J.
Carmone, "Superordinate Factorial Designs in the Analysis of
the role that segmentation can play in the Consumer Judgments." (Working paper, University of Penn-
formulation of marketing strategy for either sylvania, September 1975) and Paul E. Green, J. Douglas
Carroll and Frank J. Carmone, "Design Considerations in
products or services, consumer or industrial. Attitude Measurement." (Paper presented at the Seventh
The second part of the article suggests that Annual Conference on Attitude Research, February 1976).

FEBRUARY 1977
PAUL E. GREEN

insurance policies, data could be collected on over, it is reasonable to assume that demo-
the respondent's graphic and socioeconomic variables moderate
current insurance--amount, type o f coverage, cur- persons' choices among offerings but, in turn,
rent supplier, previous supplier, recent claims are pretty much independent of the firm's
perceptions and evaluations of alternative sup-
pliers of insurance marketing strategies.
utilities for various combinations of insurance Functional characteristics of offerings per-
policy characteristics tain to the uses or purposes to which the
evaluations of new policy concepts
insurance needs, benefits sought, problems en- products or services are put. (Included here
countered with current supplier are the symbolic values that offerings may
life style-insurance related and more general display in a social-psychological context.)
aspects Stretching the analogy a bit, functional char-
communications channel selection
demographic and socioeconomic variables. acteristics of people pertain to the ways in
which demographic and socioeconomic char-
It is banal to say that segmentation studies are acteristics shape one's style of living and one's
but a means to an end. However, one some- perceptions and values in the market place.
times loses sight of the strategic aspects of Two different structural o f f e r i n g s - f o r ex-
segmentation research when the incremental ample, a candle and table l i g h t e r - m a y serve
costs of tacking on more sections to the the single function of lighting one's cigarette.
questionnaire are so low. Conversely, a single o f f e r i n g - a lemon-lime
At any point in time, the firm markets a soft d r i n k - m a y serve two or more functions,
particular set of offerings (products or ser- such as a mid-afternoon refreshment or a
vices) to a particular set of buyers (consumer, mixer with one's favorite b o u r b o n at bedtime.
62 commercial or industrial). This all takes place The functions that an offering purports to fill
within a set of interconnected environments are part of its array of promotional appeals,
made up of competitors, distributors and claims and symbolic characteristics. Develop-
various governmental agencies. Just as obvi- ment of its image and market positioning is
ously, the firm could, in principle, be market- the job of advertising message design and
ing other kinds of offerings, to the same or to communications channel selection.
different kinds of people. By the same token, two different demo-
Figure 1 is an adaptation of the original graphic structures could display (within
Ansoff schema involving four basic competi- limits) a similar life style, benefit-seeking
tive strategies in marketing: market penetra- pattern, or brand preference profile. Con-
tion, market development, product develop- versely, a single type of demographic struc-
ment and diversification. 2 In this adaptation, ture could manifest different goal-seeking and
however, we distinguish b e t w e e n the struc- expressive behaviors.
tural and functional characteristics of offer- Functional-expressive behaviors may, in
ings and markets. turn, be arrayed hierarchically in terms of
Structural characteristics of products in- general life style, product-related needs, prob-
clude physical and chemical attribute levels, lem solving and specific brand-related percep-
packaging, distribution and price. Structural tions and preferences. (In the simplified classi-
characteristics of persons refer to their demo- fication of Figure 1 these are all called
graphic and socioeconomic characteristics. functional-expressive.)
While these latter characteristics are expected The original Ansoff schema was repre-
to change over time, they may be treated as sented by a table which did not distinguish
fixed for relatively short time intervals. More- between structural and functional characteris-
2. H. Igor Ansoff, Corporate Strategy (New York:
tics of offerings and markets. From the
McGraw-Hill Book Co., 1965). standpoint of strategy formulation, a more

BUSINESS HORIZONS
A New Approach to Market Segmentation

FIGURE 1

Alternative Marketing Strategies and Segmentation Research*

Present Segments New Segments

Structural Functional- Structural Functional-


Character- Expressive Character- Expressive
istics Behavior istics Behavior

Structural
Characteristics
Present MARKET MARKET
Offerings PENETRATION DEVELOPMENT
Functional-
Symbolic
Appeals

Structural
Characteristics 63
New PRODUCT
Offerings DEVELOPMENT DIVERSIFICATION
Functional-
Symbolic
Appeals

*Adapted from H. Igor Ansoff, Corporate Strategy (NewYork: McGraw-HillBook Co., 1965).

detailed classification might be preferred. For For example, lowering the insurance policy's
example, in the auto insurance example one premium and offering the policy b y mail may
might want to distinguish between structural markedly change the offering's functional-
product development in which the actual symbolic character. Conversely, including a
policy's characteristics are altered (increased new restriction on claims (as part of the
benefits, extended coverage, increased pre- policy's fine print) may have no effect what-
mium), and functional product development soever on the policy's image. Whatever may
in which different appeals for selling the be the case, a general problem in strategy
original policy are implemented. In turn, formulation is to achieve c o n g r u e n c e between
these appeals could be targeted to specific structure and functional-symbolic appeals. In
demographic groups, different life style seg- particular, one may wish to find a specific
ments, or both. structure that allows for a variety of believ-
Depending u p o n the empirical context, able functional-symbolic appeals--each one
the interconnections between structure and being attractive to a different market seg-
function may be loosely or tightly coupled. ment.

FEBRUARY1977
PAUL E. G R E E N

The value of the schema of Figure 1 is the of their multivariate profiles regarding such
guidance it provides for designing the segmen- characteristics as purchasing behavior or atti-
tation study in the first place. As a further tudes. Following this, the segments may then
example, consider the problem of a telephone be examined for differences in other charac-
company wishing to increase revenues for teristics, not used in the original profile
long distance usage. Clearly, little can be definition. In post hoc segmentation one does
gained by attempting to increase installed not know the number of clusters or their
sets; the market is already saturated. Simi- relative size until the cluster analysis has been
larly, little opportunity (except for the ex- completed.
ploitation ot pmture phones) exists tor drama-
As an example of a priori segmentation,
tically changing the physical structure of
one might classify all respondents according
telephones. What can be done involves such to their stated favorite brand of beer. Having
strategic aspects as done this, some technique such as multiple
increasing the variety of appeals related to why discriminant analysis might be used to deter-
people call long distance
changing the pricing or discount structure for load mine if the groups differ in terms of average
leveling and increased overall demand demographic profiles or life style variables.
appealing to special demographic groups--senior In post hoc segmentation, we prespecify
citizens, college students, military service personnel--
via special rates or gift certificates only the set of variables on which consumers
appealing to special life styles--the highly mobile, are to be clustered-benefits sought, problems
gregarious and involved person--via special appeals or encountered with the product, or whatever.
channels, such as direct mail. One then takes the consumers' response pro-
64 The reader can easily think of m a n y examples files on the whole battery of selected variables
where the permissible courses of action shape and clusters the respondents. Having done
the kind of segmentation study that should be this, some technique such as multiple discrim-
implemented. However, as basic as this point inant analysis (or simple cross-classification,
is, many omnibus-type segmentation studies for that matter) can be employed to see if the
are launched and the data analyzed before the various clusters differ with regard to demo-
central idea of formulating strategy alterna- graphics, product usage, and so on. In the
tives even comes up. preceding beer example, respondents could
first be clustered on the basis of the common-
ality of their benefit-seeking profiles. Having
TRADITIONAL APPROACHES done this, one could then see if the various
clusters differed significantly with regard to
In the past, marketing researchers have em- weekly consumption of beer, brand favorite,
ployed two basic analytical approaches to respondent age, and so on.
segmentation: In some studies a hybrid of the two
A priori segmentation, in which the re- approaches is used. For example, respondents
searcher chooses some cluster-defining de- could first be grouped according to favorite
scriptor in advance, such as respondent's brand and then a clustering procedure could
favorite brand. Respondents are then classi- be employed to see if segments evincing
fied into favorite-brand segments and are c o m m o n benefit-seeking profiles appear with-
further examined regarding their differences in each of the brand-favorite segments that
in other characteristics, such as demographics were found via the a priori approach.
or product benefits being sought. Figure 2 shows, in stylized form, hypothe-
Post hoc segmentation, in which respon- tical examples of each approach. To illustrate,
dents are clustered according to the similarity assume that we have m consumers' consump-

BUSINESS HORIZONS
A New Approach to Market Segmentation

FIGURE 2

Traditional Approaches to Segmentation

A Priori Segmentation

Brand
Brand
C o n s u m p tio n
Segment
Rates
A B C
A B C

1 4 6 8-
2 X
2 2 4 9
3 X --
3 0 15 12
4
Consumers 4 12 1 3

m 14 13 0 m X

Post Hoc Segmentation

R a t i n g s on
Product Needs Need 65
Statements Consumers Segment
1 2 3 ...... n 1 2 3 4 ...... m 1 2 3 4

1 -2 4 7 ...... 1" 1 1
2 1 3 6 . . . . . . 2 2 2 -- X
3 2 4 6 . . . . . . 1 3 Cell entries are 3 X
Consumers 4 2 3 6 ...... 2 ~ 4 dissimilarity .__..._~ 4 X
measures
• . • . • between
consumers
• , . ° .

• , • . •

m 2 2 4 ...... 3 m
m x

Hybrid Segmentation

Needs
Brands Statements Brand-Need Segment
A B C 1 2 3 ...... n A B C
1 2 1 2 3 1 2 3 4
1 4 6 8" 1 -247 ...... f" -- X --

2 2 4 9 2 1 3 6 ...... 2 2 -- -- X
3 0 15 12 3 2 4 6 ...... 1 3 -- -- X
Consumers 4 12 11 3 4 2 3 6 ...... 2 .____1~. 4 - x

m 14 13 0 m 2 2 4 ...... 3 m X --

FEBRUARY 1977
PAUL E. GREEN

tion data (in cases, over some base period) of process of doing this, information about
three brands of beer: A, B and C. Under a detailed consumption of brands or detailed
priori segmentation we may elect to group responses to needs-type statements is (willing-
people on the basis of the brand that enjoys ly) discarded.
the highest consumption rate. The second Two other points are worth making. First,
matrix is a simple transformation of the first in Figure 2 only persons were grouped into
in which an X appears under the brand for segments. This is generally the way market
which consumption is highest for each respon- segmentation proceeds, b u t it is quite possible
dent. (We note, for example, that the first to take the obverse point of view and to
two respondents are assigned to segment C.) cluster offerings. For example, in the case of a
In post hoc segmentation, an extra s t e p - large number of brands, one could just as
computation of a matrix of interperson dis- easily develop a dissimilarity measure for pairs
similarity measures-is involved. Then, the of brands across persons' consumption pro-
actual grouping process is carried out b y some files and cluster those brands that exhibit a
type of clustering algorithm. To illustrate, relatively high commonality of usage. This
assume that each of the m consumers re- may not always be useful, b u t the point is
sponds to a set of needs-type attitude seg- that a kind of symmetry exists b e t w e e n each
ments regarding beer consumption, on a facet of the response m a t r i x - o n e can cluster
seven-point agree-disagree scale. Under post market offerings just as readily as people.
hoc segmentation, the first matrix shows the Second, the choice entities need not be
original response profiles. This matrix is trans- products or services in the more narrow sense.
formed into an m x m symmetric matrix of They could be political candidates, legislative
66 dissimilarity measures in which each cell entry actions, charitable appeals, home site loca-
measures h o w disparate each pair of con- tions, or whatever.
sumers is across the whole set of needs-type
statements. This second matrix is then sub-
mitted to a cluster analysis, yielding, in this CHOOSING A SEGMENTATION BASE
illustrative case, four segments.
Hybrid segmentation starts out with b o t h In most segmentation studies, the researcher
data sets. Respondents are first clustered b y is faced with a surfeit of variables that could
favorite brand. Following this, separate and serve as the basis of segmentation. Returning
independent dissimilarity-measure computa- to the insurance policy example, eight differ-
tions and cluster analyses are carried out ent sets of data were collected, ranging from
within each favorite-brand segment. As shown the characteristics of current policies to the
in Figure 2, different numbers of clusters may respondent's demographic-socioeconomic pro-
emerge and, in general, the needs-type sub- file. Which set is to serve as the "distin-
groups found b y the hybrid approach differ guished" set or segmentation base?
from those found b y the post hoc approach. Unfortunately, there is no simple answer
As can be noted from the preceding to this question. Historically, researchers have
example, segmentation is an aggregative pro- selected some variable or battery of variables
cess insofar as matrices of individual data are and proceeded from there. For example, if a
concerned. Whether a single variable is se- needs-type attitude segmentation is being
lected for respondent assignment or whether a done, then responses to the battery of needs-
clustering is performed over a general measure type statements provide the segmentation
of interperson dissimilarity, the net effect is base. However, suppose one finds that other
to reduce the m original respondents to a variables--demographics, product usage, new
more manageable number of groups. In the concept evaluations-do not show systematic

BUSINESS HORIZONS
A New Approach to Market Segmentation

profile differences among the needs-type seg- While practitioners may continue to
ments? If so, should some other battery of choose some segmentation base b y fiat, the
variables be chosen as a base? new method can still show them h o w strongly
Relating the choice of segmentation base the clusterings obtained from alternative bases
to strategy considerations, as described ear- are related to the distinguished base. More-
lier, appears to be the most appropriate over, if one should select a benefits-seeking
approach. Even so, cases can arise in which base and find that its clustering is indepen-
several candidate bases are tenable "choices. dent of the clusterings of all other candidate
What is needed is a method that will: (1) bases-while the other bases are mutually
show h o w closely related various clusterings associated-then serious questions about the
obtained from alternative bases are; (2) test choice could be raised.
whether some clusterings are independent of In brief, the clustering comparison ap-
the remaining ones; and (3) find the clustering proach can provide new insights into market
segment relationships and statistical support
that exhibits the highest contribution to the
mutual association among the subset of clus- for the choice of some specific segmentation
terings that evince significant associations in base. Moreover, it can do this in terms of the
natural output of a segmentation analysis-
the first place. Furthermore, even if the
groupings of consumers-rather than being
researcher has settled in advance on a distin-
limited b y the linearity assumptions of such
guished set of variables to serve as the
techniques as generalized canonical correla-
segmentation base, he may be interested in
tion. 3
finding out h o w well the clustering obtained
from this distinguished set can be predicted 67
b y some function of the other clusterings.
COMPONENTIAL SEGMENTATION
As it turns out, all of these objectives are
interrelated and can be attained through Up to this point, our discussion of segmenta-
application of a new approach that utilizes tion has proceeded along more or less tradi-
multidimensional contingency table analysis. tional lines. It is now appropriate to describe,
The basic ideas of the method can be briefly briefly and nontechnically, a new .way to
described as follows: segment markets, called componential seg-
E a c h c a n d i d a t e s e g m e n t a t i o n base ( b a t t e r y o f mentation. In one sense, componential seg-
variables) is i n d e p e n d e n t l y u s e d to cluster t h e respon- mentation is disaggregative in nature. Unlike
d e n t s i n t o some specified n u m b e r of groups. T h e
n u m b e r o f clusters so f o r m e d m a y or m a y n o t b e t h e
the a priori and post h o c approaches, in
same across bases, d e p e n d i n g u p o n t h e r e s e a r c h e r ' s componential segmentation persons who
objectives. evince similar response profiles are not
E a c h c l u s t e r i n g serves as a categorical variable grouped together. On the contrary, their
(where t h e clusters r e p r e s e n t t h e categories), a n d a
m u l t i d i m e n s i o n a l c o n t i n g e n c y table is f o r m e d . individual responses are further decomposed
A v a r i e t y of tests c a n t h e n b e carried o u t o n t h e into c o m p o n e n t contributions that are per-
m u l t i w a y table to see w h i c h clusterings are r e l a t e d to son-related (demographics, life styles, and so
w h i c h o t h e r ones.
on) and contributions that are offering-related
One clustering c a n be c h o s e n as t h e d i s t i n g u i s h e d
c l u s t e r i n g a n d a m o d e l can b e d e v e l o p e d f o r p r e d i c t - (such as attribute levels of the product or
ing this c l u s t e r i n g f r o m clusterings o b t a i n e d f r o m service).
o t h e r bases. I n this way o n e can a s c e r t a i n h o w closely
a s s o c i a t e d s o m e s u b s e t o f clusterings m a y b e w i t h t h e 3. For a more technical discussion of this approach, see
d i s t i n g u i s h e d clustering. If desired, clusterings o n Paul E. Green and Frank J. Carmone, "Segment Congruence
n o n d i s t i n g u i s h e d sets c a n b e r e p e a t e d w i t h subsets of Analysis: A Method for Analyzing Association Among
t h e variables t h a t i n d i v i d u a l l y evince h i g h a s s o c i a t i o n Alternative Bases for Market Segmentation," Journal of
w i t h t h e distinguished-set clustering. Consumer Research (forthcoming, March 1977).

FEB RUARY 197 7


PAUL E. GREEN

An Example undesirable to highly desirable. Conjoint scal-


ing could then be used to estimate each
A firm engaged in the marketing of credit respondent's utility for each level of each of
cards was interested in developing a new card the five factors. 4 However, b y the application
that would maintain current patron privileges of certain experimental designs-called ortho-
but would also exhibit attractive benefits to gonal arrays-one can reduce the labor of
potential retailing establishments that might respondent evaluation from thirty-two card
wish to honor the card. Interest centered on descriptions to only sixteen, while still retain-
the establishments' utilities for various cost- ing the ability to estimate all utilities, s
benefit profiles. Five credit card attributes Orthogonal arrays represent a type of frac-
were selected for study: tional factorial design that is a generalization
d i s c o u n t r a t e - 2 V 2 % versus 6% of the more familiar latin square design.
s p e e d o f p a y m e n t - o n e d a y a f t e r r e c e i p t of w e e k ' s ("Orthogonal" means that all effects are
v o u c h e r s versus t e n days a f t e r r e c e i p t o f w e e k ' s
mutually uncorrelated.)
vouchers
c r e d i t a u t h o r i z a t i o n - - c o m p u t e r t e r m i n a l versus
toll-free t e l e p h o n e n u m b e r
m a r k e t i n g s u p p o r t to e s t a b l i s h m e n t - - 0 . 1 % o f an-
The Componential Design
n u a l billings versus 0.75% of a n n u a l billings
r e b a t e o n a n n u a l g r o w t h in b i l l i n g s - - n o n e versus a In the usual approach to conjoint scaling, a
15% r e b a t e o n billings e x c e e d i n g q u o t a . sample of retail establishments would be
chosen and respondents asked, among other
Insofar as the preceding factors are con-
things, to evaluate the hypothetical credit
cerned, the discount factor refers to the
cards. Segments may then be formed on the
68 percentage of the establishment's billings on
basis of commonality of utility functions.
the credit card that is deducted b y the credit
In componential segmentation, however,
card company for its service. Speed of pay-
one applies the same design principles to the
ment refers to h o w quickly the company pays
(less its service discount) the establishment selection of respondents--namely, fractional
factorial designs, such as orthogonal arrays--as
after receiving the establishment's vouchers.
The credit authorization procedure consists of are applied to the construction of the stimu-
two alternatives: a company-supplied com- lus cards. In the present study, four establish-
puter terminal versus a manual system involv- ment-type factors were believed to be impor-
tant background variables for segmentation:
ing credit-check calls on a toll-free telephone
t y p e of e s t a b l i s h m e n t - l o d g i n g ; retailing; restau-
number. Marketing support refers to the
rant
company's payment of local advertising on size of e s t a b l i s h m e n t w i t h i n t y p e - - s m a l l ( b e l o w
behalf of the establishment. The rebate proce- m e d i a n in sales v o l u m e ) versus large
establishment's current favorite card--card type
dure pertains to a plan in which the company
A; c a r d t y p e B; c a r d t y p e C
rebates, to the establishment, 15% of the n u m b e r of credit cards h o n o r e d b y establish-
company's charges on all billings exceeding m e n t - t h r e e or f e w e r versus f o u r or m o r e cards.
the establishment's quota, usually set at 1.25
In this case we would have 3 x 2 x 3 x 2 = 36
of the previous year's billings level. As noted,
combinations if a full factorial were used.
each of the above factors appears at two
levels, so that a full factorial design would 4. As examples of various ways to apply conjoint scaling,
entail 25 -- 32 combinations. see Paul F. Green and Vithala R. Rao, "Conjoint Measure-
ment for Quantifying Judgmental Data," Journal of Market-
Ordinarily, one would make up thirty-two ing Research (August 1971), pp. 355-363 and Richard M.
credit card combinations and have retailers Johnson, "Trade-Off Analysis of Consumer Values,"~ournal
of Marketing Research (May 1974), pp. 121-127.
evaluate them on some type of scale, such as a 5. See references in footnote 1 for technical details
one to eleven rating scale, ranging from highly regarding orthogonal arrays.

BUSINESS HORIZONS
A New Approach to Market Segmentation

However, by employing orthogonal array problem, the effect of the interaction term
principles, a design involving only nine com- was quite small (about 8% of the total),
binations can be prepared. One of the basic suggesting that this market was not highly
features of componential segmentation is that segmentable.
respondents exhibiting specific profiles are The COSEG model develops parameter
sought from some much larger frame of values for both stimulus components (the
potential respondents in order to find the credit cards) and respondent profile charac-
particular set of background profiles dictated teristics. In the former case the parameter
by the design. values represent utilities for the levels of the
If full factorials were employed in design- five components making up the stimulus
ing both stimuli and respondents, the re- cards. In the latter c a s e the parameters are
sponse matrix would consist of thirty-six saliences that indicate how much each profile
respondent types times thirty-two stimulus characteristic contributes to variation in the
types, a total of 1,152 data points. By the use evaluative responses.
of orthogonal array principles, this size is No attempt is made here to describe the
reduced to only 9 x 16 = 144 data points. In model in detail. Rather, a brief description of
practice, of course, one would obtain several the results of the COSEG analysis is offered.
replicates of each candidate respondent pro- Figure 3 shows the output for the stimulus
file so as to examine within-profile homo- part of the response matrix. In the case of
geneity and provide higher stability in the credit card utilities, the solid lines show that
parameter estimates of the model. discount rate is, by far, the major evaluative
component. (It is not surprising that respon-
dents prefer smaller discount payments to 69
Applying the Model larger ones.) Speed of payment and credit
authorization procedure are also important
Componential segmentation is a model in components, while marketing support and
which a series of decomposition techniques rebate show virtually no variation at all.
are applied to t h e basic 9 x 16 matrix of Figure 4 presents respondent results, a
averaged profile evaluations of the sixteen unique output of COSEG. The solid lines
stimulus cards. The basic objective is to indicate the contribution that each back-
decompose the response matrix into separate ground variable makes to differences in over-
parameter values for each of the levels of the all stimulus evaluations. As noted, type of
five attributes comprising the stimulus cards favorite card displays the highest salience or
and each of the levels of the four profile contribution to the overall evaluation.
characteristics describing the respondents. However, COSEG also develops, if neces-
Componential segmentation (COSEG) em- sary, a set of interaction parameters, denoted
ploys more or less standard statistical tech- by the selective dotted lines in Figures 3 and
niques, such as analysis of variance, to carry 4. In this case, only discount rate in Figure 3
out the first of the sequence of decomposi- and type of establishment and type of favor-
tions. In the process of doing this, a useful ite card in Figure 4 displayed interactive
measure is computed that shows how "seg- effects. A specific interpretation of the inter-
mentable" the market is, with respect to the action effect is that restaurants whose favorite
stimuli and background variables used in the card is type A, when evaluating credit cards
study. The segmentability index is simply a with the small discount rate, evince a higher
measure of how large the interaction sum of utility than would be implied by the sum of
squares (persons x stimuli) is, relative to the the main effects (solid lines) for this combina-
total variation in the data. In the present tion.

FEBRUARY 1977
PAULE. GREEN

FIGURE 3
Stimulus Utilities Obtained from the COSEG Model

Utilities
1.0 1.0 1.0

0 ""\\.
-1.0 f I -1.0 I I -1.0 I I
21/2% 6% 1 Day 10 Days Computer Telephone
Discount Rate Speed of Payment Credit Authorization

1.0

70
1°I
0

-1.0 I I -1.0 I I
0.1% 0.75% None 15%
Marketing Support Rebate

Main effect parameters


__-- Interaction parameters favorite card was type A, would evaluate six
different credit cards, each of which is com-
(Stimulus mode)
posable from the five basic stimulus factors,
shown earlier.
Use of COSEG COSEG provides that prediction, using
parameters estimated from the calibration
The main value of COSEG lies in its ability to sample. In practice, errors are associated with
make predictions about the relative ratings of the estimation of COSEG's parameters. A
any of the thirty-two stimulus cards by any of measure of this overall error can be obtained
the thirty-six possible respondent types, as (by standard statistical procedures) for the
based on parameters developed for the small 9 x 16 matrix that was used for parameter
subsets of each. For example, suppose a estimation in the first place. Then, having
researcher wished to learn how a large restau- estimated the standard deviation of prediction
rant, honoring three or fewer cards, whose errors, one could estimate various segments'

BUSINESS HORIZONS
A New Approach to Market Segmentation

FIGURE 4
Respondent Saliences Obtained from the COSEG Model
Saliences

1.0 -- 1.0

-1.0 I I I -1.0 I I
Lodging Retailing Restaurants Small Large
Type of Establishment Size of Establishment

1.01 1.0

71

-1.0 i I I -1.0 I I
A B C ~<3 />4
Type of Favorite Card Number of Different Cards

Main effect parameters


-- -- Interaction parameters
quency of choices for each stimulus alterna-
(Respondent mode) tive under evaluation. Hence, for a specific
respondent segment one could find the fre-
evaluations of alternative combinations of quency of first choices received by each credit
credit card service levels by Monte Carlo card composite. Similarly, for each credit card
simulation. one could find the frequency across segments
Moreover, it is a relatively simple matter in which it receives first choice. In short, once
to weight the choices of specific segments in the parameters and overall prediction error of
accordance with their relative importance in COSEG are estimated, it is relatively straight-
the population of interest. The typical kind of forward to run various sets of Monte Carlo
o u t p u t from these Monte Carlo runs entails a trials involving alternative strategies of in-
cross-classification of each segment's fre- terest to the marketing manager.

FEBRUARY1977
PAUL E. GREEN

Contribution of COSEG background variables. Moreover, Monte Carlo


simulations could be set up in either case.
The value of COSEG would appear to lie as However, COSEG seems most useful in
much in the point of view that underlies its cases where both respondents and stimuli can
construction as in the specific model itself. be further decomposed into subordinate com-
Componential segmentation focuses on the ponents and where the advantages of ortho-
building blocks of segments rather than on gonal arrays (and similar kinds of fractional
specific groups that may emerge from a factorials) can be fully realized in parameter
particular type of clustering procedure (or estimation. I suspect that a number of mar-
other means for developing segments from a keting problems could be profitably formu-
specific d a t a base). If the interaction effects lated in this way.
are small, segmentation may not be warranted The high efficiency of COSEG is based on
with respect to the stimulus domain under the use of fractional factorial designs that, in
study. If the interaction effects are large, turn, permit uncorrelated estimation of re-
segmentation could be of major interest. spondent and stimulus effects. As such,
In the illustrative example we considered a COSEG is particularly Useful in the design of
case in which both respondents and stimuli stimuli and the selection of respondents.
were capable of being decomposed into sub- However, cases will often arise where ortho-
ordinate factor levels. Two other cases are of gonal estimation of respondent background
interest where: (1) respondents, but not characteristics is not feasible. If so, the
stimuli, are decomposable; and (2) stimuli, procedures described here can still be used to
but not respondents, are decomposable. make nonorthogonal estimates of the para-
72 A number of segmentation studies- meter values of interest, albeit with lower
benefits seeking, problem-oriented segmenta- statistical precision. Thus, even if one cannot
tion, life styles, brand preferences-are carried Select respondents according to orthogonal-
out in which respondents can be selected array desiderata, the principles of COSEG can
according to specific background profiles but still be applied to obtain nonorthogonal esti-
the stimuli are ordinarily not easily decom- mates of respondent-type subordinate factors.
posable into subordinate components. For
example, in needs segmentation studies con-
sumers are asked to indicate how strongly Some Caveats
they agree or disagree with each of a set of
items that characterize various needs that a As the reader has probably surmised by now,
product might satisfy. Ordinarily, these items componential segmentation is not a variant of
cannot be described as factorial combinations market segmentation in its usual sense. Some
o f more basic attributes. COSEG can still be major differences should be noted:
applied to this type of problem, although the
leverage for using it decreases. COSEG considers the question of whether
Illustrations of the obverse case frequently or not a market is segmentable to be an
appear in the context of concept testing or empirical matter, capable of being answered
general studies involving conjoint scaling statistically for the background variables and
where respondents (not chosen according to a stimuli under study.
factorial design) evaluate designed stimuli that No groups or clusterings in the usual sense
are constructed factorially. Again, COSEG are formed. Rather, COSEG develops para-
could be applied to the response data and meters for predicting how any segment com-
post hoc examination made of the relation- posed from the basic background levels will
ship between concept ratings and various evaluate similarly composed stimuli.

BUSINESS HORIZONS
A New Approach to Market Segmentation

The size o f the segment must be estimated • The traditional Ansoff table can be
from the larger data bank from which the expanded to produce a table that distin-
specific respondent profiles (needed for the guishes between strategies at the structural
orthogonal arrays) are drawn. level versus those at the functional level. This
A number of caveats should be listed in expanded table can serve as a kind of check
considering the COSEG approach: list for strategy development within various
segments.
To be most effective, COSEG requires the
• While a myriad of segmentation bases
ability to screen a large number of possible
exist, only two fundamental a p p r o a c h e s - a
profiles for the particular subset that meets
priori and post hoc (and a hybrid of the
the design conditions. Ongoing panels where
t w o ) - h a v e been traditionally applied.
respondent background variables exist in com-
• It is not always clear what subset of
puterized form are most useful for this
segmentation variables should be chosen as
approach.
the distinguished set. However, procedures
COSEG does not employ a random sam-
have recently been developed to examine the
ple. Profiles selected to meet the design
degree of association among alternative seg-
characteristics are not necessarily "typical" of
mentation bases. Moreover, if some base is
the population as a whole. Moreover, it is one
chosen b y fiat, the same approach can be used
thing to design stimuli and quite another to
to find a functional relationship between its
select respondents from a larger universe. In
clusters and those produced b y employing
the former case one can design out (or
other bases.
randomize over) unwanted variables. In the
latter case one gets all of the respondent, not • Componential segmentation represents a
73
just the profile of interest. new approach to segmenting markets, one
When the desideratum of profile ortho- that emphasizes the prediction of new respon-
gonality must be relaxed, the high statistical dents' evaluations of new stimuli (for exam-
efficiency of COSEG diminishes. Moreover, ple, new product offerings) from a relatively
interpretation of correlated parameter values small sample of factorially selected respon-
becomes more difficult substantively. dents and factorially designed stimuli.
* In the COSEG approach, the segmen-
The preceding caveats clearly call for t h e
tability of a market can be measured for the
collection of additional data for internal cross
specific background variables and stimuli un-
validation of the COSEG model's predictions.
der study.
This should be a sine qua non for using the
• Given the fact that COSEG estimates a
model.
relatively large number of parameters on the
~ 1 1 Illl The main objective of this article basis of a relatively small number of calibra-
I]l II has been to elaborate on some tion responses, it is essential that all COSEG
rather basic beliefs: studies include an internal validation step
• Segmentation works best when the where a control sample of responses are held
study design is tied in with market strategy out of the analytical stage and later predicted
questions. b y the model.

FEBRUARY 1977

You might also like