A Theory of Market Segmentation

Author(s): Henry J. Claycamp and William F. Massy

Source: Journal of Marketing Research , Nov., 1968, Vol. 5, No. 4 (Nov., 1968), pp. 388-
394
Published by: Sage Publications, Inc. on behalf of American Marketing Association

Stable URL: https://www.jstor.org/stable/3150263

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

American Marketing Association and Sage Publications, Inc. are collaborating with JSTOR to
digitize, preserve and extend access to Journal of Marketing Research

This content downloaded from

14.139.243.18 on Sun, 21 Mar 2021 06:25:28 UTC
All use subject to https://about.jstor.org/terms
 HENRY J. CLAYCAMP and WILLIAM F. MASSY*

In this article a normative theory of market segmentation is developed that

takes account of major implementation problems. The theory is presented as a
multistage mathematical model of the full range of segmentation possibilities
from the perfectly discriminating monopolist to the mass marketer. The theory's
major implications for the philosophy and application of the market segmenta-
tion strategy are discussed in detail.

A Theory of Market Segmentation

INTRODUCTION THE ECONOMIC THEORY OF SEGMENTATION

One of the most striking developments in marketing

The concept of market segmentation was dev
is the amount of interest shown in market segmentation
in economic theory to show how a firm sellin
strategy. Nearly every issue of major marketing journals
mogeneous product in a market characteri
includes discussions of it and its implications for
heterogeneous demand could maximize profi
marketing management. theory shows that optimal profits can be achi
Despite the attention devoted to this topic, littleuses consumers' marginal responses to
the firm
progress has been made in developing a normative
i.e., price elasticities, to define mutually exclus
theory of the segmentation process. Most articles
ments on and sets price (or output) so that m
segmentation tend to be either general discussions of
profits achieved in each segment are equal. Ex
the basic concept or research reports showing differ-
of these results to include other marketing va
ences in consumption patterns among specific consumer
(besides price) is easily done.
groups. The strategy of segmentation often seems tothere
Since be seems to be little doubt that marketers
roughly equated with the act of defining subparts of in the segmentation concept because
are interested
some total market. As a result, considerable controversy
of its profit implications and because the economic
-and perhaps some confusion-exist about theory
the strat-
model shows how the concept is related to
egy's implications for the optimal allocationprofit
of scarce
maximization, it can be considered as an "ideal"
marketing resources and the requirements for or effective
"optimal" approach.
implementation. Although optimal segmentation is a simple and ap-
This article takes a fresh look at the theory of market
pealing concept, at least four kinds of problems can
segmentation and its implications for marketing man-
be identified that make it exceedingly difficult to utilize:
agement. Although the article is intended as a contribu-
1. problems
tion to marketing theory, we believe it also has impli- of defining mutually exclusive market
segments
cations for the practice of market segmentation. In
2. problems
the following sections we will (a) briefly review the of measuring response elasticities on a
segment by segment basis
theoretical foundations of the segmentation strategy
3. information constraints that affect the possibility of
and identify the major barriers to effective implementa-
reaching segments selectively (i.e., the marketer
tion, (b) develop a normative theory of the segmenta-ordinarily has only socioeconomic or demographic
tion process that recognizes major implementation information about audiences reached by promotional
difficulties, and (c) investigate some of its operational
media or areas covered by distribution outlets, and
implications. it is usually difficult to find relationships between these
variables and marginal response differentials)
* Henry J. Claycamp and William F. Massy are associate pro-
fessors of marketing, Stanford University.
I See [7] for example.
388

Journal of Marketing Research,

Vol. V (November 1968), 388-94

This content downloaded from

14.139.243.18 on Sun, 21 Mar 2021 06:25:28 UTC
All use subject to https://about.jstor.org/terms
 A THEORY OF MARKET SEGMENTATION 389

4. institutional constraints that limit the ability

existing means of reaching C = g d + qI,
segments with the d
degree of price or promotional selectivity.
where di is the product demand an
The economic theory of segmented markets
of implementing the promotiona
deal with these problems. If the obvious a
by xi , both
of optimal segmentation arefor
tothe
be ith customer
realized
is the typical cost function of manuf
to extend the theory to include fundamental
uct (a function of the sum of all c
of implementation. The primary purpose
and any fixed costs of operating th
present an extension of the segmentation pr
mathematical model. include distribution or promotio
handled elsewhere in the model.
is best rewritten as an explicit func
A MULTISTAGE THEORY OF MARKET
lable variables in the marketing m
SEGMENTATION
vi' be the vector of per unit costs
As indicated, our goal is to extend theith customer, so that qj = vilxx :
classical
microeconomic theory of market segmentation to
C
take account of problems in defining segments, and = g fi(p, Xi) + E v
i=1 i=1
the existence of institutional and informational con-
straints on managers' ability to design promotional Given reve
strategies that will reach specified segments. The prob- ing profit eq
lem of using demographic, socioeconomic, or other
"practical" variables to describe consumer groups II = R -
will also enter the analysis. Although our model uses (1)
only price and promotional variables, it could be re- - g fi (Pi, xi) - vi1x.
vised to include any of the marketing mix elements.
We assume a market with firms sufficiently de- The firm's optimal marketing mix will be obtained
coupled such that strategies can be planned without when Equation 1 is maximized with respect to pi and
direct reference to problems of possible competitivethe elements of xi (for i = 1, ... N).
retaliation, at least in the short run and for some mar- By differentiating (1) partially with respect to each
keting variables. The analysis considers profit maximi-
of the controllable marketing variables and setting the
zation strategies for a single product. The model willderivatives equal to zero,2 we get the familiar decision
be developed in five stages, representing successivelyrules:3
more aggregative and easier to apply approaches to
market segmentation. Though each stage is best con-
sidered separately, we emphasize that we are not pre-
senting five different models of market segmentation.(2) (pi - ci - MC) f = - f , i = 1 , . N,
Our "final" model is represented by Stage 4. Stage 5
shows how the procedures developed in Stages 1-4, (pi - ci - MC)8xii
-- ji =j V
= 1, N
1,...m;
if carried to the limit, lead logically to a nonsegmenta-
where MC is the cost function derivative with respect
tion or "mass market" strategy.
to total demand. This development is a direct de-
scendent of the microeconomic model of the perfectly
Stage 1: Segmentation by Perfect Discrimination Among
Customers discriminating monopolist (see [7] for example),
generalized to include both price and nonprice com-
Suppose that a firm attempts to market its product petitive variables. The (m + 1)N equations must be
to N customers, each with the demand function:
solved to determine the optimal price-promotional
mix for each market (here, an individual customer)
di = fi(pi , xi), i = 1, . . - N, supplied by the firm.
where pi is the price and, xi a vector of m nonprice This model represents the market segmentation
promotional variable offered to the ith customer. If strategy in its extreme form. Each customer is identi-
the unit cost of distribution (not including promotion) fied as a segment in his right because each person's
to the ith customer is c , the firm's gross revenue demand function may be at least slightly different from
equation can be written as: his neighbor's.
N N

2 Second-order conditions will be ignored here.

R
,;=1 i=l
= (p
3 Profit maximization subject to a constraint on the total pro- -c,
motional budget would lead to slightly different decision rules but
The firm's cost
would not change the conclusions drawn equation
from this or any of the
the costs of supplying
following models. a

This content downloaded from

14.139.243.18 on Sun, 21 Mar 2021 06:25:28 UTC
All use subject to https://about.jstor.org/terms
390 JOURNAL OF MARKETING RESEARCH, NOVEMBER 1968

tional input for the

Besides the obviously severe computational ith customer. The above is m
problems
that are likely involved in solving (2), written
concisely otherinproblems
matrix form:
ordinarily preclude this approach in practical situations,
that is, (3) xi = By, i = 1,. . N,
1. Marketers are rarely able to know the form or param- where Bi is the m X n matrix of media characteris
eters of individual customer's demand functions. parameters. Equation 3 implies that when the f
When such knowledge must be obtained by statisticalsets the variables' values in the media vector y
methods, it is almost always necessary to deal withdetermines the level of all nonprice promotional v
customer groups rather than individual customers.ables for the customers. Thus the y's, instead of the x
2. It is rarely possible to pinpoint promotional efforts
should be seen as the controllable marketing variab
to specific customers or to maintain perfect price We can clarify the meaning of (3) by using a simplif
discrimination strategies. Marketers are faced with
numerical example. Suppose that the market cons
legal constraints on pricing and price leakage for
of three customers: one with high, one with mid
resale by customers. For promotion, they must
and one with low socioeconomic status. Now, assum
usually use standing promotional vehicles, such as
advertising media, with predetermined audience there are two forms of nonprice competition (maga
characteristics. ads and the amount of shelf space the retailers al
cated to the brand) and three media available to th
The model must be generalized to account for diffi- manufacturer (a "class" or prestige consumer mag
culties. We will deal with Point 2 first because it will
zine, a "pulp" or low-status consumer magazine, a
provide the groundwork for analyzing Point 1. a trade newspaper read by retailers). Assume a
that all three customers shop at the same retail out
Stage 2: Customer Segmentation with Institutional Con-
straints (These assumptions are made only to simplify
example; they are not an essential part of the mod
Suppose that the firm faces a fixed set of promo- The table gives the matrices Bi(i = 1, 2, 3) that m
tional vehicles through which it must exercise its non-be expected to occur in the kind of situation just
price marketing efforts. The fixed set of promotionalscribed. The matrices indicate that the high-st
customer reads Magazine A, the low-status custom
vehicleswe
which will be call
shall denoted by the
"media." vector
Thus, yi are
there , y2,
n "" y7 , n
media reads B, and the middle-status customer reads neit
to reach the firm's N customers. We would expectAlso, retailers are assumed to adjust shelf spac
n < N-though the model does not specifically re- response to advertising in both the prestige consum
quire this relation. Now, the elements of the nonpriceMagazine A and the trade paper, though the las
promotional vector for the ith customer can be related
ten times more effective in this regard than is the fi
to the media by the set of equations: The level of the nonprice competition vector x, e
the number of advertising exposures or shelf facin
xij = Ti(y) j = 1, - - - m; i = 1, ... N. that stimulate each customer can now be determined

We shall assume that the I,-functions are all linearby Equation 3, using the matrices in the table.
(a reasonable assumption, as demonstrated by the Before changing the profit equation developed in
following example). Thus, we may write Stage 1 to include the existence of promotional media
with specific characteristics, we will invoke two sim-
plifying assumptions. Suppose that for legal reasons,
k=l
Xij = 1 biljkk , the firm decides not to follow a strategy of price dis-
where the "media characteristic paramaters" crimination.
bijk That is, the price to each customer is to
bepromo-
represent the contribution of the kth kind of the same so that pi = p for all i. Suppose also that

HYPOTHETICAL MEDIA CHARACTERISTIC MATRICES FOR TWO-VARIABLE, THREE-MEDIA, THREE-CUSTOMER EXA

Media (n = 3)
Customer Nonprice competition
(N = 3) variable (m = 2) Consumer maga- Consumer maga- Trade
zine A (prestige) zine B (pulp) magazine

Customer 1 (high status) Advertising exposures 1.0 0.0 0.0

Shelf space 0.1 0.0 0.0

Customer 2 (middle status) Advertising exposures 0.0 0.0 0.0

Shelf space 0.1 0.0 1.0

Customer 3 (low status) Advertising exposures 0.0 1.0 0.0

Shelf space 0.1 0.0 1.0

This content downloaded from

14.139.243.18 on Sun, 21 Mar 2021 06:25:28 UTC
All use subject to https://about.jstor.org/terms
 A THEORY OF MARKET SEGMENTATION 391

the distribution costs are the same for all customers so available even fo
that c% = c in all cases. These assumptions are made (possibly except fo
only for convenience (the notation is simpler if the sub- tions). Instead of
scripts on p and c are omitted). The model's develop- audience data are
ment does not depend on these assumptions, nor do graphic and socio
we suggest that they are realistic. most, coverage ma
Recall from the numerical example that the elements for certain key pr
of the media vector y are dimensioned in physical terms, extend our Stage 2
e.g., number of exposures or number of shelf facings. tion constraints.
Let the vector w' = wl, w2, ... w, be the per unit
Stage 3: Microsegmentation
costs of using the media. For example, if wl = one
cent, a one-unit "buy" in medium 1 (the high-status Suppose that media circulation is known only for a
magazine in the example) will cost one cent per ex- total of M mutually exclusive and exhaustive consumer
posure or ten dollars per thousand. Substitution of classes, which are defined by socioeconomic, demo-
Equation 3 for x, ,w' for v'i , and setting pi = p and graphic, or similar variables. (These classes will be
ci = c in Equation 1 yields the following Stage 2 called media descriptor classes or, alternatively, micro-
profit equation: segments.) The media characteristic coefficient matrices
now refer to the descriptor classes rather than to
II = R - C = (p - c) Efj(p, B y) individual customers--we have B, = 1,... M,
(4) i
where, for example, a given matrix might refer to
- w'y -personsg
"high-income, high-educated over 65." In (
principle, these matrices can be determined from
Equationaudience
4 must
survey information.4 Introducing descriptorbe
the n elements of
classes leads to the following modification of the Stage y
Differentiation
2 decision rule presented in Equation 5: wi
ple aggregation of
(p - c( - MC) Z blik ajf
(p - c- MCZ - -
ap (6) I ?x axij
= wk;k = 1, .; n,
(This relation is the same for all subsequent models
where the
and will not be considered further.) notation
The iel means all persons within the
derivative
with respect to a given medium Ith descriptor cell.
variable yk is:
It is obvious from Equation 6 that the constraint
II 0 = (P - c) a afi xij Wk
ing of all members in each media descriptor class.
on media audience information leads to equal weight-

y3Yk ? Oxij yYk

The term Tirl afi/axij represents the aggregate mar-
- MC af axij ginal response to be expected from all persons in de-
? ? Ox i 3yk scriptive cell 1. Thus, imposing a constraint on media
Transposing and recognizing that axij/ak = bijk,information automatically relaxes the information
requirements with respect to the individual response
we have the following set of decision rules for media:
derivatives.
We have named the segmentation level represented
i xi by Stage 3 microsegmentation because segmentation
(5) (p-c--MC).EE bikk W ,k = 1,..n.
by media descriptor classes is the least aggregative
This result differs from the one for S
degree of promotional discrimination possible given
part of Equationexisting
2)institutional
because weighte
constraints and media research
response derivatives
methods. af/jaxij are u
equations, instead of individual t
equations. Stage 4: Macrosegmentation
Stage 2 accounts for what might b
Now let us consider the problem of estimating the
tional constraints that restrict the ma
marginal response of sales to promotion. It seems
freedom of action. Note, however, tha
likely that purely judgmental methods will be insuffi-
requirements for Stage 2 are even
cient to determine the demand functions or their
than they were in Stage 1. Besides
derivatives. Some kind of empirical approach is needed
vidual response derivatives af./xiy ,
get estimates of the parameters of
4This model implicitly assumes that media cover a given de-
functions 'Ii(y). scriptive class homogeneously, i.e., problems of audience accumu-
Information at the required
lation and duplication are ignored. level o

This content downloaded from

14.139.243.18 on Sun, 21 Mar 2021 06:25:28 UTC
All use subject to https://about.jstor.org/terms
392 JOURNAL OF MARKETING RESEARCH, NOVEMBER 1968

disaggregative
if we are to obtain estimates of the response microsegments,
derivativesit seems reasonable to
in (6). use these media descriptors to build more aggregative
Empirical analyses of sales response to price or demand descriptor classes. Therefore, we define a
promotional variables may be discussed in terms of macrosegment as follows: macrosegment h consists of
either individual or aggregative demand functions. the customers in media descriptive cells leh. This defi-
The first approach is very difficult. The only substantive nition ensures that it will be possible to make media
effort to deal with individual demand functions of decisions for each segment. Since media characteristic
which we are aware is included in Duhammel's study coefficients can be found for each microsegment 1,
[1]. Though the results were interesting, they do not the media characteristics for macrosegments h can
give us confidence in the practical efficacy of a fully be found by simple aggregation.
disaggregative approach. If we are to conduct moreThe promotion rule for Stage 3 (Equation 6) is
aggregative statistical demand analyses, however, we easily modified to accommodate the higher level of
must decide how much to aggregate and on what varia- aggregation.
bles the process should be based. Equation 6 gives an
immediate answer-at least in part. According to the (p - c - MC)Z> l { blik}
results of Stage 3 we can always aggregate to the level (7) j h leh ch jet axl
of the smallest microsegment for which media informa- =wk ,k= 1, .. n.
tion is commonly available.
But aggregation to the level of (6) may not be enough.
Demand function analysis involves estimation of the The sensitivity
written termto{~eh
simply as afh/axhi E~_,that
emphasize afi/axij}
it refers might be
change in sales to be expected per unit change to inthe aggregate demand function for the hth macro-
promotion, but media audience research concerns segment.
average audience levels for descriptor classes. If
Stage 5: The "Mass Market" Concept
we consider data sources for these analyses, it be-
comes apparent that sample sizes sufficient to measure It will be useful to present another generalization of
the elements of B1 for a given descriptor class, withour market segmentation model, this one correspond-
reasonable degree of accuracy, may be insufficienting to to the case in which no segmentation strategy is
measure Ea af/laxi . As Frank and Massy [2, 3], practiced at all. Profit maximization without segmenta-
and others have shown, the estimation of response tion leads to the following decision rule for promotion:
coefficients is not easy. Even if the analysis is based
on time series data, the time series must be based on
the buying behavior of a sufficient number of families (8) j 1=1 Z=1 el (?xijl (P - c - MC) { blik }{ zf}
to avoid gross instabilities.
Since the sample sizes necessary to estimate response
= wkk = 1, ..., n,
sensitivities for a given microsegment must often where
be the first term in the brackets represent
rather large, the researcher is faced with two alterna-
impact of medium k in terms of promot
tives: (1) using the maximum number of descriptorfor all numbers of the population, and the se
cells, and hence a very large overall sample size, isorthe derivative of the total market demand function.
(2) aggregating over descriptor classes to form a smallerAs easily recognized, (8) is the same as (7) if there is
number of new classes with adequate numbers of re- only one inclusive macrosegment.
spondents in each, while keeping the total sample within
bounds. The total sample size for a study is often fixed SOME IMPLICATIONS OF THE THEORY
(as when working with syndicated panel data), in
which case the second alternative is the only feasible The theoretical models presented in the preced
one. Or perhaps the cost of data collection precludes section have several implications for both the ph
phy
using a large overall sample size. Finally, the cost of and practice of segmentation.
audience research is usually absorbed by the mediaFirst, it seems clear that segmentation shou
(or agencies), suggesting that larger sample sizes will considered
be as a process of aggregation rather tha
possible there than for product or brand specific sensi-aggregation. For example, recall that the the
five stages dealt with the full range of segment
tivity analyses, if only because the costs can be divided
among many users. possibilities. That is Stages 1 and 2 treat indiv
consumer units as segments; Stage 3 deals wit
This reasoning implies that aggregation beyond the
gregation of consumer units into microsegm
minimal descriptor class sizes dictated by available
media audience statistics will be the rule rather thanStage 4 considers aggregations of microsegm
the exception if statistical methods for estimating de- into larger groups (macrosegments), and Sta
mand sensitivities are to be used practically. How deals with the mass market or complete aggreg
of consumers.
should the aggregation be performed? Given the po-
tentialities of media research for dealing with relatively It should be obvious that if consumers have different

This content downloaded from

14.139.243.18 on Sun, 21 Mar 2021 06:25:28 UTC
All use subject to https://about.jstor.org/terms
 A THEORY OF MARKET SEGMENTATION 393

particular,
responses to the firm's we consider the question
marketing of how media a
variables
descriptorin
are no scale diseconomies cells (microsegments)
fitting should be allocated
specialize
to individual consumers, segmentation at th
to macrosegments.
individual consumer units
It is clear that(the case
if the response of
derivatives forthe
all mi- p
discriminating monopolist)
crosegments included in would yield m
each of the macrosegments
profits. However, this discussion
are identical, showed
Equation 7 is merely a factored form of
if the dubious assumption
Equation 6. For the of jth typeno scale
of nonprice eco
competition
the marketing mix is we have:
valid, other constraint
preclude this form of segmentation. For exa
of information about the response characte
1 iet aX;j h ich
groups reached by promotional media and in E blik E Z { blik
constraints on the flexibility of their use requi
gation at least to the level of fEZE fi , k=1, 1) n,
microsegments
IEh jet 3oxi)
Additional aggregation to macrosegments
mately to a single segment-may
if be required
of difficulties in measuring response differ
specific groups.
It is easy to see that addition offxij_ic*afX9ij
Sicl for all i, l*eh.
of the succes
straints and corresponding higher levels of a
must reduce the level (Recall of
that we canthe
do nothing about any possibility
firm's profit. (
ment ignores research costs
that the af//xij are heterogeneous required
within a given mi- to im
crosegment without changing audience
given level of segmentation.) That research pro- is, the app
Stage 1 segmentationcedures.)
yields more profit tha
Stage 2 more than Stage This consideration3, suggests
etc. that forThis a given kind is
of a dir
competitive activity
of the mathematical properties of constrain it will be useful to form segments
such that
unconstrained maxima. Thus the fundament
of market segmentation can be characterized
the point at which the marginal reduction
caused by the imposition variance
of ( ,janother
for leh, cons
level of aggregation,is asis small just
as possible forbalanced
each of the macrosegments.by the
reduction in research and administration costs made
This means that microsegments (1) should be assigned
possible by the constraint. to macrosegments (h) in such a way that the within-
We argued earlier that this balance will likely occur
group variances of microsegment response coefficients
at the macrosegmentation stage. Thus the basicare re-
as small as possible relative to the between-group
source allocation problem in segmentation involves variance.5 (The media characteristic coefficients blik
finding the values of the controllable marketing varia-
are not included in the variance formula because the
bles that bring the decision rule in (7) to equality.
macrosegments must be developed before a specific
The problem of finding the solution to (7) can be handled
promotional program is chosen.)
with standard mathematical programming procedures Macrosegment formation requires grouping micro-
when the necessary data have been collected and the
segments by similarities among their promotional
macrosegments have been defined. Though the problems
responses. How can information on the response de-
of finding a solution are not trivial, they will not be
rivatives for each microsegment be obtained? It will
explored here.
generally be necessary to determine the kind of pro-
The concept that segmentation is a process of ag-motional variables that are likely to be used in the
gregation implies building to a viable segmentation
marketing mix and define specific response variables
strategy rather than tearing a market apart to find one.
for each of them. (The change in the buying probability
This may be a fine point in regard to the philosophy
caused by an additional advertising exposure would
of segmentation, but it appears to be important for
be one such measure.) Often it will be necessary to
the implementation of the strategy. It is impossible
useto
surrogates for the response variables, as when the
form meaningful market segments without taking
institutional and information constraints into account
5 The average within-group variance will surely decline as more
and this means building from the point at which the
macrosegments are permitted, assuming that optimal allocations
constraints are felt, namely from persons to micro-
are made at each stage. As noted earlier, the question of determining
segments to macrosegments. the optimal number of segments must be resolved by analyzing the
trade-off between costs of research and marketing administration
Criteria for Forming Macrosegments (which rise with the number of segments considered) and the gains
to be expected from reductions in within-group variance. The analy-
Let us now consider some implications of the theory
sis will depend on factors specific to problem and product class and
for the formation of meaningful macrosegments. In is beyond the scope of this article.

This content downloaded from

14.139.243.18 on Sun, 21 Mar 2021 06:25:28 UTC
All use subject to https://about.jstor.org/terms
394 JOURNAL OF MARKETING RESEARCH, NOVEMBER 1968

response of consumer attitude research, we conclude that the

or awareness data sources and re-
measures
to changes in advertising are search usedmethods now available
in place of theare sufficient
sales to permit
response. their application. The value of our theory of market
Looser response measures, e.g., less adequate sur-
segmentation will be proved if it leads researchers to
rogates, will more usually be dictated when workingexperiment with these techniques that allow a system-
at the microsegment level than will subsequently beatic attack on the practical problems of segmenting
used when dealing with macrosegments. This willmarkets.
occur because of the many microsegments for which
response measures must be estimated and the diffi- REFERENCES
culty of obtaining data on individual microsegments.1. William F. Duhammel, "The Use of Variable Markov Pr
However, this difficulty will be reduced because the as a Partial Basis for the Determination and Analysis of M
response measures will be reestimated at the macro- Segments," Unpublished doctoral dissertation, Graduate
segment level. of Business, Stanford University, March 1966.
The last problem raised by the theory is to find a2. Ronald E. Frank and William F. Massy, "Market Segme
and the Effectiveness of a Brand's Price and Dealing Po
method for optimally grouping microsegments into Journal of Business, 38 (April 1965), 186-200.
macrosegments, assuming that information on the 3. "Estimating the Effects of Short-Term Promotional S
relevant response derivatives is available for each in Selecting Market Segments," in Patrick J. Robins
microsegment. That is, we want to make assignments Charles L. Hinkle, eds., Sales Promotion Analysis: Som
plications of Quantitative Techniques, Philadelphia: Mar
such that the resulting macrosegments consist of micro-
Science Institute (in press).
segments with large within-group and small between-4. Harper W. Boyd, "Correlates of Grocery Product Co
group homogeniety of response derivatives. tion Rates," Journal of Marketing Research, 4 (May
184-90.
Making such assignments is a problem in optimal
taxonomy, also called cluster analysis.6 It is fortunate5. Paul Green, Ronald Frank, and Patrick Robinson, "Cluster
Analysis in Test Market Selection," Management Science, 13
that such programs have recently been developed, (April 1967), B387-400.
because without them macrosegment formation would 6. William F. Massy, "The Cost of Uncertainty in Advertising
largely be guesswork. Although many detailed problems Media Selection," Graduate School of Industrial Administra-
in using these procedures must be solved by future tion, Carnegie Institute of Technology, Working Paper No. 2,
December 1966.
7. Joan Robinson, The Economics of Imperfect Competition,
6 For a description of one optimal taxonomy procedure, see [8]. London: Macmillan & Co., 1954, 179-88.
For an application of the clustering method to a marketing prob- 8. Jerrold Rubin, "Optimal Classification into Groups: An Ap-
lem, see [5]. Claycamp and Massy are doing research on using the proach for Solving the Taxonomy Problem," International
Rubin cluster analysis program to form macrosegments as indi- Business Machines Corporation, New York Scientific Center,
cated above. Results will be in a future article. May 1965.

This content downloaded from

14.139.243.18 on Sun, 21 Mar 2021 06:25:28 UTC
All use subject to https://about.jstor.org/terms