Multivariate Analysis

UNIT 14 CLUSTER ANALYSIS AND

MULTI-DIMENSIONAL SCALING
Objectives

After going through this unit on Cluster Analysis, & Multidimensional Scaling you
should be able to understand the following
• Importance of Cluster Analysis in today's market conditions
• Inputs for cluster analysis
• Different approaches to cluster analysis
• Nature of Multi-Dimentional Scaling (MDS)
• Applications of MDS
Structure
14.1 Introduction
14.2 Inputs for Cluster Analysis
143 Different Approaches to Cluster Analysis
14.4 Factors to be considered while using Cluster Analysis Techniques
14.5 Multi-dimensional Scaling
14.6 Applications of Multi-Dimensional Scaling
14.7 Summary
14.8 Self-Assessment Exercises
14.9 Further Readings

14.1 INTRODUCTION
Cluster analysis is a technique that is used in order to segment a market. The
objective is to find out a group of customers in the market place that are
homogeneous i.e., they share some characteristics so that they can be classified into
one group. The cluster/group so found out should be large enough so that the
company can develop it profitably, as the ultimate objective of a company is to serve
the customer and earn profits. The group of customers that the company hopes to
serve should be large enough for a company so that it is an economically viable
proposition for the company. This is also true for the customer as customer would not
be willing to pay beyond. a certain price for a particular product (price of course is a
function of positioning of product, cost of production etc.).

As k example, let us consider the Watch Industry. There could be many ways in
which the Watch Industry could be segmented which are as follows
a) Gender (Male/Female)
b) Technology (Digital/Analog)
c) Design Features
d) Occasion of Use (Formal/Casua/Party)
e) Price (Low/Medium/High/Jewellery)
Some of the above segmentation factors are demographic (price, gender) whereas
some are psychographic factors (occasion to use.)

This, therefore, presents a problem to the market researcher/company, as to how to

identify combination of factors that can be used to segment the market place. It is not
36 always
possible to segment a market on the basis of one single factor. Thus, a combination of Cluster Analysis and
factors must be used to segment the market place. And this is where Cluster Analysis Multi-Dimensional
technique specifically deals with how objects (people, places, products) should be Scaling
assigned to groups, so that there should be similarity within the groups; and as much
difference between the groups, as possible.

In the above example, cluster analysis has identified three distinct clusters (groups) as
follows:
Group I: People who spend less money but have more number of vacation days.
(adventure seekers, students) '
Group II : People who spend average amount of money and have average number of
vacation days (1" class train travel, 3-star hotels etc.)
Group III: People, who have very less vacation days, but spend heavily in these few
days (five-star luxury, air travel, foreign trips etc.).
These grouping would help a travel and tourism company in the following ways:
a) The travel and tourism company can study the various. clusters of customers
that have emerged, and decide which customer group it would like to serve,
depending upon the company's own resources and capabilities, the volume of
business in each cluster group that will generate sufficient business for the
company's own survival.
b) Once a cluster has been selected by the company, it can tailor various tour
programmes for its cluster of customers.
c) The company can keep profile of its customers, and identify any new emerging
group of cluster.
d) The company can decide to serve either only one group of customers (Niche
marketing), or serve all groups of customers, at one and the same time, but
having a range of travel and tour programmes.

14.2 INPUTS FOR CLUSTER ANALYSIS

Since cluster analysis is used for the purpose of segmentation of the market place, for
using the cluster analysis -technique, one has to go through the following procedure
on a very broad level:
i) Company identifies its own product capabilities, financial strength, distribution
network (Who am I?)
ii) Company identifies the basis on which the customers are to be clustered e.g.,
one factor segmentation (age, income, gender, geographic location) or two-
factor basis of segmentation of the market place (age and income, quality and
price, nutrition value and price), or multi-factor basis of segmentation of the
market place (Who am I going to serve?).
The company/market research agency has to decide which of the attributes are to be
employed that' would generate natural groupings in the market place. The choice of
variables that are 37
Multivariate Analysis to be used to cluster the objects is one of the most critical decisions analysis make.
The best way is to select those variables that make sense conceptually, rather than
using any or all variables. Ideally, the variables should be selected within the
framework of an explicitly stated theory that is used for supporting the classification.

iii) Respondents who are part of the research work are given to fill up questionnaire.
The questionnaire is typically Interval scaled (5pt, 7pt, 10pt) questionnaire.

The cluster analysis program that is run on the computer identifies the different
groups (clusters) or respondent groups such that the members. of any one • group
(cluster) are similar to each other, but different from the respondents in the other
groups. The computer program for cluster analysis searches through the answers that
the respondents give to the various questions in the questionnaire, and groups
together those respondents who have given identical ox almost similar answers of a
certain combination of the questions. These respondents whose answers are similar
are formed into one cluster. The procedure is then continued till more' cluster are
formed.

14.3 DIFFERENT APPROACHES TO CLUSTER

ANALYSIS
There are various approaches adopted for cluster analysis

One extreme approach is to cluster respondents in terms of all available information

about the respondent - although this is very thorough, but may lead to difficulty in
interpretation and marked difference between groups i.e., the clusters or groups of
respondents that are produced may be a hotch-potch, rather than clearly marked
description of the target population.

Another extreme approach involves grouping respondents only .on data that is
directly representing the specific consumer need. Example - quality, price of .a
product. This has the advantage of clarity and accuracy, but may be quite limited in
describing the target population.

The way one approaches, cluster analysis depends upon the objective of study.
Suppose the objective' of the company is physical development of an existing
product, then the consumers needs regarding the product are the most suitable
variables for forming. clusters.

Suppose an advertisement for a product is to be developed, then the attitude of the

consumer towards the product gains more importance, as there are groups of people
who have positive, negative or neutral .attitude towards product/company. And the
advertisement developed' would be targeting specific clusters of consumers,
depending upon the product/communication need.

Suppose a company wishes to develop a. new product concept/product changes in the

existing product offering, then also attitude of consumer- towards
product/company/product category gains more importance over other attributes when
developing clusters,

Therefore, depending upon the objective of study, various .characteristics can be

Used to describe the consumer, his attitudes, needs, demographic characteristics.

A general principle followed: in .cluster analysis is to omit behavioural and brand

image data. One should concentrate on characteristics describing the consumer and
the circumstances e.g., specific needs, demographic variables, general attitude and
select any variables which would interact with each other to influence consumer
behaviour. Once attributes are identified on the basis of which one is trying to form
natural groupings of consumers; one has to scale the attributes, so that one has a basis
38
to measure the similarity/differences between/within groups..
Example: Cluster Analysis and
Multi-Dimensional
Suppose one wants to understand the consumers who operate in the finance market. Scaling
With regard to the finance market, there are few outstanding aspects like Risk,
Returns, Liquidity. One could use any two aspects at one time to form clusters of
consumers or could use more than two attributes also.

There are 1'5 consumers, whose responses have" been analyzed- with respect to. Risk
and Returns in the financial market.

− Cluster' I is. a group of consumers who believe in .High Risk High Returns
instruments. e.g., Equity.

− Cluster II is a group of consumers who believe in Average Risk Average

Returns (Mutual funds, 'Company fixed - deposits)

− Cluster III is a group of consumers who believe in Low Risk Low Returns
(savings bank, fixed deposits).

An obvious way to measure similarity is to calculate the Euclidean distance between

points. In the above figure, the distance between points A, and C would calculated as

(A,C) = (Xc1 - XA1 ) 2 + (Xc 2 - XA 2 ) 2

And so on, one could calculate distance between various points. In three dimensions,
the expression for distance between A and C would be

d
(A,C)2 = (XC1 -XA1 ) 2 + (XC2 - XC2 ) 2 + (XC3 - XA 3 ) 2

39
Multivariate Analysis

The distance between any two points is an inverse measure of similarity. Thus, larger
the distance between two points , greater is the difference or dissimilarity between
the two points. The closer the distance between two points , the more the similarity
between them . Also more the number of respondents to be classified into clusters
more will be the number of distance calculations. Hence for large sample sizes of
respondents, computer would have to be used.

After calculating distances , one has to decide how to form clusters. There are various
method of doing it as follows:

i) Nodal Method

In this method one selects a point or points that will serve as nodes for forming
clusters or focal points. The rest of the points are then allocated to each cluster on the
basis of their distance from the focal point .

The above method is demonstrated as follows:

a) Choose as nodes those objects (points) that have the greatest distance between them.
b) Consider these two objects as focal nodes and allocate all the remaining objects
to one cluster or the other based on their distance from the polar nodes.
c) Split the resulting two clusters in the same way. Continue this process till the
collection of points is split into its original members.
In the above example , the points ‘D’ and ‘M’ have the biggest between them , and
they will be considered as nodes for the two clusters .Each of the remaining objects
would then be allocated to each cluster on the basis of the shortest distance to either
D or M. Thus the original cluster would be:
Group1 : DACE GK
Group2 : MBFHIJLNO
In Group 1, the least similar (greatest distance) are D and K . They would thus be
considered as new nodes, and the items in Group 1 would be allocated to each of the
new clusters on the basis of their distance from these new nodes . The new groups
would be:
Group 1A : DACG
Group1B : KE
Similarly , Group 2 would be divided (using B and J as new nodes ) to yield)
Group 2A : BFHLM
40 Group 2B : JINO
Thus, alter every such stage; one has to find out if the sub-groups that have emerged Cluster Analysis and
can be further combined on the basis of some measure of similarity between the Multi-Dimensional
objects. Scaling

ii) Factor Analysis Method

Another way of developing clusters is by using the method called as Q-factor
analysis. By this method one can determine which objects logically belong together:
This is also called as inverse factor analysis.
iii) Linkage Method
There are three methods in this Single Linkage Method, Complete Linkage Method
and Average Linkage method:
Using the previous example of analyzing financial markets, the method starts by
finding out the points with shortest Euclidean distance. Therefore, initially the
following groups emerge:
C-D - Group 1
A-G - Group2
N-O - Group3
FM - Group4
In the next stage, depending upon the cut-off" distance, remaining objects are
associated to the above clusters. The final clusters that emerge are
Group l - AGDC
Group 2 - NOKELT
Group 3 - MFLBH
The complete linkage option starts out in the same manner by clustering the two
closest points. However, the distance between two clusters is the longest distance
from a point in the first cluster to a point in the second cluster,

In the average linkage option, the distance between the two clusters is the average
distance from points in the first cluster to points in the second cluster.

Activity 1

Study the products and the consumers of your company (or the company you are
familiar with) and try to develop cluster of customers depending on the different
requirements of the customers.

…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
………………………………………………………………………………………….

14.4 FACTORS TO BE CONSIDERED WHILE USING.

CLUSTER ANALYSIS TECHNIQUE
The following aspects should be kept in mind while using cluster analysis
method:

a) A number of cluster may emerge after doing analysis . However , their is

limit to the number of cluster that a company can consider due to
− Limitation of market potential within a cluster
− Difference between clusters not sharply defined
41
b) Cluster analysis provides a way of segmenting the market but these segments are
Multivariate Analysis not water tight compartments. Products that are developed for a
particular segment may attract people from other segments too.
c) The characteristics of a cluster may change over time, as the consumers
economic status, education, lifestyle, etc., change over time then the
company has to take a relook at the market place.
d) The clusters that have been identified are used developing further
marketing strategies in the areas " like 'product developments,
advertising research, distribution strategies, pricing strategies etc,
e) The most important assumption in cluster analysis is that the basic
measure of similarity on which clustering is based is a valid measure 'of
the similarity between objects. A second major assumption is that there is
theoretical justification and basis for structuring objects into clusters. As
with other multivariate techniques, there should be theory and logic
underlying the cluster analysis.
f) The major limitation of cluster analysis is the difficulty in evaluating the
quality of the clusters. It is very, difficult to know exactly which clusters
are very similar and which objects are dissimilar, and also difficult to
select clustering criterian.
In conclusion, cluster analysis is a scientific method of help understand the consumer
groups with 'their differing needs and perceptions.

14.5 MULTI-DIMENSIONAL SCALING

Multi –dimensional scaling (MDS) technique or perceptual map or positioning map is
a multi- variate analysis technique that is particularly suitable for measuring human
perceptions and preference for products/brands. MDS is concerned with spetial
representation of relationships among behavioural data i.e., buyer perceptions and
preferences.

MDS Fundamentals

a) Metric MDS

b) Non-Metric MDS

Consider the following geographic location of various cities in our country:

The distance between any two points on the snap is given by the formula

d12 = ( X 1 − X 2 ) 2 + (Y1 − Y2 ) 2
42
The distance between various cities could be calculated and put in a tabular form as follow
 Figure 2 Distances in Kms. Cluster Analysis and
Multi-Dimensional
Cities Bombay Pune Nasik Ahmedabad Delhi Scaling

Bombay
Pune
Nasik
Ahmedabad
Delhi
In short, it is a relatively simple matter to go from the map in figure I to the set of
numerical distances in figure 2.
However, the converse or going form figure 2 to figure I is very difficult. Given
different objects and their attributes, every consumer perceives the objects in a
different manner.
So, one has to first understand the perceptions (distance,-,) in the mind of the
consumer regarding products, brands, and plot a graphical representation for the
same. The task is to work backwards. That is, we would have to find out
simultaneously, the
• Number of dimensions (can be more than two-dimension perceptual mapping)
• Configuration .(or patter4) of points in that perceptual space, so that their
computed inter-point distances most closely match the input data of figure 2.
This is the problem of metric MDS. Whenever input data is interval-scaled or
ratio-scaled, it is called as Metric MDS.
The same input data in the format of figure 2 could be transformed into a rank- order.
Suppose we take the smallest distance between two cities- and give in the Rank 1.
Then we could arrange the remaining distances between cities in ascending order and
give it ranks from 2 onwards up to Rank 13 (e.g., largest distance could be between
cities Calcutta and Bombay on the map. We therefore find with this method.

• Number of dimensions
• Configuration (pattern) of points in that dimensionality, so that the ranks on
their computed inter-point distances most closely matched the ranks of the input
data. This is the problem of non-metric MDS. Whenever input data is in the
form of norminal or ordinal form, it is called as non-metric MDS.
Whichever method for MDS one select (metric; and non-metric MDS), it does not
change the relative distances of the points.

Attribute Space - Objective Space and Perceived Space.

Any product or service can be visualized as composed of both objective and

perceived attributes or `dimensions'. A company has an objective description of its
product in physical or chemical terms. Such an objective attribute `space' in which
various brands are viewed as points positioned in space, will usually not agree with
the buyers perception of the products. The buyers perception about a product, service,
company is a sum total of not only the company which is selling the product, but also
other factors in the market like competition, competitor's products Advertising, after-
sales service etc.
Also note that the dimensions of perceived space need not be the same as, the
objective space. As an example, for many consumer goods that are highly
individualistic items like tea, coffee, toothpastes, perfumes, shampoos, -where strong
individual preferences play a key role in buying decisions, the objective space of the
company and perceived space of the consumer may be widely different. But in the
case or industrial product, the perceived configuration of brands may agree rather 43
closely with an objectively constructed configuration in which
Multivariate Analysis measurements of such characteristics as speed, reliability, efficiency can be made
rather straightforwardly.

Ultimately, the view of the consumer is the most relevant one, as this is what would
make the consumer go out and buy a particular product/brand.

Example of Multi-Dimensional Scaling

A study was conducted in a leading business school (A) among the first year MBA
students to find out how other, business schools were `similar' or `different from the
`
A' business school. Various attributes are used to evaluate a business school -
faculty, faculty/student ratio, starting salaries of the MBA students, physical facilities
like library, computers, and so on. But which of the attributes do students `actually'
use in making a competitive comparison between the business schools?

Let us consider six leading business schools in a place as A, B, C, D, E, F. Around

hundred students were given a questionnaire in which the students were asked to
respond to the questions which indicated various attributes of a management school,
and students had to mark schools that were similar/dissimilar on the different counts.

From the information that was gathered from the questionnaire and subjected to NOS
software program, what emerged was that course content and starting salary are the
most important criteria by which, students judge a business school.

The multi-dimensional scaling technique is therefore very useful in plotting the

perceived perception of the consumer about different brands/products/services. And
this analysis could be used by the marketing manager of a company for developing
further marketing strategies.
Activity 2
Consider the following products and try to develop a perceptual map of the various
brands in that particular product category. Before attempting such an exercise, the
student should understand the product category in its totality i.e., target consumer,
pricing strategy, advertisement strategy, distribution, usage and attitude for the
product, and then try to develop a positioning map, (NOS) for the same: Once a
product has been studied, the study should try to identify two dimensions along
which 'brands could be clubbed as similar/ dissimilar.
1) Toothpaste (Colgate Gel, Colgate Total, Pepsodent 2-in-1, Babool, Close-up etc.)
2) Shampoo (Kyle Herbal, Dabur Vatika, Sunsilk, Lakme, Clinic etc.).
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
…………………………………………………………………………………………
44 ………………………………………………………………………………………….
14.6 APPLICATIONS OF MULTI-DIMENSIONAL Cluster Analysis and
Multi-Dimensional
SCALING Scaling

Some of the typical marketing applications that emerge from the MDS technique are
1) Market Segmentation
Market segmentation is the technique of trying to identify groups of consumers who
exhibit commonality of perception of products and preferences, One can use MDS
techniques to identify present perceptions of products by consumers, and use it
modify the company's product, package, advertising, additional features, so that the
product offering of the company moves more and more closer to the `ideal'
requirement of the consumer.
2) Advertisement Evaluation
The MDS technique could be used at the stage of advertisement pre-testing. Once an
advertisement has been developed, it could 'be tested for similarity/dissimilarity with
other advertisements in the same product category. As the ultimate objective of an
advertisement is to communicate, with the target consumer effectively, and this is.
possible only if the advertisement is distinct in its message from the other competing
advertisements,
3) Product Re-positioning Studies
If &-company is interested in re-positioning its product/service (in the mind of the
consumer), the first and foremost activity to be done is to assess the. current
perception of the product in the mind of the consumer. The classic re-positioning
case is that of Cadbury chocolates, which kept on assessing its p9sitiolihng platform,
and successfully moved Chocolates from a product perceived. as one for children, to
a product which could be consumed by a person of any age,, at any time, of the day,
and for varied occasions.
4) New Product Development
MDS technique shows us the various perceived perceptions of the different brands.
Spaces/ Gaps in the product perceptions could be used' to, develop new offerings for
the target consumer.
5) Test Marketing
MDS technique can be used to identify cities that have similar demographic
characteristics, and one could then identify a city which could represent a national
character, and use that city for test marketing.
One can thus `observe that MDS is a very useful technique to help understand the
market place and develop strategies for the future.
Advantage of MDS,
The advantage of NOS methods is not in the measurement of physical distances, but
rather "psychological distances", also called as `dissimilarities'. In MDS, we assume
that every individual pawn has a 'metal map' of products, people, places, events,
companies, and individuals keep on evaluating their external environment on a
continuous basis.
We also assume that the respondent is able to provide either numerical measure of his
or her perceived degree of similarity/dissimilarity between pairs of objects, or can
rank pairs of objects (ordinal scale of measurement) in terms of
similarity/dissimilarity to each other.
We can then make use of methodology, of MDS to construct a physical map in one or
more dimensional whose inter-point distances (or ranks of distances) are most 45
consistent with input data.
Multivariate Analysis Now-a-days a number of software programmes are available for conducting MDS
analysis. These programmes provide for a variety of input data. Some of the widely
used softwares include MDPREF, MDSCAL SM, INDSCAL, PREFMAM, PROFIT,
KUST.

14.7 SUMMARY
In this Unit we have discussed two of the techniques of multi-variate analysis viz.,
`Cluster Analysis' and Multi-Dimensional Scaling. Cluster Analysis is mainly used
for market segmentation. Some of its other applications include test market selection
and to establish grouping of products within a product line. Various inputs required
for -cluster analysis have been discussed in the Unit. Further various application to
cluster analysis have been explained. Finally some of the important consideration
which should be kept in mind while using cluster analysis have been highlighted. In
this Unit we also discussed the Multi-dimensional Scaling technique or perceptual
map or positioning map is a multi-variate technique that is particularly suitable for
measuring human perceptions and preferences for products/brands. Its applications
include market segmentation, positioning studies, test marketing etc.

14.8 SELF-ASSESSMENT EXERCISES

1) What is cluster analysis? What are its possible applications?

2) State how the following techniques differ from each other (a) Cluster Analysis
(b) Discriminant Analysis.

3) An airlines marketing manager wants to segment his customer. How can cluster
analysis be used for this purpose?

4) Give few examples of marketing situations where cluster analysis can be used.

5) Discuss with the help of examples the areas where Multi-dimentional Scaling can
be applied for marketing.

14.9 FURTHER READINGS

Green, Tull and Albaum "Research for Mcarkeint Decision", PHI. N. Delhi.

Majumdar, R. "Marketing Research: Teat, Applications and Case Studies Wiley, New.
Delhi.

Beri, G.C. "Marketing Research", Tata McGraw Hill Publishing Company Limited,
New Delhi. Parasurama,-.A. "Marketing Research", Addison Wesley Publishing
Company Inc., U.S.A.

46