MARKETING RESEARCH

Topic 7.
MULTIVARIABLE ANALYSIS. PART 2
Lecturer : Inna Alexeeva A.
[email protected]
 7.4. Multidimensional Scale (MDS)

▪ MDS is a data analysis which represents distances – dissimilarity data – that

indicate the grade of disparity or similarity among the objects.
▪ MDS represents a structure of a set of objects according to the distance
between pairs of objects.
▪ Each object is represented through a point in a multidimensional space and
its disposition is strongly related to the similarity between pairs of objects.
▪ It builds a map of relative positions between pairs of objects on the basis of
data that quantify the differences between the two objects.

2
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.4. Multidimensional Scale (MDS)

Entry data Election of an MDS

Problem statement
acquisition procedure

Allocation of labels
Evaluation of the Decision on the
to the dimensions &
reliability and number of
interpretation of the
validity dimensions
configurations

Analysis Measured data Nature Scales

MDS Metric scale Quantitative Proportional intervals
Non-metric MDS Non-metric scale Qualitative Ordinal

3
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.4. Multidimensional Scale (MDS)

Entry data acquisition

Perception data: direct focus through application of scales
These data are known as similarity assessment. E.g., similarity assessment on
the all possible pairs of brands of toothpaste can be obtained the following
way:

Muy diferente Muy semejante

Crest – Colgate 1 2 3 4 5 6 7
Aqua-Fresh – Crest 1 2 3 4 5 6 7
Crest – Aim 1 2 3 4 5 6 7
… … … … … … … …
Colgate – Aqua-Fresh 1 2 3 4 5 6 7

4
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.4. Multidimensional Scale (MDS)

Entry data acquisition

Perception data: direct focus through application of scales

▪ The number of pairs to be evaluated is n(n-1)/2, where n is a number of

stimuli.
▪ There are also other procedures. The interviewees may be asked to order
by ranges all the possible pairs from the most to the least similar.
▪ In a different procedure the interviewees may be asked to order by ranges
(preferences) the offered brands in terms of similarity with a brand
“anchor”.

5
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.4. Multidimensional Scale (MDS)

Entry data acquisition

Perception data: direct focus - EXAMPLE

Points of similarity by pairs of toothpaste brands

6
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.4. Multidimensional Scale (MDS)
Entry data acquisition

Perception data: derived focus

These are the procedures based on the attributes, the interviewees are asked to use
Likert scale or the scale of semantic differential, to qualify the stimuli in the
identified attributes. E.g., the different toothpaste brands are qualified according to
the following attributes:

Whitens teeth Doesn’t whiten teeth

Prevents dental Doesn’t prevent
caries dental caries
*** ***
Pleasant flavour Unpleasant flavour

7
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.4. Multidimensional Scale (MDS)

Entry data acquisition

Perception data: direct or derived focus?

DIRECT Focus DERIVED Focus

• The researcher DOESN’T have to • Easy to identify the interviewees
identify specific attributes. with homogeneous perceptions.
• The interviews use THEIR own criteria • Helps to interpret the dimensions of
to do similarity assessment. the perceptual map.
• The criteria are influenced by brands or • the interviewees can be grouped
evaluated stimuli. according to their qualification of
• Difficult to determine before the attributes.
analysis of the individual assessments • Easy to allocate a label to the
should be combined and how.
dimensions.
• Difficult to assign a label to the
dimensions of the spatial map.

8
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.4. Multidimensional Scale (MDS)

Spatial map of toothpaste brands

Malhotra, “Market Research”

9
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.4. Multidimensional Scale (MDS)

Spatial map of toothpaste brands

➢ use of VECTORS of the attributes to name the dimensions

Malhotra, “Market Research”

10
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.4. Multidimensional Scale (MDS)

Spatial map of toothpaste brands

Evaluation of the stability

after removing one brand

Malhotra, “Market Research” 11

Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.4. Multidimensional Scale (MDS)

Application in Marketing:
• Analysis of the competition in the market
• Perceptions of the customers
Sport
• Preferences of the customers (high)

Spatial configuration
conjointly of the car Prestigious,
Economic
high cost
brands and preferences
of the consumers

(low)

Malhotra, “Market Research” 12

Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.4. Multidimensional Scale (MDS)

Application in Marketing:
Preference data consist of the ordered ranges of the brands according to the
preferences of the consumers.
The ideal points of the interviewees are allocated on the same spatial graph
representation. Each ideal point represents the reference locus of each individual.
This way, the interviewee 1 (labeled with I1) prefers sport cars: Porsche, Jaguar and
Audi; while the interviewee 2 (labelled with I2) prefers luxurious cars: Continental,
Mercedes, Lexus and Cadillac. Sport
(high)

Prestigious,
Economic
high cost

Malhotra, “Market Research”

(low)
13
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.5. Correspondence Analysis

▪ A technique of multidimensional scaling to scale qualitative data in market

research.
▪ Entrance data are presented as a contingence table that indicates a
qualitative association between rows and columns. Correspondence
analysis scales the rows and columns in corresponding units, so that each
of them can be exposed graphically in the same space of few dimensions.
▪ These spatial maps provide information about:
1. Similarities and differences within the rows regarding the category of one
certain column;
2. Similarities and differences within the categories of the columns regarding
a certain category of the rows;
3. Relation between rows and columns.

14
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.5. Correspondence Analysis

Represents simultaneously, in a multidimensional configuration, a

contingency table.
Discovers affinities between two sets of variables, in frequency or
mean values.
Phases:
1. Decomposition of the Chi-square of the contingency table.
2. Computing of the matrix of covariance variable-column.
3. Factorization on the basis of PCA.
4. Computing of the coordinates of variables row and column.

15
Marketing research. Topic 7: Multivariable Analysis. Part 1
 7.5. Correspondence Analysis

Example

We have 100 followers of the following Football Clubs: Barcelona, Real

Madrid, Atlético de Madrid, Athletic Club Bilbao and Valencia CF.

We ask them to choose among these 5 football clubs according to their

preference. The first one will be, of course, which they follow, the
second one – the nicest of the most friendly and the fifth one – the
most unpleasant. This will allow us analyse the relative position of the
followers of these teams.

16
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.5. Correspondence Analysis
Example
The Football Clubs: Barcelona, Real Madrid, Atlético de Madrid, Athletic Club
Bilbao and Valencia CF.
Preference: the first one will be which they follow, the second one – the nicest
of the most friendly and the fifth one – the most unpleasant.
The goal: to analyse the relative position of the followers of these teams.

Followers Barça R. Madrid At. Madrid At. Bilbao Valencia Active margin

Barça 100 0 30 60 10 200

Real Madrid 0 100 0 20 80 200
Athl. Madrid 50 0 100 40 10 200
At. Bilbao 70 5 5 100 20 200
Valencia 20 20 20 40 100 200
Active margin 240 125 155 260 220 1.000

17
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.5. Correspondence Analysis
Example
Results: The 1st dimensions, the first axis (we can say the main component, in
terms of PCA) explains at 66.3% of the inertia (= variance).
The 2nd dimension explains at 22% → the two dimensions together explain at
88.3%.

Proportion of inertia Confidence

Dimensi Eigenval Chi-
Inertia Signif. Typical Correlati
on ue square Explained Accumul.
deviation on (2)
1 .760 .577 v
.663 .663 .017 .152
2 .437 .191 v
.220 .883 .033
3 .287 .082 .095 .978
4 .139 .019 .022 1.000
Total .870 870.289 .000a 1.000 1.000
a: 16 degrees of freedom
18
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.5. Correspondence Analysis
Example
The analysis always gives the coordinates of the points of the ROW:

Dimension, points Contribution

From points to inertia From the dimension to inertia
Followers Mass Inertia
1 2 of the dimension of the point
1 2 1 2 Total
Barça .200 -.725 .402 .109 .138 .074 .734 .130 .864
Real
.200 1.483 -.121 .353 .579 .007 .946 .004 .950
Madrid
At. Madrid .200 -.810 -1.099 .208 .173 .553 .480 .509 .989
At. Bilbao .200 -.430 -.430 .105 .049 .364 .267 .661 .928
Valencia .200 .483 .483 .095 .061 .002 .371 .005 .376
Total 1.000 .870 1.000 1.000

19
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.5. Correspondence Analysis
Example
… and the coordinates of the points of the COLUMN:

Dimension, points Contribution

From points to inertia From the dimension to inertia
Followers Mass Inertia
1 2 of the dimension of the point
1 2 1 2 Total
Barça .240 -.732 .440 .131 .169 .106 .747 .156 .902
Real
.125 1.640 -.167 .292 .443 .008 .875 .005 .880
Madrid
At. Madrid .155 -.809 -1.400 .210 .133 .695 .366 .632 .998
At. Bilbao .260 -.354 .563 .071 .043 .188 .350 .509 .859
Valencia .220 .855 -.064 .166 .212 .002 .735 .002 .737
Total 1.000 .870 1.000 1.000

20
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.5. Correspondence Analysis
Example
Points of the rows and columns Preferences
2 Followers

1 At.Bilbao
At.Bilbao
Barça
Dimension 2

Barça Valencia Real Madrid

0
Valencia Real Madrid

-1 At. Madrid
At. Madrid

-2
-2 -1 0 1 2 Dimension 1
21
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.5. Correspondence Analysis

▪ Due to the similarity of algorithms, interpretation of the results

of the correspondence analysis is similar to the principal
components analysis.
▪ the correspondence analysis provides a result of grouping of
categories (activities, brands or other stimuli) allocated on the
contingency table, the same way as the PCA implies grouping of
variables.
▪ The results are interpreted in terms of proximities between
rows and columns in the contingency table. The closest
categories are more similar in the underlying structure.

22
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.5. Correspondence Analysis

ADVANTAGES DISADVANTAGES
In comparison with other techniques of The distances between sets
multidimensional scaling, the Correspondence (between the row and the
Analysis reduces the requirements of data
collection implied on the respondents, since
column) cannot be
only binary or categorical data are obtained. interpreted in a significate
way.
The respondents must only revise what
attributes are applied to each of the brands. The The Correspondence
input data are the number of affirmative Analysis is a technique of
answers for each brand in each attribute. exploratory data analysis
The brands and attributes are exposed later in which is not adequate for
the same multidimensional space. hypothesis test.

23
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.6. Conjoint Analysis

▪ It tries to determine a relative importance of that the consumers assign

to the outstanding attributes and the utilities attributed to the levels of
attributes.
▪ The information is derived from evaluations made by the brand
consumers, or brand profiles formed by these attributes and their
levels. The respondents are presented stimuli that consist in
combinations of levels of attributes. They are asked to evaluate the
stimuli in terms of suitability.
▪ Conjoint proceedings try to assign values to the levels of each attribute,
so that the resulting values or the utilities attributed to the stimuli
match as much as possible with input evaluations provided by
respondents.
▪ The basis assumption is that any set of stimuli, as products, brands or
shops, are evaluated as a set of attributes.
24
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.6. Conjoint Analysis

▪ The conjoint analysis depends on the subjective evaluations of the

respondents.
▪ The stimuli are combinations of the levels of attributes determined
by the researcher (in MDA the stimuli are products or brands).
▪ The conjoint analysis looks for development of functions of partial
value or utility the respondents attribute to the levels of each
attribute. (the goal of MDA is to develop a spatial map that would
describe the stimuli in a multidimensional space of perception or
preference.)
▪ Both techniques are complementary.

25
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.6. Conjoint Analysis
Application in Marketing

Determine the relative importance of the attributes in the election

process of the consumer

Calculate the market share of the brands that differ in levels of

attributes

Determine the composition of the preferred brand

Segment the market on the basis of similarities of preferences by the

levels of attributes

26
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis

▪ Also called Classification Analysis or Conglomerates. A technique used to

classify objects or cases in relatively homogeneous groups called
conglomerates.
▪ Ideal to identify groups the most similar among each other with the
members of other groups. The objects of each conglomerate tend to be
similar and different from the objects of other conglomerates.
▪ Like in a factorial analysis, Cluster analysis examines a set of
interdependent relations.
▪ It doesn’t distinguish between dependent and in dependent variables, but
examines interdependent relations within a complete set of variables.
▪ It doesn’t reduce the number of variables but of the objects, which are
grouped in a reduced number of conglomerates.

27
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis

An IDEAL situation of conglomeration

A REAL situation of
conglomeration

In cluster analysis there is NO information a

priori about belonging to any group of the
objects. The clusters are suggested by data.

28
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis
APPLICATION

Market segmentation: possibility to group customers according to the benefits they look
for in the purchase of a product.

Comprehend customers’ behaviour: identify homogeneous groups of customers.

Identify opportunities for new products: by clustering brands or products, it’s possible to
determine competitive sets within the market.

Choose test markets: by clustering cities/towns, it’s possible to choose comparable

markets to test various marketing strategies.

Reduce data: useful to develop subgroups of data that would be easier to handle than
individual observations.

29
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis

▪ There is a valuation of “n” objects, individuals or brands in “p”

characteristics.
▪ Distances or similarities between them are computed.

𝑋11 … 𝑋1𝑝 𝑑11 … 𝑑1𝑛

𝑋 … 𝑋2𝑝 𝑑 … 𝑑2𝑛
𝑋(𝑛,𝑝) = 21 => 𝐷(𝑟,𝑠) = 21
… …
𝑋𝑛1 … 𝑋𝑛𝑝 𝑑𝑛1 … 𝑑𝑛𝑛

30
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis

Phases

1. Selection of the variables

2. Election of the measure of distance between the

cases

3. Criteria for clustering

31
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis

Carrying out the analysis

1
• Plan the problem

2
• Select the measure of distance

3
• Choose clustering procedure

4
• Decide the number of clusters

5
• Interpret and describe the clusters

6
• Evaluate the validity of clusters

32
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis

1. Plan the problem: select the variables to include in the analysis

▪ The variables must be chosen on the basis of the previous research, theory
or consideration of the evaluated hypothesis.
▪ The set of chosen variables must describe the similarity among the objects
in relevant terms for the research problem.
• If the important variables are not included in the analysis, the
obtained results will be poor and biased.
• Inclusion of one or more irrelevant variables will distort clustering
solution.

33
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis

1. Plan the problem: select the variables to include in the analysis

Example:
Let’s consider clustering of customers on the basis of their activities towards
purchasing. From the previous research six variables of attitude were
identified. The customers were asked to express their level of agreement with
the following statements in a scale of 7 points (1 = totally disagree, 7 = totally
agree):
V1 Shopping is fun
V2 Shopping is bad for the budget
V3 When I go shopping I usually have a lunch outdoors
V4 When I go shopping I look for the best offers
V5 I’m not interested in shopping
V6 I can save much money if I compare prices
34
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis

2. Select the measure of distance

▪ The most employed measure is the square of Euclidean distance. It’s the
sum of squares of differences of all variables.

ID V1 V2
1º 4 3 𝒅 𝟏𝟐 = (𝟒 − 𝟕) 𝟐 +(𝟑 − 𝟕) 𝟐 = 𝟓
2º 7 7

▪ The formula for this distance between a point X (X1, X2, etc.) and a
point Y (Y1, Y2, etc.) is the following:

35
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis

2. Select the measure of distance

The Euclidean metric is that

which corresponds to everyday
experience and perceptions.
That is, the kind of 1, 2, and
3‐Dimensional linear metric
world where the distance
between any two points in
space corresponds to the
length of a straight line drawn
between them. The concept is
that of linear distance.

Source: pbarrett.net

36
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis

2. Select the measure of distance

We can also calculate the

Euclidean distance between
the two variables, given the
three person scores on each

Source: pbarrett.net

37
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis

2. Select the measure of distance

Ref Brand calories Sodium Alcohol price

1 Mahou 0,38 0,01 0,34 -0,46
2 Carlsberg 0,61 0,62 0,61 -0,46
3 Vall Damm 0,81 0,01 0,61 -0,11
4 Coronita 0,25 0,46 0,21 -0,39
5 San Miguel 0,12 0,62 -0,05 -0,46
Standardized data

38
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis

2. Select the measure of distance

1 2 3 4 5
1 0
2 0,49 0
3 0,38 0,53 0
4 0,24 0,32 0,75 0
5 0,59 0,67 1,40 0,11 0

Matrix of distances

𝒅 (𝒓,𝒔) = ෍(𝒙 𝒓𝒊 − 𝒙 𝒔𝒊 ) 𝟐
𝒊=𝟏
39
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis

2. Select the measure of distance

Other distance metrics:

▪ Manhattan distance: the sum of absolute differences of the values for each
variable.

▪ Chebychev distance: maximum absolute differences in the values of any

variable.

40
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis
3. Choose clustering procedure

Application of the technique of rising classification or agglomerative

with the criterion of “simple linkage or the nearest neighbour”.

1 2 3 4 5
1 0
2 0,49 0
3 0,38 0,53 0
4 0,24 0,32 0,75 0
5 0,59 0,67 1,40 0,11 0

The nearest cases

41
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis
3. Choose clustering procedure

Application of the technique of rising classification or agglomerative

with the criterion of “simple linkage or the nearest neighbour”.

1 2 3 4-5
1 0 Distances are not
2 0,49 0 modified
3 0,38 0,53 0
4-5 0,24 0,32 0,75 0

Minimal distance between each case

42
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis
3. Choose clustering procedure

Application of the technique of rising classification or agglomerative

with the criterion of “simple linkage or the nearest neighbour”.

1-4-5 2 3
1-4-5 0
2 0,32 0
3 0,38 0,53 0

1-2-4-5 3
1-2-4-5 0
3 0,38 0

43
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis
3. Choose clustering procedure

Non-hierarchical
Hierarchical Others
(k-means)

Agglomerative By division Two steps

Variance
Sequential Parallel Optimal
Linkage (Ward Centroids
threshold threshold division
technique)

Complete Average
Single linkage
linkage linkage

44
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis
3. Choose clustering procedure

Hierarchical Non-hierarchical

• Rising or agglomerative. • The researcher decides

• Groups cases forming previously the number of
clusters bigger and bigger, clusters. Algorithm of
until all the elements form Howard Harris.
one cluster. • First assign or determine a
• Algorithm of Johnson center of a cluster and them
group all the objects that
are inside of one value of a
predetermined threshold
regarding the center.

45
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis
3. Choose clustering procedure

Single linkage Complete linkage

Max.
distance

Min.
distance
Conglomerate 1 Conglomerate 2 Conglomerate 1 Conglomerate 2

Average linkage

Average
distance
Conglomerate 1 Conglomerate 2
46
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis

3. Choose clustering procedure

Ward technique - The goal is to minimize the

internal variance.
- means of all variables of
each cluster are computed.
- Then for each object
squared Euclidian distance
of the means of the cluster
are computed.

Centroid technique
A distance between two clusters is
the distance between their centroids
(the means of all variables).
Every time the objects are grouped, a
new centroid is computed.

47
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis
3. Choose clustering procedure
Dendrogram using Ward technique

Cases

Rescaled distance of the combined clusters 48

Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis
3. Choose clustering procedure

Dendrogram using Ward

technique

Cases

Source: https://docs.roguewave.com/

49
Market Research. Topic 7: Multivariable Analysis. Part 2. Rescaled distance of the combined clusters
 Sou rce: https://docs.roguewave.com/
7.7. Cluster Analysis
3. Choose clustering procedure
Dendrogram using Ward technique

S o u r c e : DOI:
10.1109/AVSS.2014.6918690

50
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis
3. Choose clustering procedure

▪ Selection of cluster procedure is interrelated with the election of a

distance metric.
▪ For example, the squares of Euclidian distances should be used
with the procedures of Ward and centroids.
▪ Many non-hierarchical techniques also use the squares of
Euclidian distances.
▪ In the procedure of two steps the Euclidian metric can be used
only hen all the variables are continuous.

51
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis
4. Decide the number of clusters

Theoretical, conceptual or practical considerations may generate

certain number of clusters.

In hierarchical procedures distances at which clusters are combined can be

used as criteria.

In non-hierarchical procedures the proportion between total variance inside

the group and the variance between groups can be graphed against the
number of clusters.

Relative sizes of clusters should be significate.

52
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis
5. Interpret and describe clusters

▪ Cluster 1 can have relatively high values in some variables, for example, V1
(go shopping is fun) and V3 (when I go shopping I have lunch outdoors). But
it can also have a low value, like in the variable V5 (I’m not interested in go
shopping). So, this cluster can be labelled as “friendly and interested
shoppers”. (the case of 20 variables used for practice in class)

Centroids of clusters
Metrics of the variables
Cluster Nº

53
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis
5. Interpret and describe clusters
▪ Cluster 2 is just the opposite: with low values V1 and V3, and high value in V5,
so, this cluster can be labelled as “apathetic shoppers”. (Cases 2, 5, 9, 11, 13
and 20.)
▪ Cluster 3 has high values in V2 (shopping unbalances my budget), V4 (I try to
find the best offers when I go shopping) and V6 (one can save much money if
you compare prices) → “thrifty shoppers”. (Cases 4, 10, 14, 16, 18 and 19.)

Centroids of clusters
Metrics of the variables
Cluster Nº

54
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis

5. Interpret and describe clusters

▪ It’s useful to describe clusters in terms of variables that were NOT used by
grouping. This includes demographic and psychographic variables and
those of the use of the product or use of the communication means, etc.
▪ For example, the clusters may be derived on the basis of the searched
profits. It’s possible to provide a more detailed description in terms of the
demographic and psychographic variables to address marketing efforts
towards each cluster.
▪ Discriminant analysis and analysis of variance of one factor allow to
distinguish clearly among the clusters.

55
Market Research. Topic 7: Multivariable Analysis. Part 2.
 7.7. Cluster Analysis
6. Evaluate the validity of clusters

• Carry out cluster analysis with the same data but with different distance metrics and
1 compare the results to determine stability of solutions.

• Use different clustering procedures and compare the results.

2

• Divide data in halves and run cluster analysis separately.

3 • Compare the centroids of the clusters of the subsamples.

• Eliminate variables (randomly). run cluster analysis on the reduced set of

4 variables.

• In non-hierarchic cluster the solution may depend on the sequence of the cases
5 in the set of data.

Malhotra, “Marketing research”

56
Market Research. Topic 7: Multivariable Analysis. Part 2.
With this presentation we finish the
Topic 7.

Thank you very much for your

attention & effort!

57