Welcome everyone to the class of marketing
research and analysis till now we have discussed about a several tools and techniques
used in the you know in the marketing space how companies utilize them and what do they
gain out of it how they make advantage out of those by using this techniques we have
discuss some of them one of this which we had recently discussed was a technique which was
an interdependence technique in which we try to bring in large number of variables to a few
ones and create factors out of it which we said as factor analysis right. So in that we had you
know reduce large number of variables may be 100 to a few meaningful ones which I always repeat the
word meaningful right once and then give a name to those factors and then try to understand how this
factors are going to determine or effect may be some other you know prediction some other kind
of a you know relationship. Today we are going to discuss similarly another technique which is
equally important right very much utilized in the market basically by marketers right. Let us say
there is a case a company wants to sell let us say some you know some kind of phones right now
a days a lot of phones are coming mobile phones so it wants to sell mobile phones and it has
got the target of about let us say 100000 okay 100000 pieces of you know phones to be sold.
Now the company want to know how should I sell them right so while selling he wants to target the
customers so while targeting the customers he has got certain variables like age of the people let
us say income of the people kind of a you know habit of a people some kind of habit let us say
so now on basis of these variables the company wants to divide a particular may be state or place
because suppose this is a whole place let us say and now they want to say are we going to target
the whole state or no if we because if we are going to target the whole state for 1 lakh pieces
of mobile it might be little difficult right. So is there a way that I can you know divide this
state in to several several groups or clusters then what it does is he tries to divide the
state not a geographically may be on certain other parameters taking these three may be on
some clusters may be right some clusters. So now it says out of this clusters they find all these
clusters have got different characteristics okay. so it finds that may be once I break this
into four clusters, cluster two is the one which is very promising it seems to the company
that it is a very promising you know cluster, why? Because the people in this cluster are once
who are would be interested to buy such kind of mobile phones which has all the characteristics
that this companies phone has got right. So in such a condition this becomes highly useful for
the company to understand how we could easily segment and then target the market so that we
can achieve better results right. So to do this we are using the technique which is called cluster
analysis, so cluster analysis as understand is to divide or you know create cluster out of a large
pool of respondents right we say respondents cases whatever you can say. So these respondents are
grouped into several similar clusters right so that each cluster has got people inside or you
know who are highly similar or highly homogenous in nature right. So what we understand is the
all the respondents within the clusters within the clusters let us say these are the respondents
all the people or all the respondent what it say there highly heterogeneous sorry homogenous in
nature that means their behaviors is more or less highly similar they are same right but
we understand the two clusters two clusters let us say cluster one and cluster four cluster
two and cluster three they all different from each other right so this is the basic understanding
of clusters analysis that within the cluster there is very minimal difference or distance
and two clusters between the clusters.
Let us say the distance is high okay so let us
see how it functions what happens in a cluster analysis so as I started so what I cluster how
to define it says it is a group of multivariate technique who is primary purpose is to group
objects right example respondents it could be products it could be entities whatever right
so you are trying to group these respondents or products are something into several clusters.
Several groups which have a similar nature right the other thing if you remember in factor analysis
we said there were trying to group the variables right so that was variables here respondents
or products or whatever other entities right, so what is saying it is a means of grouping
records based up on certain attributes.
Now what are the attributes on which one it is
grouping now the attributes in this case are for example age, income level, let s say habit.
let us say habit means spend let us say spending habit right or let us say may be technologically
savvy how technologically tech savvy are people so maybe that is an indicator of measuring right, so
you can have certain criteria certain attributes upon which you can create those similarity
right so and each similar group is a cluster right so if plotted geometrical is objects
within the clusters will be close together so that is what I was saying if you plot it then
the objects with the clusters are very close to each other that is why it is said they are
homogeneous in nature and two clusters are the variables the respondents within two different
clusters are highly heterogeneous in nature right. Then we have said more highly different
right this is what cluster basically tells you so cluster variate represents a mathematical
representation of the selected set of variables which compares the objectives object similarities
now what it is saying so mathematically you are trying to plot may be a plot or try to find out
some similarity so that it becomes easier for a marketer to understand okay which are the
clusters he or she should cater to right. It becomes very simple they say take so many example
companies of Maruti is coming up with a new car now should Maruti target everybody or should
Maruti have a specific thing specific policy in mid how to target whom to target so when it
does it has to take certain attributes on this basically attributes they finally decides okay.
So clusters versus factor analysis clustering is based on the distance matrix and factor
we said was in a correlation matrix, so if you remember in factor analysis you
said that the most important thing was.
We were trying to find out a correlation right so
we were trying to find a correlation right so when we are trying to find a correlation and this
correlation was telling us how close were the variables to each other but in cluster analysis
we do not take the correlation rather we take the distance what we take we take the distance so
the grouping is based on the distance how far are the variables from each other or how close are
the variables from each other the you know from each other. So as it says factor analysis we form
groups of variables based on the several peoples responses to those variables in contrast cluster
analysis we group people based on the responses to several variables now how people have responded
to several variables what score they have given so that tells basically their behavior that
their thinking pattern and all so by taking those variables their thinking pattern in all
these things you can classify these respondents into several groups right. Which as I said is
basically a nothing but a mental distance or a distance that is measured right so this distance
is basically mental distance sometimes we feel something are very close to us but they are
actually not where as some other things some other places might be felt that very far of but
they are actually not so far this is actually nothing but a mental perception a mental distance
right. We feel that it is very far of but actually that might be closer right so that is what
happens so where the applications are some application are other mentioned.
Field of psychiatry, biology. right information retrieval for example the world
wide web contains billions of web pages so when you type maybe for example market research
right, so all those pages which are linked to research or marketing research they will be
clubbed they will brought together right so this is nothing but they are clustering in one
way okay the climate for example understanding the earth s climate requires finding pattern in
the atmosphere and the ocean to that and cluster analysis have apply to fact fine patterns in the
atmospheric pressure of polar regions and areas of the ocean that are the significant impact so
how do you say that some places are very similar climatic conditions because this is again where
we use cluster analysis to divide the you know as per the all the distance by that two places might
be far of but the climatic at the type of climate the weather all this things are very similar
in two different places, so they can be still clubbed into one right so understanding these
things is very from more applications I can show you market segmentation grouping people.
Right so grouping people and with the willingness purchasing power the authority to buy according
to the similarity in several dimensions states segmentation can tell you what type of customers
buy what products, so customer a would buy what kind of a product that is what a helps you know
there is how the clusters is helps the marketer right so some examples are always there many
more examples city planning, insurance right geographical you know examples so all these are
basically are used to group respondents. According to certain behavior attributes and find those
clusters right once you find those clusters it becomes easy for you to polishing may for making
policy making for selling something for you know maybe understanding the kind of trend or any trait
and it behavioral trait maybe some genetics study you are wanted to do, so everywhere in fact to
tell you the use of clusters analysis was not started did not start with marketing or something
right it has it is roots here actually in taxonomy in biology. Where different kinds of spaces where
to be classified into different show in groups right so this is how it is all has started
So common role the cluster analysis can play right first is data reduction so as factor
analysis also was helpful in data reduction similarly cluster analysis also helps you in data
reduction researcher maybe face it larger number of observations that can be meaningless right
unless classify to a managerial groups so how many groups suppose you have 10,000 respondents 1 lakh
respondents but 1 lakh respondents individually try if you understand this nothing you can do so
if you either create 10 clusters out of it. Then it makes more much of better is meaning out
of it right second is hypothesis generation cluster analysis also useful when a researcher
wishes to develop hypothesis concern the nature of the data or to examine previously stated
hypothesis so cluster analysis is also useful in developing a hypothesis right to it because
it has it gives you insight the knowledge about certain things so it helps you to develop a
hypothesis right and even finally test it.
So what are the objectives so taxonomy description
so identifying groups data simplification the ability to analyze groups on similar you
know observations instead all individual which I said is very, very taxing and difficult
impossible sometimes then finally his relationship identification where it says the simplified
structure from cluster portraits relationships not revealed otherwise sometimes we can find out okay
let say now I have said there are four clusters right so are this four clusters okay.
Fine there is no doubt that this four clusters are different but this is a possibility that cluster
1 and cluster 3 are actually very closed to each other sometimes it happens in your state where you
stay that maybe one distinct the state has got 10 districts out of which two districts are extremely
similar because of their maybe a language there food habits are something or cultural habits
so they are very similar so they can sometimes if you want you can even club them and make it
like one cluster right. You can make it so this relationship identification is there important
job of cluster analysis how does work.
So three basic things are there, now what
are these three basic things so it says the primary objective of cluster analysis is
to define the structure of the data right, so by placing the most similar observation into
groups remember. If you are working with data and trying to find out similarity so that you can
form groups then there must, there could be some problems with it also. Suppose the biggest
problem that can affect cluster analysis is suppose you have a data which is got there is
few outliers in the data. Now if you have few outliers in the data then that could completely
change the you know the way the groups are formed, that could be a very important thing one should
keep in mind okay. Three things that are very important what, how do we measure the similarity
so when you are doing the cluster analysis the question comes how do I know key which groups
are similar there are four clusters.
Now how do I know which are similar to each other,
one second thing now do I form the clusters, how do we form the clusters it is not that
key the data is given to us and we just do it, there must be a way right, so how do I form
the clusters. Third, how many so the question is how do I form clusters, how do I form.
How do I form, right, how many do I form, right how many, how do I form, how many do I
form and how do I measure a similarity, how do I measure the similarity so if I can measure the
similarity then only I can do this right, so let us see but one thing is you have to keep in mind
that in the cluster analysis we are not looking at the correlation and why we are not looking at the
correlation I will explain you, see correlation could is like something like this you know suppose
two things are moving like this. The correlation might be very high among them but suppose this
two verses let us say take this you know just for understanding, understand this way. suppose
this two lines and these two lines if you look the correlation between the two both the lines pairs
let us A, B,C,D, A,B and C and D suppose you these two are one group, these two are one group is more
or less same the correlations more or less same.
But if you look at the distance is
actually not same so that is what is the basic underlying difference between cluster
and the distance and the correlation, okay.
So what is similarity represents the degree
of correspondents among objects across all the characteristics used in the analysis,
so I have several attributes as I have used let us say I have ask several variables, there
are several variables which I have used in the study some of them being like income, age and all
these things. Now taking them together when I am trying to build a kind of a similarity matrix
right, so this all this variables together will help me not one but together they will help
me in defining a cluster, right, okay.
So as I said correlation basis are less
frequently used only in case there is a special case where correlation is used where
you know when there is something called this is called a molecular distance which is used when
the certain variables do have a correlation we use it in special case I will come to that
right, otherwise most often the similarity is measured through the distance, okay.
Now this is what I was trying to explain you see if you look at it, if you look at the two charts,
the two graphs the graph 1 and graph 2 both the R=1, now what is R the correlation value so the
coefficient of correlation is 1 that means they are highly correlated right, so which implies
to have a same pattern right, but the distances are not equal, the distances between these and
the distances between these are not equal so these two will get a similar different you know
interpretation in cluster analysis but had not been enough case of a factor it would have been
it would be very similar, right So that is the, thus the basic differences okay, so graph 1
represents high level of similarity right, and graph 1 because the distance is less, why
it is saying now because if the distance is less that means they are close to each other
as good as that, right as that right if it is close the similar there will be coming trending
towards each other right if the distance is high or the correlation the trend is same but there
is a sufficient gap right so all this things are very important to understand this right.
So the distance now how do you measure the distance several ways the basic way of measuring
the distance is the Euclidean distance now Euclidean distance is the distance which we
say is the straight line right so how do we calculate so D is equal to let say (x2-x1)2
+ (y2-y1)2 right so this is what basically how we measure right so the most commonly
recognized 12 straight line distance right.
This is how you measure so the other forms
also for example square Euclidean distance which says that you take the sum of the squared
differences without taking the square root that means you only omit the square root right okay. So
the other distance is like the city block distance Manhattan distance which is not like the Euclidean
because this one takes the absolute value. And it is sometimes it is it does not work well because
of this nature of absolute value it does not work well right so one is to be very careful which you
are applying mostly if you do not understand much simply you can go for the Euclidean distance
because that is another safest way until the correlation among the variables right.
Then you have two more like Chebychev distance which is taking the maximum of the absolute
difference in the clustering variables value for example maximum of x1 x2-x1 or y2-x1 so the
absolute value is only taken right and the last which I am saying is the Mahalanobis distance
which measures accounts for the correlation among the variables in a way that each variables
equally. Now Mahalanobis distance is also very important tool which is used to also find
out, outliers let me tell you this is may be not here but suppose you are doing the
simple regression and you want to find out, outliers Mahalanobis distance is the technique
which is used to measure to find out those outliers okay so this is one way of doing it.
You remember only the condition applied is when the variables have a large correlation or high
correlation among them that time it is preferable to use a Mahalanobis distance over the other
distances and these things you will find almost every were in the software nowadays right so
you do not need to calculate it by hand okay.
Now this is how the Euclidean distance looks
like all the three one go it is showing you now this is hypothesis measure right.
So let us say this is an example now let s starts with the example now this example if you
can go through a market researcher a marketing researcher wishes to determine market segments
in a community based on patterns of loyalty to brands and stores to stores a small sample of 7
respondents is selected as a pilot two measures of loyalty store loyalty and brand loyalty V1 and V2
V1 and V2 right were measured for each respondents on a 0-10 scale right. The scores are given to
you so these are the 7 respondents A B C D E F G and the score that they are given for store
loyalty and brand loyalty is been given to you on the scale of 0-10 right now from this data let
us see how we can get into the cluster analysis okay so when I place the data on a graph
So 3 2 4 5 4 7 so if you can see this A is 3, 2 right then B is 4, 5 right C is 4,
7 D is 2, 7 so we can see here D is 2, 7 right so we have just paste them on the graph
and we place them on the graph right
Now how do we measures similarity the first
question that we have how do we measure the similarity so to do this right how do we
measure similarity I said we can use the Euclidean distance as a way of doing it now
how it has done now for example let us take the distances now let us ate two distances okay
so among between two variables now for example this was 3 2 4 5 4 7 2 7 okay. So let us take
only this much so this ADCD I am taking it. Now suppose if we want to find out the distance,
now how do we do it? Now do to do that? What is the ways simply, for between A and B.
so the distance between A and B = x2 x1 so 3- 4 right you can say 4- 3 also does not make
a difference obviously you will square right, (4 3)2 + (5 2)2 sorry either you take it this
way so(4 3)2 + (5 2)2 so what it is coming, so 1 + 9 = 10.That means it is coming something
around 3.162, similarly you can find the distance for all the other variables right, now the
distance between A and B is 3.162.
Obviously A and A would be same 1, now let us
calculate one more A and C , these two we are measuring, if you measure these two it will be 4
d, a, c (4- 3)2 + (7-2)2 so that is = 1+25 = v26 so that is = 5. Something right,
so it must be A and C IS 5.099, so for that everything you have measure right.
So after measuring the next question was how do we form the clusters? So you have measured the
distances, now find out the distance between two clusters, the minimum distance between the
two clusters, the minimum why I am saying this is the closest, they are very close to each
other. Had the distance been more they would have been far away from each other. So identify
the two most observations, who are already in the same cluster and combine them.
So what you can do is two similar clusters, 2 observation for example as I explained, this
is 1 cluster, so identify two values right and through certain ways, there are certain ways,
I will explain that also, similar linkage, average linkage, centroids method, there are
different ways how should I use right. So identify the two most similar observations, which
are already in the same cluster and combine them so that is what the objective.
So we want to form the clusters, so once we have identified this, so apply
this rule to generate the number of clusters, starting with each observation as it is own
clusters right and then combining two clusters at a time until all observations are in the single
clusters, that means what you are trying to add up the closest once, now this is one 1 and 2 are
close, so next is say 5, so 5 is closest to them, so you add up now 5 then let say 8. So you
add up 8 then let us say 6 you add up 6 so it goes on adding the nearest variable in terms of
the distance okay this method of this process is termed as hierarchical procedure why obviously
it will say hierarchical procedure because your maintaining a hierarchy right you are following
a hierarchy so when you do this ultimately all the different respondent will be clip together
to form a single cluster right single cluster. But the question is if we have a single cluster
then the whole meaning is lost right suppose I want to know to which state suppose the state
of let say or this country our country India suppose you want to do it in India then if I say
whole market is your market then it becomes very difficult for me to make an interpretation out
of it so in those cases you have to understand well what should I do this is not sufficient
for me how do I break it into several clusters first so how many clusters should India be broken
into for the marker so that it can easily scatter to those clusters so we will see that right the
process is termed hierarchical procedure
Because it moves in a step wise fashion to
form an entire range of cluster solutions right it is also agglomerative method because
clusters are formed by combining the existing clusters so what it is saying so you are
trying to form the clusters because let us say cluster one as I have drawn here.
Let us say there was a cluster one there was cluster two so cluster three so if I add up
1+2+3+4 then it becomes the whole place the big country India now the question is how
do I add up should I add up like anybody or everybody or should I have a mechanism now what
is the mechanism the mechanism is find out the two clusters which are very similar or close to
each other. And once you can find out those two clusters which are close to each other and you
start on adding and ultimately you will land up to the whole market that is India right so this is
why it is called a agglomerative method okay now let us look at this so the agglomerative process
is the process in which what we have done is if you look at it this is the minimum distance.
That you have to calculated by now we have done this through the distance now after doing
this what were the distances now distance we have arranged it so 1.414 to 2.236 now the
pairs of respondent which were related to it are the first one is E and F the second
one is EFG I hope you can recall that.
Let go back and look at it so the lowest value is
at E and F okay so E and F is 1.414 is the lowest so the next one is there are three next lowest 2
right in this line and then in this line again you have a 2 and again you have 2 so now you can take
the closest one for example in this case what we have done E and F for the first pair then second
pair we took E and G right because U was already there so the closest to E we found out and then
C and D and finally we took we started doing each right if you can see the cluster membership
here abcdefg so there are seven clusters.
So we have not done any grouping right now
after this what we did was we club E and F together here to form a single cluster so they
become 6 clusters then right then we did abcdef and g we added EFG together so 5 clusters then
we added ABCD together then EFG right so what we have done is basically we have shorten we have
reduced the number of clusters to a few so that we can finally land up into one right so by doing
this where I will explain you may be in the next session how do we calculate this part also right
this oval similarity measure the average within the cluster distance this also I will explain.
So we are identifying the number of cluster and through this we can say finally by through this
value this data how many clusters in this case we should have right so well this is what just
the introduction of the cluster analysis so we will in the next session we will get into
more details rights how one should form the clusters and how then one should interpret the
clusters okay thank you for this session.
