BUSINESS CASE

MARKET SEGMENTATION – PRODUCT SERVICE MANAGEMENT

SHYAM KISHORE TRIPATHI

PGP- BABI
NOIDA, UTTAR PRADESH

06/07/2019

1|Page
 Contents
Market Segmentation – Product Service Management....................................................................1

Project Introduction: ............................................................................................................................3

Explanatory Analysis – Managerial Report: .....................................................................................3

1.Evidence of Multicollinearity: ...............................................................................................3

2.Perform Factor Analysis by Using Four Factors.................................................................8
3.Name of the Factor................................................................................................................13
4.Perform Multiple Linear Regression with Customer Satisfaction as the dependent
variable and the four factors as the independent variables. Comment on Model
Validity.......................................................................................................................................14

2|Page
Project Introduction:

The data file Factor-Hair.csv contains 12 variables used for Market Segmentation in the
context of Product Service Management.

Variable Expansion
ProdQual Product Quality
Ecom E-commerce
TechSup Technical Support
CompRes Complaint Resolution
Advertising Advertising
Prodline Product Line
SalesFlmage Salesforce Image
Com Pricing Competitive Pricing
WartyClaim Warranty & Claims
OrdBilling Order & Billing
DelSpeed Delivery Speed
Satisfaction Customer Satisfaction

Managerial Report required for the below scenarios

1.Is there evidence of Multicollinearity?

2.Perform Factor Analysis by extracting four factors
3.Name the factors
4.Perform Multiple Linear Regression with Customer Satisfaction as the dependent variable
and the four factors as the independent variables. Comment on Model Validity

3|Page
Explanatory Analysis – Managerial Report:

1. Evidence of Multicollinearity: To find the evidence of multicollinearity in the Factor-

Hair.CSV data we need to whether any high or moderate correlation exist between
independent variables. For this evidence, we need to test the data by executing below
codes in R.

Loading file in R
hair = read.csv ("hair.csv”, header = T)

Checking dimension of data set

dim(hair)
[1] 100 13

Checking all column titles assigned to data set.

names(hair)

[1] "ID" "ProdQual" "Ecom" "TechSup" "CompRes" "Advertising”

“ProdLine" "SalesFImage" "ComPricing"

[10] "WartyClaim" "OrdBilling" "DelSpeed" "Satisfaction"

Checking structure of given data set

str(hair)

Checking summary of given data set

summary(hair)

From summary and structure, we can notice that the given data is scaled already
and there is no need to scale them.

4|Page
Assigning New column names in a matrix to use in the dataset
variables= c ("Product Quality" , "E-Commerce" , "Technical Support" , "Complaint
Resolution" ,"Advertising”, "Product Line" , "Salesforce Image", "Competitive
Pricing" ,"Warranty & Claims”, "Order & Billing" , "Delivery Speed" , "Customer
Satisfaction")

First column named "ID" is irrelevant for further analysis.

So create new data frame named hair1 with all data except the column "ID"
hairnew = hair[,-1]

Change names of variables

colnames(hairnew) = variables

Attach Data
attach(hairnew)

names(hairnew)

Checking for any missing data

sum(is.na(hairnew))

Checking for any outlier in given dataset

par(mfrow = c(2,1))
boxplot(hairnew[,-12], las = 2, names = variables[-12], cex.axis = 1)

5|Page
Finding co-relation between given data
cormtx=cor(hairnew[,-12])
cormtx

Plotting correlation plot

library(corrplot)
corrplot.mixed(cormtx,lower = "number", upper = "number",tl.col = "black",tl.pos =
"lt")

6|Page
From correlation matrix we can clearly identify that independent variables also
have correlation among themselves.

Here we have Customer satisfaction as Dependent variable which is correlated with

"Complaint Resolution" which is further correlated with"Order Billing" and
"Delivery Speed".

Hence it clearly indicates that there exists multicollinearity and factor analysis will
be applicable.

To assess multicollinearity, we will compute VIF score. If VIF score is greater than 4
for more than two or more variables, then it clearly indicate that multicollinearity
exists in independent variable.

library(car)
hairvif = vif(lm(`Customer Satisfaction` ~., data = hairnew))
hairvif

From above it is a clear indication of existence of multicollinearity as VIF value is >

4.

Check Correlation Matrix using Bartlett Test

library(psych)
cortest.bartlett(cormtx,100)

Since P value is 1.79337e-96 which is less than 0.05 so it is ideal case for dimension
reduction.

7|Page
 Kaiser-Meyer-Olkin (KMO) Test is a measure of how suited your data is for
Factor Analysis. The test measures sampling adequacy for each variable in the
model and for the complete model.

KMO(cormtx)

Since MSA value which is 0.65 and is greater than 0.5 then it clearly indicates that
we have enough samples to proceed further for Factor Analysis.

2. Perform Factor Analysis by extracting four factors

In order to perform factor analysis, we will proceed further to calculate Eigen Value
and draw a Scree plot to get the number of factors.

Calculation Eigen values for the variables to check that how the data is spread
across

evector =eigen(cormtx)
EV = evector$values
EV

[1] 3.42697133 2.55089671 1.69097648 1.08655606 0.60942409

0.55188378 0.40151815 0.24695154 0.20355327 0.13284158
0.09842702

We can clearly see that few of the Eigen Values are higher than the other Eigen
Values so this is a clear case of Dimension reduction.

8|Page
Plot scree plot of Eigen Values on a graph and check that which are all the major
influencing variables.
plot(EV, main = "Scree Plot", xlab = "Factors", ylab = "Eigen Values", pch = 20, col =
"black")
lines(EV, col = "red")
abline(h = 1, col = "black", lty = 2)

Applying Elbow Rule or Kaiser's value rule to analyze factors.

We will consider values above 1 so accordingly there are 4 factors only which are as
follows.

Factor 1 as After Sales Service which has Complaint Resolution, Order Billing and
Delivery Speed

Factor 2 as Marketing whihc has E-Commerce, Advertising and Sales F Image

Factor 3 as Customer Support which has Technical Suppot and Warrenty Claim

Factor 4 as Product which has Production Quality, Production Line and Competitive
Pricing

9|Page
Extracting PCA/FA

PCA without rotation

hair_unrotate = fa(r= hairnew[,-12], nfactors =4, rotate ="none", fm ="pa")
print(hair_unrotate)
fa.diagram(hair_unrotate)

10 | P a g e
PCA with rotation
hair_rotate = fa(hairnew[,-12], nfactors =4, rotate ="varimax")
print(hair_rotate)
fa.diagram(hair_rotate)

Competitive Pricing here is in negative which indicates that it will affect adversely
with respect to factor “MR4”.

After doing rotation we can see that there is a significant change in values so we will
consider rotation

Create a Data Structure using score values of four factors and target values
and bind the dependent variables together to proceed with supervised
learning.
hairnew2 = cbind(hairnew[,12],hair_rotate$scores)

11 | P a g e
Checking and renaming new dataset header name
head(hairnew2)

colnames(hairnew2) = c("Customer.Satisfaction", "Sales.Distribution",

“Marketing","After.Sales.Service","Value.For.Money")

head(hairnew2)

Checking class of new dataset

class(hairnew2)

Convert new data set from matrix to data.frame

hairnew2 = as.data.frame(hairnew2)
hairnew2

Plotting correlation plot for New dataset (HairNew2)

corrplot.mixed(cor(hairnew2),lower = "number", upper = "number",tl.col =
"black",tl.pos = "lt")

12 | P a g e
3. Name the factors

Factor 1 as After Sales Service which has Complaint Resolution, Order Billing and
Delivery Speed

Factor 2 as Marketing whihc has E-Commerce, Advertising and Sales F Image

Factor 3 as Customer Support which has Technical Suppot and Warrenty Claim

Factor 4 as Product which has Production Quality, Production Line and Competitive
Pricing

4. Perform Multiple Linear Regression with Customer Satisfaction as the dependent variable
and the four factors as the independent variables. Comment on Model Validity.

install.packages("caTools")
library(caTools)

Set flag for randomness

set.seed(1)

Creating two datasets one to train the model and another to test the model based on
standard split ratio of 70:30.
spl = sample.split(hairnew2$Customer.Satisf, SplitRatio = 0.70)
Train = subset(hairnew2, spl==T)
Test = subset(hairnew2, spl==F)

Check dimensions of Train and Test Data

cat (" Train Dimention: ", dim (Train))
Train Dimention: 71 5

Test Dimention : 29 5
cat (“Test Dimention : ", dim(Test))

13 | P a g e
Creating New Linear model using Train Data
newlm = lm(Customer.Satisfaction ~., data = Train)
summary(newlm)

We found that R-squared here is 0.7294

vif(newlm)

If VIF score is greater than 4 for more than two or more variables, then it clearly
indicates that multicollinearity exists in independent variable. Here we have VIF<4
so it clearly indicates that there is no multicollinearity among the independent
variable.

Predicting new linear model using Test data

pred = predict(newlm, newdata = Test)
pred

14 | P a g e
Compute R-sq. for the test data
SST = sum((Test$Customer.Satisfaction - mean(Train$Customer.Satisfaction))^2)
SSE = sum((pred - Test$Customer.Satisfaction)^2)
SSR = sum((pred - mean(Train$Customer.Satisfaction))^2)
Rsqtest = SSR/SST

cat(" SST :", SST)

SST : 36.27184

cat("SSE :", SSE)

SSE : 14.10709

cat("SSR :", SSR)

SSR : 29.47223

cat("R squared Test :" , Rsqtest)

R Squared Value: 0.8125375 which indicates that this is a good model.

Additionally, we can also do the backtracking also to verify our results.

This is not required but it will be an added check.

model1=lm(hairnew$`Customer
Satisfaction`~ProdQual+Ecom+TechSup+CompRes+Advertising+ProdLine+SalesFImage
+ComPricing+WartyClaim+OrdBilling+DelSpeed, data = hair)
summary(model1)

model2=lm(hairnew$`Customer Satisfaction`~hair_rotate$scores,data = hairnew)

summary(model2)

Predict_Model=predict(model2)
Predict_Model1=predict(model1)
plot(Predict_Model,col="Red")
lines(Predict_Model,col="Red")

plot(Predict_Model1,col="Blue")
lines(Predict_Model1,col="Blue")
lines(Predict_Model,col="Red")

15 | P a g e
From back tracking also we can see that New Model (Individual Factors) and Old model
(factors from 4 new factors) both follows the same path.

16 | P a g e