t

os
IMB 694

U DINESH KUMAR

rP
TEACHING NOTE

TESTING MARKETING HYPOTHESES AT WSES

CASE SUMMARY

yo
We Sell Everything in Software (WSES)1 Inc., sold innovative off-the-shelf products and had customers
across the world. WSES specialized in providing software solutions for different industries such as
defense, clinical research, consumer goods, capital markets, security, banks, retail, and insurance among
others. Although the products were commercial off-the-shelf, many clients required personalization and
after sales support which WSES was happy to provide.
op
WSES did not have a structured approach to taking decisions regarding chasing a sales lead. Sales team
used “gut feel” to pursue an opportunity. Therefore, it incurred high marketing cost including travel
costs, client visits, time spent by the sales team/technical experts/support staff and logistics costs; most
importantly, this list excluded the cost of advertising, which in effect meant the advertising costs were
over and above the ones mentioned. The marketing team had several beliefs about the chances of winning
a deal across different geographical locations, different domains etc. However, none of these beliefs have
been validated. Jack Williams, the CEO of WSES, was worried that despite having such a huge
tC

expenditure list, the sales conversion possibilities based on the pipeline were at best ancillary information,
as there was no substance in justifying ‘‘gut feeling’’. WSES engaged Liz Smith, a Ph.D. in Statistics, to
understand if they could use a structured approach to check whether their beliefs about marketing were
correct. Jack also believed that simple statistical analysis could help WSES with useful insights about
sales conversion.

LEARNING OBJECTIVES
No

The case can be used as a resource for teaching statistical concepts such as hypothesis tests including Z
test, t test, Z test for proportion, two sample t tests for mean, ANOVA, Pearson correlation, bi-serial
correlation, etc. The case uses the context of sales conversion and can be used to discuss how statistical
concepts can be used to derive insights in the sales and marketing context. The case can be used as part
of courses such as quantitative techniques, data science, business analytics and machine learning at
graduate level. The instructor may provide the accompanying spreadsheet (file name IMB695WSES.Xls)
to the students to work on the questions listed below:
Do

1
Names of all companies and individuals are changed to maintain the confidentiality

This Teaching note was prepared by U Dinesh Kumar, Professor of Decision Sciences and Information Systems, for the sole purpose of aiding
classroom instructors in the use of Testing Marketing Hypothesis at WSES, Case No. IMB 693. It provides analysis and questions that are
intended to present alternative approaches to deepening students’ comprehension of the business issues presented in the case and to energize
classroom discussion

Copyright © 2021 by the Indian Institute of Management, Bangalore. No part of the publication may be reproduced or transmitted in any form or
by any means – electronic, mechanical, photocopying, recording, or otherwise (including internet) – without the permission of Indian Institute of
Management Bangalore.
This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 2 of 20

t
os
SUGGESTED CASE QUESTIONS

The instructor may select a subset of questions form questions listed below depending on the session
objective. For all the tests listed below, assume that the significance value  = 0.05, unless otherwise

rP
stated.

1. Derive insights from the data using visualization.

2. The marketing team of WSES believes that the average sales value of the leads they receive is at least
8 million dollars. The marketing team believes that the standard deviation of sales value is 2 million
in the population. Use an appropriate hypothesis test to check whether the average sales value in the
population is at least 8 million dollars.
3. Change the null and alternative hypothesis used by you in Question 2 and conduct the test. What will

yo
be your conclusion based on the interchange of null and alternative hypotheses? What do you learn
from your answers to Questions 2 and 3?
4. If the actual average mean sales value of the leads is 8.2 million dollars, calculate the probability of
incorrectly concluding that the average sales value is less than 8 million dollars.
5. Jason McCullagh, senior marketing manager at WSES doubted the value of standard deviation
provided by the marketing team. Jason argued that there was no way the marketing team could have
known the population standard deviation for the sales value, since the population itself is unknown.
op
Do you agree with Jason? If yes, perform the test again using an appropriate hypothesis test.
6. Prudy Perkins, the Chief Marketing Officer (CMO) informed the board of WSES that they win at
least 50% of the sales leads they receive. Use an appropriate hypothesis testing procedure to check
whether the proportion of leads won by WSES is more than 50%.
7. Hendry Jackson, who works in the product line “learnsys”, claims that the probability of winning a
sales lead for the product “learnsys” is more than that of “Finsys”. Is there statistically significant
tC

evidence in favor of Hendry’s claim?

8. Hendry Jackson also claims that the average sales value of “learnsys” projects is higher than that of
“Finsys” projects. Check whether he is correct at 5% significance.
9. WSES decide to target customers who earn higher profit with a belief that it is easier to convert leads
from customers who make more profit. That is, they believe that the average profit of customers won
by WSES is more than the average profit of customers lost by WSES. Use an appropriate hypothesis
test to check whether this belief is correct.
No

10. Prudy Perkins, the chief marketing officer, advises her sales team that the sales value of a lead should
not be a criterion to be used before deciding to pursue the sales lead. Is there evidence to suggest
that the sales value has no influence on the ability of WSES to convert a sales lead?
11. Jack Williams, the CEO of the company believes that the sales conversions are different for different
products as well as different geographical locations. Check the validity of Jack’s belief using an
appropriate hypothesis test.
12. Joe Danby, the chief financial officer believes that the sales conversions depend on the sales value.
Use an appropriate hypothesis test to check the validity of this claim by making the following three
groups:
Do

1. Sales value less than 6 million

2. Sales value between 6 and 8 million (both inclusive)
3. More than 8 million

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 3 of 20

t
os
13. Liz was of the opinion that there was no difference in the average profit and the geographical
locations: United Kingdom, India and the Americas. Use an appropriate test to verify the same.
14. Use an appropriate statistical measure to understand the relationship between (1) Joint bid percentage
and profit; (2) Profit percentage and sales value: and (3) Joint bid and sales conversion.
15. Create two groups based on the prior classification of leads by the sales team based on “gut feel”. In
group 1, include all the leads that were classified as “excellent chance of winning”, by the sales team

rP
and in group 2 include all the leads that were classified as “low chance of winning”. Using an
appropriate hypothesis test, check whether the proportion of leads won in group 1 (leads classified as
excellent chance of winning) is higher than the proportions of leads won in group 2 (leads classified
as low chance of winning).
16. Discuss the business insights based on the data analysis.

RECOMMENDED RESPONSES TO QUESTIONS

yo
The students should have a clear understanding of different types of hypothesis tests, ANOVA and
correlation before the instructor uses this case. The case can be used to make students understand when
to use different hypothesis tests and how to interpret the hypothesis test results.

1. Derive insights from the data using visualization.

op
Response:

The instructor may start with this question. It will help the students understand the data visually. The
instructor may encourage students to come up with various visuals and insights based on the visuals. TN-
Figure 1 shows numbers of leads lost and won by WSES. The proportion of leads won and lost are very
close. The visuals in this teaching note are created using Tableau.
tC

TN-Figure 1
Number of Leads Won and Lost by WSES
No
Do

TN-Figure 2 shows the tree map of distribution of leads from various countries. The maximum of leads
has come from the UK (553), other European countries (158), America (104) and so on.

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 4 of 20

t
os
TN-Figure 2
Distribution (Count) of number of leads from various countries.

rP
yo
op
TN-Figure 3 Shows the average profit of the customer for won and lost leads. It is evident from TN-
Figure 3 that the average profit of the customer is lower for the won cases compared to lost cases.
tC

However, we must conduct a hypothesis test to check whether the difference in the average profit of the
customer between leads that are won and that are lost is statistically significant using an appropriate
hypothesis test.

TN-Figure 3
Average Profit of the customer between won and lost leads
No
Do

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 5 of 20

t
os
TN-Figure 4 shows the average sales value of lost and won leads. From TN-Figure 4, we can infer that
the average sales value for lost and won leads are very similar. Again, we must conduct an appropriate
hypothesis test to check whether there is a statistically significant difference in the sales value of leads
won and lost.

TN-Figure 4

rP
Average Sales Value of Lost and Won Leads

yo
op
TN-Figure 5 shows the percentage leads won across various countries. It is evident that the ability of
WSES to win leads from countries such as Africa, Canada, Japan and the UK are far better than countries
such as Singapore and Spain where their conversion is much lower.

TN-Figure 5
tC

Lead Conversion Percentage across Different Countries

No
Do

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 6 of 20

t
os
TN-Figure 6 shows density function of profit of the customer for won (Y = 1) and lost (Y = 0) leads.
From the plot, it is evident that the profit of the customers of leads won by WSES is lower compared to
the profit of the customers lost by WSES.

TN-Figure 6
Density of profit of the customer for won and lost leads

rP
yo
op
The instructor may encourage students to come up with more visuals to get an initial understanding of the
data. After initial understanding of the data, the instructor may get into hypothesis testing, correlation and
tC

ANOVA questions.

2. The marketing team of WSES believes that the average sales value of the leads they receive is at
least 8 million dollars. The marketing team believes that the standard deviation of sale value is
2 million dollars in the population. Use an appropriate hypothesis test to check whether the
average sales value in the population is at least 8 million dollars.
No

Recommended Answer:

In this question, we are testing the mean value of the sales leads from a sample (note that the data is a
subset of opportunities that the marketing team of WSES has identified and not the entire population of
opportunities that were available). Since the population standard deviation is given, we can use one
sample Z test for mean. Sampling distribution of the sample means will follow a normal distribution for
large samples when population standard deviation is known. In this case, Liz has taken a sample of Size
1000. So, it is safe to assume that the sampling distribution will follow a normal distribution. The
instructor may discuss central limit theorem while discussing this question.
Do

The instructor may ask the students what the null hypothesis and alternative hypothesis are in this case.
In my experience, many students struggle to decide the null and alternative hypotheses. Here, the
marketing team would like to prove that the average sales value is at least 8 million dollars, which will be
the research hypothesis and we would like to prove the same. As a convention, whatever we would like
to prove (the claim made) is taken as alternative hypothesis (also known as research hypothesis).

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 7 of 20

t
os
The null and alternative hypotheses in this case are given by:

H0:   8
HA:  > 8

rP
where  is the population mean. The Z statistic is given by:

X
Z , (1)
/ n

where X is the estimated mean from the sample,  is the population standard deviation and n is the

yo

sample size. The values of X ,  and n are given by:


X  8.04 ,  = 2 and n = 1000.

The value of Z is given by:

op

X 8.04  8
Z   0.632 .
 / n 2 / 1000

This is a right-tailed test, for  = 0.05, the Z critical value for right-tailed test is 1.64. Since the Z-statistic
is less than the Z critical value, we retain the null hypothesis, that is, H0:   8. The corresponding p-value
tC

is 0.2635.

3. Change the null and alternative hypotheses used by you in Question 2 and conduct the test.
What will be your conclusion based on the interchange of null and alternative hypotheses?
What do you learn from your answers to Questions 2 and 3?

Recommended Answer:
No

The purpose of this question is to demonstrate that the hypothesis testing is not symmetrical, that is, if
you change the null and alternative hypotheses, then it is possible that you may obtain conflicting results.
Consider the null and alternative hypotheses as shown below, which is opposite of what we used in
Question 2.

H0:   8
HA:  < 8
Do

Note that there will be no change in the Z-statistic value. However, this is a left-tailed test, and the Z-
critical value is -1.64. Since the calculated Z-statistic value is 0.632, which is greater than -1.64, the null
hypothesis is retained. The corresponding p-value is 0.7365, which is greater than the significance value
(0.05).

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 8 of 20

t
os
Now, compare decisions from Questions 2 and 3. In Question 2, we retained the null hypothesis   8;
and in Question 3, we retained the null hypothesis H0:   8, that is, we get two contradicting results.
This is because, hypothesis testing is not symmetrical. If we change the null and alternative hypotheses,
we may get contradicting results.

This question is likely to trigger much discussion about how to decide the null hypothesis. The best

rP
answer is that whatever is the current belief about the parameter should be taken as null hypothesis.
Alternative hypothesis is a research hypothesis which is something not proven so far.

4. If the actual average mean sales value of the leads is 8.2 million dollars, calculate the
probability of incorrectly concluding that the average sales value is less than 8 million dollars in
question 2 (assume  = 2).

yo
Recommended Answer:

This question can be used to teach Type II error and power of hypothesis test, which is failing to reject the
null hypothesis when it is false.

For the normal distribution with mean 8 and standard deviation 0.06325 ( 2 / 1000 ), the critical value at
 = 0.05 for a right tailed test is 8.10403. The Type II error is the probability below 8.10372 under
op
normal distribution with mean 8.2 and standard deviation 0.06325. The corresponding probability is
0.06396. That is, probability of incorrectly concluding that the average sales values is less than 8 million
dollars when the actual sales value is 8.2 million dollars is 0.0645 (Type II error). In statistics, notation 
is used for denoting Type II error and the value 1 –  is known as the power of test. That is, the power
hypothesis test in this case is 0.9354.
tC

5. Jason McCullagh, senior marketing manager at WSES doubted the value of standard deviation
provided by the marketing team for question 2. Jason argued that there is no way the
marketing team could have known the population standard deviation for the sales value, since
the population itself is unknown. Do you agree with Jason? If yes, perform the test again using
an appropriate hypothesis test.

Recommended Answer:
No

In many instances, the standard deviation of the population will be unknown; in such cases, the standard
deviation must be calculated from the sample.

The sample standard deviation (S) in this case is 1.98.

When the sample standard deviation is unknown, we must use the one sample t-test instead of Z test used
in Question 2.
Do

The null and alternative hypotheses are:

H0:   8
HA:  > 8

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 9 of 20

t
os
The t-statistic is given by:


X 8.04  8
t   0.6388 (2)
s / n 1.98 / 1000

rP
The critical t value  = 0.05 and degrees of freedom = 999 is 1.64638 (the t-distribution converges to
normal distribution for large sample size). The corresponding p-value is 0.2615, thus we retain the null
hypothesis.

6. Prudy Perkins, the Chief Marketing Office (CMO) informed the board that they win more than
50% of the sales leads they receive. Use an appropriate hypothesis testing procedure to check
whether the proportion of leads won by WSES is more than 50%.

yo
Recommended Answer:

To check this claim, we must perform one sample Z test for proportion. The null and alternative
hypotheses are:

H0: p  0.5
op
HA: p > 0.5

where p is the proportion of opportunities won in the population. The estimated value of p from the

sample is p  0.481 .
tC

The Z statistic for proportion is given by:


p p 0.481 0.5
Z   1.2016 (3)
p(1 p) 0.5 0.5
n 1000
No

The Z critical value in this case is 1.64. Since the Z statistic value is less than Z critical value, we retain
the null hypothesis. The corresponding p-value is 0.88.

7. Hendry Jackson, who works in the product line “learnsys”, claims that the probability of
winning a sales lead for the product “learnsys” is more than that of “Finsys”. Is there
statistically significant evidence in favor of Hendry’s claim?

Recommended Answer:
Do

In this question, the winning proportion for two different products is compared. We can use two sample
Z tests for proportions to test this hypothesis. The null and alternative hypotheses in this case are given
by:

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 10 of 20

t
os
H 0 : p1  p2
H 0 : p1  p2

where p1 is the proportion of “Learnsys” leads converted and p2 is the proportion of “Finsys” leads
converted. We have the following information from the dataset.

rP
n1 = Number of “Learnsys” leads received = 126
p1 = Proportion of “Learnsys” leads converted = (71/126) = 0.5635
n2 = Number of “Learnsys” leads received = 117
p2 = Proportion of “Learnsys” leads converted = (34/117) = 0.2906

The Z statistic for two proportion Z test is given by:

yo
 
( p1  p 2 )  0
Z ,
 1 1
p (1 p)    
 n1 n2 
op
where

 
 n p n p
p 1 1 2 2 .
n1  n2
Substituting the values, we get
tC

Z = 4.2908.

Since the Z statistic value is more than the critical value (1.64), we reject the null hypothesis, that is, the
conversion rate of “Learnsys” is more than that of “FinSys”. The corresponding p-value is 8.9 × 10 -6.

8. Hendry Jackson also claims that the average sales value of “learnsys” projects is higher than
that of “Finsys” projects. Check whether Jackson is correct at 5% significance.
No

Recommended Solution:

The purpose of this question is for the students to understand the use of two-sample t test for mean with
unequal variance (Welch’s t-test). In this case, null and alternative hypotheses are:

H0: 1  2
Do

H0: 1 > 2

The test statistic is given by:

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 11 of 20

t
os
  

X
 1  X 2 0
t  .
S1 S 22
2

n1 n2

rP
The degrees of freedom is:

 
 Su4 
df   ,
 (S12 / n1 )2 (S 22 / n2 ) 2 
 n 1  n 1 

yo
 1 2 

where

S12 S 22
Su   .
n1 n2
op
Substituting the values, we get the t-statistic value as 0.9319 and the corresponding t-critical value for  =
0.05 is 1.65. TN-Table 1 shows the Microsoft Excel output for the two-sample t-test. The p-value is
0.17; so, we retain the null hypothesis.

TN-Table 1
tC

t-Test: Two-Sample Test Assuming Unequal Variances for average sales value of
“Learnsys” and “Finsys”

Variable 1 Variable 2
Mean 8.03013691 7.804548784
Variance 3.277004578 3.811911322
No

Observations 126 117

Hypothesized Mean Difference 0
Df 236
t Stat 0.931988116
P(T ≤ t) one-tail 0.176147559
t Critical one-tail 1.65133585
Do

P(T ≤ t) two-tail 0.352295118

t Critical two-tail 1.970066853

We can also check whether the variance is equal for two samples in this case using test for equality of
variance.

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 12 of 20

t
os
9. WSES decide to target customers who earn a higher profit with the belief that it is easier to
convert customers who make more profit. That is, they believe that the average profit of
customers won by WSES is more than the average profit of customers lost by WSES. Use an
appropriate hypothesis test to check whether this belief is correct.

Recommended Solution:

rP
The instructor may ask the students how to construct null and alternative hypothesis in this case.
Assume that:

1 be the average profit of the customer leads won by WSES and 2 be the average profit of the
customer leads lost by WSES.

yo
H0: 1  2
HA: 1 > 2

We can conduct a two-sample t test in this case to decide whether to reject or retain the null
hypothesis. The two-sample t test with unequal variance output using Excel is shown in TN-Table 2.

TN-Table 2
op
Two Sample t test for profit of the customers

t-Test: Two-Sample Assuming Unequal Variances

tC

Profit of
Profit of Customer Customer Lost
WON Leads Leads
Mean 0.709126819 1.311637765
Variance 0.169508819 0.158438046
No

Observations 481 519

Hypothesized Mean
Difference 0
df 986
t Stat -23.49394981
t Critical one-tail 1.646400494
t Critical two-tail 1.962372845
Do

The t statistic value is -23.49, this is a right tailed test and thus we retain the null hypothesis since the
critical value at  = 0.05 is 1.64. That is, the average profit of customers won by WSES is lower than the
average profit of customers lost by WSES, which is against their belief.

This result may look counterintuitive, that is, WSES is able to convert the customer leads with
lower profit than the customer leads with higher profit. This is due to the reason that customers
who make higher profit may not like to change the system, whereas customers who make lower
profit are looking for new systems to improve their performance.
This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 13 of 20

t
os
10. Prudy Perkins, chief marketing officer, advises her sales team that the sales value of a lead
should not be a criterion to be used before deciding to pursue the sales lead. Is there evidence
to suggest that the sales value has no influence on the ability of WSES to convert a sales lead?

Recommended Solution:

rP
We can argue that if the value of sales has no influence on the conversion of an opportunity, then the
average value of sales for both converted and non-converted leads should be same. Let:

1 be the average sales value of the sales leads won by WSES and 2 be the average sales value of the
sales leads lost by WSES. The corresponding null and alternative hypothesis are:

H0: 1 = 2

yo
HA: 1 ≠ 2

We can use two sample t test with unequal variance to check this hypothesis. The corresponding Excel
output is shown in TN-Table 3.

TN-Table 3
Two Sample t test for checking equality of average sales values of won and lost opportunities
op
t-Test: Two-Sample Assuming Unequal Variances
tC

Sales Value of won Sales Value of Lost

opportunities opportunities
Mean 8.095634096 7.996531792
Variance 3.973390483 3.896553199
Observations 481 519
Hypothesized Mean
No

Difference 0
df 991
t Stat 0.789202776
P(T<=t) one-tail 0.215091135
t Critical one-tail 1.646392682
P(T<=t) two-tail 0.430182269
t Critical two-tail 1.962360677
Do

In TN-Table 3, the value of t statistic is 0.7892 the corresponding t critical value is 1.96 (this is a two
tailed test). The corresponding p-value under two tailed t test is 0.43. Since the p-value is greater than 
(= 0.05), we retain the null hypothesis. That is, there is no difference in the average sales value of
opportunities won and lost by WSES.

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 14 of 20

t
os
11. Jack Williams, the CEO of the company believes that the sales conversions are different for
different products as well as different geographical locations. Check the validity of Jack’s belief
using an appropriate hypothesis test.

Recommended Answer:

rP
You may ask the students what test should be used to answer this question. You can expect the students to
come up with answers such as multiple two-sample Z test for proportion, etc. However, the easiest and
best way to check this claim is by using chi-square test of independence. The null and alternative
hypotheses are as follows:

H0: The region and sales outcome are independent

HA: The region and sales outcome are dependent

yo
There are nine regions in the data; we must first create a contingency table. The contingency table is
shown in TN-Table 4:

TN-Table 4
Contingency Table for Chi-Square of Independence
op
Region Sales Outcome
Won (1) Lost (0) Total
Africa 55 38 93
Americas 55 49 104
Canada 4 2 6
India 17 18 35
tC

Japan 10 6 16
Other Europe 66 92 158
Singapore 6 17 23
Spain 1 11 12
UK 267 286 553
Total 481 519 1000
No

The chi-square statistic for test of independence is given by:

9 2  O  E 2 
 2    ,
ij ij

i1 j1  E 
 ij

where Oij is the observed frequency and Eij is the expected frequency. Here, i denotes the region (i = 1, 2,
…, 9) and j is the sales outcome (j = 1 for won and j = 0 for lost). The values of Oij are provided in TN-
Table 4 (calculated from the data). However, Eij can be calculated using the following formula:
Do

sum of row i sum of column j

Eij 
Total sum
Substituting the values, we get the chi-statistic value as 22.26. The corresponding chi-square critical
value is 15.50 (the degrees of freedom is 8 in this case).

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 15 of 20

t
os
Since the Chi-Square Statistic value is greater than the chi-square critical value, we reject the null
hypothesis. That is, Jack’s claim is correct that the win proportion and geographical locations are
dependent.

12. Joe Danby, the chief financial officer believes that the sales conversions depend on the sales
value. Use an appropriate hypothesis test to check the validity of this claim by making the

rP
following three groups:

a. Sales value less than 6 million dollars.

b. Sales value between 6 and 8 million (both inclusive) dollars.
c. More than 8 million dollars.

Recommended Response:

yo
Here again, we will perform the chi-square test of independence. The contingency table is shown in TN-
Table 5.

TN-Table 5
Contingency Table
op
Sales Value Sales Conversion Total
Won (1) Lose (0)
Value less than 6 million 70 86 156
Value between 6 and 8
million 163 180 343
Value more than 8 million 248 253 501
tC

481 519 1000

The chi-square statistic value is 1.09 and the corresponding critical value is 5.99. So, we retain the null
hypothesis, that is, the sales value and sales conversion are independent. The corresponding p-value is
0.5795. Answer to this question is consistent with Question 10.

13. Liz was of the opinion that there was no difference in the average profit and the geographical
No

locations: United Kingdom, India and the Americas. Use an appropriate test to verify the same.

Recommended Response:

Note that the profit is a continuous value and geographical location is a factor. We have to use one-way
ANOVA to conduct the hypothesis test.

Assume that:
Do

1 = Average profit in United Kingdom

2 = Average profit in India

3 = Average profit in the Americas

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 16 of 20

t
os
The null and alternative hypotheses are:

H0: 1 = 2 = 3

H0: Not all  values are equal.

rP
The ANOVA table using Microsoft Excel is shown in TN-Table 6.

TN-Table 6
ANOVA: Single Factor

SUMMARY

yo
Groups Count Sum Average Variance
UK 553 27981 50.59855 97.15739
India 35 1873 53.51429 114.5513
The Americas 104 5234 50.32692 102.4164
op
ANOVA
Source of
Variation SS df MS F P-value F crit
Between Groups 297.355 2 148.6775 1.504804 0.22279 3.008795
Within Groups 68074.51 689 98.8019
tC

Total 68371.86 691

It is evident from TN-Table 6 that the F-statistic value is 3.008 and the corresponding p-value is 0.22.
Hence, we retain the null hypothesis.

14. Use an appropriate statistical measure to understand the relationship between (1) Joint bid
percentage and profit; (2) Profit percentage and sales value; and (3) Joint bid and sales
No

conversion.

Recommended Answer:

This question can be used to teach Pearson correlation and bi-serial correlations and inferences based on
correlation. The students can also use visuals such as scatter plot to understand the relationship between
these variables.
Do

1. Joint bid percentage and profit

Scatter plot between joint bid percentage and profit is shown in TN-Figure 7. From scatter plot, we can
see that there is no evidence that there is a relationship between these two variables.

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 17 of 20

t
os
TN-Figure 7
Scatter Plot between Profit Vs. Proportion of Joint Bid

rP
yo
op
Both these variables are continuous variables and thus we can use Pearson product moment correlation.
Pearson correlation is given by the following equation:
tC

Cov(X ,Y )
r ,
S X  SY

where Cov(X, Y) is the covariance between X and Y and Sx and Sy are the sample standard deviations of
X and Y. The correlation between joint bid percentage and profit is -0.01256, which is very low. In fact,
the hypothesis test for correlation can be stated as follows:
No

H0:  = 0
HA:  ≠ 0

The corresponding statistic is a t-statistic and is given by:

n2
tr   0.39 .
1 r2
Do

The t-statistic value is between the t-critical values (-1.96 and 1.96) and thus we retain the null
hypothesis.

2. Profit Percentage and Sales Value

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 18 of 20

t
os
Correlation coefficient value is -0.01475. The scatter plot between profit percentage and sales value is
shown in TN-Figure 8. Again there is no evidence to suggest that there is a relationship between these
two variables.

TN-Figure 8
Scatter plot between Profit Percentage and Sales Value

rP
yo
op
3. Relative strength in the industry and sales conversion
tC

In this case, the variable relative strength is a continuous variable and sales conversion is a binary
variable. So, we cannot use the Pearson correlation coefficient. We have to use bi-serial correlation. The
bi-serial correlation coefficient is given by:

 
X 1 X 0 n0 n1
rb  ,
SX n(n  1)
No

 
where X 1 and X 0 are the mean values of relative strength for won and lost sales leads. SX is the
standard deviation of relative strength, n1 and n0 are the number of leads that are won and lost,
respectively. The value of rb = -0.3285, that is, as the relative strength increases, the chance of winning
decreases, which is a counter-intuitive relationship. The reason could be complacency of the sales team
since it is stronger product in the market.

15. Create two groups based on the classification of leads by the sales team. In group 1, include all
Do

the leads that are classified as “excellent chance of winning”, by the sales team and in group 2
include all the leads that are classified as “low chance winning”. Using an appropriate
hypothesis test, check whether the proportion of leads won in group 1 (leads classified as
excellent chance of winning) is higher than the proportions of leads won in group 2 (leads
classified as low chance of winning).

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 19 of 20

t
os
Recommended Response:

Let:
n1 = Number of leads that are classified as excellent chance of winning by the sales team
n2 = Number of leads that are classified as low chance of winning by the sales team
w1 = Number of leads won that are classified as excellent chance of winning by the sales team

rP
w2 = Number of leads won that are classified as low chance of winning by the sales team
= Proportion of leads won that are classified as excellent chance of winning
= Proportion of leads won that are classified as low chance of winning

Group 1 (Leads Classified as Excellent Chance of Group 2 (Leads Classified as Low Chance of

yo
Winning) Winning)

n1 390 n2 398
w1 190 w2 198
0.4871 0.4974

We can perform Z test for two samples. The null and alternative hypothesis are:
op
H0: p1  p2
HA: p1 > p2

The Z statistic for two proportion Z test is given by:

tC

 
( p1  p 2 )  0
Z ,
 1 1
p (1 p)    
 n1 n2 

where
No

 
 n p n p
p  1 1 2 2 = 0.4923
n1  n2

Substituting the values, we get

Z = -0.2891
Do

This is a right tailed test, since the Z statistic value is less than the critical value (1.64), we retain the null
hypothesis. That is, the win percentage of leads classified as “excellent chance of winning”, is lower than
the leads classified as “low chance of winning.

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.
 Teaching Note – Testing Marketing Hypothesis at WSES Page 20 of 20

t
os
16. Discuss the business insights based on the data analysis.

Recommended Response:

Using the hypotheses tests correlation and ANOVA, the student can arrive at many insights. Few of them
are listed below:

rP
1. Sales value has no influence on conversion of sales leads.
2. The leads won by WSES have lower profit of the customer compared to the leads lost by them.
They should focus on customers who are making lower profit.
3. Conversion probability of “Learnsys” is better than “Finsys”. Thus, if there are two leads; one
from “Learnsys” and another from “Finsys”, WSES should focus on “Learnsys”.
4. There is dependency between geographical location and sales conversion. Certain geographical

yo
locations have a higher probability of winning, so they can focus on locations where conversion
rates are high.
5. Average profit is not statistically different across geographical locations.
6. Relative strength in the industry and winning probability are negatively correlated.
7. The “gut feel” of sales team and their ability to convert a deal are not related.
op
tC
No
Do

This Teaching Note is authorized for use only by SHAILAJA REGO, Narsee Monjee Institute of Management Studies (NMIMS) until Jun 2023. Copying or posting is an infringement of
copyright. [email protected] or 617.783.7860.