Shah Imr Day 4

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 47

International Marketing Research

DAY 4
November 25, 2019
Mumbai, India

Dr. Denish Shah


Barbara & Elmer Sunday Professor of Marketing
Director, Social Media Intelligence Lab
Co-Director, (CMO) Marketing RoundTable
Robinson College of Business
Georgia State University, Atlanta, GA

INTERNATIONAL MARKETING RESEARCH

© Dr. Denish Shah 1


Why International Marketing Research?

• Growth comes from Globalization


•Not only bigger businesses but
also more complex businesses
• No two countries are alike! Differences can be
multidimensional

•If Hyundai wants to sell cars in markets outside of


South Korea, how can it do so?

Defining International Marketing Research

Marketing Research:
The function that links an organization to its market through information

Marketing Research Function:


 Specifies the information objective(s)
 Designs the method for collecting information
 Manages and implements the data collection process
 Interprets the results
 Communicates the findings and its implications

International Marketing Research:


The function that links an organization to its international markets through information

© Dr. Denish Shah 2


Global Trends Influencing Market Research

• Rapid Internet Penetration

• Consumer-Generated Marketing

• Mobile Marketing

• Advances in digital communications

What is different about


International Marketing Research?
International researchers have to deal with a number of countries that
may differ considerably in a number of important ways. Some of these
differences are:

Cultural differences: Racial differences:


Differences in races mean
Culture refers to widely shared
difference in physical features
norms or patterns of behavior
as well. Products must be
within a large group of people.
modified to meet the needs of
Cultural differences shape
attitudes that consumers have different races.
towards products.

© Dr. Denish Shah 3


What is different about
International Marketing Research? (contd.)
• Economic differences: Levels of wealth and taxation affect
consumer behavior in different countries. Eg: Norwegians consume
very little alcohol due to high taxation

• Religious differences: Certain religions have laid down specific


behavioral patterns. Eg: Middle Eastern countries prohibit alcohol
consumption

• Historical differences: Slowly evolved differences that have


profound effects on consumer behavior. Eg; Scotch whisky is
considered prestigious and trendy in Italy but old-fashioned in Scotland

• Climatic difference: Differences in climate accounts for difference


in consumer behavior.

What is different about


International Marketing Research? (contd.)
• Differences in consumption patterns: There are vast
differences in consumption patterns between regions.For
Eg: The French prefer wine, the Germans like ………, the
Russians prefer….. and the Spanish drink aperitivos.
• Differences in marketing conditions: For Eg: Japanese
do not like being contacted over the telephone for
interviews, while in Hong Kong interviews are conducted
through the grill in front door.
• Differences in actual and potential target groups: Eg: It
is easier to collect national samples in countries like
England and Germany than in Spain.

© Dr. Denish Shah 4


What is different about
International Marketing Research? (contd.)
International Marketing Researchers may also have to deal
with:

• Language differences

• Differences in market research facilities

• Differences in the criteria for assessing products or services

• Differences in product usage

Culture
Dimensions of Culture:

• Power Distance

• Individualism vs.
Collectivism

• Masculinity vs. Femininity

• Uncertainty Avoidance

• Long-term Orientation

10

© Dr. Denish Shah 5


Culture
Power Distance
The extent to which less powerful members of a
society accept and expect that power is distributed
unequally.
Example: Japan has a high power distance index
and the US has low power index.

11

Culture
Individualism Vs. Collectivism

Example: Japanese are collectivistic where as the


Americans are individualistic.

12

© Dr. Denish Shah 6


Culture
Masculinity Vs. Femininity
The dominant values in a masculine society are
achievement and success, the dominant values
in a feminine society are caring for others and
quality of life.
Example: Japan & Middle East is masculine,
while the Scandinavian countries are feminine.

13

Culture
Uncertainty Avoidance
The extent to which people feel threatened by
uncertainty and ambiguity and try to avoid these
situations.
Example: High on uncertainty avoidance are
Germany and Japan, low ones are Sweden and
Denmark.

14

© Dr. Denish Shah 7


Culture
Long-term Orientation
The extent to which a society exhibits a pragmatic
future-oriented perspective than a conventional
historic or short-term point of view.
Example: Most Asian countries, China in particular
score high on this dimension. Many western
countries have short-term orientation.

15

Cultural Influences
• Different cultures react to surveys differently.
− Eg: The concept of time in Germany, Italy, United States
and France is very different.

• In the United Sates and most developed countries:


− High rate of refusal because of the large number of
surveys
− Surveys should not be very long. (20-25 minutes in the
U.S.)
− In time conscious cultures, surveys should not last longer
than stated time

16

© Dr. Denish Shah 8


Cultural Influences (contd.)
• Interesting Facts
– The French and the Italians are very flexible about
time.
– Germans are very conscious about time. They will
seldom allow the interviewer to proceed beyond the
time stated before the survey began.
– The English are too polite to cut off the interviewer
– Scandinavians are cooperative and have lower
refusal rates.
– People in the Mediterranean except the Spanish, like
to converse.

17

Cultural Influences (contd.)


• Interesting Facts (Cont.)
– Japanese consider it impolite to be surveyed over
the telephone
– In Hong Kong, researchers are not allowed inside
the residence.
– In the Middle East Women cannot be interviewed
in their houses without the presence of men.

18

© Dr. Denish Shah 9


Qualitative and
Observational Research

Cultural Differences in Focus Groups

19

Cultural Difference In Focus Groups

• In Japan, Focus Groups are not used as people


agree with each other and therefore unbiased
information is not obtained.

• The more casual way of life in countries as


Thailand, Malaysia and Indonesia means that a
commitment to turn up for a focus group is less
likely than in west.

20

© Dr. Denish Shah 10


Cultural Difference In Focus Groups
• Women in the Middle East do not have as much
freedom as women in the Western world

• In some Asian countries, the moderator may


have to give special attention to introductions for
the discussion to be truly open and candid

• The French generally resist innovations of any


kind

21

Cultural Difference In Focus Groups


• The Japanese hesitate to criticize new product
ideas
• There is a very strong group mentality in Asian
culture while most Americans tend to be
individualistic
• Lead times tend to be longer, especially in the
Far East
• Use of external stimuli such as photos and visual
aids are not prevalent in foreign focus groups

22

© Dr. Denish Shah 11


Types of International Marketing Research
• Descriptive Research
Researcher examines in-depth the attitudes and behavior of
consumers in another country or culture.
• Comparative Research
Involves comparing attitudes and behaviors in two or more countries
or cultural contexts, with a view to identifying similarities and
differences between them.

• Theoretical Research
The researcher has a predetermined theory or model and the
objective is to cross-cultural generalizability of those theories or
models.

23

Classification of International Marketing Research

International Marketing
Research

Single- Simultaneous
country Multi-country
research research

Multi- Sequential Multi-


country country research
research
Independent
Multi-country
research

24

© Dr. Denish Shah 12


Research Pitfalls

– Too much emphasis on


statistical techniques
– Data becoming unduly
important
– Not enough data
interpretation
– Too much reliance on
computers and scanner data
– Lack of understanding
among users
– Researchers see
themselves as data
collectors rather than data
interpreters

25

Marketing Researcher of the 21st Century

Requirements include:
• Speed
• Use of Internet
• Globalization
• Ability to add value to numbers and data

26

© Dr. Denish Shah 13


Ethics in Research

Researchers should consider


• Respondents’ rights - Should be allowed to
participate voluntarily and there should be no
coercion.
• Sponsors’ rights - Researcher is morally
bound to conduct research in the manner that
has been agreed upon with the sponsor.

27

Challenges to
International Market Research

• Growing market
complexity

• Blurring of categories

• Shortened product
development cycles

• Feature overload

28

© Dr. Denish Shah 14


Things to keep in mind while conducting
Global Research
China One of the main problems of conducting research is the size
of the market. Media is heavily regulated by the government.

India The biggest problem is that of accuracy and reliability, which


are lacking in marketing research studies. The Indian market
is a sellers market, hence it is difficult to conduct market
research.

Japan Sometimes mangers are unwilling to discuss their company


or their employees because of the space constraint in
Japanese offices which forces them to work in cubicles.
Several Japanese companies are looking at cost efficiency
as a main prerogative

29

Things to keep in mind while conducting


Global Research
Germany The one misnomer that marketers have in targeting consumers
in Germany is in assuming that most of them live in urban
areas.
France The French are idealistic in their vision of current lifestyles and
often skew their responses to reflect a better state than an
actual state.
UK In term of responsiveness to purchase or participate in
international market research activities, the United Kingdom,
lags behind several other European countries.
Mexico Gifts and not money to be used as incentives, paying money is
not very appealing.
Canada One of the main problems that several companies face is that
consumer groups complain regarding the invasion of privacy.

30

© Dr. Denish Shah 15


Information Requirements for
International Marketing
Depends on Type of Decision

• Strategic Decisions
• Foreign Market Selection
• Mode of Entry
• Product/Market Portfolio Strategies
• Market Expansion Strategies
• Tactical Decisions

31

Mode of Entry

• Exporting
• Licensing and other
forms of contractual
agreements
• Joint ventures
• Wholly owned
subsidiaries

32

© Dr. Denish Shah 16


Expansion Strategy

• Sprinkler strategy:
A firm decides which markets it plans to enter and
simultaneously enter all these markets.

• Waterfall strategy:

The firm follows a roll out policy, entering the most


profitable market first.

Conservative approach.

33

Multinational Diffusion

General Example
Consider a situation where a new product is introduced
in two neighboring, but culturally different countries. We
will expect that the diffusion process in the two countries
will be different, but at the same time we expect some
interaction among the countries, especially if the two
societies mingle with each other.

34

© Dr. Denish Shah 17


Problem Orientation

Researchers face two subjective problems:


• Cultural shock or psychological jolts in
encountering a wide variety of unfamiliar value
systems

• Self Reference Criterion: Unconscious


tendency to assume that every one else has
similar values, attitudes, lifestyles, etc.

35

Equivalences in
International Marketing Research
Construct Equivalence

Measurement Equivalence
Equivalences in
International Marketing
Research
Sampling Equivalence

Analysis Equivalence

© Dr. Denish Shah 18


Establishing Equivalence in IMR
• Construct equivalence
Deals with the function of the product or service that is being
researched. Different countries that are being studied must have
the same perception or use for the product that is being
researched.
Eg: Utility of Bicycles in India & China compared to the US

• Measurement equivalence
Concerned with the scales used for measuring various aspects of
the research study. In a multi-country study, researchers need to
modify the units of measurement.
Eg: Scales used in US (5 point scale) are very different from the
ones used in France (20 point scale)

Establishing Equivalence in IMR (contd.)

• Sampling equivalence:
Concerns the relative size of the sample to obtain
reliable results.

• Analysis equivalence:
When analyzing data, the different biases that
might exist should be accounted for

© Dr. Denish Shah 19


Sentiment Analysis

Sources:
Bing Liu, 5th Text Analytics Summit, Boston
SPSS White Paper
Gartner Report

Data at the heart of


the Predictive Enterprise

Descriptive data Behavioral data


– Attributes – Orders
– Characteristics – Transactions
– Self-declared info – Payment history
– (Geo)demographics – Usage history

Source: SPSS Inc.

40

© Dr. Denish Shah 20


Data at the heart of
the Predictive Enterprise
Interaction data Attitudinal data
– Offers – Opinions
– Results – Preferences
– Context – Needs

Descriptive data Behavioral data


– Attributes – Orders
– Characteristics – Transactions
– Self-declared info – Payment history
– (Geo)demographics – Usage history
Source: SPSS Inc.

41

Market Research:
The Voice of the Customer

Solicited Feedback Unsolicited Feedback

Online Surveys Customer Complaints


Post purchase Social Media
surveys/reviews Conversations
Product Reviews

42

© Dr. Denish Shah 21


Mining unsolicited feedback

 Companies today have an unprecedented access to


consumers’ conversations across social media platforms
 Such (unstructured) data contains a wealth of information

 Information of relevance to marketers include consumers’:


 Sentiments towards a brand
 General mood or state of mind
 Verbal reaction to a product/service experience
 Unsolicited product reviews
 Likes, dislikes, aspirations, etc.
 Consequently,marketers are increasingly employing text
mining and sentiment analyses.

43

Text Mining and


Sentiment Analyses

 Two main types of textual information.


 Facts and Opinions
 Most of the text mining methods work relatively
well with factual information.
 Sentiment analysis or opinion mining
 Computational study of opinions, sentiments and
emotions expressed in text.
 Much more complicated than mining factual
information.

44

© Dr. Denish Shah 22


How does Sentiment
Analyses work?

45

Example of a comment/opinion
posted by a consumer:

 “I bought an iPhone a few days ago. It was such a


nice phone. The touch screen was really cool. The
voice quality was clear too. Although the battery life
was not long, that is ok for me. However, my mother
was mad with me as I did not tell her before I bought
the phone. She also thought the phone was too
expensive, and wanted me to return it to the shop. …”
 What do we see here?
 Opinions (sentiments), features, timelines, targets
of opinions, and opinion holders

46

© Dr. Denish Shah 23


Key Challenge:
Imposing a structure to the unstructured data

An opinion may be essentially decomposed into


five components:
(F,S,H,O,T)
Where:
‘O’ is the target object
‘H’ is the opinion holder
‘T’ is the time when the opinion is expressed
‘S’ is the sentiment value of the opinion
‘F’ is a feature of the object mentioned in the opinion

47

Structuring the unstructured

 Given an opinion:
 Discover all components of the opinions for further
analyses

 This process may be executed at different


levels and in different ways.

48

© Dr. Denish Shah 24


Sentiment Classification:
Aggregate-level
• Classify an opinion based on the overall sentiment
expressed by an opinion holder
• Example: Positive or Negative
• Assumption: Each opinion focuses on a single object and is from a single
opinion holder.
• Is the sentiment positive or negative here?
• “I bought an iPhone a few days ago. It was such a nice phone. The
touch screen was really cool. The voice quality was clear too.
Although the battery life was not long, that is ok for me. However, my
mother was mad with me as I did not tell her before I bought the
phone. She also thought the phone was too expensive, and wanted
me to return it to the shop. …”

 Aggregate level classification may be inaccurate in open ended


conversations and posts.
49

Sentiment classification:
Individual sentence level

 The sentiment from each sentence is classified as


follows:
 Subjectivity classification: Is the sentence subjective or objective?
• Objective: e.g., I bought an iPhone a few days ago.
• Subjective: e.g., It is such a nice phone.

 Sentiment classification: For each subjective sentence, classify


the sentiment as positive or negative.
• Positive: e.g., It is such a nice phone.

50

© Dr. Denish Shah 25


Potential Issues

 With Aggregate level sentiment classification:


 They do not tell what people like and/or dislike
 An overall positive opinion on an object does not mean that
the opinion holder likes everything.
 An overall negative opinion on an object does not mean that
the opinion holder dislikes everything

• With Sentence level sentiment classification:


– Not all Subjective sentences may necessarily contain a positive
or negative opinion
– Example: I think he came yesterday.
– Not all Objective sentence may be neutral.
– Example: The phone broke in two days (implies a negative opinion)

51

Feature-Based Sentiment Analyses


(Sentiment is based on each relevant feature of the object)

“I bought an iPhone a few Feature Based Summary:


days ago. It was such a
Feature1: Touch screen
nice phone. The touch Positive: 212
screen was really cool.  The touch screen was really
The voice quality was clear cool.
 The touch screen was so easy to
too. Although the battery use and can do amazing things.
life was not long, that is ok …
for me. However, my Negative: 6
 The screen is easily scratched.
mother was mad with me
 I have a lot of difficulty in
as I did not tell her before I removing finger marks from the
bought the phone. She touch screen.

also thought the phone
Feature2: battery life
was too expensive, and …
wanted me to return it to
the shop. …” Note: Opinion holders are eliminated
52

© Dr. Denish Shah 26


Visual Comparison
(feature-based sentiment analyses)
+
• Summary of
reviews of
Cell Phone 1

_
Voice Screen Battery Size Weight

• Comparison of +
reviews of
Cell Phone 1
Cell Phone 2
_
53

Feat.-based opinion summary in Bing

54

© Dr. Denish Shah 27


Integrating customer sentiments to
customer transaction data

Source: SPSS Inc. 55

Call Center Integration


1. CSR enters live
customer comments

2. When saved,
comments are
analyzed using
Predictive Text
Analytics

3. Customer churn score


is generated and
displayed in real-time

4. CSR script responds


dynamically

Source: SPSS Inc. 56

© Dr. Denish Shah 28


Threats to
Sentiment Analyses

57

An Example of Practice of Review Spam


Belkin International, Inc
• Top networking and peripherals manufacturer | Sales ~ $500 million in 2008
• Posted an ad for writing fake reviews on amazon.com (65 cents per review)

Jan 2009

58

© Dr. Denish Shah 29


Application of Sentiment
Analyses

59

Quantitative Techniques

© Dr. Denish Shah 30


Quantitative Techniques

Why are they important?

MANAGERIAL
DATA INSIGHTS
ACTIONS

QUANTITATIVE
TECHNIQUES

Popular Data Analyses Techniques

 Cluster Analysis
 Factor Analysis
 Machine Learning / Artificial Intelligence
 Linear Regression
 Logistic Regression

62

© Dr. Denish Shah 31


Cluster Analysis

Cluster Analysis
 It is a technique for segmenting customers / markets
 Cluster analysis helps to organize customers into different groups
whose members are similar in some way
 Each group is formed such that the inter-group difference and the
intra-group similarities are maximized.
 There is no distinction between dependent and independent
variables
 Often used at the initial stage as an exploratory technique.
Example:

Income

Fashion
conscious
64

© Dr. Denish Shah 32


Cluster Analysis – An Example

Consider a population of 40,000 customers that are grouped into two


clusters based on age and disposable income.

Partitioning of data

Cluster 1 Cluster 2

Customer A Customer B
20 years 60 years
$10K $40K

65

Cluster Analysis – An Example contd.

Assignment

Customer C
30 years
$25K
Cluster 1 Cluster 2
? ?

Customer A Customer B
20 years 60 years
$10K $40K

66

© Dr. Denish Shah 33


Cluster Analysis – An Example contd.

Squared distance of > Squared distance of


Customer C from the Customer C from the
centroid of Cluster 1 centroid of Cluster 2

Cluster 1 Cluster 2

Customer A
Customer B
20 years
$10K 60 years
Customer C $40 K
30 years
$25K

67

Clustering Customers @ Best Buy

Customer Clusters Best Buy Store Redesign

Database Analysis Jill (Soccer mom)

Local Census Buzz (Young tech.


Numbers enthusiast)
Barry (Wealthy
Customer Surveys
professional man)
Focus Group
Ray (Family man)
Interviews

68

© Dr. Denish Shah 34


Factor Analysis

What?
 Data Reduction technique. Used to meaningfully group
variables.

How?
 FA creates linear combinations or groups of variables
(called factors) based on the underlying correlation
structure

Why?
 Helps identify meaningful relationships amongst variables

 Helps to summarize predictors into fewer data elements

69

Factor Analysis - Example


Consider FA done on a sample of 100,000 customers taken randomly from a Direct
customer database.
Factor Analysis Output

Variable/Data Element Factors


Factor 1 Factor 2
Graduate level education √
High income √
Student Debt = Yes √
Age: 40 to 65 √
Luxury car owner √
Rent apartment = yes √
Low to medium income √
Age: Less than 30 √
Used car owner √
House owner √
Marital status: Unmarried √
70

© Dr. Denish Shah 35


Interpretation of Dimensions/Factors
Factor Analysis Output

Variable/Data Element Factors


Factor 1 Factor 2
Graduate level education √
High income √
Student Debt = Yes √
Age: 40 to 65 √
Luxury car owner √
Rent apartment = yes √
Low to medium income √
Age: Less than 30 √
Used car owner √
House owner √
Marital status: Unmarried √

 Factor 1 = High socio-economic status


 Factor 2 = Low socio-economic status

71

Can you identify the key


differences between Cluster
Analysis & Factor Analysis?

© Dr. Denish Shah 36


Machine Learning

 Originated from computer Science. Sub-set of Artificial Intelligence.


 The technique simulates the way human brain learns.
 Combines statistics with computer programming.
 Considered to be the most powerful quantitative technique.
 Used to uncover hidden patterns in data (unsupervised learning) or
develop an association between output and input (supervised
learning).
Neural Network

Hidden

Input Output

73

Machine Learning

 Steps involved:
 Train the NN using the available observed data
 The NN will establish complex relationship between input and
output based on the ‘training’ data
 The NN will then use the established relationship to predict
responses with new input data or uncover the underlying
pattern or structure in the data (depending on the objective of
the analyses).
 Major advantages:
 Can support complex non-linear relationships
 Known to have the highest predictive ability
 Can work with big data to give results in real-time
 Major drawback:
 Black-box technique

74

© Dr. Denish Shah 37


Contemporary Marketing
Applications of Machine Learning

 Most popular machine learning applications


focus on big data and/or real-time marketing
decision making. For example:
 Online product recommendation engines
 Identifying look-alike fans/followers on social media
 Brand sentiment analyses in real time
 Serving the most relevant online ad to a web page
visitor
 Predicting churn in contractual settings
 Detecting outlier behaviors and/or micro-segments
 Uncovering patterns leading to desired (or
undesired) customer behavior outcomes.

75

May 6, 2010 Flash Crash

© Dr. Denish Shah 38


19,000 Jobs
$985 Million

@ JC Penney

Recommended Approach

INTUITION INSIGHTS
= OPTIMAL
(from experience) (from data)
MANAGERIAL
ACTIONS

© Dr. Denish Shah 39


79

Correlation versus Causality

A B A B

When A changes, A causes B to


B changes change

80

© Dr. Denish Shah 40


Regression

 Most tried and trusted technique


 Has been in existence for more than a century

 Models the relationship between the key variable of


interest (Dependant Variable) and one or more other
variables (Independent Variables) that are known to
influence the variable of interest
 Linear versus Non-Linear Regression
 Simple versus Multiple Regression

 Several software packages (including Microsoft Excel)


can perform regression analysis

81

Simple Linear Regression Model


A Simple Linear Regression (SLR) model is a linear mathematical equation/line that
defines the best linear relationship between 2 variables, say x and y:
y = a + bx + u
y  Response or Dependant Variable
x  Predictor variable or Independent Variable
a  Intercept
b  Slope of the regression line
u  Error term

A simple linear regression line is


the line that best fits the data
Customer Value (y)

Income in thousands (x)


82

© Dr. Denish Shah 41


Coefficient of Determination
The coefficient of determination tells the amount of variation in the response
variable y explained by the introduction of the predictor variable x
0 < R² < 1

R² = 1

R² = 0 R² = 0.5

83

Regression Analysis Example

Suppose an Electronics store wants to measure the


impact of its Direct Marketing program on iPhone sales.

Response Variable: iPhone Sales (in thousands units)

Predictor Variables: Price ($), Advertisement,


Number of apps on iTunes,
Direct Marketing Campaign
(1 = Yes, 0 = No)

84

© Dr. Denish Shah 42


Sample Output of a Regression Analysis

Regression
Independent Variables P value
Coefficient
Intercept 1.2 0.0001
Price (in Hundreds) -0.2 0.0001
Advertisement (in TARPS) 0.7 0.1234
Number of apps
0.4 0.001
(in thousands)
Direct Marketing Campaign
0.85 0.0001
(Binary Variable)

R² = 78%

Impact of Direct Marketing = Additional 850 iPhone units sold

85

Regression Output (with interpretation)

In general, the regression coefficient indicates the amount by which the dependent
variable (or ‘Y’) will change corresponding to a unit change in the respective independent
variable (or ‘X’). It is important to keep the scale (e.g., binary, hundreds, thousands, etc.)
of the respective variables in mind while interpreting the effect size.

Independent Regression
Interpretation P value
Variables Coefficient
Intercept 1.2 Base line from where the regression line starts 0.0001
$100 increase in price corresponds to a
Price (in Hundreds) -0.2 decrease in sales of 200 units of iPod
0.0001

Advt. does not have a statistically significant


Advertisement (in
0.7 effect on iPod sales (at 95% confidence level) 0.1234
TARPS) as the p value is greater than 0.05.
Number of songs on A1 million increase in the number of songs on
iTunes 0.4 iTunes corresponds to an increase in sales of 0.001
(in millions) 400 units of iPod

Direct Marketing Direct marketing corresponds to an increase in


Campaign (Binary 0.85 sales of 850 units of iPod (as compared to no 0.0001
Variable) direct marketing)

R² = 78%
86

© Dr. Denish Shah 43


Modeling Direct Marketing Decisions

Suppose the Electronics store owner is interested


in determining whether a customer
will purchase an iPod or not?

Response Variable: Yes/No Decision


 ‘1’ = Yes, ‘0’ = No

Predictor Variables: Customer Demographics,


Psychographics, Product attributes,
Type of direct marketing campaign
etc.

87

Logistic Regression
 Extremely popular and powerful technique for analyzing individuals
across two mutually exclusive and exhaustive categories.
 Example: buy-not buy, respond-not respond to a campaign, etc.

 The logistic function “maps” or “translates” the value of the


response variable to vary from “0” to “1 ” instead of an unrestricted
range that could possibly go from “– infinity” to “+ infinity”

88

© Dr. Denish Shah 44


The Underlying Math (not for exam)

Conversion of the unrestricted continuous range (of Y) to


a restricted range (that will lie between 0 and 1):

Prob(Y=1) = exp(Y)
1+ exp(Y)

89

Illustration
Dataset A = Used for building the model
Response Income Age (years)
(1 = yes; 0 = no ($000)
1 96 22
1 86 33
LOGISITC
1 64 55 REGRESSION
1 60 47
1 26 27
0 98 48
0 62 23
0 54 48
0 38 24
0 26 42 Logistic Regression Model Output

Variable df Parameter
Interpretation: Estimate
The probability of a customer to respond INTERCEPT 1 -0.9367
increases with income and decreases with age. INCOME 1 0.0179
AGE 1 -0.0042

90

© Dr. Denish Shah 45


Summary of Quantitative Techniques
discussed in today’s class

Type of Relationship

Dependence Interdependence

Machine Factor Cluster


Regression RFM CLV
Learning Analysis Analysis

Linear

Logistic

91

Application of Popular Quantitative Techniques for


Customer Segmentation & Targeting

Primary Direct & Database Quantitative Technique(s) Remarks


Marketing Task Applicable
Typically used to segment
1. Customer Segmentation Cluster Analysis markets
and/or reduction of data
dimensions based on
Typically used to reduce the
available data comprising of
Factor Analysis dimensions of the data and/or
several variables meaningfully group variables

RFM Can help assign a


CLV quantitative/$ value to each
customer and hence guide
2. Customer selection for customer-level marketing
various DM campaigns decisions
and/or communication Regression Can help predict or explain
Logistic Regression customer behavior with
Machine Learning respect to response to
different DM initiatives

92

© Dr. Denish Shah 46


Application of Popular Quantitative Techniques for
Customer Segmentation & Targeting

Primary Direct & Database Quantitative Technique(s) Remarks


Marketing Task Applicable
Known to give the most
accurate prediction results.
1. Improve prediction, obtain
fast (real-time) analyses,
Machine Learning The computation power is
and/or uncover patterns from increasing over time with
large volume of data technological advances.

93

END OF COURSE

THANK YOU!!

www.linkedin.com/in/denishshah/

© Dr. Denish Shah 47

You might also like