BBA 304 SLM

DR.
BABASAHEB AMBEDKAR
OPEN UNIVERSITY
BBA
BACHELOR OF BUSINESS ADMINISTRATION
BBAR-304
Business Analytics
Business Analytics
ISBN 978-93-91071-43-1
Editorial Panel
Authors
Mr. Ankur Sharma
Senior Manager, TCS Limited,
Gandhinagar
Dr. Ashvin R. Dave

Professor and HOD,
Departmentof Business Administration & Commerce,
School of Liveral Studies, Pandit Deendayal Energy University. Gandhinagar
Editor
Dr Ashish Joshi
Associate Professor,
Pandit Deendayal Petroleum University,
Gandhinagar
Language Editor
Dr. Jagdish Anerao
Associate Professor,
Smt AP Patel Arts & NP Patel Commerce College, Naroda, Ahmedabad
Edition : 2021
Copyright © 2021 Knowledge Management and Research Organisation.

All rights reserved. No part of this book may be reproduced, transmitted or
utilized in any form or by a means, electronic or mechanical, including photo
copying, recording or by any information storage or retrieval system without
written permission from us.
Acknowledgment
Every attempt has been made to trace the copyright holders of material re–
produced in this book. Should an infringement have occurred, we apologize
for the same and will be pleased to make necessary correction/amendment in
future edition of this book.
The content is developed by taking reference of online and print publications
that are mentioned in Bibliography. The content developed represents the
breadth of research excellence in this multidisciplinary academic field. Some
of the information, illustrations and examples are taken "as is" and as available
in the references mentioned in Bibliography for academic purpose and better
understanding by learner.
ROLE OF SELF INSTRUCTIONAL MATERIAL IN DISTANCE
LEARNING
The need to plan effective instruction is imperative for a successful distance

teaching repertoire. This is due to the fact that the instructional designer, the
tutor, the author (s) and the student are often separated by distance and may
never meet in person. This is an increasingly common scenario in distance
education instruction. As much as possible, teaching by distance should
stimulate the student's intellectual involvement and contain all the necessary
learning instructional activities that are capable of guiding the student through
the course objectives. Therefore, the course / self–instructional material are
completely equipped with everything that the syllabus prescribes.
To ensure effective instruction, a number of instructional design ideas are
used and these help students to acquire knowledge, intellectual skills, motor
skills and necessary attitudinal changes. In this respect, students'
assessment and course evaluation are incorporated in the text.
The nature of instructional activities used in distance education self–
instructional materials depends on the domain of learning that they reinforce
in the text, that is, the cognitive, psychomotor and affective. These are further
interpreted in the acquisition of knowledge, intellectual skills and motor skills.
Students may be encouraged to gain, apply and communicate (orally or in
writing) the knowledge acquired. Intellectual– skills objectives may be met
by designing instructions that make use of students'prior knowledge and
experiences in the discourse as the foundation on which newly acquired
knowledge is built.
The provision of exercises in the form of assignments, projects and tutorial
feedback is necessary. Instructional activities that teach motor skills need to
be graphically demonstrated and the correct practices provided during
tutorials. Instructional activities for inculcating change in attitude and behavior
should create interest and demonstrate need and benefits gained by adopting
the required change. Information on the adoption and procedures for practice
of new attitudes may then be introduced.
Teaching and learning at a distance eliminates interactive communication
cues, such as pauses, intonation and gestures, associated with the face–to–
face method of teaching. This is particularly so with the exclusive use of
print media. Instructional activities built into the instructional repertoire
provide this missing interaction between the student and the teacher.
Therefore, the use of instructional activities to affect better distance teaching
is not optional, but mandatory.
Our team of successful writers and authors has tried to reduce this. Divide
and to bring this Self Instructional Material as the best teaching and
communication tool. Instructional activities are varied in order to assess the
different facets of the domains of learning.
Distance education teaching repertoire involves extensive use of self–
instructional materials, be they print or otherwise. These materials are
designed to achieve certain pre–determined learning outcomes, namely goals
and objectives that are contained in an instructional plan. Since the teaching
process is affected over a distance, there is need to ensure that students actively
participate in their learning by performing specific tasks that help them to
understand the relevant concepts. Therefore, a set of exercises is built into
the teaching repertoire in order to link what students and tutors do in the
framework of the course outline. These could be in the form of students'
assignments, a research project or a science practical exercise. Examples of
instructional activities in distance education are too numerous to list.
Instructional activities, when used in this context, help to motivate students,
guide and measure student's performance (continuous assessment).
PREFACE
We have put in lots of hard work to make this book as user–friendly as
possible, but we have not sacrificed quality. Experts were involved in
preparing the materials. However, concepts are explained in easy language
for you. We have included many tables and examples for easy understanding.
We sincerely hope this book will help you in every way you expect.
All the best for your studies from our team!
Business Analytics
Contents
BLOCK 1 : BUSINESS ANALYTICS FUNDAMENTALS

Unit 1 : Introduction to Business Analytics
Introduction, Importance of Data and its Sources, Why has
Data Suddenly become so Important, Different Sources of Data
Accumulation in the Personal and Business World, Life Cycle
of Business Analytics Process, Scope of Business Analytics –
Where Does it Fit on Business Canvas, Classification of Business
Analytics, Descriptive Analytics, Diagnostic Analytics, Predictive
Analytics, Prescriptive Analytics, Challenges in Business
Analytics
Unit 2 : Descriptive Analytics
Introduction, Introduction to Descriptive Statistics, Different
Type of Data Measurement Scales, Categorical Data, Continuous
Data, Population and Sample Size, Components of Descriptive
Statistics, The Measures of Central Tendency, Measures of
Variation
Unit 3 : Visualization Techniques for Business Analytics
Introduction, Introduction to Data Visualization, Histogram,
Bar Chart, Scatter Plot, Box Plot, Control Chart, Tree Map
BLOCK 2 : STATISTICAL CONCEPTS AND HYPOTHESIS

TESTING
Unit 1 : Discrete Probability Distributions
Introduction, Random Experiments and Probability Distributions,
Discrete Probability Distributions, Binomial Distributions,
Poisson Distribution
Unit 2 : Continuous Probability Distributions
Introduction, Probability Density Function, The Normal
Distribution, Binomial Distributions, Poisson Distribution,
Student's t–Distribution, PDF and CDF for t–Distribution,
Properties of t–Distribution
Unit 3 : Sampling and Confidence Intervals
Introduction, Introduction to Sampling Process, Important Steps
in Designing a Sampling Strategy, Sampling Methods,
Probabilistic Sampling Methods, Non-Probabilistic Sampling
Methods, Central Limit Theorem, Confidence Interval
Unit 4 : Introduction to Hypothesis Testing
Introduction, Life Cycle of Hypothesis Testing, Hypothesis
Testing Process Steps, Hypothesis Test Statistics, Two–Tailed
and One–Tailed Hypothesis Test, Concept of p–Value, Type
I, Type II Error and Power of the Hypothesis Test, Hypothesis
Testing for a Population Mean with Known Population Variance
: Z–Test, Hypothesis Testing for a Population Mean with
Known Population Variance : t-Test
BLOCK 3 : CORRELATION AND REGRESSION

Unit 1 : Covariance and Correlation Analysis
Introduction, Covariance : Statistical Relationship between
Variables, Mathematical Interpretation of the Covariance,
Relationship between Covariance and Variance, Covariance
Matrix, Relationship between Covariance and Correlation,
Spearman Rank Correlation
Unit 2 : Simple Linear Regression
Introduction, Essence of Simple Linear Regression, Introduction
to Simple Linear Regression, Determining the Equation of the
Linear Regression Line, Baseline Prediction Model, Simple
Linear Regression Model Building, Ordinary Least Square
Method to Estimate Parameters, Calculation of Regression
Parameters, Interpretation of Regression Equation, Measures
of Variation, Comparison of Two Models, Coefficient of
Determination, Mean Square Error and Root Mean Square
Error (Standard Error), Simple Linear Regression in MS Excel,
Residual Analysis to Test The Regression Assumptions
Unit 3 : Multiple Linear Regression
Introduction, Essence of Multiple Linear Regression,
Introduction to Multiple Linear Regression, Understanding the
Concept of Multiple Linear Regression with a Worked Example,
The Correlation Coefficient for Multiple Linear Regression,
Coefficient of Coefficient (R2), Adjusted R2, and Standard
Error, Multiple Linear Regression in MS Excel, The Modified
Regression Model in Excel, Residual Analysis to Test the
Regression Assumptions
BLOCK 4 : TIME SERIES ANALYSIS

Unit 1 : Introduction to Forecasting Techniques
Introduction, Forecasting : Magical Crystal Ball of Statisticians,
Time–Series Data and Components of Time–Series Data, Time–
Series Data Modelling Techniques, Additive Model of Time–
Series Modelling, Multiplicative Model of Time–Series
Modelling, Measuring Forecasting Accuracy Techniques, Mean
Absolute Error (MAE) / Mean Absolute Deviation (MAD),
Mean Absolute Percentage Error (MAPE), Mean Square Error
(MSE), Root Mean Square Error (RMSE), Factors Affecting
Forecasting Accuracy
Unit 2 : Moving Average and Single Exponential Smoothing
Techniques
Introduction, Parts of Forecasting Techniques, Naïve Forecasting
Models, Averaging Models, Simple Averages, Moving Averages,
Weighted Moving Averages, Single Exponential Smoothing
Forecasting Technique, Single Exponential Smoothing
Forecasting Technique in MS Excel
Unit 3 : Regression Methods for Forecasting
Introduction, Forecasting Techniques with a Trend, How to
Draw Trendline and Regression Equation in Time–Series Graph,
Double Exponential Smoothing Constant Technique for
Forecasting
Unit 4 : Auto–Regression (AR) and Moving Average (MA)
Forecasting Models
Introduction, Introduction to Autocorrelation, Reasons for
Autocorrelation, Impact of Autocorrelation on a Regression
Model, Ways to Detect Autocorrelation : Durbin Watson Test,
Autoregression : Remedy to Resolve Autocorrelation, Moving
Average Model MA(q)
Dr. Babasaheb BBAR-304
Ambedkar
OpenUniversity
Business Analytics
BLOCK-1 BUSINESS ANALYTICS FUNDAMENTALS
UNIT 1
INTRODUCTION TO BUSINESS ANALYTICS
UNIT 2
DESCRIPTIVE ANALYTICS
UNIT 3
VISUALIZATION TECHNIQUES FOR BUSINESS ANALYTICS
BLOCK 1 : BUSINESS ANALYTICS FUNDAMENTALS
Block Introduction
Every organization despite its size needs to measure important

business metrics like profit, sales, cost, market share, return on investment,
customer satisfaction, employee satisfaction, etc. It is utterly important to
identify relationships among these metrics and important factors which
impact these metrics directly or indirectly. Analytics is a discipline that is
made up of statistics, mathematics, and computer science. It helps us to
analyze these metrics periodically and convert all data into business
information which leads to robust decision making.
Business analytics has been used in different industries for several

decades but in the last 20 years company's dependencies on analytics have
increased exponentially. There are four main reasons for this sharp rise :
• Advanced software techniques are available e.g. advanced data

structures, advanced database systems, cloud computing, etc.
• Clean data is available, now most organizations have robust software

infrastructure which helps in capturing clean customer, vendors, and
sales data
• Advanced hardwares are available which can store huge data in such
a way that it can be easily available for analysis without any time lag.
Also, the cost is quite reasonable e.g. GPU/TPU processors, distributed
networks, etc
• Advanced business problem-solving techniques are providing new

alternatives to tackle business problems e.g. Agile and lean six sigma
frameworks for business excellence
Every hour organizations take several decisions which decide their

fate whether they will be profitable or suffer losses. Studies reveal that
organizations that take more data-backed decisions are more probable to
incur high profits. With the help of advanced analytics software and clean
input data, most of these decisions are taken by systems that make processes
more efficient and responsive. Forbes magazine stated in its famous report
that the main difference between successful and not so successful
organizations is that how much they use analytics in their decision making.
Block Objectives
After learning this block, you will be able to understand :
• Understanding how Business Analytics is changing the decision–

making process in the business world
• Different components of business analytics
• Classifications and challenges of business analytics
• Basic concepts of descriptive analytics
• Data types and scale of measurements
• Learning various types of visualization techniques
• Interpretation of visualization techniques
• Understanding the most appropriate visualization technique as per the

scenario
Block Structure
Unit 1 : Introduction to Business Analytics
Unit 2 : Descriptive Analytics
Unit 3 : Visualization Techniques for Business Analytics

Unit INTRODUCTION TO
1 BUSINESS ANALYTICS
: UNIT STRUCTURE :
1.0 Learning Objectives
1.1 Introduction
1.2 Importance of Data and its Sources
1.2.1 Why has Data Suddenly become so Important
1.2.2 Different Sources of Data Accumulation in the Personal
and Business World
1.3 Life Cycle of Business Analytics Process
1.3.1 Scope of Business Analytics – Where Does it Fit on Business
Canvas
1.4 Classification of Business Analytics
1.4.1 Descriptive Analytics
1.4.2 Diagnostic Analytics
1.4.3 Predictive Analytics
1.4.4 Prescriptive Analytics
1.5 Challenges in Business Analytics
1.6 Let Us Sum Up
1.7 Answers for Check Your Progress
1.8 Glossary
1.9 Assignment
1.10 Activities
1.11 Case Study
1.12 Further Readings
1.0 Learning Objectives :

After learning this unit, you will be able to understand :
• Understanding how Business Analytics is changing the decision–
making process in the business world
• Different components of business analytics
• Evaluation of business analytics and how it has become the new
language in the business world
• Classifications of business analytics
• Challenges of business analytics
1
Business Analytics 1.1 Introduction :
In this unit we will study basic institution about business analytics
and how it is changing the overall canvas of business world. At the end
of the unit, you will understand how different components of business
analytics are influencing decision–making capabilities. We will also see
different type of business analytics and important tools and techniques
used in each type of business analytics. We will also touch upon various
challenges organizations are facing in current era and what are different
approaches to overcome these challenges.
1.2 Importance of Data and its Sources

"Data is what you need to do analysis. Information is what you
need to do business." – John Owen
Today Businesses speak the language of analytics. We see a flood
of analytics jargon in all business presentations. Analytics has covered
a long journey from simple number crunching to solving complex business
problems to create a competitive business strategy. In the last two decades,
mostly all big and mid–size organizations have started business analytics
as one of the primary functions, and that's why business analysts hold
critical and well–paid roles in these organizations. World–renowned
magazine, HBR mentioned it in one of the articles as "The Sexiest Job
of the 21st Century".
When we apply structured and scientific tools/ approaches to
convert raw data into meaningful business information which leads
to better Business Decisions, we call it Business Analytics.
Fig 1.1 Components of Business Analytics

 Business Analytics is made up of Three Crucial Components :
1. Technology : In Business Analytics technology plays a significant
role in capturing a large amount of complex data, sharing it
simultaneously with different geographies, streaming it from various
2
sources, e.g. social media, sales systems, customer relationship Introduction to
management systems. Technology also helps in real–time data Business Analytics
backup.
2. Business Domain Knowledge : Analytics projects always revolve
around domain knowledge. An analyst with sound domain knowledge
will have a great knack for asking the right questions; it helps them
in selecting relevant data and the right tools. This would finally
result in a good analytics storyboard.
3. Data Science : This is the heart of Business Analytics; it generally
consists of statistical and machine learning concepts. It starts with
the right tools/ approaches to framing the right business problem
post that helps in analyzing data and drawing conclusions out of
that.
1.2.1 Why has Data Suddenly become so Important :
Data is becoming the most critical commodity in today's world.
It is not limited to the business world but also equally important in our
personal life. In today's boundaryless business world, there is a famous
saying, "The last two decades were dedicated to software while the next
two decades will be dedicated to Statistics". It means that in the last
two decades, all types of organizations irrespective of their size and nature
of business got the software installed in their vital departments. Even
shops in our locality like grocery stores, medical shops, hardware stores,
etc. also get the software installed to manage their inventory and customer
information. This software is collecting important data about sales,
inventory, customer demographics, vendor details, etc. But it is also
raising a huge question in front of the world about how to convert this
raw data into some meaningful information that can be further used in
strategy formation and better decision making. This is where Business
Analytics comes as a saviour. For example, if we download the feedback
report from the software then it is data but if we see that feedback of
a particular region is relatively low then it becomes information.
Management will initiate root cause analysis for finding reasons for low
feedback scores.
On the other side, the next generation of softwares are available,
which can store a very complicated and massive amount of data. Below
are a few examples of complicated or complex data :
• Emoticons, likes, and pictures, videos from social media sites
• Real–time data gathering from various points of sales at different
locations (with the help of Big Data technologies)
• Data is getting updated in real–time across geographies (with the
help of cloud technologies)
• Customer feedback from Facebook or Twitter in the form of raw
texts with wrong spellings ?
• Customer demographics and sales related data from Customer
Relationship Management (CRM) software 3
Business Analytics As currently softwares are installed in all important departments
of the organization, these softwares capture data at a very granular level.
This detailed and clean data is available for processing and analysis with
the help of recent developments in the computational power of computers
and advanced machine learning algorithms. Hence raw data is getting
converted into meaningful information on a real–time basis, which is
further consumed by decision–makers to get a competitive edge. In this
way, data has become one of the most important currencies in the business
world. Today most successful business leaders are trying to infuse analytical
capabilities throughout their organizations to start the culture of taking
data–based decisions, it also helps in optimizing business outcomes such
as maximum revenue with minimum resources, passing benefits to their
customers.
1.2.2 Different Sources of Data Accumulation in the Personal and
Business World :
 Data Sources in Personal Life :
If you are not paying for it, you're not the customer; you are the
product being sold. – Commentator Blue Beetle
We use various software in our daily life, which captures our
personal information, e.g. Facebook, Google, YouTube, Instagram, LinkedIn,
etc. These softwares get access to our personal information like places
we visit, brands we endorse, TV/radio channels we choose as entertainment/
information sources, type of blogs/ books we read, people we meet, etc.
Today softwares are the integrated part of most of the devices we
use in our personal life. This also raises the risk that our personal data
can be used for unethical purposes. Below are a few examples :
• Mobile Phones : When we install various apps, unknowingly we
accept their terms and conditions to share our contact lists, to read
our messages, to trace our locations, etc. If this data gets leaked,
then it may be dangerous for our personal and monetary security,
e.g. we have bank details, OTPs, password in our messages which
can be misused
• Health Fitness–Related Devices : These devices get access to our
locations, our diet, our sleeping and exercise patterns, etc
• TV Setup Boxes : Setup boxes get data about our entertainment
quotient, e.g. which channels we watch and for how long. They
may use this data to make customized offers for us or organizations
may share this data with their vendors for marketing purpose
• Smart Watches : Smart watches also have most of the features
of mobile phone hence above data problems may occur in the case
of smartwatches also
• Social Media Sites : Websites/ apps like YouTube/ Facebook/
Twitter/ Instagram/ LinkedIn get access to our personal information
like family photos, our views on various matters/ issues, our likes
4
or dislikes about multiple brands/ products. They also share this Introduction to
data with their customers for commercial purposes Business Analytics
These examples enlighten us about how data is being monetized
and prove that "Data is one of the most important currencies of today's
world".
Fig 1.2 Sources of Data Accumulation in Personal Life

5
Business Analytics  Data Sources in Business World :
Data is the primary fuel in the business world as all planning and
execution activities at different phases of the business life cycle generate
and consume a large amount of data. Information about customers,
vendors, competitors, employees, feedback plays an essential role in
decision making in the business world.
Fig 1.3 Sources of Data Accumulation in the Business World

6
1.3 Life Cycle of Business Analytics Process : Introduction to
Business Analytics
Business Analytics projects start with
a correctly framed business problem; analysts
convert this business problem into an
analytical problem so that problem becomes
measurable and its impact can be understood
by management in terms of money, time,
and resources. As per the nature of the
problem, analysts figure out relevant data
and tools to solve the statistical problem. In Fig 1.4 Business
the end, they again summarise their statistical Problem-Solving Process
findings in terms of Business Solutions which can be easily understood
by business executives and converted into a sustainable and replicable
solution.
The high–level data–driven business analytics process is mapped
in fig (1.5). These are very high–level guidelines; organizations use this
process as per their requirements and customize it accordingly. It is an
iterative process where we complete one analytics initiative and raise
the bar to pick up a challenge of the next level to bring more value
to business and stakeholders.
Fig 1.5 Life Cycle of Solving a Business Problem

In today's era of globalization, organizations are getting diversified
in terms of lines of business and spreading their business across
geographies. At that scale of business without a properly structured
approach to execute analytics, it would be impossible to bring efficiency
and effectiveness in their business operations. It leads to value generation
to essential stakeholders. Below is the high–level description of the
business analytics process steps :
1. Identify the opportunity for improvement to create value for business
2. Select significant sources to gather relevant data for analysis with
the help of the right set of tools, here we also clean the data and
put it in the right format
3. Post validation of data, we use suitable visualization which is easy
to understand and also convey vital information about current
business scenarios
7
Business Analytics 4. Based on prior steps we take appropriate decisions by keeping
budget, time, and resources in mind also predict the output in the
short and long run
5. To get sustainable solutions, we put suitable controls in the system
which results in optimizing outputs and helps the organization in
creating a competitive strategy
1.3.1 Scope of Business Analytics – Where Does it Fit on Business
Canvas :
Business Analytics works like a magical crystal ball that can solve
tiny day–to–day problems like reducing packaging errors to enormous and
complex business problems such as designing a space shuttle or setting
up a nuclear power plant.
Let's try to understand the applications of Business Analytics at a
different level with the help of fig 1.6
At the bottom, more and
more workforce get involved in
Business Analytics initiatives, and
they identify smaller problems in
their day–to–day operational
activities with the help of quick
applications of analytics, here
analysts help them to find
solutions and incorporate these
solutions into their way of
working. In business, we call it
Kaizen, which is a Japanese word
that means "change for good". In
the next level generally,
specialized analysts get involved Fig 1.6 - Scope of Business
in solving critical business Analytics at Different Levels
problems; typically, they tackle these in terms of business projects. The
next level is Decision Analytics, where departmental heads get engaged
in executing more significant business initiatives, here the analytics team
uses complex tools and massive data for impactful and detailed analysis.
At the top–level leadership gets involved in using comprehensive analytics
for strategy formation.
Check Your Progress – 1 :
1. Which of the below tasks can be handled by business analytics ?
a. Predicting business results
b. Finding patterns in the data
c. Validating business assumptions
d. All of the above
8
2. The business analysis process starts with : Introduction to
a. Analysis of data Business Analytics
b. Collecting data
c. Determine the need of the process
d. Predict the business outcomes
1.4 Classification of Business Analytics :

"Errors using inadequate data are much less than those using no
data at all". – Charles Babbage
Business analytics is a structured approach that brings value to the
business in a very systematic way. It is always great to start with the
basics and build analytical capabilities step by step. Analytics can be
classified into four levels which help the organizations to become mature
in terms of analytical proficiency. Below are the brief descriptions :
1. Descriptive Analytics : This
is the simplest form of
analytics; it summarises
current business status in the
way of narrative and
innovative visualization. It
emphasizes "what is going
on in the business."
2. Diagnostic Analytics : It
provides the reasons for
descriptive analytics;
generally, analysts provide Fig 1.7 Classification of
visual reasoning in terms of Business Analytics
interactive dashboards. It
emphasizes "why did it happen."
3. Predictive Analytics : It predicts the business metrics in the short
and long run, here we use advanced machine learning algorithms
to analyze massive data from different sources to predict the future
values of essential business metrics. It emphasizes "what will
happen in the future".
4. Prescriptive Analytics : It suggests all favourable business outputs
for any specified course of action, it also offers the pros and cons
for each course of action. It optimizes the business results suggested
by descriptive and predictive analytics. It emphasizes "how can we
make it happen."
Among these different types of analytics, there is no superior or
inferior to each other, but all play a significant role during different phases
of solving a business problem. These levels of analytics are sequential
and linear in nature which means an organization cannot implement
diagnostic or predictive analytics without having proper descriptive analytics
9
Business Analytics implemented throughout the organization. There is always a strong
relationship between levels of analytics used Vs business value derived.
But we need to have useful descriptive analytics to start and later
emphasize diagnostic analytics which provides a base for Predictive
Analytics and finally, prescriptive analytics tells us the way to optimize
solutions recommended by all three primary types of analytics. Let's
discuss them in detail with a few examples and tools involved :
Fig 1.8 Types of Business Analytics

1.4.1 Descriptive Analytics :
If the statistics are boring, then you've got the wrong numbers.
– Edward R. Tufte
It is the most simplistic form of analytics; it digs deeper into past
data and tells us "what has happened and when did it happen". Here,
the main objective is to summarize data into useful insights with the help
of easily interpretable visualization techniques. It highlights past trends
that lead to valuable insights for business, but we do not emphasize here
"why these trends happened". Here we try to connect the dots to make
an exciting analytics storyboard, which helps us understand the problems
and their impacts on business and stakeholders.
Descriptive statistics generally used to show the performance of an
organization against its leading Key Performance Indicators (KPIs). It also
indicates trend analysis for these KPIs against competitors and among
different geographies. Information obtained during descriptive analytics
can be used to make better business decisions.
We use Descriptive Analytics when we want to summarize the
story of an organization's performance (mostly in the form of
10
Dashboards). It provides us with a comprehensive view by joining Introduction to
different things together to highlight hidden trends and insights. Business Analytics
Information extracted from descriptive analytics helps leadership
to take actions to make things better, and now with the help of Big Data
technologies, management sees the real–time progress of various vital
business metrics. Management sees a complete picture by benchmarking
company performance against the past few years and key competitors.
Below are a few examples of knowledge extracted from descriptive
analytics :
Fig 1.9 Sample Business Dashboard

1. Men convert credit card transactions into EMI more than women;
banks should target men for EMI promotion as they are more likely
to opt for the promotional campaign.
2. Internet routers show lots of information packets drop during
4–6 PM due to high congestion, support team to provide extra
bandwidth during this time slot for seamless customer experience.
3. More cars come for servicing during monsoon due to water problems
so garage should think about hiring part–time mechanics during
monsoon to cater to the temporary demand.
4. Analysts find a strong correlation that customers quit websites on
the checkout screen and customers who use mobiles to browse
company website; it indicates customers do not see website mobile
friendly hence the company should launch a mobile version of
company website soon.
5. The health department observes a recurring hike in malaria disease
in a particular locality every year during the rainy season; they find
water bodies are open in that area which is causing mosquito
breeding.
 Essential Tools used in Descriptive Analytics :
1. Statistical Summary : It provides statistical descriptions for a given
business metric, e.g. Mean, Median, Standard Deviation, Percentile,
Interquartile range, etc. (will study these techniques in detail in
upcoming chapters).
11
Business Analytics 2. Z–Score : Z Score tells us how far (in terms of standard deviation)
is a particular value of x from its mean.
3. Coefficient of Variance : It is a ratio where we divide standard
deviation with mean. Alone mean or standard deviation are not
appropriate methods to measure to benchmark different company
performance metrics. It is important to consider both centrality and
spread of data to make it comprehensive.
4. Interquartile Range : It is an important measure to gauge the
variation in the dataset. The height of the interquartile box is the
difference between the third and first quartile of data. It is quite
powerful as it removes very small and very big data points.
1.4.2 Diagnostic Analytics :
"There is wisdom in always exploring the counterpoint – sometimes
a silver cloud has a dark lining too." – Gyan Nagpal
Diagnostic analytics
provides "Why did it happen
in my business". It is a bit
advanced where analysts
examine data in order to find
reasons for business
problems or opportunities.
One of the ways is drawing
correlation among various
business metrics as
sometimes changes
happening in one metric lead
to change in other metrics,
e.g. reduction in production
because of higher
absenteeism in that week or
drop in quality because of Fig 1.10 5-Why Diagnostic Analysis
new training batch going live in the previous week as new guys make
more errors than tenured one. There are numerous ways to perform causal
analysis. Below are a few examples :
1. A company found that employees are not completing learning
certifications, analyst diagnosed that most of the employees are stuck
at programming assignments, where programming interface was not
supportive/ flexible, and there was no way to get hints/ help to
proceed further.
2. There was a low hotel check–in feedback score; analysts diagnosed
that front office executive enters customer details which are not
required fields during check–in itself. Typing speed and system
navigation is also very slow which is resulting in a longer check–
in time.
12
3. The product return rate was very high during last month, and it Introduction to
found that out of total return items more than 60% of products Business Analytics
were supplied by two vendors only, where the vendor provided the
wrong specification about products.
 Essential tools used in Descriptive Analytics :
1. Correlation Analysis : It is a statistical measure that indicates the
strength of the relationship between two variables. It is a critical
causal analysis technique that helps in identifying reasons in terms
of relationship with other metrics.
2. 5 Why Analysis : It is a very structured approach where we try
to dig into a problem and peel it layer by layer to reach the root
cause of the problem. Solutions to root cause provide us with
sustainable solutions.
3. Cause and Effect Analysis : Here, we identify all possible reasons
for one problem then we pick up all the reasons as a problem one
by one and try to find other causes for that problem. In this way,
we create a diagram that looks like the skeleton of a fish because
of its looks. It is also known as the fishbone diagram.
1.4.3 Predictive Analytics :
"The future belongs to those who see possibilities before they
become obvious". – John Scully
Predictive analytics is the heart of business analytics, it aims to
help the organization by predicting probabilities of occurrence of a future
event or future values of any essential business metrics, e.g. sales in
the next month/ quarter, employee attrition, and product return rate, etc.
Once organizations have a
stable setup for descriptive
analytics, means data sources have
been identified, and those data
sources are supplying data about
important metrics continuously
into leadership dashboards.
Predictive analytics combines this
historical data with advanced
business protocols (policy and
rules) to forecast future values of Fig 1.11 Predictive Analytics Cycle
business events. Predictive analytics allows organizations to become
forward–looking, providing an appetite to consume calculated risk by
anticipating customer behaviour and business outcomes.
 Below is a List of Predictive Analytics Examples :
• Financial analysts predict the share value/ gold prices/crude oil
prices in the next few days or weeks with the help of predictive
modelling.
13
Business Analytics • Airline companies predict competitive airfares to extraordinary and
ordinary days also they indicate how much airfare should be increased
as per the increased customer's traffic on their websites.
• Netflix predicts the next movie customers want to watch, more than
80% of customers select their next movie from their recommendation
list. In this way, Netflix earns more rental income from regular
customers by suggesting them the next film or programs.
• IRCTC predict the probability to confirm the seat which provides
assurance to the customer about their seat confirmation, it helps
to attract more customers to their portal.
• Taxi services predict the demand during different time slots and
change their tariff accordingly.
Fig 1.12 Example of Forecasted Demand

 Important Tools used in Predictive Analytics :
1. Regression Analysis : It establishes the mathematical relationship
between input variables and output variables, which means if we
can calculate the future value of output for any given input, e.g.
sales forecast for next month.
2. Logistic Regression : It is a classification predictive analytics
technique that can predict the output class for any given set of
inputs. E.g. by providing customer demographics logistic regression
can indicate whether the customer will default bank loan in the
future or not.
3. Decision Tree : Most of the time, we use a decision tree as a
classification technique; it tells us the output probability of the
output variable for various permutations of our input variables.
Although it can be used for continuous output variables also.
14
4. Clustering Techniques : These techniques segregate our customers Introduction to
into a few logical segments so that we can create tailored offers Business Analytics
for a different type of customers as per their needs and interests.
5. Random Forest : It is another very famous business analytics
technique that uses a collaborative approach to solve the problem
by generating a large number of predictive models. Their accuracy
is generally better.
1.4.4 Prescriptive Analytics :
"Without prescriptive analytics, you are navigating with rear–view
mirror". – Martin Zync
Prescriptive analytics
solves the complex
business problem as it is
the most advanced form
of analytics, where we
have to choose the most
optimal way to increase
important business
metrics. As we discussed
earlier, all type of Business
Analytics is sequential;
therefore, perspective
analytics can be applied Fig 1.13 Prescriptive Analytics
once we have sound business knowledge from descriptive and predictive
analytics.
Descriptive and predictive analytics suggest to us various ways to
improve business performance while prescriptive analytics tells us the
pros and cons of all alternatives and try to provide the optimal outputs
by keeping minimum risk in execution.
Prescriptive analytics is not limited to predict "what will happen"
and "when will it happen" but it also tries to reveal "why it will happen"
and "what would be the impact on the business".
It relies on slightly complex machine learning algorithms, and its
success depends on the organization's maturity to ask the right questions;
otherwise, it may completely reverse the business outcomes, e.g. if
because of complex prescriptive analytical model banks make the loan
approval process too tricky then they may lose several potential customers.
 Below are Examples of Prescriptive Analytics :
• In 2019, there was a prediction of the cyclone on coastal areas
of Gujarat (by predicting changing airspeed, varying wind direction,
and mathematical relationship between low pressure in the ocean
with changes in cyclone intensity) therefore Government and disaster
management team had taken proactive actions in shifting citizens
from coastal areas to save places, and they stopped fishermen from
15
Business Analytics going to sea and arrange comfortable camps. While in a similar
situation in 1999 we lost approx. 10,000 lives due to cyclone.
• At the time of launching a new service or a product into the market,
organizations have to keep various factors into the mind like the
cost of the product, features of the product, geographies in which
they will launch first, customer segments whom they want to attract,
marketing channels for product promotion, etc. By getting analytical
results from descriptive and predictive analytics, analysts apply
prescriptive analytics to decide the right mix of all these factors
to make a product launch successful.
• In agriculture crop yield depends on various factors like rainfall,
soil type, demand in the market, etc. Analysts apply prescriptive
analytics and suggest the best kind of crop in different regions as
per the rainfall and demand forecast in that season.
• Banks use prescriptive analytics to identify investment options for
their customers to maximize their returns and minimize risk. They
balance customer's portfolio by having an optimized ratio of equity,
debt, and other types of funds.
 Important Tools used in Prescriptive Analytics :
1. Linear Programming : In linear programming, we optimize the
objective functions like revenue, market share, customer feedback
ratings by also keeping constraints in the model like budget, no.
of people deployed, etc. as linear functions.
2. Analytical Hierarchy Process : We apply these techniques in
scenarios where we have to identify the best solution among various
available options, and there is the list of criteria's to select the
solution, e.g. select best cloud service providers among top 5
organizations by keeping multiple factors into consideration like
budget, customer service, flexibility to upgrade, backup services,
maintenance cost, etc.
3. Combinational Optimization : It involves identifying optimal
solutions from a considerable number of finite solutions, e.g. the
travelling salesman problem, vehicle routing problem, etc.
1. Which type of business analytics generally uses statistical and
machine learning algorithms.
a. Descriptive b. Predictive c. Prescriptive d. Diagnostic
2. Which type of business analytics provides recommendation about
optimization of business outcomes, includes simulation, etc.
3. Which type of business analytics gain information from historic data
and provide us various reports, visualization, and scorecards, etc.
16
1.5 Challenges in Business Analytics : Introduction to
Business Analytics
Establishing profitable business analytical capabilities in any
organization requires complete cultural changes; it is one of the most
challenging tasks for leadership. Below is the list of few crucial challenges
which organizations are facing today :
• Management is comfortable in taking decisions based on their
experience and learning, and due to limited statistical knowledge
they hesitate in adapting scientific calculations to predict the outcome
for their respective business metrics.
• Investment in technological aspects for example cost of cloud
infrastructure, licenses for data science software, cost of seasoned
business analytics, etc.
• Availability of talent pool is another big challenge, as business
analysts should have very good technical, business, and statistical
skills hence it is difficult to find good professionals with the right
combination of these skills.
• Developing in house talent so that essential functions can solve
their analytical assignment themselves without taking the help of
the centralized expert team.
• Reporting is the main time–consuming activity for middle
management, to minimize its organizations have to build a culture
of clean data feeding discipline into the systems so that reports
can be published by software without human interventions.
• Most of the softwares are available in open–source form, one side
these minimize huge cost on licenses, but the flip side is that there
is minimum or no technical support hence learning curve for these
technologies is slower than other.
Soon analytics will emerge even more vital driving force for
enhancing business performance, and leadership will move to data–centric
decisions than experiential ones alone. Finding more innovative ways to
solve traditional and new business problems will be the critical success
factor for deploying business analytics among various vital functions of
the organization.
1. Which type of business analytics emphasize "What did happen."
a. Predictive analytics b. Diagnostic analytics
c. Descriptive analytics d. Prescriptive analytics
2. The main essence of Diagnostic analytics is
a. Choosing the best action among alternatives
b. Thinking forward to avoid possible consequences
c. Splitting big problem into smaller ones
d. Finding the root cause of why something happened good or bad
17
Business Analytics 3. Which of the following is NOT an important component of business
analytics ?
a. Technology b. Data Science
c. Business Domain Knowledge d. People management
4. Out of the below analytical tool which one doesn't fall under
prescriptive analytics
a. Regression analysis b. Linear programming
c. Analytic hierarchy process d. Combinational optimization
5. Which type of analytics provides visualization of all critical business
metrics
a. Predictive analytics b. Descriptive analytics
d. Descriptive analytics d. Prescriptive analytics
6. Out of the following which level of business analytics, the maximum
workforce is involved
a. Analytics for critical problem solving
b. Decision Analytics
c. Analytics for simple process improvements
d. Strategy Formation
7. Which type of analytics helps in choosing the best possible solution
among alternatives
a. Prescriptive analytics b. Descriptive analytics
c. Descriptive analytics d. All of above
8. Out of the below analytical tool which one doesn't fall under
predictive analytics
a. Decision tree b. Clustering techniques
c. Linear programming d. Regression analysis
9. Which technique establishes a mathematical relationship between
input and output variable
a. 5–Why analysis b. Z–Score
c. Regression analysis d. Linear programming
10. Which of the following sequence is correct ?
a. Data Information Decision Optimization
b. Information Data Decision Optimization
c. Optimization Information Decision Data
d. Decision Information Data Optimization
18
1.6 Let Us Sum Up : Introduction to
Business Analytics
1. In the current era of business 4.0, clean and comprehensive data
is available in all type of organization therefore the scope of
business analytics has increased many folds.
2. Business analytics is a structured approach that brings value to the
business in a very systematic way. It is always great to start with
the basics and build analytical capabilities step by step. Analytics
can be classified into four levels which help the organizations to
become mature in terms of analytical proficiency.
3. Business Analytics projects start with a correctly framed business
problem; analysts convert this business problem into an analytical
problem (writing in terms of business metrics and quantitatively
impact on business), analysts figure out relevant data and tools to
solve the statistical problem. In the end, they again summarise their
statistical findings in terms of Business Solutions which can be
easily interpreted and converted into a sustainable and replicable
solution.
4. Descriptive analytics is the most simplistic form of analytics, it
digs deeper into past data and tells us "what has happened and
when did it happen".
5. Diagnostic analytics provides "Why did it happen in my business".
It is a bit advanced where analysts examine data in order to find
reasons for business problems or opportunities.
6. Predictive analytics aims to help the organization by predicting
probabilities of occurrence of a future event or future values of
any essential business metrics, e.g. sales in the next month/ quarter,
employee attrition, and product return rate, etc.
7. Prescriptive analytics solves the complex business problem as it
is the most advanced form of analytics, where we have to choose
the most optimal way to increase important business metrics.
1.7 Answers to Check Your Progress :

1. d 2. c
1. b 2. c 3. a)
1. b 2. d 3. d 4. a 5. b
6. c 7. a 8. c 9. c 10. a
19
Business Analytics 1.8 Glossary :
Descriptive Analytics : This is the simplest form of analytics; it
summarises current business status in the way of narrative and innovative
visualization. It emphasizes on "what is going on in the business."
Diagnostic Analytics : It provides the reasons for descriptive
analytics; generally, analysts provide visual reasoning in terms of interactive
dashboards. It emphasizes on "why did it happen."
Predictive Analytics : It predicts the business metrics in the short
and long run, here we use advanced machine learning algorithms to
analyze massive data from different sources to predict the future values
of essential business metrics. It emphasizes on "what will happen in the
future".
Prescriptive Analytics : It suggests all favourable business outputs
for any specified course of action, it also offers the pros and cons for
each course of action. It optimizes the business results suggested by
descriptive and predictive analytics. It emphasizes on "how can we make
it happen."
Z–Score : Z Score tells us how far (in terms of standard deviation)
is a particular value of x from its mean.
Coefficient of Variance : It is a ratio where we divide standard
deviation with mean. Alone mean or standard deviation are not appropriate
methods to measure to benchmark different company performance metrics.
It is important to consider both centrality and spread of data to make
it comprehensive.
Regression Analysis : It establishes the mathematical relationship
between input variables and output variables, which means if we can
calculate the future value of output for any given input, e.g. sales forecast
for next month.
Logistic Regression : It is a classification predictive analytics
technique that can predict the output class for any given set of inputs.
E.g. by providing customer demographics logistic regression can indicate
whether the customer will default bank loan in the future or not.
Decision Tree : Most of the time, we use a decision tree as a
classification technique; it tells us the output probability of the output
variable for various permutations of our input variables. Although it can
be used for continuous output variables also.
Linear Programming : In linear programming, we optimize the
objective functions like revenue, market share, customer feedback ratings
by also keeping constraints in the model like budget, no. of people
deployed, etc. as linear functions.
20
1.9 Assignments : Introduction to
Business Analytics
1. Write down important tools used in descriptive analytics.
2. Write down important components of business analytics.
3. Write down important data sources in the personal and business
world.
4. Mention the scope of business analytics at different levels in an
organization.
1.10 Activities :
Suppose you have been hired as an analyst at SDFG Bank, your
manager has provided you home loan data for the last 5 years. You have
to build up an application with the help of the Data Science team to
predict potential loan defaulters.
Write down various analyses you can do on this data from a
Descriptive, diagnostic, predictive, and prescriptive business analytics
perspective.
1.11 Case Study :

Nirav Soft drinks is the largest beverages bottler and distributor
of carbonized non–alcoholic, bottled beverages in North Gujarat, and one
of the largest bottlers in western India
How ABC Image Recognition for Retail is being used :
Before using ABC's scanning and imaging technology, Nirav Soft
drinks was dependent on limited and completely manual measurements
of products/ cold drink cans in–store, as well as delayed/ lost data sourced
from either fax or phone conversations.
Nirav Soft drinks sales representatives/ manager after using ABC
Retail Execution scanning image–based technology to take pictures of
stores shopping shelves with their hand gadgets or mobile devices; these
pictures were sent to the Trax Cloud and analysed, providing actionable
reports within few minutes to sales representatives/ manager and providing
more detailed online assessments to company management.
 Value Proposition :
Real–time images of inventory allowed the sales representative to
identify performance gaps and apply corrective actions in store. Reports
on shelf share and competitive insights also allowed reps to strategize
on opportunities in–store and over the phone with store managers/
promoters.
Nirav Soft drinks gained a 7.3% market share in the western India
region within five months.
21
Business Analytics Questions :
1. Write down the various type of Descriptive, diagnostic, predictive,
and prescriptive business analytic you can imagine for Nirav Soft
drinks ?
2. Which important business metrics will improve through this
technology intervention ?
3. How this change will bring value for store owners, Nirav soft drinks,
and end customers ?
1.12 Further Readings :

• "Mathematical Methods of Statistics," Princeton University Press,
Carmer H. (1946)
• "Super Freakonomics," Penguin Press, Levitt S. D. and Dubner
S. J. (2009)
• "An Explanation of the Persistent Doctor Mortality Association"
Journal of Epidemiology and Community Hearlth, Yough F. W.
(2001)
• "Data Strategy : How to Profit from A World of Big Data, Analytics
and The Internet of Things", O'Reilly Media, Bernard Marr
• "Predictive Analytics : The Power to Predict Who Will Click, Buy,
Lie, or Die", Wiley, Eric Siegel
22
Unit DESCRIPTIVE ANALYTICS
2
: UNIT STRUCTURE :
2.1 Introduction
2.2 Introduction to Descriptive Statistics
2.3 Different Type of Data Measurement Scales
2.3.1 Categorical Data
2.3.2 Continuous Data
2.4 Population and Sample Size
2.5 Components of Descriptive Statistics
2.5.1 The Measures of Central Tendency
2.5.2 Measures of Variation
2.6 Let Us Sum Up
2.8 Glossary
2.9 Assignment
2.10 Activities
2.11 Case Study

• Basic concepts of descriptive analytics and how it is influencing
the decision process of organizations
• Data types and scale of measurements
• Understanding measures of data centrality and variability
• Important tools and techniques for data visualization
2.1 Introduction :
In this unit we will study about descriptive statistics and different
type of measurement scales used in descriptive statistics. Basic concepts
about sampling theory and how it helps business in saving huge revenue
and time to conduct the study/ analysis. We will also understand various
techniques to measures centrality and variability of data and how these
help us to take better business decisions. At the end we will touch upon
few data visualization techniques which helped us to understand the data
and analysis better.
23
Business Analytics 2.2 Introduction to Descriptive Statistics :
"Data are just summaries of thousands of stories – tell a few of
those stories to help make the data meaningful" – Chip & Dan Heath
Descriptive analytics is the foundation of any analytical project. It
focuses on "What has happened" by visualizing the current business
performance.
We develop dashboards to showcase important business metrics. By
putting together historical performances, data helps us to see hidden
inferences which lead to better business decisions. Innovative visualization
plays an important role in displaying a comprehensive picture of
organizational progress. We can also include competitor's information
which will help management take appropriate strategic actions.
Fig 2.1 Sample Dashboard
2.3 Different Type of Data Measurement Scales :

Irrespective of the
size and nature of an
organization, data is getting
generated at a high pace,
e.g. sales data, employee
data, inventory, customer–
related data, etc. This data
comes in various forms like
numeric, text, alphanumeric
and it may be captured in Fig 2.2 Usage Level of Data
different scales, e.g. sales and year of experience both are numeric data, but
they can get measured in rupees and years respectively. Different types of
analytical tools and techniques are available for different types of data. We
divide data into two important categories :
24
Type of Data Descriptive Analytics
Categorical Continious
Data Data
Nominal Ordinal Interval Ratio

Data Data Data Data
Fig 2.3 Data Types
2.3.1 Categorical Data :
This type of data is also known as discrete data. Categorical
variables have a finite number of groups, e.g. payment method, gender,
income group, property type, etc. Categorical data can be further divided
into the following two types :
a. Nominal Data : In this type of categorical variable, there is no
logical sequence among categories, which means one category is
not superior or inferior to others. Categorization is just a type of
segregation of data into different groups. For example, people
wearing a dark colour shirt or light colour shirt, north Indian city
and south Indian city, software or textile industry, etc. One of the
limitations is that we cannot perform any mathematical operation
on nominal data.
b. Ordinal Data : It is better than nominal data in terms of usage
potential for any analysis. There is a logical sequence in terms of
the superiority among categories. Ranked or ordered data generally
come into this category. E.g. pass or fail status, feedback on Likert
scale – Good, fair, bad, the risk level of investment bonds – high
and low–risk bonds, etc. Mathematical operations are not possible
on ordinal data.
2.3.2 Continuous Data :
Data that can be measured on a continuous scale–like height,
weight, money, time, etc. It can be divided into halves any number of
times. Continuous data can be further divided into the following two
types.
a. Interval Data : Interval data is always numeric, and there is an
equal distance between consecutive interval data points. Data comes
from an interval range like e.g. temperature, percentage return of
a share, percentage change in the gold price, intelligence quotient
scores for an exam, etc. Another important point is that zero is
like another number (it does not mean that value is none or
missing). For example, a zero degree is a valid temperature. On
such data, we can do summation and subtraction as mathematical
operations, but division and multiplication wouldn't make sense.
The difference in heat between 90oC and 70oC is the same as the
difference between 50oC and 30oC, but we cannot say that 80oC
25
Business Analytics is not twice as hot as 40oC or somebody with IQ scores 70 is not
twice as smart as another student with IQ score 35.
b. Ratio Data : In terms of statistical relevance, this is the highest
level of data which is desirable for the application of statistical tools
and techniques. On such data, we can do all types of mathematical
operations. For example, if product A's sale is ` 50,000 and product
B's sale is ` 25,000, then we can interpret that sale of product A
is twice the sale of product B.
2.4 Population and Sample Size :

For a statistical problem if we have access to all possible data sets
and the entire dataset is used for analysis, then it is an analysis of
population data. Below are a few examples
• The government analyses the base of census data to calculate
important indices like per capita income, employment rate, GDP,
etc.
• The election commission uses entire voters' data to provide
information on the proportion of graduate voters, how many voted
in the last election etc. Generally, the size of the population is huge
in terms of data hence it requires large space to store and lots of
computer memory to run the analytical queries.
• Big organizations like Indian Railway, Indian Army, etc. use entire
data of their employees (which is in millions) to analyze their salary
data.
It is very expensive, time–consuming, and requires advanced
computation power to analyze population data. Therefore, it is not advisable
to analyze population data until there is no other way to get the information.
The sample is a logical subset of the population, which mimics
the population. Selecting a relevant sample out of the population is
challenging, but it makes analysis faster, precise, and economical. There
are standard guidelines from statisticians to calculate the relevant sample
size, appropriate sampling methodology, and tool to analyze sampled data.
Below are a few examples of sample–based analysis :
• During exit polls, companies want to predict the election winner
much before the result is declared by the election commission, they
ask the opinion of few sampled voters and declare their predictions.
• Companies ask preferences of a few sampled candidates to decide
features for their upcoming mobile phones.
• Companies call few sampled customers in order to understand the
area of improvements about their product or services.
26
Check Your Progress – 1 : Descriptive Analytics
1. Final grades (A, A+, B, etc.) in university exams is an example
of :
a. Nominal data b. Ordinal data
c. Ratio data d. None of above
2. Generally, we prefer to analyze the ___________ data as it is not
advisable to analyze the entire data even if we have access to it
as it leads to consuming more time, resources and efforts.
2.5 Components of Descriptive Statistics :

The focus of Descriptive analytics is on two important dimensions
of data : Central tendency of data and dispersion of data.
Either central tendency or dispersion alone cannot provide a complete
picture of the dataset. E.g. if one river is 4.5 feet deep towards its banks
and 5 feet deep for the next two meters from the bank and 6 feet deep
in the middle. Let's say the average depth of the river is 5.3 feet. Can
anybody with 5.6 feet of height cross the river (if he/she doesn't know
swimming) ? Not ! Therefore, having information only about central
tendency is not enough to make a better business decision we need to
know the data dispersion also.
Mean
Central Tendency Median
Mode
Descriptive
Statistics
Range
Dispersion Variance
Standard Deviation
Fig 2.3 Components of Descriptive Analytics

2.5.1 The Measure of Central Tendency :
One of the most important measures used to describe a dataset
statistically is the measure of central tendency. It provides us with
information about the centre point of the dataset in terms of a single
value. Mean, median, and mode are the most important measures, but
frequently percentile and quartile are used as a measure of central
tendency.
27
Business Analytics 1. Mean (Average) Value :
Arithmetic mean is the most frequently used measure of central
tendency. For simplicity, we refer to arithmetic mean as "mean". It is
the summation of all numbers divided by the total numbers. The population
means is represented by a Greek letter while the sample means is
represented by X whereas N is the number of total data in population
while n is the total number of data points in the sample.
Σx x1 + x 2 + x 3 + x 4 + ... + x n
=
µ =
N N
MS Excel formula for the mean is AVERAGE (array of numbers).
The array represents all numbers in sequence.
Example 2.1 : Calculate the mean for a student who has scored
90, 76, 62, 91 and 56.
Solution : Step1 : Add all numbers, which is 375
Step2 : Divide by the total number of data points, which is 5 here
90 + 76 + 62 + 91 + 56
=µ = 75
5
Example 2.2 : Data is available in the form of frequency
Age 16 18 20 25 28
Total Students 2 1 5 2 10
Weighted Average : If a few data points amongst the data set are
to be given more importance than others, the weighted average method
can be applied. In the below example, few subjects are more important.
In that scenario, we calculate the weighted average; it's calculated the
same as we have calculated the mean (average) from the frequency table
in the above example.
Example 2.3 : Data is available in the form of frequency
Sr. No. Subject Name Credit (Weight) Score (out of 10)
1 Statistical Analysis 5 7
2 Data Mining 4 8
3 Logical Reasoning 4 9
4 English 3 8
5 × 7 + 4 × 8 + 4 × 9 + 3 × 8 127
Weighted Average
= = = 7.94
(5 + 4 + 4 + 3) 16
2. Median Value :
Median value is the midpoint of the data sets when data is sorted
in ascending/descending order. For the odd number of data points, the
28
(n + 1) Descriptive Analytics
median value will be observation while for the even number
2
of data points, an average of the middle two observations after sorting
the data in ascending order.
MS Excel formula for mean is median (array of numbers).
Example 2.4 : Calculate median for following income data of 11
executives in company ABC Limited.
Annual Income
62000
64000
49000
324000
1264000
54330
64000
51000
55000
48000
53000
Solution :
Step1 : Arrange all data in ascending or descending order
Step2 : Check if total observations are odd or even
Step3 : As there is an odd number of data points hence median
th
 n + 1
will be  observation.
 2 
th
 n + 1
In this case, n = 11 hence  observation will be the median.
 2 
Therefore 6th observation, which is 55000 is the median.
Example 2.5 : Calculate median for following income data of 12
executives in company ABC Limited.
Annual Income
62000
64000
49000
324000
29
Business Analytics 1264000
54330
64000
51000
55000
48000
49000
53000
Solution : Step 1 : Arrange all data in ascending or descending
order
Step 2 : Check if total observations are odd or even
Step 3 : As there are an even number of data points hence the
median will be the average of the middle two observations
(54, 330 + 55, 000) 1, 09, 330

= = 54, 665
2 2
3. Mode :
It is the value that most often occurs in the data; it can be applicable
for both numeric and categorical data.
MS Excel formula for Mode is MODE (array of numbers).
Example 2.5 : Burger prices of 10 different food outlets in Vadodara
and Ahmedabad
Food Outlet Number Vadodara Ahmedabad
1 35 23
2 55 25
3 65 36
4 45 39
5 35 42
6 62 49
7 48 55
8 35 67
9 40 72
10 42 80
Solution : As we can see 35 is the most repetitive price of Burger
in Vadodara; hence it is the mode. In the case of Ahmedabad, all values
are unique; hence we can say there is no mode. Although we can claim
that there are 10 modes in the case of Ahmedabad, that will not make
30
any business sense. Therefore, there can be 0 or any number of modes Descriptive Analytics
in the data. For example, there are two modes, 25 and 30 in the below
data as both numbers occur thrice in the data set :
25
30
36
25
25
30
30
45
Now the question arises which measure of central tendency is best,
and how can we determine which one is most appropriate in any given
situation ?
Mean is the best measure of central tendency as it includes all
data points of the dataset, but it is highly sensitive towards outliers (data
points that are significantly different from others).
In the case of outliers, the median will be more appropriate as
it does not matter what minimum or maximum observations are. The
Median is always calculated by middle observation(s).
Mode is generally used when we want to estimate the central
tendency, and we do not have the opportunity to calculate the mean or
median of data.
To understand the distribution of our data, whether it is homogenous
(all data points are close to each other and do not have outliers), we
plot the data distribution. One of the easy ways to visualize data distribution
is seeing a Histogram of data. We will cover it in detail in the next unit.
2.5.2 Measure of Dispersion of Data :
There are three most important measures for dispersion of data
which are Range, Variance, and Standard Deviation.
1. Range : Range is the difference between the maximum and minimum
observation in a dataset. For example, in example 2.4, the range
of annual salary is :
12,64,000 (maximum data point) – 48,000 (minimum data point)
= 12,16,000
2. Variance : Variance is a measure of the distance of data points
from its mean. Population variance is generally represented by 2
and the formula is as below :
n (Xi − µ)2
∑
i =1 N
31
Business Analytics While the variance of a sample is given by :
n (Xi − µ)2
∑
i =1 N − 1
MS Excel formula for variance for population, sample is var.p(array)

and var.s(array) respectively.
In statistics, we have a different formula for the population (when
we can analyze entire data) and sample (when we can access limited
data to save time and effort). In the case of the population, we are 100%
sure about the measures like mean, variance, etc. While the sample is
an approximation of the population parameters as if we get 10 different
samples from the same population, we will get 10 different measures.
Hence statisticians tried to solve this by adjusting the formula. This will
be clearer in the next block when we introduce the concept of Degree
of Freedom.
Variance tells us how far data points on an average are from the
mean. In a way, it tells us the spread or dispersion of the dataset. Let's
try to understand why we do the square of all distances in the variance
formula without going much in–depth into its mathematical interpretation.
There are two reasons; first, it magnifies the distance and second is that
it helps to escape from positive and negative terms cancellations. For
example, suppose the mean is 16, and there are two points 12 and 20.
Now distance of 12 from 16 is –4 while the distance of 20 from 16
is +4. Now if we see the total distance of these two points from mean
16 then it will become 0, which is incorrect as both data points are 4
units away from the mean 16. Hence, we do square as it helps in getting
rid out of negative signs.
Example 2.6 : Calculate the variance of the below data points,
consider it as a sample (not population)
No of Toys
1
1
1
2
3
4
5
5
5
5
32
Solution : Descriptive Analytics
No. of Toys (x) Xi – (Xi – )2
1 1 – 3.2 = –2.20 4.84
1 1 – 3.2 = –1.96 4.84
1 1 – 3.2 = 1.00 4.84
2 2 – 3.2 = 2.00 1.44
3 3 – 3.2 = 3.00 0.04
4 4 – 3.2 = 4.00 0.64
5 5 – 3.2 = 5.00 3.24
5 5 – 3.2 = 5.00 3.24
5 5 – 3.2 = 5.00 3.24
5 5 – 3.2 = 5.00 3.24
x = 32 (Xi – )2 = 29.6
Mean = 3.2
The formula for sample variance
n(Xi − µ)2
∑
i =1 N − 1
29.6
Variance = = 3.29
9
29.6
While if it had been population, then variance would be = 2.96
10
3. Standard Deviation :
Standard deviation is the square root of variance. The formula for
the standard deviation of the population ( ) and sample standard deviation
(s) is as below :
The standard deviation for a population :
n
(Xi − µ )2
σ= ∑ N
i =1
The standard deviation for a sample :
n
(Xi − µ)2
s= ∑ N −1
i =1
33
Business Analytics The standard deviation of no of toys given in example 2.6 is as
below :
= 1.72 s = 1.81
Variance is squared; hence it is quite large compared to the data
points. Another challenge is it is square the unit of measurement. Because
of that, most of the time, we use standard deviation instead of variance.
One challenge with variance and mean is that we can't compare
these measures of different datasets as these depend on absolute values
of datasets. For example, we can't compare the share price of one company
with another company as one company may have a share price of less
than 100 while others may have more than 20,000. To counter this, we
get another measure known as the coefficient of variance.
4. Coefficient of Variance :
It is a ratio of standard deviation and means. It is unitless; hence
we can compare the coefficient of variance of two completely different
datasets. Similar to other measures, we have a different formula for the
coefficient of variance for population and sample.
σ S
CV = × 100%; CV = × 100
µ x

1. If we have only 4 data points, then which one would be the best
way to calculate the variation in the data
a. Variance b. Standard deviation
c. Range d. All of above
2. Coefficient of variance is a better measure as it includes both
___________ and ___________ of the data
1. Calculate the mean and median of below six numbers :
13, 3, 8, 10, 8, and 6.
a. Mean = 8; Median = 10 b. Mean = 8; Median = 8
c. Mean = 10; Median = 8 d. Mean = 10; Median = 10
2. The approximate value of the below expression :
n
f (x)
= ∑ (Xi − X)
n =1
a. Zero b. Mean
c. Variance d. Standard deviation
34
3. Which one of the following measures is appropriate for Nominal Descriptive Analytics
scale variables ?
a. The average age of students in a class
b. Average marks of students
c. Gender of students
d. Height of students
4. Calculate the variance of the below numbers; consider it as population
data :
6, 8, 2, 9, 7, 3, 1, 4
a. 5 b. 40 c. 6 d. 7.5
5. Calculate the standard deviation of the below numbers; consider
it as sample data :
6, 8, 2, 9, 7, 3, 1, 4
a. 2.4 b. 2.93 c. 3 d. 4.5
6. What are NOT true characteristics of mean
a. It includes all data points
b. It gets impacted with extreme values
c. It depends only on the central value of an ordered list
d. It is equivalent to the arithmetic mean
7. Mean, Median and Mode are best to represent :
a. Shape of dataset b. Location of dataset
c. Both d. None of the above
8. If most frequently occurring data is an outlier, then in that case :
a. Mode is a poor measure
b. Mode is a good measure
c. We can't say
d. The mode can be calculated in that case
9. Which of the following is best to compare the variability of two
completely different data–set ?
a. Standard deviation b. Variance
c. Coefficient of variance d. Range
10. We have extracted a sample of 200 people from a city, which one
of below is correct :
a. The standard deviation of the sample is always greater than the
standard deviation of the population
b. The standard deviation of the sample is always less than the
standard deviation of the population
c. We can't say whether the Standard deviation of the sample is
lesser, or the Standard deviation of the population is lesser
d. 200 people sample is too less to calculate standard deviation
35
Business Analytics 2.6 Let Us Sum Up :
1. Business analytics projects start with descriptive analytics, which
includes data summarization, aggregation, descriptive statistics, and
visualization
2. Analysts run queries to fetch appropriate data and consolidate
required data from different sources
3. Descriptive analytics focus on information gathering from raw data
which tells us what happened in the past
4. Visualization helps in storytelling through data and appropriate
graphs as per the data types
5. Data scientists calculate central tendency, variation, and shape of
data to understand the characteristics such as variance, central point,
and skewness
6. Descriptive analytics prepare field for further in–depth analysis like
diagnostic and predictive analytics
1. b 2. Sample
1. c 2. Centrality, variation
1. b 2. d 3. d 4. a 5. b
6. c 7. a 8. c 9. c 10. a
2.8 Glossary :
Descriptive Analytics is the simplest form of analytics; it summarises
current business status in the way of narrative and innovative visualization.
It emphasizes "what is going on in the business."
The sample is a logical subset of the population, which mimics
the population.
Categorical Data : This type of data is known as discrete data.
Categorical variables have a finite number of groups, e.g., payment
method, gender, income group, property type, etc.
Continuous Data : This type of data can be measured on a
continuous scale–like height, weight, money, time, etc. It can be divided
into halves any number of times.
Nominal Data : In this type of categorical variable, there is no
logical sequence among categories, which means one category is not
superior or inferior to others. Categorization is just a type of segregation
of data into different groups.
Ordinal Data : It is better than nominal data in terms of usage
potential for any analysis. Ranked or ordered data generally come into
this category.
36
Ratio Data : In terms of statistical relevance, this is the highest Descriptive Analytics
level of data which is desirable for the application of statistical tools
and techniques.
Mean (Average) Value : It is the summation of all numbers divided
by the total numbers.
Median Value : Median value is the midpoint of the data sets when
data is sorted in ascending order. For odd data points, the median value
(n + 1)
will be observation while for even data points, an average of
2nd
middle two observations after sorting the data in ascending order.
Mode : It is the value that most often occurs in the data; it can
be applicable for both numeric and categorical data.
Variance : Variance is a measure of the distance of data points
from its mean.
Standard Deviation : Standard deviation is the square root of
variance.
Coefficient of Variance : It is a ratio of standard deviation and
means. It is unitless; hence we can compare the coefficient of variance
of two completely different datasets.
2.9 Assignments :
1. Define various ways to calculate the dispersion of data, write their
strengths and limitations.
2. Write down one scenario where mean is better than median and
vice-versa.
3. Explain how the coefficient of variance can help in ranking important
mutual funds for your customers.
2.10 Activities :
Supertronic motors have published month–wise sales data in Gujarat.
Month Number of Units Sold
1949–01 112
1949–02 118
1949–03 132
1949–04 129
1949–05 121
1949–06 135
1949–07 148
1949–08 148
1949–09 136
1949–10 119
37
Business Analytics Calculate mean, median, mode, range, variance, standard deviation,
and coefficient of variance of above data.
MEAN 129.8
Median 130.5
Mode 148.0
Variance 153.7
Standard Deviation 12.4
Range 36.0
Coefficient of Variance 0.1
2.11 Case Study :

AWQS Constructions is a premium construction company in North
Gujarat. They construct villas and flats. They want to understand which
unit is giving them more consistent revenue.
Monthly Revenue from Monthly Revenue from
Flats Unit (` crores) Villas Unit (` crores)
123 128 123 102
124 123 225 97
125 129 226 129
123 121 123 121
156 154 111 102
178 174 98 215
129 122 121 205
Questions :
1. Which unit has more consistent revenue ?
2. Which unit has a better coefficient of variance ?
3. Which unit has a higher mean and median ?

Carmer H. (1946)
S. J. (2009)
Journal of Epidemiology and Community Hearlth, Yough
F. W. (2001)
38
Unit VISUALIZATION TECHNIQUES
3 FOR BUSINESS ANALYTICS
: UNIT STRUCTURE :
3.1 Introduction
3.2 Introduction to Data Visualization
3.3 Histogram
3.4 Bar Chart
3.5 Scatter Plot
3.6 Box Plot
3.7 Control Chart
3.8 Tree Map
3.9 Let Us Sum Up
3.11 Glossary
3.12 Assignment
3.13 Activities
3.14 Case Study

• Learning various types of visualization techniques
• Interpretation of visualization techniques
• Understanding the most appropriate visualization technique as per
the scenario
3.1 Introduction :
In this unit we will learn various type visualization techniques and
scenarios in which we use specific type of visualization techniques. At
the end we will also look into the benefits and limitation of each type
of graphs. We have also included few case studies so that interpretation
part can also be explained in detail.
3.2 Introduction to Data Visualization :

"Visualization is the soul of business analytics; it is the language
which everyone understands completely without saying much".
– Anonymous 39
Business Analytics Data Visualization : Data visualization makes descriptive analytics
so important and easy to understand. It helps in taking appropriate and
prompt decisions for the betterment of the business. There are various
charts and graphs which assist decision–makers in extracting important
insights, and hidden knowledge lies in data. The few important types of
charts which we will cover in courses are Bar charts, histograms, pie
charts, box–plot, radar charts, line charts, area charts and scatter plots.
Data visualization is the very first step towards starting an analytical
project. There are various softwares available for visualization, few most
important are Microsoft Excel, Open Office Maths, tableau, Microsoft
PowerBI, etc.
3.3 Histogram :
Histogram applicable for continuous data (Ratio and Interval data
types), there are consecutive but non–overlapping bars where each bar
shows the frequency distribution of the data. Histogram assesses the
probability distribution of the data.
In MS–Excel, a histogram is not part of graphs; we have to activate
the "Data Analysis" pack.
Below is an example of sales data from different customers in the
last financial year. Total there are 33 sales figures from these customers,
and sales amount varies from ` 14,786 to ` 29,29,278.
Fig 3.1 Histogram

Most statistical software automatically determines the number of
bins (Intervals). Although there are formulas suggested by different
statisticians by which we can decide the count of bins. So, there can
be a different number of bins for the same data if we are using different
statistical software. In MS Excel, we have to determine the number of
bins manually. For the above histogram, six bins were supplied :
One of the most frequently used approaches to determine the number
of bins is as below :
40
` 4,85,749 Visualization Techniques
for Business Analytics
` 9,71,497
` 14,57,246
` 19,42,995
` 24,28,743
` 29,14,492
X max − X min
Number of bins, N =
W
Where Xmax is the maximum value in the dataset while Xmin is
the minimum value in the dataset, W is the width of the bins : Famous
statistician Strurges (1926) suggested below equation in this research
paper :
Width of the bins (W) = 1 + 3.332 log10(n)
Here, n is the total number of observations in the data set.
In the above example, we had n = 33 hence the number of bins
suggested by the above formula is 6 (round off). Therefore,
(2929278 − 14786)
= 485749
6
Now all data points in the range of 0–485749 will be in the first
bin (interval); similarly, we will have six bins of width 485749 while
the last bin will have all remaining data points also.
The histogram is one of the most important charts in statistical
visualization treasure. Statisticians and business analysts use the histogram
for the following reasons :
(a) It helps in identifying the outlier in the data set (if few bins are
quite far from the remaining bins then it can be considered as an
outlier)
(b) It helps in assessing the probability distribution of the data by
judging the shape of the distribution
(c) It suggests about central tendency of the datasets like mean, median,
and mode
(d) It helps in assessing the variability of the data set
(e) It also helps in assessing other important statistical measures like
skewness
41
Business Analytics
Fig 3.2 Histogram of Machines Sold Per Day

Above is another example of a histogram that depicts per day sale
of machines. Here clearly, we can see 72 is an outlier. It is also suggesting
that the mean is close to 50, which is also median and mode in this
case. This is also telling us the probability distribution; here, data is
following a normal distribution. In the second block, we will cover
probability distributions in detail.
3.4 Bar Chart :

It is a chart for categorical or qualitative variables. It helps in
comparing and assessing categorizes within a data set. Below is an
example of the sector–wise contribution of the GDP of India.
Fig 3.3 Bar Chart

Various variants for bar charts are available, we can show multiple
bars for each category like each sector's revenue for the financial year
2017, the financial year 2018 and the financial year 2019, etc. We can
check the Microsoft website to learn about these modifications in detail.
We can showcase subcategories also in the bar chart.
42
Visualization Techniques
Fig 3.4 Detailed Bar Chart

All these charts are also available in the open office which is
available without any license fee. A bar chart is among the frequently
used charts which executives use daily to showcase various important
business metrics.
3.5 Scatter Plot :

In the business world, often we need to see the relationship between
various variables, for example, whether the repo rate impacts on Sensex
index or whether there is a relationship between the gold price and crude
oil in the international market.
Below is an example where the scatter plot is showing the relationship
between the area of the property in square feet and Price in INR
Fig 3.5 Scatter Plot

 Case Study :
Case study to analyze student performance with the help of Scatter
Plot and Coefficient of Variance :
Let's say, we have a class of 25 students and the professor records
their monthly test scores on a spreadsheet for the Business Analytics
subject. Data is stored in the below format :
43
Business Analytics
Month_10
Month_1
Month_2
Month_3
Month_4
Month_5
Month_6
Month_7
Month_8
Month_9
Student
Roll No
1 86 99 97 84 88 84 87 94 100 83
2 62 73 50 50 52 50 58 57 53 51
3 64 62 61 60 73 73 74 79 75 64
| | | | | | | | | | |
| | | | | | | | | | |
| | | | | | | | | | |
| | | | | | | | | | |
25 91 84 83 84 80 97 91 89 99 89
Solution : We can calculate the average score in the last column

(next to Month_10), and in the next cell, we can calculate the standard
deviation. Hence, we will have centrality and variation of each student.
We can create a scatter plot for these two measures on both the axis
in such a way that the centre of the graph will have a midpoint of mean
and standard deviation. In this way, we can segregate the entire class
into four quadrants, and each quadrant will reveal important information
about the progress of these students. With the help of domain knowledge,
we can also name these quadrants so that we can think about customized
help for each set of students.
Fig 3.6 Scatter Plot – Avg Marks Vs Standard Deviation

 Interpretation of Scatter Plot :
Students fall in Q1 : Here Mean is High, which is good in this
scenario while the Standard Deviation is also high, which always lesser
is better. Students who fall in this quadrant are most of the time scoring
above the average, but they have high fluctuation means sometimes they
score very high while sometimes they score quite low. Hence teachers
can deep dive into their performance so that they can identify the topics
where these students are weak and arrange revision class only on those
particular topics.
44
Students fall in Q2 : Here Mean is high, and the Standard deviation Visualization Techniques
is low. Best students fall under this quadrant as they are most consistent. for Business Analytics
They score high, and there is very little fluctuation within their performance.
We can benchmark these students; try to understand how they prepare
for exams and what are other best practices they follow due to which
their performance is consistently good.
Students fall in Q3 : Here mean is low and the standard deviation
also low. These students are consistently low performers. They score
poorly in most of the exams. The instructor can arrange extra classes
for these students, or they can make one good, one poor student pair
to clarify their poor concepts.
Students fall in Q4 : Here means is low, and the standard deviation
is high. These students have the worst performance as most of the time,
their scores are below average, but also, they have high fluctuations.
Conclusion : This is a very powerful technique as it provides very
insightful information about the different types of student's cohorts (groups).
It provides a very specific improvement area for each group without
generalizing solutions like arrange extra classes etc. This study is very
generic and can be replicated for share analysis, e.g. generally blue–chip
shares fall under Q2 where the mean is very high, and the standard
deviation is very low, in other words very consistent shares give a high
and steady return. Similarly, it can be replicated for mutual fund analysis.
ICC also conducts a similar analytical approach to find out the most
consistent batsman, who scores high runs and has comparatively less
variation among his performance (consistent performance).
1. Histogram and scatter plots are applicable for :
a. Continuous data b. Discrete data
c. Both type of data
2. A Scatter plot can be used only to show a linear relationship
between the variables
a. True b. False
3. Y–axis of histograms shows ___________ of the dataset
3.6 Box Plot :

Box plot is a smarter way to assess
variability in the numerical data by
splitting data into four quartiles. It is
an effective way to identify the outliers
in the data sets.
Box plot splits data into four equal
quartiles (Quarter 1 quartile means 25th
percentile). If 25% of data points are
Fig 3.7 Box-Plot Graph
45
Business Analytics below a data point, then we call it the 25th percentile, or these lowest
25% data sets constitute the first quartile.
The middle part is known as the interquartile range (IQR), which
is the difference between quartile 3 and quartile 1. These are 50% of
data points from the middle (25% of data points are below the interquartile
range box, and 25% are above it).
Any data point which is above Q3 + 1.5*IQR or Q1 – 1.5*IQR
is considered as an outlier, which means these data points are not as
per normal nature. There may be some special reasons for occurring these
data points which need further investigation to understand the root cause.
Box plot is very important for comparing datasets. For example,
if we want to compare the sales for the last three financial years or
productivity of a company from its various branches etc. It is also very
useful in decision making like IQR represents the middle class if we
consider it as a population of any state or country. Policymakers try to
draft special policy for the betterment of people fall in Q1 (low–income
group), IQR (middle and upper–middle–class) and Q3 (rich or elite class).
Fig 3.8 Multiple Box-Plot Graphs

In the above figure, we can see that although overall sales are high
in 2014 but variability is also high. High variability reduces our capability
for robust planning. It is difficult to predict sales for next year if variability
in last year's data is high. Box plot is not available in Microsoft Excel
or Open office graphical unit, but with a small modification to the stacked
bar graph, we can create it in Excel or open office easily.
3.7 Control Charts :

A control chart is a special statistical chart that shows changes in
a process over time. The X–axis of a control chart is always time, and
its centreline is the mean of data. We can also call it a time series plot
with control limits. It is also a powerful tool to identify the outliers in
the data.
46
Visualization Techniques
Fig 3.9 Control Charts

This is a simple control chart, also known as the Individual control
chart, as all data points are individually plotted on the mean line (central
line). Instead of plotting the average as a central line, we can also draw
a target as a central line. There is a various advanced type of control
charts like Individual – Moving Range Control Chart (I–MR Chart) or
X bar Range Chart, etc. but those are not in the scope of our course.
1. Boxplots provide us information about spread of data as well as
___________ in the data.
2. Center point of box in boxplot graph is ___________.
3.8 Tree Map :

A Tree map chart is very appropriate to represent hierarchical data
in a tree–like structure. The Treemap is a strong visualization technique
when we have to show so many categories.
Fig 3.10 Tree Chart

1. Any observation below Q1 – 1.5 IQR in a box plot is :
a. Median b. Potential outlier
c. Mean d. Minimum value
47
Business Analytics 2. A bar chart is appropriate for which type of data
a. Quantitative data b. Qualitative data
c. Ordinal data d. Interval scale data
3. The upper control limit in the control chart is :
a. Mean + 3 Standard deviation
b. Median + 3 Standard deviation
c. Mean – 3 Standard deviation
d. Mean + 2 Standard deviation
4. Histogram can be used for which type of data
a. Quantitative data b. Qualitative data
c. Ordinal data d. Interval scale data
5. The main objective of the Scatter chart is :
a. Highlight outliers in the data
b. Showing the relationship between quantitate data sets
c. Showing the relationship between qualitative data sets
d. Showcase variation in the data
6. Which graph is relevant for hierarchal data ?
a. Box plot b. Bar graph c. Histogram d. Treemap
7. Y–axis in the histogram represents :
a. Probability b. Area under curve
c. Frequency d. Area
8. The central line of control charts represents :
a. Mean or target b. Standard deviation
c. Median d. Mode
9. The central line of the IQR box of box plot represents :
a. Mean b. Variation c. Median d. Mode
10. The formula for the coefficient of variance is :
a. Standard deviation / Median b. Mean/standard deviation
c. Median / Standard deviation d. Standard deviation / Mean
3.9 Let Us Sum Up :
1. Data visualization simplifies the information understanding hidden
in raw data
2. Data visualization helps analysts in deciding the next step by
displaying pictorial relationship among various business metrics
3. As per the data type, we should choose the right type of graph
so that maximum information can be understood with minimum
efforts
4. Visualization becomes more effective by adding features in data
visualization like right axis scale, legends, combing important data
together, or split data into sensible sub–parts
48
5. Histograms, boxplot, scatter plot, control charts are important graphs Visualization Techniques
can be made on continuous data for Business Analytics
6. Pie chart, bar chart, Treemap are important graphs for categorical
graphs

1. (a) 2. False 3. Frequency
1. Outliers 2. Median
1. b 2. b 3. a 4. a
5. b 6. d 7. c 8. a
9. c 10. d
3.11 Glossary :
Bar Chart : It is a chart for categorical or qualitative variables.
It helps in comparing and assessing categorizes within a data set. Here
x–axis represents categories while the y–axis represents the count or
percentage of data.
Histogram : It is a kind of area graph. It looks similar to a bar
graph of data that buckets a range of data into columns along the x–
axis while the y–axis represents the frequency (number of data count)
or percentage of occurrences in the data for each column. It is applicable
for continuous data only.
Scatter Plot : It is used to show representation between two
continuous variables, generally we put the input variable on the x–axis
while the output variable on the y–axis. It also helps in identifying outliers
in the dataset.
Box–Plot : It is a smarter way to assess variability in the numerical
data by splitting data into four quartiles. It is an effective way to identify
the outliers in the data sets.
Control Chart : A control chart is a special statistical chart that
shows changes in a process over time. The X–axis of a control chart
is always time, and its centreline is the mean of data. We can also call
it a time series plot with control limits. It is also a powerful tool to
identify the outliers in the data.
Tree Map : A Tree map chart is very appropriate to represent
hierarchical data in a tree–like structure. The Treemap is a strong
visualization technique when we must show so many categories.
3.12 Assignments :
1. Which graph to be used if you want to show the relationship
between two continuous variables, explain with an example.
49
Business Analytics 2. How interquartile range helps in taking decisions, explain it with
an example.
3. Write down the importance of tree diagram, mention two scenarios
where you can implement a tree diagram.
3.13 Activities :
Online training firms collected the training hours for their 16 trainers
and their feedback. Draw a scatter plot and see where there is the
relationship between training hours and feedback scores. Also, draw a
quadrant graph like provided in the chapter (fig 3.6)
Training Hours Feedback Training Hours Feedback
75 8 94 5
69 5 81 5
104 5 86 9
106 4 99 3
92 8 110 7
111 2 76 4
64 6 101 4
92 4 125 4
3.14 Case Study :

Pokymon trading limited is an investment advisory organization,
they have active customers across central part of India. They have listed
top 40 company's stocks and their value in rupees :
292 790 354 318 345 216 362 533 256 260
233 349 467 477 50 419 565 590 522 466
247 293 273 545 566 411 470 235 247 288
364 429 569 597 248 505 309 347 353 417
(a) Make a histogram for the above dataset. List down important
insights from the data
(b) Draw a box plot and see how different all four quadrants from each
other, also see if there are any outlier in the data
(c) Draw a control chart of the above data and see if any trend is visible

Carmer H. (1946)
S. J. (2009)
50 F. W. (2001)
BLOCK SUMMARY
Business analytics has become the brain or spinal cord of businesses
nowadays. Like the human body depends on brain/ spinal code for decision
making and other important functions similarly entire organizational
strategy is dependent on business analytics. Executives want to make all
decisions based on data only. Nowadays business units are integrated with
the help of enterprise–wide softwares hence businesses have a better view
of the entire value chain instead of seeing opportunities or problems in
individual business units. Business analytics is a very structured approach
where it starts with descriptive statistics where the focus is on visualization
and "What did happen" then we see all good and bad phenomenon in the
business with the help of outcomes of descriptive statistics like a report
card, dashboards, executive summaries, KPI metrics, etc. Then we move
to diagnostics statistics to understand the root cause for all good and bad
things so that we can replicate the good scenarios while minimizing the
probability of happening bad things again in the business. These two types
of analytics provide a solid foundation for the next level of analytics like
predictive and prescriptive where focus on the future. We try to predict the
future values of important business metrics and prepare a continuity plan.
51
Business Analytics BLOCK ASSIGNMENT
Short Answer Questions :
1. Explain how business analytics is a systematic approach to be
established in an organization.
2. Write down few important challenges in establishing business analytics
in an organization
3. Write down the important difference between descriptive and diagnostic
analytics
4. Explain the importance of visualization in the business analytics
5. Why we say that "central tendency alone does not show us the
complete picture, we should also consider variation also". Justify the
statement with one example ?
6. Write down important components of business analytics
7. Explain the life cycle of business analytics, mention one example
8. Why should we study sample only even if we have access to entire
population data ?
9. Write one scenario for the usage of mean and median as a measure
of central tendency
10. Why do we call mode a weak measure of central tendency ?
11. Explain briefly the difference between bar chart and histogram
12. Write a short note on the degree of freedom
52
Long Answer Questions :
1. Write a note on how predictive analytics is different from prescriptive
analytics. Mention important tools/ techniques under each type of
analytics ?
2. Why data become so important in the last 10 years, write down few
success stories from the business world about data–driven decision
making ?
3. Write short note on different data measurement scales, give two
examples for each data type ?
4. How control charts and boxplot help in the analysis of variance in
the data. Give suitable examples ?
53
Business Analytics  Enrolment No. :
1. How many hours did you need for studying the units ?
Unit No. 1 2 3
No. of Hrs.
2. Please give your reactions to the following items based on your reading
of the block :
3. Any other Comments

.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
54
Ambedkar
OpenUniversity
Business Analytics
BLOCK-2 STATISTICAL CONCEPTS AND HYPOTHESIS

TESTING
UNIT 1
DISCRETE PROBABILITY DISTRIBUTIONS
UNIT 2
CONTINUOUS PROBABILITY DISTRIBUTIONS
UNIT 3
SAMPLING AND CONFIDENCE INTERVALS
UNIT 4
INTRODUCTION TO HYPOTHESIS TESTING
BLOCK 2 : STATISTICAL CONCEPTS AND HYPOTHESIS
TESTING
Block Introduction
Business analytics try to reduce the uncertainty in the businesses so
that executives can make better decisions backed by data. Descriptive and
diagnostic statistics provide them important insights about business progress
in terms of important business metrics and lay down the foundation blocks
for the next level of analytics. Here onwards probability theory start playing
an important role. Probability theory and concepts of probability distributions
are essential building blocks of business analytics. These concepts help us
to study the hid-den/obvious trends in the business and provide us
alternatives that have fewer risks and enhance the decision-making process.
Probability distributions and understanding the theory of hypothesis testing
are important pillars on which predictive and prescriptive analytics models
are built.
Business analysts generally avoid predicting point (single value)
estimates rather they prefer to estimate the confidence interval which
reduces the uncertainty e.g. instead of saying next month sales will be Rs
856 crores, analysts say sales will be in the interval of (Rs 800 Cr to 912
Cr). Confidence intervals also tell us information about population parameter
instead of just sample which result into more robust decision making. Here
business analysts use a very important concept called hypothesis testing
which is used to validate the business as-sumptions based on sample data.
It reduces the ambiguity in decision making and provides us with better
opportunities.
Block Objectives
• Understanding concepts and applications of the probability distribution
• Understanding important discrete probability distribution functions
• Difference between probability mass function and probability density

function
• Applications of the continuous probability distributions
• Different types of sampling methodologies and techniques
• Significance of central limit theorem and its applications
• Application of confidence interval and confidence level
• Understanding the basics of hypothesis testing and its applications in

decision making
• Understand the concept of significance (α) and Type 1 and Type 2

error
• Understand the significance of p-value and its application in concluding

hypothesis test
Block Structure
Unit 1 : Discrete Probability Distributions
Unit 2 : Continuous Probability Distributions
Unit 3 : Sampling and Confidence Intervals
Unit 4 : Introduction to Hypothesis Testing

Unit DISCRETE PROBABILITY
1 DISTRIBUTIONS
: UNIT STRUCTURE :
1.1 Introduction
1.2 Random Experiments and Probability Distributions
1.3 Discrete Probability Distributions
1.3.1 Binomial Distributions
1.3.2 Poisson Distribution
1.4 Let Us Sum Up
1.6 Glossary
1.7 Assignment
1.8 Activities
1.9 Case Study

• Understanding concepts and applications of the probability
distributions
• Concept of Probability mass functions and cumulative density
functions
• Understanding important discrete probability distribution functions
• Expected values and variance of discrete probability distributions
1.1 Introduction :
Unit Introduction : In this unit, we will study the basic concepts
of probability distributions. We will see various examples of how we
apply theory of these distributions in the business world to take important
decisions. Further, in the unit we will see how each outcome of a random
experiment is mapped to a probability distribution know as probability
mass function while the cumulative distribution function is the probability
that a random variable can take values less than or equal to xt. As a
part of an application, we will see the expected values and variance of
these discrete probability distributions.
1.2 Random Experiments and Probability Distributions :

"Probability theory is nothing but common sense reduced to
calculation." – Pierre–Simon Laplace
55
Business Analytics Probability distribution consists of all possible values a random
variable can take; these values tell us about the minimum and maximum
values between which a random variable can vary. Post discussion on

a range of the values; now the point is how frequently these possible
values can occur within the potential range. Generally, it depends on two
critical factors – mean (expected value) and standard deviation. Although
there are other two factors also which are comparatively less critical are
skewness and kurtosis.
Let's demonstrate it with the help of an example :
x is a likely outcome of the throw of dice; x = [1, 2, 3, 4, 5, 6]

while P(x) is the probability of occurring 3. We represent it as
1
P(X = 3) = . As P(x) represents the probability of any possible
6
outcome, we call it the probability function P(x). So probability distributions
represent the likelihood of an outcome depending on how frequently it
is featured in the given sample space.
Let's consider one more example to understand this concept better;
an online shopping company collected feedback for their packaging facility.
They collected feedback from 108 customers. Feedback was collected on
the Likert scare of 1–5 where 1 was 'very dissatisfied' while 5 means 'very
satisfied'. We can represent these five outcomes in terms of numbers like
x = 1, 2, 3, 4, 5 while P(x) represents the probability of each outcome
There are two types of probability distribution functions :
• Discrete probability distribution functions
• Continuous probability distribution functions
Let's understand these distributions in detail with the help of their
definitions and applications in the business world.
56
1.3 Discrete Probability Distributions : Discrete Probability
Distributions
If a random variable can obtain only discrete (countable) outcomes,
for example, 0, 1, 2, 3 etc but it cannot take values like 2.3 or 3.5 etc.
then we say it is a discrete random variable that follows discrete probability
distribution. In other words, it tells us how probabilities are assigned
to every outcome of a random variable as we saw in the above two
examples. As we studied earlier that probability distributions are defined
by two important factors, the expected value (average) and standard
deviation. For a discrete distribution, the expected value is the mean of
the random variable (average of expected outcome); we represent it as
follows :
E(X) = = x P(x)
Let's calculate the expected value of the feedback data :
So we can see the expected value of the packaging facility is closer

to "satisfied." Please note discrete random variables cannot obtain decimal
values, but the expected value can be represented in decimals. Now let's
discuss variability (how far is a random variable from its mean). For
any discrete variable, the variance is calculated by the below formula :
Var (x) = 2 = (x – µ)2 P(x)
Standard deviation = Var(x) = σ = ( Σ(x − µ)2 P(x))
Let's understand a few essential terminologies regarding the discrete

probability distribution – Probability mass function (PMF) and cumulative
distribution function (CDF).
57
Business Analytics Probability Mass Function
(PMF) is a graphical representation of
probabilities for all possible values of
a discrete random variable. This is also
known as a frequency function.
If we see an example of dice,
then possible outcomes are x = [1, 2,
3, 4, 5, 6] and there is equal probability,
1
= .167
6
There is an equal probability of each possible value of random
variable x.
Cumulative Distribution
Function (CDF) is another way to
represent the distribution of a
random variable, but unlike PMF,
it is not limited to individual
discrete variables only. At any
point, it represents probability up
to that point.
P(X < 3) = P(X = 0) + P(X = 1) + P(X = 2)
Y–axis of CDF goes up to 1 as probability cannot be more than 1.
Let's consider one more example to understand this concept further,
suppose I have a special dice where I can't roll 3 or 4. There x = 1,
2, 5, 6. Therefore there is blank space in PDF graphs (there is no "mass"
in the probability mass function). In CDF, we can see the probability
of having 2 or less is the same as 4 or less. Once we have mass at
5 in the PDF graph hence CDF value at 5 get change.

1. Which of below are characteristics of a discrete probability distribution
a. Mean (expected value) b. standard deviation
c. Skewness d. Kurtosis e. All of the above
58
2. Probability mass function can be used to define a discrete probability Discrete Probability
function only Distributions
a. True b. False
1.3.1 Discrete Probability Distributions – Binomial Distribution :
"Life is a school of probability" – Walter Bagehot
In the binomial distribution, there are only two possible outcomes.
Here prefix "bi" indicates two. For example the probability of "yes" or
"no", "pass" or "fail", "heads" or "tails" etc.
There are below pre–requisites of binomial distribution :
1. Only two possible outcomes per trial
2. The probability of success or failure will remain constant across
all the trials
3. There are a fixed number of trails
4. Each trial is entirely independent of each other
 The General Formula for Binomial Probability :
Probability of k out of n ways :
n!
P(k out of n)
= p k (1 − p)(n − k)
k!(n − k)!
• Expected Value (Mean) of Binomial Distribution : np
Here n is the no of trials while p is the probability of success
• Variance and Standard Deviation of Binomial distribution :
np(1 – p)
• Standard Deviation : np(1 − p)

Example – 1.1 : Survey shows from Ambedkar University only
8% of students have tried paragliding. Now if we are picking randomly
10 students.
(a) What is the probability that all 10 tried paragliding
(b) What is the probability none of the students tried paragliding
(c) What is the probability exactly 2 students tried paragliding
(d) What is the probability at least 2 students tried paragliding
(e) What is the expected number of students who tried paragliding ?
(f) What is the standard deviation of the number of students who tried
paragliding in the sample
Solution : Here n = 10 and p = 0.08
(a)
59
Business Analytics The probability of the first student tried paragliding is .08.
Probability of second student also tried paragliding is .08 .08 =
.0064. Similarly, the probability of all 10 students tried paragliding
is (.08)10.
P(X = 10) = (.08)10 = 1.07 10–11
(b)
Probability of the first student NOT tried paragliding is .92.

Probability of second student also not tried paragliding is .92 .92
= .8464. Similarly, the probability of all 10 students not tried
paragliding is (.92)10 = P(X = 0) = .4344
(c) It seems that the probability of 2 students tried paragliding is (.08)2
while the remaining 8 students have not tried paragliding is (.92)8.
So the answer would be (.08)2 (.92)8.
But it is NOT RIGHT, because there can be the first two students
who have tried paragliding or the last two students or maybe
alternate one etc. Therefore there can be 10C2 ways in which we
can have 2 students out of 10, who have tried paragliding.
P(X = 5) = 10C2 (.08)2(.92)8 = 45 .0000022 = .1478
(d) At least two means, two or more.
One way we can calculate all these
probabilities and add them up
(P(X = 2) + P(X = 3) + ..... +
P(X = 10). (All those black bars in
the left graph).
But the easy approach can be, we can calculate P(X = 0) +
P(X = 1) and subtract it from 1.
1 – P(X 1) = 1 – P(X = 1) – P(X = 0) = 1 – .378 – .434 = 0.188
(e) The expected value of students out of a total sample of 10, who
have tried paragliding
E(X) = np, n is the number of trials while p is the probability of
success
Here, E(X) = 10 0.08 = .8
Therefore .8 is the mean of the sampling distribution
(f) Standard deviation : np(1 − p) = 10 × .08 × .92 = .858

 Microsoft Excel Ealculations for Binomial Distribution :
We calculate excel formula to calculate the probability of exact 2
students tried paragliding as we did in the last example, 2.1(c)
60
The formula is BINOM.DIST(2, 10, .08, False), here the first Discrete Probability
argument is required individual discrete outcome (2 in this example) then Distributions
the second argument is total numbers of trial, the third argument is the
probability of success, and the last argument will be False for NOT
cumulative while True for cumulative.
Similarly, answer to example 2.1(d) would be 1–BINOM.DIST(1,
10, .08, True) = 0.188.
Please Note : cumulative functions always work in one direction
(equal to or lower). Therefore if we want to calculate for a greater than
region, then we can subtract the result from 1, as we did above.
1.3.2 Discrete Probability Distributions – Poisson Distribution :
Poisson distribution is named French mathematician, Siméon Denis
Poisson. He discovered it in 1938. In his surname "I" remain silent; hence
it sounds like Pozzon. One of the critical assumptions of the binomial
distribution is that number of events (trails) must be fixed, but in case
that is not true then we use Poisson distribution.
When we cannot calculate the probability of success and failure
both, we use Poisson distribution–for example, the number of potholes
in a 10 km long road. Here we cannot calculate the probability of success
and failure both. We can calculate only one either success or failure.
For example, in a 10 km road stretch, we can only count potholes (which
can be a measure of failure), but we cannot say how many potholes could
be possible in this 10 km stretch (total number of trials).
Another example, number of calls received at a call centre in an
hour, here theoretically there can be numerous calls of small duration
are possible, but we cannot estimate the total number of possible calls,
we can only calculate the average number of calls received in an hour
(by analysing the historical calls data).
So Poisson distribution describes the average number of events
occurring in a fixed interval of time or any continuum region of
opportunity (which can be measure on a continuous scale, like length,
money, distance, weight etc.). It requires only one parameter, or
, which tells the average number of events in a given interval.
Poisson distribution bounded by 0 and . For example, there can
be 0 calls in a day, or there can be an infinite number of calls in
a day at a call centre.
The expected value (mean), E(X) and variance, V(X) of Poisson
distribution are the same, .
Example : On average, three
customers need a wheelchair per day
( / ) at a local grocery store. In the
below graph, these bars go up to
infinity although probability will
become minuscule, so for convenience, I have put up to 10 customers
only. 61
Business Analytics  Important Assumptions of Poisson Distribution :
1. Events occur at a constant rate, which means there should be equal
chances of the happening number of events in a one–time interval
to any other interval of time
2. The occurrence of one event must be independent of any other event
(i.e. Events are independent)
Please Note : These assumptions may not hold good in reality or
may not aligned with the business problem, which we are trying to solve
hence we need to be sure whether it is appropriate and to apply Poisson
distribution.
 PMF for Poisson Distribution :
It is the probability of each discrete outcome (height of each bar
in the above graph).
e−λ λ x
P(X
= x)
=
x!
Here e is Euler's constant whose value is 2.71.
 CDF for Poisson Distribution :
Γ (  x + 1 , λ)
P(X ≤ x) =
 x!
We don't need to understand this formula as it may be difficult to
understand for undergraduate students of this course. Instead, we will see
the excel formula to calculate CDF.
 Excel Formula to Calculate PMF and CDF for Poisson
Distribution :
To calculate PDF, POISSON.DIST(X, , FALSE) while calculating
CDF POISSON.DIST(X, , TRUE)
Example 1.2 :
(a) Calculate the probability of arrival for 5 customers who need a
wheelchair in a day, while = 3 is given.
(b) Calculate the probability of arrival of at least 5 customers who need
a wheel–chair in a day
Solution :
e−3 35
(a) P(X
= 5)
= = 0.101
5!
(b) POISSON.DIST(5, 3, TRUE) = 0.916
62
 Graphical Representation of Poisson Distributions for Different Discrete Probability
Values : Distributions
Need to remember that can be a decimal value also

Example 1.3 : One watch selling company sells an average of 12
watches per day through their website.
(a) Find the probability of selling 10 watches in a day
(b) Find the probability of selling at least 10 watches in a day
(c) Find the probability of selling more than 1 watch in the first hour
of the day
Solution :
(a) Probability of selling exactly 10 sales in a day
The desired bar is highlighted with white colour in the above graph.
e−121210
P(X
= 10)
= = 0.105
10!
POISSON.DIST(10, 12, FALSE) = 0.105
So there is approx. 10.5% probability of selling exactly 10 watches
in a day
63
Business Analytics (b) Probability of selling at least 10 sales in a day, probability of 10
or more in a day
The desired bars are highlighted with white colour in the above
graph.
P(X > = 10) = 1 – P(X < 9) = POISSON.DIST(9, 12, TRUE)
= 0.758.
So there is approx. 76% probability of selling at least 10 watches
in a day
(c) probability of selling more than 1 watch in the first hour of the
day
We have the value of in terms of sales per day while we need
to calculate it in terms of sales per hour. Therefore we need to change
our (mean) accordingly
12
λ= = 0.5 sales per hour
24
The desired bars are highlighted with white colour in the above
graph.
1 – POISSON.DIST(1, 0.5, TRUE) = 0.090. So there is approx.
a – 9% probability of selling more than 1 watch during the first hour.
Please Note : we solved the above example mathematically, that
is correct but if we re–think the assumptions of Poisson distribution that
there should be a constant rate of occurring events through all intervals
but here in the case of online watch sales it is very unlikely that somebody
will purchase a watch from late night to early morning. So in a way,
it is breaching the assumption of Poisson distribution. Therefore, before
applying statistical tools or techniques, we have to be very careful whether
our use case fulfil all required assumptions or not. Otherwise, we will
not get any error theoretically, but our inferences would be very misleading.
64
Poisson distribution can be an excellent approximation to the Discrete Probability
normal distribution, but only in case when the probability of success (p) Distributions
must be significantly small compared to the number of trials (n). One
thumb rule says Poisson distribution can be an excellent approximation
to the binomial distribution, in case n > 20 and np < 10. We will cover
the normal distribution detail in the next chapter.
1. Poisson distribution can be an excellent approximation to the normal
distribution, but only in case when the probability of success (p)
must be ___________ compared to the number of trials (n)
2. Cumulative distribution function (CDF) can be used to define a
discrete probability function only
a. True b. False
1. Which of the following is the formula for variance for a discrete
probability distribution ?
a. Np b. x P(x)
c. (x – )2P(x) d. Σ(x − µ)2 P(x)
2. The expected value of Binomial distribution

a. Np b. x P(x)
c. (x – )2P(x) d. Σ(x − µ)2 P(x)
3. The variance of Binomial distribution

a. np(1 – p) b. Np
c. x P(x) d. (x – )2P(x)
4. Which one of the following is NOT a valid assumption of Binomial
distribution
a. Only two possible outcomes per trial
b. The probability of success or failure will remain constant across
all the trials
c. There are an infinite number of trails possible
d. Each trial is entirely independent of each other
5. What is the expected value of the Poisson distribution
a. Np(1 – p) b. (mean)
c. x P(x) d. Σ(x − µ)2 P(x)
65
Business Analytics 6. Which of the following statement(s) are correct about the Poisson
distribution ?
a. Events occur at a constant rate, which means there should be
equal chances of the happening number of events in a one–time
interval to any other interval of time
b. The occurrence of one event must be independent of any other
event
c. Both (a) and (b) are correct
d. Only (a) is correct
7. Calculate the probability of arrival for 7 customers who need a
wheelchair in a day, while = 3 given.
a. 0.988 b. 0.082 c. 0.022 d. 0.052
8. What of following is the formula for the standard deviation for
Poisson distribution
a. b. np c. np(1 – p) d. λ
9. One car showroom sells 60 cars per month (historical average)
which following Poisson distribution, the analyst wants to calculate
the probability of selling 3 cars on the first day of the month. What
can be in this case
a. 30
b. 2
c. Not sufficient information to calculate the expected value
d. 3
10. In an exam, there are 25 multiple choice questions, each question
has four probable answers but only one answer can be correct if
a student is guessing every question. calculate the probability of
getting 10 questions correct ?
a. .042 b. .625
c. 2.17 d. None of the above
1.4 Let Us Sum Up :

1. In statistics probability distribution for a given random variable
explains how the probabilities are distributed over the possible
values of the random variable.
2. The shape of a probability distribution depends on two critical
factors – mean (expected value) and standard deviation. Although
there are other two factors also which are comparatively less critical
are skewness and kurtosis.
3. In the binomial distribution, there are only two possible outcomes.
Here prefix "bi" indicates two. For example the probability of "yes"
or "no", "pass" or "fail", "heads" or "tails" etc.
66
4. One of the critical assumptions of the binomial distribution is that Discrete Probability
number of events (trails) must be fixed, but in case that is not Distributions
true then we use Poisson distribution.
5. When we cannot calculate the probability of success and failure
both, we use Poisson distribution–for example, the number of
potholes in 10 km long road
6. Poisson distribution can be an excellent approximation to the
normal distribution, but only in case when the probability of success
(p) must be significantly small compared to the number of trials (n)

1. e 2. a
1. Very small 2. b
1. c 2. a 3. a 4. c 5. b
6. c 7. c 8. a 9. b 10. a
1.6 Glossary :
Random Probability Distribution : A random probability
distribution is a statistical function that explains all the possible values
a random variable can take within a given minimum and maximum range.
Discrete Probability Distribution : If a random variable can obtain
only discrete (countable) outcomes, for example, 0, 1, 2, 3 etc. then we
say it is a discrete random variable that follows discrete probability
distribution.
Probability Mass Function (PMF) : It is a graphical representation
of probabilities for all possible values of a discrete random variable. This
is also known as a frequency function.
Cumulative Distribution Function (CDF) : It is another way to
represent the distribution of a random variable, but unlike PMF, it is
not limited to discrete variables only. At any point, it represents probability
up to that point.
Binomial Distribution : In the binomial distribution, there are only
two possible outcomes. Here prefix "bi" indicates two. For example the
probability of "yes" or "no", "pass" or "fail", "heads" or "tails" etc.
Poisson Distribution : When we cannot calculate the probability
of success and failure both, we use Poisson distribution–for example,
the number of potholes in a 10 km long road.
67
Business Analytics 1.7 Assignments :
1. What do you mean by a probability distribution, what are different
types of probability distributions ?
2. What is the generic formula of variance and standard deviation for
a discrete probability distribution ?
3. What is the relationship between probability mass function and
probability cumulative function, explain it with an example ?
4. Explain the MS Excel formula for Binomial distribution and its
important arguments.
1.8 Activities :
1. Suppose email come to a customer care inbox follows a Poisson
distribution and the average number of emails per hour is 20
(a) What is the probability that exactly 10 emails will arrive in an hour ?
(b) What is the probability of arriving more than 15 emails in an hour ?
(c) What is the probability of arriving less than 5 emails in an hour
Ans. (a) 0.006
(b) 0.84
(c) 0.000017l,
1.9 Case Study :

Myclick.com is a leading online sales website in Northern India.
They have expanded themselves exponentially in the last three quarters.
Post adding new segments for women and kids garments they have faced
a hike in product returns. By analysing the past few months data, business
analysts informed the leadership team that about 10% of their customers
return the apparel because of various reasons. Three important reasons
are size, colour, material. On a particular week, there were 20 customers
purchased apparel from myclick.com.
Questions :
1. Calculate the probability that exactly 7 customers will return the
product
2. Calculate the probability that a maximum of 12 customers will
return the product
3. Calculate the probability that a minimum of 3 customers will return
the product
4. The average number of customers who will return their product
during this week
5. The variance and the standard deviation of the number of the returns
68
1.10 Further Readings : Discrete Probability
Distributions
Carmer H. (1946)
S. J. (2009)
F. W. (2001)
• "Data Strategy : How to Profit from A World of Big Data, Analytics
and The Internet of Things", O'Reilly Media, Bernard Marr
69
Business Analytics
Unit CONTINUOUS PROBABILITY
2 DISTRIBUTIONS
: UNIT STRUCTURE :
2.1 Introduction
2.2 Probability Density Function
2.3 The Normal Distribution
2.3.1 Binomial Distributions
2.3.2 Poisson Distribution
2.4 Student's t–Distribution
2.4.1 PDF and CDF for t–Distribution
2.4.2 Properties of t–Distribution
2.5 Let Us Sum Up
2.7 Glossary
2.8 Assignment
2.9 Activities
2.10 Case Study

• Understanding concepts and applications of the continuous probability
distribution
• Difference between probability mass function and probability density
function
• Applications of the continuous probability distribution
• Important properties and assumptions for continuous probability
distributions
2.1 Introduction :
In this unit, we will study the theory and application of continuous
probability distributions. We will see how these distributions help us to
take important business decisions about continuous business metrics like
time, money, weight etc. Through various examples from the business
world, we will understand the concepts of the probability density function
and cumulative density functions. In the end, we will understand the
assumptions and decision criteria to select the most appropriate distribution
as per the experimental data.
70
2.2 Probability Density Function : Continuous Probability
Distributions
"I believe that we don't know anything for certain, but everything
probably." – Christiaan Huygens
A Continuous Probability distribution can obtain any value in a
given range. Unlike a discrete variable, the probability of a particular
point in continuous distribution is always 0 because it is one point out
of infinite possible outcomes. Hence continuous probability can't be
represented in tabular form. There is always a formula or equation to
express a continuous probability distribution. This equation to describe
a continuous probability distribution is known as the probability density
function. It is a counterpart of the probability mass function for the
discrete probability distribution. Example of continuous variables is height,
length, money, weight, distance, time, temperature, blood pressure, etc.
As we can split these things into half infinite times.
2.3 The Normal Distribution :

Normal distribution is among the most important distribution in
statistics. It has a wide variety of applications. Most of the naturally
occurring measures in the real/ business world follow normal distributions.
To understand the probability density function (PDF) for normal
distribution, let's have sample data of the height of women working in
Ahmedabad. To see the shape of the distribution, an easy way to create
a histogram.
We can connect the top part of this histogram to see the approximate
shape of the probability density function. This bell shape graph represents
Normal Distribution.
Here we can see that mean of this height data is approx.–165 cm.
The Bell shape in front of the histogram is a pictorial representation of
the probability density function, but, in the probability density function
graph, Y–axis is not frequency like in the histogram; instead of that
Y–axis is always a probability. A normal distribution can be defined
by two parameters, (mean) and (standard deviation) of data.
Like we draw cumulative distribution function for the discrete
probability distribution; similarly, we can draw CDF in case of continuous
probability distribution also.
71
Business Analytics
The formula for PDF represents by f(x) and CDF represents F(x)
for the normal distribution is as follows :
2
1 x − µ 
1 −  
=f (x) e 2 σ  , − ∞ < x < + ∞
σ 2π
x 1  t − µ 2
1 −  
=F(x)
∫σ
−∞
2π
e 2  σ  dt, − ∞ < x < +∞
We can ignore these equations due to their complexity as we can

calculate these with the help of a standard normal distribution table and
excel formula. But we can notice that the above equations have only
two variables (mean) and (standard deviation).
We studied that any distribution can be transformed into standard
normal distribution :
X−µ
Z=
σ
This is a typical normal distribution with mean = 0 and standard
deviation = 1. The value of Z tells us that the calculated point is how
many standard deviations away from the mean.
Above standard normal distribution equation can be rewritten in
terms of random variable X :
X = + Z
 Important Properties of Normal Distribution :
1. The total area for a normal distribution is always 1 as it is a
continuous probability distribution
2. It is always symmetric bellshaped around its mean; it has a mirror
image about its mean
3. Theoretically, normal distribution never touches the x–axis; it is
defined from – to + . This property of probability distribution
is known as asymptotic
72
4. For any normal distribution, always area between specific values Continuous Probability
remain constant (in terms of and ) Distributions
5. Linear transformation of any normal distribution is also normal
distribution. If X is a normal random variable, then its linear
transformation AX + B (where A and B are constants)
6. Two independent normal distributions X1 and X2 with means 1
and 2 and variance 21 and 22 respectively. New transformed
distribution X1 + X2 will also follow the normal distribution with
mean 1 + 2 and variance 21 + 22
7. If we extract many samples from a normal distribution and draw

the distribution of their means, then this distribution of means is
likely to follow the normal distribution. This property is known as
the central limit theorem, will study the central limit theorem in
unit 3.
PDF in normal distribution tell us the height of the normal distribution
at different values of variable X so instead of using complex PDF formula;
An easy approach is that we
can transform variable X to a standard
normal distribution. Let's understand
this concept in more detail.
The entire area under standard
normal distribution is 1, and it is
symmetric about its mean.
Therefore half area under
standard normal distribution is always .50
 Steps to Solve Normal Distribution Problem :
• Step 1 : If we have data for variable X is given, then calculate
the mean and standard deviation of the data
• Step 2 : Draw standard normal distribution for better visualizing
the problem
• Step 3 : Calculate the Z score
73
Business Analytics • Step 4 : See the value of the Z score in the Standard Normal
Distribution (Z) table
• Step 5 : Answer will be in terms of probability, multiply it by
100 to get an answer in percentage
2.3.1 How to Check Z Score in Standard Normal Distribution Table
(Z Table) :
Suppose we want to see the value of z = –3.33 in the Z table.
Now we want to see the value of –3.33 in the Z table. Up to one decimal
place we have to see in the left–most column (under z) so –3.3 we will
see in the first column, and for the second decimal point, we have to
see in the first row. Hence probability value for z = –3.33 is .0004 which
means .04% data is left to z value –3.33 (area shown in below standard
normal distribution is .04% only). In other words, 99.96% data points
are right to z = –3.33 value (unshaded area).
 Excel Formula for Calculating PDF for

Normal Distribution :
As PDF for continuous distribution does
not make sense, therefore, we calculate CDF
only
NORM.DIST(X, , , TRUE). The formula
will provide direct probability (no need to
calculate z value and corresponding probability
from a standard normal distribution).
Example 2.2 : Below is a sample data time
taken by a bank employee to process the loan
file. Bank has set a target to complete the
processing of a file to 135 minutes.
Let's solve the below questions with the
help of the steps mentioned above to solve a
normal distribution example.
74
(a) How many files can be processed under 125 mins ? Continuous Probability
(b) How many files can be processed under 115 mins ? Distributions
(c) How many files processes took more than 140 mins
(d) How many files took time between 120 mins to 125 mins
(e) What should be the target that they should be able to complete
95% files within the set target
Solution : Here mean = 125 and standard deviation = 9
(a) Z (125
= = 9
− 125)
9
The value of 0 in the Z table is .5
hence 50% of files are Processed within
125 mins.
(b) (115 − 115)

Z= = − 1.11
9
Value of –1.11 is 0.1333 hence 13.33%
files processed under 115 mins
(c) P(X > = 140) = 1 – P(X < = 140)
1 − (140 − 125
Z =
9
= 1 – .9525 = 0.048
Hence 4.8% files process took more than 140 minutes.
(d) P(120 < = X > = 125)

= P(X > =125) – P(X > = 120)
= .5 – .2893 = .2107
Hence 21.07% of files took time between 120 mins and 125 mins
(e) Here we know the probability
beforehand hence we need to do the
reverse calculation
So we can see the Z table that for
which value of Z, we can see the
probability .95.
We can see that at z = 1.65, we are getting approx. 0.95
(X − 125)
1.65 =
9
X = 1.65 9 + 125 = 139.5 minutes
Therefore, if the bank will set 139.5 mins target then almost 95%
of files will be processed within the set target.
75
Business Analytics Check Your Progress – 1 :
1. For Standard normal distribution mean is always ___________ and
standard deviation is ___________.
2. Two independent normal distributions X1 and X2 with means 1
and 2 and variance 21 and 22 respectively. New transformed
distribution X1 + X2 will also follow the normal distribution with
mean ___________ and variance ___________.
2.4 Student's t–Distribution :

One of the limitations of the normal distribution is that we should
know the population standard deviation. Especially in the case of a small
sample. In statistics, t–distribution plays a very significant role. For a
random variable with mean X and S standard deviation of the sample.
Then the random variable t can be represented as :
t=
(X − µ)
s
n
above equation follows t–distribution with n–1 degree of freedoms. Here

the degree of freedom is lost due to the estimation of standard deviation
from sample data.
Please Note : Degree of freedom can be calculated as the number
of observations in the sample minus the number of estimates made using
the data (sample). For example, in the above t–distribution formula,
standard deviation needs to be estimated hence lost one degree of
freedom. Similarly, for any data, if we have to estimate mean and standard
deviation both then in that case degree of freedom will be n–2.
76
2.4.1 PDF and CDF for t–Distribution : Continuous Probability
PDF and CDF formula for t–distribution are very complex; hence Distributions
we will see only Microsoft Excel function to calculate.
MS Excel formula is T.DIST(x, degree of freedom, true). Here
x is the t value we have calculated with the above formula.
2.4.2 Properties of t–Distribution :
1. The mean of a t–distribution with 2 or more degrees of freedom
is always 0
n
2. The variance of t–distribution is for 2 or more degrees
(n − 2)
of freedom
3. As we increase the degree of freedom, PDF of t–distribution will
approach PDF for standard normal distribution
4. For a small sample, t–distribution has a bell curve shape, but it
will be fatter than the normal distribution. But if the sample increase
more than 30 observations, t–distribution start mimicking standard
normal distribution
5. t–distribution play an important role in hypothesis testing of means
of a population and comparing means of two populations
We will study about import properties of t–distribution in the next
unit, "Hypothesis testing" and in the next two blocks, "Regression analysis"
and "Time series analysis".
1. We apply t–distribution when population ___________ is unknown
and sample size is ___________.
2. Shape of t–distribution depends on ___________.
1. The weight of bags in a box are normally distributed with mean
µ and standard deviation of 6.5 given that 20% of bags are less
than 250 gms, the value of :
a. 255.46 b. 250.49
c. 244.54 d. None of the above
2. A normal distribution can be defined by :
a. Mean
b. Standard deviation
c. Both
d. Either mean or standard deviation
77
Business Analytics 3. The best scenarios to use t–distribution
a. Population standard deviation is not known
b. The population is normally distributed
c. The sample size is small
d. All of the above
4. t–distribution is defined by :
a. Mean b. Standard deviation
c. Degree of freedom d. None of the above
5. The total area under normal distribution :
a. It depends on observations b. 1
c. .5 d. None of the above
6. % of data points between the mean 2 standard deviations :
a. Approx. 68% b. Approx. 95%
c. Approx. 99.7% d. None of the above
7. Two standard normal distribution X1 and X2 with mean 1 and
2 and 2 respectively, then what we can say
2 and variance 1 2
about new distribution X1 + X2 :
a. X1 + X2 will also be normally distributed
b. Mean of X1 + X2 will be 1 + 2
c. Variance of X1 + X2 will be 2 + 2
1 2
d. All of the above
8. Mean of t–distribution with more than two degrees of freedom :
a. Always 0
b. Depends on the population mean
c. Depends on the sample mean
d. Always 1
9. The variance of t–distribution with more than two degrees of
freedom :
n n n
a. b. c. d. Always 1
2 (n + 2) (n − 2)
10. PDF graph of a standard normal distribution will vary from :

a. – to +
b. –3 standard deviation to +3 standard deviation
c. 0 to
d. – to 0
78
2.5 Let Us Sum Up : Continuous Probability
Distributions
1. Unlike a discrete variable, the probability of a particular point in
continuous distribution is always 0 because it is one point out of
infinite possible outcomes.
2. Continuous probability distribution is defined by an equation called
probability density function
3. A normal distribution can be defined by two parameters, (mean)
and (standard deviation) of data.
4. The standard normal distribution has zero mean and 1 standard
deviation
5. The standard normal distribution is always symmetric bell–shaped
around its mean; it has a mirror image about its mean. Theoretically,
normal distribution never touches the x–axis; it is defined from –
to + . This property of probability distribution is known as asymptotic
6. The degree of freedom can be calculated as the number of observations
in the sample minus the number of estimates made using the data
(sample).
7. The mean of a t–distribution with 2 or more degrees of freedom
n
is always 0 while the variance of t–distribution is for 2
(n − 2)
or more degrees of freedom
8. As we increase the degree of freedom, PDF of t–distribution will
approach PDF for standard normal distribution

1. Zero, one 2. + 2 2
1 2, 1 + 2
1. standard deviation, small 2. Degree of freedom
1. a 2. c 3. d 4. c 5. b
6. b 7. d 8. a 9. c 10. a
2.7 Glossary :
Probability Density Function (PDF) : PDF defines as the probability
of the continuous variable coming within a given range of values. It
generally tells us the height of the distribution at a given point
Cumulative Density Function (CDF) : Derivative probability density
function is known as the cumulative density function. It provides us with
the area from – to a particular point
Standard Normal Distribution : It is a special normal distribution
whose mean is always 0 and the standard deviation is always 1
79
Business Analytics Degree of Freedom : Degree of freedom can be calculated as the
number of observations in the sample minus the number of estimates
made using the data (sample).
t–distribution : t–distribution has a bell curve shape, but it will be
fatter than the normal distribution. It is defined by the degree of freedom,
for a large sample size its shape start mimicking like a normal distribution
2.8 Assignments :
1. What is the probability of a particular point in a continuous probability
distribution ?
2. How many data points occur between mean ± 1 standard deviation,
mean ± 2 standard deviation and mean ± 3 standard deviations
for a normally distributed sample data.
3. TRYC bank takes an admission test (TCAT) to hire freshers. Scores
on the TCAT are normally distributed with a mean of 353 and a
standard deviation of 80. What is the probability of an individual
scoring above 250 in the TCAT exam ?
2.9 Activities :
The teacher surveyed 200 students to know how many hours
students study Business Analytics per day. The study shows a sample
mean of 68 minutes and a standard deviation of 12 minutes. Assume
study hours follow the normal distribution.
(a) How many students study Business Analytics less than 90 minutes
per day
(b) How many students study less than 60 minutes
(c) How many students study between 50 minutes and 80 minutes
Ans. (a) 25.46% – approx. 51 students
(b) 96.64% – approx. 193 students
(c) 77.45% – approx. 155 students
2.10 Case Study :
There are approx. 5000 tickets raised for a technical helpdesk in
an Akshay software company. The technical helpdesk team resolves a
ticket in 25 hours with a standard deviation of 3 hours. Let's say the
technical helpdesk manager said that they would award ` 100 to associates
whose tickets will take more than 28 hours. Calculate the monthly
technical helpdesk expense due to penalty.
Question :
Now manager calculates the expense and realized that it is going
out of budget, suggest to him the penalty threshold so he should not
pay penalty to more than 2% of associates.
80
2.11 Further Readings : Continuous Probability
Distributions
Carmer H. (1946)
• "Super Freakonomics," Penguin Presss, Levitt S. D. and Dubner
S. J. (2009)
F. W. (2001)
• "Data Strategy : How To Profit From A World Of Big Data,
Analytics And The Internet Of Things", O'Reilly Media,
Bernard Marr
81
Business Analytics
Unit SAMPLING AND
3 CONFIDENCE INTERVALS
: UNIT STRUCTURE :
3.1 Introduction
3.2 Introduction to Sampling Process
3.2.1 Important Steps in Designing a Sampling Strategy
3.3 Sampling Methods
3.3.1 Probabilistic Sampling Methods
3.3.2 Non-Probabilistic Sampling Methods
3.4 Central Limit Theorem
3.5 Confidence Interval
3.6 Let Us Sum Up
3.8 Glossary
3.9 Assignment
3.10 Activities
3.11 Case Study

• Understanding the requirements of sampling and its importance in
business decisions
• Different types of sampling methodologies and techniques
• Significance of central limit theorem and its applications
• Understand the concept of interval estimate
• Application of confidence interval and confidence level
3.1 Introduction :
In this unit, we will study samples and their importance in the
business world. Sampling helps in making decisions faster also reduce
the cost of the experiment. There are different types of sampling
methodologies as per the business scenarios. We will also see how
estimating a range of values know as confidence intervals are better than
point estimation (average, variance etc.). In the end, we will see the
importance of the confidence level of a study and how it influences our
study in terms of its effectiveness and efficiency.
82
3.2 Introduction to Sampling Process : Sampling and
Confidence Intervals
"Statistical analysis in cases involving small numbers can be
particularly helpful because on many occasions intuition can be highly
misleading." – Sandy Zabell
Sampling is one of the important tools in the business world as
it is directly aligned with the three most important business metrics –
cost, efforts and time. In the real business world, it is very difficult to
analyze entire data even if we have access to it because accessing entire
data is very expensive and time–consuming. A correct sampling
methodology by fulfilling all its assumptions can help us to complete
endeavours within time and budget.
Sampling is a technique by which we collect only a few data points
from the population to reveal information and insights about the population
parameters like average, standard deviation, variance, proportion etc.
Most of the time, sampling works as two edge sword where right
sampling can save huge money and time, but wrong sampling can result
in disastrous results. Sampling involves various steps; hence it is important
we have to be very cautious at each step to get desired results.
3.2.1 Important Steps in Designing a Sampling Strategy :
Fig 3.1 Important Steps in Sampling Process

To extract the right sample from a population, we need to follow
the above five–step sequential process. Sometimes people think, a larger
sample will lead to better predictions, but it is not correct. If we have
not followed the right process, then a larger sample will give us more
misleading results. Below are important steps in designing a robust
sampling methodology :
1. Identify the Right Population as per Business Problem : It is
utterly important that we should identify the right target population;
the definition must be written and aligned with the business problem.
For example, we want to understand the feedback about a restaurant
from young diners. In this case term "young diners" is vague, it
must be mentioned that diners between the age of 24 to 30 years
old.
2. Select the Sampling Frame : The sampling frame describes the
source from where samples have to be extracted; this can be a
customer database, directory, company's employees' database, social
media portals like Zomato, eat panda etc. It may also be possible
that we can use more than one sampling frame for our study.
83
Business Analytics 3. Select a Suitable Sampling Technique : The right sampling technique
plays an important role in achieving the research objective. Various
sampling techniques broadly can be distinguished between
probabilistic and non–probabilistic techniques. We will study these
techniques in the next section.
4. Determine the Right Sample Size : Data collection can be expensive
and time–consuming; hence it is very important to calculate the right
sample size, which is sufficient for achieving the research objective
within budget and time frame. Sample size depends on the required
level of confidence, effect size, variation and margin of error.
5. Execute the Sampling Process : All the above steps must be
executed as the right process, and we must be agile to keep it aligned
with the business problem for which we are drafting this sampling
strategy. Proper execution requires a nicely identified population,
correct sampling frames, appropriate sampling technique and sufficient
sample size.
3.3 Sampling Methods :

There are two types of the sampling methodology, probabilistic and
non–probabilistic. Which sampling method will be most appropriate always
depends on the research objective, constraints etc. Below are various
sampling techniques :
Fig 3.2 Sampling Techniques Tree

84
Portability based sampling techniques are based on probability Sampling and
distributions. Below are important probabilistic sampling techniques : Confidence Intervals
3.3.1 Probabilistic Sampling Methods :
1. Random Sampling : Random sampling is the most popular and
frequently used sampling technique; here, each observation from the
population has equal chances to get selected as part of the sample.
This is a very good method when the population is homogenous means
all observations are of the same kind. For example – from a bulb box
of 100 bulbs, we pick up 10 bulbs to check (quality inspection). There
is two type of random sampling – with replacement and without
replacement. But if we have some type of group different like the
population of a city, then random sampling can be a dangerous pick.
Exit polls companies do not pick their sample through random
sampling techniques or Mobile Phones Company also do not pick their
sample customers for feedback on a random basis.
2. Stratified Sampling : In stratified sampling we divide the population
into some mutually exclusive groups based on some factors, for
example, the population of a city can be divided by urban areas,
low–income group areas, slums, middle–income group areas etc.
or population of a city can be a divide based on their education
level, age, marital status or type of their jobs etc. The clusters
formed in stratified sampling are known as strata.
Within each stratum, then we should apply the random sampling
method as now data points within each stratum are homogenously
distributed. Another important point is we should see the proportion
of each stratum in the population and in that proportion only we
should extract the sample.
Below are important steps in designing stratified sampling :
1. Identify the strata in the population, in the figure on the left we
have considered age as a strata
2. Calculate their proportion in the population
3. Calculate the total sample size, now recalculate the sample size
for each stratum in the same proportion these are in the population
4. Create the final sample by combining individual samples of each
stratum
3. Cluster Sampling : Cluster
sampling also divides the
population into a few groups as
we did in the case of stratified
sampling. An important difference
between these two types of
sampling techniques is that in
stratified sampling, we represent
all strata in the sample while in
Fig 3.3 Cluster Sampling
the cluster, we can select a few
85
Business Analytics clusters only. Second, in stratified each stratum is homogenous while
in cluster sampling each cluster is generally heterogeneous (not in
similar nature) hence generally we select few important clusters than
all because of few reasons contribute significantly than other hence
it is making sense to understand few important ones by analyzing
their samples, it saves cost and very convenient.
4. Systematic Sampling : Here we pick up samples after a regular
interval. For example, products are moving on the production belt
in a manufacturing firm, and we are picking up every 20th product
for quality checking.
5. Bootstrap Aggregating Sampling (Bagging) : It is a very useful
sampling technique generally used in machine learning. Here we
do sampling with replacement. Bagging, generally used in the case
where we develop various models based on each sample (with
replacement) and final prediction made based on voting.
3.3.2 Non–Probabilistic Sampling Methods :
1. Convenience Sampling : Here researcher used samples based on
his/her convenience; for example, he is picking up samples from
his organization or locality. It is generally a time–saving approach
where the result does not get impacted due to biased sample size.
2. Judgement Sampling : The researcher selects the sample based
on their judgement, again this is a very quick and cost–efficient
approach, but as it is also a non–probabilistic approach hence it
is very difficult to calculate sampling error (variation in samples).
Another drawback it is not comparable with any other researcher
as everybody may have a different opinion about the scope of the
study.
3. Quota Sampling : Quota sampling is quite close to stratified
sampling. Here again, the quota is selected based on a few strata
like education qualification, age, gender etc. but in quota sampling
researchers do not use a probabilistic approach to gather data for
each stratum. The quota for each stratum is also in the same
proportion these are in the population. Quota sampling is important
with time and budget are constraints.
4. Snowball Sampling : In the case of snowball sampling, respondents
are chosen as per referrals from other respondents. When a researcher
wants to collect information beyond his/her domain, then researchers
want to meet with somebody who has good knowledge about that
subject and further leads also get from the first respondents. It is
a very effective way to gather information where we collect
information in terms of the survey.
86
Check Your Progress – 1 : Sampling and
1. In which sampling technique, samples are picked up after a regular Confidence Intervals
interval :
a. Quota sampling b. Systematic sampling
c. Cluster sampling d. Snowball sampling
2. Bootstrap sampling is a type of ___________ sampling techniques
while quota sampling is a type of ___________ sampling technique.
3.4 Central Limit Theorem :

The central limit theorem plays a very important role in entire
statistics due to its wide applications in almost every field of statistics,
especially in hypothesis testing. It states that if a population is normally
distributed (if the histogram is symmetric about mean then distribution
said to normally distributed, we will study about various probability
distributions in next chapter) and we take few samples of size n then
means of these samples will always be normally distributed. The Centre
limit theorem works for all types of distribution, either continuous or
discrete. An important assumption for the centre limit theorem is that
samples are extracted from identical distribution, and samples drawn are
independent of each other. Let's see the below simulation to understand
the centre limit theorem :
We want to plot the outcome of a six–phase dice throw; we are
considering a very small sample size of 4. Let's say the first sample
is [2, 4, 5, 5], and its mean is 4. Similarly [2, 4, 4, 5] and its average
is 3.75, in the same way, we extracted means of 10 samples, below are
10 sample means :
4
3.75
5.75
4.25
3.75
3.5
3.25
1.75
2.5
4
Fig 3.4 Histogram of 10 Sample Means

Now let's see the shape of distribution post 30 means
87
Business Analytics

Now we are seeing, the shape of the distribution is shaping up like
normal distribution. Below are histograms of samples 150 and 16,000.
Fig 3.7 Histogram of 16,000 Sample Means

Therefore, we saw, as we are increasing the size of our sample,
we are getting perfect normal distribution. This is true for any type of
population distribution. There are below very important implications of
the central limit theorem :
88
1. A bigger sample leads to a smaller spread. We saw when we had Sampling and
only 10 means in the sample; our bars were varying from 1 to Confidence Intervals
6 while in the case of the 150 means sample, most of the data
points are varying from 2.3 to 4.9. Irrespective of the type of
population distribution, a relatively large sampling distribution
(n > 30) will follow the normal distribution. This distribution will
have mean same as the population meanwhile standard error (standard
σ
deviation of the sample) will be
n
2. The spread of sampling distribution will be lesser than the spread
of population distribution (from which this sample has been drawn)
(X − µ )
3. The variable
σ
will always have a
mean 0 and standard
error = 1. This
distribution of mean 0
and standard deviation
1. This distribution is
known as the standard
normal distribution.
This calculation is true
for any dataset. If we subtract
the entire data set from its
mean and divide it by
Fig 3.8 Stanardization
standard deviation then for
transformed data, the mean will always be 0, and the standard deviation
always is 1. This is known as the Standard Score.
This helps us to compare different datasets which are measured
in different units like one data is in Kgs while another is in currency.
When we compare two datasets with the help of a standard score
is known as standardization as we standardize all data points. Each data
is no standard deviation away from its mean. For example, – .05 means
.05 standard deviation less than its mean.
3.5 Confidence Interval :

"Confidence comes not from always being right but from not fearing
to be wrong." – Peter McIntyre
We studied sampling and sampling error (standard deviation of the
sample divided by the root of sample size) in the last section, these two
things are building blocks for the formula of the confidence interval.
Let's say we are studying the weight of Kesar mangoes from Saurashtra
as we can't measure all mango's weight hence we take a sample and
measure their weights, but the important thing to note is that we do our
89
Business Analytics study on the sample while we conclude decisions about the population.
So we will extract different samples of mangoes, then every time our
average weight will be different. This is called sampling error or variation
due to sampling. We can't overcome the sampling error, here confidence
intervals help us in this situation. Instead of mean, we should give a
confidence interval of the mean. A confidence interval tells us how precise
(accurate) our sample estimates are. In other words, the confidence
interval states the range in which the weight of mangoes will be.
Confidence interval depends on two factors : variation in the
population and sample size. If all mangoes have very little variation in
their weight then our sample will also have very little variation hence
our confidence interval will be narrow, which means more confidently
we can say that weight of mangoes will vary in this small range. The
second factor on which the width of the confidence interval depends is
the sample size. If we have a small sample, then our sample will not
nicely represent our population; hence we will be less confident about
our inferences. If we extract different samples, then every sample will
tell us quite a different average weight. In contrast, larger samples will
be more similar to each other. The effect of sampling error will be reduced
in the case of larger samples.
We don't know the variation of the population; hence we estimate
with the help of variation in the sample. As we keep increasing our sample
size, we keep getting more information about the population; hence our
confidence interval keeps getting narrow, which means we are more
confident about our inferences. So, if we want to be more confident (say
90% confidence level), then our confidence interval will be wider (we
can say more confidently that the weight of mangoes varies from 125
gms to 175 gms. But if we want to be more confident (confidence level
95%) then our confidence interval will be even wider (let's say 110 gms
to 190 gms). Somebody can say that I am 99.99% confident that the
weight of mango will vary from 10 gms to 400 gms, but that kind of
confidence interval does not make any business sense.
Note : Confidence level tells us how confident we want to be about
our inferences (study output). It is generally taken as 90% or 95%, which
means we want to be 90% or 95% confident about study inferences. 100%
– confidence level is known as , alpha (error %). In other words,
we are ok with a 5% error in our estimates or study output. Let's
understand it with the help of confidence interval for the mean formula :
S S
Confidence Interval= X − t α / 2, n −1 ≤ µ ≤ X − t α / 2, n −1
n n
Here the value of t depends on sample size and confidence level;
there would be a wider confidence level for bigger values of t. Here we
are using t /2, n–1 because of the two–tailed test, in unit 3 we will study
one–tailed or two–tailed tests in detail. Below are some important t values
for different confidence levels and sample sizes.
90
• In the above formula Sampling and
X is sample mean Confidence Intervals
• t × s n is called the
margin of error
• The confidence
interval provides us with the range we are quite sure of (as per
confidence level) for the mean of our population.
• At 95% confidence level, the confidence interval does not tell us
that 95% of mangoes will weigh within this range rather it will
tell us that there are 95% chances that the average weight of
mangoes will lie in this range.
• Therefore if we take lots of samples and create 95% confidence
intervals for them, then we can be assured that 95% of them will
contain the true population mean although 5% of these confidence
intervals will not contain the true population mean
• The confidence interval can be calculated for any population measure
(parameter) like median, standard deviation etc
• Excel formula for calculating confidence interval with the help of
excel is CONFIDENCE.T (alpha, standard deviation, size). The
value of t /2, n–1 can be calculated using the excel formula
T.INV( /2, n–1) or another excel formula T.INV.2T( , n–1).
There is another formula in excel TINV( , n–1) while
TINV(2 , n–1) provides value for a one–tailed T–test.
Example 3.1 : A sample of 100 students from the Business Analytics
class was taken to estimate their daily study hours. The sample mean
is 4.8 hours and the standard deviation of the population is given as
1.4 hours. Solve the below questions :
(a) 90% confidence interval for the population mean
(b) 95% confidence interval for the population mean
Solution :
(a) Confidence interval at 95% confidence level
Here we have n = 100, X = 4.8, s = 1.4, t0.025,99 = 1.98
Confidence interval = X ± t α / 2,n − 1 × s / n
=4.8 ± 1.98 × 1.4 / 100 =4.8 ± .28

Therefore 95% confidence interval for the study hours is (4.52,
5.08)
(b) Confidence interval at 90% confidence level
Here we have n = 100, X = 4.8, s = 1.4, t0.05,99 = 1.66
91
Business Analytics
Confidence interval = X ± t α / 2,n − 1 × s / n
=4.8 ± 1.66 × 1.4 / 100 =4.8 ± .23

Therefore 95% confidence interval for the study hours is (4.57, 5.03)
An important point to observe is that we can conclude a narrow
confidence interval (4.57, 5.03) at a relatively low confidence level (90%)
while when we are increasing confidence level to 95% confidence interval
becomes wider (4.52, 5.08).
1. In the formula of the confidence interval, term is known as :
a. Error % b. Confidence level
c. Sampling error d. Margin of error
2. Confidence interval and confidence level are ___________
proportional to each other.
1. What is the main objective of the analysis of sample instead of
population :
a. Reduce cost b. Reduce time
c. Reduce resources d. All of above
2. Which of the following is a probabilistic sampling method
a. Convenience sampling b. Snowball sampling
c. Quota sampling d. Stratified sampling
3. Which of the following is a non–probabilistic sampling method
a. Random sampling b. Quota sampling
c. Cluster sampling d. Stratified sampling
4. If there are 2000 employees in an organization and the mean height
is 168 cm, then the estimated mean of the sampling distribution
will be approx.
a. 168 cm b. 168 / 200 cm
c. 168 200 cm d. None of the above
5. Which of the following is not a valid assumption of the central
limit theorem is
a. Samples are drawn from the identical distribution
b. Samples are independent of each other
c. None of the above
d. Both (a) and (b)
92
6. When we increase the confidence level, how does it impact the Sampling and
confidence interval Confidence Intervals
a. Confidence interval gets narrow
b. Confidence interval gets wider
c. Does not impact the confidence interval
d. None of the above is a correct statement
7. For a standard normal distribution
a. The sample mean is always higher than the population and the
standard deviation is lesser than the population
b. Mean is always 50 and the standard deviation is always 10
c. Mean, and standard deviation always depends on the population
from where the sample extracted
d. Mean is always ZERO while standard deviation is always ONE
8. The relation between significance (alpha) and confidence level is :
a. Both are valid only for a two–tailed test
b. The summation is both is always 100%
c. Both are valid only for a one–tailed test
d. None of the above
9. One of the below factors does not impact the width of the confidence
interval :
a. Population standard deviation b. Sample size
c. Sample mean d. Confidence level
10. How the width of the confidence interval will change. If we increase
sample size and also increase the confidence level at the same time
a. We cannot conclude anything about the width of the confidence
interval
b. The length of the confidence interval will be increased
c. The length of the confidence interval will be decreased
d. The length of the confidence interval will remain the same
3.6 Let Us Sum Up :

1. It is very difficult to analyze entire data (population) even if we
have access to it because accessing entire data is very expensive
and time–consuming therefore we use sample instead of entire
population data
2. There are two types of the sampling methodology, probabilistic and
non–probabilistic. Which sampling method will be most appropriate
always depends on the research objective, constraints etc.
3. If we subtract the entire data set from its mean and divide it by
standard deviation then for transformed data, the mean will always
93
Business Analytics be 0, and the standard deviation always is 1. This is known as the
Standard Score.
4. Point estimation of a population parameter gives us a unique value
while the confidence interval provides us with the range, we are
quite sure of (as per confidence level) for the mean of our population.
Hence analysts prefer confidence interval for estimation
5. The confidence level tells us how confident we want to be about
our inferences (study output). It is generally taken as 90% or 95%,
which means we want to be 90% or 95% confident about study
inferences.

1. b 2. Probabilistic, Non–Probabilistic
1. d 2. Inversely
1. d 2. d 3. b 4. a 5. c
6. a 7. d 8. b 9. c 10. a
3.8 Glossary :
Sampling : Sampling is a technique by which we collect only a
few data points from the population to reveal information and insights
about the population parameters like average, standard deviation, variance,
proportion etc.
Probabilistic Sampling Methods : When the researcher picks up
a sample from the population based on probability theory. e.g. random
sampling, stratified sampling, bootstrapping, systematic sampling, cluster
sampling etc
Non–Probabilistic Sampling Methods : When the researcher picked
sample without any probability–based method e.g. snowball sampling,
Convenience sampling, judgement sampling etc
Center Limit Theorem : It states that if a population is normally
distributed and we take few samples of size n then the means of these
samples will always be normally distributed.
Standardization : When we compare two datasets with the help
of a standard score, it is known as standardization
Confidence Interval : It is a measure of certainty in terms of a
probability value that population parameter will fall within a range of
values around the mean
94
3.9 Assignments : Sampling and
Confidence Intervals
1. Write down important steps involved in designing a sampling
strategy.
2. In an organization, there is a total of 3,000 employees. A random
sample of 90 engineers reveals that the average sample age is 29
years. Historically, the population ? of the age of the company's
engineers is approx. 7 years. Construct a 95% confidence interval
to estimate the average age of all the employees in this company.
3. What are the important differences between probabilistic and non-
probabilistic sampling techniques ? Write important techniques
under each of these categories.
3.10 Activities :
A sample of 70 customers from an online shopping company was
taken. The sample mean is 24 transactions as per their transaction history
and the standard deviation of the population is given as 35. Calculate
a 95% confidence interval for the population mean.
Ans. : (23.09, 24.91)
3.11 Case Study :
ABC Bank limited is ranked 6th in an important customer satisfaction

survey conducted by a market analysis firm. Bank wants to penetrate
further into the semiurban and rural part of Southern India. Management
put customer satisfaction as the focus strategy for next year to attract
more customers in these new territories. One of the important customer
concerns is the time taken to solve their queries. Analysts have extracted
a sample from the database (assume the population is following a normal
distribution). The variance of resolution time is 144 mins based on 750
complaints. Management wants to understand the additional requirement
of manpower so that problems can be resolved within promised time
to resolve a problem.
Questions :
(a) Calculate the 95% confidence interval for the resolution time.
(b) Calculate the 90% confidence interval for the resolution time.
95
Business Analytics 3.12 Further Readings :
Carmer H (1946)
• "Super Freakonomics," Penguin Press, Levitt S D and Dubner
S J (2009)
F W (2001)
• "Data Strategy : How To Profit From A World Of Big Data,
Analytics And The Internet Of Things", O'Reilly Media,
Bernard Marr
96
4 HYPOTHESIS TESTING
: UNIT STRUCTURE :
4.1 Introduction
4.2 Life Cycle of Hypothesis Testing
4.2.1 Hypothesis Testing Process Steps
4.3 Hypothesis Test Statistics
4.4 Two–Tailed and One–Tailed Hypothesis Test
4.5 Concept of p–Value
4.6 Type I, Type II Error and Power of the Hypothesis Test
4.7 Hypothesis Testing for a Population Mean with Known Population
Variance : Z–Test
Variance : t-Test
4.9 Let Us Sum Up
4.11 Glossary
4.12 Assignment
4.13 Activities
4.14 Case Study
4..15 Further Readings

• Understanding the basics of hypothesis testing and its applications
in decision making
• Learn to set up a hypothesis test and concluding results in terms
of business context
• Understand the concept of significance ( ) and Type 1 and Type
2 error
• Understand the significance of p–value and its application in
concluding hypothesis test
4.1 Introduction :
In this unit, we will study the concept of hypothesis testing and
how it helps us to make robust decisions about the future based on a
study conducted on sample data. We will see the various types of errors
associated with hypothesis testing techniques. We will also understand
97
Business Analytics the concept of p–value and its application in decision–making with the
help of various examples.
4.1 Life Cycle of Hypothesis Testing :

"Beware of the problem of conducting too many hypotheses as more,
we torture the data, more likely they are to confess, but confession under
duress may not be acceptable in the court of scientific opinion"
– Stephen Stigler
Hypothesis testing is one of the most critical tools in statistic's
quiver. We can't analyze the entire population because of time, and budget
constraints; so most of the time we analyze the sample only then the
hypothesis test runs on those results in order to conclude whether the
results will remain valid for the population or not. One of the metrics
to check the authenticity of any analysis is whether results will remain
valid for unseen data.
A hypothesis can be understood as a claim made by someone. The
claim is generally regarding the population parameters like mean, standard
deviation or proportion etc. by seeking evidence from the sample study. For
example, the average time to grant a home loan is 22 days. Hypothesis testing
is a process that rejects or retain this claim by analyzing the sample data.
 Null and Alternate Hypothesis Statements :
The null hypothesis, H0 is generally the commonly accepted fact
which researchers want to disprove with their research data. We are going
to believe that null hypothesis statement unless it is proved wrong. Null
does not mean "nullify", but it means "nothing going to change or things
will remain the same" as these were before the research. Opposite to
null hypothesis (H1/Ha) is known as alternate hypothesis statement. These
two statements are mutually exclusive and exhaustive statements which
mean collectively both these statements cover all possible scenarios.
Below are different examples of null and alternate hypothesis statements
1. AZH company wants to check whether new training helped them
in reducing call handling time significantly or not :
H0 : There is no change in average call handling time after a
training
H1 : Average call handling time changed significantly after training
2. New engine helped in improving the mileage of a car brand
H0 : µ < = 15
H1 : µ > 15
3. A retail store wants to see whether there is a relation between time
spent in–store and gender
H0 : There is no relation between time spent in–store and gender
H1 : Avg time spent in the store is significantly different for one
gender than other
98
Writing a correct null and alternate hypothesis is an important step Introduction to
as it set the research objective clear and provide direction towards data Hypothesis Testing
collection, sampling methodology and tool and techniques applied to
analyze the collected sample data. We will see it more during problem–
solving in the next section.
Hypothesis testing generally plays two roles in analytics or research,
it helps to find the new hypothesis for the research as we found during
root cause analysis that asking and processing of unnecessary documents
for a home loan cause the delay in granting the home loan to the
customers. This is a new finding (which we call hypothesis in statistical
term) that need to prove with the help of sample data collected. Or we
already have a hypothesis like a furniture shop claim that the height of
their particular model wardrobe is 175 cm.
Here the furniture shop claimed
that the height of their particular model
wardrobe is 175 cm. Again, we can
validate this claim with the help of data
collected for a few wardrobes. So how
hypothesis testing helps us to validate
these claims. It can be written in terms
of hypothesis statements :
H0 : = 175
H1 : 175 Fig 4.1 Rejection Region
If our sample mean is 170 cm, can we say that shop's claim is
valid ? Or if our sample mean is 165 cm, then the shop's claim is not
valid ? So, at what point we can say that our sample mean is close to
the population mean (claim) or quite far away from the population mean.
Hypothesis testing techniques help us to remove this ambiguity from
rejecting or retaining hypothesis claims. We know that if we have extracted
the right sample (by following all prerequisites of sampling theory), then
our sample means should be equal to the population mean. We know
there is always some random variable due to which it may not be exactly
equal to the population mean, so hypothesis tests help us to draw a
boundary beyond which we can say that sample mean is different
(significantly) from the population mean.
99
Business Analytics 4.2.1 Hypothesis Testing Process Steps :
There are eight important steps in the hypothesis testing process :
Fig 4.2 Hypothesis Testing Process Steps

1. Define Problem Statement : Understand the business problem and
covert that business problem into a statistical problem. Describe
hypothesis clearly in terms of population parameters like mean,
variance, standard deviation, proportion etc. Write clear sentences
to avoid ambiguity in understanding the scope of the analysis.
2. Write Up Hypothesis Statements : We studied that the null
hypothesis is considered to be true until it is not proved wrong
with the help of evidence. We should try to write hypothesis
statements in terms of the equation if possible.
3. Determine the Suitable Statistical Test : The researcher has to
identify the most suitable hypothesis test to conduct statistical
analysis. Here we have to identify the test statistic to test the validity
of the null hypothesis. It depends on the probability distribution
that sample data follows. We will study the test statistics in the
next section.
4. Set the Level of Significance : As we conduct statistical analysis
on sample data; hence there will always be an embedded risk in
our analysis's result, one type of error is the level of significance
or Alpha ( ), which is also known as the significance level. P
is the probability of alpha error. Alpha error is also known as False
positive. It is the error of rejecting the null hypothesis statement
when, in fact, it is TRUE. For example, we are saying a healthy
person is sick.
The value of significance is always set by the researcher before
collecting data. 100% – significance level is called the confidence level.
In other words, the researcher said that I want to be 95% sure about
my research result as we cannot be 100% sure from sample analysis.
So it always depends on the researcher how stringent he/she wants to
be sure about his analysis result.
Alpha is the highest level of p that we are willing to tolerate and
say that a change/difference in the sample is "Statistically significant".
100
Introduction to
Hypothesis Testing
Fig 4.3 Probability of Error

As we increase the confidence level, the required sample size will
also increase exponentially. Conventionally, researchers take confidence
level at 95% or 5% alpha error. Reducing alpha also increase the
error; we will study error in the next section.
Alpha and p are cumulative
probability, the area under the
curve.
Left–hand side curve
showing at 5% for a two–tail
Fig 4.3 Critical Error
test. For one tail test entire 5%
will be a shaded area on one side only. This shaded area
Alpha defines the critical
values for test statistics like t,
z etc. The below diagram
shows the relationship between
critical value and confidence
level.
Let's say our sample of
height has a mean of 175 cm.
In the last section, we saw these
calculations in detail. By
converting the Z value of
Fig 4.4 Critical Value and
± 1.960 into real value in cm,
Confidence Limits
we will get the confidence
interval for the mean. It is coming out 170 cm and 180 cm for the left
and right sides of the confidence interval.
5. Set the Appropriate Decision Rule : The researcher selects the
value of , which we calculate the critical values. If the computed
value of test statistics falls beyond critical values, then we reject
the null hypothesis statement; otherwise, we fail to reject the null
hypothesis (accepted).
101
Business Analytics
Fig 4.5 Decision Rule in Hypothesis Testing

6. Identify Sampling Methodology and Collect Sample Data : At
this stage, we apply the appropriate sampling methodology and
calculate the sample size required statistically and then collected
data used for calculating the test statistic. We should have completed
the first five steps before starting the data collection. We already
discussed the sampling methods for data collection in the last unit.
7. Calculate Test Statistics and Write Statistical Conclusions : We
calculate the test statistic; here important steps are selecting the
appropriate probability distribution like for a sample size less than
30 we should use t–distribution instead of z–distribution.
8. Write a Report in Terms of the Business Implications of
Hypothesis Testing : The researcher concludes based on the
hypothesis testing result. We reject or fail to reject the null hypothesis
testing statement and convert this statistical statement into business
language and draw the predictions accordingly.
4.3 Hypothesis Test Statistics :

We create a probability density function for sample difference
consider the null hypothesis is true (? = 0). Its mean is always zero
and standard deviation = 1 as it is a standardized difference between
the estimated value of parameter being tested calculated from the sample
and the hypothesis value. It is a standardized difference between X and
if we are testing the mean).
A test statistic is a standardized value used for calculating the
p–value (probability) in support of the null hypothesis. We set the value
of ? as per our risk–taking ability; then we draw the critical value beyond
which we have rejection region(s). Here in the below graph, we considered
= 5%. As below is the two–tailed Test; hence we will calculate two
critical values one for the left side and the other for the right side. We
know each side there is a .025% region and if we see the value of .025
102
in the Z table, then the .025 value comes for Z = – 1.96. Similarly .9750 Introduction to
comes for + 1.96. Therefore ± 1.96 are critical values at both sides, and Hypothesis Testing
beyond these rejections, the region starts.
Fig 4.6 Test Statistic

Test statistics has an associated probability distribution. There are
four commonly used test statistics : z, t, F and 2 (Chi–Square). We will
z and t and in this chapter in upcoming sections. A higher value of test
statistic indicates that the sample is likely to be more accurate as a
representative of the population.
In case the calculated value of the test statistic is greater than
the critical value (test statistic value lies in the rejection region), we
conclude that there is a statistically significant difference. We reject
the Null Hypothesis statement.
4.4 Two–Tailed and One–Tailed Hypothesis Test :

A researcher wants to check the hypothesis that the productivity
of a Surat factory of a toy manufacturing company is different from their
Vapi factory. µSurat and µVapi are the average productivity of these two
locations. In this scenario, the rejection region can be either side of the
mean. As the rejection region is both sides of the distribution; hence,
it is a two–tailed test. Hypothesis testing statements can be written as
below :
H0 : Surat = Vapi
H1 : Surat Vapi
Fig 4.7 Two–Tailed Test
103
Business Analytics In scenarios where the rejection region lies only on one side of
the distribution, we called it a one–tail test. If the rejection region lies
on the left side, then we call it a left–tailed Test otherwise if the rejection
region lies only on the right side then we call it a right–tailed test. Below
are the examples :
For a right–tailed test, the average salary of a business analyst in
an organization is at least ` 65,000
H0 : salary <= 65000
H1 : salary > 65000
Fig 4.7 Right Tailed Test

For the left–tailed Test, the average salary of a business analyst
in an organization is lesser than ` 65,000
H0 : salary >= 65000
H1 : salary < 65000
Fig 4.8 Left Tailed Test

In Microsoft Excel NORM.S.INV( ) can be used to find the critical
value for a left–tailed test while NORM.S.INV(1– ) for the right–tailed
Test. NORM.S.INV( /2) and NORM.S.INV(1 – /2) for a two–tailed
test.
4.5 Concept of p–Value :

In layman terms, the p–value is the probability value that
indicates how likely it is that a result occurred by chance alone. The
p–value is a conditional probability of observing the sample statistic
value when the null hypothesis is true. P–value (probability) is
evidence in support of the null hypothesis statement.
104
Let's try to understand it with the help of an example. The average Introduction to
salary of young analysts in the Analytics and Insight department of Hypothesis Testing
company Circa Ltd. is at least ` 55,000.
Here, H0 : <= 55000
He has collected sample salary information from the Human resource
department and found that the sample mean ( X ) is ` 55,000. Suppose
the standard deviation of the population is known, and the standard error
 σ 
of the sampling distributions is 2500  = 2500  .
 n 
The standardized difference between the estimated sample mean
value and hypothesized salary is :
(55000 − 50000)
=2
2500
Now we will find the probability of calculating this sample mean
if the null hypothesis is true. Remember, a large standardized distance
between hypothesized mean and sample mean will result in a low p–
value. This is a right–tailed test. Therefore, we will see Z value in the
Z table (.9772) which is an area up to Z = 2 from left most point, but
we are interested in calculating area beyond Z = 2 hence our interested
p–value is 1–.9772 = .0228
Fig 4.9 Decision Criteria

Important to note that we never say "Null hypothesis accepted"
as we cannot prove it in all possible scenarios. We just find evidence
where the alternate hypothesis is not valid (it may be possible that
the alternate hypothesis statement is not valid in a few scenarios while
it may be valid in other scenarios). As soon as we find enough evidence
against the alternate hypothesis statement, we declare that we fail
to reject the null hypothesis.
4.6 Type I, Type II Error and Power of the Hypothesis Test :

Hypothesis testing results in below two decisions :
1. Reject the null hypothesis statement
2. Retain the null hypothesis (fail to reject)
Below two types of error can happen during concluding hypothesis
testing :
105
Business Analytics 1. Type I Error : It is a conditional probability of incorrectly rejecting
the null hypothesis testing. Here we incorrectly support the claims
made by the alternate hypothesis; therefore, it is also known as
False Positive. The significance value is the probability of type
I error.
Type I Error ( ) = P (Rejecting null hypothesis | H0 is true)
Sometimes students are confused between significance value ( )
and p–value. Significance value ( ) is an error due to repetitive
sampling, while p–value is the evidence for the null hypothesis.
Type 1 error is also known as the producer's risk as a quality team
rejects a good product or services indicated by the null hypothesis.
2. Type II Error : It is a conditional probability of retaining a null
hypothesis when the alternate hypothesis is true; therefore, it is also
known as False Negative. It is represented by .
Type II Error ( ) = P (Retain null hypothesis | H0 is false)
The power of the hypothesis test can be calculated as 1 – .
Type II error is also known as consumer's risk as a quality team
pass a faulty product. Later this faulty product will be purchased
by the consumer.
Fig 4.10 Decision Matrix

1. In Hypothesis testing result is called "statistically significant" when :
a. The null hypothesis statement is true
b. The alternate hypothesis statement is true
c. P–value is less than or equal to the significant level
d. P–value is greater than the significant level
106
2. As per hypothesis testing theory significance test based on a small Introduction to
sample may not provide a significant output even if the correct value Hypothesis Testing
differs significantly from the null hypothesis statement value. This
type of incorrect result is known as :
a. Alpha value (the significance level of the test)
b. 1 – (the power of the hypothesis test)
c. a Type 1 error
d. a Type 2 error

Variance : Z–Test :
Z–Test is also known as the one–sample Z test because this hypothesis
is carried out with one sample only. In Z–Test, we claim population
parameters like mean or proportion when the population variance is
known. We studied in the last unit that according to the central limit
theorem (CLT) a large sample extracted from a normally distributed
population follow the same mean µ as the population mean and standard
σ
deviation where is the standard deviation of the population.
n
For Z–test, a test statistic is as below :
µ
Z − Statistics =
X−
σ
n
The critical value will always depend on the level of significance

.
Example 4.1 : A company claims that they complete the full and
final settlement of an employee within 30 days of quitting the company.
The researcher collected data for 40 employees. Assume population standard
deviation is 12.5 days. Conduct an appropriate hypothesis test at significance
level = .05 to verify the company's claim.
28 26 38 24 16 23 23 41 27 21
16 16 30 37 25 22 19 35 27 32
22 25 24 32 33 28 31 18 29 28
34 28 24 35 24 21 32 29 24 35
Solution :
Hypothesis statements are as below :
H0 : >= 30
H1 : < 30
107
Business Analytics From the above data points, we calculated the sample mean,
X = 27.05 days
σ 12.5
Standard deviation of sampling distribution
= = = 1.974
n 40
µ 30
Z–Statistics =
X− 27.05 −
= −1.4926
=
σ 12.5
n 40
The critical value of left–tailed test for = 0.05 is –1.644. Since

the critical value is less than the Z–Statistic value, we fail to reject the
null hypothesis. P–value for Z = – 1.4926 is 0.06777, which is greater
than the value of , so retain the null hypothesis.
Fig 4.11 Critical Value of One Tailed Test

Variance : t-Test :
We studied in the second unit that if we extracted a small sample
(less than 30) from a normally distributed population with an unknown
standard deviation, then sample distribution follows t–distribution with
n–1 degree of freedoms. In the case of t–distribution, we have to estimate
the variance using the sample itself. Suppose the standard deviation of
the sample is S.
µ
t − Statistics =
X−
S
n
Here n is the sample size; as we are estimating the value of standard

deviation, hence we lost one degree of freedom. The above t–statistic
formula follows t–distribution with n–1 degree of freedom.
Example 4.2 : An insurance company wants to understand the claim
data for one of their products X. They believed that on average there
are 500 claims every year. Assume claim data follows the normal
distribution. Conduct an appropriate hypothesis test at = 0.05 to check
company's belief about an annual claim is correct.
108
632 457 335 252 667 636 286 444 636 292 Introduction to
Hypothesis Testing
601 627 330 364 562 353 583 254 528 470
762 439 599 708 530 402 729 593 601 408
125 60 101 110 60 252 281 227 484 402
Solution :
Here n = 40, S = 195.0337 and X = 429.55

Hypothesis statements are as below :
H0 : <= 500
H1 : < 500
Test statics calculation will be as follows :
µ 429.55 − 500
t − Statistics =
X− = − 2.2845
=
S 195.0337
n 40
As this is a one–tailed test (right–tailed) and the critical t–value

at = 0.05 is 1.6848
[In Microsoft Excel, we can calculate the critical value, TINV(2 ,
degree of freedom) TINV(.1,39) = 1.6848]. Since the t–statistics value
is less than the critical t–value, we will retain the null hypothesis.
Fig 4.11 Critical Value for a Two Tailed Test

1. Which one is the correct Null and alternate hypothesis statement
for the below sentence :
The average age of an Indian to get married in 25 years
a. H0 : > 25; Ha : 25 b. H0 : = 25; Ha : > = 25
c. H0 : > = 34; Ha : 34 d. H0 : = 25; Ha : 25
2. Sentence 1 – Type I Error : We conclude that the average age
at the time of marriage is NOT 25 years when it is 25 years
Sentence 2 – Type II Error : We conclude that the average age
at the time of marriage is 25 years when in fact it is NOT 25 years
109
Business Analytics a. Only sentence 1 is correct b. Only sentence 2 is correct
c. Both sentences are correct d. None of the sentences is correct
1. A claim made about the population parameter for analysis purpose
is called
a. Test–Statistic b. Statistic
c. Hypothesis d. Level of Significance
2. A researcher is conducting hypothesis testing at = 0.10. P–value
is 0.06 then what is the most appropriate answer :
a. Reject the null hypothesis
b. Fail to reject the null hypothesis
c. Information is incomplete to make a decision
d. value is too high
3. The null hypothesis gets rejected beyond a point. What we call
this point
a. Significant Value b. Critical Value
c. Rejection Value d. Acceptance Value
4. If the critical region is split into two parts then which option is
correct
a. The Test is two–tailed b. The Test is one–tailed
c. The Test is zero–tailed d. The Test is the three–tailed Test
5. When will type I error occurred
a. We accept H0 if it is True b. We reject H0 if it is False
c. We accept H0 if it is False d. We reject H0 if it is True
6. Power of Test can be calculated as
a. b. c. 1 – d. 1 –
7. Confidence level can be calculated as
a. 100% b. 100%
c. (1 – ) 100% d. (1 – ) 100%
8. What can be another name of the alternate hypothesis
a. Composite hypothesis b. Simple Hypothesis
c. Null Hypothesis d. Research Hypothesis
9. Another name for type II error
a. Producer's risk b. P–value
c. Consumer's risk d. Confidence interval
110
10. = 30, this statement can be considered as Introduction to
a. Alternate hypothesis Hypothesis Testing
b. Null hypothesis
d. It depends on the scenario; it can be either a null or alternate
hypothesis
4.9 Let Us Sum Up :

1. Hypothesis means assumption and validating these assumptions with
the help of sample data is known as hypothesis testing
2. Most of the time we do analysis based on sample data, but we
want results to be validated for the entire population therefore
hypothesis testing techniques help us in this process
3. Hypothesis testing is an important part of various predictive analytics
techniques like linear regression, logistic regression etc
4. The central limit theorem is used to calculate the test statistics in
the case of t–test and z–test
5. P–value plays important role in concluding the hypothesis testing
process, it is the evidence in support of null hypothesis statement
6. We retain or reject the null hypothesis statement by comparing
p–value with significance value ( )
7. Rejecting a null hypothesis when in fact it is true is known as type
I error or producer risk
8. Retaining the null hypothesis when it is false is called type II error
or consumer's risk. It is also denoted by .
9. we never say "Null hypothesis accepted" as we cannot prove it in
all possible scenarios. We just find evidence where the alternate
hypothesis is not valid (it may be possible that the alternate hypothesis
statement is not valid in a few scenarios while it may be valid in
other scenarios). As soon as we find enough evidence against the
alternate hypothesis statement, we declare that we fail to reject the
null hypothesis.
10. The value of 1– is also known as the power of test which means
how sensitive is our hypothesis testing in rejecting the null hypothesis
when it is false

1. c 2. d
1. d 2. c
111
Business Analytics Check Your Progress – 3 :
1. c 2. a 3. b 4. a 5. b
6. d 7. c 8. d 9. c 10. b
4.11 Glossary :
Hypothesis Testing : It is a process of validating a claim (hypothesis)
based on analysis of sample data
Null Hypothesis Statement : It is a claim about research that is
assumed to be true unless it is proved incorrect with the help of collected
sample data
Alternate Hypothesis Statement : It is opposite to the null hypothesis
statement. It is the interested claim of a researcher which he wants to
prove correct with the help of collected data
P–Value : The p–value is the probability value that indicates how
likely it is that a result occurred by chance alone. The p–value is a
conditional probability of observing the sample statistic value when the
null hypothesis is true. P–value (probability) is evidence in support of
the null hypothesis statement.
Type–I Error : When we reject a null hypothesis when in fact
it is true
Type–II Error : When we fail to reject (accept) null hypothesis
when in fact it is false
Power of Test : The value of 1– is known as the power of test
which means how sensitive is our hypothesis testing in rejecting the null
hypothesis when it is false
4.12 Assignments :
1. Define Hypothesis testing, risk, null and alternate hypothesis
testing statements and p–value with example.
2. What is the difference between one–tail and two-tail hypothesis
tests, explain with an example.
3. Why type–I error in hypothesis testing is known as producer's error.
Explain with an example.
4. Write down null and alternate hypothesis testing statements for the
below scenarios :
a. A company claims that they complete the full and final
settlement of an employee within 30 days of quitting the
company.
b. The passport office claimed that they issue passports within
30 days after submission of all required documents.
112
4.13 Activities : Introduction to
Hypothesis Testing
Peacock training institute invented new technological ways to teach
programming to their students. They picked up 20 students and checked
their monthly study hours before and after training as per new technology.
Below is the table that represents study hours before and after. Conduct
a t–test to see whether new technology is motivating students to study
more hours. Assume = 0.05
Before New Technology After New Technology
349 335
449 344
378 318
359 492
469 531
329 417
389 358
497 391
493 398
268 394
445 508
287 399
338 345
271 341
412 326
335 467
470 408
354 439
496 321
351 437
Ans. : The value of test statistics is 0.5375 and the critical value of
t–test when = 0.05 and degree of freedom 19 is 1.7291. since
the t–statistics value is less than the critical value hence we will
retain the null hypothesis and the difference is studying hours
is not significant.
113
Business Analytics 4.14 Case Study :
ABC limited company has appointed a new analytics and insight
department head to streng–then its presence in India and the Asian market.
He wants to analyze whether the marketing cost for units A and B for
the financial year 2018 and financial year 2019 is significantly different.
He asked the MIS team to provide him with sales data and marketing
costs in the last 10 years :
Year Marketing Cost for Unit A Marketing Cost for Unit B
B2009 41 31
2010 56 36
2011 28 37
2012 42 35
2013 31 38
2014 38 41
2015 39 44
2016 39 45
2017 51 47
2018 43 49
Answer the below Questions :
a. Draw scatter plot of above data
b. Conduct hypothesis testing at a 95% confidence level and conclude
your result
c. Will, there be any change in your recommendations if you will run
the hypothesis testing at a 90% confidence level

Carmer H (1946)
• "Super Freakonomics," Penguin Press, Levitt S D and Dubner
S J (2009)
F W (2001)
114
BLOCK SUMMARY
In the business world, we have various business metrics which tells
about the health of our organization. We should see these metrics over some
time so that we can see the trends. Business wants to predict the values with
the help of probability theory. One such tool is a probability distribution.
Based on the nature of business metrics we classify them among continuous
or discrete metrics. We saw how we apply the goodness of fit test to check
whether data follows a specific distribution or not. This is done with the
help of hypothesis testing. On the other hand, most of the time we do not
want to predict the point estimation (e.g. future share price of our
organization) instead of that we want to estimate the interval range in which
this value can fall. The study of confidence interval helps us to calculate
it more precisely and also advise businesses on how they can be ready for
extreme situations. Businesses can plan continuity measures in case things
will be extremely good or not so good.
In the business world, time always remains a critical resource. Most
of the time we want to conclude our research fast so that results can be
beneficial for businesses and their end customers. Therefore, we conduct
the sample study at the end of the research whether it is meeting with the
business objective and there is no adverse effect. This can be done with the
help of hypothesis techniques. It helps us to conclude whether the sample
result will hold in the future also. Hypothesis testing is the soul of
inferential statistics, it plays important role in concluding research outputs.
115
1. Write short notes on PDF, PMF and CDF
2. Explain the difference between the binomial and Poisson probability
distribution
3. Explain the difference between the normal and student–t probability
distribution
4. Write important assumption of Poisson distribution
5. Why do we say that we need to be very cautious about applying
Poisson distribution to solve a real–life business problem ?
6. Write a short note on "How to check Z score in Standard Normal
Distribution Table (Z Table)"
7. Why confidence interval is better than point estimation, explain with
an example
8. Write a short note on the margin of error in the calculation of the
confidence interval
9. Is there any relationship between sample size and confidence level,
explain with an example ?
10. Why do we need to define null and alternate hypothesis statements ?
11. Write short notes on critical value and rejection region in the
hypothesis testing process
12. Write differences between type I and type II error, draw a decision
matrix to explain the concept
116
1. Write down the basic difference between discrete and continuous
probability distributions. Write examples with their formulae
2. Explain important properties of normal distribution, draw diagram
wherever possible to highlight the important characteristic of each
property
3. Write important properties of t–distribution
4. Explain important steps in designing a sampling strategy
5. Explain two important types of sampling techniques, write down
important sampling techniques under each category
6. Write down Hypothesis Testing Process Steps, explain these steps
briefly
117
Unit No. 1 2 3 4
No. of Hrs.
of the block :

.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
118
Ambedkar
OpenUniversity
Business Analytics
BLOCK-3 CORRELATION AND REGRESSION
UNIT 1
COVARIANCE AND CORRELATION ANALYSIS
UNIT 2
SIMPLE LINEAR REGRESSION
UNIT 3
MULTIPLE LINEAR REGRESSION
BLOCK 3 : CORRELATION AND REGRESSION
Block Introduction
Finding the relationship among various business metrics and helping

businesses in predicting these metrics for the next period is utterly essential
for robust business decisions. Correlation and regression provide this
strength to analysts. Regression is the soul of business analytics as it
remains the foundation stone for various vital tools and techniques in the
analytics world. In the business world, things do not happen in isolation;
there is always a story behind all critical incidents; things depend on various
factors. Correlation helps in establishing a relationship among business
incidents, while regression helps in providing mathematical relationships
among these incidents. If we know the value of one metric, then we can
predict the value of others. These techniques allow us to find crucial
business questions like :
• If the marketing budget is X, then what can be sales in the near future
• Which service model will attract the customers most
• What will be the sales in next month/quarter/year
• Which customer is likely to churn
• Who will be the probable customer for email marketing
• Which age group will be the right customer segment for the upcoming
product
Various predictive/ prescriptive analytical models are required to

answers the above questions; these models directly or indirectly depend on
correlation and regression techniques. In today's era of digitization, businesses
collect various data points due to regulatory and industry requirements.
Correlation and regression select the most critical variables out of this
plethora of data.
Block Objectives
• Understand the concept of covariance and correlation

• Applications of covariance and correlation in the business world
• Mathematical interpretation of covariance and correlation
• Difference between correlation and causation
• Visual interpretation of correlation analysis
• Spearman rank correlation
• Understand the concept of simple linear regression and its
mathematical interpretation
• Various stages in the regression model building
• Essential assumptions of linear regression
• Learn Ordinary–Least–Square method for estimating regression
parameters
• Validation techniques for a regression model
• Application of simple linear regression in machine learning and
Block Structure
Unit 1 : Covariance and Correlation Analysis
Unit 2 : Simple Linear Regression
Unit 3 : Multiple Linear Regression

Unit COVARIANCE AND
1 CORRELATION ANALYSIS
: UNIT STRUCTURE :
1.1 Introduction
1.2 Covariance : Statistical Relationship between Variables
1.2.1 Mathematical Interpretation of the Covariance
1.2.2 Relationship between Covariance and Variance
1.3 Covariance Matrix
1.4 Relationship between Covariance and Correlation
1.5 Spearman Rank Correlation
1.6 Let Us Sum Up
1.8 Glossary
1.9 Assignment
1.10 Activities
1.11 Case Study

• Understand the concept of covariance and correlation
• Mathematical interpretation of covariance and correlation
• Applications of covariance and correlation in the business world
• Difference between correlation and causation
• Visual interpretation of correlation analysis
• Spearman rank correlation
1.1 Introduction :
In this unit, we study the statistical relationship between continuous
variables in terms of covariance and correlation. We will discuss the
different scenarios in which one of these techniques will be more
appropriate than another. We will see the limitation of these measures
and their visualization techniques. In the end, we will also touch upon
the correlation of a ranked data.
119
Business Analytics 1.2 Covariance : Statistical Relationship between Variables :
Covariance is one of the techniques we use in the business world
to measure the linear relationship between two variables. Other measures
from the same family are correlation and linear regression, which we will
study in this block later. Covariance means – "Co–Vary," variables that
behave in pairs.
This behaviour can be positive (covariance value is a positive

number) means if the value of one variable increases, other variable's
value will also increase, e.g., temperature and sales of ice cream, we
know in summer sales of ice cream goes up, or temperature and electricity
bill in North and West India.
Covariance can be negative (covariance value is a negative number)

means if the value of one variable increases other decreases, e.g.,
temperature and sales of woollen clothes or electricity bills in North and
West India.
Or there can be no covariance (covariance value is very close to
zero), e.g., movement of the stock market when the temperature goes
up. E.g. Relationship between temperature and NIFTY or BSE index. In
this case, if one variable increase then second remain constant or
significantly less change.
120
Covariance and
Correlation Analysis
1.2.1 Mathematical Interpretation of the Covariance :

Let's try to understand the mathematical formula for covariance with
the help of an example of total study hours in a week and study hours
for a Business analytics subject.
As we have studied in the last section, the objective of covariance

is to find out the relationship between these study activities. There is
one fundamental thing to observe that one column has comparatively
high numbers than the second column. So it is not very intuitive to
compare these in this form. One way to tackle this problem is to find
the deviation of these two columns from their mean.
Now we can compare numbers in columns D and E. It is quite

a trial to understand that if both numbers in these columns have the
same sign, then these variables are positively correlated. While if there
would have been different signs in lots of rows, then that means both
variables are negatively correlated. The same thing also revealed from
column F that if the number (product of column D and column E) is
positive, then these variables are positively correlated, while if there are
negative numbers, then negatively correlated.
The same can be derived from the mathematical formula for
covariance:
Σ ( (x − x) (yi − y) )
COV(x, y) =σ xy = i
n −1
121
Business Analytics Please Note : In the above formula, we have n – 1 in the denominator
as we lost one degree of freedom for estimating the mean to the sample.
35
COV(x, y) =σ xy = = 8.75
4
So our both variables, "total study hours" and "study hours for
business analytics," are positively correlated. Here the vital point to note
is that a covariance score of 8.75 does not tell us the strength of a
relationship. It depends on the variable range. If we had both variables
in lacs, then the value of the covariance coefficient would also be in
lacks. So we cannot compare the covariance coefficient of two different
studies. It just tells us whether there is positive covariance, negative
covariance, or no covariance at all.
Suppose there would have been lots of negative values in column
F of the above spreadsheet. In that case, the covariance value might also
be negative, which indicates a negative relationship between the variables.
1.2.2 Relationship between Covariance and Variance :

In the last block, we studied variance. Below is the formula:
Σ(x − x)2
VAR(x) =σ 2x = i
n −1
We can notice that formula for covariance and variance is quite
similar; we can also rewrite the variance formula as below:
Σ ( (x i − x) (x i − x) )
VAR(x) =σ2x =
n −1
if we replace xj with yi and x in second them with y then the formula of
variance will turn into the formula of covariance. In other words, covariance
is nothing but a variance formula with two variables x and y.
Σ ( (x − x) (yi − y) )
COV(x, y) =σxy = i
n −1
122
Check Your Progress – 1 : Covariance and
1. Which statement about covariance is NOT correct Correlation Analysis
a. If variables x and y increase simultaneously, then covariance

between x and y is positive
b. Zero covariance means we can not calculate the covariance of
x and y
c. Zero covariance means both variables are not related
d. If variable x increase when y decrease, then covariance between
x and y is Negative
2. The covariance between rain in centimetres and occupancy percentage
of movie theatres is –0.76. this indecases ___________ relationship
between these two variables.
Example – 1.1 : ABC company wishes to study the relationship
between the number of employees (x) and the number of cardboard
manufactured (y). Below is the sample data, each one hour in length
from the production floor.
Solution : It is always advisable to make a scatter plot to see the

relationship visually.
We can calculate measures to put in covariance formula.
123
Business Analytics
962.4
COV(x, y) =σ xy = =106.93
9
Here it is a positive covariance between the number of employees
and cardboard manufactured, so if the company wants to increase
productivity, then they should increase the workforce.
1.3 Covariance Matrix :

In case we have various variables, then we can create a covariance
matrix; we calculate covariance for all possible pairs of variables. Below
is the covariance matrix for four variables x1, x2, x3 and x4
Below is the covariance matrix, where we have covariance for all

possible pairs of variables. Diagonally, there is the covariance of each
variable with itself, which is a variance of each variable.
124
The covariance matrix shows us a comprehensive picture of all Covariance and
variables included in the study. The lower diagonal and upper diagonal Correlation Analysis
of the covariance matrix is always identical as the covariance of variable
x and y is the same as the covariance of variable y and x.
1. Which statement about the covariance matrix is NOT correct
a. Covariance is advisable if we have more than two variables in
the study
b. Off diagonal elements provides covariance between each pair
of variables
c. In the covariance matrix, lower triangular and upper triangular
elements are the same
d. Values always vary between 1 and 5
2. Diagonal elements of covariance matix shows ___________ of each
variable
1.4 Relationship between Covariance and Correlation :

Correlation values are independent of the scale of data, which
means whether we have two variable's values in millions, still, their
correlation will always vary between –1 and +1 because correlation is
the standardized score by the standard deviation of both the variables.
The symbol r represents correlation; the below formula represents it:
σxy
r=
σx σ y
Therefore correlation is covariance divided by the standard deviation

of both variables x and y.
So if we put values covariance and standard deviation of x and
y from the above formula:
8.75
=r = 0.86
5.70 × 1.79
Correlation also tells us the strength of the relationship as it always

varies between –1 and 1. Where –1 indicates a strong negative correlation,
while +1 indicates a strong positive correlation between variables. It is
always advisable to see the scatter plot of variables before calculating
the value of correlation coefficients. As correlation is only applicable
for a linear relationship if the scatter plot is showing curvilinear or any
other non–linear trend, then we should not use the above formula to
calculate the correlation. It may be possible that the value of the correlation
coefficient is close to zero for a non–linear relationship, while they may
be a strong relationship between both variables.
125
Business Analytics Please Note : Covariance tells only the direction (no covariance,
positive covariance, and negative covariance) of the linear relationship
between two variables. In contrast, correlation tells us both direction and
strength. The sign of the correlation coefficient is the same as the sign
of covariance.
The above correlation formula can be rewritten as below:
n
∑ (x i − x) (yi − y)
i =1
n n
∑ (x i − x) 2 ∑ (yi − y) 2
=i 1=i 1
n ∑ x i yi − ∑ x i ∑ yi
⇒
2 2
n ∑ x i2 − ( ∑ x i ) n ∑ yi2 − ( ∑ yi )
Statistician Karl Pearson discovered the correlation coefficient; hence

it is also known as Pearson's coefficient of correlation. In Microsoft Excel,
the formula for the correlation coefficient is CORREL(array1, array2).
If the correlation between two variables is .7, then it can say that
these two variables are explaining a total of 49% variation in the study.
In other words, 49% variation is explained by these two variables, while
the remaining 51% variation is due to different variables that are not
included in the study.
Example – 1.2 : In a manufacturing firm, the maintenance team
collected data for the number of batches runs on each machine daily,
and the number of the month the machine works without breakdown.
No. of batches
per day (x) 25 35 10 40 85 75 60 45 50
Machine life in
months (y) 63 68 72 62 65 46 51 60 55
Check whether there is a correlation between the number of batches
and machines life.
Solution : We can create a table to calculate all components of
the coefficient of the correlation formula:
126
Covariance and
Correlation Analysis
Hence there is 61% that the number of batches influences the life
of the machine before breakdown.
rxy NΣ x i y i − Σx i y i
=
NΣx i2 − (Σx i ) 2 NΣyi2 − (Σyi )2
9.24640 − 425.542
=
9.24525 − 180625 9.33188 − 293764
− 8590
= = − 0.61
40100 4928
1.5 Spearman Rank Correlation :

Pearson correlation is more appropriate when variables show a
linear relationship and both from either interval or ratio scale. When
variables show the non–linear relationship or there are outliers in the
data or variables are of an ordinal scale.
6Σd i2
rs= 1 −
n(n 2 − 1)
Here, di is the difference between the rank of xi and yi
127
Business Analytics Example – 1.3 : An advertisement company collected data for
'Number of hours TV watch' and 'IQ level.' Below is the data–calculated
Spearman rank correlation to check whether there is a relationship between
TV watching hours and IQ level.
IQ 106 100 86 101 99 103 97 113 112 110
TV Watching
hours / Week 7 27 2 50 28 29 20 12 6 17
Solution : Let's make a scatter plot to see the relationship
The first thing we have to calculate a rank for both x and y variables.
Here di2 = 194 and n = 10
6Σdi2
ρ= 1−
n(n 2 − 1)
to give
6 × 194
ρ= 1−
10(102 − 1)
128
− 29 Covariance and
Finally, ρ = = − 0.17576 Correlation Analysis
165
The value is very close to zero, which indicates no correlation
between "TV–watching hours" and "IQ level."
1. The covariance of two numeric variables varies:
a. –1 to +1
b. –50 to + 50
c. It depends on the median of data
d. It depends on the distribution of data
2. Which statement is NOT correct regarding covariance of data
a. Covariance depends on the sample size
b. It is applicable for numeric data only
c. It always varies between –1 and +1
d. All of the above
3. Which of the following is NOT correct about the correlation
a. Correlation value always depends on the scale of variables
b. The correlation value is independent of the scale of the variables
c. It always varies between –1 and +1
d. All of the above
4. If the correlation between two variables is .8, then how much
variance explained by these two variables
a. 16%
b. 40%
c. 64%
d. Can't calculate variance if the correlation is provided between
two variables
5. Rajesh found a covariance level of 21.98 between the two variables.
What we can say about this:
a. It means there is a strong negative between two variables
b. It means there is a strong positive correlation between the two
variables
c. It means there is no correlation between two variables
d. It means there is a weak positive correlation between two
variables
6. Mahesh created a scatter plot diagram for two variables for which
we want to assess the relationship; he could see a few outliers in
his graph. What would you like to suggest as the most appropriate
technique to analyze the data:
129
Business Analytics a. He should calculate the covariance
b. He should calculate spearman rank correlation
c. He should calculate the variance of both variables individually
and then compare
d. He should calculate the correlation coefficient for both the variables
7. Which one is the most appropriate option regarding the covariance
matrix
a. We should fill all the cells of the covariance matrix, including
the upper triangular and lower triangular matrix
b. We should create a covariance matrix only in case when we have
more than two variables in our study
c. The covariance matrix is vital for the non–linear relationship
d. We should always calculate the variance of each variable
individually before creating the covariance matrix
8. If the correlation coefficient between study hours and the final score
is .9, how much variance in the final score is not accounted for
by study hours
a. 19% b. 18% c. 45% d. 9%
9. Which statement about covariance is NOT correct:
a. Covariance can be defined as an unstandardized version of the
correlation coefficient
b. It is a measure of the linear relationship between two variables
c. It is a synonym of the correlation coefficient
d. It is dependent on units of measurement of the variables
10. Statement 1 : Covariance tells about the direction of the linear
relationship between two variables
Statement 2 : Correlation tells about the direction and strength of
the linear relationship between two variables
a. Only statement 1 is true b. Only statement 2 is true
c. Both statements are true d. None of the statements is true
1.6 Let Us Sum Up :

1. It is always recommended to see the relationship between the
variables with the help of a scatter plot before applying a mathematical
formula.
2. Covariance explains the linear relationship between two continuous
variables
3. Covariance tells us only about direction (whether a relationship is
positive, no relation, or negative), but it does not tell anything about
the strength of the relationship
130
4. Covraince score depends on the unit of variables; therefore, we Covariance and
can not compare the covariance score of two different studies Correlation Analysis
5. Correlation is a standardized relationship score (covariance
standardized by the standard deviation of both variables); therefore,
its value is always between –1 to +1. We can compare the correlation
score of two different studies
6. Correlation explains the only association of relationship but not
guarantees the causal relationship between the variables
7. Pearson correlation is appropriate to apply if both variables are
continuous (either interval or ratio scale). In the case, variables are
in ordinal scale or if there are outlier or relationship is non–linear
then Spearman rank correlation would be a better choice
1.7 Answers for Check Your Progress :

1. b 2. Negative
1. d 2. Variance
1. d 2. c 3. a 4. c 5. d
6. b 7. b 8. a 9. c 10. c
1.8 Glossary :
Covariance : It is a linear relationship between continuous variables;
it explains only direction, not the strength of the relationship. Covariance
scores always depend on unit of variables; hence we can not compare
covariance scores of two independent studies
Pearson Correlation Coefficient : It is a standardized linear
relationship between two variables; therefore, values always come in
between –1 and +1. Correlation does not guarantee causation
Covariance Matrix : In case there are more than two variables
in the study, then it is always advisable to see the linear relationship
between all possible pairs of variables. This combination of the relationship
among various variables is known as a covariance matrix.
Spearman Rank Correlation : If variables are on an ordinal scale,
or there are some outliers, or the relationship is not non–linear, the
Spearman rank correlation is more appropriate than Pearson's correlation.
Here we calculate the correlation between the ranks of the variable instead
of raw form.
131
Business Analytics 1.9 Assignment :
1. Why do we call covariance means co–vary, explain it with an
example ?
2. What is the significance of covariance metrics visualization ?
3. If two variables are measured in different scales why correlation
is better to measure for the relationship than covariance.
1.10 Activities :
1. In the below table, there are dividends provided by two banks for
the last five years. Calculate covariance and correlation for the
dividends of these two banks. Describe the relationship between
the shares.
Year 1 Year 2 Year 3 Year 4 Year 5
Bank A 5% 7% 2% –5% 3%
Bank B –1% 0% 5% –1% 3%
2. Use the following tabular data to answer the below questions
Variable A Variable B Variable C Variable D
10 110 92 930
11 120 94 900
12 115 97 1020
13 128 98 990
11 137 100 1100
10 145 102 1050
9 150 104 1150
10 130 105 1120
11 120 105 1130
14 115 107 1200
a. Make covariance matrix of the above data
b. Calculate the correlation between variable A and variable B
Ans. : 1. Covariance = 1.075 Correlation = 0.0904
2. (b) –0.51
1.11 Case Study :

department head to strengthen its presence in India and the Asian market.
He wants to analyze sales and advertisement costs to examine whether
different advertising channels are providing enough sales or not. He asked
the MIS team to provide him with sales data and marketing costs in the
last 10 years :
132
Year Sales Marketing Cost Covariance and
(in a million rupees) (in a million rupees) Correlation Analysis
2009 2064 31
2010 2389 36
2011 2418 37
2012 2509 35
2013 2608 38
2014 2706 41
2015 2802 44
2016 2905 45
2017 3056 47
2018 3189 49
a. Draw scatter plot of above data
b. Calculate the variance of Sales and Marketing Cost data
c. Calculate covariance and correlation of sales and marketing cost
d. Suggest whether Pearson correlation or Pearson rank correlation
would be better in this case, justify your answer

Carmer H (1946)
• "Super Freakonomics," Penguin Presss, Levitt S D and Dubner
S J (2009)
F W (2001)
133
Business Analytics
Unit SIMPLE LINEAR REGRESSION
2
: UNIT STRUCTURE :
2.1 Introduction
2.2 Essence of Simple Linear Regression
2.2.1 Introduction to Simple Linear Regression
2.2.2 Determining the Equation of the Linear Regression Line
2.3 Baseline Prediction Model
2.4 Simple Linear Regression Model Building
2.5 Ordinary Least Square Method to Estimate Parameters
2.5.1 Calculation of Regression Parameters
2.5.2 Interpretation of Regression Equation
2.6 Measures of Variation
2.6.1 Comparison of Two Models
2.6.2 Coefficient of Determination
2.6.3 Mean Square Error and Root Mean Square Error (Standard
Error)
2.7 Simple Linear Regression in MS Excel
2.7.1 Residual Analysis to Test The Regression Assumptions
2.8 Let Us Sum Up
2.10 Glossary
2.11 Assignment
2.12 Activities
2.13 Case Study

• Understand the concept of simple linear regression and its
• Various stages in the regression model building
• Essential assumptions of linear regression
• Learn Ordinary–Least–Square method for estimating regression
parameters
134 • Validation techniques for the regression model
• Application of simple linear regression in machine learning and Simple Linear Regression
2.1 Introduction
Unit : In this unit, we will study Simple Linear regression that
explains the relationship between two continuous variables and how it
is more effective than correlation and covariance. There are various ways
to calculate linear relationships, we will touch upon the most fundamental
technique know as least square optimization. We will also touch upon
various validating techniques for linear regression. In the end, we will
see various examples where linear regression helps us to make better
decisions and empowered us with predicting capabilities.
2.2 Essence of Simple Linear Regression :

In the last unit, we discussed covariance and correlation as a
measure of the linear relationship between variables. On similar lines,
simple linear regression is a statistical technique for finding a relationship
between variables, but unlike correlation, it also provides us with a
mathematical equation between these two variables. If the value of one
variable is known, then we can estimate the value of other variables.
Organizations use various metrics to measure their performance, e.g.,
sales revenue, employee tenure, feedback score, cost of goods sold,
growth rate, market share, return on investment, etc. Performance of these
matrics depends on various influential factors such as:
• Price of product
• Competitor's price
• Market share by competition
• Marketing budget
• Marketing channels
• New product introduction
• Promotional strategy
• Macroeconomic variables such as GDP, unemployment index,
inflation rate, inflation, etc.
In order to make robust business decisions, organizations want to
understand the relationship of their essential business metrics with these
influencing factors.
2.2.1 Introduction to Simple Linear Regression :
Regression analysis is the statistical technique of developing a
statistical model, which can be used to predict the value of one variable
(output variable) with the help of another variable (input variable). The
output variable is also known as the dependent variable as the value
of the output variable depend on the value of the input variable, generally,
we denote output by alphabet Y. In the business world, there is always
a cause behind every business outcome. The variable on which our output
135
Business Analytics variable (metric) depends is known as the input variable. Regression
analysis helps us to calculate the impact of this input variable on the
output variable. To see the clear impact of input variable on output
variable, regression has an important assumption that input variables
should not have any relationship among themselves otherwise we will
not be able to calculate the influence of each input variable on output
variable. As input variables do not show any relationship, therefore, these
are also known as independent variables. In regression analysis, the
dependent variable is also known as regressed or explained or response
variable. At the same time, the independent variable is also be defined
as a regressor or predictor or explanatory or input variable. In simple
linear regression analysis, a mathematical relationship between two variables
is explained by a straight line, for example, predicting sales revenue based
on marketing expenditure or the number of years of experience of employees
and salary package or temperature and water consumption, etc. The
relationship between one dependent variable and various independent
variables is known as multiple linear regression; we will study multiple
regression in detail in the next unit.
2.2.2 Determining the Equation of the Linear Regression Line :
Simple linear regression is established on the slope–intercept equation
of a line. We have studied the equation of line as:
y = b + mx
Here, m is the slope (angle with x–axis) of the line, and b is the
y–intercept of the line (the point where the best–fit line crosses the
y–axis).
But this is a mathematical expression that can be defined for a finite

dataset. In case we want to write it for a sample, then there is a score
for random error. In that case, instead of a mathematical expression (also
known as a deterministic model), we write a statistical expression (also
known as the probabilistic model). This can be expressed as below:
y = b + mx + random error
Here we can think as y is a function of x: y = f(x);
If the value of x (independent/input variable) is known, then we
can estimate the value of y (dependent/ output variable).
136
Simple linear regression can be expressed in a functional form as Simple Linear Regression
follows:
Y = 0 + 1X +
For a dataset with n observations, we can write the above functional
form for each of these observations as follows:
Yi = 0 + 1Xi + i
Where
Yi is the ith observation (data) of the dependent (output) variable
in the dataset
Xi is the ith observation of the independent variable in the dataset
i is the random error
and 1 are the regression parameters (or regression coefficient)
0
The expected value of simple linear regression is as follows:
E(y) = 0 + 1x
So the value of 1 determines the slope of the relationship between

variables.

1. In simple linear regression there is ___________ output variable
and ___________ input variable.
2. Regression is important technique in ___________ analytics.
3. Linear regression implies that the relationship between output variable
and input variable is ___________.
137
Business Analytics 2.3 Baseline Prediction Model :
A restaurant chain collected feedback scores for its customers. They
have received a "Very satisfied" rating for six days but suppose they have
not counted the total number of feedbacks received. Now data for "Very
Satisfied" customers are as below:
Day Count Day 1 Day 2 Day 3 Day 4 Day 5 Day 6
Count of
5 17 11 8 14 5
Very Satisfied Feedback
Here we do not have the input variable (total number of feedbacks).
Now can we predict the count of "Very Satisfied" feedback on the 7th
day?
Let's visualize our data with the help of a simple line graph.
So one way to predict the "Very Satisfied" feedback score is

considering the average of these six days feedback count. The average
of six days of "Very Satisfied" feedback is 10. So for one variable best
prediction can be average.
The next step is to validate how good is this prediction as we can
see, all feedback counts are either more than 10 or less than 10. Let's
see the difference of each day count from this predicted count of 10
feedbacks. This difference is an error; in statistics, it is known as residuals.
So if we sum up all the errors in our prediction model, then it

will be zero.
138
So to get rid out of these negative differences, we can square all Simple Linear Regression
these residuals. It helps in two ways:
1. It makes all differences positive
2. It emphasizes larger deviations
If we sum all these squared residuals, then this term is known as
the sum of squared errors (SSE), which is 120 in this case.
So we have established a basic prediction model and calculate the
error in that model. Now the goal of simple linear regression is to create
a model that can minimize this sum of squared error (SSE). The regression
model will try to minimize this error by introducing an independent
variable in the model and draw a best fit line, so that difference between
the dependent variable value and line best is minimum; this error explained
by the regression model is known as Regression error.
1. If we have only dependent variable then we can predict the next
value by taking ___________ of all the observeations.
2. Difference between predicting line and each observation is know
as ___________.
3. If we add all residuals, then its sum will be very close to _________.
2.4 Simple Linear Regression Model Building :

To understand the statistical relationship between output and input
variables, we develop a simple linear regression model. A very high–
level framework for linear regression model building, it generally works
for the type of industries irrespective of their size.
Regression models can lead to previously unknown relationships,

therefore leading to a new hypothesis. Therefore it is a very systematic
approach where we see the relationship between important business
metrics and important factors that influence those metrics. By control
those factors, we can improve the performance of those business metrics.
2.5 Ordinary Least Square Method to Estimate Parameters :

We understood in the earlier section that the overall objective of
the regression model is the minimize the residuals by estimating the best
fit–line.
139
Business Analytics Where
Yi = observed value of dependent variable
Yi = predicted (estimated) value of the dependent variable
Difference between Yi and Yi is residual. There can be infinite

lines that can go through these data points; our task is to find the best
fit which yields a minimum sum of residuals squares.
2.5.1 Calculation of Regression Parameters :

It is relatively easy to calculate regression parameters if we can
put them all in the form of a table.
615
b1
= = 0.1462 → slope of line
4200
b0 =y − b1 × x =10 − 0.1462 × 74 =− 0.8188 → y − intercept

140
so the regression equation is as below: Simple Linear Regression
yi = 0.1462x – 0.8188
Please Note : Best–fit line always cross from the centroid
(intersection point of the mean of x and mean of y)
2.5.2 Interpretation of Regression Equation :
yi = 0.1462x – 0.8188
The above regression equation tells us that for every new addition
in feedback, there will be a .1462 addition to the "Very satisfied" category.
In other words, for any new feedback, there are 15% chances of having
it in the "Very Satisfied" category. If there is zero feedback, then there
are –.8188 "Very satisfied" feedback, which does not make any sense;
therefore, intercept may or may not make sense in terms of its business
value. So the important point is how the dependent variable changes in
relation to one unit change in the independent variable.
In this way, the regression model tells us the relationship between
important business metrics and factors that influence that metric. An
important point to remember is that the regression model can only predict
the output variable only in the range of input variables supplied. For
example, in the above example, input variable rage from 34 to 108; hence
technically, we can predict the number of "Very Satisfied" feedback only
if total feedbacks is in the range of 34 to 108. If the total number of
feedbacks is significantly high or lower than this range, then the regression
model may give us misleading predictions.
2.6 Measures of Variation :

Still, one question which is unanswered yet that how good is this
regression model? We saw in the earlier section that in the case of
baseline model total error, SSE, which was 120. In that case, it was also
a total sum of error (SST). Similarly, we have to measure in case of
the regression model where instead of the average line, we have the best–
fit line. Again we will calculate the distance of each observation from
this best–fit line, and we will square those distances to get rid out of
negative numbers and also to emphasize large deviations.
Total variation can be segregated into two parts: variation due to
relationship between x and y variable, it is referred to as regression sum
of squares (SSR), other is due to unexplained variation due to factors
which are not part of the regression model and is referred to as error
of sum of squares (SSE).
141
Business Analytics Total sum of squares (SST) = Regression sum of squares (SSR)
+ Error sum of squares (SSE)
• Total sum of squares (SST) is the difference between the mean of

y and each observed value (yi)
Σ (y i − y)2
Total Sum of Squares = SST =
• The regression sum of squares (SSR) is the difference between the

mean of y and each predicted value ( yi )
Σ(y i − y)2
Regression Sum of Squares = SSR =
• Error sum of squares (SSE) is the difference between each observed

value of y and each predicted value ( yi )
Σ(y i − y i )2
Error Sum of Squares = SSE =
We can calculate yi value by substituting the value of each

observation in the above regression equation as follows:
We can calculate SSE by substituting their corresponding values

in the above equations
142
Simple Linear Regression
So when we conducted regression, the SSE decreased from 120

to 30.075. That is, 30.075 of the sum of squares was unexplained by
the regression model or allocated to error.
The 89.925 is the sum of squares error due to regression.
2.6.1 Comparison of Two Models :
Below is the performance of two models, the first one without
regression (with the help of dependent variable only) and the second one
by applying regression calculations.
2.6.2 Coefficient of Determination :

Coefficient of determination measures that how well the regression
equation fits our data
We saw in the earlier section that SST further splits into SSE and
SSR. The coefficient quantifies this ratio as a percentage.
SSR
Coefficient of determination = r2 =
SST
Coefficient of determination for the above feedback example:
89.925
Coefficient of determination = r2 = = 0.7493 or 74.93%
120
We can conclude that 74.93% of the total sum of squares can be
explained by using the estimated regression equation to predict the "Very
Satisfied" feedback. The remainder is an error. For simple linear regression,
r2 is the square of correlation coefficient r. In the above survey feedback
143
Business Analytics example, the correlation coefficient is .866; if we square it, then it will
be equal to the coefficient of determination r2, which we calculated
through the sum of squares method.
2.6.3 Mean Square Error and Root Mean Square Error (Standard
Error) :
MSE, s2 is an estimate of 2 the variance of the error, . In other
words, how spread out the data points are from the regression line. MSE
is SSE divided by its degree of freedom; in the above example, it will
be n–2 as we lost two degrees of freedom as we estimated slope and
intercept. In simple linear regression, it is always n–2, while in the case
of multiple linear regression, it will be different; we will study that in
the next unit.
2 SSE
MSE
= s=
n−2
This is the reason MSE is not a simple average of residuals. But

in case we are using a population instead of a sample then we will use
n in the denominator, which is a simple average of residuals.
The standard error of the estimates (also known as the standard
error) is the standard deviation of the error term, .
If we calculate standard error for the above feedback example:
S= 7.5187 = 2.742
So the average distance of the data points from the fitted line is
about 2.74 feedbacks. We can think of s as a measure of how well the
regression model makes the predictions.
144
Check Your Progress – 3 : Simple Linear Regression
1. A total sum of squares is the summation of sum of squares
regression and ___________.
2. Cofficient of determination is the ratio of _________ and _________.
3. If SSE is 100 and there are 27 observations in the sample then
standard error would be _________.
2.7 Simple Linear Regression in MS Excel :

Microsoft Excel provides an option to conduct simple linear
regression,
• In MS Excel, under the "Data" tab "Data Analysis" package

(if it is not visible then go to File Options Activate data
analysis package
• Select the input and output data range
• Select all graphs options available towards the lower side of the
pop–up window
Detailed videos "How to perform a regression analysis in MS
Excel" are available on Microsoft's website.
Below is MS Excel output for regression analysis.
Below is the explanation for the essential components of the

regression output:
145
Business Analytics 1 The correlation coefficient for both the variables
SSR
2 The coefficient of determination (R2) .749 74.9%
SST
variation in output variable due to input variable
 SSE 
3 The standard error   2.742
n − 2
4 Regression sum of squares (SSR) 89.925

5 Error sum of squares (SSE) 30.749
6 Total sum of squares (SST) 120
7 The overall significance of the model, if the value is less than .05
at 95% confidence level, then the overall model showing a significant
relationship between input and output variables
8 y–intercept –0.820
9 Slope 0.146
10 The P–value for each variable, if it is less than .05 at 95% CL,
then the input variable is significant in the model
2.7.1 Residual Analysis to Test The Regression Assumptions :
Residual analysis is mainly used to test the assumptions of the
regression model. Below two are important residual analyses provided
by MS Excel:
1. The Linearity of the Regression Model : The linearity of the
regression model can be obtained by plotting the residual on the
vertical axis against the corresponding xi values of the independent
variable on the horizontal axis. There should not be any apparent
pattern in the plot for a fit regression model.
2. Normality of Error : The normality probability plot of the residuals

should roughly follow a straight line for meeting the assumption
of normality. A straight line correcting all the residuals indicates
that the residuals are normally distributed. It means that residuals
are random in nature, and there is not any hidden trend. A curve
in the tail is an indication of skewness.
146
Simple Linear Regression
There are some more advanced assumption validations, which are

not part of MS Excel like Constant error variance (Homoscedasticity),
which means constant error variance. Independence of Error, which tells
us that there is no relationship between error and independent variable;
this effect is known as autocorrelation.
1. The ___________ is to mean as the standard error is to the
regression line.
2. If SSR is 90 and SST is 130, coefficient of determination will be
___________.
3. In a regression output if P–value is 0.029 at 95% confidence level,
we can say that input variable is ___________ for the study.
1. In a simple linear regression, if the input variable will change by
one unit, how much the output variable will be changed (impacted)
a. By intercept b. By its slope c. No change d. By one unit
2. In a regression equation, Y = 0 + 1X + . 0 and 1 refer to:
a. Intercept and slope b. Error and Intercept
c. Slope and Error d. Intercept and Error
3. In simple linear regression, how many coefficients do we need to
estimate
a. 1 b. 2 c. 0 d. Can't say
4. To evaluate a regression model, which of the following metrics can
be used:
a. R squared b. Standard Error
c. Residual analysis d. All of the above
5. In simple linear regression, what does the slope represents
a. The value of Y when the input variable is zero
b. The estimated change in the input variable per unit change in
the output variable
c. The estimated change in output variable per unit change in the
input variable
d. Variation around the best–fit line
147
Business Analytics 6. How can we explain the standard error
a. It is defined as a variation around the regression line
b. Total error in regression model divided by its standard deviation
c. Total error in the regression model
d. Total variation in output and input variable together
7. How can we define the coefficient of determination in simple linear
regression
a. SSR/SST
b. Square of the coefficient of correlation
c. The proportion of variation in the output variable explained by
the input variable
d. All of the above
8. How can we define residual in a regression model
a. Difference between the variation of input and output variable
b. Difference between the predicted value and observed value
c. Difference between the input and output variable
d. All of the above
9. Which of the following statements are correct
a. Residual in a regression model must be distributed normally
b. Residuals must have constant variance
c. There should not be any significant relation between the input
variable and residuals
d. All of the above
10. Which of the below option is correct about the below two statements:
Statement 1 : Constant variance among residuals is known as
homoscedasticity
Statement 2 : Relationship between the input variable and residuals
is known as autocorrelation
a. Only 1 is correct b. Only 2 is correct
c. Both are correct d. Both are incorrect
2.8 Let Us Sum Up :

• Regression is an essential technique in the business world to predict
business metrics
• Simple linear regression helps to understand the relationship between
input and output variables and also provide mathematical equation
so that by supplying a value of input variable, we can estimate the
value of the output variable
• Regression parameters can be estimated using the method of ordinary
least squares under the assumption that the residuals follow a normal
148
distribution. The method of ordinary least squares gives the best Simple Linear Regression
linear unbiased estimate.
• The efficacy of the regression model can be validated by measures
like R2, standard error, significance F value, etc.
• Residual analysis is performed to check whether the model satisfies
assumptions such as normality of residuals, homoscedasticity,
autocorrelation, etc.

1. one, one 2. predictive 3. linear
1. mean 2. residuals 3. zero
1. Sum of squares error (SSE)
2. SSR, SST 3. 2
1. Standard deviation 2. 0.6923 3. significant
1. b 2. a 3. b 4. d 5. c
6. a 7. d 8. b 9. d 10. c
2.10 Glossary :
Simple Linear Regression : It is an important statistical technique
for finding the linear (straight line) relationship between an input and
output variable. Regression gives us a mathematical expression by which
we can predict the value of the output variable by supplying a value
of the input variable.
Total Sum of Squares (SST) : Total sum of squares in a model
can be measured as the summation of total variation explained by the
regression model (SSR) and unexplained variation due to random error
(SSE)
Coefficient of Determination (R2) : It is a measure of fit of the
regression model. It is a ratio of SSR to SST. The value of the coefficient
of determination varies between 0 and 1.
Standard Error : Standard error can be understood as a standard
deviation around the regression line. A large standard error indicates a
large amount of variation or scatters around the regression line
Residual Analysis of Regression Model : Residual or error analysis
is a vital step to check whether the assumptions of regression models
have been satisfied
149
Business Analytics Homoscedasticity : The assumption of homoscedasticity means
constant error variance across the range of input variable
Autocorrelation : The input variable must be independent of error
terms. The Durbin–Watson statistic can measure the autocorrelation effect.
2.11 Assignment :
1. What is residual in simple linear regression.
2. What do we mean by the baseline prediction model in simple linear
regression ?
3. Write down important steps in simple linear regression model
calculation.
4. In simple linear regression equation, Yi = 23x – 7. Explain the
relationship between the x and y variable.
2.12 Activities :
Company Nirja Plc has produced its expenses (input variable) and
revenue (output variable) for the last 10 years:
Year Expenses (In million $) Revenue (In million $)
2009 3506 4108
2010 3518 4190
2011 3609 4278
2012 3689 4295
2013 3712 4307
2014 3754 4389
2015 3798 4401
2016 3803 4456
2017 3874 4497
2018 3967 4501
2019 3994 4508
a. Make a scatter plot and check whether it is showing the linear
relationship, and are there a few outliers ?
b. Calculate the Pearson's correlation coefficient and coefficient of
determination (r2)
c. Calculate the regression equation
d. Calculate standard error, the total sum of squares, the sum of square
error, and a regression sum of squares
e. Validate the regression equation through residual analysis
150
Ans. : Simple Linear Regression
• Yes, the scatter plot is showing a linear relationship. There are not
any outliers visible in the scatter plot
• Pearson's correlation coefficient = .9617 and coefficient of

determination (r2) = 0.9248
• Revenue = 1403.10 + 0.788 expenses
• Standard error = 38.60, SST = 178372.18, SSE = 13412.05,
SSR = 164960.13
The residual plot is not showing any visible trend
Errors are in a straight line with a slight curve, need to understand

the reason for the curve in the residual plot
2.13 Case Study :

department head to strengthen its presence in India and the Asian market.
He wants to analyze sales and advertisement costs to examine whether
different advertising channels are providing enough sales or not. He asked
the MIS team to provide him with sales data and marketing costs in the
last 10 years:
151
Business Analytics Year Sales Marketing Cost
(in a million rupees) (in a million rupees)
2009 2064 31
2010 2389 36
2011 2418 37
2012 2509 35
2013 2608 38
2014 2706 41
2015 2802 44
2016 2905 45
2017 3056 47
2018 3189 49
a. Make a scatter plot and check whether it is showing the linear
relationship, and are there a few outliers ?
b. Calculate the Pearson's correlation coefficient and coefficient of
determination (r2)
c. Calculate the regression equation
d. Calculate standard error, the total sum of squares, the sum of square
error, and a regression sum of squares
e. Validate the regression equation through residual analysis

• Ken Black, "Business Statistics: Contemporary Decision Making",
John Wiley and Sons (2009)
• U Dinesh Kumar, "Business Analytics: The Science of Data–Driven
Decision Making", Wiley (2017)
• Naval Bajpai, "Business Research Methods", Pearson (2013)
152
Unit MULTIPLE LINEAR REGRESSION
3
: UNIT STRUCTURE :
3.1 Introduction
3.2 Essence of Multiple Linear Regression
3.2.1 Introduction to Multiple Linear Regression
3.3 Understanding the Concept of Multiple Linear Regression with
a Worked Example
3.4 The Correlation Coefficient for Multiple Linear Regression
3.5 Coefficient of Coefficient (R2), Adjusted R2, and Standard Error
3.6 Multiple Linear Regression in MS Excel
3.6.1 The Modified Regression Model in Excel
3.6.2 Residual Analysis to Test the Regression Assumptions
3.7 Let Us Sum Up
3.9 Glossary
3.10 Assignment
3.11 Activities
3.12 Case Study

• Understand the concept of multiple linear regression and its
• Understand the concept of coefficient of multiple determination and
adjusted coefficient of multiple determination
• Interpretation of multiple regression parameters and the concept of
partial regression coefficients
• Understand the effect of multicollinearity and auto–correlation on
the regression model
• Residual analysis for multiple regression model
• Running multiple regression on MS Excel and interpretation of the
model output
3.1 Introduction :
In this unit, we will study multiple linear regression where we see
the relationship between one output variable and multiple input variables.
It resembles business problems that we encounter often in the corporate
153
Business Analytics world. We will see various techniques to validate the multiple linear
regression. In the end, we will see various examples to see the effectiveness
of multiple linear and its influence in decision making.
3.2 Essence of Multiple Linear Regression :

Multiple linear regression is an important statistical technique for
finding the relationship between one dependent variable and multiple
independent variables. In business scenarios generally, any performance
metric depends on various factors; hence multiple regression is used most
often than simple linear regression. Besides finding the relationship, it
also helps us to find the most important factors influencing the business
metric so that instead of controlling various factors, we can concentrate
on vital few factors. For example, the percentage salary hike of an
employee may depend on his/her performance rating, working hours,
office environment, technology, salary, career growth options etc. or
mileage of a car can be dependent on the age of the car, engine power,
fuel type, engine technology, city/countryside driving, driver proficiency
etc. For multiple regression model output variable should be continuous
(ratio or interval) and independent variables can be either continuous or
discrete (nominal or ordinal scale).
Ideally, in the regression model, there should not be any relationship

between independent variables because if there is a relationship between
independent variables, then we can not calculate their actual impact on
the dependent variable. When dependent variables show a relationship,
then we call it multicollinearity. We will discuss the ways to tackle
multicollinearity in the next section.
On the other hand, too many independent variables explain the
higher variance of the dependent variable than actually, it holds but also
increases the risk of over–fitting of the regression model. This means
the model will work very nicely on training data but will not work fine
on the test and unknown data. We will discuss the over–fitting problem
in detail in the next section. Here point to be noted is that having more
variables will not solve the purpose instead few variables with a significant
relationship with dependent variables will constitute a better regression
model.
154
3.2.1 Introduction to Multiple Linear Regression : Multiple Linear
In simple linear regression, we discussed the probabilistic regression Regression
model as:
Yi = 0 + 1Xi
+ i
In the case of multiple regression as we have more than one
independent variable. Hence above probabilistic model can be extended
to multiple probabilistic regression models as:
Yi = 0 + 1X1 + 2X2 + 3X3 + 4X4 + ..... +
+ i kXk
Where, Yi is the ith value of the dependent variable, 0 the
y–intercept, 1 the slope of y with input variable X1 by keeping X2,
X3, X4……, Xk constant. Similarly, k the slope of y with input variable
Xk by keeping X1, X2, X3……, Xk–1 constant. i is a random error in
y for ith observation.
In multiple regression analysis, i is the slope of y with independent
variable xi keeping all other independent variables constant. This is also
referred to as the partial regression coefficient for the independent variable
xi.
To predict the value of y, we have to calculate the value of 0,
1, 2,……… k. But this can be done only if we have access to the
entire population, which is not possible most of the time in the business
world. Hence we use sample data to predict the value of these regression
coefficients. So the equation for estimative y with the sample information
is given as:
y = b0 + b1x1 + b2x2 + b3x3 + ............. + bkxk

For example,
So in the above example if x1 will increase by one unit, then y

will increase by .014 units provided x2 and x3 are constant.
1. Multiple linear regression has___________ independent variables
and ___________ dependent variable.
2. If we have large number of independent variables then it may lead
to the ___________ problem.
3. If there is a significant relationship between two independent
variables, it is known as ___________.
155
Business Analytics 3.3 Understanding the Concept of Multiple Linear Regression with
a Worked Example :
Let's consider an example of a supply chain of an online retail
company; their logistic arm tries to deliver the packages on the same
day. They try to optimize their delivery associate's trips by strategizing
trips with the help of city maps to reduce time and fuel costs. Below
is the data for 10 random trips, each trip has three important pieces of
information – Total miles travelled, number of deliveries, daily gas price
and a total time of travel in hours.
As an analyst, you want to estimate how long delivery will take
(dependent variable) based on two inputs – total distance, number of
deliveries and daily gas (fuel) price (independent variables).
MileTraveled numDeliveries gasPrice travelTime(hrs)
89 4 3.84 7
66 1 3.19 5.4
78 3 3.78 6.6
111 6 3.89 7.4
44 1 3.57 4.8
77 3 3.57 6.4
80 3 3.03 7
66 2 3.51 5.6
109 5 3.54 7.3
76 3 3.25 6.4
Below is the pictorial view of the relationship between the dependent
and independent variables.
Before starting calculations, it is always recommended to use the

scatter plot to visualize the relationship of every independent variable
with the dependent variable separately.
156
Multiple Linear
Regression
Here we saw miles travelled and the number of deliveries showing

a significant relationship while gas price does not show any relationship
with the dependent variable (travel time).
As gas price does not appear to be correlated with travel time,
we should not use it in a multiple regression model.
But we are keeping it for the learning process so that we can see
the consequences of having a non–significant variable in the model.
Now we should also see the relationship among independent
variables.
So miles travelled, and the number of deliveries is showing a strong

relationship; hence it is a problem (multicollinearity). Ideally, we should
not use BOTH in the regression model; instead, we should use either
of them, which makes more sense from the business point of view.
But for learning purposes, let's keep both in the regression model.
We will remove it in the later stage of the model building.
It is recommended that we should also see the correlation coefficient
between these variables in case it is not visible in the graph then we
can rely on the numerical coefficient better.
157
Business Analytics
So correlation coefficients tell us clearly, which variables have to

include in the model and which one has to remove.
1. In multiple linear regression, there is ___________ relationship
between dependent and independent variables
2. For a good multiple linear regression Scatter plot should
show___________ relationship between each individual variable
and dependent variable.
3. If an independent variable is found to be highly significant in a
regression model, we can conclude that the changes in ___________
variable are associted to the changes in ___________ variable.
3.4 The Correlation Coefficient for Multiple Linear Regression :

In the case of simple linear regression, we calculated the relationship
between one independent and one dependent variable. In the case of
multiple linear regression, we see the relationship between the dependent
variable and the cumulative impact of all independent variables. Here
we see the relationship between the dependent variable (y) and predicted
variable ( yi ). Let's say below are the coefficient of the regression model.
So it is a correlation between column D (y) and column F ( yi ) :

A B C D E F
Mile num gas Time travelTime_Predicted travelTime
Traveled Deliveries Price travel Predicted
89 4 3.84 7 6.21+0.01*A3+0.38*B3–0.61*C3 6.67
66 1 3.19 5.4 6.21+0.01*A4+0.38*B4–0.61*C4 5.59
78 3 3.78 6.6 6.21+0.01*A5+0.38*B5–0.61*C5 6.17
111 6 3.89 7.4 6.21+0.01*A6+0.38*B6–0.61*C6 7.72
44 1 3.57 4.8 6.21+0.01*A7+0.38*B7–0.61*C7 5.05
77 3 3.57 6.4 6.21+0.01*A8+0.38*B8–0.61*C8 6.28
80 3 3.03 7 6.21+0.01*A9+0.38*B9–0.61*C9 6.65
66 2 3.51 5.6 6.21+0.01*A10+0.38*B10–0.61*C10 5.78
109 5 3.54 7.3 6.21+0.01*A11+0.38*B11–0.61*C11 7.52
76 3 3.25 6.4 6.21+0.01*A12+0.38*B12–0.61*C12 6.46
158
Correlation Coefficient travelTime(hrs) (y) travelTime Predicted Multiple Linear
Regression
travelTime(hrs) (y) 1
travelTime Predicted 0.9459 1
Therefore correlation coefficient between independent variables and
the dependent variable is .9459. If we calculate the square of it, then
it will become a coefficient of determination (R2), which is 0.8947 in
the above example.
3.5 Coefficient of Coefficient (R2), Adjusted R2, and Standard

Error :
In the last unit, we calculated SSR, SSE and SST for simple linear
regression. Calculations remain the same in the case of multiple linear
regression also:
• Total sum of squares (SST) is the difference between the mean
of y and each observed value (yi)
Σ (y i − y)2
Total Sum of Squares = SST =
• The regression sum of squares (SSR) is the difference between the

mean of y and each predicted value ( yi )
Σ(y i − y)2
Regression Sum of Squares = SSR =
• Error sum of squares (SSE) is the difference between each observed

value of y and each predicted value ( yi )
Σ(y i − y i )2
Error Sum of Squares = SSE =
Mile num gas travel
(y i - y¯) (yi - y¯)2 (y î - y¯) (y î - y¯)2 (y i - y î) (yi - yî)2
Traveled Deliveries Price Time (y)
89 4 3.84 7 0.61 0.3721 0.2817 0.0793 0.3283 0.1078
66 1 3.19 5.4 -0.99 0.9801 -0.7983 0.6373 -0.1917 0.0367
78 3 3.78 6.6 0.21 0.0441 -0.2204 0.0486 0.4304 0.1853
111 6 3.89 7.4 1.01 1.0201 1.3283 1.7644 -0.3183 0.1013
44 1 3.57 4.8 -1.59 2.5281 -1.3395 1.7943 -0.2505 0.0627
77 3 3.57 6.4 0.01 0.0001 -0.1072 0.0115 0.1172 0.0137
80 3 3.03 7 0.61 0.3721 0.2627 0.0690 0.3473 0.1206
66 2 3.51 5.6 -0.79 0.6241 -0.6093 0.3712 -0.1807 0.0327
109 5 3.54 7.3 0.91 0.8281 1.1292 1.2751 -0.2192 0.0481
76 3 3.25 6.4 0.01 0.0001 0.0728 0.0053 -0.0628 0.0039
Average 6.39 Sum 6.769 6.0561 0.7129
• Σ(y i − y)2 =
Total Sum of Squares = SST = 6.769
• Σ(y i − y)2 =
Regression Sum of Squares = SSR = 6.0561
• Σ(y i − y i )2 =
Error Sum of Squares = SSE = 0.7129
In the case of simple linear regression, R2 explains the total

variation in output (dependent) variable explained by an independent
159
Business Analytics variable. In the case of multiple linear regression, R2 is the variation
in the dependent variable explained by the combination of independent
variables.
SSR SSE
R2 = =1−
SST SST
SSR SSE 6.0561 .7129

R2 = =1− = =1− = 0.8947
SST SST 6.769 6.769
In the last section, we calculated the multiple correlation coefficient
= .9459. Therefore if we square this multiple correlation coefficient, then
it is 0.8947.
Adding any new variable in the regression model does not change
the total sum of squares (SST) while it is likely to increase the SSR.
Sometimes despite the newly added variable not making the regression
model better but still it increases the R2 value, hence considering the
R2 value along as efficacy of the regression model is wrong.
This problem can be tackled by another measure called adjusted
R2; it considers additional information added by new variable(s) and
changed in the degree of freedom.
SSE /(n − k − 1)
Adjusted R 2 = 1 −
SST /(n − 1)
.7129 /(10 − 3 − 1)
1−
= 0.8420
=
6.769 /(10 − 1)
Where n is the sample size taken and k is the number of independent

variables.
If more insignificant variables are added to the regression model,
the difference between R2 and adjusted R2 will be larger. In the case
of adjusted R2 total variation in the dependent variable is adjusted by
the sample size and number of independent variables.
The standard error is the standard deviation of residuals around the
regression line. In the case of multiple linear regression, there is a slightly
different formula for standard error which is adjusted by sample size
(number of data points and number of independent variables in the model)
and number of independent variables.
Standard error
= SSE /(n − k − 1)
= .7129 /(10 =
− 3 − 1) 0.3447
160
Check Your Progress – 3 : Multiple Linear
1. In multiple linear regression, correlation coefficient is calculated Regression
between dependent variable and ___________.
2. Adjusted R2 is always ___________ than R2 value.
3. Standard error in multiple linear regression is adjusted by
___________ and ___________.
3.6 Multiple Linear Regression in MS Excel :

For the same example, estimate how long delivery will take
(dependent variable) based on three inputs – total distance, number of
deliveries and daily gas (fuel) price (independent variables).
Below is the output of MS Excel for the above example:
Below is the explanation for the essential components of the

regression output:
1 The correlation coefficient for the independent variable and all
independent variables, .9459
SSR
2 The coefficient of determination (R2) .8947 ' 89.47%
SST
variation in output variable due to input variables
SSE /(n − k −1) .7129 /(10 − 3 − 1)

3 Adjusted R 2 → 1 − =1− = 0.8420
SST /(n − 1) 6.769 /(10 − 1)
4 The standard error SSE /(n=

− k −1) .7129 /(10
= − 3 −1) 0.3447
5 Regression sum of squares (SSR) 6.0561
161
Business Analytics 6 Error sum of squares (SSE) 0.7129
7 Total sum of squares (SST) 6.6769
8 The overall significance of the model, if the value is less than .05
at 95% confidence level, then the overall model showing a significant
relationship between input and output variables
9 y–intercept 6.21
10 Cofficient for first independent variable (MileTraveled), x1 0.01

11 Cofficient for second independent variable (numDeliveries),
x2 0.3831
12 Cofficient for third independent variable (gasPrice), x3 –0.606
13 The P–value for the first independent variable (MileTraveled), is
less than .05 at 95% CL, which indicates input variable (MileTraveled)
is NOT significant in the model
14 The P–value for the second independent variable (numDeliveries),
is less than .05 at 95% CL, which indicates the input variable
(MileTraveled) is NOT significant in the model
15 The P–value for the third independent variable (numDeliveries), is
less than .05 at 95% CL, which indicates the input variable (gasPrice)
is NOT significant in the model
We have observed that the overall model is significant (point number
8) while all three independent variables are not significant. It is a problem
due to multicollinearity. Below are two issues with the above model:
1. The first two independent variables are highly correlated
2. The third variable is not correlated with the dependent variable
1. In multiple linear regression, F statistics tells us about ___________
of overall model.
2. At 95% confidence level, p–value of each independent variable
should be less than ___________ .
3. At 95% confidence level, if overall model has p–value less than
.05 but each independent variable has p–value more than .05. In
that scenario, there can be an issue of ___________ in the regression
model.
3.6.1 The Modified Regression Model in Excel :
It is always recommending that we should apply the above remedies
to improve the model in a stepwise manner, so let's remove the third
variable (gasPrice) from the model and rerun to see how it is performing:
162
Multiple Linear
Regression
Still, we are observing that the overall model is significant, but

both variables are non–significant while individually, they are positively
correlated with the dependent variable.
So now we have to go with a single linear regression and check
the model for each variable in order to determine the best predicting
independent variable. The below table shows a comparable study for all
significant combinations of independent variables.
2
F P- Standard R Mile Traveled numDeliveries gasPrice Model
Value Value Error Adjusted Significant Significant Significant Selected
49.77 <.001 .3423 84.42% v vv
41.96 <.001 .3680 81.99% v v
0.62 .455 .8864 0.00% X X
23.72 <.001 .3526 59.95% X X X
22.63 <.001 .3599 68.11% X X X
27.63 <.001 .3297 71.76% X X X
16.99 .002 .3447 57.49% X X X X
So, in this case, simple linear regression with MileTraveled as an

independent variable makes the most significant model. So it is not
always necessary that for the same dependent variable, multiple linear
regression will make a better model than simple linear regression.
3.6.2 Residual Analysis to Test the Regression Assumptions :
Residual analysis is mainly used to test the assumptions of the
regression model. Below two are important residual analyses provided
by MS Excel:
1. The Linearity of the Regression Model : The linearity of the
regression model can be obtained by plotting the residual on the
vertical axis against the corresponding xi values of the independent
variable on the horizontal axis. There should not be any apparent
pattern in the plot for a fit regression model.
163
Business Analytics
2. Normality of Error : The normality probability plot of the residuals

should roughly follow a straight line for meeting the assumption
of normality. A straight line correcting all the residuals indicates
that the residuals are normally distributed. It means that residuals
are random in nature, and there is not any hidden trend. A curve
in the tail is an indication of skewness.
There are some more advanced assumption validations, which are

not part of MS Excel like Constant error variance (Homoscedasticity),
which means constant error variance. Independence of Error, which tells
us that there is no relationship between error and independent variable;
this effect is known as autocorrelation.
Question 1 – 5 : Use below data to answer the questions
Analytical Score Hours Study Final Score
125 30 100
104 40 95
110 25 92
105 20 90
100 20 85
100 20 80
95 15 78
95 10 75
85 0 72
90 5 65
164
1. The value of R2 for the above data is: Multiple Linear
a. 0.9515 b. 0.9053 c. 0.8782 d. 3.8751 Regression
2. The value of SSR and SST for the above data is:
a. 1004.484 and 1109.6 b. 105.1158 and 1109.6
c. 502.2421 and 1004.484 d. 33.4459 and 1109.6
3. Value of Standard error for the above data is:
a. 0.9515 b. 0.9053 c. 0.8782 d. 3.8751
4. Value of Adjusted R2 for the above data is:
a. 0.9515 b. 0.9053 c. 0.8782 d. 3.8751
5. By visualizing scatter plot and regression output, we can conclude
that:
a. Data have multicollinearity issue
b. Data has overfitting issue
c. Data has both multicollinearity and overfitting issues
d. Data has neither multicollinearity nor overfitting issue
6. If there is multicollinearity in the model, it may result in:
a. The regression coefficient may be in the opposite sign
b. May add a new variable that is not significant
c. Removing a significant variable from the data
d. All of the above
7. If we have added a new variable in the model then:
Statement 1 : R2 will always increase
Statement 2: Adjusted R2 will always increase
a. Statement 1 is always true
b. Statement 2 is always true
c. a and b both statements are correct
d. a and b both statements are incorrect
8. In the case of multiple linear regression, the correlation coefficient
is:
a. Correlation between the dependent variable and predicted variable
b. The square root of the coefficient of determination
c. a and b both options are correct
d. a and b both options are incorrect
165
Business Analytics 9. In the below formula of standard error, n and k represent:
Standard error= SSE /(n − k − 1)

a. n and k are sample size and number of independent variables
respectively
b. n and k are the number of independent variables respectively
and sample size respectively
c. n is the degree of freedom and k is the sample size
10. In the case of two independent variables:
a. Running two separate simple linear regression is always better
than multiple linear regression
b. It depends on the relationship of each independent variable and
dependent variable
c. Running multiple linear regression would always be better than
two separate simple linear regression
3.7 Let Us Sum Up :

• Multiple linear regression is a statistical extension of simple linear
regression where two or more independent variables used to predict
variance in one dependent variable.
• In a multiple regression model, each coefficient is interpreted as
the estimated change in the dependent variable (y) corresponding
to a one–unit change in the independent variable when all other
variables are held constant.
• Too many independent variables may cause to overfitting of the
regression model as these variables keep increasing coefficient
determination R2.
• The correlation coefficient in the case of multiple linear regression
is calculated between the dependent variable and predicted variable.
• R2 adjusted helps in determining whether the newly added variable
is adding value to the model or not; it is a better measure than
R2 alone.
• Multicollinearity happens when some independent variables are
correlated with each other.
• In the case of multiple linear regression, R2 explains the total
variation in output (dependent) variable explained by an independent
variable while in the case of adjusted R2, the total variation in the
dependent variable is adjusted by the sample size and number of
independent variables.
• In multiple linear regression, the overall model significance is
calculated by F–test.
166
• If their a multicollinearity issue in the model then there are chances Multiple Linear
to remove the significant independent variable. Regression
• Residual normality plot should show a straight line while there
should not be any relationship between residuals and independent
variables.

1. Two or more, one 2. overfitting
3. multicollinearity
1. Many to one 2. Linear
3. Independent, dependent
1. Predicted variable 2. less
3. Sample size, number of independent variables
1. Significance 2. 0.05 3. multicollinearity
1. b 2. a 3. d 4. c 5. d
6. c 7. a 8. c 9. a 10. b
3.9 Glossary :
Multiple Linear Regression : Multiple linear regression is an
important statistical technique for finding the relationship between one
dependent variable and multiple independent variables.
Multiple R (Multiple Correlation Coefficient) : In multiple linear
regression, correlation coefficient represents the relationship between the
dependent variable and predicted dependent variable.
Overfitting : In multiple linear regression if there are too many
significant variables in the model, then there would be a higher fictitious
coefficient of determination. Also, if we have not split and validated the
model on test data, then also model may be overfitted on train data.
Multicollinearity : If independent variables show a strong
relationship among themselves, then it is difficult to measure the actual
impact of an independent variable on the dependent variable as an
independent variable also has an impact on another independent variable.
Residual Analysis : Residual analysis validates the regression
assumptions like residuals are random in nature, and there is not any
hidden trend. Another assumption is the residuals do not show any
relation with the independent variable.
167
Business Analytics Adjusted R2 : Coefficient of determination (R2) increase whenever
we add a new variable in the model irrespective of its significance, hence
sometimes R2 tells us fictitious model accuracy. Adjusted R2 is a measure
where R2 get adjusted by the degree of freedom (sample size and the
number of independent variables).
Regression Equation : It is a mathematical expression that represents
the mathematical relationship between the dependent variable and
independent variables. If we have values of all independent variables then
with the help of the regression equation, we can predict the value of
the dependent variable.
3.10 Assignment :
1. Why it is always recommended to see a scatter plot before applying
multiple linear regression. Explain with an example ?
2. What do we mean by non-linear regression, explain with an example ?
3. What do we mean by y-intercept, explain its role in the interpretation
of regression equation ?
3.11 Activities :
Below is the data for car sales. Here "Price" is the dependent
variable, while all others are independent variables. Calculate SST, SSR,
SSE, standard error, R2 and adjusted R2, regression equation for the below
data. Also, justify the difference between R2 and adjusted R2.
engine_s horsepower wheelbase width length curb_w gt fuel_cap mpg price
1.8 140 101.2 67.3 172.4 2.639 13.2 28 21.5
3.2 225 106.9 70.6 192 3.47 17.2 26 27.3
3.2 225 108.1 70.3 192.9 3.517 17.2 25 28.4
3.5 210 114.6 71.4 196.6 3.85 18 22 42
1.8 150 102.6 68.2 178 2.998 16.4 27 23.99
2.8 200 108.7 76.1 192 3.561 18.5 22 33.95
4.2 310 113 74 198.2 3.902 23.7 21 62
2.5 170 107.3 68.4 176 3.179 16.6 26.1 26.99
2.8 193 107.3 68.5 176 3.197 16.6 24 33.4
2.8 193 111.4 70.9 188 3.472 18.5 24.8 38.9
3.1 175 109 72.7 194.6 3.368 17.5 25 21.975
3.8 240 109 72.7 196.2 3.543 17.5 23 25.3
3.8 205 112.2 73.5 200 3.591 17.5 25 27.885
3.8 205 113.8 74.7 206.8 3.778 18.5 24 31.965
3 200 107.4 70.3 194.8 3.77 18 22 31.01
4.6 275 108 75.5 200.6 3.843 19 22 39.665
4.6 275 115.3 74.5 207.2 3.978 18.5 22 39.895
5.7 255 117.5 77 201.2 5.572 30 15 46.225
1 55 93.1 62.6 149.4 1.895 10.3 45 9.235
1.8 120 97.1 66.7 174.3 2.398 13.2 33 13.96
168
Ans. : Multiple Linear
Regression Regression
Statistics
Multiple R 0.9460
R Square 0.8950
Adjusted R
Square 0.8187
Standard Error 4.9892
Observations 20
ANOVA
df SS MS F Significance F
Regression 8 2334.10 291.76 11.72 0.00
Residual 11 273.81 24.89
Total 19 2607.91
Standard
Coefficients Error t Stat P-value Lower 95%
Intercept -116.159 78.407 -1.481 0.167 -288.731
engine_s -6.055 4.886 -1.239 0.241 -16.809
horsepow 0.203 0.065 3.133 0.010 0.060
wheelbas 1.310 0.520 2.521 0.028 0.166
width 0.261 0.954 0.274 0.789 -1.840
length -0.359 0.287 -1.250 0.237 -0.990
curb_wgt -3.136 10.581 -0.296 0.772 -26.425
fuel_cap 1.831 1.203 1.521 0.156 -0.818
mpg 0.475 0.661 0.718 0.488 -0.980
3.12 Case Study :

Ranco automobile has established a business for car repair and
service in Ahmedabad. They are receiving a few random complaints about
car mileage. To understand the factors affecting car mileage (mpg), they
collected data for 20 cars for various attributes. Below is the data:
engine_s horsepower wheelbase width length curb_wgt fuel_cap mpg
2.5 163 103.7 69.7 190.9 2.967 15.9 24
2.5 168 106 69.2 193 3.332 16 24
2.7 200 113 74.4 209.1 3.452 17 26
3.5 253 113 74.4 207.7 3.564 17 23
3.5 253 113 74.4 197.8 3.567 17 23
2 132 105 74.4 174.4 2.567 12.5 29
2.5 163 103.7 69.1 190.2 2.879 15.9 24
2.5 168 108 71 186 3.058 16 24
8 450 96.2 75.7 176.7 3.375 19 16
2.4 150 113.3 76.8 186.3 3.533 20 24
2.5 120 131 71.5 215 3.557 22 19
3.9 175 109.6 78.8 192.6 4.245 32 15
3.9 175 127.2 78.8 208.5 4.298 32 16
5.2 230 115.7 71.7 193.5 4.394 25 17
5.2 230 138.7 79.3 224.2 4.47 26 17
2 110 98.4 67 174.7 2.468 12.7 30
2 107 103 66.9 174.8 2.564 13.2 30
2.5 170 106.5 69.1 184.6 2.769 15 25
2.5 119 117.5 69.4 200.7 3.086 20 23
3.8 190 101.3 73.1 183.2 3.203 15.7 24
169
Business Analytics Consider mpg (mileage per gallon) as predicted variable and answers
for the below questions:
1. Calculate the correlation coefficient between the dependent variable
(mpg) and independent variables
2. Calculate R2 and adjusted R2
3. Calculate SSR, SSE, SST and standard error
4. Calculate the regression equation
5. Scatter plot between the dependent variable and top 3 independent
variables

• Ken Black, "Business Statistics: Contemporary Decision Making",
John Wiley and Sons (2009)
• U Dinesh Kumar, "Business Analytics: The Science of Data–Driven
Decision Making", Wiley (2017)
• Naval Bajpai, "Business Research Methods", Pearson (2013)
170
BLOCK SUMMARY
Correlation plays a vital role in the corporate world as decision
making about any business metric always depends on various factors.
Therefore we can not take decisions in silos; we need to make a robust
strategy about all positive and negative consequences of our decision. At
one side correlation validates all these relationships while we take the help
of simple and multiple linear regression to establish mathematical relations
between these business metrics (dependent and independent variables).
Regression models help us to predict the value of output metrics like sales,
revenue, productivity, time etc. based on various factors (independent
variables) like manpower requirement, marketing budget, number of sales
pitches etc.
We have to be cautious at the time of doing regression analysis as
model accuracy and validity depends on selecting the right independent
variables. We should not unnecessary try to add independent variables as
one side; it rises the problem of overfitting another side practically it is
difficult to control too many input variables in the model. Regression also
has two critical assumptions about independent variables – firstly, there
should be a linear relationship between the variables (relationship should
not be curvilinear or zig–zag); secondly, independent variables should not
have a significant relationship among themselves (multicollinearity problem).
Regression is also very sensitive towards residual analysis, therefore, it is
utterly crucial to observe and analyze residuals as it may lead to
autocorrelation (the relationship between residual and independent variables)
or heteroscedasticity (error variance is not constant).
We should not supply the entire data into the regression model as it
leads to overfitting. The model works very well for trained data but behaves
poorly for unobserved data. Therefore, it is vital to split the entire data into
training and test set (80% training and 20% test). Before exposing the
model to the external world, it should be validated on the test dataset.
Regression is also an important foundation for various important tools and
techniques in predictive and prescriptive analytics. Currently, most machine
learning algorithms use regression as a base methodology.
171
Short Questions
1. Explain the fundamental difference between covariance and correlation
2. Write down three scenarios where Spearman rank correlation can be
better than the Pearson correlation coefficient
3. Regression analysis plays a vital role in decision making. Explain this
statement
4. Explain the concept of SST, SSE and SSR in a regression model
5. Explain the importance of the coefficient of multiple determination
(R2) in interpreting the multiple regression output
6. Why adjusted R2 is better than the coefficient of multiple determination
(R2), explain with an example
172
Long Questions
1. Explain the relationship between variance and covariance also derive
the Pearson correlation formula from the covariance formula
2. Write down the assumptions of simple linear regression analysis,
write down the consequences of these assumptions that are not
validated by the model
3. Explain the concept of standard error and coefficient of determination
and their significance in the regression model
4. Explain the importance of residual analysis in the regression model,
define different types of residual assumptions
5. Explain the steps of carrying out multiple linear regression and the
significance of each step in decision making
173
Unit No. 1 2 3
No. of Hrs.
of the block :

.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
174
Ambedkar
OpenUniversity
Business Analytics
BLOCK-4 TIME SERIES ANALYSIS
UNIT 1
INTRODUCTION TO FORECASTING TECHNIQUES
UNIT 2
MOVING AVERAGE AND SINGLE EXPONENTIAL SMOOTHING
TECHNIQUES
UNIT 3
REGRESSION METHODS FOR FORECASTING
UNIT 4
AUTO–REGRESSION (AR) AND MOVING AVERAGE (MA)
FORECASTING MODELS
BLOCK 4 : TIME SERIES ANALYSIS
Block Introduction
One of the main objectives of business analytics is to forecast

important business metrics like sales, raw material, manpower, budget etc.
Therefore, forecasting is among the most important and fre–quently
addressed in analytics. Forecasting is the process of predicting the future
values of any business metric based on historic values for the same metric.
For example, forecasting of sales data for next month based on historic sales
data for the last 24 months. For forecasting, we use Time Series Analysis
techniques. In the current era of globalization, the supply chain plays the
most important role in the success of any organization and forecasting or
time series analysis is the backbone of supply chain analytics which helps
the organizations in predicting resources at each milestone of the entire
supply chain (organizational value chain). In case the forecasting of an
organization is not up to the mark then it will have a severe impact on the
organization's top–line (revenue earned) and bottom–line (cost incurred in
running the businesses). It can be understood with the help of an example
like if forecasting is not correct and we have predicted less demand than
actual then either we will not have enough goods in the market hence it
would impact on the organization revenue or in case we have predicted
more than required demand then we unnecessary we would to purchase
extra raw material, deployed extra manpower and utilized extra space in
warehouses, all this will impact on the cost of the organization. Hence it
is utterly important to forecast the demand for a service/ product as
accurately as possible.
All organizations despite their size and industry prepare a short term
(weekly, monthly or quarterly) and long–term planning (half–yearly, yearly
or five–yearly etc.) and forecasting remain the most important integral part
of all these planning activities. It directly impacts the revenue and cost of
the organization.
Block Objectives
• Understand the significance of forecasting and how it influences

decision making
• Understand various important components of time series analysis
• Learn various forecasting accuracy techniques
• Learning smoothing techniques like naïve forecasting models, average
models and exponential smoothing techniques
• Application of regression theory in time series analysis
• Forecasting time series data in influence with seasonal variation
• Learn the concept of autocorrelation and how it impacts the time
series analysis
• Understanding of Auto–regressive (AR) forecasting models
Block Structure
Unit 1 : Introduction to Forecasting Techniques
Unit 2 : Moving Average and Single Exponential Smoothing Techniques
Unit 3 : Regression Methods for Forecasting
Unit 4 : Auto–Regression (AR) and Moving Average (MA) Forecasting

Models
1 FORECASTING TECHNIQUES
: UNIT STRUCTURE :
1.1 Introduction
1.2 Forecasting : Magical Crystal Ball of Statisticians
1.3 Time–Series Data and Components of Time–Series Data
1.4 Time–Series Data Modelling Techniques
1.4.1 Additive Model of Time–Series Modelling
1.4.2 Multiplicative Model of Time–Series Modelling
1.5 Measuring Forecasting Accuracy Techniques
1.5.1 Mean Absolute Error (MAE) / Mean Absolute Deviation
(MAD)
1.5.2 Mean Absolute Percentage Error (MAPE)
1.5.3 Mean Square Error (MSE)
1.5.4 Root Mean Square Error (RMSE)
1.6 Factors Affecting Forecasting Accuracy
1.7 Let Us Sum Up
1.9 Glossary
1.10 Assignment
1.11 Activities
1.12 Case Study

• Understand the significance of forecasting and how it influences
decision making
• How forecasting is an integral part of supply chain management
and the overall performance of the organization
• Understand various important components of time series analysis
• Time series data modelling techniques – additive methods and
multiplicative methods
• Learn various forecasting accuracy techniques such as mean absolute
error, mean absolute percentage error, mean square error and root
mean square error
1.0 Introduction :
In this unit, we will study the basic concepts of time series analysis
and its role in decision making. We will understand the impact of time
175
Business Analytics series analysis in an organization with the help of various industrial
examples. Later we will study various components of time series analysis
and how these components influence the overall performance of an
organization. In the end, we will see important techniques to measure the
accuracy of a time series analysis technique and how to determine which
forecasting modelling is best for a given scenario.
1.2 Forecasting : Magical Crystal Ball of Statisticians :

"Those who know don't predict. Those who predict don't have
knowledge" – Lao Tzu
Forecasting plays important role in all major departments in an
organization including man–power hiring, sales, production, quality control,
supply chain, finance etc. Therefore, time series analysis holds a very
strong role in the predictive analytics tool kit. Time series analysis impacts
both the revenue side (top–line performance) and expenditure side (bottom–
line performance) of an organization. Following are time series forecasting
examples from the business world:
• Organizations want to predict the manpower requirement for the
upcoming financial year
• The finance department wants to estimate the budget for procuring
hardware and software units for the new plant
• Municipality predicts the water shortage in Ahmedabad during May
and June month
• The weather department predicts normal rainfall in northern India
• The aviation industry estimates a downfall in demand due to the
corona pandemic in 2022 also
• There will be a significant increment in demand for the electric
vehicle due to income tax relaxation by the government
• The life insurance market will grow at the rate of 5% in the next
five years
Besides severe application in the business world, time series also
plays important role in personal life. For example, somebody wants to
invest in a mutual fund, and he must choose between two brands. Below
is the data for the last ten years return of two renowned mutual funds:
Year Average return of brand A Average return of brand B
2011 6.2 6.57
2012 6.5 6.90
2013 6.8 7.25
2014 7.2 7.61
2015 7.5 7.99
2016 7.9 8.39
2017 5.9 7.38
2018 6.4 7.75
2019 6.8 8.14
176 2020 7.1 8.55
Selection will depend on the prediction about their returns in the Introduction to
upcoming years which is an application of time series analysis. As studied Forecasting Techniques
in block 1, time–series measure the centrality and variance in the data
to estimate the future value but it is not limited to these two factors
only, we will see it in detail in upcoming topics and units.
Time series data is always collected over a period which means
the x–axis of a time series data is always time (day, date, month etc).
Line charts are the most preferred graph to visualize time–series data.
1.3 Time–Series Data and Components of Time–Series Data :

In time–series data, the x–axis always denotes time while the
y–axis represents the response variable, Yt, such as share price, productivity,
revenue, cost etc at different time points t. The variable Yt is a random
variable and generally, it is collected at regular intervals of time and
arranged in chronological order. If time–series data contains data only
for one variable, then it is called univariate time series for example daily
gold price or the number of employees who join an organization every
month. While if time–series involves more than one variable then it is
known as multivariate time–series data, for example per employee revenue
earned, per employee cost, per employee logged in hours etc collected
for each month.
Month Per employee Per employee Per employee
revenue cost logged in hours
Jan'20 1,70,000 1,15,000 176
Feb'20 1,72,000 1,17,000 176
Mar'20 1,65,000 1,16,000 172
Apr'20 1,83,000 1,19,000 175
May'20 1,79,000 1,18,000 173
Jun'20 1,84,000 1,21,000 179
177
Business Analytics Most statisticians believed that time–series data from a forecasting
perspective can be categorized into four components: trend component,
seasonal component, cyclic component, and irregular component. Most
of the time all these components are not present in time–series data.
Below is the description of each of these components:

1. Trend Component (Tt) : The trend
component depicts long term overall
downward or upward movement of time–
series data. Generally, the trend does not
repeat in time–series data, it shows the
overall slope of the time–series data. There
are two main ways to find a trend in the
data either by simply plot the data e.g.
line plot, run chart or control chart etc. another way is the
decomposition of time–series data but this is not included in the
scope of this text. For example, in Gold price in the last three years,
there are many fluctuations in the daily price but overall, there is
an increasing trend.
2. Seasonal Component (St) : When time–
series data shows repetitive downward or
upward movement/ fluctuations from the
trend then we say there is a seasonal influence
in our data. These repetitions are generally
shorter (less than a year) for example, daily,
weekly, monthly, quarterly etc. But the
frequency of this repetition is less than a year most of the time.
There can be various reasons for seasonality in the time–series data
like a weekly footfall in the shopping mall or theatres, school
holidays, festivals or end of season sales etc. The time after which
seasonal trend repeats is known as the periodicity of seasonal
variation and it repeats over a period.
3. Cyclical Component (C t ) : Cyclical
component is the movement over a longer
period (generally more than a year). These
movements or fluctuations are because of
macro economic changes while in the case
of seasonal component movement observed
within a year and depend on micro–economic factors like festivals
178
and customs that depend on society, the end of season sale etc. Introduction to
Another big difference between cyclical and seasonal components Forecasting Techniques
is that cyclical movement has not fixed time between fluctuations,
it is random in nature while seasonal component has a fixed time
period within a year generally. In another world periodicity of
cyclical fluctuations is not constant while it is constant for a
seasonal component.
4. Irregular Component (It) : Irregular
component consists of random
movements in the time series and follow
a normal distribution with mean 0 and
constant variance. Distribution with
mean zero and constant variance is
known as white noise. From time–series
data if we remove the trend, seasonal and cycle components then
residual time–series data is an irregular component.
1.4 Time–Series Data Modelling Techniques :

Data modelling plays important role in business analytics. Modelling
means if we have Time–series data as an input then we can forecast
future value for this Time–series, this predicting future capability based
on input data is known as modelling in statistics. There are two important
techniques to model Time–series data either through the addition of all
or few Time–series components or product of these components. Let's
discuss these in brief:
1.4.1 Additive Model of Time–Series Modelling :
In the Additive model we add these components like below
Yt = Tt + Ct + St + It
The basic assumption behind the additive model is that cyclical
and seasonal components are completely independent of the trend
component. Which is generally not the case in real–life scenarios hence
additive models are not very popular for forecasting purpose. An additive
model can be un–derstood with an example, there are few extra weekend
classes in an institute and there are fixed students enrolled in those classes
hence extra food demand in the canteen can be forecasted through an
additive model.
1.4.2 Multiplicative Model of Time–Series Modelling :
Multiplicative models are more common in business scenarios as
they do not have the assumption that seasonal component should be
uncorrelated with trend component. The general structure of the
multiplicative model is as below:
Yt = Tt Ct St It
The cyclical component needs long term data to be used in the
forecasting model, which is difficult to have all the time. In case long
179
Business Analytics term (few years data) is not available then we use a modified version
of the above equation:
Yt = Tt St
A multiplicative model better fits the data that's why it is more
common than the additive model.
1.5 Measuring Forecasting Accuracy Techniques :

There are several forecasting techniques, which analysts can use
to predict the future value of their desired KPI/ business measure. It is
relatively a difficult question to select the most appropriate technique as
it depends on various factors and business situations, we will cover these
techniques in detail in upcoming topics and units.
One of the major factors of choosing the right forecasting technique
is calculating the accuracy measure. We calculate accuracy measures for
different forecasting techniques and compare the forecasting error, whichever
method gives the least error become the strong contender for choosing
forecasting technique, although this approach can be computationally
expensive and time consuming for very large data sets. Besides accuracy
measure, business objective, analysis software capability and analysts
command on these techniques also plays important role in selecting the
most relevant forecasting technique for a given business scenario.
Forecasting error or deviation is the difference between forecasted
value and actual value of a Time–series. It is also known as residuals.
et = xt – Ft
Where
et = Forecasted error/ deviation for given time t
xt = Actual value at time t
Ft = Forecasted value at time t
There are four important forecasting accuracy measures used
frequently:
1. Mean Absolute Error (MAE) / Mean Absolute Deviation (MAD)
2. Mean Absolute Percentage Error (MAPE)
3. Mean Square Error (MSE)
4. Root Mean Square Error (RMSE)
1.5.1 Mean Absolute Error (MAE) / Mean Absolute Deviation (MAD):
This is the average error in forecast without considering the sing
of the error. The formula is as below:
n |x − F |
t t
MAE = ∑
t =1 n
We saw during the calculation of variance, some of the deviations

from the mean line were positive and the rest were negative hence
180
variances of the different signs were cancelling each other to overcome Introduction to
that we squared all variance terms. Another way to overcome this sign Forecasting Techniques
(direction) problem is taking absolute value instead of squaring these
terms. It will tell us the overall error in the forecasted values. Let's see
the examples of the forecasted value of an organization for the last 10
years.
Year Sales Forecast Error Value Absolute error
1 1336 -
2 1392 1477 -85 85
3 1487 1532 -45 45
4 1547 1460 87 87
5 1610 1626 -16 16
6 1689 1729 -40 40
7 1741 1669 72 72
8 1798 1760 38 38
9 1760 1752 8 8
10 1714 1733 -19 19
Sum 0 410
410
MAD = = 2.878
9
Here n = 9 (forecasting is available only for 9 observations).
This example explains that some of the forecasted errors are positive
while others are negative. If add all these error values, then most of
the time outcome will be either 0 or very close to 0 as these negative
and positive errors are cancelling each other. By taking the absolute value
we can resolve this problem. In the above example, the summation of
all error terms is zero but it is not mandatory that we will get error
summation all the time zero but for sure it will remain close to zero
all the time until there are lots of outliers in our data.
One of the important limitations of mean absolute error is that we
cannot compare the MAD score of two different studies as it directly
depends on the absolute value of the Time–series data. If one time series
in hundreds, share prices of an organization is in range of ` 600 to
` 850 while the share price of another organization is in range of
` 45,000 to ` 56,000 then MAD in case of the second organization will
be higher as their absolute values are higher. Another accuracy measure
means absolute percentage error resolve this problem.
1.5.2 Mean Absolute Percentage Error (MAPE) :
Mean absolute percentage error consider the average of absolute
percentage error instead of the average of just absolute value of the error
terms. The formula is as follows:
1 n | x t − Ft |
MAPE = ∑
n t =1 xt
181
Business Analytics As MAPE is dimensionless hence it can be used to compare two
different Time–series irrespective of the magnitude of their values.
Absolute values of
Year Sales Forecast Error Value errors divided by
actual values
1 1336 -
2 1392 1477 -85 0.061
3 1487 1532 -45 0.030
4 1547 1460 87 0.056
5 1610 1626 -16 0.010
6 1689 1729 -40 0.024
7 1741 1669 72 0.041
8 1798 1760 38 0.021
9 1760 1752 8 0.005
10 1714 1733 -19 0.011
Sum 0 0.259
.259
MAPE = × 100 = 2.878
9
1.5.3 Mean Square Error (MSE) :
When we calculate error terms in Time–series forecasting, we had
already seen the method of taking the absolute value to get rid out of
negative values. Another way to do the same thing is to calculate the
square of error terms.
n (x t − Ft )2
MSE = ∑
t =1 n
Year Sales Forecast Error Value Square of Error Terms

1 1336 -
2 1392 1477 -85 7225
3 1487 1532 -45 2025
4 1547 1460 87 7569
5 1610 1626 -16 256
6 1689 1729 -40 1600
7 1741 1669 72 5184
8 1798 1760 38 1444
9 1760 1752 8 64
10 1714 1733 -19 361
Sum 0 25728
25728
MSE
= = 2859
9
182
1.5.4 Root Mean Square Error (RMSE) : Introduction to
Square root of the mean square error is known as root mean square Forecasting Techniques
error (RMSE). It is the standard deviation of errors.
n (x − F ) 2
t t
RMSE = ∑
t =1 n
Year Sales Forecast Error Value Square of Error Terms

1 1336 -
2 1392 1477 -85 7225
3 1487 1532 -45 2025
4 1547 1460 87 7569
5 1610 1626 -16 256
6 1689 1729 -40 1600
7 1741 1669 72 5184
8 1798 1760 38 1444
9 1760 1752 8 64
10 1714 1733 -19 361
Sum 0 25728
25728
RMSE = = 2859 = 53.47
9
RMSE and MAPE are the two most popular forecasting accuracy
measures. Generally, the organization put it as a target.
Worked Example : An organization has below sales and forecasting
figures. Calculate MAD, MAPE, MSE and RMSE.
Year Actual Forecast

1 27.580 27.580
2 25.950 26.765
3 26.080 26.015
4 26.360 26.220
5 27.990 27.175
6 29.610 28.800
7 28.850 29.230
8 29.430 29.140
9 29.670 29.550
10 30.190 29.930
11 31.790 30.990
12 31.980 31.885
Solution :
In the earlier section, we have seen the formula and significance
of different forecasting accuracy measures. Here Time–series data is
available for 12 data periods. Generally, we do not have forecasting
figures available for the first period or sometimes we consider actual
183
Business Analytics sales itself forecasting number. Generally, a line chart is preferred to
visualise time–series data.
Absolute
error value
Square of
Error Value Absolute Value divided by
Year Actual Forecast Error Terms
x t - Ft |xt - Ft| actual
(xt - Ft)2
values
|xt - Ft|/xt
1 27.580 27.580 0.000 0 0.000 0.000
2 25.950 26.765 -0.815 0.815 0.664 0.031
3 26.080 26.015 0.065 0.065 0.004 0.002
4 26.360 26.220 0.140 0.14 0.020 0.005
5 27.990 27.175 0.815 0.815 0.664 0.029
6 29.610 28.800 0.810 0.81 0.656 0.027
7 28.850 29.230 -0.380 0.38 0.144 0.013
8 29.430 29.140 0.290 0.29 0.084 0.010
9 29.670 29.550 0.120 0.12 0.014 0.004
10 30.190 29.930 0.260 0.26 0.068 0.009
11 31.790 30.990 0.800 0.8 0.640 0.025
12 31.980 31.885 0.095 0.095 0.009 0.003
Sum 2.2 4.59 2.9679 0.160
Here n = 12
n |x − F | 4.59
t = t
MAD/MAE
= ∑ = 0.383
t =1 n 12
n (x − F )2 2.968
t = t
MSE
= ∑ = 0.247
t =1 n 12
n (x − F ) 2 2.968
RMSE
= ∑ t= t = 0.497
t =1 n 12
1 n | x t − Ft | .160
MAPE = ∑ = × 100 = 1.329
n t =1 xt 12
184
1.6 Factors Affecting Forecasting Accuracy : Introduction to
Forecasting Techniques
Forecasting always remain challenging as there are so many factors
that influence it, we apply various fixes to improve the forecasting
accuracy. Below are few important factors:
A. The high volume of data helps in better forecasting accuracy
: High volumes always help in attaining better forecasting accuracy
as it covers various possible business scenarios also smoothes the
impact of natural variance in the data. Therefore, forecasting a
superstore is always easier than a local confectionary or grocery
store. High volume is the prerequisite for most of the data science
techniques as large data explains its variation well.
B. Aggregation helps in improving forecasting accuracy : Forecasting
at an individual product/ service level is difficult but if we aggregate
data into a few logical groups then it becomes easy. For example,
forecasting at each course in a university may be difficult but if
we try to forecast graduate students in science streams by aggregating
individual specialization in the science department then it would
be relatively easy.
C. Short period forecasting works better than long term : A long
term period consists of various business scenarios which are not
so regular therefore longer forecasting has relatively lower accuracy.
D. Forecasting is relatively easy for a stable business : In a start–
up or new business, the unit has unstable demand which makes
forecasting challenging while on the other hand, a mature business
has regular customers and a stronger brand which helps in superior
accuracy.
Check Your Progress :
1. In Time–series data y–axis denotes the business measures while
x–axis denotes ___________.
2. Seasonality component is relatively easier to estimate while cyclical
component is ___________ to estimate.
3. Root mean square error and ___________ forecasting accuracy
measures can be compared across the industries.
4. Mean absolute error is also known as ___________.
 Multiple Choice Questions :
1. Cyclic components in a Time–series is caused due to:
a. Random events occur in local geography
b. Festivals in society
c. Macroeconomic changes
d. Changes in customer behaviour
185
Business Analytics 2. Which option is NOT a valid component of Time–Series data ?
a. Irregular component b. Seasonal component
c. Trend component d. Noise
3. White noise is:
a. Errors with mean and the standard deviation both 1
b. Uncorrelated errors with a mean value of 0 and constant standard
deviation
c. Errors with constant mean and standard deviation
d. Completely random errors
4. Which forecasting accuracy measures can be measured across different
industries
a. MSE and RMSE b. MAD and MAPE
c. Only MAPE d. RMSE and MAPE
5. Which forecasting model will be more appropriate if seasonality
is completely independent of the trend
a. Additive model
b. Multiplicative model
c. Both additive and multiplicative can be used
d. Neither additive nor multiplicative can be used
6. Which Time–Series component depicts long term upward and
downward movement of the data
a. Seasonality b. Cyclic c. Irregular d. Trend
7. Which component depends on local festivals and events in the
society
a. Trend b. Seasonality c. Cyclic d. Irregular
8. Which forecasting accuracy measure is the standard deviation of
errors
a. MAPE b. MAD c. RMSE d. MSE
9. Which of the below factors help in better forecasting ?
a. Large data set
b. Predicting for a shorter period
c. Forecasting for a stable business
d. All the above
10. What is modelling in statistics ?
a. Predicting output variable based on input variables
b. Predicting input variables
c. Predicting error terms
d. To check errors following a normal distribution or not
186
1.7 Let Us Sum Up : Introduction to
1. Time–series analysis plays an important role in descriptive analytics
as it shows timely movement (fluctuations) of important KPIs over
the period
2. Time–series forecasting is the backbone of predictive analytics as
all organizations need to forecast their important business KPIs (key
performance indicators)
3. Time–series data always have an x–axis as time while the y–axis
is the output variable, for which we want to show the movement
over the time
4. Time–series has four important components, trend, seasonality,
cyclical and irregular
5. Trends showcase overall upward or downward movement of data
throughout the time period
6. When time–series data shows repetitive downward or upward
movement / fluctuations from the trend then we say there is a
seasonal influence in our data. These repetitions are generally
shorter (less than a year) for example, daily, weekly, monthly,
quarterly etc.
7. A cyclical component is a movement over a longer period (generally
more than a year).
8. An irregular component consists of random movements in the time
series and follow a normal distribution with mean 0 and constant
variance.
9. One of the major factors of choosing the right forecasting technique
is calculating the accuracy measure, technique which shows the
least error generally become the choice of forecasting, but it is not
the only factor. Businesses consider several other factors also to
identify the right forecasting technique
10. There are four important forecasting accuracy measures used
frequently, these are Mean Absolute Error (MAE) / Mean Absolute
Deviation (MAD), Mean Absolute Percentage Error (MAPE), Mean
Square Error (MSE), Root Mean Square Error (RMSE)
11. The value of MAD and MSE depends on the magnitude of data
hence these cannot be used for benchmarking across the department
or businesses
12. MAPE and RMSE can be used as a target for benchmarking as
these are represented as ratio and percentage respectively
13. Forecasting is relatively easy for large data, stable businesses,
aggregated data and shorter period
187
Business Analytics 1.8 Answers for Check Your Progress :
1. Time 2. Difficult
3. MAPE 4. Mean absolute deviation
1. c 2. c 3. b 4. d
5. a 6. d 7. b 8. c
9. d 10. a
1.9 Glossary :
Time–Series : Time–series analysis represents the output variable
on the y–axis while the x–axis always remains time (generally time
interval remain constant).
Forecasting Technique : Predicting future value of output variable
based on historic data is known as a forecasting technique.
Component of Time–Series : Time series can be decomposed into
four important parts namely, trend, seasonality, cyclical and irregular.
These parts of time–series are known as components of time series.
Forecasting Error : The difference between the actual value and
forecasting value is known as a forecasting error.
Forecasting Accuracy Measure : The approach for calculating the
forecasting error in a time series is known as the forecasting accuracy
measure.
Mean Absolute Error (MAE) / Mean Absolute Deviation (MAD):
This is the average error in forecast without considering the sing of the
error.
Mean Absolute Percentage Error (MAPE) : Mean absolute
percentage error consider the average of absolute percentage error instead
of the average of just absolute value of the error terms.
Mean Square Error (MSE) : It is the average of the square of
the error term.
Root Mean Square Error (RMSE) : Square root of the mean
square error is known as root mean square error (RMSE).
1.10 Assignments :
1. Why forecasting is known as one of the most important functions
of business analytics.
2. Write down a few examples of the application of forecasting
techniques in personal life.
3. Write down various components of time series analysis and their
importance in brief.
4. Write important scenarios where additive and multiplicative
forecasting models can be used. Explain with a few examples.
188
1.11 Activities : Introduction to
One of the leading organizations in the retail store provided their
west India store's revenue (in crores rupees) of last 12 months. Calculate
the MAD, MSE, RMSE and MAPE for this data.
1 26.580 26.580
2 24.950 25.765
3 25.080 25.015
4 25.360 25.220
5 26.990 26.175
6 28.610 27.800
7 27.850 28.230
8 28.430 28.140
9 28.670 28.550
10 29.190 28.930
11 30.790 29.990
12 30.980 30.885
Ans. MAD – 0.383, MSE – 0.247, MAPE – 1.377, RMSE – 0.497
1.12 Case Study :

Pokymon construction company uses more than thousands of different
small and big machinery parts which range from ` 7 to ` 5 million.
They supply ready to mix and other intermediate materials require in
the construction business. Consistent supply of these parts are required
for their customers as in absence of these parts, they may face severe
losses on the other hand if the Pokemon industry keeps surpassing
inventory then it blocks their working capital as well as there is always
a risk of part become obsolete. Therefore, forecasting these parts is very
critical for the Pokymon industry.
Below is the actual quantity and forecasted of one of the small
parts:
1 321 398
2 345 382
3 543 375
4 424 408
5 390 411
6 321 407
7 422 390
8 362 396
9 516 389
10 462 415
11 462 424
12 459 432
189
1. Suggest the most appropriate forecasting accuracy measure
2. Plot the actual quantity with trend line
3. Which forecasting accuracy measures, Pokymon industry can be
compared with their competitors

• "Time series analysis and Control," Holden Day, Box and
Jenkins (1970)
• "How to get a better forecast, Harvard Business Review",
Parker G, Segura E (1971)
• "An introduction to Time series analysis and forecasting with
applications of SAS and SPSS", Management science journal,
Yaffee R, McGee M (2000)
190
MOVING AVERAGE AND
Unit SINGLE EXPONENTIAL
2 SMOOTHING TECHNIQUES
: UNIT STRUCTURE :
2.1 Introduction
2.2 Parts of Forecasting Techniques
2.3 Naïve Forecasting Models
2.4 Averaging Models
2.4.1 Simple Averages
2.4.2 Moving Averages
2.4.3 Weighted Moving Averages
2.5 Single Exponential Smoothing Forecasting Technique
2.6 Single Exponential Smoothing Forecasting Technique in
MS Excel
2.7 Let Us Sum Up
2.9 Glossary
2.10 Assignment
2.11 Activities
2.12 Case Study

• Understand smoothing techniques for time–series data
• Graphical representation of smoothing techniques
• Learning smoothing techniques like naïve forecasting models, average
models and exponential smoothing techniques
• Learning optimization of smoothing constants
• Establishing the relationship between smoothing constant and
accuracy measures
2.1 Introduction :
In this unit, we will study various forecasting techniques also known
as smoothing techniques as these techniques smooth out the irregular
fluctuation effects in the time–series data. We will analyse the graphical
representation of these smoothing techniques and will see the ways to
optimize smoothing constants. In the end, we will see the relationship
between smoothing constant and accuracy measures.
191
Business Analytics 2.2 Parts of Forecasting Techniques :
Forecasting techniques for stationary time series data (no significant
level of seasonal, trend, cyclical component) are also known as smoothing
techniques as smooth (minimize) the fluctuation due to irregular effects
in the time–series. Below are important forecasting techniques:
2.3 Naïve Forecasting Models :

These are simple methods where we consider the most recent
periods of data to represent the forecast or prediction of future outcomes.
Here we do not consider anything like trend or seasonality. Shorter
frequency data like hourly, daily or weekly naïve model generally works
better. The simplest form of a naïve model is as follows:
Ft = xt–1
Where,
xt–1 = The actual value at time t–1
Ft = Forecast value for time t
So, in simple words, if last month incoming call volume for a help
desk was 1150 then the naïve forecasting model would forecast that the
helpdesk will receive 1150 calls this week also.
This data is for the 12th month, now if we want to predict incoming
call volume for the 13th month then it will remain the same as for the
12th month, 1150 calls.
192
Month Incoming calls Moving Average and
January 1052 Single Exponential
February 979 Smoothing Techniques
March 936
April 927
May 1004
June 1125
July 1241
August 1029
September 1053
October 965
November 996
December 1150
January ???
Another way to use the naïve forecasting model is that we can

use the value of last year during the same week on month. For example,
if last year January sale was ` 37 lakh then this year January we will
consider sales of ` 37 lakh only (here we will not consider the sale
of last month, December).
Month Sales in Lakh
January 37
February 30
March 58
April 77
May 93
June 80
July 66
August 38
September 33
October 41
November 45
December 41
January
193
Business Analytics
2.4 Averaging Models :

Averaging models smooths the fluctuation by taking an average of
several data points in order to calculate the forecast for the next time
period. There are various ways to calculate the averaging models, below
three are the most important ones:
2.4.1 Simple Averages :
Most fundamental averaging model is a simple average model. Here
forecast for time t is calculated by averaging the value for a given number
of previous time periods. Below is the formula for the same:
x t −1 + x t − 2 + ..... + x t − n
Ft =
n
Below is the data for a year, given in the form of weekly sales
data.
Week Sales (in Thousands) Week Sales (in Thousands)
1 297 27 374
2 437 28 315
3 317 29 306
4 421 30 452
5 297 31 378
6 434 32 438
7 276 33 366
8 427 34 379
9 292 35 365
10 359 36 369
11 324 37 470
12 281 38 318
13 441 39 342
14 364 40 540
194
15 344 41 421 Moving Average and
16 369 42 387 Single Exponential
17 380 43 318 Smoothing Techniques
18 385 44 419
19 384 45 418
20 364 46 359
21 332 47 513
22 333 48 459
23 364 49 431
24 309 50 459
25 460 51 456
26 382 52 354
Here if want to calculate the forecast for the 53rd week then we
can calculate the average of all 52 weeks.
x 52 + x 51 + ...... + x1
=F53 = 380
52
In case we have data for several months, for example, 30 months
then instead of calculating the average of all 30 months we can also
calculate the average of 12 months. Generally, statisticians take decisions
by consulting with domain experts.
2.4.2 Moving Averages :
Moving average is another simple and interesting way to forecast
the time series data. It forecast the future value of a time series using
an average of the last 'N' observations. Below is the formula for calculating
a simple moving average:
1 t
Ft + 1 = ∑ Yk
N k = t +1− N
Here we take the average of the last N observations.

Worked Example : Below is the sales data of an organization for
the last 12 months. Use a 4–month moving average to calculate the
forecast for Jan 2021. Also calculate the MAD, MAPE, MSE and RMSE.
195
Business Analytics Year Sales (in Thousands)
1 297
2 437
3 317
4 421
5 297
6 434
7 276
8 427
9 292
10 359
11 324
12 281
Solution :
We can calculate the moving average by calculating the average
of 4 consecutive terms, first forecasting term will be
(297 + 437 + 317 + 421)
= 368 similarly second forecasting term will
4
(437 + 317 + 421 + 297)
be = 368 so on so for..
4
Absolute
Square
Absolute values of
Sales (in Moving value
Month Error error Errors Divided
Thousands) Average of
value by Actual
Error
values
Jan’20 297
Feb’20 437
Mar’20 317
Apr’20 421
May’20 297 368 -71 71 5041 0.239
Jun’20 434 368 66 66 4356 0.152
Jul’20 276 367 -91 91 8327 0.331
Aug’20 427 357 70 70 4900 0.164
Sep’20 292 359 -67 67 4422 0.228
Oct’20 359 357 2 2 3 0.005
Nov’20 324 339 -15 15 210 0.045
Dec’20 281 351 -70 70 4830 0.247
Jan’21 314
Average 56 4011 0.176
292 + 359 + 324 + 281

Forecasted value (Jan'21) = = 314
4
MAD – 56
MSE – 4011
RMSE – 22
MAPE – 0.176
196
Moving Average and
Single Exponential
Smoothing Techniques
Besides so many benefits of moving averages, there are few

disadvantages also as determining the optimal length of time periods for
moving average is always challenging also moving average does not
adjust the effects of trend, cycles or seasonality.
2.4.3 Weighted Moving Averages :
When we calculate moving average, we give equal weights to all
the terms but in the weighted moving average, we assign varying weights.
Generally, we give higher weights to most recent terms and relatively
lower weights to older data points (observations). The general formula
is as follows:
t
=Ft + 1 ∑ Wk × Yk
k = t +1− N
Where Wk is the weight given to the value of Y at time K (Yk)

t
and ∑ Wk = 1.
k = t +1 − N
The summation of all weights is equal to 1.

There are two ways to keep the summation of all weights to 1
either we keep all the weights in decimal in such a manner that summation
of all these weights in decimal must be equal to 1 or we can multiply
different data points by different weights and divided by the summation
of all weights. Let's see both with the help of the examples.
Scenario 1 – weights are in decimals and their summation is 1.
Ft+1 = .4(Ft) + .3(Ft–1) + .2(Ft–2) + .1(Ft–3)
Scenario 2 – weights are in integers and we divide the entire term

by summation of all weights
5 × Ft + 3 × Ft −1 + 2 × Ft − 2 + 1 × Ft − 3
Ft + 1 =
11
197
Business Analytics Worked Example : Calculate a 3–month weighted moving average
for a retail firm. Using weights of 3 for the recent month, 2 for the month
prior and 1 for the month before that. Calculate forecast for Jan 2021
also calculate MAD, MSE, RMSE and MAPE. 12 months data as below:
Month Sales (in Thousands)
Jan’20 496
Feb’20 339
Mar’20 427
Apr’20 421
May’20 398
Jun’20 434
Jul’20 276
Aug’20 427
Sep’20 397
Oct’20 359
Nov’20 424
Dec’20 361
Solution :
First moving average will be calculated on the first three values
(Jan, Feb and Mar data).
(3 × 496 + 2 × 337 + 1 × 427)

= 432
6
Absolute
Forecasting Absolute Square
Sales (in values of
Month (Moving Error error value of
Thousands) Errors Divided
Average) value Error
by Actual values
Jan’20 496
Feb’20 339
Mar’20 427
Apr’20 421 432 -11 11 125 0.027
May’20 398 382 16 16 256 0.040
Jun’20 434 420 14 14 191 0.032
Jul’20 276 416 -140 140 19460 0.505
Aug’20 427 390 37 37 1394 0.087
Sep’20 397 380 17 17 283 0.042
Oct’20 359 347 13 13 156 0.035
Nov’20 424 406 18 18 336 0.043
Dec’20 361 389 -28 28 775 0.077
Jan’21 381
Average 35 2856 0.108
3 × 359 + 2 × 424 + 1 × 361

Forecasted value (Jan'21) = = 381
6
MAD – 35
MSE – 2856
RMSE – 18
MAPE – 0.108
198
Moving Average and
Single Exponential
Smoothing Techniques
Deciding on the optimal length of moving average remain a challenge

for weighted moving average also besides that deciding weights is also
very difficult and remain subjective.
2.5 Single Exponential Smoothing Forecasting Technique :

Exponential smoothing techniques are quite like the weighted moving
average technique; the only difference is that here weights decrease their
importance exponentially. Here we have a smoothing constant , generally
a value between 0 and 1. The general form of single exponential smoothing
forecasting equation is:
Ft+1 = xt + (1– ) Ft
Where
xt = The actual value at time t
Ft = Forecast value for time t
Ft+1 = Forecast value for time t+1
= Exponential smoothing constant, a value between 0 and 1.
Here the value of is decided by the analyst. They may require
consulting with domain experts.
As we have a general equation for a single exponential smoothing
equation, we can write an equation for Ft as below:
Ft = xt–1 + (1– ) Ft–1
We can substitute the value for Ft in the above equation for Ft+1
Ft+1 = xt + (1– ) { xt–1 + (1– ) Ft–1}
Above equation can be simplified as follows:
Ft+1 = xt + (1– ) xt–1 + (1– )2 Ft–1
Similarly, if we can substitute the value of Ft–1 = xt–2 +
(1– ) Ft–2 in the above equation, it can be further expanded as below:
Ft+1 = xt + (1– ) xt–1 + (1– )2 { xt–2 + (1– ) Ft–2}
Ft+1 = xt + (1– ) xt–1 + (1– )2 xt–2 + (1– )3 Ft–2
199
Business Analytics Continuing the above process explains the weights on the older time
period and forecasts include (1– )n (exponential values). Because of
exponential values impacts higher emphasis on recent values and rapidly
decreasing emphasis on older time periods.
The basic idea about exponential smoothing techniques is that the
forecasted value of every next time period is a combination of current
actual and forecasted value. If the value of is chosen less than .5 then
there would be less weight on actual values and more weight on forecasted
value while if the value of is more than .5 then there would be more
weight on actual value and less weight on forecasted values.
is known as the smoothing constant and (1– ) is known as the
dumping factor.
Worked Example : 12 months sales figures are given in the table

below. Apply exponential smoothing to predict the values of each month.
Calculate forecast for January 2021 using = .3, .5 and .9
Month Sales (in Thousands)

Jan’20 496
Feb’20 339
Mar’20 427
Apr’20 421
May’20 398
Jun’20 434
Jul’20 276
Aug’20 427
Sep’20 397
Oct’20 359
Nov’20 424
Dec’20 361
200
Solution : Moving Average and
We can create a table with forecast and error values for each alpha Single Exponential
value. Generally, the forecast is not available for the very first data point Smoothing Techniques
hence we cannot calculate a forecasted value for the second time period.
Therefore, we use the actual value of the first time period as the
forecasted value for the second period and start the forecasting process.
Applying the exponential smoothing formula to calculate the forecasted
value for third and fourth data points (for = .3) :
F3 = .3(339) + .7(496) = 449
F4 = .3(427) + .7(449) = 442
F5 = .3(421) + .7(442) = 436
F6 = .3(398) + .7(436) = 425
Similarly, we can calculate forecasted values with other values
also. The below table has all calculated values.
201
Business Analytics
Conclusion :
The smaller value of ? (larger the damping factor), the smoothers
the entire time–series. On the other hand, the larger ? (smaller the damping
factor), forecasted values follow time–series data.
Benefits and Drawbacks of Single Smoothing Forecasting
Techniques:
Benefits :
The single exponential smoothing technique performs very well in
most scenarios. Its important benefits are as follows:
1. It is simple in calculation and works well with the data if there
is no strong trend or seasonal patterns in the time series
2. This technique uses all historic data unlike moving average forecasting
techniques
3. It assigns progressively decreasing weights which ensure more
importance to recent data
Drawbacks :
1. For every large data set, the forecast becomes less sensitive to
changes in the data
2. It always lags the trend and it is based on past observations. The
longer the time period n, the greater the last as it is slow to recognize
the shifts in the level of the data points
3. It doesn't work well if data has trend and seasonality influence
2.6 Single Exponential Smoothing Forecasting Technique in

MS Excel :
In the single exponential smoothing forecasting technique, we have
only one smoothing constant that's why we call it single exponential
202
smoothing. There are advanced versions of smoothing techniques like Moving Average and
the double exponential smoothing technique and triple exponential Single Exponential
smoothing technique. The single exponential smoothing technique can Smoothing Techniques
also be carried out in MS Excel. Below are the important steps:
Step – 1 : We can create a simple line graph to visualise the
fluctuations in the time series.
Step – 2 : Select the "Data Analysis" pack under the Data tab
Step – 3 : Select the option, "Exponential smoothing" and click

on the OK button
Step – 4 : Provide the reference of the input range, enter damping

factor (1– ), output cell reference and click on the OK button
203
Business Analytics
Step – 5 : Calculate the forecasted values for different values of

.
Notice that there is always a lag because forecasts depend on the

last observations. Lower the alpha value smoother the forecast.
1. Naïve forecasting methods works well with ___________ frequency
time period
2. In simple moving average, there are ___________ weights assigned
for all data period
3. Summation of all weights in weighted moving average is always
___________.
4. Value of smoothing constant in single exponential smoothing constant
is generally between ___________ and ___________.
204
5. If is the smoothing constant in exponential smoothing time series Moving Average and
forecasting technique, then 1– is known as ___________. Single Exponential
 Multiple Choice Questions : Smoothing Techniques
1. If smoothing constant in single smoothing is less than .5 then it
means, there will be more emphasis on
a. Actual value
b. Residual or error term
c. Forecasted value
d. The absolute value of the error term
2. If is small, then the value of the dumping factor will be
a. Small b. Large
c. It doesn't matter d. None of above
3. How smoothing techniques transform the time series
a. It doesn't change anything to the time series data
b. It makes time series more fluctuating
c. It reduces the peak part of the time series
d. It smooths the fluctuations of a time–series data
4. Single smoothing exponential technique has:
a. Single smoothing constant b. Single trend line
c. Both options are correct d. None of the above
5. The value of the smoothing constant always lies between
a. –1 to +1 b. 0 and the average value
c. 0 and 10 d. 0 and 1
6. One of the main differences between moving average and single
exponential smoothing technique is:
a. single exponential smoothing technique uses a constant value
always more than .5
b. single exponential smoothing technique uses all the historic data
c. Both above options are correct
7. Which forecasting technique uses progressively decreasing weights
to historic data
a. Naïve forecasting technique
b. The single exponential smoothing technique
c. Moving average
d. Weighted moving average
205
Business Analytics 8. What is the main motive behind the progressively decreasing weights
in the single exponential smoothing technique ?
a. Recent actual values will have higher weights
b. Recent actual values will have lower weights
c. Time series will smooth all the fluctuations
d. Forecasting accuracy measure will be better
9. One of the parameters to select the appropriate forecasting technique
is
a. Forecasting accuracy measure
b. Business domain
c. Knowledge of statistician about the forecasting technique
10. One of the drawbacks of moving average is
a. It is computationally very expensive
b. It provides the same importance to all data points
c. It is difficult to choose the optimal length of moving average
2.7 Let Us Sum Up :

1. Forecasting techniques for stationary time series data are also known
as smoothing techniques as smooth (minimize) the fluctuation due
to irregular effects in the time series.
2. Naïve forecasting techniques are simple methods where we consider
the most recent periods of data to represent the forecast or prediction
of future outcomes
3. In the simple average forecasting method forecast for time t is
calculated by averaging the value for a given number of previous
time periods.
4. Moving average is a simple and interesting way to forecast the time
series data. It forecast the future value of a time series using an
average of the last 'N' observations.
5. In the weighted moving average forecasting technique, we assign
varying weights. Generally, we give higher weights to most recent
terms and relatively lower weights to older data points (observations).
6. In the exponential smoothing technique, weights decrease their
importance exponentially. Here we have a smoothing constant ?,
generally a value between 0 and 1.
7. Single exponential smoothing technique uses all historic data unlike
moving average forecasting techniques, it assigns progressively
decreasing weights which ensure more importance to recent data
206
8. Single exponential smoothing technique, there is always a lag Moving Average and
because forecasts depend on the last observations. Lower the alpha Single Exponential
value smoother the forecast. Smoothing Techniques

1. Shorter 2. Same 3. One
4. 0 and 1 5. Damping factor
1. c 2. b 3. d 4. a
5. d 6. b 7. b 8. a
9. a 10. c
2.9 Glossary :
Forecasting Technique : Predicting future value of output variable
based on historic data is known as a forecasting technique
Moving Average Forecasting Techniques : Moving average is a
forecasting technique that forecasts the future value of a time series data
using average (or weighted average) of past 'N' observations
Single Exponential Smoothing Forecasting Technique : This
technique uses exponentially reducing weights with the help of a smoothing
constant alpha. Unlike moving average methods, it uses all historic data
Damping Factor : It is used to smooth the time series and
progressively assign weights to historic data. Its value is 1–
2.10 Assignment :
1. What is the basic difference between naïve, averaging models and
exponential smoothing forecasting techniques ? Write down the
important type of forecasting techniques under these categories.
2. "It is not necessary that more complex mathematical forecasting
technique will always give better forecasting accuracy". Explain this
statement with appropriate examples.
3. What is the basic difference between moving average and weighted
moving average forecasting techniques ?
2.11 Activities :
One of the leading organizations in the retail store provided their
west India store's revenue (in crores rupees) of last 12 months. Calculate
the forecast using a 3–month moving average and single exponential
smoothing technique (use = .3). Also calculate MAD, MSE, RMSE
and MAPE for this data.
207
Business Analytics Year Revenue
1 26.580
2 24.950
3 25.080
4 25.360
5 26.990
6 28.610
7 27.850
8 28.430
9 28.670
10 29.190
11 30.790
12 30.980
2.12 Case Study :

"Delicious bowl" is a famous restaurant chain in North–East India.
They receive their raw materials from various parts of Assam and Sikkim.
It is the area where rain is very high compared to the rest of India.
Sometimes heavy rain impacts their sales hence they always prepare their
dishes as per the weather forecasts and procurement also goes with the
same logic. Below is the table where their restaurant chain daily revenue
and rain in cms.
1 321 398
2 345 382
3 543 375
4 424 408
5 390 411
6 321 407
7 422 390
8 362 396
9 516 389
10 462 415
11 462 424
12 459 432
Questions :
1. Develop a forecasting model using 5–days and 7 days moving
average and single exponential smoothing technique for alpha value
.3 and .8
2. Visualise the forecasted values
3. Calculate the MAD, MSE, RMSE and MAPE also suggest which
measure is most appropriate for their restaurant chain
4. Which forecasted model will you recommend for their business
208
2.13 Further Readings : Moving Average and
Single Exponential
• "Time series analysis and Control," Holden Day, Box and Smoothing Techniques
Jenkins (1970)
Praker G, Segura E (1971)
• "Time Series Based Predictive Analytics Modelling: Using
MS Excel", Glyn Davis, Branko Pecar; 1st edition (2016)
209
Business Analytics
Unit REGRESSION METHODS FOR
3 FORECASTING
: UNIT STRUCTURE :
3.1 Introduction
3.2 Forecasting Techniques with a Trend
3.3 How to Draw Trendline and Regression Equation in Time–Series
Graph
3.4 Double Exponential Smoothing Constant Technique for
Forecasting
3.5 Let Us Sum Up
3.7 Glossary
3.8 Assignment
3.9 Activities
3.10 Case Study

• Application of regression theory in time series analysis
• Forecasting time series data in influence with seasonal variation
• Industry examples to understand the interpretation of regression–
based forecasting outputs
• Application of regression methods in business decisions
3.1 Introduction :
In this unit, we will study the forecasting methods when we can
have extra information about time series. For example, besides sales data,
we also have information about marketing expenses, competition
information, consumer's demographic information etc. Here we can also
include information about seasonality variation. In the last, we will study
a few worked examples to understand how regression–based forecasting
methods help in robust decision making in various industries.
3.2 Forecasting Techniques with a Trend :

Forecasting techniques discussed in unit 2 have an assumption that
trend is not included in the time–series data. In this unit, we will discuss
the possible ways to determine a trend in the data. One of the most
effective ways to determine trend components in time–series data is
210
regression analysis. Here we consider the output variable as the metric Regression Methods for
we are determining on a time scale or the metric we want to forecast Forecasting
while the x–axis represents time.
Regression is a more powerful forecasting technique when the time–
series has values of various independent variables also besides the dependent
variable Yt. The forecasting techniques we discussed in the last units have
only output variables that's why their application in the real world is
comparatively less as most of the time we have various independent
variables which influence output variables. Therefore, forecasting only
based on output variables is less effective. For example, if we want to
predict the sales price alone then it may not be so effective but if we
also include information like marketing cost, whether a promotion was
going on, nearby festival etc then our forecasting will be more accurate.
The general form of a regression equation is as follows:
Ft = 0 + 1X1t + 2X2t + ..... + nX3t + t
Where
Ft = The forecasted value at time t
X1t, X2t etc are the predictor variables measures at time t
t = Error or Irregular component at time t
Worked Example : Below are crude oil per barrel prices in the
Indian market (IN USD). See if a trend is visible, can we predict the
price for 17th August 2021
Day Crude Oil Price per Barrel in India

1 $73.62
2 $73.95
3 $71.26
4 $70.56
5 $68.15
6 $69.09
7 $68.28
8 $66.48
9 $68.29
10 $69.25
11 $69.09
12 $68.44
13 $67.29
Solution : We can use regression analysis under MS Excel "Data

Analysis" tab
211
Business Analytics SUMMARY OUTPUT
Regression Statistics
Multiple R 0.76187804
R Square 0.580458147
Adjusted R
0.542317979
Square
Observations 13
ANOVA
df SS MS F Sig F
Regression 1 35.58 35.58 15.22 0.00
Residual 11 25.72 2.34
Total 12 61.30
Standard P- Lower Upper Lower Upper

Coefficients t Stat
Error value 95% 95% 95.0% 95.0%
Intercept 72.6142 0.8996 80.720 0.000 70.634 74.594 70.63 74.59
Day -0.4421 0.1133 -3.9012 0.003 -0.6916 -0.1927 -0.691 -0.192
Ft = 72.6142 – 0.4421 Xt
Here, Ft is forecasted per barrel price of crude oil and Xt is the
time period
So, if we want to forecast the sales for the next time period then
the value will be:
F14 = 72.6142 – 0.4421 X14 = 72.6142 – 0.4421 14 = $66.42
Here can interpret it like every unit increase in the time period
Xt, per barrel price of crude oil will decrease by 0.442 USD. Y–intercept
72.6142 indicates that the time period before the very first time period
(at Day 0) per barrel price of crude oil was $ 72.61.
The P–value for the independent variable "Day" is 0.0025 represents
a significant relationship between input and output variables. Below is
the visual representation of the trend.
212
3.3 How to Draw Trendline and Regression Equation in Time–Series Regression Methods for
Graph : Forecasting
We can add a trend line to our time–series graph to visualize a

trend. Below are the important steps to create a trend line and publish
a regression equation, R square value etc.
1. Select the dependent and independent variables and create a line
graph, time variable must be on X–axis and the output variable
on Y–axis
2. Select the graph, right hand side three features will appear. Select
the + icon on the right–hand top side of the graph
3. On clicking on the + icon, one menu bar will be open
4. Select the second last option, "Trendline", which will create a trend
line as per the data. There is various type of trendline like linear
(straight line), exponential, Linear forecast etc. In this course, we
have included only Linear, but others are also quite intuitive, please
refer to the Microsoft excel help website to learn about these
different trendlines. By default, Excel will create a Linear trendline
only (the very first option).
213
Business Analytics
5. Select the trendline and right–click, to choose the "format trendline

option"
Here we can select a different types of trendlines and select the

options provided below, "Display Equation on chart" or "Display
R–squared value on chart".
6. We can also extend the trendline either forward or backwards to
forecast the data points
214
Regression Methods for
Forecasting
3.4 Double Exponential Smoothing Constant Technique for

Forecasting :
One of the drawbacks of single exponential smoothing is that the
model does not do well if a trend is visible in the data. This can be
improved by introducing an additional equation for capturing the trend
in the time–series data. Double exponential smoothing uses two equations
to forecast the future values of the time series, one for forecasting the
level (short term average value) and another for capturing the trend. Below
are the two equations:
Level (or Intercept) equation (Lt) :
Lt = Yt + (1– ) Ft
Equation for the trend (Tt) :
Tt = (Lt – Lt–1) + (1– ) Tt–1
Where
and are smoothing constant for level and trend, respectively
and 0 < < 1 and 0 < < 1.
The forecast at time t + 1 is given by:
Ft+1 = Lt + Tt
Ft+n = Lt + (n)Tt
This technique does not work very well if the seasonality component
is also available besides the trend. In that situation, we use the triple
exponential smoothing technique which is not in the scope of this text.
Worked Example : Below is the sales data of ABC Limited for
42 months. Use double exponential smoothing technique to forecast sales
for the next 6 months.
215
Business Analytics Month Sales Month Sales
Apr-17 9.75 Jan-19 11.46
May-17 10.09 Feb-19 11.06
Jun-17 10.27 Mar-19 10.76
Jul-17 8.98 Apr-19 10.32
Aug-17 8.79 May-19 9.34
Sep-17 8.23 Jun-19 8.39
Oct-17 8.24 Jul-19 7.01
Nov-17 8.97 Aug-19 6.71
Dec-17 9.03 Sep-19 5.93
Jan-18 9.31 Oct-19 5.07
Feb-18 10.25 Nov-19 5.21
Mar-18 9.5 Dec-19 6.46
Apr-18 10.05 Jan-20 6.35
May-18 9.73 Feb-20 5.36
Jun-18 9.48 Mar-20 5.25
Jul-18 9.38 Apr-20 4.94
Aug-18 9.44 May-20 5.13
Sep-18 10.02 Jun-20 5.63
Oct-18 10.15 Jul-20 5.35
Nov-18 10.55 Aug-20 5.74
Dec-18 11.13 Sep-20 5.26
Solution :
Step 1 : Let's assume value for smoothing constants, = 0.7 and
= 0.6. Later we will see how we can optimize these values with the
help of the "Solver" functionality of MS Excel.
Step 2 : We must initialize the very first values of Level, trend
and forecast. One of the popular ways to initialize the level value is
considering the same actual value for that period While the trend is the
difference between the current and last actual value.
Month Sales Level Trend Forecast Error

Apr-17 9.75
May-17 10.09 10.09 0.34
Jun-17 10.27 10.43 -0.16
Jul-17 8.98
Aug-17 8.79
Sep-17 8.23
Oct-17 8.24
216
The initial value of forecast value is the summation of Level and Regression Methods for
trend while Error is the difference between the actual value and forecasted Forecasting
value.
Step 3 : Now we must calculate the level value for the row "Jun–
17" Yt + (1– ) Ft.
LevelJun–17 = 0.7 10.27 + (1 – 0.7) (10.09 + 0.34) = 10.318
Step 4 : Now we must calculate the trend value for the row "Jun–
17" (Lt – Lt–1) + (1– ) Tt–1
TrendJun–17 = 0.6 (10.318 – 10.090) + (1 – 0.6) (0.340) =
0.273
Step 5 : Copy the same formula for entire column of "Level" and
"Trend" (up to the cells actual values are available). Also copy the formula
for "Forecast" and "Error" column
Month Sales Level Trend Forecast Error
Apr-17 9.75
May-17 10.09 10.090 0.340
Jun-17 10.27 10.318 0.273 10.43 -0.16
Jul-17 8.98 9.463 -0.404 10.59 -1.61
Aug-17 8.79 8.871 -0.517 9.06 -0.27
Sep-17 8.23 8.267 -0.569 8.35 -0.12
Oct-17 8.24 8.077 -0.341 7.70 0.54
Nov-17 8.97 8.600 0.177 7.74 1.23
Dec-17 9.03 8.954 0.283 8.78 0.25
Jan-18 9.31 9.288 0.314 9.24 0.07
Feb-18 10.25 10.056 0.586 9.60 0.65
Mar-18 9.5 9.842 0.107 10.64 -1.14
Apr-18 10.05 10.020 0.149 9.95 0.10
May-18 9.73 9.862 -0.035 10.17 -0.44
Jun-18 9.48 9.584 -0.181 9.83 -0.35
Jul-18 9.38 9.387 -0.190 9.40 -0.02
Aug-18 9.44 9.367 -0.088 9.20 0.24
Sep-18 10.02 9.798 0.223 9.28 0.74
Oct-18 10.15 10.111 0.277 10.02 0.13
Nov-18 10.55 10.502 0.345 10.39 0.16
Dec-18 11.13 11.045 0.464 10.85 0.28
Jan-19 11.46 11.475 0.443 11.51 -0.05
Feb-19 11.06 11.317 0.083 11.92 -0.86
Mar-19 10.76 10.952 -0.186 11.40 -0.64
Apr-19 10.32 10.454 -0.373 10.77 -0.45
May-19 9.34 9.562 -0.684 10.08 -0.74
217
Business Analytics Jun-19 8.39 8.536 -0.889 8.88 -0.49
Jul-19 7.01 7.201 -1.157 7.65 -0.64
Aug-19 6.71 6.510 -0.877 6.04 0.67
Sep-19 5.93 5.841 -0.753 5.63 0.30
Oct-19 5.07 5.076 -0.760 5.09 -0.02
Nov-19 5.21 4.942 -0.384 4.32 0.89
Dec-19 6.46 5.889 0.415 4.56 1.90
Jan-20 6.35 6.336 0.434 6.30 0.05
Feb-20 5.36 5.783 -0.158 6.77 -1.41
Mar-20 5.25 5.362 -0.316 5.62 -0.37
Apr-20 4.94 4.972 -0.361 5.05 -0.11
May-20 5.13 4.974 -0.143 4.61 0.52
Jun-20 5.63 5.391 0.193 4.83 0.80
Jul-20 5.35 5.420 0.095 5.58 -0.23
Aug-20 5.74 5.672 0.189 5.51 0.23
Sep-20 5.26 5.441 -0.063 5.86 -0.60
Step 6 : Enter the numbers of the forecast horizon. It is just the

number under the "Trend" column that starts with 1. Here we use the
same last value (fixed excel cell) of Level and Trend for all next
forecasted values, formula for the forecast is Ft+n = Lt + (n)Tt. Here
n is the forecast horizon.
Step 7 : Calculate the forecasted accuracy like MAPE and RMSE.

MAPE value for the above data is 6.86.
Step 8 : Optimize the error value by adjusting different values of
and . It can be done using MS Excel functionality, solver. It is
available under the "Data" tab.
218
Regression Methods for
Forecasting
Here we want to reduce the MAPE value (we selected as a minimum)

by changing the values of and . We also put the constraint that these
values can't be less than 0 and more than 1. Click the "Solve" button
and excel using a linear programming approach to find the minimum value
MAPE.
Select "Keep solver solution" and click on the OK button to see

the improved MAPE value and changed values of and smoothing
constants.
ABS(Error)/Actual
Trend Forecast Error α β
Value
0.906461 0.360152
0.340
0.288 10.43 -0.16 0.016
-0.069 5.82 -0.56 0.106
1 5.24 -5.24
2 5.17 -5.17
3 5.11 -5.11
4 5.04 -5.04
5 4.97 -4.97
6 4.90 -4.90
MAPE 6.332
Here we can see MAPE has been reduced to 6.332 by adjusting
= .906461 and = .360152.
219
Business Analytics Check Your Progress :
1. Regression based forecasting techniques are advisable if clear
___________ is visible in the time series data
2. In regression equation, Ft = 0 + 1X1t; 0 is known as ___________
while 1 is known as ___________.
3. If a significant value under the ANOVA table in regression output
is 0 then ___________ shows significant relationship between input
and output variables.
4. Double smoothing forecasting techniques has two smoothing
constants, one for ___________ and other for ___________.
5. Smoothing constants have values between ___________ and
___________.
1. Double smoothing forecasting techniques are applicable if the below
effect is NOT visible:
a. Trend b. Irregular or random
c. Seasonality d. Level
2. Which of the below statements are true ?
I. Both smoothing constants must have the same value
II. Both smoothing constants can have a value between 0 and .5
a. Only statement 1 is correct b. Only statement 2 is correct
c. Both statements are correct d. Both statements are incorrect
3. Complete the Level (or Intercept) equation (Lt) : Lt = Yt +
_________ from below options:
a. (1– ) Ft b. Ft
c. (1– ) Ft d. Ft
4. If the R–Square value is .72 in an MS Excel output of regression
analysis, then it means:
a. Input variables justify 72% variation in an output variable
b. The researcher needs more input variables in the study
c. Regression output cannot be trusted
d. The researcher can use a regression model at a 72% confidence
level only
5. Complete the trend equation (Tt) : Tt = (Lt – Lt–1) + _________
from below options:
a. (1– ) Ft b. Ft
c. (1– ) Ft d. Ft
220
6. As per unit content, MAPE/ RMSE value can be minimized by using Regression Methods for
the below functionality in MS Excel : Forecasting
a. Pivot table b. Solver
c. Both the functionality d. None of the above
7. Solver optimises MAPE/ RMSE to :
a. Maximum b. More than 1
c. Minimum d. Less than 0
3.5 Let Us Sum Up :

1. Regression is a more powerful forecasting technique when the time–
series has values of various independent variables also besides the
dependent variable Yt.
2. Sometimes simple forecasting techniques results better than complex
techniques hence various forecasting models must be tried before
selecting the final model.
3. The effectiveness of regression is generally more than other forecasting
techniques as it includes both input and output variables.
4. MS Excel help us to visualize the trend line clearly, it also shows
the regression equation and R–Square value in the graph itself.
5. The residual analysis must be performed before selecting the
regression model.
6. Double exponential smoothing uses two equations to forecast the
future values of the time series, one for forecasting the level (short
term average value) and another for capturing the trend.
7. The double exponential smoothing technique uses two smoothing
constants, (for level/ intercept) and (for trend).
8. The final forecast comes as a summation of trend and level
components both.
9. By using MS Excel Solver functionality, the forecasting error can
be reduced.
10. Solver adjusts the values of smoothing constants in order to optimize
the forecasting error.

1. Trend 2. Intercept, slope 3. Overall model
4. Level, trend 5. 0, 1
1. c 2. d 3. a 4. a
5. c 6. b 7. c
221
Business Analytics 3.7 Glossary :
Regression Model for Forecasting : When information about input
and output variables are available then regression–based forecasting
techniques work better than other techniques which are based on output
only.
Double Exponential Smoothing Forecasting Techniques : When
the clear trend is visible in the time–series data then double exponential
smoothing techniques generally works better, here we use two equations,
one for level (intercept) and the other for trend. Final forecasts come
as a summation of both equation.
Optimization of Smoothing Constants : We use linear programming
techniques to adjust the values of smoothing constants and to
minimize the forecasting error. In MS–Excel there is an optimizing
technique in form of Solver.
Decomposition of Time–Series : Separating the impact of trend,
seasonality and irregular components from a time series is known as
decomposition of a time series.
3.8 Assignment :
1. What is the basic difference between regression-based and other
forecasting techniques, why most of the time regression-based
forecasting techniques works better in terms of accuracy of forecasts.
2. Write down the important steps to draw a trend line in a forecasting
graph in MS Excel.
3. What is the basic difference between single and exponential
forecasting techniques ? Write down the important scenarios where
these techniques work better than other.
3.9 Activities :
Below are the sales (in ` Crore) data of Contesa Ice cream. Use
double exponential smoothing technique to forecast sales for the next 5
days.
222
Date Sales Date Sales Regression Methods for
January 1, 2012 8.75 January 22, 2012 10.26 Forecasting
January 2, 2012 9.09 January 23, 2012 9.86
January 3, 2012 9.27 January 24, 2012 9.56
January 4, 2012 7.98 January 25, 2012 9.12
January 5, 2012 7.79 January 26, 2012 8.14
January 6, 2012 7.23 January 27, 2012 7.19
January 7, 2012 7.24 January 28, 2012 5.81
January 8, 2012 7.97 January 29, 2012 5.51
January 9, 2012 8.03 January 30, 2012 4.73
January 10, 2012 8.31 January 31, 2012 3.87
January 11, 2012 9.25 February 1, 2012 4.01
3.10 Case Study :

OTT based entertainment platforms become quite popular during
the corona virus–based pandemic as most of the cinema halls and theatres
were closed. Most of the big entertainment companies launched their OTT
based platforms also big production houses and artists signed contracts
with OTT platforms as these were gaining popularity during the pandemic.
Below are TRP ratings of a famous web series.
Episode TRP Episode TRP

1 7.98 11 8.5
2 9.8 12 9.03
3 9.54 13 9.82
4 7.28 14 9.77
5 9.62 15 10.77
6 9.8 16 9.46
7 7.9 17 9.57
8 8.26 18 9.81
9 8.17 19 9.92
10 8.36 20 9.98
223
1. Is there any clear trend visible in the data, use MS Excel to visualize
the trend also show R–Square and regression equation in the same
graph ?
2. Which technique do you think will be more appropriate between
double exponential smoothing technique or regression? Justify your
selection
3. Use the initial value of both and constants as .6 but optimize
them using solver functionality of MS Excel

Jenkins (1970)
224
AUTO–REGRESSION (AR) AND
Unit MOVING AVERAGE (MA)
4 FORECASTING MODELS
: UNIT STRUCTURE :
4.1 Introduction
4.2 Introduction to Autocorrelation
4.2.1 Reasons for Autocorrelation
4.2.2 Impact of Autocorrelation on a Regression Model
4.2.3 Ways to Detect Autocorrelation : Durbin Watson Test
4.3 Autoregression : Remedy to Resolve Autocorrelation
4.4 Moving Average Model MA(q)
4.5 Let Us Sum Up
4.7 Glossary
4.8 Assignment
4.9 Activities
4.10 Case Study

• Learn the concept of autocorrelation and how it impacts the time
series analysis
• Understanding of Auto–regressive (AR) forecasting models
• Identification of the order of autoregression through MS Excel
• Understanding and application of moving average (MA) forecasting
models
4.1 Introduction :
In this unit, we will study the concepts of autocorrelation and how
autore–gression models help to overcome the problem of autocorrelation.
We will see the difference between moving average based forecasting
techniques studied in the last units and the moving average process of
time–series modelling.
4.2 Introduction to Autocorrelation :

Business data generally collected in chronological order means older
date's data comes first and then we have more recent data. This data
cannot be completely independent of each other. Productivity of today
must be inclined with yesterday's production similarly gold price, crude
oil price or share prices etc. always depends on older values. Now this
older impact may be of one day, two days or even much older. In other
words, these metrics correlate with themselves which is a breach of the
225
Business Analytics regression assumption that there should not be any correlation among
independent variables. When an independent variable shows correlation
with itself then we called it autocorrelation. In the case of autocorrelation
error terms of the regression model are correlated. As error terms in the
case of autocorrelation are not random, this creates several problems in
regression analysis. In the below example, share price shows the
autocorrelation because error terms are no longer random. If we know
the error term at a time t, we can predict the error term for next time
t+1.
Share Residuals
Price
Time
Time
4.2.1 Reasons for Autocorrelation :

• Missing of the Important Variable(s) from the Regression Model:
The important variable(s) is not part of the regression model then
the impact of this missing variable will be seen in error terms. For
example, in the above example share price also depends on the
market so including that variable may reduce the impact of
autocorrelation from the study
So, the inclusion of the market index (S&P500) variable must

resolve the autocorrelation problem, now error terms are relatively
more random
• The Incorrect Functional form of a Regression Model : In case,
the scatter plot is not showing a linear relationship while we are
writing an equation of linear regression equation then it may cause
autocorrelation. For example, the maximum weight a person can
lift shows a curvilinear model, young age we can lift less weight
while maximum at the middle ages and again less weight during
old age. But if we try to build a linear model then there may be
an effect of autocorrelation.
226
Auto–Regression (AR)
and Moving Average
(MA) Forecasting Models
A here linear equation is not appropriate to fit the data points, it

seems that a quadratic model may fit the data better.
A quadratic regression model is fitting data well and removed the

effect of autocorrelation from the model.
Autocorrelation can be a problem of non–time–series data also, few
statisticians have this myth that autocorrelation problem occurs only in
the case of time–series data.
4.2.2 Impact of Autocorrelation on a Regression Model :
Below are the important consequences of autocorrelation:
• The estimation of regression coefficients is not correct as they do
not show minimum variance property
• The variance of the error term is greatly underestimated so it may
look like a good regression model but, it is because of the
autocorrelation problem and generally, these models work very
badly for new (future) data
• The standard deviation of the regression coefficient may be calculated
very less than actual
• Autocorrelations make the overall regression model unreliable (unlike
multicollinearity, which does not impact the R2 value of the overall
model) as confidence interval and hypothesis tests using F and t
distribution have also become unreliable
Therefore, we need to correct the autocorrelation before evaluating
the model performance.
227
Business Analytics 4.2.3 Ways to Detect Autocorrelation : Durbin Watson Test :
There are two important ways to detect autocorrelation, first one
is the "Durbin Watson test" which detects only first–order autocorrelation
(only consecutive error terms are correlated) and the second one is the
"Breusch–Godfrey test", which can detect autocorrelation of any order
(error terms are corrected with any past error order of error terms). In
this text, we will consider only the Durbin–Watson test as the second
one is more complex mathematically and rarely used in day to day
business analytics.
Durbin Watson Test : This test was developed by two statisticians
in 1950. Here we capture the error terms in a regression model. We try
to see the successive error terms if these are related.
So, if a positive error term is correlated with the very next positive
error term then we call it positive autocorrelation. Similarly, if a positive
error term is related to the next negative error term then it is known
as negative autocorrelation. There may be the cases when error term at
the kth position is correlated with the K+2nd error term or any other error
of higher–order but the Durbin–Watson test is only effective for the first
order of autocorrelation.
(e2 − e1 ) 2 + (e3 − e 2 ) 2 + ... + (en − en −1 ) 2

dw statistics =
e12 + e2 2 + ... e n 2
in the simple and generic form above equation can be rewritten as:
n
∑ (e t − e t −1)2
t =2
dw statistics =
n
∑ (et )2
t =1
We can observe that in numerator one term is less than the

denominator.
Durbin–Watson statistics give us a value between 2 and 4. If the
value is more towards 2 then it is positively autocorrelated and if it is
more towards 4 then it is more negatively autocorrelated error terms.
228
and Moving Average
These cut–off values of the Durbin–Watson test depend on two

factors:
• Number of data points (n)
• Number of input variables in the regression model (k)
Worked Example : A leading laptop manufacturing company
manufacture laptops and special circuits which require in few models.
Check with the help of the Durbin–Watson test, if the data shows the
autocorrelation at = 0.05.
Number of Number of
No of No of
Month Special Ics Month Special Ics
Laptops Laptops
Manufactured Manufactured
1 17194 18005 13 14342 16597
2 17144 17769 14 13694 17123
3 17298 16241 15 13324 17783
4 17376 14585 16 13120 17714
5 17758 15945 17 12930 18169
6 17942 14967 18 12904 18276
7 17360 14586 19 12504 18161
8 16698 15239 20 11762 18022
9 16280 15899 21 11644 18158
10 15226 16141 22 11602 18561
11 14710 16486 23 11492 18227
12 14834 16450 24 11362 18251
Solution : Using MS Excel we can write the predicted regression

equation. Below is the coefficient section of Excel regression output.
Regression Coefficients
Intercept 23350.34888
No of Laptops –0.436671535
229
Business Analytics Number of Special Ics Manufactured = 23350.34888 +
(–0.436671535 * No. of Laptops)
For the first month, an error term can be calculated as:
Actual – Predicted (Regressed) = 18005 – 15842 = 2163
Difference of the first pair of consecutive error terms:
eSecond month – eFirst month = –258
Similarly, we can calculate the error terms for the rest of the data
points
Month Y Predicted et e2t et - e t-1 (e t - et-1 ) 2
1 15842 2163 4677624 - -
2 15864 1905 3628827 -258 66478
3 15797 444 197309 -1461 2133798
4 15763 -1178 1387082 -1622 2630688
5 15596 349 121846 1527 2331144
6 15516 -549 300949 -898 805780
7 15770 -1184 1401219 -635 403406
8 16059 -820 672084 364 132440
9 16241 -342 117194 477 227979
10 16702 -561 314259 -218 47634
11 16927 -441 194402 120 14323
12 16873 -423 178729 18 329
13 17088 -491 240694 -68 4603
14 17371 -248 61290 243 59067
15 17532 251 62932 498 248434
16 17621 93 8608 -158 24990
17 17704 465 216052 372 138408
18 17716 560 314116 96 9148
19 17890 271 73328 -290 83908
20 18214 -192 36948 -463 214379
21 18266 -108 11609 84 7136
22 18284 277 76682 385 147963
23 18332 -105 11050 -382 145950
24 18389 -138 19013 -33 1074
Sum 14323847 9879058
n
− e t −1)2
∑ (e t
t =2 9879058
dw=
statistics = = 0.6897
n 14323847
∑ (et )2
t =1
230
and Moving Average
Here, we have k = 1 as we have only one input variable, n = 24

(number of data points/observations) and = 0.05 (given). We can see
critical values from the Durbin–Watson table for = 0.05, k = 1 and
n = 24. The table gives us two critical values, one for the upper critical
value (dU) and the other is the lower critical value (dL). For k = 1 and
n = 24, dL = 1.273 and dU = 1.446. For right–hand side critical values,
we can subtract these values from 4, we can get dL = 2.554 and
dU = 2.727.
In the case of DW statistics value would have been in between
dL and dU then we could not conclude anything. Below is the sample
DW statistics table for = .05 similar tables are available on the internet
for different values of .
231
Business Analytics
4.3 Autoregression : Remedy to Resolve Autocorrelation :

Autoregression resolves the autocorrelation problem by creating a
multiple regression model of a variable on itself measured at different
time periods (different time periods are known as lags). Here independent
variables are the input variable itself and the same variable at lag1, lag2,
leg3 etc. up to lag n. The optimum value of lags is determined by
examining the regression output. A general equation for an autoregression
model can be written as :
Y = 0 + 1 Yt–1 + 2 Yt–2 + … + n Yn
232
Worked Example : For below "Number of special ICs" data, try Auto–Regression (AR)
to fit the third–order autoregression model if require fitting second or and Moving Average
first–order regression model (Given = .05). Also, forecast the value (MA) Forecasting Models
th
for the 25 month.
Number of Number of
Month Special Ics Month Special Ics
Manufactured Manufactured
1 13541 13 12133
2 13305 14 12659
3 11777 15 13319
4 10121 16 13250
5 11481 17 13705
6 10503 18 13812
7 10122 19 13697
8 10775 20 13558
9 11435 21 13694
10 11677 22 14097
11 12022 23 13763
12 11986 24 13787
Solution : We can create data for lagged 1, lagged 2 and lagged

3 (as below)
Two
Number of One Period Three Period
Period
Month Special Ics Lagged Lagged
Lagged
Manufactured Yt-1 (X1) Yt-3 (X3)
Yt-2 (X2)
1 13541 - - -
2 13305 13541 - -
3 11777 13305 13541 -
4 10121 11777 13305 13541
5 11481 10121 11777 13305
6 10503 11481 10121 11777
7 10122 10503 11481 10121
8 10775 10122 10503 11481
9 11435 10775 10122 10503
10 11677 11435 10775 10122
11 12022 11677 11435 10775
12 11986 12022 11677 11435
13 12133 11986 12022 11677
14 12659 12133 11986 12022
15 13319 12659 12133 11986
233
Business Analytics 16 13250 13319 12659 12133
17 13705 13250 13319 12659
18 13812 13705 13250 13319
19 13697 13812 13705 13250
20 13558 13697 13812 13705
21 13694 13558 13697 13812
22 14097 13694 13558 13697
23 13763 14097 13694 13558
24 13787 13763 14097 13694
13787 13763 14097
13787 13763
13787
A regression model can be developed only for the rows where data
is available for input variables X1, X2 and X3. Hence, we can remove
the first three rows and the last three rows. We will consider the first
column as the output variable and the remaining three columns for input
variables. Below is the regression output:
SUMMARY
OUTPUT
Regression
Statistics
Multiple R 0.8896
R Square 0.7914
Adjusted R Square 0.7545
Observations 21.0000
ANOVA
df SS MS F Significance F
Regression 3 28432295 9477432 21 0
Residual 17 7495900 440935
Total 20 35928195
Coefficients Standard Error t Stat P-value

Intercept 1254.631 1565.633 0.801 0.434
X Variable 1 0.957 0.215 4.452 0.000
X Variable 2 -0.086 0.298 -0.290 0.775
X Variable 3 0.036 0.221 0.161 0.874
Here the higher value of R2 and adjusted R2 shows decent

predictability of the regression model. Here we can see that only the first
X variable (One period lagged) is showing a significant relationship
between output and input variable.
234
It is always advisable that lagged period should be chosen by Auto–Regression (AR)
consulting with domain/ business experts and visualizing past data. If and Moving Average
data has been captured on daily basis then 7 days lagged may make sense (MA) Forecasting Models
as there may be some cyclical effect during different weekdays and
weekends. Similarly, if data is captured quarterly then 4 quarter lagged
data may make more sense as 4 quarters complete one yearly cycle.
4.4 Moving Average Model MA(q) :

Moving average (MA) processes are regression models in which
the past residuals are used for forecasting future values of the time–series
data. The moving average process is different from the moving average
technique discussed in earlier units except that the regression model of
the MA process can be considered as a weighted moving average of
past residuals. Moving average process of lag 1, MA (1) is as follows:
Y t+1 = + et–1 + t
MA (1) process uses the previous residual, t to forecast the next
value of the time series. The reasoning behind the MA process is that
the error at the current period, t, and the error at the next period, t+1,
drive the next value of the time series Yt+1, A moving average process
with q lags, MA(q) process, is given by
Yt+1 = + 1et + 2et–1 + …. + qet–q+1 + t+1
1. AR (1) models depends only on past values of the time series,
these input variables of past values are known as ___________
2. Yt = 0 + 1Yt–1 + et is an example ___________ autoregression
model.
3. Moving average Forecast model based on past ___________
4. Range of Durbin–Watson statistics is always between ___________
and ___________.
5. Durbin–Watson is useful to identify only first order ___________
in the time–series.
1. The problem created by autocorrelation:
a. One variable correlates with itself which denies regression
assumptions
b. Error terms are not random
c. Both of above
235
Business Analytics 2. What are important reasons for autocorrelation
a. Missing of the important variable(s) from the regression model
b. The incorrect functional form of a regression model
d. Both options a and b are correct
3. Due to autocorrelation, we cannot estimate the regression coefficients
because of:
a. Minimum variance property
b. Non–randomness of the error term
c. Seasonality component in the time series
4. AR models can be used when:
a. There is no autocorrelation effect in the data
b. Data is stationary
c. When there is more than one variable
5. Auto–regressive models are regression models where:
a. Yt is the output variable
b. Yt–1, Yt–2 etc are independent variable
c. The error term is not an independent variable
d. All the above are correct
6. For not showing autocorrelation, Durbin–Watson statistics should
be around:
a. Statistics value should be around 2
b. Statistics value should be around 0
c. Statistics value should be around 4
d. Statistics value should be less than 1
7. In the Durbin–Watson test if the value of the statistics is in between
dU and dL
a. There is no autocorrelation
b. Result is inconclusive
c. There is a strong autocorrelation
d. There is weak autocorrelation
236
8. In the Durbin–Watson test, which of the below statements are true: Auto–Regression (AR)
a. If the test statistics value is around zero, there is a positive and Moving Average
autocorrelation (MA) Forecasting Models
b. If the test statistics value is around four, there is a negative
autocorrelation
c. If the test statistics value is around two, there is no autocorrelation
d. All the above are true
9. For moving average of lag 1, MA (1) model
a. Past residual is the independent variable
b. Error terms correlate with one error term before
c. Both a and b statements are wrong
d. Both a and b options are correct
10. If data is showing an autoregressive model of third–order then
which test can be used to validate it
a. Durbin–Watson test
b. Breusch–Godfrey test
c. Both above test
d. None of the tests given in options a and b
4.5 Let Us Sum Up :

1. Most of the economic phenomenon shows autocorrelation where
data of the current day depend on historic data.
2. If today's data depends on yesterday's data, then it is known as
first–order auto–correlation but if today's data depends on the day
before yesterday's data then it is second–order autocorrelation
3. If autocorrelation is there, then we cannot rely on the regression
coefficient
4. The variance of the error term is greatly underestimated so it may
look like a good regres–sion model but, it is because of the
autocorrelation problem and generally, these models work very
badly for new (future) data
5. The standard deviation of the regression coefficient may be calculated
very less than actual. The first–order autocorrelation can be detected
by the Durbin–Watson test
6. Any order autocorrelation can be detected by the Breusch–Godfrey
test
7. Moving average models, MA(q) use past error terms as independent
variables
237
Business Analytics 4.6 Answers for Check Your Progress :
1. Lags 2. First–order 3. Errors
4. 1 and 4 5. Autocorrelation
1. c 2. d 3. a 4. b
5. d 6. a 7. b 8. d
9. d 10. b
4.7 Glossary :
Autocorrelation : Autocorrelation is a mathematical representation
where a time–series data shows correlation with a lagged version of itself.
Durbin–Watson Test : This test helps to detect autocorrelation of
the first order when a time series data correlate with lag 1 data of itself.
Breusch–Godfrey Test : This test helps to detect autocorrelation
of any order.
Autoregressive (AR) Models : AR Models forecast time–series data
using a linear combination of its lagged values. Here the output variable
is Yt while input variables are Yt–1, Yt–2 etc.
Moving Average MA(q) Models : Moving average models use past
forecast errors as input variables. MA (q) models are also known as the
moving average process.
4.8 Assignments :
1. What could be important consequences of ignoring the auto-correlation
problem in a forecasting technique.
2. Write down two important reasons for autocorrelation in a forecasting
technique.
3. What are two important ways to detect auto-correlation, write down
the appropriate scenarios where these techniques can be utilized.
4. What is the basic difference between positive and negative
autocorrelation, explain with an example.
4.9 Activities :
Laburnum chemicals limited has issued their production data for
one of the agricultural pesticide products for the last 26 years.
238
Year Unit Produced Year Unit Produced Auto–Regression (AR)
1 1196 14 2330 and Moving Average
2 1536 15 3792
3 1726 16 2756
4 1636 17 3152
5 1682 18 2920
6 2280 19 3178
7 2430 20 3512
8 2818 21 3446
9 2526 22 3780
10 2468 23 3166
11 2892 24 3350
12 2672 25 3744
13 2486 26 3690
Check the first–order autocorrelation in the above data with the
help of the Durbin–Watson test also develop a one–period (lag 1) two–
period (lag 2) AR model and check if there is significant autocorrelation.
4.10 Case Study :

Vedic Fabrics Pvt Limited conducts an employee satisfaction survey
every year, below is the data for the last 32 years. Below are two
important factors – Salary satisfaction and Motivation at work
Salary Motivation at Salary Motivation at
Year Year
Satisfaction work Satisfaction work
1989 8.5 2005
1990 7.8 2006
1991 4.1 2007
1992 2.3 2008
1993 3.7 2009
1994 2.3 2010
1995 2011
1996 2012
1997 2013
1998 2014
1999 2015
2000 2016
2001 2017
2002 2018
2003 2019
2004 2020
239
1. Is there any clear trend visible in the data, use MS Excel to visualize
the trend also show R–Square and regression equation in the same
graph ?
2. Calculate a Durbin–Watson test to check if there is a first–order
autocorrelation at = 0.05
3. Develop an autoregressive model with a one–period lag and then
develop a model with two–period lag. Compare the results and write
your observations.
Jenkins (1970)
240
BLOCK SUMMARY
Business analytics is the brain of the organization while forecasting
techniques are the spinal cord. Forecasting was one of the fundamental
requirements to establish business analytics as a separate department in a
leading organization. Better forecasting is the key difference between
successful and struggling organizations, in the last two decades supply
chain management has emerged as one of the most important departments
and time–series forecasting is the heart of successful supply chain
management. Forecasting techniques like MAPE and RMSE can be used
to benchmark different industries, these parameters have gained huge
importance in most quality standard certifications like ISO, CMMI etc.
Most of the fundamental features for forecasting are available in MS Excel
while few advanced softwares make forecasting quite easy, for example,
e–views, Minitab, SAS etc. Forecasting techniques are important for both
short term and long term planning in the organization. Good forecasting
helps us to reduce warehouse cost, additional manpower, additional
bandwidth and server capacity. It also helps in optimizing manpower cost,
budgeting, revenue management etc.
241
1. What are the different components of a time series ?
2. How forecasting techniques impact both the top and bottom line of
an organization
3. Write a short note on additive and multiplicative forecasting models
and write down the ideal scenarios in which we can apply each of
these models
4. Write down the difference between weighted average and moving
average forecasting techniques
5. Write a short note on autoregression and what are important reasons
for it
6. Describe the basic concept of the Durbin–Watson test to detect the
first–order autocorrelation
7. Write a short note on the Moving average, MA(q) forecasting model
8. Write down different factors affecting forecasting accuracy of a time–
series
242
1. Explain the benefits of forecasting in the business world with the help
of few examples in the banking, retail and textile industries
2. Write down different forecasting accuracy techniques, which
techniques are suitable for benchmarking across different industries
3. Explain the intuition behind single and double exponential smoothing
techniques. Take data from with assignment question and use MS
Excel solver the optimize the smoothing constants
4. Explain the difference between autoregressive models and the moving
average MA (q) forecasting model. Write down the similarity
between moving average techniques and moving average forecasting
model MA (q)
5. Write down impacts of autocorrelation
243
Unit No. 1 2 3 4
No. of Hrs.
of the block :

.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
.......................................................................................................................
244
DR.BABASAHEB AMBEDKAR
OPEN UNIVERSITY
'Jyotirmay' Parisar,
Sarkhej-Gandhinagar Highway, Chharodi, Ahmedabad-382 481.
Website : www.baou.edu.in

BBA 304 SLM

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

BBA 304 SLM

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

BBA 304 SLM

Uploaded by

Copyright:

Available Formats

DR.

Dr. Ashvin R. Dave

Copyright © 2021 Knowledge Management and Research Organisation.

The need to plan effective instruction is imperative for a successful distance

BLOCK 1 : BUSINESS ANALYTICS FUNDAMENTALS

BLOCK 2 : STATISTICAL CONCEPTS AND HYPOTHESIS

BLOCK 3 : CORRELATION AND REGRESSION

BLOCK 4 : TIME SERIES ANALYSIS

BLOCK-1 BUSINESS ANALYTICS FUNDAMENTALS

Every organization despite its size needs to measure important

Business analytics has been used in different industries for several

• Advanced software techniques are available e.g. advanced data

• Clean data is available, now most organizations have robust software

• Advanced business problem-solving techniques are providing new

Every hour organizations take several decisions which decide their

After learning this block, you will be able to understand :

• Understanding how Business Analytics is changing the decision–

• Different components of business analytics

• Classifications and challenges of business analytics

• Basic concepts of descriptive analytics

• Data types and scale of measurements

• Learning various types of visualization techniques

• Interpretation of visualization techniques

• Understanding the most appropriate visualization technique as per the

Unit 1 : Introduction to Business Analytics

Unit 2 : Descriptive Analytics

Unit 3 : Visualization Techniques for Business Analytics

1.0 Learning Objectives :

1.2 Importance of Data and its Sources

Fig 1.1 Components of Business Analytics

Fig 1.2 Sources of Data Accumulation in Personal Life

Fig 1.3 Sources of Data Accumulation in the Business World

Fig 1.5 Life Cycle of Solving a Business Problem

1.4 Classification of Business Analytics :

Fig 1.8 Types of Business Analytics

Fig 1.9 Sample Business Dashboard

Fig 1.12 Example of Forecasted Demand

1.7 Answers to Check Your Progress :

1.11 Case Study :

1.12 Further Readings :

2.0 Learning Objectives :

Fig 2.1 Sample Dashboard

2.3 Different Type of Data Measurement Scales :

Nominal Ordinal Interval Ratio

2.4 Population and Sample Size :

2.5 Components of Descriptive Statistics :

Central Tendency Median

Fig 2.3 Components of Descriptive Analytics

(54, 330 + 55, 000) 1, 09, 330

MS Excel formula for variance for population, sample is var.p(array)

The standard deviation for a sample :

Check Your Progress – 2 :

2.11 Case Study :

2.12 Further Readings :

3.0 Learning Objectives :

3.2 Introduction to Data Visualization :

Fig 3.1 Histogram

Fig 3.2 Histogram of Machines Sold Per Day