Data Science and Analytics for Ordinary People

Dr. Jeffrey Strickland is the proprietor of

Simulation Educators, Inc. He has been
performing modeling and simulation for
over 20 years, both in the private sector
Data Science and
and in defense. He has earned advanced
degrees in mathematics, and has taught
mathematics and operations research at
several institutions. He has authored a
Analytics for
number of technical books on this subject.
He resides in Colorado. Ordinary People
This book is about really about Big Data Analytics. Yet, it is
written for people who do not have a strong background in data
analysis or statistics. So one might think of this as a survey of
analytics for ordinary people. This is not to say the the rest us
(Geeks) are extraordinary. I suppose I use ordinary because I do
not care for the “Dummy” series of books. Consquently, you will
not find any formulas or tutorials on tools like SAS, SPSS and R.
If you are looking for this content, you need "Predictive Analytics
using R".

Data science and analytics have been been used as synonyms

on occasion. In reality data science includes data modeling, data
mining, data analysis, database architecture and so on. Analytics
is what we do to make sense of the data. That is, we take data
and turn it into information for business decision makers. This,
our course, implies that we translate our data science jargon into
English, so that ordinary people can use the results. After all, all
of our work boils down to, "Is it useful to decision makers?".

Data Science and Analytics for Ordinary People is a collection of

blogs I have written on LinkedIn over the past year. As I continue
to perform big data analytics, I continue to discover, not only my
weaknesses in communicating the information, but new insights
into using the information obtained from analytics and
communicating it. These are the kinds of things I blog about and
are contained herein.

If you are a manager, marketer or any other ordinary person, this

book is for you.
Jeffrey Strickland

Jeffrey Strickland
ISBN 978-1-329-28062-5
90000

ID: 16909563
9 781329 280625 www.lulu.com
Data Science and Analytics for Ordinary People

By

Jeffrey S. Strickland
Data Science and Analytics for Ordinary People

Copyright 2014 by Jeffrey S. Strickland.

All rights reserved. This book or any portion thereof may not be
reproduced or used in any manner whatsoever without the express
written permission of the publisher except for the use of brief quotations
in a book review.

Printed in the United States of America

First Printing, 2015

ISBN 978-1-329-28062-5

Published by Lulu, Inc.

 Table of Contents
TABLE OF CONTENTS .......................................................................................... VII
ACKNOWLEDGEMENTS....................................................................................... XVII
PREFACE .......................................................................................................... XIX
PART I – BIG DATA ......................................................................................... 1
LIFE BEFORE BIG DATA ......................................................................................... 1
Headlines .................................................................................................... 1
Background ................................................................................................ 2
NASA’s Bigger Data .................................................................................... 2
Other Bigger Data ...................................................................................... 2
Conclusion .................................................................................................. 3
Works Cited ................................................................................................ 3
BIG DATA, SMALL DATA, HORSES AND UNICORNS .................................................... 4
I didn’t know it was sick… ........................................................................... 4
The end came quickly… .............................................................................. 5
Life after the horses… ................................................................................. 5
PART II - ANALYTICS ....................................................................................... 6
WHAT IS BUSINESS ANALYTICS? ............................................................................. 6
Types .......................................................................................................... 6
Descriptive analytics .......................................................................................... 7
Predictive analytics ............................................................................................ 7
Prescriptive analytics ......................................................................................... 7
Applications ................................................................................................ 8
PREDICTIVE ANALYTICS & MODELING ...................................................................... 9
Definition .................................................................................................. 10
 Not Statistics............................................................................................ 10
Types ........................................................................................................ 11
Predictive models ............................................................................................ 11
Descriptive models .......................................................................................... 12
Decision models ............................................................................................... 12
Applications ............................................................................................. 13
Analytical customer relationship management (CRM) .................................... 13
Clinical decision support systems .................................................................... 13
Collection analytics .......................................................................................... 14
Cross-sell .......................................................................................................... 14
Customer retention ......................................................................................... 14
Direct marketing .............................................................................................. 15
Fraud detection ............................................................................................... 15
Portfolio, product or economy-level prediction .............................................. 16
Risk management ............................................................................................ 16
Underwriting .................................................................................................... 16

Technology and big data influences ........................................................ 17

Analytical Techniques .............................................................................. 17
Regression techniques ..................................................................................... 18
Classification and regression trees .................................................................. 18
Machine learning techniques .......................................................................... 18

Criticism ................................................................................................... 18
Tools ........................................................................................................ 20
References ............................................................................................... 20
A DANGEROUS GAME WE PLAY .......................................................................... 22
What are our assumptions? .................................................................... 22
Can we really predict the Unpredictable?................................................ 23
What are models really?.......................................................................... 23
Can models be “good enough”? .............................................................. 24
 Did we build the “Right” Model? .............................................................. 25
Is it still dangerous?.................................................................................. 26
PREDICTIVE ANALYTICS: THE SKILLS AND TOOLS ........................................................ 27
What is Predictive Analytics? ................................................................... 27
Statistical Modeling & Tools..................................................................... 28
Data Processing ........................................................................................ 29
PART III – MODELS AND MODELS AND MORE MODELS................................ 30
HOW DO YOU BUILD A MODEL? ............................................................................ 30
Define the problem ................................................................................... 31
Define the Business case .......................................................................... 31
Define the Model Objective ...................................................................... 31
Determine the requirements .................................................................... 31
Gather the Data ....................................................................................... 32
Process the Data....................................................................................... 32
Build the Model ........................................................................................ 32
Interpret the Model Results ...................................................................... 32
Validate the Model for Production ........................................................... 33
Perform an Economic Analysis ................................................................. 33
Present the Results ................................................................................... 33
Follow-up .................................................................................................. 33
Conclusion ................................................................................................ 34
WHAT ARE PREDICTIVE MODELS ANYWAY? ............................................................ 35
How old is Data Science? .......................................................................... 35
What is Predictive Modeling?................................................................... 36
What is a Predictive Model?..................................................................... 36
Examples of Predictive Models ................................................................. 37
What Predictive Models have I Built?....................................................... 40
References ................................................................................................ 41
MATHEMATICAL MODELING ............................................................................... 42
Conclusion................................................................................................ 43
IF YOU BUILD IT THEY WILL BUY... ....................................................................... 44
A Field of Dreams..................................................................................... 44
Rocker Boxes ............................................................................................ 45
Propensity Models ................................................................................... 46
UPLIFT MODELS IN PLAIN ENGLISH, SORT OF... ....................................................... 48
WHAT’S THE DIFFERENCE? ................................................................................. 50
Descriptive models ................................................................................... 50
Predictive models..................................................................................... 50
Prescriptive models .................................................................................. 51
Analytics .................................................................................................. 51
Terms ....................................................................................................... 51
WHAT IS A PROPENSITY MODEL?......................................................................... 52
Model 1: Predicted customer lifetime value ............................................ 53
Model 2: Predicted share of wallet .......................................................... 53
Model 3: Propensity to engage ................................................................ 53
Model 4: Propensity to unsubscribe ........................................................ 54
Model 5: Propensity to buy ...................................................................... 54
Model 6: Propensity to churn .................................................................. 55
Conclusion................................................................................................ 55
WHAT ARE CLUSTERING MODELS? ....................................................................... 56
Predictive model 1: Behavioral clustering ............................................... 56
Predictive model 2: Product based clustering (also called category based
clustering) ................................................................................................ 57
Predictive model 3: Brand based clustering ............................................ 58
WHAT IS A SIMULATION MODEL? ........................................................................ 59
Monte Carlo Simulation Model ................................................................ 60
 Dynamic Simulation Models ..................................................................... 60
Discrete Time Simulation Models ............................................................. 61
Discrete Event Simulation Models ............................................................ 61
Simulation Architectures .......................................................................... 62
Conclusion ................................................................................................ 62
WHAT ARE STOCHASTIC MODELS ......................................................................... 63
Is there a chance?..................................................................................... 63
Stochastic ................................................................................................. 63
Deterministic ............................................................................................ 64
References ................................................................................................ 64
WHAT ARE NEURAL NETWORKS? ......................................................................... 65
Are you neural? ........................................................................................ 65
References ................................................................................................ 68
WHAT IS DISCRETE EVENT SIMULATION? ............................................................... 70
WHAT IS PREDICTIVE ANALYTICS MISSING? ............................................................ 72
Advantages of using simulation ............................................................... 72
What are the Insurance and Financial industries missing? ...................... 73
What could they do differently? ............................................................... 73
We are doing just fine .............................................................................. 74
The One-eye Man in the Kingdom of the Blind ......................................... 74
Break the rules! ........................................................................................ 75
A little at a time ........................................................................................ 75
Will it crash? ............................................................................................. 75
Conclusion ................................................................................................ 76
PART IV – STATISTICAL MATERS ................................................................... 77
WHY YOU SHOULD CARE ABOUT STATISTICS ............................................................ 77
Who is Benjamin Disraeli? ........................................................................ 77
Can we lie statistics? ................................................................................ 78
 Can we lie with data? .............................................................................. 78
Can we lie with statistical models? .......................................................... 79
Can we be confident in models? .............................................................. 79
Conclusion................................................................................................ 80
WHY IS ANALYSIS LIKE HIKING? ........................................................................... 81
Planning ................................................................................................... 81
Execution ................................................................................................. 82
Relating your Story .................................................................................. 83
Conclusion................................................................................................ 83
12 WAYS NOT TO PLEASE YOUR CUSTOMER .......................................................... 85
ANALYTICS AND STATISTICS: IS THERE A DIFFERENCE?............................................... 89
What is Statistics? ................................................................................... 89
What is Analytics? ................................................................................... 90
Conclusion................................................................................................ 91
References ............................................................................................... 92
ARE STATISTICIANS A DYING BREED? .................................................................... 93
STATISTICS IS OBSOLETE ..................................................................................... 96
Why do we no longer need Statistics? ..................................................... 96
What’s wrong with this Picture? ............................................................. 97
How do we fix the Picture? ...................................................................... 99
MATH, PHYSICS AND CHEMISTRY ARE OBSOLETE................................................... 100
PART V – DATA SCIENCE CONCERNS........................................................... 102
DATA SCIENTISTS ARE DEAD, LONG LIVE DATA SCIENCE! ........................................ 102
Call it what it is ...................................................................................... 102
Mimic Mathematical Sciences? ............................................................. 103
A Data Science Taxonomy?.................................................................... 104
Fun with Machine Learning ................................................................... 104
Conclusion.............................................................................................. 105
 SO YOU THINK YOU ARE A DATA SCIENTIST? .......................................................... 106
Assistant Scientist I/II ............................................................................. 108
Scientist I/II ............................................................................................. 108
Senior Scientist I/II .................................................................................. 109
Principal Scientist I/II .............................................................................. 109
5 SIGNS THAT YOU MIGHT BE A DATA SCIENTIST .................................................... 111
HOW CAN I BE A DATA SCIENTIST? ...................................................................... 113
Johns Hopkins University – Data Science ....................................................... 113
University of Illinois at Urbana-Champaign – Data Mining ............................ 113
SAS and Hadoop for Learning ......................................................................... 114

WHY YOU MIGHT NOT WANT TO BE A DATA SCIENTIST ............................................ 115

Saturation............................................................................................... 115
Internet of Things (IoT) ........................................................................... 116
Imposters ................................................................................................ 116
Alternatives ............................................................................................ 117
Count the Costs ...................................................................................... 118
I AM AN ANALYST, REALLY! ................................................................................ 119
ALL THINGS DATA............................................................................................ 121
PYTHON PLUS R EQUALS DATA SCIENTIST? ........................................................... 123
WHY YOUR CLIENT MIGHT NOT BE LISTENING ........................................................ 125
What is the Problem? ............................................................................. 125
Data Analyst Jargon ............................................................................... 125
Client Interaction .................................................................................... 126
Language is Important ........................................................................... 126
A Real Example - Sort of ......................................................................... 127
Conclusion .............................................................................................. 128
PART VI - APPLICATIONS ............................................................................ 129
BIG DATA ANALYTICS AND HUMAN RESOURCES?................................................... 129
 Hogwash!............................................................................................... 130
What is Analytics anyway? .................................................................... 130
Root Causes. .......................................................................................... 130
Where did the Human go?..................................................................... 131
What can Analytics do? ......................................................................... 131
Conclusion.............................................................................................. 132
CALL CENTER ANALYTICS: WHAT'S MISSING? ...................................................... 133
What is the Problem? ............................................................................ 133
What is Discrete Event Simulation? ....................................................... 134
Can We Simulate the Call Center? ......................................................... 134
WIN, LOSE OR DRAW…IS THIS A GAME? ............................................................. 136
Decision Theory ..................................................................................... 136
General Equilibrium Theory ................................................................... 137
Mechanism Design Theory .................................................................... 137
Example ................................................................................................. 138
References ............................................................................................. 140
PART VII – THE POWER OF OPERATIONS RESEARCH................................... 141
WHAT IS OPERATIONS RESEARCH? ..................................................................... 141
Where do you find them? ...................................................................... 142
What are their tools? ............................................................................. 143
What are they built upon? ..................................................................... 143
What is their history? ............................................................................ 144
GETTING THE QUESTION RIGHT ......................................................................... 145
Reference ............................................................................................... 147
HOLISTIC ANALYSIS AND OPERATIONS RESEARCH .................................................. 148
Introduction ........................................................................................... 148
History of Operations Research ............................................................. 149
What do Operations Research Analysts do?.......................................... 150
 Reference................................................................................................ 151
PART VIII - TOOLS ....................................................................................... 152
WHAT ARE R USERS SAYING? ............................................................................ 152
WHY YOU MIGHT USE SAS............................................................................... 158
When to use open-source ....................................................................... 158
How I use SAS ......................................................................................... 158
Why I use SAS ......................................................................................... 159
Open-source and SAS ............................................................................. 159
Conclusion .............................................................................................. 159
WHAT IS BOARD BEAM ?............................................................................... 160
Disclaimer ............................................................................................... 160
What is it? .............................................................................................. 160
What does it do? .................................................................................... 161
How can I get it?..................................................................................... 162
R, SPSS MODELER AND HALF-TRUTHS ................................................................ 163
The Ad .................................................................................................... 163
The Claim ................................................................................................ 163
The Survey .............................................................................................. 164
IMB’s Motive? ........................................................................................ 164
How is SPSS Modeler? ............................................................................ 164
Half-Truths ............................................................................................. 165
Open-Source versus Commercial ............................................................ 165
Conclusion .............................................................................................. 165
AN ANALYTICS BEST KEPT SECRET....................................................................... 166
WHAT IS LINEAR PROGRAMMING? ..................................................................... 168
What Is a Linear Programming Problem? .............................................. 168
MATLAB Code ......................................................................................... 170
Conclusion .............................................................................................. 170
PART IX - ADVICE........................................................................................ 172
SEVEN ANALYTICS LESSONS LEARNED.................................................................. 172
HERE’S ONE MORE THE THING…ADVICE FOR YOUNG PEOPLE .................................... 174
INDEX ......................................................................................................... 177
 Acknowledgements

The author would like to thank colleagues Adam Wright, Adam Miller,
Matt Santoni. And Olaf Larson of Clarity Solution Group. Working with
them over the past two years has validated the concepts presented
herein.

A special thanks to Dr. Bob Simmonds, who mentored me as a senior

operations research analyst.

xvii
 Preface

Ordinary people included anyone who is not a Geek like myself. This
book is written for ordinary people. That includes manager, marketers,
technical writers, couch potatoes and so on.

Data Science and Analytics for Ordinary People is a collection of blogs I

have written on LinkedIn over the past year. As I continue to perform big
data analytics, I continue to discover, not only my weaknesses in
communicating the information, but new insights into using the
information obtained from analytics and communicating it. These are
the kinds of things I blog about and are contained herein.

Data science and analytics have been been used as synonyms on

occasion. In reality data science includes data modeling, data mining,
data analysis, database architecture and so on. Analytics is what we do
to make sense of the data. That is, we take data and turn it into
information for business decision makers. This our course implies that
we translate our data science jargon into English.

The book is organized into nine logical sections: Big Data, Analytics,
Models, Statistical Matters, Data Science Concerns, Applications,
Operations Research, Tools and Advice.

xix
 PART I – Big Data

Life before Big Data

Headlines

Big-Data Computing: Creating Revolutionary Breakthroughs in
Commerce, Science and Society.

— Randal E. Bryant, Randy H. Katz, and Edward D. Lazowska (Dec 2008)


Big data: The next frontier for innovation, competition, and
productivity.

— James Manyika, et al (May 2011)

The term “big data” was included in the most recent quarterly online
update of the Oxford English Dictionary (OED). So now we have a most
1
authoritative definition of what recently became big news: “data of a
very large size, typically to the extent that its manipulation and
management present significant logistical challenges.” The term,
however, may have appeared as early as 1944. [1]

Background
Predictive modeling and analytics are getting a lot of attention these
days, perhaps in light of corporate America’s “discovery” of Big Data.
Well, I have news for you, “big data” has been around longer than you
think and corporate anyone did not “discover it”.

NASA’s Bigger Data

The Space Race and Cold War gave us big data, and NASA and DoD had
the supercomputers to amass and store incredible amounts of data. Of
note, on February 11th, 2000 NASA, launched the shuttle Endeavour on
an 11-day mission. The mission, The Shuttle Radar Topography Mission
(SRTM) was a joint project between the National Geospatial-Intelligence
Agency (NGA) and the National Aeronautics and Space Administration
(NASA). The objective of this project is to produce digital topographic
data for 80% of the Earth’s land surface (all land areas between 60° north
and 56° south latitude), with data points located every 1-arc second
(approximately 30 meters) on a latitude/longitude grid. The absolute
vertical accuracy of the elevation data will be 16 meters (at 90%
confidence). [2]

One of the products, Global (Longitude 180 W-180 E, Latitude 90 N-90

S), including SRTM data for Longitude 180 W-180 E, Latitude 60 N-56 S
region, contains 173 GB for the elevation raw data mosaic with gaps
filled. Follow-on missions collected additional data and certainly “bigger
data” in terms of storage (hundreds of gigabytes).

Other Bigger Data

While working at the Missile Defense Agency’s Threat Modeling Center
we would build 6 degrees of freedom (DoF) threat missiles and a single
missile could generate over a terabyte of data for a simulation. Most

2
scenarios included more than one threat missile. I think we once had a
data file that was on the order of 50 terabytes.

Surveillance data from space also exceeds most people’s concept of big
data, and we have been collecting it since the Cold War. I once had to
write an algorithm that would take the enormous amount of data and
filter it (along with some data reduction) using things like value of
information (VOI), or else it would over-saturate our networks and be
useless.

Conclusion
The point is, “bigger data” has been around a lot longer than “big data”.
The growth of data corresponds with the growth in storage capacity,
which corresponds to the growth in technology, and if NASA had not
raced to get men on the moon, we might still be drawing histograms on
graph paper.

Works Cited
1. Gil Press, “A Very Short History Of Big Data”, Forbes, May 9, 2013
2. Strickland, J. (2011). Using Math to Defeat the Enemy: Combat
Modeling for Simulation. Lulu.com. ISBN 978-1-257-83225-23

3
 BIG DATA, small data, Horses and
Unicorns
Every so often a term becomes so beloved by media that it moves from
‘instructive’ to ‘hackneyed’ to ‘worthless,’ and Big Data is one of those
terms….

— Roger Ehrenberg

Somehow, I missed the funeral, but BIG DATA is dead. The phrase “big
data” now has no value ... it's completely void of meaning. For those of
us who have been around long enough, the mere mention of the phrase
is awful enough to induce a big data migraine — please pass the big data
Tylenol, no not strong enough, the big data Vicodin.

I didn’t know it was sick…

It wasn’t too long ago that Big Data was healthy as a horse. Then the
horse flies got really bad. They brought disease and the horses got sick.
But the media, I mean the horse flies, continued to buzz around, and the
illness got worse.

4
The end came quickly…
After the horse flies finished their work, the worms came and the horses
got sicker. I do not have anything against venders, I mean worms, but
they wouldn’t let a downed horse recover. Then one day the horses
died…all of them. We were not left with ponies—they had been
forgotten. So it would seem that we were left no horses of any kind.

Life after the horses…

With the horses all gone, the unicorns emerged. They were smarter than
the horses…like smart horses with a horn. The unicorns were so smart
that they could predict the future…kind of like predictive unicorns. I
hope they are real ... and the unicorn flies stay away.

5
 Part II - Analytics

What is Business Analytics?

Business analytics (BA) refers to the skills, technologies, practices
for continuous iterative exploration and investigation of past
business performance to gain insight and drive business planning.
Business analytics makes extensive use of statistical analysis,
including descriptive and predictive modeling, and fact-based
management to drive decision making. In recent years, prescriptive
modeling has also taken a role in BA. It is therefore closely related
to management science and operations research. Business analytics
can answer questions like why is this happening, what if these
trends continue, what will happen next (that is, predict), what is the
best that can happen (that is, optimize).

Types
Business analytics is comprised of descriptive, predictive and
prescriptive analytics, these are generally understood to be

6
descriptive modeling, predictive modeling, and prescriptive
modeling.

Descriptive analytics

Descriptive models quantify relationships in data in a way that is

often used to classify customers or prospects into groups. Unlike
predictive models that focus on predicting a single customer
behavior (such as credit risk), descriptive models identify many
different relationships between customers or products. Descriptive
analytics provides simple summaries about the sample audience
and about the observations that have been made. Such summaries
may be either quantitative, i.e. summary statistics, or visual, i.e.
simple-to-understand graphs. These summaries may either form the
basis of the initial description of the data as part of a more extensive
statistical analysis, or they may be sufficient in and of themselves
for a particular investigation.

Predictive analytics

Predictive analytics encompasses a variety of statistical techniques

from modeling, machine learning, and data mining that analyze
current and historical facts to make predictions about future, or
otherwise unknown, events. Predictive models are models of the
relation between the specific performance of a unit in a sample and
one or more known attributes or features of the unit. The objective
of the model is to assess the likelihood that a similar unit in a
different sample will exhibit the specific performance. This
category encompasses models that are in many areas, such as
marketing, where they seek out subtle data patterns to answer
questions about customer performance, such as fraud detection
models.

Prescriptive analytics

Prescriptive analytics not only anticipates what will happen and

when it will happen, but also why it will happen. Further,
prescriptive analytics suggests decision options on how to take

7
advantage of a future opportunity or mitigate a future risk and
shows the implication of each decision option. Prescriptive
analytics can continually take in new data to re-predict and re-
prescribe, thus automatically improving prediction accuracy and
prescribing better decision options. Prescriptive analytics ingests
hybrid data, a combination of structured (numbers, categories) and
unstructured data (videos, images, sounds, texts), and business
rules to predict what lies ahead and to prescribe how to take
advantage of this predicted future without compromising other
priorities

Applications
Although predictive analytics can be put to use in many
applications, I list a few here:

 Analytical customer relationship management (CRM)

 Clinical decision support systems
 Collection analytics
 Cross-sell
 Customer retention
 Direct marketing
 Fraud detection
 Portfolio, product or economy-level prediction
 Risk management
 Underwriting

This is expansion one of my previous posts, "Predictive Analytics

for Almost Everyone."

8
 Predictive Analytics & Modeling
Predictive analytics—sometimes used synonymously with predictive
modeling—is not synonymous with statistics, often requiring
modification of functional forms and use of ad hoc procedures, making
it a part of data science to some degree. It does however, encompasses
a variety of statistical techniques for modeling, incorporates machine
learning, and utilizes data mining to analyze current and historical facts,
making predictions about future.

In business, predictive models exploit patterns found in historical and

transactional data to identify risks and opportunities. Models capture
relationships among many factors to allow assessment of risk or
potential associated with a particular set of conditions, guiding decision
making for candidate transactions. Predictive models are not restricted
to business, for they are used to predict anything from the reliability of
an electronic component to the success of a manned lunar landing.
These model, however, are usually stochastic models that can be used
in a simulation.

Predictive analytics is used in actuarial science (Conz, 2008), marketing

(Fletcher, 2011), financial services (Korn, 2011), insurance,
telecommunications (Barkin, 2011), retail (Das & Vidyashankar, 2006),

9
travel (McDonald, 2010), healthcare (Stevenson, 2011), pharmaceuticals
(McKay, 2009), defense (Strickland, 2011) and other fields.

Definition
Predictive analytics is an area of Data Science that deals with extracting
information from data and using it to predict trends and behavior
patterns. Often the unknown events of interest is in the future, but
predictive analytics can be applied to any type of unknown whether it
be in the past, present or future. For example, identifying suspects after
a crime has been committed, or credit card fraud as it occurs (Strickland
J., 2013). The core of predictive analytics relies on capturing
relationships between explanatory variables and the predicted variables
from past occurrences, and exploiting them to predict the unknown
outcome. It is important to note, however, that the accuracy and
usability of results will depend greatly on the level of data analysis and
the quality of assumptions.

Not Statistics
Predictive analytics uses statistical methods, but also machine learning
algorithms, and heuristics. Though statistical methods are important,
the Analytics professional cannot always follow the “rules of statistics to
the letter.” Instead, the analyst often implements what I call “modeler
judgment”. Unlike the statistician, the analytics professional—akin to
the operations research analyst—must understand the system,
business, or enterprise where the problem lies, and in the context of the
business processes, rules, operating procedures, budget, and so on,
make judgments about the analytical solution subject to various
constraints. This requires a certain degree of creativity, and lends itself
to being both a science and an art,

For example, a pure statistical model, say a logistic regression, may

determine that the response is explained by 30 independent variables
with a significance of 0.05. However, the analytics professional knows
that 10 of the variables cannot be used subject to legal constraints
imposed for say a bank product. Moreover, the analytics modeler is
aware that variables with many degrees of freedom can lead to
overfitting the model. Thus, in their final analysis they develop a good
10
model with 12 explanatory variables using modeler judgment. The
regression got them near to a solution, and their intuition carried them
to the end.

Additionally, the Analytics professional does not always look for a

hypothesis a priori. Consequently, they may use a machine learning
algorithm, such as Random Forests, which does not depend upon
statistical assumptions, but instead they "learn" from the data.

Types
Generally, the term predictive analytics is used to mean predictive
modeling, “scoring” data with predictive models, and forecasting.
However, people are increasingly using the term to refer to related
analytical disciplines, such as descriptive modeling and decision
modeling or optimization. These disciplines also involve rigorous data
analysis, and are widely used in business for segmentation and decision
making, but have different purposes and the statistical techniques
underlying them vary.

Predictive models

Predictive models are models of the relation between the specific

performance of a unit in a sample and one or more known attributes or
features of the unit. The objective of the model is to assess the likelihood
that a similar unit in a different sample will exhibit the specific
performance. This category encompasses models that are in many areas,
such as marketing, where they seek out subtle data patterns to answer
questions about customer performance, such as fraud detection models.
Predictive models often perform calculations during live transactions,
for example, to evaluate the risk or opportunity of a given customer or
transaction, in order to guide a decision. With advancements in
computing speed, individual agent modeling systems have become
capable of simulating human behavior or reactions to given stimuli or
scenarios.

The available sample units with known attributes and known

performances is referred to as the “training sample.” The units in other
sample, with known attributes but un-known performances, are

11
referred to as “out of [training] sample” units. The out of sample bear
no chronological relation to the training sample units. For example, the
training sample may consists of literary attributes of writings by
Victorian authors, with known attribution, and the out-of sample unit
may be newly found writing with unknown authorship; a predictive
model may aid the attribution of the unknown author. Another example
is given by analysis of blood splatter in simulated crime scenes in which
the out-of sample unit is the actual blood splatter pattern from a crime
scene. The out of sample unit may be from the same time as the training
units, from a previous time, or from a future time.

Descriptive models

Descriptive models quantify relationships in data in a way that is often

used to classify customers or prospects into groups. Unlike predictive
models that focus on predicting a single customer behavior (such as
credit risk), descriptive models identify many different relationships
between customers or products. Descriptive models do not rank-order
customers by their likelihood of taking a particular action the way
predictive models do. Instead, descriptive models can be used, for
example, to categorize customers by their product preferences and life
stage. Descriptive modeling tools can be utilized to develop further
models that can simulate large number of individualized agents and
make predictions.

Decision models

Decision models describe the relationship between all the elements of a

decision—the known data (including results of predictive models), the
decision, and the forecast results of the decision—in order to predict the
results of decisions involving many variables. These models can be used
in optimization, maximizing certain outcomes while minimizing others.
Decision models are generally used to develop decision logic or a set of
business rules that will produce the desired action for every customer or
circumstance.

12
Applications
Although predictive analytics can be put to use in many applications, we
outline a few examples where predictive analytics has shown positive
impact in recent years.

Analytical customer relationship management (CRM)

Analytical Customer Relationship Management is a frequent commercial

application of Predictive Analysis. Methods of predictive analysis are
applied to customer data to pursue CRM objectives, which involve
constructing a holistic view of the customer no matter where their
information resides in the company or the department involved. CRM
uses predictive analysis in applications for marketing campaigns, sales,
and customer services to name a few. These tools are required in order
for a company to posture and focus their efforts effectively across the
breadth of their customer base. They must analyze and understand the
products in demand or have the potential for high demand, predict
customers’ buying habits in order to promote relevant products at
multiple touch points, and proactively identify and mitigate issues that
have the potential to lose customers or reduce their ability to gain new
ones. Analytical Customer Relationship Management can be applied
throughout the customer lifecycle (acquisition, relationship growth,
retention, and win-back). Several of the application areas described
below (direct marketing, cross-sell, customer retention) are part of
Customer Relationship Managements.

Clinical decision support systems

Experts use predictive analysis in health care primarily to determine

which patients are at risk of developing certain conditions, like diabetes,
asthma, heart disease, and other lifetime illnesses. Additionally,
sophisticated clinical decision support systems incorporate predictive
analytics to support medical decision making at the point of care. A
working definition has been proposed by Robert Hayward of the Centre
for Health Evidence: “Clinical Decision Support Systems link health
observations with health knowledge to influence health choices by
clinicians for improved health care.” (Hayward, 2004)

13
Collection analytics

Every portfolio has a set of delinquent customers who do not make their
payments on time. The financial institution has to undertake collection
activities on these customers to recover the amounts due. A lot of
collection resources are wasted on customers who are difficult or
impossible to recover. Predictive analytics can help optimize the
allocation of collection resources by identifying the most effective
collection agencies, contact strategies, legal actions and other strategies
to each customer, thus significantly increasing recovery at the same time
reducing collection costs.

Cross-sell

Often corporate organizations collect and maintain abundant data (e.g.

customer records, sale transactions) as exploiting hidden relationships
in the data can provide a competitive advantage. For an organization
that offers multiple products, predictive analytics can help analyze
customers’ spending, usage and other behavior, leading to efficient
cross sales, or selling additional products to current customers. This
directly leads to higher profitability per customer and stronger customer
relationships.

Customer retention

With the number of competing services available, businesses need to

focus efforts on maintaining continuous consumer satisfaction,
rewarding consumer loyalty and minimizing customer attrition.
Businesses tend to respond to customer attrition on a reactive basis,
acting only after the customer has initiated the process to terminate
service. At this stage, the chance of changing the customer's decision is
almost impossible. Proper application of predictive analytics can lead to
a more proactive retention strategy. By a frequent examination of a
customer’s past service usage, service performance, spending and other
behavior patterns, predictive models can determine the likelihood of a
customer terminating service sometime soon (Barkin, 2011). An
intervention with lucrative offers can increase the chance of retaining
the customer. Silent attrition, the behavior of a customer to slowly but
steadily reduce usage, is another problem that many companies face.

14
Predictive analytics can also predict this behavior, so that the company
can take proper actions to increase customer activity.

Direct marketing

When marketing consumer products and services, there is the challenge

of keeping up with competing products and consumer behavior. Apart
from identifying prospects, predictive analytics can also help to identify
the most effective combination of product versions, marketing material,
communication channels and timing that should be used to target a
given consumer. The goal of predictive analytics is typically to lower the
cost per order or cost per action.

Fraud detection

Fraud is a big problem for many businesses and can be of various types:
inaccurate credit applications, fraudulent transactions (both offline and
online), identity thefts and false insurance claims. These problems
plague firms of all sizes in many industries. Some examples of likely
victims are credit card issuers, insurance companies (Schiff, 2012), retail
merchants, manufacturers, business-to-business suppliers and even
services providers. A predictive model can help weed out the “bads” and
reduce a business's exposure to fraud.

Predictive modeling can also be used to identify high-risk fraud

candidates in business or the public sector. Mark Nigrini developed a
risk-scoring method to identify audit targets. He describes the use of this
approach to detect fraud in the franchisee sales reports of an
international fast-food chain. Each location is scored using 10 predictors.
The 10 scores are then weighted to give one final overall risk score for
each location. The same scoring approach was also used to identify high-
risk check kiting accounts, potentially fraudulent travel agents, and
questionable vendors. A reasonably complex model was used to identify
fraudulent monthly reports submitted by divisional controllers (Nigrini,
2011).

The Internal Revenue Service (IRS) of the United States also uses
predictive analytics to mine tax returns and identify tax fraud (Schiff,
2012).

15
Recent advancements in technology have also introduced predictive
behavior analysis for web fraud detection. This type of solution utilizes
heuristics in order to study normal web user behavior and detect
anomalies indicating fraud attempts.

Portfolio, product or economy-level prediction

Often the focus of analysis is not the consumer but the product,
portfolio, firm, industry or even the economy. For example, a retailer
might be interested in predicting store-level demand for inventory
management purposes. Or the Federal Reserve Board might be
interested in predicting the unemployment rate for the next year. These
types of problems can be addressed by predictive analytics using time
series techniques (see Chapter 18). They can also be addressed via
machine learning approaches which transform the original time series
into a feature vector space, where the learning algorithm finds patterns
that have predictive power.

Risk management

When employing risk management techniques, the results are always to

predict and benefit from a future scenario. The Capital asset pricing
model (CAM-P) and Probabilistic Risk Assessment (PRA) examples of
approaches that can extend from project to market, and from near to
long term. CAP-M (Chong, Jin, & Phillips, 2013) “predicts” the best
portfolio to maximize return. PRA, when combined with mini-Delphi
Techniques and statistical approaches, yields accurate forecasts (Parry,
1996). @Risk is an Excel add-in used for modeling and simulating risks
(Strickland, 2005). Underwriting (see below) and other business
approaches identify risk management as a predictive method.

Underwriting

Many businesses have to account for risk exposure due to their different
services and determine the cost needed to cover the risk. For example,
auto insurance providers need to accurately determine the amount of
premium to charge to cover each automobile and driver. A financial
company needs to assess a borrower's potential and ability to pay before
granting a loan. For a health insurance provider, predictive analytics can

16
analyze a few years of past medical claims data, as well as lab, pharmacy
and other records where available, to predict how expensive an enrollee
is likely to be in the future. Predictive analytics can help underwrite
these quantities by predicting the chances of illness, default, bankruptcy,
etc. Predictive analytics can streamline the process of customer
acquisition by predicting the future risk behavior of a customer using
application level data. Predictive analytics in the form of credit scores
have reduced the amount of time it takes for loan approvals, especially
in the mortgage market where lending decisions are now made in a
matter of hours rather than days or even weeks. Proper predictive
analytics can lead to proper pricing decisions, which can help mitigate
future risk of default.

Technology and big data influences

Big data is a collection of data sets that are so large and complex that
they become awkward to work with using traditional database
management tools. The volume, variety and velocity of big data have
introduced challenges across the board for capture, storage, search,
sharing, analysis, and visualization. Examples of Big Data sources include
web logs, RFID and sensor data, social networks, Internet search
indexing, call detail records, military surveillance, and complex data in
astronomic, biogeochemical, genomics, and atmospheric sciences.
Thanks to technological advances in computer hardware—faster CPUs,
cheaper memory, and MPP architectures—and new technologies such
as Hadoop, MapReduce, and in-database and text analytics for
processing big data, it is now feasible to collect, analyze, and mine
massive amounts of structured and unstructured data for new insights
(Conz, 2008). Today, exploring big data and using predictive analytics is
within reach of more organizations than ever before and new methods
that are capable for handling such datasets are proposed (Ben-Gal I.
Dana A., 2014).

Analytical Techniques
The approaches and techniques used to conduct predictive analytics can
broadly be grouped into regression techniques and machine learning
techniques. [Condensed]

17
Regression techniques

Regression models are the mainstay of predictive analytics.

 Linear regression model

 Ridge regression
 LASSO (Least Absolute Shrinkage and Selection Operator)
 Logic regression
 Quantile regression
 Multinomial logistic regression
 Probit regression

Classification and regression trees

 Hierarchical Optimal Discriminant Analysis (HODA)

 Classification and regression trees (CART)
 Decision trees
 Multivariate adaptive regression splines (MARS)

Machine learning techniques

Machine learning, a branch of artificial intelligence, was originally

employed to develop techniques to enable computers to learn.

 Neural networks
 Multilayer Perceptron (MLP)
 Radial basis function (RBF)
 Naïve Bayes
 K-Nearest Neighbor algorithm (k-NN)

Criticism
There are plenty of skeptics when it comes to computers and algorithms
abilities to predict the future, including Gary King, a professor from
Harvard University and the director of the Institute for Quantitative
Social Science. People are influenced by their environment in
innumerable ways. Trying to understand what people will do next
assumes that all the influential variables can be known and measured
accurately. “People’s environments change even more quickly than they

18
themselves do. Everything from the weather to their relationship with
their mother can change the way people think and act. All of those
variables are unpredictable. How they will impact a person is even less
predictable. If put in the exact same situation tomorrow, they may make
a completely different decision. This means that a statistical prediction
is only valid in sterile laboratory conditions, which suddenly isn't as
useful as it seemed before.” (King, 2014)

19
Tools
Tools change often, but SAS appears to be the industry standard, and I
relay heavily on SAS Enterprise Modeler for my job. Be that as it may, I
use R a great deal and find SPSS (particularly SPSS Modeler) useful for
some things. Personally, I prefer R.

References
Barkin, E. (2011). CRM + Predictive Analytics: Why It All Adds Up. New
York: Destination CRM. Retrieved 2014, from
http://www.destinationcrm.com/Articles/Editorial/Magazine-
Features/CRM---Predictive-Analytics-Why-It-All-Adds-Up-
74700.aspx

Hayward, R. (2004). Clinical decision support tools: Do they support

clinicians? FUTURE Practice, 66-68.

Nigrini, M. (2011). Forensic Analytics: Methods and Techniques for

Forensic Accounting Investigations. Hoboken, NJ: John Wiley & Sons
Inc.

Parry, G. (1996, November–December). The characterization of

uncertainty in Probabilistic Risk Assessments of complex systems.
Reliability Engineering & System Safety, 54(2-3), 119–1. Retrieved
2014, from
http://www.sciencedirect.com/science/article/pii/S095183209600
0695

Schiff, M. (2012, March 6). BI Experts: Why Predictive Analytics Will

Continue to Grow. Renton: The Data Warehouse Institute. Retrieved
2014, from http://tdwi.org/Articles/2012/03/06/Predictive-
Analytics-Growth.aspx?Page=1

Strickland, J. (2013). Introduction toe Crime Analysis and Mapping.

Lulu.com. Retrieved from http://www.lulu.com/shop/jeffrey-
strickland/introduction-to-crime-analysis-and-
mapping/paperback/product-21628219.html

20
Strickland, J. (2013). Introduction toe Crime Analysis and Mapping.
Lulu.com. Retrieved from http://www.lulu.com/shop/jeffrey-
strickland/introduction-to-crime-analysis-and-
mapping/paperback/product-21628219.html

21
 A Dangerous Game We Play
We, the scientists who perform predictive modeling, hold a crystal ball
in our hands, making concise and accurate predictions based on quality
data and scientific procedures. Or, do we?

What are our assumptions?

We build models based on these assumptions:

 The data is good

 The statistical processes are failsafe
 The models are exact duplicates of the real system or
phenomenon

But here is the reality:

 The data is rarely good

 We rarely implement the statistics correctly
 The models are simplistic abstractions of real stuff

22
Perhaps this is why Professor Box said, “all models are wrong, but some
are useful.”

Can we really predict the Unpredictable?

Can we really predict the unpredictable, embodied in a time and space
not yet obtained? How good is our crystal ball? Professor Box made his
livelihood and reputation on predictive models, as do I. So how do we
succeed with “bad” models?

Well, first the model are not “bad”. They are just models. Modeling in
general is to pretend that one deals with a real thing while really working
with an imitation. In operations research the imitation is a computer
model of the simulated reality. A flight simulator on a PC is also a
computer model of some aspects of the flight: it shows on the screen
the controls and what the “pilot” (the youngster who operates it) is
supposed to see from the “cockpit” (his armchair). A statistical model
predicting who has the propensity to buy a certain product, or a machine
learning algorithm with the same objective are also models.

What are models really?

In order to understand what it means to model a phenomena or process,
we must first understand the term “model” and understand its
limitation. A model is a physical, mathematical, or otherwise logical
“representation” of a system, entity, phenomenon, or process.
“Representation” or “imitation” are key in understanding models.

A model can also be thought of as an abstraction of the real world, or an

approximation of it. If you think about the problem of modeling a human
being, or just the mind of a human, you can immediately see the
limitations of modeling. We can use the term “system” to encompass
systems, entities, phenomenon, or process.

Since a model is a representation, abstraction, or approximation of the

“system” being modeled, we must understand that it is not an “exact”
representation, i.e., we can’t model every aspect of the system. First, we
don’t know everything we need to know in order to model the system.
We may not be able to define a process of the system with mathematical
23
precision, or with heuristic algorithms, and many of the processes may
not appear logical. Second, the data that we use for a statistical model,
for instance, may not be perfect (probably isn’t), and the model is only
as good as the data upon which it is built. Third, even if we were to know
everything about the system, we may not have enough computing
power to model every process, at least for complex systems, e.g., a
human being, the earth’s ecosystem, etc.; we may not have enough
money to spend on it; and we may not have enough time to accomplish
it.

Can models be “good enough”?

Even with their limitations, models are a good way to gain understanding
of how a system operates. We cannot model every aspect or process of
a real system (at least for most real systems). However, we can measure
some aspects or processes. We cannot build a good model of the global
economy. It is simply too complicated to reduce to an abstraction by
simplifying assumption. But, we can build one for the American (US)
economy, and another for the British economy and another for the
Russian economy. Together, these separate model may provide insight
to a global economy, but certainly not a very accurate one.

The question we often ask is, “How good is your model”? What we
should ask is, “Is your model's performance better than no model at all?”
The answer to the former is probably, “Not very good.” The answer to
the latter is “Probably better.” The graph below represents the
performance of a statistical model with predicted values, actual values
and random values (or no model at all). Assuming the two curves above
the diagonal line (representing no model) are statistically different from
random chance, then the model would be better than having no model.

24
Did we build the “Right” Model?
Clearly a model in this instance is better than no model. But this requires
qualification: is the model valid? This is different than "is the model a
good one". It is not a matter of whether we built the model right,
although that is equally important. No, the question is: “Did we build the
right model?" If our statistical implementations were correct, then the
model depicted in the figure would seem to be a good model. It captures
about 70% of the audience at the 7th pentile. It is relatively smooth and
is about 2.5 times better that no model at the 7th pentile. The question
remains: is it the right model?

Would I use this model to predict? Yes, and I already do so. It is not a
perfect model, but it is a useful one. The model addresses the client’s
business case, and is used to predict the propensity of a customer to
engage in action X (I cannot elaborate on the client or model any more
than this). Yet, though it is used for prediction, it will still occasionally
“miss the mark” and be wrong, due to inherent statistical errors and to
data that may not be “perfect.” So the equation for a useful model might
look like:

“good statistical procedures” + “relatively good data” – “an exact

representation of reality” = “a model that is wrong but useful”

25
Is it still dangerous?
Thus, prediction is a dangerous game we play. We do not really know for
certain what will occur in the months to come, but an educated guess—
based on a model—is probably much better than a random shot in the
dark. Our crystal ball may not work so well after all, but we can still dimly
see what might come to pass.

26
 Predictive Analytics: the skills and
tools
Several weeks ago I posted an article called, What is Predictive Analytics,
describing what it is. In the present article, I want to talk about the skills
and tools that one should have to perform predictive analytics.

I am always at a loss in describing the skills of analytics, for there are

many. I am working on another book about analytics that has a different
approach than Predictive Analytics using R, though I am using material
from three chapters. The new book is an operations research approach
to analytics, covering a different set of methods, skill and tools.
Combined, the two books are over 1000 pages, so perhaps you can see
my dilemma. Hence, this article is going to touch the very basics.

What is Predictive Analytics?

In case you missed my previous article, this is a high level description.
Predictive analytics—sometimes used synonymously with predictive
modeling—is not synonymous with statistics, often requiring
modification of functional forms and use of ad hoc procedures, making
27
it a part of data science to some degree. It does however, encompasses
a variety of statistical techniques for modeling, incorporates machine
learning, and utilizes data mining to analyze current and historical facts,
making predictions about future. Beyond the statistical aspect lies a
mathematical modeling and programming dimension, which includes
linear optimization and simulation, for example. Yet analytics goes even
farther by defining the business case and requirements, which are not
covered here. I discussed those in How to Build a Model.

Statistical Modeling & Tools

This assumes that you already know the basics of parametric and a little
bit of nonparametric statistics. If you are not familiar with these terms,
then you are missing a prerequisite. However, this is gap you can fill with
online courses from Coursera. Though I have never taken one, I have
many colleagues who swear by them.

By statistical modeling I am referring to subject matter that would be

covered beyond material in a statistics for engineering or business
course(s). Here we are concerned with linear regression, logistic
regression, analysis of variance (ANOVA), multivariate regression and
clustering analysis, as well as goodness of fit testing, hypotheses testing,
experimental design and my friends Kolmogorov and Smirnoff.
Mathematical Statistics could be a plus, as it will take you into the
underlying theory.

The tools one would/could use are a myriad and are often the tools our
company or customer has already deployed. SAS modeling products are
well-established tools of the trade. These include SAS Statistics, SAS
Enterprise Guide, SAS Enterprise Modeler, and others. IBM made its mark
on the market with the purchase of Clementine and its repackaging as
IBM SPSS Modeler. There are other commercial products like Tableau. I
have to mention Excel here, for it is all many will have to work with. But
you have to go beyond the basics and into its data tools, statistical
analysis tools and perhaps its linear programming Solver, plus be able to
construct pivot tables, and so on.

Today, there a multitude of open source domain tools that have become
popular, including R and its GUI, R-Studio; the S programming package;

28
and the Python programming language (the most used language in
2014). R, for example, is every bit as good as its nemesis SAS, but I have
yet to get it to leverage the enormous amount of data that I have with
SAS. Part of this is due to server capacity and allocation, so I really don't
know how much data R can handle.

Data Processing
For the forgoing methods, data is necessary and it will probably not be
handed to you on a silver platter ready for consumption. It may be
“dirty”, in the wrong format, incomplete, or just not right. Since this is
where you may spend an abundant amount of time, you need the skill
at tools to process data. Even if this is a secondary task--it has not been
for me--you will probably need to know Structured Query Language
(SQL) and something bout the structure of databases.

If you do not have clean, complete, and reliable data to model with, you
are doomed. You may have to remove inconsistencies, impute missing
values, and so on. Then you have to analyze the data, perform data
reduction, and integrate the data so that it is ready for use. Modeling
with “bad” data results in a “bad” model!

Databases are plentiful and come in the form of Oracle Exadata,

Teradata, Microsoft SQL Server Parallel Data Warehouse, IBM Netezza,
and Vertica. The Greenplum Database builds on the foundations of open
source database PostgreSQL. Or you may need to use a data platform
like Hadoop. Also, Excel has the capacity to store "small amounts" of
data across multiple worksheets and built in data processing tools.

29
 Part III – Models and Models
and More Models

How do you build a model?

Customers and modelers, this is for you. Years ago I learned the seven
steps to model building as a budding Operations Research Analyst. As I
grew from sapling to tree, I realized that something was missing, but I
did not really know what link was absent until I plowed into analytics.
Now I am prepared to tell you, from an Operations Research (OR)
perspective what I believe is a more complete set of steps. [Thanks to
Yifan Gu for suggesting the following quote.]


If I had an hour to solve a problem I'd spend 55 minutes thinking about
the problem and 5 minutes thinking about solutions

—Albert Einstein

30
Define the problem
Defining the problem is necessary. Defining the “right” problem is
absolutely critical, otherwise you're just wasting everyone's time.
However, the customers you are working for may not know how to
express the real problem or may not know what the problem really is.
The Operations Research (OR) Analyst must ask the right questions and
draw the right problem out form where it may be hiding.

Define the Business case

Once you identify the real problem, you have to help answer the
question, “Is the problem worth solving (or at least worth the OR
analyst’s involvement)?”. That may sound odd, but if the problem
statement is not translated in to a business case, then the business does
not need the problem solved badly enough to warrant the time and
effort to solve it. We have to ask, “What is it worth to the business?”
Increased profit? Improved production? Savings of marketing dollars?...

Define the Model Objective

It takes a well-defined problem and solid business case to ensure that
we build the right model or provide the right solution. In my earlier
years, I often found myself halfway through a project before realizing I
was not building the right model. A model objective that answers the call
of the business case is essential. It will keep you on course.

Determine the requirements

We have a well-defined problem, a business case with a supporting
model objective, so now let’s go and build the model. Slow down! What
are the requirements? Do you have the tools you need, the team you
need, and the budget to support it? If data is to be used, do you have
access to it, has it been processed, and is it in the right format? We have
to determine the requirements or we may fail due to some oversight.

31
Gather the Data
If data is necessary, it will probably not be handed to you on a silver
platter ready for consumption. It may be “dirty”, in the wrong format,
incomplete, of just not right. If you have fully identified the
requirements, you will already know what has to be accomplished at this
stage.

Process the Data

This is where you may spend an abundant amount of time. If you do not
have clean, complete, and reliable data, you are doomed. You may have
to remove inconsistencies, impute missing values, and so on. Then you
have to analyze the data, perform data reduction, and integrate the data
so that it is ready for use. Modeling with “bad” data results in a “bad”
model!

Build the Model

So, you have done the hard part and it’s time to have fun. Building the
model is an iterative process. I doubt that anyone ever uses the first
model they build. If they do, they probably should not. There is as much
art as there is science in model development. Modeler judgment is as
important as p-values, and Chi-square tests. If the model does not make
sense to you—is not intuitive—then it is probably not the final model.

Interpret the Model Results

Here is where we often have the most difficulty as analysts. We
understand the mathematical and statistical interpretation of results,
but we have to go beyond that and translate it to the customer’s
language. They are not going to get the significance of the odds ratio of
variable X. If you start thinking about that here, it will make life less
painful later on.

[Thanks to David Nixon for suggesting this addition.] Whatever the

model suggests probably takes a “leap of faith” within the organization.
After all, what the model may have identified is a bunch of correlations

32
that fly in the face of conventional wisdom. The validation step is an
extremely important piece of the puzzle whereby a company may have
to perform a series of controlled experiments to confirm for sure that “x
contributes to the explanation of y”. This leap of faith extends to
potentially creating the necessary data in the first place if it doesn’t
already exist and may involve short-term losses for the promise of a
much larger long-term gain.

Validate the Model for Production

This is purely a scientific task. The model must be documented every
step of the way and then stand up to the scrutiny of stress testing, peer
review, and so on. This is where you find out if you really did build the
right model, and you built it right.

Perform an Economic Analysis

Here is where you answer the call of the business case. You must
translate the results to dollars, the economic benefit of the solution, or
another metric if savings or profit are not objectives. For instance, the
military OR may be saving lives instead of dollars.

Present the Results

If you bothered to interpret the results and translate them to “customer
speak”, this part is easy. But, you are not just presenting results, you are
selling the solution. If your work and your recommendations are not
implemented, you just wasted a lot of time, effort and money.

Follow-up
Too often we fix the problem and go back to our lives or onto another
problem, but following up with the customer to evaluate the
effectiveness of the solution you provided is just as important as the
solution that you provided. It is especially important if you want to ever
see that customer again.

33
Conclusion
So there is my view of model building. I have built just about every kind
of model: linear programs, nonlinear programs, machine learning
models, statistical models, Markov transition models, and so on. I have
built models of unmanned aerial vehicles, space launch vehicles,
communications networks, software systems, propensity to buy a
product, call center operations, combat phenomena, and so on. And I
have, more than I would like to admit, built the wrong model. Hopefully,
this will help the modeler step out on the right foot, and help the model
recipient know what to look for when asking for a modeling solution.

34
 What are Predictive Models
Anyway?
Predictive modeling does not lie solely in the domain of Big Data
Analytics or Data Science. I am sure that there are a few “data scientist”
who think they invented predictive modeling. However, predictive
modeling has existed for a while and at least since World War II. In
simple terms, a predictive model is a model with some predictive power.
I will elaborate on this later.

I have been building predictive models since 1990. Doing the math, 2015
– 1990 = 25 years, I have been engaged in the predictive modeling
business longer that data science has been around. My first book on the
subject, "Fundamentals of Combat Modeling (2007), predates the "Data
Science" of 2009 (see below).

How old is Data Science?

It is really a trick question. The term was first used in 1997 by C. F. Jeff
Wu. In his inaugural lecture for the H. C. Carver Chair in Statistics at the
University of Michigan, Professor Wu (currently at the Georgia Institute
of Technology), calls for statistics to be renamed data science and
statisticians to be renamed data scientists. That idea did not land on solid

35
ground, but the topic reemerges in 2001 when William S. Cleveland
publishes “Data Science: An Action Plan for Expanding the Technical
Areas of the Field of Statistics.” But it is really not until 2009 that data
science gains any significant following and that is also the year that Troy
Sadkowsky created the data scientists group on LinkedIn as a companion
to his website, datasceintists.com (which later became
datascientists.net). [1]

What is Predictive Modeling?

It is not a field of statistics! Yes, we do predictive modeling in statistics,
but it is really a multidisciplinary field and is based more in mathematics
than in other fields. Now, if you consult the most authoritative source of
factual information available to the world, Wikipedia, you will find an
incorrect view of predictive modeling (of course, I do not believe what I
said about Wikipedia). It was formed by people with too much time on
their hands and too little exposure to other disciplines, such as physics
and mathematics.

Predictive modeling may have begun as early as World War II in the

Planning of Operation Overlord, the Normandy Invasion, but was
certainly used in determining air defenses and bombing raid sizes (it may
have appeared as early as 1840 [2]). Now, this is not an article about the
history of operations research, so suffice it to say that the modern field
of operational research arose during World War II. In the World War II
era, operational research was defined as “a scientific method of
providing executive departments with a quantitative basis for decisions
regarding the operations under their control.”[3]

What is a Predictive Model?

The answer is easy: a model with some predictive power. I say that with
caution, and use the word “some”, because more often than not,
decision makers think that these model are absolute. Of course, they
become very disappointed when the predictions do not occur as
predicted. Rather than expand on my simplistic definition, I think some
examples my help.

36
Examples of Predictive Models
The taxonomy of predictive models represented here is neither
exhaustive of exclusive. In other words, there are other ways to classify
predictive models, but here is one.

Times Series Models/Forecasting Models. This kind of model is a

statistical model based on time series data. It uses “smoothing”
techniques to account for things like seasonality in predicting or
forecasting what may happen in the near future. These models are based
on time-series data.

Regression Models. Time series model are technically regression

models, but machine learning algorithms like auto neural networks have
been employed recently in Time Series Analysis. Here I am referring to
logistic regression models used in propensity modeling, and other
regression models like linear regression models, robust regression
models, etc. These models are based on data.

37
Physical models. These models are based on physical phenomena. They
include 6-DoF (Degrees of Freedom) flight models, space flight models,
missile models, combat attrition models (based on physical properties
of munitions and equipment).

38
Machine Leaning Models. These include auto neural networks (ANN),
support vector machines, classification trees, random forests, etc. These
are based on data, but unlike statistical models, they “learn” from the
data.

Weather models. These are forecasting models based on data, but the
amount of data, the short interval of prediction windows and the
physical phenomena involved make them much different that statistical
forecasting models.

Mathematical Models. These are usually restricted to continuous time

models based on differential equations or estimated using difference
equations. They are often used to model very precise processes like the
dynamics solid fuel rockets, or to approximate physical phenomena in
the absence of actual data, like attrition coefficients approximation or
direct fire effects in combat models.

39
Statistical Models. The first two examples, Time Series and Regression
models, are statistical models. However, I list it separately because many
do not realize that statistical models are mathematical models, based on
mathematical statistics. Things like means and standard deviations are
statistical moments, derived from mathematical moment generating
functions. Every statistic in Statistics is based on a mathematical
function.

What Predictive Models have I Built?

I have built predictive models in all example categories except weather
models. Models I have built include Reliability, Availability and
40
Maintainability (RAM) models for Unmanned Aerial Vehicle design;
unspecified models involving satellites (unspecified because they are
classified); unspecified missile models; combat attrition models; 6-DoF
missiles models; missile defense models; propensity to purchase,
propensity to engage, and share or wallet models regression models;
time-series forecasting models for logistics; uplift (net-lift models)
marketing models; ANN models as part of ensembles, classification
trees, and random forests marketing models. I have also worked on
descriptive and prescriptive models.

Models I have consulted on include the NASA Ares I Crew Launch Vehicle
Reliability and Launch Availability; The Extended Range Multi-Purpose
(ERMP) Unmanned Aerial Vehicle RAM Model, The Future Combat
Systems (FCS) C4ISR family of models; FCS Logistic Decision Support
System Test-Bed Model; Unspecified models (unspecified because they
are classified).

References
Press, G. “A Very Short History Of Data Science”, Forbes, May 28, 2013
@ 7:09 AM, Retrieved 05-29-2015.

P. W. Bridgman, The Logic of Modern Physics, The MacMillan Company,

New York, 1927.

Operational Research in the British Army 1939–1945, October 1947,

Report C67/3/4/48, UK National Archives file WO291/1301. Quoted
on the dust-jacket of: Morse, Philip M, and Kimball, George E,
Methods of Operations Research, 1st Edition Revised, pub MIT Press
& J Wiley, 5th printing, 1954

41
 Mathematical Modeling
Again, there are prerequisites like differential and integral calculus and
linear algebra. Multivariate calculus is a plus, particularly if you'll be
doing models involving differential equations and nonlinear
optimization. The skills you need to acquire beyond the basics include
mathematical programming--linear, integer, mixed, and nonlinear. Goal
programming, game theory, Markov chains, and queuing theory, to
name a few, may be required. Mathematical studies in real and complex
analysis, and linear vector spaces, as well as abstract algebraic concepts
like group, fields and rings, can reveal the foundational theory.

Simulation modeling, including Monte Carlo, discrete and continuous

time, plus discrete event simulation can be applied in analytics--I have
not seen this as common practice in business analytics, but it certainly
has its place. These models may rely heavily upon queuing theory,
Markov chains, inventory theory and network theory.

The corporate mainstay is the powerhouse combination of MATLAB and

Simulink. MATrix LABoratory or MATLAB (that is why it is spelled with all
caps!). Other noteworthy commercial products include Mathematica

42
and Analytica. Otave is an open-source mathematical modeling tool that
reads MATLAB code and there are add-on GUI environments (like R-
studio for R) floating around in hyperspace. I recently discovered the
power of Scilab and the world of modules (packages) that are available
for this open-source gem.

For simulation, Simulink works "on top of" MATLAB functions/code for a
variety of simulation models. I wrote the book "Missile Flight
Simulation", using MATLAB and Simulink. ExtendSim is an excellent tool
for discrete event simulation and the subject of my book "Discrete Event
Simulation using ExtendSim". In Scilab, I have used Xcos for discrete
event simulation and Quapro for linear programming. Both are featured
in my next book.

There is a general analytics tool that I do not know much about yet.
BOARD, in its newest release, boasts a predictive analytics capability. I
will be speaking on predictive analytics at the BOARD User Conference
during April 13th-14th in San Diego. Again, I would be remiss not to
mention Excel, and particularly the Solver add-in for mathematical
programming. Another 3rd-party add-in to consider to @Risk.

Conclusion
If you aspire to become an analytics consultant or scientist, you have a
lot of open-source tools, free training and online tutorials at your
fingertips. If you are already working in analytics, you can easily
specialize in predictive analytics. If you are already working in predictive
analytics, you have what you need to become an expert. All of the tools
will either work with your PC's native processing power or through a
virtual machine, for example, when using Hadoop, or remote server.

43
 If You Build It They Will Buy...
Most everyone knows that using a propensity model to predict
customers’ likelihood of buying, engaging, leaving and so on, produces a
larger target audience. That is too bad... for what "everyone knows" is
actually incorrect.

A Field of Dreams
In the movie File of Dreams, actor Kevin Costner (whom people often
confuse me with), plays an Iowa corn farmer struggling to pay the
mortgage on his homestead. He is out farming one day and hears a voice,
like a loud whisper in the wind:

If you build it they will come.

And in the end he does build it and they do come...

Models are not like that. In fact, they reduce the size of the customer
population that has a high propensity to do whatever it is that you are
modeling. If you heard a voice in this context it would say:

If you build it, fewer will come.

44
Wow, that is really bad! So why do we use models?

Actually, it is not bad—that is what we want models to do. But you must
think I am outside my mind for saying so. However, modeling the
propensity of customers to buy, engage, etc. is like prospecting for gold.
There is a lot of rock and sand that has to be filtered out to get the “pure”
gold dust.

Rocker Boxes
A rocker box (also known as a cradle) is a gold mining implement for
separating alluvial placer gold from sand and gravel which was used in
placer mining in the 19th century. It consists of a high-sided box, which
is open on one end and on top, and was placed on rockers.

The inside bottom of the box is lined fles (usually glass of plastic) and a
usually a carpet similar to a sluice. On top of the box is a classifier sieve
or grizzly (usually with half-inch or quarter-inch openings) which
screens-out larger pieces of rock and other material, allowing only finer
sand and gravel through. Between the sieve and the lower sluice section
is a baffle, which acts as another trap for fine gold and also ensures that
the aggregate material being processed is evenly distributed before it
enters the sluice section. It sits at an angle and points towards the closed
back of the box. Traditionally, the baffle consisted of a flexible apron or
“blanket riffle” made of canvas or a similar material, which had a sag of
45
about an inch and a half in the center, to act as a collection pocket for
fine gold. Later rockers (including most modern ones) dispensed with the
flexible apron and used a pair of solid wood or metal baffle boards. These
are sometimes covered with carpet to trap fine gold. The entire device
sits on rockers at a slight gradient, which allows it to be rocked side to
side.

Propensity Models
A propensity model is like a rocker box: what comes out with a high
propensity score is your gold dust customers—the ones that are rich
with propensity to do whatever it is that you are modeling. The debris
scores low in the model. If you have 10,000,000 customers, chances are
that only 10,000 (or fewer) have a high propensity to buy product X. But
those customer are your gold dust. (A friend, Steve Cartwright, just
posted an article on microtargeting.) I also use the analogy of filtered
funnel, where the top has a large opening and the bottom a much
smaller one, and where my "gold dust" customers are small enough to
pass through the tiny funnel opening at the bottom. Many go in, but few
come out.

46
It is incumbent on us modelers to manage the expectations of our
customers. Though quite a few already understand this model
phenomenon, we must establish with the customer what to anticipate
as an outcome at the binning of a model project, even before we think
about data, model functional form and so on.

So, the voice that said, "If you build it they will come", was not talking to
me (even though I strongly resemble Kevin Costner). Instead, my voice
said, "If you build it fewer, but richer will come."

47
 Uplift Models in Plain English, sort
of...
I have used the term "Uplift" when speaking of class of predictive
models, but now I want to provide a plain English description. Before I
do that , however, do not get your hopes up. Uplift models are hard to
construct and even harder to maintain--they are not a "cure all" solution
to marketing problems.

Uplift or netlift refers to what marketing analysts call "incremental lift"

or "incremental response". It is the lift (response) you get from touching
people with a campaign message, that would not have responded
without it--the "Persuadables"..

Uplift models require data from a controlled experiment, that is, a

campaign design with a treated group and a control group. If you use
direct mail as the media for your campaign, your treated group would
receive the direct mail and you control group would not. Among the
customers in the control group are those who will purchase what you
are promoting without receiving the campaign message. These are
called the "buy anyways" and are a very important group in uplift
modeling (some of these are in the treated group as well). They are the
customers that you do not need to spend marketing dollars on, and the
difference in the responses from these individuals and the responses
from the treated group represents the incremental response. So, the

48
netlift is equal to responses from the treated group minus responses
from the control group during the campaign acquisition window,
roughly.

Now, what makes it hard to maintain an uplift model? The model is

super-sensitive to any changes in the campaign upon which it was built.
In other words, if you alter the campaign message too much--and "too
much" is relative--then the uplift model may be a poor predictor. Thus
the campaign message needs to remain fairly consistent over time. If you
cannot do this, then an uplift model is probably not your solution.

Now for a point about data collection. You do not necessarily have to
track who is responding to the campaign. You only need to see who
purchases the product you are marketing during the acquisition window.
Then for both the treated and control group, if a purchase is made during
the campaign acquisition window it counts as a response. If your
customer database is set up so that you track products on a regular
basis--daily, weekly, monthly, and so on--then if the acquisition window
is 60 days, you merely look at the product count before the 60-day
window and the product count after the 60-day window, and measure
the difference. Without going into the technical detail, the modeler or
marketing analyst usually does this, and only one response is counted
regardless of the number of products purchased. The non-technical
reason for this is that we are measuring "response to a campaign," not
"number of products purchased.

Finally, in order to validate an uplift model, you have to have multiple

campaigns with that "consistent" campaign message I talked about.
Again, without going into the technical detail, the model is used to score
the other campaigns and the performance results are compared, using
netlift (usually by deciles or 10 groups) as the metric.

If you are eager for the technical details, or if you are suffering from
insomnia and wish to sleep, download my free e-Book entitled
"Predictive Modeling and Analytics". Uplift modeling is found in Chapter
17. Of course you are always welcome to purchase a copy at Lulu.com
or Amazon.com, thus contributing to my gas money.

49
 What’s the Difference?
I have been writing a little bit about analytics and models. Here is the
differences between three types of models I have discussed that may
need clarification.

Descriptive models
These models describe what has already occurred and in analytics they
are usually statistical. They help the business know how thing are going.
Clustering models might fall into this category.

Predictive models
Predictive models tell us what will probably happen in the future as a
result of something that has already happened. They help the business
forecast future behavior and results. These may be comprised of
statistical models, like regression, or machine learning models, like
neural networks. "Propensity to buy" models fall into this category.

50
Prescriptive models
These models are like descriptive-predictive models. They go an extra
step and tell us why and help the business prescribe a next course of
action. That is, they not only tell us what probably will happen, but also
why it may happen. Form here we can look at what's next, for if we know
the why, we can better know how to mold things for the future.

Analytics
According to The Institute for Operations Research and the
Management Science (INFORMS) Analytics is defined as the scientific
process of transforming data into insight for making better decisions.
To perform good analytics, companies should employ all three types of
modeling in order to cover the entire spectrum of their business.

Terms
I use words like “probably” and “may” because there is always a bit of
uncurtaining in predictive and prescriptive models. You usually hear
about this in phrases like, “95 percent confidence”, or perhaps in terms
like, “Type II error.”

51
 What is a Propensity Model?
Predictive analytics was a topic of one of my recent posts, and I have
been often asked how specifically marketers can use predictions to
develop more profitable relations with their customers. Yesterday, I
talked specifically about predictive modeling. There are three types of
predictive models marketers should know about, but I will only talk
about the first one in this article:

 Propensity models (predictions)

 Clustering models (segments)
 Collaborative filtering (recommendations)

Propensity models are what most people think of when they hear
“predictive analytics”. Propensity models make predictions about a
customer’s future behavior. With propensity models you can anticipate
a customers’ future behavior. However, keep in mind that even
propensity models are abstractions and do not necessarily predict
absolute true behavior (see, “What is Predictive Modeling?”). I’ll go
through six examples of propensity models to explain the concept.

52
Model 1: Predicted customer lifetime value
CLV (Customer Lifetime Value) is a prediction of all the value a business
will derive from their entire relationship with a customer. The Pareto
Principle states that, for many events, roughly 80% of the effects come
from 20% of the causes. When applied to e-commerce, this means that
80% of your revenue can be attributed to 20% of your customers. While
the exact percentages may not be 80/20, it is still the case that some
customers are worth a whole lot more than others, and identifying your
“All-Star” customers can be extremely valuable to your business.
Algorithms can predict how much a customer will spend with you long
before customers themselves realizes this.

At the moment a customer makes their first purchase you may know a
lot more than just their initial transaction record: you may have email
and web engagement data for example, as well as demographic and
geographic information. By comparing a customer to many others who
came before them, you can predict with a high degree of accuracy their
future lifetime value. This information is extremely valuable as it allows
you to make value based marketing decisions. For example, it makes
sense to invest more in those acquisition channels and campaigns that
produce customers with the highest predicted lifetime value.

Model 2: Predicted share of wallet

Predicted share of wallet refers to the amount of the customer's total
spending that a business captures in the products and services that it
offers. Increasing the share of a customer's wallet a company receives is
often a cheaper way of boosting revenue than increasing market share.
For example if a customer spends $100 with you on groceries, is this 10%
or 90% of their grocery spending for a given year? Knowing this allows
you to see where future revenue potential is within your existing
customer base and to design campaigns to capture this revenue.

Model 3: Propensity to engage

A propensity to engage model predicts the likelihood that a person will
engage in some activity, like unethical behavior or post purchases. For

53
example, a propensity to engage model can predict how likely it is that
a customer will click on your email links. Armed with this information
you can decide not to send an email to a certain “low likelihood to click”
segment.

Model 4: Propensity to unsubscribe

A propensity to unsubscribe model tells you which customers not to
touch: if there are high value customers you are at risk of losing to
unsubscribe, you need to find other ways to reaching out to them that
are not by email. For example, you can predict how likely it is that a
customer will unsubscribe from your email list at any given point in time.
Armed with this information you can optimize email frequency. For “high
likelihood to unsubscribe” segments, you should decrease send
frequency; whereas for “low likelihood to unsubscribe” segments, you
can increase email send frequency. You could also decide to use
different channels (like direct mail or LinkedIn) to reach out to “high
likelihood to unsubscribe” customers.

Model 5: Propensity to buy

The propensity to buy model tells you which customers are ready to
make their purchase, so you can find who to target. Moreover, once you
know who is ready and who is not helps you provide the right aggression
in your offer. Those that are likely to buy won’t need high discounts (You
can stop cannibalizing your margin) while customers who are not likely
to buy may need a more aggressive offer, thereby bringing you
incremental revenue.

For example, a “propensity to buy a new vehicle” model built with only
data the automotive manufacturer has in their database can be used to
predict percent of sales. By incorporating demographic and lifestyle data
from third parties, the accuracy of that model can be improved. That is,
if the first model predicts 50% sales in the top five deciles (there are ten
deciles), then the latter could improve the result to 70% in the top five
deciles.

54
Model 6: Propensity to churn
Companies often rely on customer service agents to "save" customers
who call to say they are taking their business elsewhere. But by this time,
it is often too late to save the relationship. The propensity to churn
model tells you which active customers are at risk, so you know which
high value, at risk customers to put on your watch list and reach out.
Armed with this information, you may be able to save those customers
with preemptive marketing programs designed to retain them.

Often propensity models can be combined to make campaign decisions.

For example, you may want to do an aggressive customer win back
campaign for customers who have both a high likelihood to unsubscribe
and a high predicted lifetime value.

Conclusion
Predictive analytics models are great, but they are ultimately useless
unless you can actually tie them to your day-to-day marketing
campaigns. This leads me to the first rule of predictive analytics:

Always make sure that your predictive analytics platform is directly

integrated with your marketing execution systems such as your email
service provider, web site, call center or Point of Sell (POS) system.

It is better to start with just one model that you use in day-to-day
marketing campaigns than to have 10 models without the data being
actionable in the hands of marketers.

55
 What Are Clustering models?
Clustering is the predictive analytics term for customer segmentation.
Clustering, like classification, is used to segment the data. Unlike
classification, clustering models segment data into groups that were not
previously defined. Cluster analysis itself is not one specific algorithm,
but the general task to be solved. It can be achieved by various
algorithms that differ significantly in their notion of what constitutes a
cluster and how to efficiently find them.

With clustering you let the algorithms, rather than the marketers, create
customer segments. Think of clustering as auto-segmentation.
Algorithms are able to segment customers based on many more
variables than a human being ever could. It’s not unusual for two clusters
to be different on 30 customer dimensions or more. In this article I will
talk about three different types of predictive clustering models.

Predictive model 1: Behavioral clustering

Behavioral clustering informs you how people behave while purchasing.
Do they use the web site or the call center? Are they discount addicts?
How frequently do they buy? How much do they spend? How much time
will go buy before they purchase again? This algorithm helps set the right
tone while contacting the customer. For instance, customers that buy
56
frequently but with low sized orders might react well to offers like ‘Earn
double rewards points when you spend $100 or more.

Behavioral clustering can also informs us on other behaviors, such as

crime and is used in performing crime analysis. In the example below
there are three crime clusters (only the top ten are shown in the table):

Graphically, the clusters appear as follow:

Cluster plot created in R-Studio

Predictive model 2: Product based clustering

(also called category based clustering)

57
Product based clustering algorithms discover what different groupings
of products people buy from. See the example below of a category (or
product) based segment or cluster. You can see people in one customer
segment ONLY buy Pinot Noir, whereas those in another customer
segment buy different types of Varietal products, such as Champagne,
Chardonnay, Pinot Grigio and Prosecco – but never Cabernet Sauvignon,
Malbec or Espumante. This is useful information when deciding which
product offers or email content to send to each of these customer
segments.

Predictive model 3: Brand based clustering

Brand-based clustering, on the other hand, focuses on the brand of
items customers purchase. Marketers can use this information to project
what other brands those customers are likely to buy. Customers are then
ordered according to Nike, Adidas, Under Armour, etc. Now you know
what specific brands to pitch to certain customers. When a brand
releases new products – you know who is likely to be interested.

58
 What is a Simulation Model?
The question is actually misleading. A simulation can be comprised of
one model or many interacting models. For instance we might take a
statistical model and simulate it using random variates. In this case we
are simulating one model. But suppose we want to simulate the
effectiveness of a ballistic missile defense system. This system is
comprised of models for sensors, interceptors, guidance systems,
command and control, and so on. So in this instance we are simulating a
system of models.

Perhaps I should just answer the question, “What is simulation?”

Simulation is the process of executing a model over time (or another
quantity) to see how a system performs. Yet this very answer is
misleading as well. For a single run of a model does not generally yield
reliable results. This is due to a number of inherent challenges of
simulation, which I will not address here, but suffice it to say that we
generally want to run a model over and over and see how it behaves on
average.

There are several types of simulation models, some of which you may
have heard.

59
Monte Carlo Simulation Model
Monte Carlo simulation is used to estimate stochastic processes where
either the underlying probability distributions are unknown or difficult
to calculate by exact computations. Monte Carlo simulations sample
probability distributions for each variable to produce hundreds or
thousands of possible outcomes. The results are analyzed to get
probabilities of different outcomes occurring. For example, a
comparison of a spreadsheet cost construction model run using
traditional “what if” scenarios, and then run again with Monte Carlo
simulation and Triangular probability distributions may show that the
Monte Carlo analysis has a narrower range than the “what if” analysis.
This is because the “what if” analysis gives equal weight to all scenarios,
while Monte Carlo method hardly samples in the very low probability
regions. The samples in such regions are called "rare events". Monte
Carlo simulations may be used to:

 To make inventory-ordering decisions in the face of uncertain

demands
 To price bids for contracts when competitors’ bids are uncertain
 To insure the consistent quality of manufactured goods when
the sources of defects are uncertain
 To schedule and manage large projects when the durations of
individual tasks are uncertain.
 To make investment decisions in sometimes volatile markets

Dynamic Simulation Models

Dynamic Simulation models include continuous system simulation and
discrete system simulation. If time plays a substantive role, these are
called continuous time models. Continuous systems simulations are
typically solved using systems of differential equations. Discrete systems
simulations are usually in the form of discrete time model or discrete
event models.

60
Discrete Time Simulation Models
In Dynamic Time simulation, the models are iterated of fixed time
increments. For example, we may want to experiment to the operation
of a car wash over time and increment the model every ten minutes to
observe the frequency that it is used through the working day. We might
make decisions about its operation based on the frequency of use in
peak hours. Or we might want to see how an interceptor missile
performs over time as it attempts to destroy a theater ballistic missile,
making observations every second. We can use these simulations to:

 To make inventory-ordering decisions in the face of uncertain

demands
 To study the reliability, maintainability, and availability of
machinery and systems
 To study the waiting times of customers is a certain service
process

Discrete Event Simulation Models

With a Discrete Event simulation time plays a role, but the model is not
iterated base on time. Rather, the model is iterated when an event
occurs. Events are generally classified as arrivals, services, and
departures (from queuing theory). The time between events is not
constant, but are drawn from probability distributions. For example, we
may know that the interarrival times of customers at a bank are
exponentially distributed. Then, we would use random numbers to
generate random arrival times. Likewise, we would have random service
times. If the service times are significantly greater than the interarrival
times, we might need to hire more bank tellers, for example. In the case
of the ballistic missile intercept, we might want to observe its behavior
every time its guidance system causes a change in course, rather than
every second. We can use these simulations to study:

 Manufacturing plants
 Communications networks
 Transportation systems
 Combat operations
 Other systems that involve service, queuing, and processing
61
Simulation Architectures
Simulation architectures exist for running mixed time simulations, such
as discrete time and discrete events simulations. One of the earliest
architectures is known as Distributed Interactive Simulation or DIS. The
High Level Architecture (HLA) is another example.

Conclusion
We use simulations to experiment with systems and perform analyses
without having to use the actual systems. For example, it would be very
disruptive to a call center if we performed experiments with it during its
normal operating hours, which in many cases is 24-7. However, if we can
construct a reasonable model of the call center, then we can use it to
simulate call center operations and perform our experiments using the
simulation.

62
 What are Stochastic Models
Is there a chance?
In my previous post on simulation, I used the term stochastic. What does
that mean? This article is about the meaning of “stochastic” and its
counterpart, “deterministic”.

Stochastic
Stochastic is synonymous with “random.” The word is of Greek origin
and means “pertaining to chance” (Parzen 1962, p. 7). It is used to
indicate that a particular subject is seen from point of view of
randomness. Any system or process that must be analyzed using
probability theory is stochastic at least in part. Stochastic systems and
processes play a fundamental role in mathematical models of
phenomena in many fields of science, engineering, and economics.
Familiar examples of processes modeled as stochastic (or stochastic time
series) include stock market and exchange rate fluctuations, signals such
as speech, audio and video, medical data such as a patient's EKG, EEG,
blood pressure or temperature, and random movement such as
Brownian motion or random walks. Stochastic models are sometimes
referred to as “probabilistic models”.

63
 Plot of S&P Composite Real Price Index, Earnings, Dividends, and
Interest Rates using data from
http://irrationalexuberance.com/shiller_downloads/ie_data.xls

Deterministic
A deterministic system is one whose resulting behavior is entirely
determined by its initial state and inputs, and which is not random or
stochastic. Processes or projects having only one outcome are said to be
deterministic their outcome is 'pre-determined.' A deterministic
algorithm, for example, if given the same input information will always
produce the same output information. An example of a deterministic
system is the physical laws that are described by differential equations,
even though the state of the system at a given point in time may be
difficult to describe explicitly. An example of a deterministic model is a
logistic regression model of customers’ propensity to buy life insurance
products. A stochastic system is one that is non-deterministic.

References
Parzen, E. Stochastic Processes. Oakland CA: Holden Day, p. 7, 1962.

64
 What Are Neural Networks?
Are you neural?
What I really meant to say is, “Are your models neural?” Should they be?
Could they be? I have built some crazy models, but nothing like a neural
network or auto-neural network (ANN). Maybe we should ask “Is you
modeler neural?” Be that he may be, here is a layman's
explanation...hopefully (just remember I am a modeler and cut me some
slack).

We have been building these since the advent of computers…wait…my

alter ego does not agree… really? It appears that the first artificial
neuron was produced in 1943 by the neurophysiologist Warren
McCulloch and the logician Walter Pits. But the technology available at
that time did not allow them to do too much. So, this happened before
computers? I thought this was a machine learning algorithm… Oh. It
seems as though ANN were first described as an information processing
paradigm that was inspired by the way biological nervous systems, such
as the brain, process information.

The key element of this paradigm was the novel structure of the
information processing system. It was composed of a large number of
highly interconnected processing elements (neurons) working in unison
65
to solve specific problems. ANNs, like people, learn by example. An ANN
is configured for a specific application, such as pattern recognition or
data classification, through a learning process. Learning in biological
systems involves adjustments to the synaptic connections that exist
between the neurons. This is true of ANNs as well.

There was an initial surge in ANNs with support from neuroscience. In

1958 Frank Rosenblatt stirred considerable interest in ANNs when he
designed and developed the Perceptron. The Perceptron had three
layers with the middle layer known as the association layer. This system
could learn to connect or associate a given input to a random output
unit. Other development emerged during this period but by 1969,
Minsky and Papert wrote a book in which they generalized the
limitations of single layer Perceptron to multilayered systems. In the
book they said: "...our intuitive judgment that the extension (to
multilayer systems) is sterile", which resulted in funding for these
projects drying up and a period of “frustration” was incurred.

There was a re-emergence of interest and research in the late 1970s and
early 1980s. However, in the 1990s, neural networks were overtaken in
popularity in machine learning by support vector machines and other,
much simpler methods such as linear classifiers. Yet the neurons would
not dies and between 2009 and 2012, work at the Swiss AI Lab IDSIA
produced award winning neural networks in competitions in pattern
recognition and machine learning.

At any rate, An ANN is typically defined by three types of parameters:

1. The interconnection pattern between the different layers of

neurons
2. The learning process for updating the weights of the
interconnections
3. The activation function that converts a neuron's weighted input
to its output activation

Without going into the mathematical detail (my book, “Predictive

Modeling and Analytics”, does that), the main attraction in neural
networks is the possibility of learning. Given a specific task to solve, and
a class of functions, learning means using a set of observations to find

66
optimal function (or optimal solution) in this set of functions, or a
function that solves the task in some optimal sense. The simple ANN can
consists of three layers as depicted below. Within the hidden layer,
learning occurs. In very simple pseudo-code, where f is an input node h
is a hidden node and f* is an output node, this might look like:

Neural networks, with their remarkable ability to derive meaning from

67
complicated or imprecise data, can be used to extract patterns and
detect trends that are too complex to be noticed by either humans or
other computer techniques. A trained ANN can be thought of as an
"expert" in the category of information it has been given to analyze. This
expert can then be used to provide projections given new situations of
interest and answer "what if" questions. ANNs are universal
approximators, and they work best if the system you are using them to
model has a high tolerance to error. You would not want to use a neural
network to balance checkbook! However they work very well for:

 capturing associations or discovering regularities within a set of

patterns;
 where the volume, number of variables or diversity of the data
is very great;
 the relationships between variables are vaguely understood; or,
 the relationships are difficult to describe adequately with
conventional approaches.

Another instance where they might be useful is when the available

response data is sparse. I have also found them to be useful in ensemble
models along with logistic regression models. The tendency is to make a
good preforming logistic regression model perform even better as a
component model of the ensemble. In other words, it makes something
that already tastes good taste even better.

Before using an ANN model, you should have (or if you are not the
modeler, the modeler should have) an understanding of the
mathematically complexity of ANNs. To some degree, they “black box”
models; that is a model in which you cannot see the internal workings.
In other words, using our pseudo-code, we cannot see how h is learning.

References
http://www.kurzweilai.net/how-bio-inspired-deep-learning-keeps-
winning-competitions 2012 Kurzweil AI Interview with Jürgen
Schmidhuber on the eight competitions won by his Deep Learning
team 2009–2012

68
McCulloch, Warren; Walter Pitts (1943). "A Logical Calculus of Ideas
Immanent in Nervous Activity". Bulletin of Mathematical Biophysics
5 (4): 115–133. doi:10.1007/BF02478259

Minsky, M.; S. Papert (1969). An Introduction to Computational

Geometry. MIT Press. ISBN 0-262-63022-2.

Rosenblatt, F. (1958). "The Perceptron: A Probalistic Model For

Information Storage And Organization In The Brain". Psychological
Review 65 (6): 386–408. doi:10.1037/h0042519

Strickland, J. (2014). “Predictive Modeling and Analytics”. Lulu Inc. ISBN

978-1-312-37544-4

69
 What is Discrete Event Simulation?
I do not go to the bank very much anymore. My handy iPhone app allows
me to deposit checks online and the ATM gives me fast cash. But, have
you ever waited in line for service at a bank on your lunch hour? Enter
Discrete Event Simulation...

Discrete Event Simulation (DES) is the process of codifying the behavior

of a complex system as an ordered sequence of well-defined events. In
this context, an event comprises a specific change in the system's state
at a specific point in time (arrival at the bank, service by a teller, etc.).
Rather than stepping based on a time increment, like every second, DES
advances based on events—events that may or may not be equally
spaced in time.

Common applications of DES include:

 stress testing ( a form of deliberately intense or thorough testing

used to determine the stability of a given system or entity)
 evaluating potential financial investments
 studying call center operations
 modeling procedures and processes in various industries, such
as manufacturing and healthcare
 studies that support system design

70
  studying reliability, availability and maintainability (RAM)
 just about anything that involves queuing (arrivals, waiting time,
service time)

As an example of a situation that lends itself to DES, consider the

operations of a call center. When a customer calls, you have an arrival
event. If they have to wait for a representative they enter a queue until
such time as their call is answered, which begins a service event. If their
call was rerouted upon entering the system--the initial arrival event--the
rerouting creates a new arrival event for the call center to which it was
transferred, and so on. One can study where bottle-necks occur,
rerouting efficiency, waiting times, balking (customers deciding not to
join the queue if it is too long), reneging (customers leave the queue if
they have waited too long for service), etc.

An effective DES process must include, at a minimum, the following

characteristics:

 Predetermined starting and ending points, which can be discrete

events or instants in time (arrivals and departures, for instance).
 A method of keeping track of the time that has elapsed since the
process began (waiting time, for instance).
 A list of discrete events that have occurred since the process
began (begin service, for instance).
 A list of discrete events pending or expected (if such events are
known) until the process is expected to end.
 A graphical, statistical, or tabular record of the function for
which DES is currently engaged (plot of waiting times and service
times, for instance).

DES is commonly used to monitor and predict the behavior of

investments; the stock market is a classic example. DES can also help
administrators predict how a network will behave under extraordinary
conditions, such as the Internet or claims call center during a major
disaster, like a wildfire or earthquake.

71
 What is Predictive Analytics
Missing?
Simulation! Oh sure, there are pockets of analysts using simulation in
predictive analytics and pockets of companies that use the results, but
the use of simulation to predict behavior (e.g., propensity to do
something) is not widespread.

I am taking a break today from the “how to” and “X reasons why” articles
and offer instead a little editorial contemplation.

Several industries rely heavily upon simulation models: Aeronautics,

Aerospace, Defense, Transportation, Medical and Maritime for example.
So, why do these industries use simulation, while the Financial and
Insurance industries do not?

Advantages of using simulation

Simulation provides a method for checking your understanding of the
world around you and helps you produce better results, faster. It is an
important tool you can use to:

 Predict the course and results of certain actions

72
  Understand why observed events occur
 Identify problem areas before implementation
 Explore the effects of modifications
 Confirm that all variables are known
 Evaluate ideas and identify inefficiencies
 Gain insight and stimulate creative thinking
 Communicate the integrity and feasibility of your plans

What are the Insurance and Financial

industries missing?
I see, read about, and create statistical models for predictive analytics in
this realm. It is amazing there are so many people who will accept a
predictive model based on internal or third party data. Even if they
realize how bad their data really is, they’re still locked into statistical
models. I know of at least two instances (two different companies)
where their whole approach to predictive analytics is based upon time-
series analysis—nothing else is considered.

So, your financial institution may be using time-series analysis solely to

predict propensity, deploying the model and operating based on its
results. Wow! What faith it takes to do something like that! There is very
little understanding of why certain behaviors or events occur and very
little insight into potential problems before implementation. And the
data is blindly trusted—“it must be good since we collected it!”

What could they do differently?

Perhaps there is a fear of the unknown, and perhaps the unknown is
anything outside of the realm of logistic regression or time-series
analysis. This is of course what I do, so I am talking about impacting my
livelihood. Moreover, I know, as George Box would echo if her were still
with us, that my models are wrong! In order to get the closer to “right”
(that is like trying to reach infinity), I know I need simulation. Simulate
to ensure all the variables are known. Simulate to know the variables are
reasonably distributed and experiment with alternatives. Simulate to
gain insight into a particular propensity phenomenon. Simulate to…

73
We are doing just fine
The response to this might be that they already understand these
phenomena well and they already know what variables are important.
They probably did at one time. However, I know for sure my spending
behavior, engaging behavior, borrowing behavior and so on, has been
altered drastically by a struggling economy. “Oh, but our economy is
healthy,” you might say. So then, why do products and services cost
more and I get paid less. Ten years ago, a Senior Operations Research
Analyst was worth $X thousand, but today he or she is worth $(X – Y)
thousand, but that energy bill keeps rising without bound. Should we not
even test the idea that the world has changed and what we used to know
is now just a mystery? However, we are doing just fine, aren’t we?

The One-eye Man in the Kingdom of the Blind

You may have read a post by this title or at least seen the quote that this
comes from, but for those of you who have no idea what I am talking
about, here is an explanation. Kurt Vonnegut, author of Player Piano,
said this in his book:

“Almost nobody’s competent, Paul. It’s enough to make you cry to see
how bad most people are at their jobs. If you can do a half-assed job of
anything, you’re a one-eyed man in the kingdom of the blind.”

The Operations Research analyst should have one good eye and hence
lead the blind to safety. I argued before about how OR analysts must
have a holistic approach to problem solving, and that includes
simulations. “Oh, but I have not delved into simulation before”, one
might say. My response is, “what are you waiting for?” LPs, IPs, and
MILPs (if you are an OR analyst you know what these are) are not going
to solve every problem we face, but it seems to be the focus in many OR
educational programs. Get out of your comfort zone and get busy being
a holistic problem solver.

74
Break the rules!
Every OR analyst should have a little bit of rebel inside them. Our
approaches must be novel in many cases and the rules that say
regression models must only be broken if we are to be worth our weight
in salt (since we are not on a gold standard). When our models are not
producing much lift (or net lift), something may be broken. It could be
the data or the preconceived notion of customer propensity. Simulation
can help find the root causes, for you can simulate a system or
phenomenon without ever touching the real, operational system.

A little at a time
That financial industry might not be ready to pour the weight of their
analytics resources into simulation, and that is understandable. So,
perhaps start with a pilot study, get some results and show them how
simulation can help them. Imagine where we might be if the financial
industry had been doing this kind of “What if” analysis in the early part
of this century. Would we have seen the events of 2008-ish coming?

Will it crash?
We do not fly launch vehicles in space without simulation, nor launch
test missiles without simulation. It is too expensive—in people and
material—to launch and fail. In our early years of space exploration, we
experienced many crashes before the first human was launched into
space, and we have had some failures since then. But a lot of time, effort
and money go into simulation for these programs. We used simulation
before executing Operation Iraqi Freedom, and one senior officer was
quoted saying that we had fought that battle over and over in
simulation, prior to deploying, which made the fight look easy (even
though there is no “easy” in warfare). Imagine how many lives were
saved by simulation. Will it crash? Of course it will, but to simulate the
crash is much better than experiencing the crash in real time.

75
Conclusion
I have thus far failed to impress my financial institution customer with
the importance of simulation. Technically, I am a failure as an OR analyst.
Sure, I have given my customer the product they asked for, but
sometimes the customer does not know what they really need. It is
incumbent on us to help them to ask the right questions, develop the
best business cases, and use the correct solution method—and that may
be simulation.

76
 Part IV – Statistical Maters

Why you should care about Statistics

I read a recent post that used the old adage, “Lies, More Lies, and
Statistics.” The author claims to have a few statistic courses and was
apparently influenced by a professor who has never done anything that
was “real”. The author claimed that you can lie with statistics. He is
wrong!


There are three types of lies — lies, damn lies, and statistics.

― Benjamin Disraeli

Who is Benjamin Disraeli?

Benjamin Disraeli was one of the great British politicians of the
nineteenth century, Disraeli served twice as Tory Prime Minister (1868
and 1874 - 1880) and was also a prominent figure in opposition. He is
most famous today for the bitter hatred between himself and his

77
political rival William Gladstone. He enjoyed the favor of Queen Victoria,
who shared his dislike of Gladstone. Benjamin was not a statistician!

Can we lie statistics?

The average of a sample is a statistics, as is the shape of the data’s
distribution. Statistics tell you what is in the data—it describes the data
and uses data to predict future events. Statistics is based on
mathematical theory. If we have data comprised of the ages 12, 13, 11,
and 10, the expression (12+13+11+10)/4 = 11.5 is the mathematical
average of the ages. The calculation is absolute, based on the data. That
is why “real” statistician are required to study mathematical statistics.
The mathematics does not lie, so the statistics do not either.


I couldn't claim that I was smarter than sixty-five other guys--but the
average of sixty-five other guys, certainly!

― Richard P. Feynman, Surely You're Joking, Mr. Feynman!:

Adventures of a Curious Character

Can we lie with data?

However, you can manipulate the data. You can collect the data that
only supports your position or hypothesis. You can select specific data
from a database that supports your position or hypothesis. You can
make up data that supports your position or hypothesis. The statistics
will still tell you what is in your data, and it will do so using the
mathematical foundations that it uses for all data. Suppose we do not
want the average age to be 11.5 (perhaps decimals scare us). We could
replace the actual age “10” with “12” (even though the true age is 10).
Then we would have (12+13+11+12)/4 = 12. Age 12 is the mathematical
average of the ages. The problem is not the statistics, it is the data!


Outside of the killings, DC has one of the lowest crime rates in the
country.

― Marion Barry
78
Can we lie with statistical models?
We modelers say that our models are only as good as the data they are
built upon. We try to select data that is predictive, without the biases of
our positions or theories. Sometimes the phenomena we are trying to
predict, say the propensity to shed a product, is just not there. It could
be that our intuition is wrong. It could be that the data is wrong. It could
be that we do not have all the pertinent data. Hence, we are often left
with the conclusion that we need more data, better data, or we cannot
build a predictive model. Again, there is nothing wrong with statistical
methods, say linear regression. The problem lies with the data: not
enough, too much, missing data, and so on.


One of the first things taught in introductory statistics textbooks is that
correlation is not causation. It is also one of the first things forgotten.

― Thomas Sowell, The Vision of the Anointed: Self-Congratulation as a

Basis for Social Policy

Can we be confident in models?

No matter how hard we try to be unbiased, the data does not always
allow us this luxury. Often, biases are built into the data itself, through
sampling, collection and entry errors. However, statistics give us the
ability to check for biases, correlation between variables, collinearity,
etc. If it were not for statistics, we could not use data for anything—we
could not turn data into information. Averages, variances, distributions,
and so on are all statistics. Having 10 million test scores is useless, if you
cannot describe the distribution of the scores, for example.


We may at once admit that any inference from the particular to the
general must be attended with some degree of uncertainty, but this is
not the same as to admit that such inference cannot be absolutely
rigorous, for the nature and degree of the uncertainty may itself be
capable of rigorous expression.

― Sir Ronald Fisher

79
Conclusion
Perhaps we should change the old adage to, “Lies, More Lies, and
Manipulated Data.”

80
 Why is Analysis like Hiking?
I was in the Rocky Mountains last week, near Buena Vista. I went on
several hikes along the Colorado Trail and over some of the roughest
terrain I have ever hiked, due to the rainfall and resulting washouts.
Afterward, I went to my customer’s site for a few days. I missed the
hiking, but as I reflected on it the reason was not just because I was at
work again. It was because it was like doing analytics.

Planning
Hiking requires extensive planning. My longest hike was 16 kilometers.
Part of the hike (toward the end) was on the Colorado Trail. The
beginning was up a set of switchbacks that went from about 2740 meter
to 3050 meters. At the top, I had to go cross-country until picking up the
main trail. During this first portion I would probably not see another
human being. I would also be exposed to the intense sun. Since I was
hiking alone, I needed to let someone know my route and the time I
expected to return. If I were incapacitated on the isolated portion of my
hike, it might be days before I was found. I also had to plan my load,

81
which needed to provide protection from the sun and mosquitos,
protection from potential rainfall, water, and so on.

The descent would be on the Colorado Trail using a set of switchback.

This portion of the trail has loose rocks and gravel with a steep descent.
Thus, I had to plan for a means of stabilizing myself and my 50 lb. load,
as well and a means for recovery should I slip. In all, I planed my route
and communicated it to someone. I also planned a load that included
one meal, a means for staying dry and warm, trekking poles, 3.0 liters of
water, a means for cutting timber, rope for recovery, sunscreen and so
on. The planning included contingencies for storms and staying on the
ridge overnight.

Planning an analytics project is much the same. You have to consider the
path that your analytic solution will take you down, where you might go
astray, different contingencies, bottlenecks, time, tools required and
how you would finish.

Execution
Executing the hike started with walking up a dirt road for half a mile to
the base of the switchbacks, ascending to the top of the ridge, going
north to the Colorado Trail, the west along the top of the ridge to the
descent, and back to a dirt road for a 1.5 mile walk to the basecamp.

Although I was familiar with the route, rainfall in May had altered he
appearance of the terrain and the path to the Colorado Trail was difficult
to follow, so I had to dead-reckon until I came upon a cairn (a small pile
of rock used to mark a trail). Once I found the trail, there was a gradual
ascent to about 3050 meters. Here I stop for lunch and rested my legs
that felt like jelly. The descent down the Colorado Tail would require
strong legs and concentration. Often going down is much more difficult
than going up, especially if you are carrying a load. Fortunately, I did not
encounter any rain or storms and made it back to camp an hour early.

Project execution is similar. Often the first part of a project is not

exciting, like walking up a dirt road. You have to define requirements
with the client and stakeholder, acquire data, clean the data, explore the
data, and so on. Once you start implementing your analytic approach,

82
you start climbing over rough terrain, attempting to a solution at the
apex of your approach. Once you get your solution, you start your decent
of post-processing and evaluating your solution. Finally you have to
interpret and explain your solution on your way toward project
completion.

Relating your Story

Back at basecamp, I wanted to relate my experience of hiking the
Colorado Trail to numerous people, some of which were neither hikers
nor outdoorsmen. I recognized the fact that some of the people I want
to relate my experience to had taken the route by horseback a few years
back. This knowledge helped by explain what I experience on the train
from a ground perspective. I could talk about key intersections, the
steepest switchback and so on, because I understood the context that
they saw it in. I also talked about feet and miles instead of the metric
measure on my map, using their language. For instance about 10,000
feet, instead of 3050 meters. I did not use terms like trekking poles,
alpine zone, bivouac, bushwhack, cairn, or gendarme. Instead I used
hiking poles, above tree-line, temporary camp, hiking off-trail, small pile
of rocks, and steep-sided rock formation along a ridge. Finally, I allowed
them to fell the weight of my load (backpack).

Relating the analysis story is similar. It almost requires a story teller who
either has never hiked or has hiked and also travel by horseback. The
storyteller has to be in the culture of their listeners and in the proper
context. Obviously, we would not use terms like mean and type I error,
but use average and false positive. We may believe that everyone know
what a “mean” value is, but many in fact do not. We have to allow the
stakeholder to “experience” our analytic journey within their culture and
context.

Conclusion
The two words 'information' and 'communication' are often used
interchangeably, but they signify quite different things. Information is
giving out; communication is getting through.

—Sydney J. Harris
83
If we are not ensuring that our information is getting through to the
client, we not only fail to relate our analysis journey, we also fail to
deliver our analytic solution, for delivery requires both deployment
implementation. Thus, as we hike, or climb, or ride, we also analyze.

84
 12 Ways Not to Please Your
Customer
I guess that "250.5 ways NOT to please your customer", seemed a bit
daunting. Thus, I am going to cut it to the top twelve—the first six are
additions to the original 6 (of 250.5). If you do not have a direct
customer, then interpret customer as boss or supervisor. Also, if your
spouse is your boss—as mine is, but I am not implying she is bossy—then
interpret customer as spouse or significant other, which could be your
Great Dane. I had intended to add a few every week, since there are 1‫א‬
(pronounced aleph-1 and you’ll have to read my post on Jessica Rabbit
and the Number System) ways too NOT please your customer. Now I
think I will just write twelve and most others are variations. I have
committed most of the taboo actions at one time or another, so this is
based on lessons learned by a slow learner.

1. Do not get to work after your customer, unless your customer goes
to work a 2:30 AM, or has scolded you several times for being too
early. I always want to be available to my customer, especially if
they come in early and want to get thing accomplished be people
typically show up for work. Sometimes when they know you are

85
 going to be there for them, they turn to you as someone they can
depend upon.
2. Do not go home before your customer, unless they are a hardcore
workaholic. At least check with your customer before you leave for
the day, to make sure they have no pressing deadlines, or
something just came up that they need help with. While others are
having a beer at Bubba’s Bar, you will be gaining the trust of your
customer. However, do this within reason if you have a family. Your
family is more important than going the extra mile for your
customer.
3. Do not blindly follow the instruction of your customer. Wait, will
that not displease them? Not if you do it selectively and with tact.
You have the responsibility to help you customer do the right thing
and make them successful. It is not your job to shine. It is your job
to make sure them shine. If you want glory, change careers—
become an actor, a singer, a professional quarterback, etc.
Continued business and a paycheck is your reward. Ask questions
that will cause the customer to consider, or reconsider courses of
action, and do so with great humility. “Mam, I know you are
probably right about this, but have you considered X?” In the end
when the customer says execute, do so as I it was your own idea,
unless ethics are involved.
4. Do not let your customer fail due to lack of planning. Make sure
they have (or you have for them) considered redundancy and
contingencies. Remember Charlie Beckwith, always take an extra
helicopter (look up Operation Eagle Claw). Sub-rules: (1) Always
use pencil—things change often, and (2) Do things faster, harder,
and smarter
5. Never work on company business or personal stuff on the
customer’s time. Even if you are not charging, if this occurs during
normal work hours, this could appear as unethical—perception is
everything, and this is a quick way to lose trust, and perhaps your
contract. Sometimes your contract will specify things that you can
do on the customer’s time, like company ethics training. If in doubt
ask—questions do not cost as much as breaches of integrity.
6. Do not withhold information from your customer (unless it is
proprietary and you have signed an agreement saying that you will
not). This is not the same as lying, but it is related, and will destroy
trust just as quickly as lying. Don’t use information as power. Give
them information to leverage as power.
86
7. Do not tell them what they can’t do. Our job is to help do what
they want to do, to achieve what they want to achieve. We have to
help them fulfill their vision. That does not mean that we cannot
advise them about risks and issues—we need to be diligent in
doing this—but we have to also help them mitigate the risks and
find a solution to get there. If it is NASA—well, at least before the
next lunar exploration program was cancelled—then we have to
help them reach the moon.
8. Do not tell them that something they want is out of scope. At least
do not do this directly and without discussing scope with your
contract representative. Going to the moon with a new launch
platform (like the Ares I) is probably not one of those “out of
scope” tasks. However, if they want you to build the Starship
Enterprise, at least without Mr. Scott on your team, and maybe a
Vulcan, then it might be out-of-scope. Often scope can be changed
to help the customer get to here they want to go.
9. Do not tell them they are wrong. Your customer is never wrong—
and if your boss is your spouse and your spouse is your wife, she is
never wrong, unless you have a really nice doghouse in the back
yard. Your customer’s idea, desire, command, etc., may be
misguided (out of alignment with their vision), and our job is to
also help them align tasks, project and programs with their vision
and to provide a roadmap to get there. Now, they may be wrong,
but if you help guide them, they will come to this realization
themselves, usually.
10. Do not—never ever ever—lie to your customer. In fact, never lie to
anyone. It is perfectly okay to say that you cannot reveal
information due to its security classification of proprietary nature,
but never lie. If you did not cross a “t” or dot an “i”, then own up to
it and do it. This will, at a minimum, fulfill your contract obligation
to complete a task, but it may also say to your customer, “this
person can be trusted, because they always tell me the truth.”
Now, you may be incompetent and truthful at the same time, at
the incompetence mat get your fired, but at least you maintain
your integrity. The author of the Piano Player [Kurt Vonnegut] said
this: "Ninety percent of people are incompetent at what they do.
You would be surprised at how really bad they are. If you are
marginally competent, you may go far." Thus, if you a marginally
competent, non-liar, you at least have a chance of pleasing your
customer.
87
11. Do not berate your customer behind their back—or to their side, or
over them, or in a parallel universe. They WILL “hear it through the
grapevine.” I have been a supervisor/manager and as such suffered
the natural phenomenon of beratization—add that to your
dictionary—but sooner or later, I always here about it, even if I
have to sneak around and eavesdrop. Do not think your customer
will not do the same. Nothing good can come from it!
12. Do not mention your customer on social media, even if it is in
praise, unless it is part of a task on your contract to do so. Without
belaboring the point, nothing good can come from it!

So, twelve makes a dozen—maybe I will do the more later—so perhaps

we should title the post “Half-a-Dozen Ways NOT to Please Your
Customer”, but 250.5 just sounds better to me. Moreover, I have
changed my mind about half of a way not to please your customer:

88
 Analytics and Statistics: Is there a
difference?
Is there a difference between statistics and analytics? In previous posts,
I have claimed that there is a difference. Here, I will attempt to explain
my reasoning for approaching this conclusion.

What is Statistics?
Statistics is the study of the collection, analysis, interpretation,
presentation, and organization of data [1]. In applying statistics to, e.g.,
a scientific, industrial, or societal problem, it is necessary to begin with
a population or process to be studied. Populations can be diverse topics
such as “all persons living in a country” or “every atom composing a
crystal”. It deals with all aspects of data including the planning of data
collection in terms of the design of surveys and experiments [1].

Two main statistical methodologies are used in data analysis: descriptive

statistics, which summarizes data from a sample using indexes such as
the mean or standard deviation, and inferential statistics, which draws
conclusions from data that are subject to random variation (e.g.,
observational errors, sampling variation) [2]. Descriptive statistics are
most often concerned with two sets of properties of a distribution

89
(sample or population): central tendency (or location) seeks to
characterize the distribution's central or typical value, while dispersion
(or variability) characterizes the extent to which members of the
distribution depart from its center and each other. Inferences on
mathematical statistics are made under the framework of probability
theory, which deals with the analysis of random phenomena. To make
an inference upon unknown quantities, one or more estimators are
evaluated using the sample.

Standard statistical procedure involve the development of a null

hypothesis, a general statement or default position that there is no
relationship between two quantities. Rejecting or disproving the null
hypothesis is a central task in the modern practice of science, and gives
a precise sense in which a claim is capable of being proven false. What
statisticians call an alternative hypothesis is simply a hypothesis that
contradicts the null hypothesis. Working from a null hypothesis two
basic forms of error are recognized: Type I errors (null hypothesis is
falsely rejected giving a “false positive”) and Type II errors (null
hypothesis fails to be rejected and an actual difference between
populations is missed giving a “false negative”). A critical region is the
set of values of the estimator that leads to refuting the null hypothesis.
The probability of type I error is therefore the probability that the
estimator belongs to the critical region given that null hypothesis is true
(statistical significance) and the probability of type II error is the
probability that the estimator doesn’t belong to the critical region given
that the alternative hypothesis is true. The statistical power of a test is
the probability that it correctly rejects the null hypothesis when the null
hypothesis is false. Multiple problems have come to be associated with
this framework: ranging from obtaining a sufficient sample size to
specifying an adequate null hypothesis.

What is Analytics?
Actually, analytics and statistics share a common thread: they both use
statistical procedures and analyses. However, unlike statistics, the
analytics scientist often deals with analyses where there is no assumed
null hypothesis, and subsequently employs machine learning algorithms
in the analyses. This is an important distinction in understanding my
point of view. For instance, I very seldom approach predictive modeling

90
with a null hypothesis. On the other hand, I do perform statistical
analyses with a null hypothesis at the forefront—there is a subtle but
distinct difference.

I will not go into great detail here, since my last post described Predictive
Analytics, but I will summarize a description of analytics. Analytics is the
discovery and communication of meaningful patterns in data. Especially
valuable in areas rich with recorded information, analytics relies on the
simultaneous application of statistics, machine learning, computer
programming and operations research to quantify performance or
predictions. Analytics often favors data visualization to communicate
insight [3].

Firms may commonly apply analytics to business data, to describe,

predict, and improve business performance. Specifically, arenas within
analytics include predictive analytics, enterprise decision management,
retail analytics, store assortment and stock-keeping unit optimization,
marketing optimization and marketing mix modeling, web analytics,
sales force sizing and optimization, price and promotion modeling,
predictive science, credit risk analysis, and fraud analytics. Since
analytics can require extensive computation, the algorithms and
software used for analytics harness the most current methods in
computer science, statistics, and mathematics [4].

I emphasized the use of optimization in analytics applications because it

is a method used in operations research, not in statistics. Optimization
may be performed either mathematically or heuristically. The heuristic
approach often “violates” the principals of mathematics (in terms of
proving them out) in arriving at a “close enough” solution. As a student
years ago, I worked under a department head that said, “This is really
good stuff, but it is not mathematics.” Moreover, many of the
applications of analytics are not stochastic—not based on probability.

Conclusion
If you accept the descriptions of statistics and analytics that I have
presented here, then there is an obvious difference between the two
disciplines, thought they do share some things. This is not meant to
belittle the work of the statistician. The statistician is very professional,

91
understands their discipline extremely well and provides valuable,
focused analyses. While the statistician is a “focused” of “specialized”
discipline, analytics is much more general—the analytics professional is
to some degree a “jack of all trades”. Often, the analytics professional
will turn to the statistics professional for help with analyses where
statistical hypotheses are paramount.

References
Dodge, Y. (2006) The Oxford Dictionary of Statistical Terms, OUP. ISBN 0-
19-920613-9

Lund Research Ltd. "Descriptive and Inferential Statistics".

https://statistics.laerd.com/statistical-guides/descriptive-
inferential-statistics.php. Retrieved 2015-01-26.

Strickland, J. (2014) Predictive Modeling and Analytics. Lulu. ISBN 978-1-

312-37544-4.

Strickland, J. (2015) Predictive Analytics using R. Lulu. ISBN 978-1-312-

84101-7.

92
 Are Statisticians a Dying Breed?

“Facts are stubborn things, but statistics are pliable.”

― Mark Twain

I am not a statistician, though I perform statistical analyses almost every

day; I am an Operations Research Analyst. I have seen a lot of discussion
regarding the title of this article, and as someone who does not have a
comprehensive background in statistics, I would say loudly, “I need
statisticians”.


“Every time I sit with our general manager at a baseball game, and
there's number-cruncher and statistician guy - I'm sitting around - they
start talking about stuff, and I say, 'What's that? I've never heard of
that one before.'”

― George Brett

People have been espousing the idea that statisticians have no role in
analytics and data science. Almost all of what I do in predictive analytics
is statistical in nature. One of the most important statistical procedure is

93
design of experiments, both for collecting the data and for performing
the analysis. Statisticians are very good at this.


“To consult the statistician after an experiment is finished is often
merely to ask him to conduct a post mortem examination. He can
perhaps say what the experiment died of.”

― Ronald Fisher

Statisticians do more than regression analysis, hypothesis testing and

analysis of variance (ANOVA). More often than not, statisticians are
cleaning data, performing data reduction and other prepossessing
functions. Today, the statistician does more than they have in the past
and their service is crucial to developing unbiased information to
decision makers.


“Definition of a Statistician: A man who believes figures don't lie, but
admits than under analysis some of them won't stand up either.”

― Evan Esar

Can statisticians do everything that is required in analytics and data

science? For the most part, probably not, but neither can I. I am sure
there are a few super-humans out there that can, but these fields require
multi-discipline teams in order to provide a holistic approach. Depending
on the analyses at hand, teams should be comprised of statisticians,
psychologists, marketing professional, data architects, operations
research analysts, and so on.


“The greatest value of a picture is when it forces us to notice what we
never expected to see.”

– John Tukey

We need to change the conversation from who is not needed to who is

needed and why. The fray at hand is counter-productive.

94
 
“I couldn't claim that I was smarter than sixty-five other guys--but the
average of sixty-five other guys, certainly!”

― Richard P. Feynman, Surely You're Joking, Mr. Feynman!:

Adventures of a Curious Character

95
 Statistics is Obsolete

Experts often possess more data than judgment.

— Colin Powell

Hey, we no longer need statistics! Cancel all college and university

statistics programs…fire all the statistics faculty…shut down SAS…stop
wasting time on R and Python…the end has come…there is a new
beginning…ALL DATA.

Why do we no longer need Statistics?


Being deeply knowledgeable on one subject narrows one's focus and
increases confidence, but it also blurs dissenting views until they are no
longer visible, thereby transforming data collection into bias
confirmation and morphing self-deception into self-assurance.

― Michael Shermer

I am sure you have heard of BIG DATA, but have you heard of ALL DATA?
Probably not, because I think I may just invented the term. Apparently

96
there are some major players who actually believe that when they have
"all of the data", they no longer have to take samples, and that statistics
only applies to samples. Since we do not need to take samples anymore,
then we do not need statistics anymore. By the way, Statisticians, we do
not need you anymore. Go find a nice job as a greeter at Walmart, not
that there is anything wrong with Walmart greeting—I kind of like the
special touch.

What’s wrong with this Picture?


Not everything that can be counted counts, and not everything that
counts can be counted.

— Albert Einstein

First, the CEO’s, CTO’s, and other O’s that are espousing this view
absolutely have no earthy idea (or one from a parallel universe) of what
they are talking about. I would call them naïve, but they are really
somewhere between “so full of themselves” and just plain oblivious to
reality! I think their issue is their lack of understanding statistics,
statistics anxiety, or a combination of the two. But, ALL DATA will take
care of that, since it will not require any statistics.

Second, ALL DATA does not equal PERFECT DATA. I deal with colossal
data sets. In my experience roughly 80% of the data is useless when
building a model of some phenomenon, like the propensity to buy
product X or the probability that the O-ring on component Y of the latest
space launch vehicle will fail. Moreover, regarding the 80% of useless
data, about half of that is really bad data. Of the twenty percent that is
useable, only about 10% of it actually contains strong predictors. And
this stuff is pretty easy to figure out using Weight of Evidence (WOE) and
Information Value (IV) algorithms. But never fear, ALL DATA has
morphed into ALL-PERFECT DATA.

Third, I think it is great that we have ALL-PERFECT DATA, but how are
you going to process it? I know, you’ll just push a button get a prediction.
And inside that “black box” there are no statistics involved. There is no
regressing, no averaging, no transforming, no imputing, on so on. The

97
data is just speaking for itself. Oh wait, we have gone from ALL-PERFECT
DATA to something much better, TALKING DATA.


One person's data is another person's noise.

― K.C. Cole

Fourth, this implies we do not have to

analyze the results. I suppose that is due to
our new TALKING DATA. It speaks the
results itself! No more histogram, means,
variances, Skewness, etc., because we do
not care. Our TALKING DATA will tell us
what decisions to make. Wait a minute…if
our TALKING DATA is making decisions,
then it is really THINKING DATA. So now we
do not need any decision makers! Fire all
the CEO’s, CTO’s, and other O’s. Wait
another minute! Who is going to
implement the decisions made by our THINKING DATA? I know, it’s the
MAGICAL DATA (probably in the form of unicorns). Wait another minute,
if we have MAGICAL DATA, we do not need anyone else, for magic can
do it all. Layoff everyone. No one ever needs to work again, thanks to
MIGICAL DATA!

98
Fifth, I do not really have a fifth so I am going to ask my BIG, ALL,
PERFECT, TALKING, MAGICAL DATA for its opinion. What was that? Our
UNICORN said, "all those people who think we do not need statistics are
not learned!" Now, do not get angry with me. I did not say that. It came
for the UNICORN you created…

How do we fix the Picture?


Statistics are like bikinis. What they reveal is suggestive, but what they
conceal is vital.

— Aaron Levenstein

That is an easy one, fire all the brainless CEO’s, CTO’s and other O’s who
wouldn’t know a statistic if it slapped them in the face. But that would
not work. Who would create the UNICORN?

99
 Math, Physics and Chemistry are
Obsolete
I mentioned in “Statistics are Obsolete” that there are some pretty
influential people who reject statistics as the underlying science of data
analysis with respect to big data, or in their words ALL DATA (specifically,
“We have all of the data.”). I think the essence of the issue for those
people is that simply do not understand statistics, or for that matter any
science that has “formulas”, like math, physics and chemistry.

The ALL DATA proponents seem to cluster with those who have softer
quantitative degrees or degrees in management, psychology, and so on.
Of course, cluster analysis is obsolete, as are principle component
analysis, regression analysis and so on (and probably any kind of analysis
that involves using the scientific method). So just ignore the clustering.

If the ALL DATA people prevail, then I suppose Math, Physics and
Chemistry are obsolete, as well. Therefore, let’s be proactive and shut-
down all the mathematics, physics and chemistry programs. All you
math, physics and chemistry majors, you should switch majors to
anything that does not require mathematics. All you tenured professors
out there, well, you just lost your tenure, but I hear MacDonald’s is
hiring.

100
Parents, stop teaching your kids how to count. I know it’s cute, but it’s
also futile in light of ALL DATA. We do not need anything numerical
anymore. In fact, we do not need anyone to think anymore. These
absolutely brilliant ALL DATA supporters and their THINKING DATA will
think for us (they are not really brilliant, just cleaver).

Bummer, ten years of formal mathematics, physics and chemistry

education is down the drain. Does that call for a toga party or a road
trip?

So I believe I have narrowed down the requirements for one work in

Data Science or Big Data Analytics with our new ALL DATA:

 Obtain a High School Diploma with at least a 2.0 grade point

average
 Take at least one Coursera class in anything except math,
physics, chemistry or statistics
 Obtain a bachelor’s degree in Under–Water Weaving or
equivalent
 Score at least 25 on your SATs
 Have at least one year experience doing just about anything

If you exceed these requirements, then you are over-qualified. After all,
with the possession of ALL DATA you just need a chimpanzee to push a
button on a black box and predict anything you care to predict, because
it is in the data.

Could I interest you in some ocean-front property in Arizona?

101
Part V – Data Science Concerns

Data Scientists are Dead, Long Live

Data Science!
I know, that is pretty confusing. What we have, however, is a distorted,
all-over-the-place, inadequate, confusing, and too broad a definition of
“data scientist”. I do not know where we went wrong, but we have the
equivalent of “bridge scientist” if you happen to work on a bridge,
whether you design it, build it, resurface it, and who knows, maybe just
drive across it.

Call it what it is
I think that life would be better in all things data if we just called thing
what they are, like calling mathematicians by their function: number
theorist, math teacher, mathematical modeler, algebraist, etc. How
about datatician? Will that cause confusion with statistician?

102
Mimic Mathematical Sciences?
Maybe “data science” is okay and the problem is “data scientist”. We
have “mathematical sciences” (note the plural), but few mathematicians
would call themselves a “mathematical scientist”, even though that
sounds pretty cool. The problem is it doesn’t describe what one does in
the field of mathematics. Algebraist, however, says that one works in the
mathematical area of algebraic structures. The mathematical sciences
might looks like this (I borrowed these from the National Science
Foundation):

 Algebra
 Number Theory
 Analysis
 Applied Mathematics
 Combinatorics
 Computational Mathematics
 Foundations
 Geometric Analysis
 Mathematical Biology
 Probability
 Statistics
 Topology

This is not an all-inclusive list. Note that statistics falls under

mathematical sciences at the NSF. And “real” statisticians know from the
mathematical statistics courses why this might be the case.

103
A Data Science Taxonomy?

Perhaps under “data science”, we eliminate the term “data scientist”

and call out titles by what we do:

 Data architect
 Data processor
 Data miner
 Data modeler
 Data analyst
 Data explorer
 Database administrator
 Database developer
 Algorithm developer
 etc.

Of course, the guru of data analysis and data modeling is the

“statistician”. But perhaps statistician can be divided in like manner, as
the mathematician: regression analyst, exploratory data analyst, etc.

Fun with Machine Learning

What about machine learning?

104
  Machine learning architect
 Neural networker
 Classification tree farmer
 Random forest ranger
 (now I am just being silly)

Conclusion
Somehow, “data scientists does not cut it for me, and there are many in
the industry and the field that have similar issues. I have read and ben
told that many organization will no longer hire data scientist, partly
because there is no standard for what a data scientist is or does. I do not
think we will reach a standard, so let’s label a person who works with
data by what they actually do. Then hire them accordingly.

105
 So you think you are a Data
Scientist?
Many are writing articles about the skills required to be a data scientist.
I disagree with most of them. Before you can be a data scientist, you
must qualify to be a scientist. A degree in management does not qualify
you to be a Chemical Scientist or Mathematical Scientist, nor does it
qualify you to be a Data Scientist. A degree in Management Science or
Operations Research could qualify you, depending on the course
content.

This article represents my opinion and will hopefully add substance to

an ongoing dialog regarding what constitutes a data scientist. Also, there
is nothing derogatory about being a data analyst, data processor, data
architect, database administrator, business analyst, marketing analyst,
analytics consultant and so on, but doing these jobs do not make you a
data scientist. Neither does a few courses from Coursera.

106
This guy might qualify to be a data scientist

In general, scientists perform research toward a more comprehensive

understanding of nature, including physical,
mathematical and social realms. Scientists
are also distinct from engineers, those who
design, build and maintain devices for
particular situations. When science
is performed with a goal toward practical
utility, it is called applied science. An applied
scientist may not be designing something in
particular, but rather is conducting research
with the aim of developing new technologies
and practical methods.

If you don’t understand these formulas, you might not be a data

scientist.

The following are requirements for scientists from a General Services

Administration (GSA) Schedule related to machine learning
(www.gsa.gov/), for instance (this is the closest I could find for data
science). Many who call themselves Scientists (i.e., Data Scientists) are

107
not qualified to be an Assistant Scientist. Though I may be qualified to
be a Principal Data Scientist, I do not use the title. Instead, I use Big Data
Analytics Consultant, being careful not to misrepresent myself.

Assistant Scientist I/II

 BS in mathematics, science, engineering or equivalent
 One to three years research experience or equivalent
course/project work
 Strong foundation in object-oriented analysis and design
 Knowledge of and experience with a variety of software
languages and platforms
 Demonstrated research skills
 Demonstrated ability to understand appropriate algorithms and
concepts
 Demonstrated ability to write clear, grammatically-correct
documents
 Demonstrated ability to effectively communicate and work with
other members of project team
 Strong analytical skills
 Strong fit with Company goals and values (work ethic, attitude,
interest in work)

Scientist I/II
 MS (or BS plus three years of experience) in mathematics,
science, engineering or equivalent
 Two to five years research experience or equivalent
course/project work
 Demonstrated object-oriented analysis and design skills
 Knowledge of and experience with a variety of software
languages and platforms
 Proven research skills
 Proven ability to understand appropriate algorithms and
concepts
 Demonstrated ability to employ a hierarchical writing process to
generate clear, grammatically correct documents
 Demonstrated ability to provide technical guidance to interns
and other more junior team members
108
  Strong analytical skills
 Strong fit with Company goals and values (work ethic, attitude,
interest in work)

Senior Scientist I/II

 PhD (or MS plus four years of experience) in mathematics,
science, engineering or equivalent
 Three to six years research experience or equivalent
course/project work
 Strong object-oriented analysis and design skills
 Knowledge of and experience with a variety of software
languages and platforms
 Strong research skills proven by publication record or equivalent
references from the research community
 In-depth experience with appropriate technologies
 Strong ability to employ a hierarchical writing process to
generate clear, grammatically-correct documents
 Two years of experience leading teams
 Demonstrated program management skills
 Strong analytical skills
 Strong fit with Company goals and values (work ethic, attitude,
interest in work)

Principal Scientist I/II

 PhD (or MS plus five-years of experience) in mathematics,
science, engineering or equivalent
 Four plus years research experience or equivalent
course/project work Strong object-oriented analysis and design
skills
 Knowledge of and experience with a variety of software
languages and platforms
 Superior research skills proven by extensive publication record
or equivalent references from the research community
 Expert knowledge of appropriate technologies
 Demonstrated ability to write winning proposals
 Three years of experience managing teams
 Strong analytical skills
109
  Strong fit with Company goals and values (work ethic, attitude,
interest in work)

Regarding the education requirements for mathematics, science,

engineering or equivalent, the disciplines for "equivalent" are not
mentioned. However, in my experience these would generally include
statistics.

No wonder companies who are

hiring data scientist are being
disappointed in their
performance and contribution.
Too many people are watering-
down the requirements. As a
community of professionals, we
need to define the role of data
scientist in terms of
requirements, rather than trying
to be all inclusive for people who
work with data. Moreover, the requirements might possible vary from
country to country, so we should look at common requirements. For
instance a Master’s degree in Economics or Informatics in Europe in not
necessarily the same as one in the United States. What really matters in
the end is the scientist's role in turning data into information and
information into "useful" information. Oh, and you should have one of
those white lab coats...wireframe glasses are optional.

110
 5 Signs that you might be a Data
Scientist
Why are people so down on Data Scientists? Probably because they
don't know what one looks like. There are a lot of folks out there calling
themselves Data Scientist just because they had some introductory
statistic courses, and maybe they can make really cool charts in Excel. All
that makes you is someone who had some introductory statistic courses
and can make some really cool charts in Excel. So what might a real Data
Scientist look like? Here are some signs that you might "really" be a Data
Scientist.

1. You are a scientist. In order to be a data scientist, you must first be

a scientist. The Government Services Administration (GSA) defines a
scientist as a subject matter expert who: “Provides technical
knowledge and analysis of highly specialized applications and
operational environment, high-level functional systems analysis,
design, integration, documentation and implementation advice on
exceptionally complex problems that require graduate level
knowledge of the subject matter for effective implementation.”
Thus, you might write charts diagrams and formulas on a chalk board
and contemplate them for hours.

111
2. You have a graduate degree in a science or related field, i.e.,
Analytics, Predictive Modeling, Risk Analysis, Applied Statistics,
Database Architecture, etc. Most occupations where the title
“Scientist” is used, requires a doctorate degree. I am taking the
liberty to soften the requirement to a Masters of Science degree.
Sorry, a MBA does not make the cut. Alternatively, you professional
certification like Certified Analytics Professional (CAP).
3. See 1 and 2. Moreover, you include some or all or the following skills
in your profile: Data Scientist, Analytics Scientist, SPSS Modeler, SAS
Programming, Regression Analysis, Statistical Modeling, Logistics
Regression, Machine Learning, SQL, Python, etc.
4. See 1 and 2. Additionally, your boss or customer listens to you and
takes action when you present your analysis and recommendations.
Moreover, profit was made or money was saved (or for non-profits
some measure of performance was achieved) based on your work.
5. See 1 and 2. Not only can you crunch numbers with the best of them,
you can logically present the results of your analysis in plain
language, or at least your customer’s language. Additionally, you
make contributions to the body of knowledge through scientific
experimentation or studies, resulting in blogs, articles, papers, or
presentations.

The real scientists worked very hard to become one; they did not just
attach a label to their title. That does not mean that only scientists work
with data. Beside data scientists, other people who work with data may
include data analysts, data consultants, data architects, data miners,
analytics consultants, and so on. However, to be a data scientist you
need the prerequisites outlined in 1 and 2. For hiring managers, project
managers, human resource professional, etc., when you hire a data
scientist, make sure you actually hire a scientist who works with data.
Otherwise, do not call them "scientists". You might be less disappointed
in their work.

112
 How can I be a Data Scientist?
Several aspiring data scientist and analytics scientist have asked me how
to achieve their knowledge and training goals. Though there are many
ways to accomplish this, below are two options offered through
Coursera.

Johns Hopkins University – Data Science

 The Data Scientist’s Toolbox

 R Programming
 Getting and Cleaning Data
 Exploratory Data Analysis
 Reproducible Research
 Statistical Inference
 Regression Models
 Practical Machine Learning
 Developing Data Products
 Data Science Capstone

University of Illinois at Urbana-Champaign – Data Mining

 Pattern Discovery in Data Mining

 Text Retrieval and Search Engines

113
  Cluster Analysis in Data Mining
 Text Mining and Analytics
 Data Visualization
 Data Mining Capstone

SAS and Hadoop for Learning

If you need SAS for training or self-study, you can get the university
addition through the link below. You do not have to be student or
faculty, but you cannot use it for work.

 http://www.sas.com/en_us/software/university-edition.html

If you need to use hadoop for training try:

 http://hortonworks.com/products/hortonworks-
sandbox/#install

114
Why you might not want to be a Data
Scientist
Data Scientist was one of the world’s most popular technical jobs, at
least in 2014. From what I have been told, of all the major countries, it
was second only in the United States. Universities everywhere are
creating, or have already created, graduate programs in Data Science.
Data is becoming more and more available and in large quantities.
Recently, President Obama appointed a US Chief Data Scientist (Dr. DJ
Patil). Data science is at center stage!

So, why would you not want to be a data scientist? The following reflects
my opinion only, and should not be construed as advice, merely
something to consider.

Saturation
We are putting more and more data scientists into the field every day.
Personally, I am asked several times a month to consider filling a data
scientist role here or there. Today, there are ample opportunities for
data scientists. It may be different tomorrow. If you are thinking about
becoming a data scientist is some way, shape or form, especially if you
have just started college or will start soon, consider the idea that the job

115
market may be saturated by the time you are qualified. Ask yourself if
there are alternative courses of study and course of action.

Internet of Things (IoT)

I think the jury is still out deliberating guilt or innocence. The Internet of
Things is a difficult concept to define precisely. In fact, there are many
different groups that have defined the term, although its initial use has
been attributed to Kevin Ashton, an expert on digital innovation. Each
definition shares the idea that the first version of the Internet was about
data created by people, while the next version is about data created by
things. In 1999, Ashton said it best in this quote from an article in the
RFID Journal:

If we had computers that knew everything there was to know about

things - using data they gathered without any help from us - we would
be able to track and count everything, and greatly reduce waste, loss and
cost. We would know when things needed replacing, repairing or
recalling, and whether they were fresh or past their best.

Data without a human touch? I do not know, nor does my crystal ball
show me, a glimpse of the impact that IoT could have, but it is something
to consider.

Imposters
I am sure I will get in trouble here, but it would seem to me that before
you can be a data scientist, you have to first be a Scientist. The terms
Scientist, Engineer, Analyst, and so on, have specific meanings. For
instance, the Government Services Administration defines a scientist as
a subject matter expert who:

 Provides technical knowledge and analysis of highly specialized

applications and operational environment, high-level functional
systems analysis, design, integration, documentation and
implementation advice on exceptionally complex problems that
require graduate level knowledge of the subject matter for
effective implementation.

116
  Applies principles, methods and knowledge of the functional
area of capability to specific task order requirements, advanced
mathematical principles and methods to exceptionally difficult
and narrowly defined technical problems in engineering and
other scientific applications to arrive at automated solutions.

They define an Engineer a subject matter expert who:

 Provides technical knowledge and analysis of highly specialized

applications and operational environment, high-level functional
systems analysis, design, integration, documentation, and
implementation advice on exceptionally complex problems that
necessitate high-level knowledge of the subject matter for
effective implementation.
 Participates as needed in all phases of software development
with emphasis on the planning, analysis, modeling, simulation,
testing, integration, documentation and presentation phases.

They list the following occupational specialties separately:

 Data Architect
 Data Communication Manager
 Data Warehousing Administrator
 Data Warehousing Analyst
 Data Warehousing Programmer
 Data/Configuration Management Specialist
 Database Analyst/Programmer
 Database Manager/Administrator

If you are a hiring manager, watch out for imposters. If you are an
aspiring data scientist, make sure you are qualified before you make
your claim. Imposters have the potential for damaging the reputation
and expectations of the occupation.

Alternatives
There are really some cool alternatives to becoming a data scientist and
still being involved with data science. I do not call myself a data scientist.
Instead, I claim to be a "predictive analytics scientist". However, by prior

117
training and occupation, I am really an Operations Research Analyst (or
Scientist). Operations Research (OR) has been around since World War
II and OR analysts did things like plan the Normandy invasion (Operation
Overlord). The International Forum for Operations Research and
Management Science (INFORMS) now sponsors the Certified Analytics
Professional (CAP) program. Many universities offer degree program or
concentrations in OR. ORs work for cruise lines, airlines, the
entertainment industry, the wholesale and retail sales industries, the
trucking and train industries, the automotive industry, manufacturing
industry, and the departments of Defense, Transportation and others. In
addition to data "stuff", they perform optimization, predictive
modeling, inventory control, and simulation.

OR Analyst (along with other disciplines like Statistician) may be a viable

alternative to Data Scientist.

Count the Costs

Whatever you consider doing, evaluate the potential cast and outcome.
If you do pursue becoming a "real" data scientist, and all of my "points
to consider" actually come to pass, you will still be a highly skill
individual. It might not be a great leap to acquire some additional
knowledge and become an OR Analyst or an Actuary, for instance.
However, if you are unwilling or unable to achieve the requirements for
becoming a scientists, then perhaps data analysts or data miner would
be appropriate fields. Just think before you leap...

118
 I am an Analyst, really!
Many people say they are analysts, as I do. But are we really analysts?
Do we perform analysis? That is, do we take problems and break them
into smaller pieces to solve, and then develop an aggregate solution.
Analyst is a term for someone who uses an analytic approach to problem
solving. The opposite problem solving approach is intuition.

Analysts come is a variety of flavors: intelligence analysts, operations

research analysts, risk analysts, text analysts, and so on. But all analysts
have this in common: they use the analytic approach, breaking large
problems down into smaller ones and solving them. Someone who uses
the alternative approach is not an analyst, unless intuition is used in
conjunction with the analytic approach.

Definition: an analyst is an individual who uses the analytic approach to

problem solving, in any discipline.

I am confronted regularly with "fake analysts", who typically do not

realize they are posing as analysts in title only. If an employer needs an
analyst, then they should seek out the real ones, and save those who are
not for other positions.

119
I have often written about data scientists and what actions and skills
they manifest most. However, not all Data Scientist are analysts--they
may not be engaged in problem solving. At any rate, it should be clear
that Analytics is strongly correlated to the problem solving process.

So, who are the real analysts, the ones we may be looking to hire, who
approach problem-solving by disaggregating problems into their smaller
parts? These are people who have specific training and experience,
either qualitative or quantitative, in the science and logic of a field that
many are calling Analytics.

Analytics is often used synonymously with quantitative modeling, like

statistical modeling, mathematical programming, simulation, and so on.
However, qualitative modeling such as text analysis, is just as important
in Analytics as its quantitative cousin.

There does exist, however, the situation where the intuitive problem
solvers call themselves analyst, and this may not necessarily be the case.
There is a need for this, but not in Analytics.

On the other hand, some who are statisticians or mathematicians call

themselves analysts when "real problem solving" is not their forte. For
instance, describing or characterizing data is important, but it is not
Analytics if it is not performed as part of the problem solving process.

You could face a situation where you have a sociologist and a statistician
both claiming to be analysts. Suppose the latter routinely characterizes
data and then passes that information to someone else for problem
solving. We could argue that the statistician is not the analysts.

With the sociologist, we might be confronted with someone who takes

social problems, say unemployment, breaks them into their component
parts, and then solves each piece though non-quantitative means. We
could argue that this person is an analyst.

Before you start throwing analyst or analysis on your resume, ask

yourself if you are an integral part of problem solving possess using the
analytic method. If you are not, do not call yourself an analyst, rather
describe what you actually do.

120
 All Things Data
Data Scientist, Analytic Scientist, Statistician, Operation Research
Analyst, Predictive Modeler, … They are all very different… or are they?

We often focus on differences to delineate our profession and what we

can bring to the analytical table to feast upon. Yes there are differences,
but let’s try something novel. Let’s focus on the similarities. What do we
all have in common? Data!

Is it a surprise that the Administration has appointed a Chief Data

Scientist (congratulation Dr. Patil), or is it more surprising that it took so
long? Everyone else has a chief data scientist or a “chief someone who
uses data to do good stuff with”. After all, it doesn't really matter that
you have a bunch of data, or that you have a fancy title. What matters is
whether you can turn that data into useful information that is used and
results in making the business, the industry, the government and/or the
country better off tomorrow than it was yesterday.

Ask not what your data can do for you; ask what you can do with your
data. After all, to turn that data into useful information, you have to
scientisize it, or statisticize it, or researchacize it, or modelize it (all of

121
these are new words now). You might even have to analyze it, or
synthesize it, perform some other –izes upon it. And is that not what
statisticians, modelers, analysts, scientist who work with data do from
day to day?

So, instead of “how to become a data scientist” and “statisticians are not
data scientists”, why not “we are all a bunch of professionals trying to
turn tons of data into useful, applicable and actionable information and
this is how you can join us”?

I do not label myself as a data scientist, but as an analytic scientist. But

day-to-day I live in data, trying to make sense of it, trying to manipulate
it, cleanse it, impute it, transform it and model it. Is the statistician doing
anything much different? Or is the operations researcher, or the data
scientist? No, we are all Hadooping, and SASing, and R-ing, and
Tableauing, and Pythoning, and other –ings every day to a boatload (a
very big boat like Royal Caribbean’s Oasis of the Seas) of data. I work
with all of these disciplines every day and I see us all doing the same
stuff, maybe from a different angle. And who’s to say what the best
angle is, except all of them combined. After all, it does take eight 45%
angles to make a circle, at least in this solar system.

Like Alan Turing, who had to get his machine to determine millions of
different combinations of Enigma settings to help in the fight against
Nazi Germany, so we too have to sift through millions of combinations
of variables to find useful information that will aid in our fight against
stupidity. That’s right, we have been stupid for too long. Regardless of
what CNN or FOX News tells you, we are not doing just fine. Look around
your neighborhood, or mine. We need to set aside our differences and
join our collective intellects to make this enormous amount of data that
is available to us, work for us and make us better tomorrow than we are
today.

If you are expecting our new Chief Data Scientist to solve all of our ills,
don’t hold your breath. The Department of Education didn’t work, other
than give jobs to people who have very little talent for getting things
done; the House and the Senate isn’t doing anything but fighting a
standoff and keeping the status quo; the IRS isn’t doing anything other
than than taking your hard earned dollars and redistributing them. Let’s
just fire them all and give Dr. Patil an army of “professionals who do
122
great stuff with data.” Well, that’s not going to happen, but still the
answer may be in All Things Data.

Python plus R equals Data Scientist?

I have read a number of articles and posts that claim that if you had
courses in linear algebra and multivariate calculus, and know R and
Python, then you are a data scientist. That sounds really good. I have
courses in embryology, histology, and physiology, so I must be an OBGYN
doctor—I also stayed in a Holiday Inn Express last night.

Other things I have read indicate that data scientists are carrying a
blemish which causes them not to be needed. No wonder! It appears as
if I give $10 to Vincent ‘Vinnie’ Antonelli [Steve Martin in My Blue
Heaven], then he can make me a data scientist. I have talked to
recruiters who have been disappointed in the actual qualifications of
some who label themselves a data scientist.

Are these people real? Two college courses plus R and Python? Although
my view may not be popular, a monkey can learn to use tools. Tools do
not make you anything other than a tool user. A deeper understanding
of data structure, data analysis, data modeling, data … is what one needs

123
to be Data Scientist. Conceptual understanding, not tools. If you have
the former, you can pick up the latter in no time.

However, I will go on to say that conceptual understanding is not

enough. When I see the data scientist depicted in cartoons or on TV, I
see a reclusive geek with no social skills and a propensity to immerse
themselves in data at the expense of all else. But that is not a data
scientist.

The term “scientist” refers to someone who is an expert in their science

and uses the scientific method. As a profession the scientist of today is
widely recognized. Scientists include theoreticians who mainly develop
new models to explain existing data and predict new results, and
experimentalists who mainly test models by making measurements —
though in practice the division between these activities is not clear-cut,
and many scientists perform both tasks.

Then what makes a data scientist? I would say that coursework leading
to a degree, conceptual understanding, a sprinkle of research, a dash of
new development, an occasional bath, plus “effectiveness” makes a data
scientist. When a client uses the product produced by the data scientist,
whether it be a database, a data architecture, a data model, or data
analysis, then we have “effectiveness”, and they person who did all of
this is a data scientist.

If you are a pilot of a Boeing 757 and you cannot land the aircraft (it may
be a skill you never acquired), then you are not a pilot. You are just an
airline employee with wings on your chest. Landing the plane is pretty
important, and “landing” your solution with the big data customer is
pretty important. When you talk as a data scientist, then clients should
listen and take action based on what you tell them.

This is where we are missing the boat. We can ‘python-ate’ or ‘statistic-

ate’ [two new verbs] all day long, but at the end of the day when the
client is no longer listening, then what are we?

124
 Why your client might not be
listening
What is the Problem?
To effectively communicate, we must realize that we are all different in
the way we perceive the world and use this understanding as a guide to
our communication with others.

—Anthony Robbins

Have you ever worked your fingers to the bone developing a model or
other analytic solution to a problem for your client or department, and
then observed that your solution was never used? Is the boss or
customer listening to your advice? If the answers are Yes and No,
respectively, then there is a problem, but it is a common problem. There
a number of root causes for this, the the most common is
communication.

Data Analyst Jargon

As big data analysts, we have our own language. This is sometimes called
jargon. We understand it, but other do not. Others may include
125
managers, hiring managers, intelligence analysts, chief technology
officer, etc. When we talk they are probably nodding their head and do
not have a clue what we are saying.

The two words 'information' and 'communication' are often used

interchangeably, but they signify quite different things. Information is
giving out; communication is getting through.

—Sydney J. Harris

Client Interaction
There is a group on YouTube called Studio C (If you have not watched
them before, you are missing out on some good comedy). I watched one
yesterday called Poke Face,
https://www.youtube.com/watch?v=XQ6_GdODuww. Each player is
showing their poker face and we get to hear each one of their thoughts.
One guy does not know the game at all—not the rules, the poker chips,
etc.—and when it is time to show his hand, he is showing UNO cards.
However, he kept his poker face the whole time. This is your client. Just
because you are an outstanding analyst, providing excellent analytic
solutions, does not guarantee that people will listen to you. You have to
speak their language.

Language is Important
Don't use words too big for the subject. Don't say infinitely when you
mean very; otherwise you'll have no word left when you want to talk
about something really infinite.

― C.S. Lewis

There was a certain Christian missionary who was trying to reach a

certain native tribe in a remote part of the world with his message. The
tribe killed him. Later, the tribe was converted to Christianity, and it was
not until then they understood what they had done was wrong. There
had been a communication gap of several layers. One was language, one
was culture, and another was belief system. The tribal chief is your client.

126
You have to speak his language, understand his culture, and recognize
his belief system.

A Real Example - Sort of

In the land of Gibberish, the man who makes sense, the man who
speaks clearly, clearly speaks nonsense.”

― Jarod Kintz, This Book Has No Title

I said this is a real example, but it is really a real example with the client
using their jargon with the analyst, rather than the other way around.
Recently, a colleague asked for some advice for an upcoming interview.
The telephone interview was with the CTO and this is what was taken
away from that interview. In answering what kind of projects would be
expected, the CTO said that the following could be expected

1. Message testing
2. Spent-time model of customers that are at risk and at grow
3. Atomization of models/algorithms/etc.

He is what I draw out of the CTO’s description:

1. Message testing probably involve constructing an acquisition

model, like a propensity to buy or engage. These models are
typically logistic regression models. Once you build a model, say
in SAS, you can deploy it and make operational on a main frame
for example. The the client could run the model on a regular
basis. The output would be a score of customers’ propensity to
buy or engage. Using the score file, they would market/message
the top deciles. This way they are spending marketing dollars on
those whom the message would have the most likelihood of a
response.
2. Customers at risk is also a propensity model: the propensity to
shed. This is also a logistic regression model and the same
principles from above apply, except the scores would be high for
people who are at risk.
3. Automation could take a number forms, but the most common
is model deployments as I described above. For example, when

127
 I build a model in SAS Enterprise Miner, I take the score code
and wrap it in an execution macro and then give that code to a
guy who deploys it for production on a mainframe. The same
guy runs the model each month along with about 20 other
models during the first week of the month. The process is
automated and the scores are sent to a marketing directory for
access. He also runs a model performance report on each model,
which is partially automated. An emerging process is real-time
modeling. The same models would be used, but they might be
written in Python in a Hadoop environment and the data would
be refresh a regular short time periods. This could be used to
direct traffic to call center. For example, if a caller has a high
score for propensity to shed a mortgage, that customer could be
directed to the mortgage company and provided some incentive
not to shed their mortgage.

Conclusion
The ability to simplify means to eliminate the unnecessary so that the
necessary may speak.

—Hans Hofmann

The CTO is speaking a different language. To understand the CTO’s

language you must first understand his culture, belief system and
context. Not knowing these, I am guessing at my interpretation above.
However, if you interview for a job you should understand these things
first. The same is true for our communication with clients. We must
understand their culture, belief system—their corporate belief system,
not religious—the relative context, and then their language they
understand. If we do not do these things first, our analysis results land
on deaf ears and we fail. Our solutions are neither understood nor
implemented.

128
 Part VI - Applications

Big Data Analytics and Human

Resources?
The ability to capture and analyze big data is recognized by many
enterprises as an accelerator. It has enabled many enterprises to
increase revenues by better understanding, more accurately target
customers and cut costs through improved business processes.

Big data also has attracted the attention of enterprise managers and
their human resource (HR) managers. Many believe that they can now
analyze mountains of structured and unstructured data to answer
important questions regarding workforce productivity, the impact of
training programs on enterprise performance, predictors of workforce
attrition, and how to identify potential leaders. For the purpose of this
article, HR will refer to the enterprise function and human resource will
refer to the humans who are resources for the enterprise.

129
Hogwash!
The HUMAN in HR barely exists as it is. Now I hear that HR is going to
identify leaders through big data analytics (I am trying not to laugh too
hard). You would think that one would need to know what defines a
leader, but with all the “junk” articles I read on leadership, most people
including HR do not know what leadership is! Many use management
and leadership synonymously, and they are mistaken.

What is Analytics anyway?

Analytics is slightly better than an educated guess and much better than
a random choice. I am an analytics scientist, making a living from
performing analytics, but I declare that is not a “fool-proof” science.
Good big data analytics is based on the quality data. However, big data
does not imply good data. In fact, most of the so-called big data I have
worked with is very poor data. Quantity does not equal quality, so stop
saying that big data is the solution to every problem that arises, including
HR problems.

Root Causes.
I am not picking on HR. This is a common problem across disciplines, and
I know many competent, caring and professional HR personnel. Having
stated that, I often use a quote from Kurt Vonnegut, author of Player
Piano:

“Almost nobody’s competent, Paul. It’s enough to make you cry to see
how bad most people are at their jobs. If you can do a half-assed job of
anything, you’re a one-eyed man in the kingdom of the blind.”

Analytics does not make incompetent people more competent—

incompetent people performing analytics produces “garbage”! If you
want to improve HR, then put the HUMAN back into HR—transform HR
personnel into those competent, caring and professional ones I know
and work with.

130
Analytics also does not make bad products better, poor services better,
and so on. It may provide insight into why products and services are
perceived as being poor, but the only way to fix these problems is to
“make a better product” and “provide better services”.

Where did the Human go?

I remember a time when HR was different than it is today. I have not
performed analytics on HR, so I would be guessing as to what happened
to it, how it became less human. So, for now I will guess. Human
resources implies managing humans who are resources for your
enterprise. There was also a time when enterprises (in the US) where
flourishing, having fewer constraints in terms of capital resources,
regulations, and so on. Somewhere around 2007 that changed. Human
resources (the resources, not HR) became surplus in an over-saturated
market. Instead of struggling to find good talent to fill a growing
enterprise, HR seems to have been forced into constraining resources.
What was already in the pipeline was a resource, and everyone else was
a potential element that would break the pipeline. So, guarding the
enterprise became more important than managing its resources, and HR
became more involved in the decision making of enterprises than should
have been allowed. The service-oriented HR personnel became
constrainers, and the HUMAN disappeared. Now we treat human
resources (the humans who are the resource) as dollars or paper and
pencils.

Now, with the perception that HR is “broken” or that something in the

enterprise related to human resources is diminished, we aim to fix it with
analytics… Perhaps we should find the HUMAN and put it back in HR first.

What can Analytics do?

I have mentioned what analytics cannot do. Here are some ways that
analytics could help the enterprise and its HR functions:

 Recruitment cost per hire

 New hire failure factors
 Employee turnaround rates
 E-learning abandonment rates
131
  Bonus compensation rates

In spite of the fact that I have been downplaying the need for analytics
in HR, when we know these things kinds of things, we can actually put
the HUMAN back in HR. I know, it is a paradox. The point is do not use
big data analytics to fix HR, rather, use it to enhance HR functionality. If
an enterprise feels it needs to fix its HR, then it should fix its HR before
adding big data analytics to the sauce. If HR is broken, adding analytics
is just going to produce more of the same sauce, and you will not like its
flavor.

Conclusion
HR is an important function for the enterprise, unless you replace
humans completely with machines, robots and androids. Improving the
enterprise’s HR function is a worthy undertaking and big data analytics
may help improve its functionality, but only if it is already functional!

132
 Call Center Analytics: What's
Missing?
What is the Problem?
My customer frequently asks if this model or that model can be used to
direct call center traffic. Usually the models are projecting an acquisition
or engagement window with too much of a gap to perform this function,
and this is hardly part of their use case as acquisition models. However,
I can perceive a model that does perform this function. Yet a model is
not entirely what they need.

Models seem to be the magic “fix-all” in the Financial Services and

Insurance Industry (FSI). Customers think that models will fix their
product, marketing strategy and call center. The reality is: anything that
pertains to analytics seems to be a model to them. That is an article for
another day. For this one, I want to talk about the call center problem,
namely, directing calls more efficiently.

A model might be the answer they are looking for. However, you cannot
just build a model and leap into its use blindly, without testing. The best
way to test in this situation is using discrete event simulation.

133
What is Discrete Event Simulation?
Discrete Event Simulation (DES) is the process of codifying the behavior
of a complex system as an ordered sequence of well-defined events. In
this context, an event comprises a specific change in the system's state
at a specific point in time (arrival at the bank, service by a teller, etc.).
Rather than stepping based on a time increment, like every second, DES
advances based on events—events that may or may not be equally
spaced in time.

Common applications of DES include:

 stress testing ( a form of deliberately intense or thorough testing

used to determine the stability of a given system or entity)
 evaluating potential financial investments
 studying call center operations
 modeling procedures and processes in various industries, such
as manufacturing and healthcare
 studies that support system design
 studying reliability, availability and maintainability (RAM)
 just about anything that involves queuing (arrivals, waiting time,
service time)

Can We Simulate the Call Center?

One of the most applicable situations that lends itself to DES is the
operations of a call center. When a customer calls, you have an arrival
event. If they have to wait for a representative they enter a queue until
such time as their call is answered, which begins a service event. If their
call was rerouted upon entering the system–the initial arrival event–the
rerouting creates a new arrival event for the call center to which it was
transferred, and so on. One can study where bottle-necks occur,
rerouting efficiency, waiting times, balking (customers deciding not to
join the queue if it is too long), reneging (customers leave the queue if
they have waited too long for service), etc.

An effective DES process must include, at a minimum, the following

characteristics:

134
  Predetermined starting and ending points, which can be discrete
events or instants in time (arrivals and departures, for instance).
 A method of keeping track of the time that has elapsed since the
process began (waiting time, for instance).
 A list of discrete events that have occurred since the process
began (begin service, for instance).
 A list of discrete events pending or expected (if such events are
known) until the process is expected to end.
 A graphical, statistical, or tabular record of the function for
which DES is currently engaged (plot of waiting times and service
times, for instance).

Using DES, we can test the model with multiple parameter, or test
multiple models, without disrupting the operations of a call center. Once
we determine which model (or which parameters) work the best, we can
then use that particular model to direct call, for instance. Continued
monitoring and used of the DES for simultaneous parallel testing can be
used to maintain and improve efficiency.

135
Win, Lose or Draw…Is this a Game?
What economists call game theory psychologists call the theory of social
situations, which is an accurate description of what game theory is
about. Although game theory is relevant to parlor games such as poker
or bridge, most research in game theory focuses on how groups of
people interact. There are two main branches of game theory:
cooperative and non-cooperative game theory. Non-cooperative game
theory deals largely with how intelligent individuals interact with one
another in an effort to achieve their own goals. That is the branch of
game theory I will discuss here.

In addition to game theory, economic theory has three other main

branches: decision theory, general equilibrium theory and mechanism
design theory. All are closely connected to game theory.

Decision Theory
Decision theory can be viewed as a theory of one person games, or a
game of a single player against nature. The focus is on preferences and
the formation of beliefs. The most widely used form of decision theory
argues that preferences among risky alternatives can be described by

136
the maximization of the expected value of a numerical utility function,
where utility may depend on a number of things, but in situations of
interest to economists often depends on money income. Probability
theory is heavily used in order to represent the uncertainty of outcomes,
and Bayes Law is frequently used to model the way in which new
information is used to revise beliefs. Decision theory is often used in the
form of decision analysis, which shows how best to acquire information
before making a decision.

General Equilibrium Theory

General equilibrium theory can be viewed as a specialized branch of
game theory that deals with trade and production, and typically with a
relatively large number of individual consumers and producers. It is
widely used in the macroeconomic analysis of broad-based economic
policies such as monetary or tax policy, in finance to analyze stock
markets, to study interest and exchange rates and other prices. In recent
years, political economy has emerged as a combination of general
equilibrium theory and game theory in which the private sector of the
economy is modeled by general equilibrium theory, while voting
behavior and the incentive of governments is analyzed using game
theory. Issues studied include tax policy, trade policy, and the role of
international trade agreements such as the European Union.

Mechanism Design Theory

Mechanism design theory differs from game theory in that game theory
takes the rules of the game as given, while mechanism design theory
asks about the consequences of different types of rules. Naturally this
relies heavily on game theory. Questions addressed by mechanism
design theory include the design of compensation and wage agreements
that effectively spread risk while maintaining incentives, and the design
of auctions to maximize revenue, or achieve other goals.

137
Example
One way to describe a game is by listing the players (or individuals)
participating in the game, and for each player, listing the alternative
choices (called actions or strategies) available to that player. In the case
of a two-player game, the actions of the first player from the rows, and
the actions of the second player the columns, of a matrix. The entries in
the matrix are two numbers representing the utility or payoff to the first
and second player respectively.

The Prisoner’s Dilemma (PD): The problem can be traced back to von
Neumann and Morgenstern [von Neumann, 1944] and, of course, John
Nash [Nash, 1953], you and a friend have committed a crime and have
been caught. You are being held in separate cells so that you cannot
communicate with each other. You are both offered a deal by the police
and you have to decide what to do independently. Essentially the deal is
this.

 If you confess and your partner denies taking part in the crime,
you go free and your partner goes to prison for ten years.
 If your partner confesses and you deny participating in the
crime, you go to prison for ten years and your partner goes free.
 If you both confess you will serve six years each.
 If you both deny taking part in the crime, you both go to prison
for six months.

What will you do? The game can be represented by the following matrix
of payoffs

Note that higher numbers are better (more utility). If neither of you
confesses, you both go free, and split the proceeds of their crime which
we represent by 5 units of utility for each suspect. However, if one of
you confesses and the other does not, the one who confesses testifies

138
against the other in exchange for going free and gets the entire 10 units
of utility, while the one who did not confess goes to prison and which
results in the low utility of -4. If both of you confess, then both are given
a reduced term, but both are convicted, which we represent by giving
each 1 unit of utility: this is called mutual defection.

Mutual defection is a Nash equilibrium of PD [it is a crucial concept in

non-cooperative games, and it won Nash the Nobel Prize in economics
in 1994]. Informally, a set of strategies is a Nash equilibrium if no player
can do better by unilaterally changing his or her strategy. Mutual
defection is the unique Nash equilibrium of PD, which means that it is
the only stable solution to this game. In real world scenarios, however,
a Nash equilibrium is not necessarily played. Some conditions to
guarantee that the Nash equilibrium is played are:

1. The players aim to maximize their own payoffs.

2. The players know the Nash equilibrium strategy of all players.
3. The players believe that a deviation in their own strategy will not
cause deviations by any other players.
4. There is common knowledge that all players know these
conditions.

This game has fascinated game theorists for a variety of reasons. First, it
is a simple representation of a variety of important situations. For
example, instead of confess/not confess we could label the strategies
“contribute to the common good” or “behave selfishly.” This captures a
variety of situations economists describe as public goods problems. An
example is the construction of a bridge. It is best for everyone if the
bridge is built, but best for each individual if someone else builds the
bridge. This is sometimes referred to in economics as an externality.
Similarly this game could describe the alternative of two firms
competing in the same market, and instead of confess/not confess we
could label the strategies “set a high price” and “set a low price.”
Naturally it is best for both firms if they both set high prices, but best for
each individual firm to set a low price while the opposition sets a high
price.

A second feature of this game, is that it is self-evident how an intelligent

individual should behave. No matter what a suspect believes his partner
is going to do, it is always best to confess. If the partner in the other cell
139
is not confessing, it is possible to get 10 instead of 5. If the partner in the
other cell is confessing, it is possible to get 1 instead of -4. Yet the pursuit
of individually sensible behavior results in each player getting only 1 unit
of utility, much less than the 5 units each that they would get if neither
confessed. This conflict between the pursuit of individual goals and the
common good is at the heart of many game theoretic problems.

A third feature of this game is that it changes in a very significant way if

the game is repeated, or if the players will interact with each other again
in the future. Suppose for example that after this game is over, and the
suspects either are freed or are released from jail they will commit
another crime and the game will be played again. In this case in the first
period the suspects may reason that they should not confess because if
they do not their partner will not confess in the second game. Strictly
speaking, this conclusion is not valid, since in the second game both
suspects will confess no matter what happened in the first game.
However, repetition opens up the possibility of being rewarded or
punished in the future for current behavior, and game theorists have
provided a number of theories to explain the obvious intuition that if the
game is repeated often enough, the suspects ought to cooperate.

References
Nash J. Two-Person Cooperative Games. Econometrica 21: 128-140,
1953.

von Neumann J., Morgenstern O. Theory of Games and Economic

Behavior. Princeton NJ: Princeton Univ. Press, 1944.

140
 Part VII – The Power of
Operations Research

What is Operations Research?

Many people probably never heard the phrase “Operations
Research” used. Operations Research (OR), or operational research in
the U.K, is a discipline that deals with the application of advanced
analytical methods to help make better decisions. The terms
management science and analytics are sometimes used as synonyms for
operations research. Yet, in my experience OR extends far beyond
either. The figure shows a hierarchy of operations research activities,
and I’ll let you decide if they are also performed in analytics.

 Data Mining and Machine Learning

 Artificial Intelligence and Expert Systems
 Financial Engineering
 Games, Decision, and Strategic Planning
 Marketing Research
 Investment Science

141
  Experimental and Engineering Design
 Manufacturing and Production
 Logistics and Transportation
 Supply Chain Management
 Enterprise Resource Planning

Modeling Systems and Optimization Services is an interface part that

bridges OR modeling with OR tools. When implemented smoothly, it is
the part that is not noticed by modelers or users.

Where do you find them?

Operation Research Analysts work in many industries, including
maritime, space operations, defense, airlines, train lines, financial
service, entertainment and many more. Wherever operation occur,
operations research analysts are usually there. The following lists several
key functional areas for operations research analysts.

 Communications
 Interfaces
 Networks

142
  Scheduling
 Routing
 Manpower
 Modeling

What are their tools?

Underlying Tools is the level that is typically regarded as what uniquely
defines Operations Research.

 Mathematical Programming
 Computing Technology
 Probability and Statistics
 Stochastic Simulation
 Systems Analysis
 Organization Theory
 Accounting Principles
 Engineering Economics
 Decision Analysis
 Game Theory
 Heuristics
 Computer Programming
 Numeric Methods
 Stochastic Analysis
 Queuing Theory
 Evolutionary Algorithms
 Dynamic Programming

What are they built upon?

Foundations upon which OR are built include:

 Mathematical Theory
 Statistical Theory
 Computing Theory
 Economic Theory

143
What is their history?
Operational Research was born during the early year of WWII and
matured rapidly. One of its primary function was the planning of
Operation Overlord or the Normandy Invasion. It has its foundations in
mathematics, computing and economic theories, on which basic tools in
optimization and simulation are built. Today OR’s are employed by
airlines, train lines, logistic systems, delivery systems (e.g., FedEx),
defense systems, military, oil companies, insurance companies, financial
institutions, manufacturing, marketing and many more.

144
 Getting the Question Right
As an analysts and modeler I have a variety of customers with a variety
of problems in business. We usually frame their “question” as a business
case. The question of course is something they require an answer for it
usually stems from a problem or a desire to have more Share of Wallet
or other metric. What is interesting here is though they know they have
a question that needs to be answered, they often do not know how to
state their question.

In "Making Analysis Relevant", Mr. Vince Roske reiterated a recipe

provided by John D. (Dave) Robinson, MG USA, Ret, when he was the
Director of Joint Staffs J-8. The recipe is provided below:

 What’s the question?

 What’s the “real” question?
 What do the final slides look like?
 What do I already know?
 How do I get the remaining information that I need?

So, this idea of what is the real question is not new and it is not unique
to business.

145
As an analyst, we have to draw that real question out through dialogue
with our customers. Doing this is often as much an art as it is a science.
If we do not do it, we stand a good chance delivering a well-built solution
that happens to be the wrong one. I have done this at least twice in the
last three years, knowing that getting the question right was paramount.
In spite of my failures I have found some keys to help with formulating
the business case.

1. Tie it to a metric. Even if currency is not at stake, you have to have

a metric to focus on. Otherwise you can fail to see the forest for the
trees. As your project rolls along you will forget what you are
building to without a target. It can be Return on Investment, Share
of Wallet, Key Performance Parameter, or something else, but you
need a metric.
2. Do not mention the solution. Do not state your analytic solution
method in the business case. The customer may want a model and
a model may be the right analytic solution for answering the
question, but most customers do not understand what models do
and don’t do. The business case should focus on the
question/problem, not the solution. If you allow it to, the customer
will start building your model for you, without the expertise to do
so. I have seen this happen.
3. Use timing. Guide the customer to include timing in their business
case, i.e., achieve metric A with X months, detect behavior B at least
Y months in advance, and so on. With a timing “device” you can help
the customer realize the temporal nature of an analytic solution. On
one hand, the solutions do not work forever, but many believe
otherwise. On the other hand, solutions apply only to events and
behaviors in a specified time period. Behavior farther out from the
phenomenon or event are quite different from behavior closer to it.
4. Tie it to the business unit. That might sound like a no-brainer, but I
have seen business units responsible for web activity or call center
activity want to tie the business case to a product. If you are trying
to improve call center operations, you are trying to improve call
center operations, period. Now, you are probably doing so with
product sales as a downstream goal, but you cannot put the cart
before the horse. Fix your call center problem and then ask a new
question.

146
I am sure there is a lot more that can be said about this subject, but this
is what I have learned over time and I am just one analyst. The fact
remains that if we do not get the “real question”, then we may not
provide the “right solution.”

There are other interesting pieces of Roske’s Recipe I did not address
here. Perhaps another article?

Reference
Mr Vincent P. Roske, Jr., “Making Analysis Relevant,” PHALANX Volume
31, Number 1, March 1998

147
 Holistic Analysis and Operations
Research
Introduction
I am not really sure what holistic analysis is, so I will define it. Our English
word comes from the Greek ὅλος (holos, meaning “all, whole or entire”).
Reductionism may be viewed as the complement of holism.
Reductionism analyzes a complex system by subdividing or reduction to
more fundamental parts. For businesses, knowledge and know-how,
know-who, know-what and know-why are part of the whole business
economics. Having a holistic view keeps us from missing the forest due
to the trees.

148
Operations Research (OR), or operational research in the U.K, is a
discipline that deals with the application of advanced analytical methods
to help make better decisions. The terms management science and
analytics are sometimes used as synonyms for operations research. Yet,
in my experience OR extends far beyond either. The list below is a
collection of operations research activities – I’ll let you decide if they are
also performed in analytics.

 Data Mining and Machine Learning

 Artificial Intelligence and Expert Systems
 Financial Engineering
 Games, Decision, and Strategic Planning
 Marketing Research
 Investment Science
 Experimental and Engineering Design
 Manufacturing and Production
 Logistics and Transportation
 Supply Chain Management
 Enterprise Resource Planning

History of Operations Research

Operational Research was born during the early year of WWII and
matured rapidly. One of its primary functions was the planning of
Operation Overlord or the Normandy Invasion. It has its foundations in
mathematics, computing and economic theories, on which basic tools in
optimization and simulation are built. Today OR’s are employed by

149
airlines, train lines, logistic systems, delivery systems (e.g., FedEx),
defense systems, military, oil companies, insurance companies, financial
institutions, manufacturing, marketing and many more.

What do Operations Research Analysts do?

The Operations Research Analyst is a jack-of-all-trades (one guy's
opinion), or at least that has been my experience. Some tend to
specialize in a particular area, like mathematical optimization, but I think
this is a mistake. One can find people who specialize in a particular
methodology or discipline, but they would be challenged to find a good
Operations Research analyst without a holistic view of the problem
space.

Figure 1. The OR Problem Space (I made this up yesterday)

An OR’s view of the problem space is really what defines them and
describes what they do. The list above displayed some of the activities
that ORs engage in, but not without a holistic view of the problem space.
Figure 1 depicts the entire problem space. Mathematically, we could
look at it like this:
150
({(Analysis Space)⊂Research Space}⊂Operations Space)⊂Problem
Space

The OR Analyst must enter the problem space with the following in
mind: (1) the potential operational domains, (2) the types of research
that may be used, and (3) the types of analyses that may be appropriate.
If one goes in having done nothing more than math programming for 10
years, that analyst is NOT an operations research analyst—they are just
a math programmer.

Operations research analysts provide this holistic view, which then

allows for the definition of the right problem within any domain, and
application of the most appropriate research methodology, using the
most appropriate analyses. You cannot build a house with just a
screwdriver, unless you are MacGyver[1].

If we look at the historical context of OR, as we discussed earlier, we

should be able to ascertain that anything short of a holistic point of view
may have resulted in operational chaos much worse than missed
dropped zones. Operation Overlord—the most complex operation ever
executed—could have easily failed.

Reference
[1] MacGyver is an American action-adventure television series created
by Lee David Zlotoff. Henry Winkler and John Rich were the executive
producers. The show follows secret agent Angus MacGyver, played by
Richard Dean Anderson, who works as a troubleshooter for the fictional
Phoenix Foundation in Los Angeles and as an agent for a fictional United
States government agency, the Department of External Services (DXS).
Resourceful and possessed of an encyclopedic knowledge of the physical
sciences, he solves complex problems with everyday materials he finds
at hand, along with his ever-present duct tape and Swiss Army knife.

151
 Part VIII - Tools

What are R Users Saying?

Partial Results of the 2011 Survey, Copyright © 2011 Rexer Analytics

The following is a excerpt from 5th Annual Survey (2011) conducted by

Rexer Analytics. This is a partial list of insights from R users. The full
survey results can be found by following the survey link above. The
following is entirely the work of Rexer Analytics. The surveys they
conduct contain more categories, such as Overcoming Data Mining
Challenges, Analytic Success Measurement and The Positive Impact of
Data Mining. I have included the more substantial comments. The
details of the most previous years (2012, 2013 and 2014) are not yet
available.

"Over the years, respondents to the 2007-2011 Data Miner Surveys have
shown increasing use of R. In the 5th Annual Survey (2011) we asked R
users to tell us more about their use of R. The question asked, “If you
use R, please tell us more about your use of R. For example, tell us why
you have chosen to use R, why you use the R interface you identified in
the previous question, the pros and cons of R, or tell us how you use R
in conjunction with other tools.” 225 R users shared information about

152
their use of R. They provided an enormous wealth of useful and detailed
information. Below are the verbatim comments they shared."

 Best variety of algorithms available, biggest mindshare in online

data mining community, free/open source. Previous con of 32-bit
in-memory data limitation removed with release of 64-bit. Still
suffers some compared to other solutions in handling big data. Stat-
ET plugin gives best IDE for R, using widely used Eclipse framework,
and is available for free. RapidMiner add-in that supports R is
extremely useful, and increasingly used as well.
 The main reason for selecting R, back in the late 1990’s, was that it
had the algorithms I needed readily available, and it was free of
charge. After getting familiar with R, I have never seen a reason to
exchange it for anything else. R is open-source, and runs on many
different system architectures, so distributing code and results is
easy. Compared to some latest commercial software I’ve evaluated,
R is sluggish for certain tasks, and can’t handle very large datasets
(mainly because I do not have a 64-bit machine to work with). On
top of that, to be really productive with R, one needs to learn other
languages, e.g., SQL, but that’s just how things are. Besides,
knowledge of those other languages is needed anyway.
 I’ve migrated to R as my primary platform. The fact that it’s free,
robust, comprehensive, extensible, and open. More than offsets its
shortcomings – and those are disappearing over time anyway. I’m
using Revolution’s IDE because it improves my productivity. R has a
significant learning curve, but once mastered, it is very elegant. R’s
shortcomings are its in-memory architecture and immature data
manipulation operations (i.e. lack of a built-in SQL engine and
inability to handle very large datasets). Nevertheless, I use R for all
analytic and modeling tasks once I get my data properly prepared. I
usually import flat files in CSV format, but I am now using
Revolution’s XDF format more frequently.
 Why I use R: Initial adoption was due to the availability of the
randomForest package, which was the best option for random forest
research, as rated on speed, modification possibility, model
information and tweaking. Since adopting R, I have further derived
satisfaction from the availability of open source R packages for the
newest developments in statistical learning (e.g. conditional
inference forests, boosting algorithms, etc.). I use the R command
line interface, as well as scripting and function creation, as this gives
153
 me maximum modification ability. I generally stay away from GUIs,
as I find them generally restrictive to model development,
application and analysis. I use R in conjunction with Matlab, through
the statconnDCOM connection software, primarily due to the very
slow performance of the TreeBagger algorithm in Matlab. Pros of R:
Availability of state-of-the-art algorithms, extensive control of
model development, access to model information, relatively easy
access from Matlab. Cons of R: Lack of friendly editor (such as
Matlab’s Mlint; although I am considering TinnR, and have tried
Revolution R); less detailed support than Matlab.
 I employ R-Extension in the Rapid Miner interface. I use R for its
graphing and data mining capability. I use R for its ability to scrape
data from different sources (FTP, ODBC) and implement frequent
and automated tasks Pros: -Highly customizable -Great potential for
growth and improved efficiencies -Fantastic selection of packages –
versatility -Growing number of video tutorials and blogs where users
are happy to share their code. CONS: -x64 not fully integrated yet -
Steep learning curve -Limited number of coding and output
examples in package vignettes -Difficult to setup properly
(Rprofile.site examples are scarce online) -Memory constraints with
i386 32-bit -Output – reporting design limitations suitable for the
business environment.
 I have about 12 years experience with R. It’s free, the available
libraries are robust and efficient. I mostly work with the command
line, but I am moving towards R-Studio because it’s available both as
a desktop application and a browser-based client-server tool set. I
occasionally use Rcmdr. The only other tool I use as heavily as R is
Perl – again, lots of experience with it, it’s free, and there are
thousands of available library packages.
 The analytics department at my company is very new. Hence
we haven’t yet decided which analytics-tool we’ll be using. R is a
great interim solution as it is free and compatible with a reasonable
amount of commercial analytics tools. The reason I use the standard
R GUI and script editor is simply because I haven’t invested the time
in trying out different GUI’s, and I haven’t really had any reason to.
The advantage of using R (at least compared to SAS, which was the
analytics-tool at my old job) is mainly that you have much more
control over your data and your algorithms. The main problem with
R is that it can’t really handle the data sets we need to analyze.

154
 Furthermore, your scripts can tend to be a bit messy if you are not
sure what kind of analysis or models you are going to use.
 I use R for the diversity of its algorithms, packages. I use Emacs for
other tasks and it’s a natural to use it to run R, and Splus for
that matter. I usually do data preparation in Splus, if the technique
I want to use is available in Splus I will do all the analysis in
Splus. Otherwise I’ll export the data to R, do the analysis in R, export
results to Splus where I’ll prepare tables and graphs for
presentations of the model(s). The main drawback to R, in my
opinion, is that R loads in live memory all the work space it is linked
to which is a big waste of time and memory and makes it difficult to
use R in a multi-users environment where typical projects consist of
several very large data sets.
 We continue to evaluate R. As yet it doesn’t offer the ease of use
and ability to deploy models that are required for use by our
internationally distributed modeling team. “System” maintenance
of R is too high a requirement at the moment and the enormous
flexibility and range of tools it offers is offset by data handling
limitations (on 32 bit systems) and difficulty of standardizing quick
deployment solutions into our environment. But we expect to
continue evaluation and training on R and other open source
tools. We do, for instance, make extensive use of open source ETL
tools.
 I use R extensively for a variety of tasks and I find the R GUI the most
flexible way to use it. On occasion I’ve used JGR and Deducer, but
I’ve generally found it more convenient to use the GUI. R’s strengths
are its support network and the range of packages available for it
and its weaknesses are its ability to handle very large datasets and,
on occasion, its speed. More recently, with large or broad datasets
I’ve been using tools such as Tiberius or Eureqa to identify important
variables and then building models in based on the identified
variables.
 I’m using R in order to know why R is “buzzing” in analytical areas
and to discover some new algorithms. R has many problems with
big data, and I don’t really believe that Revolution can effectively
support that. R language is not mature for production, but really
efficient for research: for my personal researches, I also use SAS/IML
programming (which is for me the real equivalent for R, not
SAS/STAT). I’m not against R, it’s a perfect tool to learn statistics,
but I’m not really for data mining: don’t forget that many techniques
155
 used in data mining comes from Operational Research, in
convergence with statistics. Good language, but not really
conceived for professional efficiency.
 R is used for the whole data loading process (importing, cleaning,
profiling, data preparation), the model building as well as for
creating graphical results through other technologies like
Python. We use it also using the PL/R procedural language to do in-
database analytics & plotting.
 Utilize R heavily for survey research sampling and analysis and
political data mining. The R TextMate bundle is fantastic although
RStudio is quickly becoming a favorite as well. Use heavily in
conjunction with MySQL databases.
 I use R in conjunction with Matlab mostly, programming my
personalized algorithms in Matlab and using R for running statistical
test, ROC curves, and other simple statistical models. I do this since
I feel more comfortable with this setting. R is a very good tool for
statistical analysis basically because there are many packages
covering most of statistical activities, but still I find Matlab more easy
to code in.
 I greatly prefer to use R and do use it when working on more
“research” type projects versus models that will be reviewed and
put into production. Because of the type of work that I do, SAS is the
main software that everyone is familiar with and the most popular
one that has a strong license. Our organization needs to be able to
share across departments and with vendors and governmental
divisions. We can’t use R as much as I would like because of how
open the software is – good for sharing code, bad for ensuring
regulators that data is safe.
 Main reasons for use: 1) Strong and flexible programming language,
making it very flexible. 2) No cost, allowing me to also having it on
my personal computer so that I can test things at home that I later
use at work. I use RODBC to get the data from server and let the
server do some of the data manipulation, but control it from within
R. Have also started to use RExcel with the goal as using that as a
method to deploy models to analysts more familiar with Excel than
R.
 Personally I find R easier to use than SAS, mostly because I am not
constrained in getting where I want to go. SAS has a
canned approach. I see using GUI’s as a “sign of weakness” and as
preventing understanding the language at its core. I have not found
156
Rattle to be particularly helpful. I have also tried JGR and Sciviews
and found I could not surmount the installation learning curve. Their
documentation did not produce a working environment for me.

157
 Why You Might Use SAS
I write a good bit of content about using open-source tools for analytics
and operations research. However, my workhorse happens to be SAS.
Actually, I use SAS Enterprise Guide (EG) and SAS Enterprise Miner (EM).

When to use open-source

I have argued for the use of open-source tools, but they have their place.
If you have a limited budget, open-source is a good path to journey
down. Also, if you are teaching or are a student at a university, open-
source seems like a logical option.

How I use SAS

I perform predictive modeling in the Financial and Insurance (FSI)
industry. My method requires me to use SAS EG to retrieve variables
from multiple data sets (between 10 and 20 sometimes), resulting in
about 2000 variables. Once these variables are merged into one data set
(for up to 11 million customers), I run an information value algorithm to
determine which variables have the most predictive power for the
response variable. After eliminating variables that cannot be used for
marketing bank products (fair lending acts and so on), I import between
150 and 300 variables into SAS EM. In EM, I perform data partitioning,

158
data imputations, and data transformations prior to running a model,
like logistic regression. When I get an adequate model, I take the scoring
code from EM and port it to EG, wrapped in a macro. I run the macro in
EG to measure model performance. When I get a model that performs
well enough against challenger models, I develop model production
code in EG, which includes the EM scoring code.

Why I use SAS

SAS EG and EM are trusted by the FSI that I service. I actually began using
SAS in 1990 on a mainframe computer. It has been the statistical
software of choice by many industries for many years. With external
model governance and validation, the models I have developed in SAS
EG and EM have been stress tested and proven valid for use by the FSI.
This does not imply that models created with open-source tools are any
less valid, as I will discuss next.

Open-source and SAS

Once I build a predictive model in SAS, I attempt the build the same
model in R, for example. The models in R are similar enough to show that
the models in SAS are indeed valid models. For instance, I constructed
an uplift model in SAS using logistic regression and an uplift model for
the same acquisition situation using random forest in R. The overall net
lift was identical, although the distribution among deciles was slightly
different. Anyone can use SAS EG for learning, regardless of your status
(non-student, student, professor, professional, etc.) at
http://www.sas.com/en_us/software/university-edition.html.

Conclusion
I will not argue that either SAS or open-source tools are better than the
other. Instead I will state that they each have their use in various
situations. Yet, SAS is tried and true…and I will continue to use both.

159
 What is BOARD BEAM ?
Is it a beam supporting a roof? Is it a board made from Canadian timber?
Is it capable of leaping a tall building in a single bound? No, it is a tool
for performing analytics! BOARD is the company and software platform
and BEAM is a software module: BOARD Enterprise Analytics Modeling
(BEAM).

Disclaimer
I am not a spokes-person, customer, or user of BOARD BEAM, but I think
that it is worthy enough of our attention to write about it here.

What is it?
You could probably compare BOARD BEAM most closely with Tableau,
but I will not do so here. I'll just tell you what I have seen with BOARD
BEAM. At first glance, it integrates advanced and predictive analytics
with business intelligence and performance management. It's
capabilities range from analytics reporting to predictive analytics using
time series models. It is also easy to use. I have spoken with project
managers who use it for powerful business insights and take action, and

160
data analyst who use it for reporting medical procedure information in
healthcare to insurance providers.

BOARD BEAM was conceived and developed in collaboration with IDSIA,

a Swiss research institute recognized amongst the top 10 world-players
in the Artificial Intelligence space. It is an answer for two key business
needs. First, the necessity to embed more advanced analytical and
predictive capabilities into companies’ business processes ensuring the
agility necessary to operate at the rapid pace of today’s businesses.
Second, is the need to do this in a simpler way, allowing business users
to take the lead without having to depend on an army of experts as
necessary with the traditional data-mining and analytics products to
manually build analytical data models. Of course, that could potentially
leave me unemployed!

What does it do?

BOARD BEAM provides a solution that covers many analytical areas
through three different modules: Predictive Analytics, Clustering and
Analytical Functions.

BOARD BEAM possesses automated predictive modelling to execute

extremely accurate forecasts in a fraction of the time compared to
traditional analytic solutions. Using time-series models for smooth,
intermittent and discontinuous Time-series data.

You can automatically group your customers, products etc. into clusters
and immediately use them as analysis dimensions in your Business
Intelligence environment. Using k-means as its basis, clustering can be
performed with a few clicks of the mouse. There is no coding involved,
and if you can spell "K-means" correctly, you can probably perform it as
well.

It provides Out-of-the-Box statistical functions to increase the value of

your business analyses, reports, and dashboards and to build new
powerful analytical models. These functions range from the traditional
min/max, average, standard deviation, to algorithms specifically
designed for business analysis such as frequency, recency, dormancy
and nascency.

161
How can I get it?
You can test-drive BOARD BEAM by visiting
http://www.board.com/us/.

162
 R, SPSS Modeler and Half-Truths
The Ad
"Please join us on May 26th, 2015 at 9am PDT for our latest Data Science
Central Webinar Event: 7 Reasons to combine SPSS Statistics and R
sponsored by IBM."

The Claim
The ad does on to claim: “According to the Rexer survey,* R is the
analytic software of choice for data scientists, business analysts, and
data miners in the corporate world. Despite R's popularity, adoption of
R has lagged due to a few limitations like:”

 Deployment - Using R to integrate predictive outputs into an

operational environment can be difficult.
 User Interface - R does not have a modern graphical user
interface, which makes it difficult for those who are not R
programmers to use it.
 Learning curve - R is not easy to learn for everyone. Not
everyone is a programmer.

163
  Data Complexity - R does not easily connect to databases
natively.
 Output - Production of publish-ready output is difficult.
 Performance & Scalability - R can very quickly consume all
available memory.
 Collaboration - R makes sharing work among an analyst team
difficult, especially when team members do not have the same
level of R knowledge.
 Enterprise security - The security of the packages that you
download is not assured.

The Survey
I posted partial results of Rexer survey yesterday on bicorner.com and
today on LinkedIn. I have read the entire survey of about 225 responses.
I would not draw the same conclusions entirely, unless I worked for IBM.

IMB’s Motive?
It is of course in IBM’s best interest to draw unsupported conclusions to
support their product, SPSS Modeler. However, what they do not tell you
is that SPSS Modeler is not that user friendly either. Modeler alone lacks
functionality that could be provided by SPSS Stats, but you have to pay
extra for that. Consequently, building a model for measuring the
information value of variables is cumbersome in SPSS Modeler, but is
quite easy in SAS Enterprise Guide.

How is SPSS Modeler?

Admittedly, SPSS Modeler is a nice tool and I do use it from time to time;
however, I prefer SAS Enterprise Miner (EM) over it. SAS EM is more
intuitive to use and the available documentation is much better than
what is available for SPSS Modeler. Formerly Clementine, IBM did not
improve documentation when they acquired the tool.

164
Half-Truths
What bothers me about this is the half-truths that are espoused by the
ad (and to me half of a truth is the equivalent of a lie). You could read
the results of the survey and use them to justify purchasing a red
Mustang convertible just as easily. This is not to say that there are not
some fundamental issues with relying on R solely for your analytics. But,
some of their “bold” conclusions are not accurate. For instance, “R is not
easy to learn for everyone. Not everyone is a programmer,” is true;
however, to do anything other than very basic modeling with SPSS
Modeler, you have to write scripts, and the scripting language is more
difficult that R’s programming language. Plus the documentation on the
scripting language is almost non-existent.

Open-Source versus Commercial

Moreover, the statement, “R makes sharing work among an analyst
team difficult, especially when team members do not have the same
level of R knowledge,” is also partially true, with the added especially.
But the same is true for SPSS Modeler, “especially when team members
do not have the same level of” SPSS Modeler “knowledge.” Without their
qualification, “especially”, the whole idea behind open-source is about
sharing. But IBM certainly does not want to share the market with R. So
as R gains momentum in the analytic community, why not put forth the
idea that R is good but SPSS Modeler makes it better (perhaps the
opposite is more truthful).

Conclusion
Nevertheless, it will probably be a good webinar. It is just lame that you
have to tell half-truths to advertise it. Finally, if you want SPSS
Modeler to be better, include SPSS Stats functionality for the modeling
blocks included in Modeler, and work on you documentation, a lot!

165
 An Analytics Best Kept Secret
I have written about R. I have told you about Octave. Little did I know
that I had not found the motherlode. Thanks to my friend and colleague
Dan Pompea, I now have discovered SciLab.

If MATLAB has any advantage over this open-source gem, I have yet to
find it (at least for use in operations research and analytics). Not even
Simulink is worthy of mention in the same breath. I have used MATLAB
extensively. It was the foundation of my bestselling book, Missile Flight
Simulation. Now I must turn my back to this industry-standard
workhouse and its ferocious price tag to this newcomer and it's
incredibly lost cost: a donation if you choose.

I have yet to explore its vast capabilities but what I read is remarkable.
However, I have carefully stressed its modeling and simulation features,
and I say, "Farewell poor Simulink, I knew you so well." I am so sorry for
your loss, dear Octave, for you were becoming a shining star. But now
there is a new kid in town, whose glimmer is brighter than your own.

To R, I say, "Work hard in your niche or you will be overcome." Though

SciLab boasts of it statistical modeling capabilities, I have thus far not
found the equalizer. Thus my dear R, you still have life abundantly.

166
To ExtendSim, I say, "I enjoyed our time together, and the role you
played in my second best-seller, Discrete Event Simulation using
ExtendSim." To Arena Simulation, which I introduced to Army
Operations Research ten year ago, I say, "Farewell trusted companion. I
hope your future is bright."

To SciLab, I say "Welcome, jewel of the Earth."

167
 What is Linear Programming?
Linear programming (LP) is a tool for solving optimization problems. In
1947, George Dantzig developed an efficient method, the simplex
algorithm, for solving linear programming problems (also called LP).
Since the development of the simplex algorithm, LP has been used to
solve optimization problems in industries as diverse as banking,
education, forestry, petroleum, and trucking. In a survey of Fortune 500
firms, 85% of the respondents said they had used linear programming.
As a measure of the importance of linear programming in operations
research, approximately 70% of this book will be devoted to linear
programming and related optimization techniques.

Optimization simply means to find the optimal of best as possible

solution to a problem. These are often framed as maximums or
minimums. For example, maximize profit and minimize costs.

What Is a Linear Programming Problem?

Here I will introduce linear programming and define important terms
that are used to describe linear programming (optimization) problems.

Algorithm. Optimization modeling.

168
Step 1. Find the decision variables, i.e. find out what are the variables
whose values you can choose.

Step 2. Find the objective function, i.e. find out how your objective to
be minimized or maximized depends on the decision variables.

Step 3. Find the constraints, i.e. find out the (in) equalities that the
decision variables must satisfy.(Don’t forget the possible sign
constraints!)

This is not the best example in the world, because we can solve this
without a linear program. But it is one that makes illustration of the
concept simple. Let’s say we want to maximize Profit on a certain
product. Profit would be our decision variable. In simplest terms, profit
is a function of revenue minus costs, so we could write the objective
function as

Maximize: Profit = Revenue – Cost

Now, do we ever have anything constraining revenue and cost? What

about our budgets? So, say we cannot spend any more than $500 on
advertising the product and it cost $20 to make each product, and for
simplicity, those are our only costs. Let’s also say that we can only
produce 250 of this product, which would constrain revenue. Let’s
further say that we can only charge $40 for each product.

Our constraints are:

 Marketing cost must be less than or equal to $500

 Production cost equal $20 per item time 250 items
 Revenue is less than or equal to $40 time 250 items.
 All of the values are positive numbers or zero

So, we write this in mathematical shorthand as, let P = Profit, R =

Revenue and C = Costs, (s.t. is shorthand for subject to, which identifies
the constraints) then

max P = R – C

169
s.t.:

1. C(marketing) <= $500

2. C(production) = $20*(250)
3. R <= $40*(250)
4. C >= 0 and R >= 0

We would then code this into a software program like Excel, LINDO,
MATLAB, etc. and let the software solve the LP. Though there are many
other technical details associated with arriving at a solution, this is
basically what linear programing is all about. We could make this
problem more interesting by allowing the number of items produce vary
between zero and its upper bound of 250.

MATLAB Code
This is an LP with six variables and seven constraints. The code consist of
defining three matrices, A, b and c. The code that executes the LP is the
line at the bottom. It happens to be a minimize LP. glpk is the function
that executes the LP, the zeros are lower bounds of the variables, "L"
indicates the constrains have lower bounds, "C" indicates the variable
are continuous, and "1" tells the program to minimize. The c matrix
represents the objective function, While A and b constitute the
constraints. This also runs in an open source program called Octave.

c=[-1, 0, 0, 0, 0, 1];
A=[-1, 0, 1, 0, 0, 0;
-1, 1, 0, 0, 0, 0;
0, 0, -1, 0, 1, 0;
0, 0, -1, 1, 0, 0;
0, 0, 0, -1, 1, 0;
0, 0, 0, 0, -1, 1;
0, -1, 1, 0, 0, 0];
b=[6; 9; 8; 7; 10; 12; 0];
[x_min,z_min,status,extra]=glpk(c,A,b,[0;0;0;0;0;0],[
],"LLLLLLL","CCCCCC",1)

Conclusion
LPs are used for optimizing schedules, networks, transpiration
systems, delivery systems, manufacturing, investing, and so on.

170
These will often be comprised of many variables and constraints.
LPs can be used to optimize a project schedule in Project
Management. Operations research analysts are specifically trained
to formulate and solve these kinds of problems.

171
 Part IX - Advice

Seven Analytics Lessons Learned

I have now been a Senior Analytics Consultant with Clarity Solution
Group for three years (from 22 May 2012). This was a career change of
sorts, moving from modeling, simulation and analysis in aerospace and
defense industry, to predictive modeling in the FSI industry. Here are
some things I have learned (not the ones on the chalk board):

1. When dealing with “big data” (#big#data), about 2/3 of you project
time is spent getting access to the data, getting the right data,
preprocessing the data, and exploratory data analysis, prior to
modeling building.
2. Presentation of the results are as important as the analytics
performed. If you cannot convey the results clearly and succinctly,
you’re not adding value for your customer. In particular, your results
must be translated to economic value or a similar metric.
3. Everything benefits from peer reviews, including the 2/3 work
performed before modeling (See #1). You should develop a template
for conducting peer reviews so that they are thorough and
consistent.

172
4. You should attempt to validate everything. This includes any
requirements, data provided (response data, etc.), macros provided,
etc. That is to say, not just your model.
5. Listen carefully to the customer and help them express their “real”
problem. Often the customer knows they have a problem but either
cannot express it or they do not identify the root problem.
6. Ensure that the variables in a model are intuitive and limit them to
the most predictive variable. Rule of thumb: not more than ten
variables.
7. Document your work as you progress. Either keep a notebook for
your project or record it electronically. I use a spreadsheet with a tab
for each task or sub-task. This may be used later for a writing
development document or providing information for validation.

I am certain that I could provide more, but seven sounds good. In

addition to the lessons learned, it has been fun to step back from being
a technical director to a modeler/analyst. It has also been great to work
for a company that values its employees and treats them with respect. Ii
has also been a pleasure sharing knowledge (non-proprietary) with you,
my peers.

About the picture: The lessons learned on the picture are seven of ten
staff rules I learned from the executive officer when I was a personnel
officer in 3rd Squadron 2nd Armored Cavalry Regiment. Note: Charlie
Beckwith commanded Desert One during Operation Eagle Claw (Google
it).

173
 Here’s one more the thing…advice
for young people
In this article I want to add to "Here’s the thing…more advice for young
analysts", based on questions I continue to receive through LinkedIn
messages and e-mail. I also want to extend this beyond advice to
analysts, and include anyone who may have an ear to hear.

Here’s the thing…Everyone wants to be noticed, to stand-out from the

crowd. One of the best ways to do this is blogging, like posting articles
here on LinkedIn. It also helps to have your own blog. But keep it
professional. If analytics is your area of expertise, write about analytics.
If it is human resources, then write about that. Yet, only write what you
know about, or if you are just stimulating conversation and feedback,
then be honest. People will see right through a fake.

Here’s the thing…Not everyone can write. If you are one of these, then
read and comment on other people's articles/blogs. Better yet,
comment and share both on LinkedIn and Twitter. Every LinkedIn post
you read has an icon (under the title) for LinkedIn, Twitter, Google+ and
Facebook. Sharing helps the author get more exposure, and the person
who shares. Also, reading more may also make you a better writer.

174
Here’s the thing…There are two ways to do things: right and over. Even
the best of us make mistakes. I can make a mistake in the blink of an eye,
even though I have been at my profession for over 20 years. So when I
do not get it right, I own up to my mistake and do it (whatever task I was
trying to perform) OVER! I suppose that leads to some advice for not so
young people: allow your young employees to make mistakes and have
them do it over. This leads to learning and better workers.

Here’s the thing…I have to repeat this one: some jobs require a Ph.D.,
like a professor at an accredited university. (You can teach without a
Ph.D., just not as a professor—I know a really great guy who has been
doing this for years.) Most jobs do not require a Ph.D. However, some
people feel like they will never be of much worth without it. And here is
the crux of the matter:

“If you are not god enough without it, you will never be good enough
with it.”

Here’s the thing…Some of the very best Data Scientists and Analytic
Professionals that I know do not have Ph.D.’s, and hardly anyone
addresses me a Doctor (usually just “Hey you!”).

Here’s the thing…This one is also worth repeating: there are many
online courses and tutorials that can provide you with the training you
need without getting another degree. For young analysts, I refer you to
another bicorner.com article I wrote recently: Online Education in
Analytics and Data Science. In general, Coursera offers many courses
with flexible scheduling at no cost. The key here is to become a "lifelong
learner", constantly seeking self-improvement. Oh, and it shows a little
initiative, not waiting on the company to train you. Most of them do not
offer much training anyway.

Here’s the thing…Know who you are competing with and (here's the
shocker) help them! This can back-fire, but a good supervisor will take
note of the one who is helping their colleagues and might say, "I see
leadership potential there". There are two ways of doing this (and they
are not "right" and "over"): with fanfare and without it. One way to
openly do this is searching for people on LinkedIn who have the “title”
or “professional headline” that you are interested in. Study profiles to

175
see what these people are doing and have done and connect with them.
Oh, and I mean CONNECT with them, not just add someone to your
network. In fact, go back to my second "Here's the thing..." and share
what other people are doing. At work, I would do it with less fanfare--
not that I would hide it, but I wouldn't go to extremes in making sure my
help was noticed.

Here’s the thing…and this one is difficult and counter-intuitive: Be

selfless! That's right, I did not say selfish, rather SELF-LESS. I think that
literally means "without self", so as an adjective describes the one who
acts without self in the forefront. Face it, we are all a little selfish--it is
human nature. But, the ability to suppress our nature and act out of
regard for others first is also a characteristic that will not go unnoticed
by most people.

176
 Index

@
@Risk .......................................................................................................... 17, 44
“
“what if” analysis .............................................................................................. 61
A
acquisition ............................................................... 14, 18, 50, 54, 128, 134, 160
ad hoc procedures ...................................................................................... 10, 28
Alan Turing ...................................................................................................... 123
All Data ...................................................................................................... 97, 101
alternative hypothesis ...................................................................................... 91
Analytical Customer Relationship Management .................................... 9, 14, 21
analytical solution ............................................................................................. 11
Analytics .. 3, 7, 8, 9, 11, 14, 15, 16, 17, 18, 28, 29, 31, 43, 44, 51, 52, 56, 74, 76,
82, 83, 90, 91, 92, 93, 94, 95, 107, 113, 114, 131, 132, 133, 134, 142, 150,
155, 157, 159, 161, 162, 166, 167, 173, 175
ANN ..................................................................................... 40, 42, 66, 67, 68, 69
attrition ................................................................................... 15, 39, 40, 42, 130
auto-neural network ............................................................................... See ANN
B
Big Data ..........................................................2, 3, 4, 5, 18, 36, 97, 102, 109, 130
BOARD....................................................................................... 44, 161, 162, 163
bottlenecks ....................................................................................................... 83
Business analytics ............................................................................................... 7
business case .................................................................. 26, 29, 32, 34, 146, 147
business rules ................................................................................................ 9, 13

177
C
call center .................................... 35, 56, 57, 63, 71, 72, 129, 134, 135, 136, 147
campaign ...............................................................................................49, 50, 56
Capital asset pricing model .............................................................................. 17
channels ................................................................................................16, 54, 55
Clarity Solution Group .................................................................................... 173
classification trees .......................................................................................40, 42
Clinical Decision Support Systems .................................................................... 14
clustering ......................................................................... 29, 57, 58, 59, 101, 162
Colorado Trail ........................................................................................82, 83, 84
control group...............................................................................................49, 50
Coursera ............................................................................ 29, 102, 107, 114, 176
cross-sale .......................................................................................................... 15
cross-sell ........................................................................................................... 14
customer base .............................................................................................14, 54
Customer Lifetime Value .................................................................................. 54
customers .. 8, 9, 12, 13, 14, 15, 16, 18, 26, 29, 33, 34, 45, 47, 48, 50, 53, 54, 55,
56, 57, 59, 72, 76, 77, 82, 86, 87, 88, 89, 113, 125, 126, 129, 134, 135, 147,
161, 173, 174
D
data mining ..................................................................... 114, 115, 142, 150, 153
data patterns .................................................................................................8, 12
data reduction ................................................................................... 4, 30, 33, 95
Data Science ii, 10, 11, 29, 36, 37, 42, 94, 95, 102, 103, 104, 105, 108, 114, 116,
118, 164, 176
data scientist .... 36, 104, 105, 106, 107, 108, 111, 112, 113, 114, 116, 117, 118,
119, 122, 123, 124, 125
decision making......................................................................... 7, 10, 12, 14, 132
decision models................................................................................................ 13
decision trees ................................................................................................... 19
descriptive analytics ........................................................................................... 8

178
descriptive models .................................................................................. 8, 13, 51
deterministic ............................................................................................... 64, 65
deterministic model .......................................................................................... 65
direct marketing.......................................................................................... 14, 16
discrete event simulation ........................... 43, 44, 61, 62, 71, 72, 134, 135, 168
Distributed Interactive Simulation .................................................................... 63
E
explanatory variables .................................................................................. 11, 12
ExtendSim ................................................................................................. 44, 168
F
Facebook ......................................................................................................... 175
false positive ............................................................................................... 84, 91
File of Dreams ................................................................................................... 45
Financial Services and Insurance ...............................................................See FSI
forecasting model ........................................................................... 12, 38, 40, 42
fraud...................................................................................... 8, 11, 12, 16, 17, 92
FSI............................................................................................ 134, 159, 160, 173
funnel ................................................................................................................ 47
G
game theory ...................................................................................... 43, 137, 138
general equilibrium theory ..................................................................... 137, 138
George Box........................................................................................................ 24
Google+ ........................................................................................................... 175
H
Hadoop.................................................................................. 18, 30, 44, 115, 129
heuristics ......................................................................................... 11, 17, 25, 92
Hierarchical Optimal Discriminant Analysis ...................................................... 19
High Level Architecture ..................................................................................... 63
Human Resources ................................................................... 130, 131, 132, 133

179
I
inferential statistics .......................................................................................... 90
Informatics ..................................................................................................... 111
Information Value ............................................................................................ 98
INFORMS ...................................................................................................52, 119
integration .......................................................................................112, 117, 118
Internet of Things ........................................................................................... 117
J
jargon ......................................................................................................126, 128
Jessica Rabbit ................................................................................................... 86
John Nash ....................................................................................................... 139
K
Key Performance Parameter .......................................................................... 147
K-Nearest Neighbor .......................................................................................... 19
Kurt Vonnegut .....................................................................................75, 88, 131
L
likelihood ........................................................... 8, 12, 13, 15, 45, 54, 55, 56, 128
linear programming.............................................................................29, 44, 169
linear regression ............................................................................................... 19
LinkedIn ............................................................................... 37, 55, 165, 175, 176
logic regression ................................................................................................ 19
logistic regression .......................................... 11, 19, 29, 38, 65, 69, 74, 128, 160
M
machine learning ..... 8, 10, 11, 12, 17, 18, 24, 29, 35, 38, 51, 66, 67, 91, 92, 105,
108
maintainability ....................................................................................62, 72, 135
MapReduce ...................................................................................................... 18
marketing ...... 8, 9, 10, 12, 14, 32, 42, 49, 50, 54, 56, 92, 95, 107, 128, 129, 134,
145, 151, 159, 171
mathematical modeling ..............................................................................29, 44

180
MATLAB ...................................................................................... 43, 44, 167, 171
mean ................................................................................................................. 84
metric ............................................................................ 34, 50, 84, 146, 147, 173
model 8, 10, 11, 12, 13, 16, 19, 24, 25, 26, 27, 30, 31, 32, 33, 34, 35, 36, 37, 38,
40, 47, 48, 50, 54, 55, 56, 57, 60, 61, 62, 63, 69, 74, 98, 123, 125, 126, 128,
134, 136, 138, 147, 154, 156, 157, 160, 165, 174
modeler judgment ................................................................................ 11, 12, 33
Monte Carlo ................................................................................................ 43, 61
Multivariate adaptive regression splines .......................................................... 19
mutual defection............................................................................................. 140
N
Naïve Bayes ....................................................................................................... 19
NASA ................................................................................................... 3, 4, 42, 88
Nash equilibrium ............................................................................................. 140
neural networks .................................................................. 19, 38, 40, 51, 67, 68
null hypothesis .................................................................................................. 91
O
open-source .............................................................. 44, 154, 159, 160, 166, 167
Operation Eagle Claw ................................................................................ 87, 174
Operation Overlord ........................................................... 37, 119, 145, 150, 152
operational research ...................................................... See operations research
operations research .. xvi, 7, 11, 24, 28, 31, 32, 34, 37, 42, 52, 75, 76, 77, 92, 94,
95, 107, 119, 120, 142, 143, 144, 145, 150, 151, 152, 159, 167, 168, 169
optimization ...............................12, 13, 29, 43, 92, 119, 143, 145, 150, 151, 169
Otave................................................................................................................. 44
P
Point of Sell ....................................................................................................... 56
predictive analytics 8, 9, 10, 11, 12, 14, 15, 16, 17, 18, 19, 28, 44, 53, 56, 57, 73,
74, 92, 94, 118, 161
predictive modeling.................. 7, 8, 10, 12, 23, 28, 36, 37, 53, 91, 119, 159, 173
predictive models .................. 8, 10, 12, 13, 15, 24, 36, 38, 41, 49, 51, 52, 53, 80

181
prescriptive analytics ......................................................................................8, 9
Probabilistic Risk Assessment .......................................................................... 17
probit regression .............................................................................................. 19
problem solving .................................................................................75, 120, 121
problem space .........................................................................................151, 152
profit .................................................................................... 32, 34, 113, 169, 170
propensity model ............................................................................................. 45
propensity models............................................................................................ 53
propensity to buy ........................................................ 24, 35, 47, 55, 65, 98, 128
propensity to churn .......................................................................................... 56
propensity to engage ..................................................................................42, 54
propensity to unsubscribe................................................................................ 55
Python ......................................................................... 30, 97, 113, 124, 129, 157
Q
quantile regression........................................................................................... 19
R
R 154, 164
random forests ............................................................................................40, 42
regression .............................. 12, 18, 19, 29, 38, 42, 51, 69, 76, 80, 95, 101, 105
regression trees ................................................................................................ 19
reliability ....................................................................................... 10, 62, 72, 135
requirements ................................... 29, 32, 33, 83, 102, 108, 111, 118, 119, 174
retention .................................................................................................9, 14, 15
Return on Investment .................................................................................... 147
revenue ........................................................................................ 54, 55, 138, 170
Rexer Analytics ............................................................................................... 153
ridge regression ................................................................................................ 19
R-Studio ...............................................................................................29, 58, 155

182
S
SAS ................... 21, 29, 30, 97, 113, 115, 128, 129, 155, 156, 157, 159, 160, 165
SAS Enterprise Modeler .............................................................................. 21, 29
scientist .................. 23, 36, 37, 106, 108, 113, 116, 119, 121, 123, 124, 125, 164
Scilab ................................................................................................................. 44
SCILAB ..................................................................................................... 167, 168
segmentation .............................................................................................. 12, 57
Share of Wallet ................................................................................. 54, 146, 147
simulation ..... 3, 10, 29, 44, 60, 61, 62, 63, 64, 73, 74, 75, 76, 77, 118, 119, 121,
145, 150, 167, 173
SPSS Modeler ...................................................................... 21, 29, 113, 165, 166
SQL .................................................................................................... 30, 113, 154
statistical model ........................................................................ 11, 24, 25, 38, 60
statistical techniques ...................................................... 8, 10, 11, 12, 23, 29, 80
statistician ................................................. 11, 79, 92, 94, 95, 103, 105, 121, 123
Statistics .. 8, 10, 11, 23, 28, 29, 36, 37, 41, 78, 79, 80, 90, 91, 92, 93, 94, 97, 98,
100, 101, 102, 104, 111, 156, 157
Steve Cartwright ............................................................................................... 47
stochastic processes ................................................................................... 61, 64
Structured Query Language ..................................................................... See SQL
T
target ................................................................................................................ 16
The Prisoner’s Dilemma .................................................................................. 139
time-series model ............................................................................................. 38
training sample ................................................................................................. 12
treated group .................................................................................................... 49
Twitter............................................................................................................. 175
U
unstructured data ................................................................................. 9, 18, 130
Uplift model ................................................................................................ 49, 50

183
V
validation ..................................................................................... 20, 26, 141, 160
Vince Roske .................................................................................................... 146
W
waiting time ......................................................................................72, 135, 136
Weight of Evidence .......................................................................................... 98

184