Anda di halaman 1dari 94

A Course in Data Discovery and Predictive Analytics 16 Nov 2013

David M. Levine, Baruch College—CUNY  A definition of business analytics


A Course in  Broad categories of business analytics
Kathryn A. Szabat, La Salle University
Data Discovery What Are We (INFORMS 2010-2011)
and Predictive David F. Stephan, Two Bridges Instructional Talking About?
Technology  Business analytics continues to become
Analytics increasingly important in business and therefore
in business education
analytics.davidlevinestatistics.com
DSI MSMESB session, November 16, 2013

1 2

 Addresses a topic of growing interest


 Introduces methods of problem description and  Technology use should not hamper students
decision-making not seen elsewhere in the ability to learn concepts
Course business statistics curriculum  Emphasize application of methods (business
Justification  Assumes a pre-requisite introductory course that Guiding students are the audience)
and Starting covers descriptive statistics, confidence intervals Principles  Compare and contrast with decision-making
Points and hypothesis testing, and simple linear using traditional methods where possible.
regression  Capitalize on insights gained teaching related
 Presents methods that have antecedents in subjects such as CIS and OR/MS
introductory course

3 4

How David  Have sought to make statistics useful to students


How Our
Levine’s majoring in the functional areas of accounting,
Teaching As a team, our varied backgrounds and economics/finance, management, and marketing
Teaching
Experience interests contribute to shaping our choices
Experience  Have changed my focus as changes in
Informs Us technology occurred over time
Informs Us

5 6

2013‐Levine‐Szabat‐Stephan‐DSI‐MEMESB‐slides.pdf 1
A Course in Data Discovery and Predictive Analytics 16 Nov 2013

Late 1980s/early
Early 1980s – 1990s – Started
Integrated to focus on
software such  Enabled me to begin focusing on results rather
than calculations
software with
as SAS, SPSS, enhanced user Saw how this would make statistical tools more
 Helped me realize that students trained to use
and Minitab statistical programs would have increased
interfaces that accessible to novice students, in particular.
into opportunities in business replaced older,
introductory programming-
oriented
course
interfaces

7 8

Early 1990s – Late 1990s –


Integrated Pondered the
Deming’s Total  Through consulting work, learned the use of  Realized Excel needed to be modified for
Quality importance of organizational culture and the
difficulty of implementing change Microsoft classroom use
Management
philosophy and  This had limited long term impact as coverage
Excel, by then  Crossed paths and discovered shared interests
prevalent in with David Stephan
practices into the of this topic migrated to operations management
introductory business
course. schools

9 10

 Crossed path and discovered shared interests Overarching guiding principle:


with Kathy Szabat. Statistics plays a role in problem solving and
 Realized this is our best opportunity to make decision making.
Current Day – Kathryn
business statistics critical to the success of
Reflected on majors in the functional areas Szabat’s Statistics – the methods that help transform data into
analytics  Believe this represents an opportunity to
Experience useful information for decision makers
 Provides support for gut feeling, intuition,
develop new majors in analytics and revise experience
majors in business statistics (CIS, et. al.)  Provides opportunity to gain insight

11 12

2013‐Levine‐Szabat‐Stephan‐DSI‐MEMESB‐slides.pdf 2
A Course in Data Discovery and Predictive Analytics 16 Nov 2013

Have
consistently
emphasized Continual outreach to colleagues in different Have used  Without compromising understanding of logic
applications of departments within the school of business to technology of formulas
statistics to better understand how statistics is used in the extensively in  Advocating the importance of “using a tool” to
various functional areas generate results
functional the course
areas of
business

13 14

Have
Have
increased, over
increased, over
time, focus on
time, focus on
problem- With attention to “formulating the problem” Someone has to tell the story at the end
interpretation
solving and
and
decision-
communication
making

15 16

Have recently
been engaged in  Effort as a response to the technology and data-
developing a driven changes in business today
 Visualization has always been a theme in my
new,  Outreach to practitioners to better understand David work and interests
interdisciplinary “business analytics” as an emerging field
academic Stephan’s  Context-based learning advocate
 Developed an introductory presentation on
department, business analytics to be used by all faculty in the
Experience  Witnessed and taught about several generations
of information technology
Business introductory statistics course (as well as
Systems and introductory IS and operations courses)
Analytics

17 18

2013‐Levine‐Szabat‐Stephan‐DSI‐MEMESB‐slides.pdf 3
A Course in Data Discovery and Predictive Analytics 16 Nov 2013

 The story of the textbook that omitted the


How things  Do you remember the ALU and CU? dBASE language
Relational Accept “Last Name:” to lastname
work versus  CP/M or DOS—Which is the better choice?
Database Input “Grade:” to grade
how to work  When is the last time someone asked you about @5,10 SAY Trim(lastname) + grade PICTURE 99.9
Debate
with things the ASCII table?  Should database examples use one relation or
two or more?

19 20

 Simpler things can be used to teach operating  If you don’t teach {formulas, computations, fully explain
principles and simulate more complex things Challenge: methods, widgets, whatever}, students will not
understand “anything.”
 Large-scale things can be imagined from small- Finding the
Lessons from  How many helpful “black boxes” do you already use
scale things right level of without explanation?
the Debate
 Don’t fuss over technology choices—in the abstraction to  The Microsoft Excel xls file format

 Don’t try to reveal/decompose all complex systems


long-run, your choice will most likely not be teach.  Can end up discussing parts that, at a later time, get use as an
future-proof! integrated whole

21 22

 “Volume, velocity, and variety” How to address


these data characteristics often associated with
analytics?
New  Semi-subjective analysis of outputs (e.g., 3D Best Topics to Teach
scatterplots or cluster plots) Seeking
Challenges to Best Technology to Use
Course “Bests”
Address  Examining patterns before testing hypotheses Best Context to Deliver Instruction
 Need to determine when to assign causality (to
relationships) as part of the analysis versus
testing a hypothesized causality

23 24

2013‐Levine‐Szabat‐Stephan‐DSI‐MEMESB‐slides.pdf 4
A Course in Data Discovery and Predictive Analytics 16 Nov 2013

 Descriptive analytics/data discovery: most likely  Experience teaches us not to be overly


to be seen, builds on and extends introductory concerned about choice!
descriptive methods. Can be used to raise and
 No one program, application, or package is best
“simulate” volume and velocity issues. “Best”
“Best” Topics in 2013
 Predictive not prescriptive analytics. The latter Technology to
to Teach  Best technology combines most accessible with
brings into play management insight, judgment, Use what bests illustrates the concept
and wisdom. (Predictive combines traditional
statistical analysis with data mining, as defined  Our choice: mix of Microsoft Excel, Tableau
earlier.) Public, and JMP

25 26

 A broad case that represents an enterprise of


“Best” Context suitable complexity, yet one that can be Course
to Deliver understandable on a casual level Description
Instruction  Our choice: a theme park with several different In-Depth
parts (“lands”) and an integrated resort hotel

27 28

 Introduction (2)
 Descriptive Analytics (2)
 Preparing for Predictive Analytics (1)  How We Got Here: Evolutionary changes that
Topic List  Multiple regression including residual analysis, have led to more widespread usage of analytics
(with dummy variables, interaction terms, and Introduction (2  How analytics can change the data analysis and
influence analysis (1.5-2) decision-making processes
suggested weeks)
 Logistic regression (1)  Basic vocabulary and taxonomy of analytics
weeks)
 Multiple regression model building including  Technology requirements and orientation
transformations, collinearity, stepwise
regression, and best subsets (1.5-2)
 Predictive Analytics (4-5)
29 30

2013‐Levine‐Szabat‐Stephan‐DSI‐MEMESB‐slides.pdf 5
A Course in Data Discovery and Predictive Analytics 16 Nov 2013

Provide information about the current status of a business or


business activity in a form easy to comprehend and review.

 Summarizing volume and velocity


 “Sexiness” versus usefulness issue Summarizing
Descriptive
 Levels of summary: drill down, levels of volume and
Analytics (2
hierarchy, and subsetting velocity:
weeks)
 Information design principles that inform Dashboards
descriptive methods

31 32

Example: combining a numerical measure with a


Sexiness categorical group Sexiness
versus  Which one looks more “sexy,” appealing, versus
usefulness: interesting, etc.? usefulness:
Gauges vs.  Which one best facilitates comparisons? Gauges vs.
bullet graphs  What if the answers to the two questions are bullet graphs
different?

33 34

 Which one looks more “sexy,” appealing, interesting, Drill-down sequence example (using Excel)
etc.?
 Which one best facilitates comparisons?
Sexiness  What if the answers to the two questions are different? Levels of
versus summary: drill
usefulness: down, levels of
Gauges vs. hierarchy, and
bullet graphs subsetting

35 36

2013‐Levine‐Szabat‐Stephan‐DSI‐MEMESB‐slides.pdf 6
A Course in Data Discovery and Predictive Analytics 16 Nov 2013

Visual drill-down using a tree map


Financial example showing another level of drill-down

Levels of Levels of
summary: drill summary: drill
down, levels of down, levels of
hierarchy, and hierarchy, and
subsetting subsetting

37 38

Subsetting using “slicers” (Excel)

 Fostering efficient and effective communication


and understanding
Levels of
 Provide context for data in a compact
summary: drill Information presentation
down, levels of design
 Add additional “dimensions” of data
hierarchy, and principles
 Misuse raises issues beyond “typical” statistical
subsetting concerns: visual perception, artistic
considerations

39 40

Tree Map of Retirement Fund Assets Colored by 10-Year


Return Percentage, By Fund Type (JMP) Sparklines example (Excel)
Does this tree
map provide
Does this table
context for data
provide context
in a compact
for data in a
presentation?
compact
presentation?
Add additional
“dimensions”
of data?

GROWTH FUNDS VALUE FUNDS 41 42

2013‐Levine‐Szabat‐Stephan‐DSI‐MEMESB‐slides.pdf 7
A Course in Data Discovery and Predictive Analytics 16 Nov 2013

Nobel Laureates Graph (Accurat information design agency)


Tree Map of Number of Social Media Comments
Colored by Tone, By “Land” (Excel)

Information
Information
design tree
design
map example
principles:
with simpler
“infographics”
data

43 44

Detail of Nobel Prize Laureates Graph

Information Preparing for  Confidence intervals


design Predictive  Hypothesis testing
principles: Analytics (1
 Simple linear regression
“infographics” week)

45 46

 Normal distribution  Basic Concepts of hypothesis testing


Confidence  Sampling distributions  p-values
Hypothesis
intervals  Confidence intervals for the mean and  Tests for the differences between means and
proportion
testing proportions

47 48

2013‐Levine‐Szabat‐Stephan‐DSI‐MEMESB‐slides.pdf 8
A Course in Data Discovery and Predictive Analytics 16 Nov 2013

 Developing the multiple regression model


 The simple linear regression model
 Inference in multiple regression
 Interpreting the regression coefficients Multiple
Simple linear  Residual analysis
 Residual analysis
Regression
regression  Dummy variables
 Assumptions of regression (1.5-2 weeks)
 Interaction terms
 Inferences in simple linear regression
 Influence analysis

49 50

 Interpreting the coefficients  Testing the overall model


Developing the
Inference in  Testing the contribution of each independent
multiple  Coefficients of multiple determination
multiple variable
regression  Coefficients of partial determination
regression  Adjusted r2
model  Assumptions

51 52

Using categorical independent variables in a


 Plots of the residuals vs. independent variables regression model:
Residual  Plots of the residuals vs. predicted Y
Dummy  Defining dummy variables
analysis variables  Interpreting dummy variables
 Plots of the residuals vs. time (if appropriate)
 Assumptions in using dummy variables

53 54

2013‐Levine‐Szabat‐Stephan‐DSI‐MEMESB‐slides.pdf 9
A Course in Data Discovery and Predictive Analytics 16 Nov 2013

Examining the effect of individual observations


 What they are on the regression model
Interaction  Why they are sometimes necessary
Influence  Hat matrix elements hi
terms analysis  Studentized deleted residuals ti
 Interpreting interaction terms
 Cook’s Distance statistic Di

55 56

“Predicting the likelihood of upgrading to a premium


credit card based on the monthly purchase amount and
whether the account has multiple cards”
Predicting a categorical dependent variable
 Cannot use least squares regression Logistic
Logistic  Odds ratio
regression
regression (1  Logistic regression model
 Predicting probability of an event of interest example using
week)
 Deviance statistic an Excel add-in
 Wald statistic

57 58

Multiple  Transformations
 Purposes
Regression  Collinearity
Transformations  Square root transformations
Model Building  Stepwise regression
 Logarithmic transformations
(1.5-2 weeks)  Best subsets regression

59 60

2013‐Levine‐Szabat‐Stephan‐DSI‐MEMESB‐slides.pdf 10
A Course in Data Discovery and Predictive Analytics 16 Nov 2013

 History
 Effect on the regression model
Stepwise  How it works
Collinearity  Measuring the variance inflationary factor (VIF)
regression  Limitations
 Dealing with collinear independent variables
 Use in an era of big data

61 62

METHOD FOR 
METHOD Prediction Classification Clustering Association
Classification and
 How it works regression trees  
Predictive
Best subsets  Advantages and disadvantages vs. stepwise (1-1.5 weeks)
Analytics (4-5 Neural networks
regression regression
(1-1.5 weeks)
  

 Mallows Cp statistic
weeks)
Cluster analysis

(1 week)
Multidimensional
 
scaling (1week)

63 64

“Predicting the likelihood of upgrading to a premium credit


Decision trees that split data into groups based on the card based on the monthly purchase amount and whether the
values of independent or explanatory (X) variables. account has multiple cards” (same example used in logistic
 Not affected by the distribution of the variables regression)
 Splitting determines which values of a specific
independent variable are useful in predicting the
Classification dependent (Y) variable present
 Using a categorical dependent Y variable results in a Classification
and regression classification tree
tree example
trees  Using a numerical dependent Y variable results in a
regression tree
 Rules for splitting the tree
 Pruning back a tree
 If possible, divide data into training sample and
validation sample

65 66

2013‐Levine‐Szabat‐Stephan‐DSI‐MEMESB‐slides.pdf 11
A Course in Data Discovery and Predictive Analytics 16 Nov 2013

“Predicting sales of energy bars based on price and promotion


“Predicting the likelihood of upgrading to a premium credit expenses” (could be multiple regression example, too)
card based on the monthly purchase amount and whether the
account has multiple cards” (same example used in logistic
regression)

Classification Regression tree


tree example example

67 68

“Predicting the
likelihood of upgrading
 Constructs models from patterns and relationships to a premium credit
uncovered in data card based on the
monthly purchase
 Computations that begin with inputs and end with Neural net amount and whether
Neural nets outputs
example 1 the account has
 Uses a hyperbolic tangent function multiple cards” (same
example used for
 Divide data into training sample and validation sample logistic regression and
classification tree)

69 70

“Predicting sales of Classifies data into a sequence of groupings such that


energy bars based on
price and promotion
objects in each group are more alike other objects in
expenses” (same their group than they are to objects found in other
Neural net example used in Cluster groups.
example 2 regression tree) analysis  Hierarchical clustering
 k-means clustering
 Distance measures
 Types of linkage between clusters

71 72

2013‐Levine‐Szabat‐Stephan‐DSI‐MEMESB‐slides.pdf 12
A Course in Data Discovery and Predictive Analytics 16 Nov 2013

“Perception of sports based on a survey of these attributes: movement


speed, rules, team orientation, amount of contact”

Visualizes objects in a two or more dimensional


space, or map, with the goal of discovering patterns
Cluster Multi- of similarities or dissimilarities among the objects.
analysis dimensional  Types of multidimensional scaling
 Distance measures
example scaling  Stress statistic – measure of fit
 Challenge in interpreting dimensions

73 74

“Perception of sports based on a survey of these “Perception of sports based on a survey of these
attributes: movement speed, rules, team orientation, attributes: movement speed, rules, team orientation,
amount of contact” amount of contact”

Multi- Multi-
dimensional dimensional
scaling scaling
example using example using
JMP add-in JMP add-in

75 76

 Microsoft Excel (latest versions equipped Apps for Office)


 Good for selected dashboard elements (treemap, gauges, sparklines) and
 Could add some of the descriptive analytics into
illustrating drill-down (with PivotTables) and subsetting (with Slicers) the introductory course
 Extend with third-party add-ins to perform logistic regression
 Drill down and subsetting
 Tableau Public (web-based, free download) Can I  Perhaps one graph that summarize volume and
 Good for descriptive analytics (bullet graph, treemaps)
 Drag-and-drop interface that can be taught in minutes
Incorporate velocity
Software  “Premium” version (not free) extends utility of software to many other
methods, although this server-based version is more geared to business
Any of This  Show-and-tell to illustrate information design and/or
“sexiness” versus usefulness issue
Resources  JMP
Into the
 Could add binary logistic regression if your
 Many displays have drill-down built into them Introductory course covers multiple regression and mentions
 Good for regression trees, neural nets, cluster analysis, and
multidimensional scaling (with additional free add-in) Course? binary logistic regression, but this will not be
 Requires SAS or R for some processing; user interface contains some
quirks for new and casual users (most of which could be eliminated feasible in most cases
through the use of custom add-ins)
 Future versions promise additional capabilities.  “Funny, you should ask that question….”

77 78

2013‐Levine‐Szabat‐Stephan‐DSI‐MEMESB‐slides.pdf 13
A Course in Data Discovery and Predictive Analytics 16 Nov 2013

 Berenson, M. L., D. M. Levine, and K. A. Szabat. Basic Business Statistics 13th


edition. Upper Saddle River: Pearson Education, forthcoming January 2014.
 Breiman, L., J. Friedman, C. J. Stone, and R. A. Olshen. Classification and
Regression Trees. London: Chapman and Hall, 1984.
 Cox, T. F., and M. A. Cox. Multidimensional Scaling, Second edition. Boca Raton,
FL: CRC Press, 2010.  Contact us at analytics@davidlevinestatistics.com
 Everitt, B. S., S. Landau, and M. Leese. Cluster Analysis, Fifth edition. New York:
John Wiley, 2011.  Visit analytics.davidlevinestatistics.com for
 Few, S. Information Dashboard Design: Displaying Data for At-a-Glance Further  Today’s slides including references
Monitoring, Second edition. Burlingame, CA: Analytics Press, 2013.
References  Hakimpoor, H., K. Arshad, H. Tat, N. Khani, and M. Rahmandoust. “Artificial Information or  A preview of some of our current work in this area
Neural Network Application in Management.” World Applied Sciences Journal,  Coming soon WaldoLands.com
2011, 14(7): 1008–1019. Contact
 R. Klimberg, and B. D. McCullough. Fundamentals of Predictive Analytics with
JMP. Cary, NC: SAS Press. 2013
 Look for our (very occasional) tweets using
 Lindoff, G., and M. Berry. Data Mining Techniques: For Marketing, Sales, and #AnalyticsEducation
Customer Relationship Management. Hoboken, NJ: Wiley Publishing, Inc., 2011.
 Loh, W. Y. “Fifty years of classification and regression trees.” International
Statistical Review, 2013, in press
 Tufte, E. Beautiful Evidence. Cheshire, CT: Graphics Press, 2006.

79 80

2013‐Levine‐Szabat‐Stephan‐DSI‐MEMESB‐slides.pdf 14
David M. Levine, Baruch College—CUNY
A Course in
Kathryn A. Szabat, La Salle University
Data Discovery
and Predictive David F. Stephan, Two Bridges Instructional
Technology
Analytics

analytics.davidlevinestatistics.com
DSI MSMESB session, November 16, 2013

16 Nov 2013 2013‐Levine‐Szabat‐Stephan‐DSI‐MEMESB‐slides.pdf 1
 A definition of business analytics
 Broad categories of business analytics
What Are We (INFORMS 2010-2011)
Talking About?  Business analytics continues to become
increasingly important in business and therefore
in business education
 Addresses a topic of growing interest
 Introduces methods of problem description and
decision-making not seen elsewhere in the
Course business statistics curriculum
Justification  Assumes a pre-requisite introductory course that
and Starting covers descriptive statistics, confidence intervals
Points and hypothesis testing, and simple linear
regression
 Presents methods that have antecedents in
introductory course
 Technology use should not hamper students
ability to learn concepts
 Emphasize application of methods (business
Guiding students are the audience)
Principles  Compare and contrast with decision-making
using traditional methods where possible.
 Capitalize on insights gained teaching related
subjects such as CIS and OR/MS
How Our
Teaching As a team, our varied backgrounds and
Experience interests contribute to shaping our choices
Informs Us
How David  Have sought to make statistics useful to students
Levine’s majoring in the functional areas of accounting,
Teaching economics/finance, management, and marketing
Experience  Have changed my focus as changes in
Informs Us technology occurred over time
Early 1980s –
Integrated
software such  Enabled me to begin focusing on results rather
than calculations
as SAS, SPSS,
 Helped me realize that students trained to use
and Minitab statistical programs would have increased
into opportunities in business
introductory
course
Late 1980s/early
1990s – Started
to focus on
software with
enhanced user Saw how this would make statistical tools more
interfaces that accessible to novice students, in particular.
replaced older,
programming-
oriented
interfaces
Early 1990s –
Integrated
Deming’s Total  Through consulting work, learned the
Quality importance of organizational culture and the
Management difficulty of implementing change
philosophy and  This had limited long term impact as coverage
practices into the of this topic migrated to operations management
introductory
course.
Late 1990s –
Pondered the
use of  Realized Excel needed to be modified for
Microsoft classroom use
Excel, by then  Crossed paths and discovered shared interests
prevalent in with David Stephan
business
schools
 Crossed path and discovered shared interests
with Kathy Szabat.
Current Day –  Realized this is our best opportunity to make
business statistics critical to the success of
Reflected on majors in the functional areas
analytics  Believe this represents an opportunity to
develop new majors in analytics and revise
majors in business statistics (CIS, et. al.)
Overarching guiding principle:
Statistics plays a role in problem solving and
decision making.
Kathryn
Szabat’s Statistics – the methods that help transform data into
Experience useful information for decision makers
 Provides support for gut feeling, intuition,
experience
 Provides opportunity to gain insight
Have
consistently
emphasized Continual outreach to colleagues in different
applications of departments within the school of business to
statistics to better understand how statistics is used in the
various functional areas
functional
areas of
business
Have used  Without compromising understanding of logic
technology of formulas
extensively in  Advocating the importance of “using a tool” to
the course generate results
Have
increased, over
time, focus on
problem- With attention to “formulating the problem”
solving and
decision-
making
Have
increased, over
time, focus on Someone has to tell the story at the end
interpretation
and
communication
Have recently
been engaged in  Effort as a response to the technology and data-
developing a driven changes in business today
new,  Outreach to practitioners to better understand
interdisciplinary “business analytics” as an emerging field
academic  Developed an introductory presentation on
department, business analytics to be used by all faculty in the
Business introductory statistics course (as well as
Systems and introductory IS and operations courses)
Analytics
 Visualization has always been a theme in my
David work and interests
Stephan’s  Context-based learning advocate
Experience  Witnessed and taught about several generations
of information technology
How things  Do you remember the ALU and CU?
work versus  CP/M or DOS—Which is the better choice?
how to work  When is the last time someone asked you about
with things the ASCII table?
 The story of the textbook that omitted the
dBASE language
Relational Accept “Last Name:” to lastname
Database Input “Grade:” to grade
@5,10 SAY Trim(lastname) + grade PICTURE 99.9
Debate
 Should database examples use one relation or
two or more?
 Simpler things can be used to teach operating
principles and simulate more complex things
Lessons from  Large-scale things can be imagined from small-
scale things
the Debate
 Don’t fuss over technology choices—in the
long-run, your choice will most likely not be
future-proof!
 If you don’t teach {formulas, computations, fully explain
Challenge: methods, widgets, whatever}, students will not
understand “anything.”
Finding the  How many helpful “black boxes” do you already use
right level of without explanation?
abstraction to  The Microsoft Excel xls file format

 Don’t try to reveal/decompose all complex systems


teach.  Can end up discussing parts that, at a later time, get use as an
integrated whole
 “Volume, velocity, and variety” How to address
these data characteristics often associated with
analytics?
New  Semi-subjective analysis of outputs (e.g., 3D
Challenges to scatterplots or cluster plots)
Address  Examining patterns before testing hypotheses
 Need to determine when to assign causality (to
relationships) as part of the analysis versus
testing a hypothesized causality
Best Topics to Teach
Seeking
Best Technology to Use
Course “Bests”
Best Context to Deliver Instruction
 Descriptive analytics/data discovery: most likely
to be seen, builds on and extends introductory
descriptive methods. Can be used to raise and
“simulate” volume and velocity issues.
“Best” Topics
 Predictive not prescriptive analytics. The latter
to Teach
brings into play management insight, judgment,
and wisdom. (Predictive combines traditional
statistical analysis with data mining, as defined
earlier.)
 Experience teaches us not to be overly
concerned about choice!
 No one program, application, or package is best
“Best” in 2013
Technology to
 Best technology combines most accessible with
Use what bests illustrates the concept
 Our choice: mix of Microsoft Excel, Tableau
Public, and JMP
 A broad case that represents an enterprise of
“Best” Context suitable complexity, yet one that can be
to Deliver understandable on a casual level
Instruction  Our choice: a theme park with several different
parts (“lands”) and an integrated resort hotel
Course
Description
In-Depth
 Introduction (2)
 Descriptive Analytics (2)
 Preparing for Predictive Analytics (1)
Topic List  Multiple regression including residual analysis,
(with dummy variables, interaction terms, and
influence analysis (1.5-2)
suggested
 Logistic regression (1)
weeks)
 Multiple regression model building including
transformations, collinearity, stepwise
regression, and best subsets (1.5-2)
 Predictive Analytics (4-5)
 How We Got Here: Evolutionary changes that
have led to more widespread usage of analytics
Introduction (2  How analytics can change the data analysis and
weeks) decision-making processes
 Basic vocabulary and taxonomy of analytics
 Technology requirements and orientation
 Summarizing volume and velocity
 “Sexiness” versus usefulness issue
Descriptive
Analytics (2  Levels of summary: drill down, levels of
hierarchy, and subsetting
weeks)
 Information design principles that inform
descriptive methods
Provide information about the current status of a business or
business activity in a form easy to comprehend and review.

Summarizing
volume and
velocity:
Dashboards
Example: combining a numerical measure with a
Sexiness categorical group
versus  Which one looks more “sexy,” appealing,
usefulness: interesting, etc.?
Gauges vs.  Which one best facilitates comparisons?
bullet graphs  What if the answers to the two questions are
different?
Sexiness
versus
usefulness:
Gauges vs.
bullet graphs
 Which one looks more “sexy,” appealing, interesting,
etc.?
 Which one best facilitates comparisons?
Sexiness  What if the answers to the two questions are different?

versus
usefulness:
Gauges vs.
bullet graphs
Drill-down sequence example (using Excel)

Levels of
summary: drill
down, levels of
hierarchy, and
subsetting
Financial example showing another level of drill-down

Levels of
summary: drill
down, levels of
hierarchy, and
subsetting
Visual drill-down using a tree map

Levels of
summary: drill
down, levels of
hierarchy, and
subsetting
Subsetting using “slicers” (Excel)

Levels of
summary: drill
down, levels of
hierarchy, and
subsetting
 Fostering efficient and effective communication
and understanding
 Provide context for data in a compact
Information presentation
design
 Add additional “dimensions” of data
principles
 Misuse raises issues beyond “typical” statistical
concerns: visual perception, artistic
considerations
Tree Map of Retirement Fund Assets Colored by 10-Year
Return Percentage, By Fund Type (JMP)
Does this tree
map provide
context for data
in a compact
presentation?

Add additional
“dimensions”
of data?

GROWTH FUNDS VALUE FUNDS
Sparklines example (Excel)

Does this table


provide context
for data in a
compact
presentation?
Tree Map of Number of Social Media Comments
Colored by Tone, By “Land” (Excel)

Information
design tree
map example
with simpler
data
Nobel Laureates Graph (Accurat information design agency)

Information
design
principles:
“infographics”
Detail of Nobel Prize Laureates Graph

Information
design
principles:
“infographics”
Preparing for  Confidence intervals
Predictive  Hypothesis testing
Analytics (1
 Simple linear regression
week)
 Normal distribution
Confidence  Sampling distributions
intervals  Confidence intervals for the mean and
proportion
 Basic Concepts of hypothesis testing
 p-values
Hypothesis
 Tests for the differences between means and
testing proportions
 The simple linear regression model
 Interpreting the regression coefficients
Simple linear  Residual analysis
regression
 Assumptions of regression
 Inferences in simple linear regression
 Developing the multiple regression model
 Inference in multiple regression
Multiple  Residual analysis
Regression
 Dummy variables
(1.5-2 weeks)
 Interaction terms
 Influence analysis
Developing the  Interpreting the coefficients
multiple  Coefficients of multiple determination
regression  Coefficients of partial determination
model  Assumptions
 Testing the overall model
Inference in  Testing the contribution of each independent
multiple variable
regression  Adjusted r2
 Plots of the residuals vs. independent variables
Residual  Plots of the residuals vs. predicted Y
analysis
 Plots of the residuals vs. time (if appropriate)
Using categorical independent variables in a
regression model:
Dummy  Defining dummy variables
variables  Interpreting dummy variables
 Assumptions in using dummy variables
 What they are
Interaction  Why they are sometimes necessary
terms
 Interpreting interaction terms
Examining the effect of individual observations
on the regression model
Influence  Hat matrix elements hi
analysis  Studentized deleted residuals ti
 Cook’s Distance statistic Di
Predicting a categorical dependent variable
 Cannot use least squares regression
Logistic  Odds ratio
regression (1  Logistic regression model
week)  Predicting probability of an event of interest
 Deviance statistic
 Wald statistic
“Predicting the likelihood of upgrading to a premium
credit card based on the monthly purchase amount and
whether the account has multiple cards”

Logistic
regression
example using
an Excel add-in
Multiple  Transformations
Regression  Collinearity
Model Building  Stepwise regression
(1.5-2 weeks)  Best subsets regression
 Purposes
Transformations  Square root transformations
 Logarithmic transformations
 Effect on the regression model
Collinearity  Measuring the variance inflationary factor (VIF)
 Dealing with collinear independent variables
 History
Stepwise  How it works
regression  Limitations
 Use in an era of big data
 How it works
Best subsets  Advantages and disadvantages vs. stepwise
regression regression
 Mallows Cp statistic
METHOD FOR 
METHOD Prediction Classification Clustering Association
Classification and
regression trees  
Predictive (1-1.5 weeks)
Analytics (4-5 Neural networks
  
(1-1.5 weeks)
weeks)
Cluster analysis

(1 week)
Multidimensional
 
scaling (1week)
Decision trees that split data into groups based on the
values of independent or explanatory (X) variables.
 Not affected by the distribution of the variables
 Splitting determines which values of a specific
independent variable are useful in predicting the
Classification dependent (Y) variable present
 Using a categorical dependent Y variable results in a
and regression classification tree
trees  Using a numerical dependent Y variable results in a
regression tree
 Rules for splitting the tree
 Pruning back a tree
 If possible, divide data into training sample and
validation sample
“Predicting the likelihood of upgrading to a premium credit
card based on the monthly purchase amount and whether the
account has multiple cards” (same example used in logistic
regression)

Classification
tree example
“Predicting the likelihood of upgrading to a premium credit
card based on the monthly purchase amount and whether the
account has multiple cards” (same example used in logistic
regression)

Classification
tree example
“Predicting sales of energy bars based on price and promotion
expenses” (could be multiple regression example, too)

Regression tree
example
 Constructs models from patterns and relationships
uncovered in data
 Computations that begin with inputs and end with
Neural nets outputs
 Uses a hyperbolic tangent function
 Divide data into training sample and validation sample
“Predicting the
likelihood of upgrading
to a premium credit
card based on the
monthly purchase
Neural net amount and whether
example 1 the account has
multiple cards” (same
example used for
logistic regression and
classification tree)
“Predicting sales of
energy bars based on
price and promotion
expenses” (same
Neural net example used in
example 2 regression tree)
Classifies data into a sequence of groupings such that
objects in each group are more alike other objects in
their group than they are to objects found in other
Cluster groups.
analysis  Hierarchical clustering
 k-means clustering
 Distance measures
 Types of linkage between clusters
“Perception of sports based on a survey of these attributes: movement
speed, rules, team orientation, amount of contact”

Cluster
analysis
example
Visualizes objects in a two or more dimensional
space, or map, with the goal of discovering patterns
Multi- of similarities or dissimilarities among the objects.
dimensional  Types of multidimensional scaling
 Distance measures
scaling  Stress statistic – measure of fit
 Challenge in interpreting dimensions
“Perception of sports based on a survey of these
attributes: movement speed, rules, team orientation,
amount of contact”

Multi-
dimensional
scaling
example using
JMP add-in
“Perception of sports based on a survey of these
attributes: movement speed, rules, team orientation,
amount of contact”

Multi-
dimensional
scaling
example using
JMP add-in
 Microsoft Excel (latest versions equipped Apps for Office)
 Good for selected dashboard elements (treemap, gauges, sparklines) and
illustrating drill-down (with PivotTables) and subsetting (with Slicers)
 Extend with third-party add-ins to perform logistic regression

 Tableau Public (web-based, free download)


 Good for descriptive analytics (bullet graph, treemaps)
 Drag-and-drop interface that can be taught in minutes
Software  “Premium” version (not free) extends utility of software to many other
methods, although this server-based version is more geared to business
Resources  JMP
 Many displays have drill-down built into them
 Good for regression trees, neural nets, cluster analysis, and
multidimensional scaling (with additional free add-in)
 Requires SAS or R for some processing; user interface contains some
quirks for new and casual users (most of which could be eliminated
through the use of custom add-ins)
 Future versions promise additional capabilities.
 Could add some of the descriptive analytics into
the introductory course
 Drill down and subsetting
Can I  Perhaps one graph that summarize volume and
Incorporate velocity
Any of This  Show-and-tell to illustrate information design and/or
“sexiness” versus usefulness issue
Into the
 Could add binary logistic regression if your
Introductory course covers multiple regression and mentions
Course? binary logistic regression, but this will not be
feasible in most cases
 “Funny, you should ask that question….”
 Berenson, M. L., D. M. Levine, and K. A. Szabat. Basic Business Statistics 13th
edition. Upper Saddle River: Pearson Education, forthcoming January 2014.
 Breiman, L., J. Friedman, C. J. Stone, and R. A. Olshen. Classification and
Regression Trees. London: Chapman and Hall, 1984.
 Cox, T. F., and M. A. Cox. Multidimensional Scaling, Second edition. Boca Raton,
FL: CRC Press, 2010.
 Everitt, B. S., S. Landau, and M. Leese. Cluster Analysis, Fifth edition. New York:
John Wiley, 2011.
 Few, S. Information Dashboard Design: Displaying Data for At-a-Glance
Monitoring, Second edition. Burlingame, CA: Analytics Press, 2013.
References  Hakimpoor, H., K. Arshad, H. Tat, N. Khani, and M. Rahmandoust. “Artificial
Neural Network Application in Management.” World Applied Sciences Journal,
2011, 14(7): 1008–1019.
 R. Klimberg, and B. D. McCullough. Fundamentals of Predictive Analytics with
JMP. Cary, NC: SAS Press. 2013
 Lindoff, G., and M. Berry. Data Mining Techniques: For Marketing, Sales, and
Customer Relationship Management. Hoboken, NJ: Wiley Publishing, Inc., 2011.
 Loh, W. Y. “Fifty years of classification and regression trees.” International
Statistical Review, 2013, in press
 Tufte, E. Beautiful Evidence. Cheshire, CT: Graphics Press, 2006.
 Contact us at analytics@davidlevinestatistics.com
 Visit analytics.davidlevinestatistics.com for
Further  Today’s slides including references
Information or  A preview of some of our current work in this area
 Coming soon WaldoLands.com
Contact
 Look for our (very occasional) tweets using
#AnalyticsEducation