H T T P : / / W W W. A N A LY T I C S - M A G A Z I N E . O R G
DRIVING BETTER BUSINESS DECISIONS
J ULY / AUGUST 2016 BROUGHT TO YOU BY:
ALSO INSIDE: • Predicting 2016 U.S. presidential election • Quantifying missing link in sales analytics • Navigating big data analytics SaaS terrain • Forecasting software: new tools & trends
Rx FOR HEALTHCARE ANALYTICS • Machine learning & value-based medicine • Fighting fraud, waste, abuse with analytics • Technologies converging but hurdles remain Executive Edge Mather Economics chief Matt Lindsay on combining statistical regression with visualization
INS IDE STO RY
Prediction problems “It’s tough to make predictions, especially about the future.” A number of sources have been credited with making some version of that statement, from Danish physicist Niels Bohr to legendary baseball player/ philosopher Yogi Berra. Berra and Bohr are both right: Making predictions about the future is tough. But that’s never stopped folks from making predictions, and since we’re knee-deep in the U.S. presidential campaign, modern-day Nostradamuses and political pundits are coming out of the woodwork to make predictions regarding the outcome of the race to the White House between presumptive Republican and Democratic presidential nominees Donald Trump and Hillary Clinton. How do you predict a presidential election when one of the presumptive nominees is perhaps the most unorthodox, unpredictable candidate in recent history, when both of the leading candidates have record-high negative polling numbers, and, as of this writing, when both of the presumptive nominees face intra-party strife that could damage or even scuttle their respective campaigns by the time their party conventions are concluded? And what about the impact
2
|
A N A LY T I C S - M A G A Z I N E . O R G
that independents and third- and fourthparty candidates might have on the outcome? College professor and quantitative historian Allan Lichtman has correctly predicted the national popular vote outcome of every U.S. presidential election since 1984, and his method barely takes into consideration the candidates, their personalities, their positions on the issues or their poll numbers. Rather, Lichtman’s prediction is based on the answers to 13 key questions – the keys to the White House. Doug Samuelson, a longtime member of INFORMS, interviewed Lichtman for his take on this year’s presidential election, the turning of the “keys” and the science behind the “keys.” For more on the story, see page 28. Meanwhile, Sheldon Jacobson, another college professor and longtime member of INFORMS, also has his analytical eyes on the race for the presidential prize. Sheldon and his students maintain an Election Analytics website that tracks and analyzes polling data to forecast who will win the presidency and which party will secure control of the United States Senate. The analytics are based on Bayesian statistics and operations research methodologies. ❙
W W W. I N F O R M S . O R G
Boost Planning Agility and Watch Your Business Grow AIMMS puts the insights you need to drive innovation and business improvement at your fingertips
YOUR ERP DATA MADE ACTIONABLE
ERP systems have their limitations. The AIMMS Prescriptive Analytics platform works side by side with your ERP system and other tools to help you: • Drive Business Innovation
• Reduce Costs
• Increase Speed to Realization
• Improve Efficiency
• Get Flexibility and Scalability
• Perform Rapid Scenario Analyses
Multiple data sources, one portal for effective planning and decision support
Contact us for a demo
Email us at info@aimms.com or call 425-458-4024 Companies that add agility to their ERP systems with AIMMS include:
C O N T E N T S
DRIVING BETTER BUSINESS DECISIONS
JULY/AUGUST 2016 Brought to you by
FEATURES 28
WHO HOLDS THE KEYS TO THE WHITE HOUSE? Predicting the 2016 U.S. presidential election: What the “13 Keys” forecast, what to watch for and why they might not matter. By Douglas A. Samuelson
38
QUANTIFYING MISSING LINK IN SALES ANALYTICS Exploring the softer side of sales analytics, where data, timely messaging and the human element converge for success. By Lisa Clark
44
NAVIGATING BIG DATA ANALYTICS SaaS TERRAIN Focus on data, not infrastructure: Three things to look for, three things to avoid when starting down big data analytics path. By Brad Kolarov
48
ANALYTICS, MACHINE LEARNING AND HEALTHCARE Two essentials for success in value-based medicine to manage health, improve outcomes and keep costs under control. By Steve Curd
54
THE RADIATION BADGE FOR HEALTHCARE FWA Fighting fraud, waste and abuse using data and analytics tools is a critical step toward making healthcare affordable for all. By Rodger Smith
58
FORECASTING: 2016 SOFTWARE SURVEY New tools, new capabilities and new trends: Survey of 26 forecasting software packages from 19 vendors. By Chris Fry and Vijay Mehrotra
28
38
44
48 4
|
A N A LY T I C S - M A G A Z I N E . O R G
W W W. I N F O R M S . O R G
XLMINER®: Data Mining Everywhere Predictive Analytics in Excel, Your Browser, Your Own App
XLMiner® in Excel – part of Analytic Solver® Platform – is the most popular desktop tool for business analysts who want to apply data mining and predictive analytics. And now it’s available on the Web, and in SDK (Software Development Kit) form for your own applications.
Forecasting, Data Mining, Text Mining in Excel. XLMiner does it all: Text processing, latent semantic analysis, feature selection, principal components and clustering; exponential smoothing and ARIMA for forecasting; multiple regression, k-nearest neighbors, and ensembles of regression trees and neural networks for prediction; discriminant analysis, logistic regression, naïve Bayes, k-nearest neighbors, and ensembles of classification trees and neural nets for classification; and association rules for affinity analysis.
have in Excel, and generate the same reports, displayed in your browser or downloaded for local use.
XLMiner SDK: Predictive Analytics in Your App. Access all of XLMiner’s parallelized forecasting, data mining, and text mining power in your own application written in C++, C#, Java, R or Python. Use a powerful object API to create and manipulate DataFrames, and combine data wrangling, training a model, and scoring new data in a single operation “pipeline”.
Find Out More, Start Your Free Trial Now. Visit www.solver.com to learn more, register and download Analytic Solver Platform or XLMiner SDK. And visit www.xlminer.com to learn more and register for a free trial subscription – or email or call us today.
XLMiner.com: Data Mining in Your Web Browser. Use a PC, Mac, or tablet and a browser to access all the forecasting, data mining, and text mining power of XLMiner in the cloud. Upload files or access datasets already online. Use the same Ribbon and dialogs you
The Leader in Analytics for Spreadsheets and the Web Tel 775 831 0300 • info@solver.com • www.solver.com
DRIVING BETTER BUSINESS DECISIONS
REGISTER FOR A FREE SUBSCRIPTION: http://analytics.informs.org INFORMS BOARD OF DIRECTORS
22
68
DEPARTMENTS
2 8 14 18 22 68 74 78
Inside Story Executive Edge Analyze This! Healthcare Analytics INFORMS Initiatives Conference Preview Five-Minute Analyst Thinking Analytically
Analytics (ISSN 1938-1697) is published six times a year by the Institute for Operations Research and the Management Sciences (INFORMS), the largest membership society in the world dedicated to the analytics profession. For a free subscription, register at http://analytics.informs.org. Address other correspondence to the editor, Peter Horner, peter.horner@mail.informs.org. The opinions expressed in Analytics are those of the authors, and do not necessarily reflect the opinions of INFORMS, its officers, Lionheart Publishing Inc. or the editorial staff of Analytics. Analytics copyright ©2016 by the Institute for Operations Research and the Management Sciences. All rights reserved.
6
|
A N A LY T I C S - M AGA Z I N E . O RG
President Edward H. Kaplan, Yale University President-Elect Brian Denton, University of Michigan Past President L. Robin Keller, University of California, Irvine Secretary Pinar Keskinocak, Georgia Tech Treasurer Sheldon N. Jacobson, University of Illinois Vice President-Meetings Ronald G. Askin, Arizona State University Vice President-Publications Jonathan F. Bard, University of Texas at Austin Vice President Sections and Societies Esma Gel, Arizona State University Vice President Information Technology Marco Lübbecke, RWTH Aachen University Vice President-Practice Activities Jonathan Owen, CAP, General Motors Vice President-International Activities Grace Lin, Institute for Information Industry Vice President-Membership and Professional Recognition Susan E. Martonosi, Harvey Mudd College Vice President-Education Jill Hardin Wilson, Northwestern University Vice President-Marketing, Communications and Outreach Laura Albert McLay, University of Wisconsin-Madison Vice President-Chapters/Fora Michael Johnson, University of Massachusetts-Boston INFORMS OFFICES www.informs.org • Tel: 1-800-4INFORMS Executive Director Melissa Moore Meetings Director Laura Payne Director, Public Relations & Marketing Jeffery M. Cohen Headquarters INFORMS (Maryland) 5521 Research Park Drive, Suite 200 Catonsville, MD 21228 Tel.: 443.757.3500 E-mail: informs@informs.org ANALYTICS EDITORIAL AND ADVERTISING
Lionheart Publishing Inc., 506 Roswell Street, Suite 220, Marietta, GA 30060 USA Tel.: 770.431.0867 • Fax: 770.432.6969
President & Advertising Sales John Llewellyn john.llewellyn@mail.informs.org Tel.: 770.431.0867, ext. 209 Editor Peter R. Horner peter.horner@mail.informs.org Tel.: 770.587.3172 Assistant Editor Donna Brooks donna.brooks@mail.informs.org Art Director Alan Brubaker alan.brubaker@mail.informs.org Tel.: 770.431.0867, ext. 218 Advertising Sales Aileen Kronke aileen@lionhrtpub.com Tel.: 770.431.0867, ext. 212
Su pp ac Pow orts he e T Sp r B abl ar I a ea k n u ExcelBig Dd , at a
Ap
ANALYTIC SOLVER PLATFORM ®
From Solver to Full-Power Business Analytics in
Solve Models in Desktop Excel or Excel Online.
Plus Forecasting, Data Mining, Text Mining.
From the developers of the Excel Solver, Analytic Solver Platform makes the world’s best optimization software accessible in Excel. Solve your existing models faster, scale up to large size, and solve new kinds of problems. Easily publish models from Excel to share on the Web.
Analytic Solver Platform samples data from Excel, PowerPivot, and SQL databases for forecasting, data mining and text mining, from time series methods to classification and regression trees and neural networks. And you can use visual data exploration, cluster analysis and mining on your Monte Carlo simulation results.
Conventional and Stochastic Optimization. Fast linear, quadratic and mixed-integer programming is just the starting point in Analytic Solver Platform. Conic, nonlinear, non-smooth and global optimization are just the next step. Easily incorporate uncertainty and solve with simulation optimization, stochastic programming, and robust optimization – all at your fingertips.
Find Out More, Download Your Free Trial Now. Analytic Solver Platform comes with Wizards, Help, User Guides, 90 examples, and unique Active Support that brings live assistance to you right inside Microsoft Excel. Visit www.solver.com to learn more, register and download a free trial – or email or call us today.
Fast Monte Carlo Simulation and Decision Trees. Analytic Solver Platform is also a full-power tool for Monte Carlo simulation and decision analysis, with 50 distributions, 40 statistics, Six Sigma metrics and risk measures, and a wide array of charts and graphs.
The Leader in Analytics for Spreadsheets and the Web Tel 775 831 0300 • info@solver.com • www.solver.com
EXE CU TIVE E D G E
A picture is worth a thousand words. A regression is worth a few pictures. Combining visualization with statistical regression to identify causation.
BY MATT LINDSAY
8
|
Data visualization tools provide unprecedented access to data analysis and facilitate collaboration and the sharing of insights throughout an organization. Dashboards present data in beautiful charts and graphs that are compelling aesthetically and enable a user to quickly dive into data with filters and selection criteria. All is good, right? Maybe. Regardless of the tools being used, to produce valid predictions and effective recommendations, it is important to identify causal relationships within the data. Causality is when a change in one variable causes a change in another. Changes in price affect the quantity sold, for example. Correlation exists between variables that may not have any relationship to each other but appear to be related given similar changes over time, such as the population of the United States and the price of vodka (both are increasing over time.) When visualizing data, a graph or chart may reflect coincidental alignment, or more importantly,
A N A LY T I C S - M A G A Z I N E . O R G
W W W. I N F O R M S . O R G
Your Analytics App – Everywhere
Use Solver, Risk Solver, XLMiner in Excel Online, Google Sheets Or Turn YOUR Excel Model into a Web or Mobile App in Seconds
The easiest way to build an analytic model – in Excel – is now the easiest way to deploy your analytic application to Web browsers and mobile devices – thanks to the magic of Frontline Solvers® and our RASON® server.
Use our Analytics Tools in your Web Browser. Solve linear, integer and nonlinear optimization models with Frontline’s free Solver, and run Monte Carlo simulation models with our free Risk Solver® tool, in Excel Online and Google Sheets. Use our free XLMiner® Analysis ToolPak tool for statistical analysis, matching the familiar Analysis ToolPak in desktop Excel.
Build Your Own Apps with RASON Software. RASON – RESTful Analytic Solver® Object Notation – is a new modeling language for optimization and simulation that’s embedded in JSON (JavaScript Object Notation). With support for linear, nonlinear and stochastic optimization, array and vector-matrix operations, and dimensional tables linked to external databases, the RASON language gives you all the power you need.
Your Excel Model Can Be a Web/Mobile App. The magic begins in Excel with Frontline Solvers V2016: Our Create App button converts your Excel optimization or simulation model to a RASON model, embedded in a Web page, that accesses our cloud servers via a simple REST API. You’re ready to run analytics in a browser or mobile device! Or if you prefer, run your RASON model on your desktop or server, with our Solver SDK®. Either way, you’re light-years ahead of other software tools.
Find Out More, Sign Up for a Free Trial Now. Visit www.solver.com/apps to learn more, and visit rason.com to sign up for a free trial of RASON and our REST API. Or email or call us today.
The Leader in Analytics for Spreadsheets and the Web Tel 775 831 0300 • info@solver.com • www.solver.com
EXE CU TIVE E D G E
the effect of an unobserved third variable that is missing from the visual representation. A graph of police officers and crimes committed by jurisdiction would likely show positive correlation. The important factor that is omitted from this chart is population density, as both police and crime generally grow as population increases. This is a simple example, but there are subtler examples of how businesses can mistake correlation for causality and act on those interpretations of the data. The workhorse of analytics is regression analysis. Invented in the early 1800s, statistical regression isolates the effect of one variable on another separate from the effect of other variables. For instance, regression analysis is often used to predict the acceptance rate of a sales offer. Factors that affect an offer’s acceptance rate include price point, subscription length, payment method, the acquisition channel and customer demographics. A regression model that includes these explanatory variables could accurately measure the effect of the price point on offer acceptance unbiased by the effect of the other variables. If we exclude an important factor from the regression, our estimate of the relationship between price and the acceptance rate could be inaccurate since the model would conflate 10
|
A N A LY T I C S - M A G A Z I N E . O R G
the effect of the missing variable with the price effect. Similar to the approach taken by an analyst using regression analysis, an analyst studying offer acceptance rates using a data visualization tool must isolate the effect of one variable on another. However, instead of using model specification (the inclusion or exclusion of certain variables from a regression), the analyst using a visualization tool must accomplish the discovery of causal relationships using data filters and selection criteria. For example, if an analyst graphed the relationship between offer acceptance and price point, he may conclude that price is the most important factor for offer success. If the analyst filters that chart to only include offers made through direct mail to high-income households, the difference in acceptance rates by price point will be much smaller. Knowing what data filters are important to uncover a true causal relationship is the paramount challenge using visualization tools. It is often necessary to explore the data using alternative filters across several data fields, which can be a time-consuming process. Another example of how conclusions reached from observing data graphically can be misleading is presented using data on subscriber retention for a W W W. I N F O R M S . O R G
magazine. Graph 1 shows the percentage of subscribers that remain active over time following the start of their subscription up to 1,500 days. From the chart it is clear that there is a significant difference in retention between customers across income groups. The highest income group has about 70 percent retention at 500 days follow- Graph 1: Retention by income tier. ing their subscription start, while the lowest income tier has about 50 percent retention. The temptation to conclude that income is the most important factor for predicting retention is compelling. This publisher could elect not to solicit subscriptions among low-income households as a result. However, a plot of retention data by income group for starts from the insert channel in Graph 2 Graph 2: Retention by income tier for Insert starts. shows that differences in retention by wealthy income group in this channel income for this channel are relatively has about 80 percent retention at 500 small. In this channel, the low-income days, a much smaller difference than group has about 75 percent retention the overall retention across income levat 500 days, higher than the wealthy els. This insight suggests that income income group in the first chart. The does not have as much of an influence A NA L Y T I C S
J U LY / A U G U S T 2 016
|
11
EXE CU TIVE E D G E
on retention as the first chart indicates, and that a chart of retention by income alone is not an accurate representation of the relationship between these two variables. To further investigate the relative effects of income and channel on subscriber retention, we can plot retention curves for high-income subscribers Graph 3: Retention for high-income subscribers by acquisition channel. acquired through different channels. In Graph 3, we see that high- academia and research organizations, income subscribers acquired through the data visualization has expanded the direct channel have much lower retenpower of data across the economy. As tion than subscribers acquired in other with all tools, using it appropriately is imchannels. In addition, it is clear that the portant or the promise of the technology variation in retention by channel for highwill not be realized. The cases described income subscribers is at least as great as here provide an example of why an anathe variation in retention across income lyst using a visualization platform must levels. This plot confirms that acquisition be as careful and thorough as one using channel, in particular the direct channel, statistical regression to reach the correct is an important determinant of retention conclusions about how variables affect in addition to household income. one another. â?™ CONCLUSION Data visualization is a powerful tool that enables organizations to leverage data analytics to improve business operations. Just as increasing computing power enabled regression analysis to become widely adopted outside of 12
|
A N A LY T I C S - M A G A Z I N E . O R G
Matt Lindsay, Ph.D., is president of Mather Economics, a global consulting firm that applies proprietary analytical tools and hands-on expertise to help businesses better understand customers and, in turn, develop and implement pricing strategies. Lindsay has more than 20 years of experience in helping businesses improve performance and drive revenue through economic modeling. He holds a doctorate in economics from the University of Georgia.
W W W. I N F O R M S . O R G
THE
NATION’S FIRST
Associate in Applied Science (A.A.S.) degree in Business Analytics on campus or online.
Credential options • Enroll in one or several: • AAS degree
Why Study Business Analytics?
The Business Analytics curriculum is designed to provide students with the knowledge and the skills necessary for employment and growth in analytical professions. Business Analysts process and analyze essential information about business operations and also assimilate data for forecasting purposes. Students will complete course work in business analytics, including general theory, best practices, data mining, data warehousing, predictive modeling, project operations management, statistical analysis, and software packages. Related skills include business communication, critical thinking and decision making.The curriculum is hands-on, with an emphasis on application of theoretical and practical concepts. Students will engage with the latest tools and technology utilized in today’s analytics fields.
Accelerated Executive Program
Our accelerated learning options allow students to complete certificate credentials in two semesters part time or one semester full time. Accelerated options are available for the Business Intelligence and the Business Analyst certificates.
Questions? Tanya Scott
Director, Business Analytics
919-866-7106 tescott1@waketech.edu
• Certificates: Business Intelligence, Business Analyst, Finance Analytics, Marketing Analytics, and Logistics Analytics
Flexibility • Open-door enrollment • Courses are offered in the fall and spring • Courses can be taken online or on campus • Competitively priced tuition
Gain skills in: • Data gathering • Collating • Cleaning • Statistical Modeling • Visualization • Analysis • Reporting • Decision making
Use data and analysis tools: • Advanced Excel • Tableau • Analytics Programming • SAS Enterprise Guide • SAS Enterprise Miner • SPSS Modeler • MicroStrategy
• Presentation
Funded in full by a $2.9 million Dept. of Labor Trade Adjustment Assistance Community College & Career Training (DOLTAACCCT) grant.
businessanalytics.waketech.edu
ANALY ZE TH I S !
Mathematics: The gift that keeps on giving 30-year reunion reinforces appreciation for popular major at small, Midwestern liberal arts college.
Why was mathematics so popular at St. Olaf? While there are many reasons, the root cause is surely the faculty.
BY VIJAY MEHROTRA
14
|
In the late 20th century, I was an undergraduate at St. Olaf, a small, Midwestern liberal arts college. Earlier this summer, I returned to campus for a class reunion, excited to see my classmates and to hear about where their life journeys had taken them. We had hoped for blue skies, but instead we got mostly puffy gray clouds and scattered showers with the occasional patch of sunshine. Still, even when the sun was hidden, our old school seemed to shimmer, not only with its elegant white limestone buildings, tall trees and green grass, but also with the memories that flashed out from behind nearly every corner. On that glorious June weekend none of us felt all that far removed from college. Had it really been 30 years since we had graduated? In our graduating class of approximately 700, more than 100 of us majored in mathematics. Why was mathematics so popular at St. Olaf? While there are many reasons, the root cause is surely the faculty. Since the 1970s, the St. Olaf mathematics department
A N A LY T I C S - M A G A Z I N E . O R G
W W W. I N F O R M S . O R G
has been full of enthusiastic, energetic teachers who believed that math classes should be fun, that mathematical literacy is an essential part of a liberal arts education, and that all students should be encouraged to take math classes, even those who have no intention of becoming “real” mathematicians by going to graduate school. Over time, the results have been remarkable. From 1980 to 1989, St. Olaf graduated more than 500 students with mathematics majors, and more than 50 of us from those graduating classes ultimately completed Ph.D.s in mathematical sciences fields. This strong tradition continues; mathematics remains one of the most popular departments on campus, with about 10 percent of each graduating class majoring in mathematics. Returning to campus for this year’s reunion, I wondered about my classmates, especially my fellow math majors. Where had they gone over the past 30 years? How had the digital revolution shaped our lives and careers? Where did software, data and models fit into all of those vocational adventures? For some of us, the math department’s influence on our career was obvious. Thomas Halverson (Wisconsin) and Tamara Olson (Courant Institute) went A NA L Y T I C S
to graduate school in mathematics and became math professors, Tom at Macalaster College and Tamara at Michigan Tech. Many other classmates became math teachers at the elementary and secondary levels, including Susan Ahrendt who now educates future math teachers at the University of WisconsinRiver Falls. Like me, Hai Chu (Clemson) and Karen Donohue (Northwestern) studied operations research in graduate school. Hai’s distinguished career has featured stints with American Airlines/Sabre, Amazon.com and most recently the Walt Disney Company. Karen has been a business school faculty member, first at Wharton and most recently at the University of Minnesota. Today, Hai leads an internal consulting team at Disney that focuses on analytics and revenue management, helping bring solutions to both traditional problems and new domains. Meanwhile, Karen’s research is in behavioral operations management, sustainability and supply chain coordination, topics that were barely on the O.R. agenda when we were graduate students but are now viewed as important and fertile research areas. Another classmate gave me a visceral sense of how advances in supply chain management have impacted the J U LY / A U G U S T 2 016
|
15
ANALY ZE TH I S !
business world. Joel Anderson is the CEO of Five Below (www.fivebelow. com), a publicly traded discount retail chain, the most recent leadership role in his 20+ year-career in retail. In addition to discussing the rise of e-commerce (he had recently completed a stint as CEO of Walmart.com), Joel also described just how much real-time information is delivered to him today (“I get real time sales data delivered to my phone three times each day for all of our SKUs and all 400+ stores”) and how this data supports supplier collaboration and inventory management (“So much stuff that used to require a pile of spreadsheets before is really easy today – and optimized too.”). This observation produced a wry grin from my old roommate Jim Ford. After starting his career with Anderson Consulting (before it was rebranded as “Accenture”), Jim has spent much of the past few decades running complex IT projects. He knows just how challenging it can be to implement the systems that make things look “easy” for executives like Joel. Jim noted that while he had cut his teeth on customized corporate system development projects early in his career, his projects today feature more standardized commercial packages and/ or open source components integrated, producing different flavors of technical and managerial complexity. 16
|
A N A LY T I C S - M A G A Z I N E . O R G
John Haugen, another college math classmate, has spent the last 25 years at General Mills, where he has seen a steady shift away from mass marketing, monolithic distribution channels and high barriers to entry – and toward specialized food brands and direct-to-customer channels. His current job is vice president and general manager of 301 Inc., General Mills’ venture capital arm, where he leads General Mills’ strategic investments in emerging start-ups. The existence of such funds – several of General Mills’ traditional competitors have similar groups – reflects the fact that in today’s world smaller companies are far better equipped to more rapidly develop new products, to reach statistically identified customer groups through social media, and to leverage specialty (on- and off-line) distribution channels. None of this stuff existed when we graduated from college. Another classmate, Paul James, is an industrial designer who has been running his own studio [1] for the past 14 years. While CAD software and 3D rendering models have been around for a long time, Paul pointed out that the way in which designers use them has changed dramatically. “It used to be that software like CAD was mostly for documentation of what had been designed and produced. But today, this W W W. I N F O R M S . O R G
kind of software is central throughout the design process, from initial idea generation through manufacturing, which often features mass customization and/or 3D printing.” He credited his mathematics training not only for helping him reframe and solve technical design challenges, but also for enabling him to effectively organize logical explanations for elegant design solutions. “It’s a lot like writing a proof,’ he mused. Long before the New York Times [2] and the Harvard Business Review [3] mentioned it, we were regularly told by the St. Olaf math faculty that what they were teaching us would be valuable, directly or indirectly, in careers we could not yet imagine and in ways that we could not anticipate. And long before there was a plethora of academic programs in data science and analytics, the St. Olaf mathematics faculty was partnering with companies to provide its students with practicum projects that brought us into the joys, and terrors, of real-world applications. And while some of us have put our
training to use in traditional roles, many classmates who chose different career paths have found their undergraduate math training unexpectedly valuable in today’s increasingly digitized and datadriven world. To my classmates, especially the math majors, congratulations on making it this far in your life’s journey. I’m proud to be in your company. Best wishes for the next 30 years and beyond. To my professors, especially the math teachers, thanks for those precious few years. Looking back from the middle of middle age, I am incredibly grateful. ❙ Vijay Mehrotra (vmehrotra@usfca.edu) is a professor in the Department of Business Analytics and Information Systems at the University of San Francisco’s School of Management and a longtime member of INFORMS.
REFERENCES 1. http://www.gunlocke.com/DesignerPages/ PaulJames.html 2. http://www.nytimes.com/2009/08/06/ technology/06stats.html 3. https://hbr.org/2012/10/data-scientist-thesexiest-job-of-the-21st-century/
Request a no-obligation INFORMS Member Benefits Packet For more information, visit: http://www.informs.org/Membership
A NA L Y T I C S
J U LY / A U G U S T 2 016
|
17
HEALT H CARE A N A LY T I C S
Technologies converging but hurdles remain Deployment and adoption of technology solutions, especially digital health solutions, still struggle to pick up momentum.
With 2016 half over, we have seen many developments in the political and business world during the first six months, including the most recent Microsoft acquisition of LinkedIn. Meanwhile, the healthcare analytics space has stayed quite vibrant. According to a report published by Rock Health, a start-up accelerator turned venture fund, investment in healthcare analytics and digital health during the first quarter alone grew to nearly a billion dollars, of which $307 million was invested in big data analytics and population health management technology companies. However, deployment and adoption of technology solutions, especially digital health solutions, is still struggling to pick up momentum. In this article I will share my thoughts about the hurdles faced by healthcare organizations despite advancements and convergence of various technologies. FIVE TECHNOLOGY FORCES MODEL
BY RAJIB GHOSH
18
|
Five technology forces are now shaping our experience of the real world: the Internet of Things (IoT), social networks, mobile and big data, along with
A N A LY T I C S - M A G A Z I N E . O R G
W W W. I N F O R M S . O R G
artificial intelligence. These technologies interact with each other and will eventually converge. We are seeing the advent of more sensors and cloud-enabled intelligence within devices from thermostats on the wall to smart clothing we wear. The promise of the IoT is every humancreated object would be capable of interacting with each other. Cloud computing powered by massive server farms of Google, Amazon, Microsoft, Rackspace and many others are delivering intelligence to the mobile devices we use every day: phones, smart watches, tablets, lights at home or work, sprinkler systems, smoke detectors and appliances. The Google Fit app on an Android phone or watch can now measure steps as we take them and the types of exercise as we do them, as long as we keep the phone in our pocket or the smart watches on our wrist. Data captured through the built-in sensors are analyzed in the cloud, and glance-able dashboards present feedback seamlessly and instantaneously. Apple iPhone and Apple Watch can do similar things as well. Our phones and smart watches have become the points of convergence for mobility, IoT, cloud and analytics to measure our personal health condition and fitness activity. Big data with analytics is now everywhere. Market research firm IDC A NA L Y T I C S
Figure 1: Five forces model. predicts that revenues from the sale of big data and analytics hardware, software and services will increase by 50 percent between 2015 and 2019. By 2019, IDC forecasts sales will reach $187 billion globally. While big data is poised for a rapid growth in retail, banking and finance, uptake in healthcare is still questionable. ARTIFICIAL INTELLIGENCE The emergence of artificial intelligence (AI) is a more recent phenomenon in the consumer technology space. AI, however, is not a new technology. The first “expert system� named MYCIN that used a rule-based technology was developed in the early 1970s at Stanford University, albeit it was never used for real-life diagnosis. Advancement in computing power and focused efforts by global technology giants such as J U LY / A U G U S T 2 016
|
19
HEALT H CARE A N A LY T I C S
IBM, Google, Microsoft and Facebook have now started to move AI to the consumer world. Google and Microsoft recently open sourced their AI technology stack for the global developer community. IBM made the first big splash with Watson, its super computer-powered AI software, which they now offer as a cloud-based service. Watson is at the center of IBM’s analytics platform strategy (branded as “Platform for Cognitive Business”), which includes healthcare analytics. Google and Microsoft’s AI strategy do not include healthcare yet, but hopefully they will include it in the future. AI-based automation is a key development. McKinsey & Company estimates that as much as 45 percent of the tasks currently performed by people can be automated using existing technologies. This includes tasks within healthcare as well. HEALTHCARE A HARD NUT TO CRACK Despite many advancements and the convergence of technology forces, entrepreneurs and big companies alike are finding healthcare and biology domains difficult for disruption. We have seen many failures for digital health technologies already. Some worked initially or received media hype but then 20
|
A N A LY T I C S - M A G A Z I N E . O R G
faded or adoption died. The saga of Theranos is a sobering reminder that healthcare is different from other industry verticals. It is structurally complicated, highly regulated and layered with emotions. A nurse or a physician is not just a service provider; patients make emotional ties with them. Patients like to interact with a good doctor or a compassionate nurse rather than interacting with a computer with builtin intelligence. Even when we get our care through virtual care technologies like email, we expect the human compassionate touch of a care provider. Interestingly, the healthcare industry in the United States has always embraced cutting-edge technologies in the form of medical devices. That’s not true for information technologies. Digitization is a recent phenomenon. The true value of data has only recently been understood. Physicians now agree that data and analytics are crucial for delivering appropriate care to the right patients in a timely fashion. Demands for data are very different across departments within a healthcare delivery organization. Physicians want to have meaningful analytics delivered to them at the point of care. However, they have productivity requirements to meet, which makes most encounters only 15 minutes long. W W W. I N F O R M S . O R G
During this limited timeframe a physician has to engage in a conversation with the patient, document relevant information in an electronic health record (EHR) system using a poorly designed interface, prescribe medications or order a lab test. Where is the time to check an analytics dashboard during that process? Even if they could, what analytics would be required? What if an analytics tool is not integrated seamlessly with the legacy EHR? Quality improvement groups, on the other hand, need to examine process analytics across the organization or clinical quality measures by site and by physician. Case management groups need risk profiles of all high-risk patients and their recent healthcare utilization trends. With bigger health systems it is possible to fulfill such demands, but for smaller institutions and clinics this is an enormous challenge. Off-the-shelf solutions do not meet all requirements, and they cannot find or afford data scientists to develop analytics in house. Who will create and maintain the enterprise data warehouse? Without strong support from data scientists and data stewards, adoption fizzles out after the initial euphoria. With the change in payment models, organizations are also struggling to adequately combine cost with clinical data to generate actionable A NA L Y T I C S
insights. This is a huge undertaking for many. The healthcare industry is in the midst of experiencing big tectonic shifts. This will surely bring a lot of upheavals in the form of mergers, acquisitions and consolidation. The dust hasn’t settled yet; in fact, the dust storm has just begun. It is unclear when this behemoth of a 2.8 trillion-dollar industry will settle down with a new model of care, payment and organizational structure, and where the five technology forces will eventually converge just like they have in other industry verticals. ❙ Rajib Ghosh (rghosh@hotmail.com) is an independent consultant and business advisor with 20 years of technology experience in various industry verticals where he had senior-level management roles in software engineering, program management, product management and business and strategy development. Ghosh spent a decade in the U.S. healthcare industry as part of a global ecosystem of medical device manufacturers, medical software companies and telehealth and telemedicine solution providers. He’s held senior positions at Hill-Rom, Solta Medical and Bosch Healthcare. His recent work interest includes public health and the field of IT-enabled sustainable healthcare delivery in the United States as well as emerging nations.
Subscribe to Analytics It’s fast, it’s easy and it’s FREE! Just visit: http://analytics.informs.org/
J U LY / A U G U S T 2 016
|
21
INFO RM S IN I T I AT I VE S
Team competition, CAP and good deals The goal is to provide participants with practical experience in the complete O.R./analytics decision process.
22
|
COMPETITION FOR TOMORROW’S LEADERS IN O.R. & ANALYTICS INFORMS recently announced a new International O.R. & Analytics Team Competition, a unique student competition that provides a real-world workplace experience. Sponsored by Syngenta, the competition will feature teams of university students using identical data sets and software systems to solve a challenging business problem via an O.R./analytics approach. A panel of industry experts will judge the submissions. The goal of the competition is to provide participants with practical experience in the complete O.R./analytics decision process, including communication, leadership and teamwork skills. Teams can be comprised of undergraduate and master’s degree students associated with a university operations research or analytics degree program. The teams will be given two weeks to review the problem and, if needed, to correspond with company representatives with clarifying questions. All teams will approach the business problem as if they are
A N A LY T I C S - M A G A Z I N E . O R G
W W W. I N F O R M S . O R G
employed by the organization. Finally, each team must convince the panel of judges, acting as “management,” in a 15-minute formal presentation of the merits of their solution. Teams will be judged on a written report, team organization and planning, and the presentation. The competition will begin in September and end with team presentations at the INFORMS Conference on Analytics & Operations Research in April 2017. For more information, click here. INFORMS OFFERS FREE, THREE-YEAR CERTIFICATION RENEWAL Certified Analytics Professionals (CAP®) can renew their certification for another three years for free, and it only takes a few minutes. Simply update and/ or input your professional development units by logging into your account at https://acgi.informs.org/sso_login.php. If you ever find that your employer is asking for CAP certification (and more and more are doing it), you’ll already have it. If you want to change employers, it might help you get an interview. If you plan to retire and become a part-time consultant, it can impart a measure of trust to prospective clients.
A NA L Y T I C S
Having already spent time, effort and money acquiring the CAP, doesn’t it make sense to keep it? It won’t cost you a penny today, but it might cost you hundreds if you need it in the future. For more information, contact info@ certifiedanalytics.org. 2017 SYNGENTA CROP CHALLENGE IN ANALYTICS OFFERS $5,000 PRIZE Put your analytics skills to the test and win $5,000 in the 2017 “Syngenta Crop Challenge in Analytics” administered by INFORMS and sponsored by the Analytics Society of INFORMS. As the world population increases and arable land decreases, it becomes vital to improve the productivity of the agricultural land available. Companies such as Syngenta strive to provide varieties of their crops to meet this need. Every year farmers have to make decisions about which soybean seeds to plant given information about different soybean varieties and knowledge about the soil and climate at their respective farms. These annual decisions are critical; after a variety is planted, the decision is irreversible. Unusual weather patterns can have disastrous impacts on crops. A highly desirable variety may be in short supply and unavailable for farmers.
J U LY / A U G U S T 2 016
|
23
Photo Source: Syngenta
INFO RM S IN I T I AT I VE S
Crop Challenge: Predict which soybean seed variety or mix of varieties is more likely to be chosen by farmers. To ensure there is enough seed of the desired varieties for farmers, it is critical to evaluate which variety or varieties are more likely to be chosen by farmers from a growing region. Using the provided variety, growing region and exogenous datasets, Crop Challenge participants are challenged to predict which soybean seed variety or mix of up to five varieties in appropriate proportions is more likely to be chosen by farmers from a growing region. Timeline: Data for the challenge will be available no later than Sept. 1, 2016. 24
|
A N A LY T I C S - M A G A Z I N E . O R G
The deadline for submissions is Jan. 16, 2017. Finalists will be announced Feb. 24, 2017. Finalist presentations will be held at the INFORMS Conference on Business Analytics & Operations Research, April 2-4, 2017, in Las Vegas. For more information, click here. HOW TO GET BETTER DEALS THAN ‘MINIMUM ADVERTISED PRICE’ Certain products like Bose headphones and Sonos speakers never seem to be advertised below a certain price. That is because many manufacturers insist W W W. I N F O R M S . O R G
on a minimum advertised price (MAP) for their products. Ever wondered if it is worth searching for a lower price for such products? Using data from manufacturers across a range of industries, a forthcoming article in the INFORMS journal Marketing Science finds that it may actually pay to search as prices below the MAP are fairly common, not just across unauthorized retailers, but even across authorized retailers. The research, conducted by Ayelet Israeli of Harvard University and Eric Anderson and Anne Coughlan of Northwestern University, uses data from more than 1.25 million observations of daily online prices over a one-year period from around a thousand online retailers for over 200 product SKUs from a manufacturer with products in the electronics and music category. They find that unauthorized retailers price below MAP as much as half the time. Surprisingly, even authorized retailers price below the MAP about 20 percent of the time. And the median discount from the MAP for authorized and unauthorized retailers is around 5 percent and 13 percent, respectively. Manufacturers often use MAP to control their brand image – as A NA L Y T I C S
aggressive price cutting by retailers can negatively impact consumer perceptions of the brand. To that end, the authors surveyed manufacturer
J U LY / A U G U S T 2 016
|
25
beliefs about how retailers respond to MAP and compare this with actual retailer pricing in the market. Manufacturers rightly believe that unauthorized retailers are far more likely to price below MAP than authorized retailers. “What surprised us is that unauthorized retailers do comply with MAP about 50 percent of the time, across a variety of manu- Finding a price below MAP may not be as good for the consumer as it might appear at first glance. facturers, despite the fact that they are not bound by the manufachave only a weak association with viturers’ policies,” Anderson says. olations in the unauthorized channel,” Other manufacturer beliefs about Israeli says. “MAP enforcement efforts MAP compliance by retailers were intherefore need to be separately targetaccurate. For example, manufacturers ed toward both the authorized and the believed that retailers sell below MAP unauthorized channel.” at sites such as Amazon or Ebay, but The authors caution that finding a not at their own websites. This turned price below MAP may not be as good for out to be untrue. Manufacturers also the consumer as it might appear at first believed that MAP violations by unauglance. “Violating MAP means lowering thorized retailers created pricing presprices and hence margins. But having sure on authorized retailers to violate attracted consumers by offering them MAP. Hence, ensuring compliance by lower prices on the core product, these unauthorized retailers would induce auretailers then gain back the lost margin thorized retailers to fall in line. “But we through overcharging the consumer on find that authorized and unauthorized shipping,” Coughlan explains. retailers are largely separate, and that This means that the seemingly cheapviolations in the authorized channel est option can end up costing more. ❙ 26
|
A N A LY T I C S - M A G A Z I N E . O R G
W W W. I N F O R M S . O R G
Photo Courtesy of 123rf.com | iqoncept
INFO RM S IN I T I AT I VE S
Everyone talks about empowering business users. But only FICO delivers compelling, visually rich solutions in hours - powered by the world’s leading optimization modeling and solving platform, Xpress. Whether you run your business using complex spreadsheets or advanced analytics, FICO optimization can turbocharge your decision engine. And that makes your business users really empowered. www.fico.com/optimization
Optimize, at the speed of thought. Now it’s possible with FICO. Visit www.fico.com/Xpresstrial to download a free trial. © 2016 Fair Isaac Corporation. All rights reserved.
PO L ITIC S & A N A LY T I C S
Who holds the keys to the White House? Predicting the 2016 U.S. presidential election: What the “13 Keys” forecast, what to watch for and why they might not matter.
BY DOUGLAS A. SAMUELSON onald Trump could win the popular vote for president this year, not get elected and lead his party to disaster. The nastiest campaign in recent memory could also have one of the strangest outcomes. According to quantitative historian Allan Lichtman, Trump has a serious chance to win the popular vote. Based on his “13 Keys” model [5], Lichtman called the 2016 election inconclusive as of May, with three of the 13 variables yet
D
28
|
A N A LY T I C S - M A G A Z I N E . O R G
to be determined. “The 13 Keys model has been remarkably stable through all kinds of variations that led people to say, ‘This election is different,’ ” Lichtman says. “This record of reliability and the implications of which variables made it into the model still indicate that presidential elections are about governance, not campaigning.” Lichtman, a professor of history at American University in Washington, D.C., has been the subject of several feature articles in OR/MS Today and
W W W. I N F O R M S . O R G
Photo Courtesy of publicdomainpictures.net
Does the 13 Keys model hold the keys to the 2016 presidential election and thus the White House?
Analytics magazines [8, 9]. More significantly, he has attracted quite a bit of coverage in the mainstream media, his books continue to sell well, and it is evident that campaign strategists take his model into consideration in their planning. His model deserves to be taken seriously, as it has correctly predicted the popular vote outcome of every U. S. presidential election since 1984, including George H. W. Bush’s
A NA L Y T I C S
Donald Trump
Hillary Clinton
Photos Courtesy of Gage Skidmore | Wikipedia Commons
comeback from nearly 20 percent behind in the polls in 1988, Al Gore’s narrow popular vote win in 2000 and Obama’s 2012 reelection two years in advance [3]. His predictions are based on 13 questions, each with a “true” or “false” answer.
J U LY / A U G U S T 2 016
|
29
PO L ITIC S & A N A LY T I C S
“True” answers favor the incumbent party. If five or fewer answers are “true,” the incumbent party retains the presidency; if six or more are “false,” the challenger wins. Interestingly, with few exceptions, the 13 Keys have little or nothing to do with the perceived strengths or weaknesses of the presumed presidential nominees. (Trump’s “unorthodox” campaign will no doubt put the 13 Keys theory to a stress test this year.) THE 13 KEYS TO THE PRESIDENCY Following are the 13 Keys and Lichtman’s assessment of how they turn (as of May): 1. After the midterm election, the incumbent party holds more seats in the U.S. House of Representatives than it did after the preceding midterm election. (FALSE) 2. The incumbent-party nominee gets at least two-thirds of the vote on the first ballot at the nominating convention. (UNDETERMINED) 3. The incumbent-party candidate is the sitting president. (FALSE) 4. There is no third-party or independent candidacy that wins at least 5 percent of the vote. (UNDETERMINED) 5. The economy is not in recession during the campaign. (TRUE) 6. Real (constant-dollar) per capita economic growth during the term 30
|
A N A LY T I C S - M A G A Z I N E . O R G
equals or exceeds mean growth for the preceding two terms. (TRUE) 7. The administration achieves a major policy change during the term, on the order of the New Deal or the firstterm Reagan “revolution.” (FALSE) 8. There has been no major social unrest during the term, sufficient to cause deep concerns about the unraveling of society. (TRUE) 9. There is no broad recognition of a scandal that directly touches the president. (TRUE) 10. There has been no military or foreign policy failure during the term, substantial enough that it appears to undermine America’s national interests significantly or threaten its standing in the world. (TRUE) 11. There has been a military or foreign policy success during the term substantial enough to advance America’s national interests or improve its standing in the world. (UNDETERMINED) 12. The incumbent-party candidate is charismatic or is a national hero. (FALSE) 13. The challenger is not charismatic and is not a national hero. (TRUE) Again, if six or more of these statements are false, the incumbent party loses. W W W. I N F O R M S . O R G
For this election, Lichtman says the Democrats have lost Key 1 (the 2014 mid-term election was a setback), Key 3 (incumbency), Key 7 (major policy change) and Key 12 (the incumbent-party candidate is not very charismatic or a national hero). Key 2 (no incumbent-party contested nomination), Key 4 (significant third-party candidacy) and Key 11 (major military of foreign policy success) are still “in play.” This leaves the Republicans two keys short of what they need, with three keys undecided. In short, no prediction yet,
but those three keys will determine the ultimate winner of the popular vote, according to the 13 Keys. Key 2 looks as if it will fall against the Democrats, as Sen. Sanders has more than one-third of the delegates and seems determined to fight all the way to the convention. Lichtman also asserts that Key 2 seems to be the best single predictor. “The last incumbent-party candidate who won after a serious contest for his party’s nomination was James Garfield in 1880,” he points out. “That’s a long time.”
BIG DATA. BIG OPPORTUNITIES. 100% ONLINE
Advance your career in the world’s data-driven job market by earning your Master of Science degree or Graduate Certificate in Data Analytics. These rigorous programs are delivered 100 percent online by Oregon State University Ecampus – a national leader in online education. Learn from Oregon State faculty members who are experts in their fields and have extensive experience working with real, complex data.
DATA ANALYTICS GRADUATE PROGRAMS • Master of Science • Graduate Certificate
ecampus.oregonstate.edu/data
A NA L Y T I C S
J U LY / A U G U S T 2 016
|
31
PO L ITIC S & A N A LY T I C S
This may help to explain efforts by President Obama and Hillary Clinton to persuade Bernie Sanders to drop or at least mute his opposition to her nomination before the convention. Key 4 appeared solid for the incumbents, but the recent entry of former New Mexico Gov. Gary Johnson as a Libertarian could tip Key 4 if Johnson can get 5 percent of the vote. As of this writing, polls showed him getting as much as 10 percent, but polls can change quite a bit in five months, especially for third-party candidates. Lichtman notes that third-party candidacies generally hurt the incumbent party, regardless of the previous affiliation of the third-party contestant. Johnson is a Republican, but like John Anderson in 1980, also a Republican, he appeals to disaffected voters from both major parties. Key 11 depends essentially on how well President Obama can make the case that the Iran nuclear deal or the Paris agreement on carbon emission controls to combat climate change are major accomplishments. It is most likely no coincidence that both presumptive nominees keep talking about foreign policy accomplishments or lack thereof. Still, Lichtman observed, “For a president who was supposed to be such a great communicator, Obama hasn’t been all that 32
|
A N A LY T I C S - M A G A Z I N E . O R G
effective at communicating the benefits of his claimed successes.” As for Key 13, Lichtman says, “No matter how strong a candidate’s appeal to one-third of the electorate, if the other two-thirds despise him, he doesn’t turn the charisma key.” On the other hand, he notes, deep divisions within the challenging party – even including the replacement of the nominee, if it were somehow to happen – do not affect the forecast. THE SCIENCE BEHIND IT It is worth re-emphasizing that the 13 Keys model is based on a statistical pattern recognition algorithm. The emergence of these 13 variables implies that governance is more important than campaign characteristics. “Political consultants hate the Keys,” Lichtman says. “They keep telling me, ‘Give us something we can influence!’ But that’s not what the model indicates.” Of course, he adds, this does not mean that a candidate favored by the Keys could simply go home and await the results. Clearly the model assumes the usual sorts of campaign activities by both sides. Given that, however, the model also implies that much more of the same won’t change the outcome. It is also important to note that while massive spending on media blitzes and local organizations may not affect the W W W. I N F O R M S . O R G
Dynamic Ideas llc
The Analytics Edge provides a unified, insightful, modern and entertaining treatment of analytics. The book covers the science of using data to build models, improve decisions, and ultimately add value to institutions and individuals. Most of the chapters start with a real world problem and data set, then describe how analytics has provided an edge in addressing that particular problem. The topics covered in the book include:
• • • • • • •
IBM Deep Blue and IBM Watson Sports Analytics and the MIT Blackjack Team The Framingham Heart Study and Kidney Allocation Google’s Search Engine and Recommendation Systems Fraud Detection and Predictive Policing Revenue Management and Emergency Room Operations Asset Management and Options Pricing
In the last chapters of the book, we give a brief overview of the analytics techniques used throughout the chapters, and provide exercises to help the reader put analytics into action.
www.dynamic-ideas.com Phone: 603 882-399 Fax: 603 386-6197 Email: orders@altlog.net
DI_ANALYTICS_AD.indd 1
4/8/16 8:07 PM
PO L ITIC S & A N A LY T I C S
national popular vote much, it certainly can tip the result in a few closely contested swing states and in Senate and House races, and has substantially affected the nomination primaries and caucuses. The Keys model does not address these effects. The model’s success also underscores the unimportance of poll results this far ahead of the election. Lichtman declares bluntly, “Polls more than two months before the election are meaningless.” Indeed, as he points out, George H. W. Bush trailed Michael Dukakis by 17 points three months before the 1988 election, and Gore trailed George W. Bush in every poll up to the weekend before the 2000 election, and then only edged ahead in one, the Zogby poll. But Gore won the popular vote, as the model predicted. OTHER THINGS TO WATCH Even if the popular vote is headed Trump’s way, as the Keys model appears to imply, the electoral vote is still in play, and many Senate and House races are hotly contested as well. So here are some things that do matter: The electoral vote is different from the popular vote. The Democrats start with a group of mostly big states that have gone Democratic in every presidential election since 1992, and those states add up to 242 electoral votes [11]. It only 34
|
A N A LY T I C S - M A G A Z I N E . O R G
takes 270 electoral votes to win. Watch Florida (29 electoral votes), Pennsylvania (20), Ohio (18), North Carolina (15) and Virginia (13). The Republicans would need to carry Florida and at least two more of these states. They also need to pick up some usually Democratic states such as Wisconsin (10) and Colorado (9). Voter access, encouragement and discouragement, and differences in access to voting. In Ohio in 2004 and 2006, allocations of voting machines to precincts and provisions for replacing malfunctioning machines contributed to long waiting times, mostly in poor and minority-heavy urban areas. A few years ago, analytics-based efforts to improve the process were outlined in OR/MS Today [10]. However, the concern continues in Ohio and elsewhere, as do efforts to tighten requirements for voter registration and actual voting. Multiple lawsuits allege that these efforts are really directed at suppressing legitimate voting. Vote counting rules. In addition to his work on the Keys model, Lichtman has testified as an expert witness in more than 75 cases of alleged wrongdoing in counting votes. In Florida in 2000, he found that the way the rules for counting ballots were applied was a critical factor: Ballots with a hole punched for a candidate and the same candidate’s name written in were disqualified as double W W W. I N F O R M S . O R G
votes, even though the voter’s intent was clear. This rule resulted in the disqualification of more than 120,000 ballots, disproportionately from black voters [6]. “If black voters’ ballots had been rejected at the same rate as white voters’ ballots,” Lichtman concluded, “there would have been 50,000 more black votes statewide.” Assuming the disqualified votes would have split similarly to the black votes that did get counted, this would have tipped the election to Gore. Clearly, counting rules are also a continuing issue that will command attention [4].
A NA L Y T I C S
Ethnic blocs. In particular, Latino voters look like the swing bloc in several key states where the races look close; Florida, Ohio, Colorado, New Mexico and Nevada are most frequently mentioned by commentators on this topic. Political analysts estimate that the Republican candidate needs about 40 percent of the Latino vote to win [12]. George W. Bush drew 44 percent of the Latino vote in 2004, John McCain received 31 percent in 2000 and Mitt Romney collected just 27 percent in 2012. Recent polls have Trump below 20 percent. If this pattern holds, the anti-anti-immigration
J U LY / A U G U S T 2 016
|
35
backlash could cost the Republicans a number of Senate seats, as well as the electoral votes of enough states to cost Trump the presidency. Targeting appeals to selected groups is a wellknown and long-standing (more than 50 years old) campaign method Big spending could create a backlash from people over-saturated with ads. that has had considerable impact, particularly in more localized house-to-house canvassers. Tracking elections, such as those for House dis“how much is enough and how much is tricts [7]. In some cases, however, targettoo much� will be an interesting challenge. ing also raises questions of propriety and The self-serving professional class. of possible undermining of the integrity of One reason candidates keep raising and the political process. Negative messages, spending more and more money, for what either localized or on a large scale, risk a the Keys model indicates is of little benefit, phenomenon known in sales as market is because all the fund-raisers, political saturation; people simply get disgusted consultants and pundits keep telling them with both sides. There is a stronger, more they have to. This does, however, generspecific backlash against negative mesate considerable benefits for the fundsages subsequently proven to be false. raisers, political consultants and pundits. The effects of massive spending. Eugene Burdick, a highly regarded politiBig money spent on media blitzes and on cal scientist in addition to his stellar career organizational activity may not affect the as a novelist, pointed out most of the applinational popular vote by much, but it can cable ethical issues in his 1956 and 1964 and does influence primaries, Senate and novels [1, 2]. Naturally, these people are House races and the outcomes in a few also among the first and most committed closely contested swing states. Beyond a to argue against any legislation limiting the certain point, however, big spending could money or its uses in advertising, just as the also create a backlash from people overmajor manufacturers of voting machines saturated with the ads, phone calls and are among the first and most committed 36
|
A N A LY T I C S - M A G A Z I N E . O R G
W W W. I N F O R M S . O R G
Photo Courtesy of 123rf.com | Sunagatov Dmitry
PO L ITIC S & A N A LY T I C S
to raise all kinds of creative objections to any strict standards on the reliability and tamper-resistance of voting machines.
1. Burdick, Eugene, 1956, “The Ninth Wave,” Houghton Mifflin.
CONCLUSIONS
2. Burdick, Eugene, 1964, “The 480,” McGrawHill.
While political methods and tactics continue to change, some factors seem fairly reliable over the long term in enabling us to predict who will win. Barring major new events, these factors favor Donald Trump’s election, at least in the popular vote. However, Clinton seems somewhat the better bet to win the electoral vote, and both parties are expected to campaign intensely in the half-dozen swing states. TV advertising is likely to have little practical effect other than to annoy most voters. Issues of voter access and bloc voting could substantially affect the electoral vote, however, and potentially cast doubt on the fairness of the election. OR/MS analysts, regardless of political preferences, would do well to learn about the analytical methods and issues that contribute to improving elections and keeping them credible. ❙ Doug Samuelson (samuelsondoug@yahoo.com), a frequent contributor to OR/MS Today, is president of InfoLogix, Inc., a consulting company in Annandale, Va. Samuelson worked as a paid campaign staffer in a U.S. Senate campaign in Nevada in 1970, as a county coordinator in a gubernatorial campaign and targeting analyst for a local campaign in California in 1974, and as a Federal Civil Service policy analyst from 1975 to 1982. He is a longtime member of INFORMS.
A NA L Y T I C S
REFERENCES
3. Csetlar, Maralee, July 12, 2010, “Scholar’s 13 Keys’ Predict Another Obama Win,” http:// www.american.edu/media/news/20100712_ Lichtman_Predicts_Obama_Wins_ Reelection_2012.cfm 4. Kyle, Susan, Samuelson, D., Scheuren, F., and Vincinanza, N., Winter 2007, “Explaining the Differences Between Official Votes and Exit Polls in the 2004 Presidential Election,” Chance. 5. Lichtman, Allan J., 2016, “Predicting the Next Presiden: The Keys to the White House,” 2016 Edition, Rowman & Littlefield Publishers, Lanham, Md. 6. Lichtman, Allan, January 2003, “What Really Happened in Florida’s 2000 Presidential Election,” Journal of Legal Studies, Vol. 32, p. 221. 7. Lichtman, A. J., and Keilis-Borok, V. I., 1981, “Pattern Recognition Applied to Presidential Elections in the United States, 1860-1980: Role of Integral Social, Economic and Political Traits,” Proceedings of the National Academy of Science, Vol. 78, No. 11, pp. 7230-7234. 8. Samuelson, Doug, June 2011, “Election 2012: The 13 Keys to the White House,” OR/MS Today. 9. Samuelson, Doug, September-October 2012, “Election 2012 Update: The 13 Keys’ to the White House,” Analytics. 10. Samuelson, Doug, Ted Allen and Michael Bernshteyn, December 2007, “The Right Not to Wait: Forecasting and Simulation Reduce Waiting Times to Vote,” OR/MS Today. 11. www.270toWin.com 12. www.nbcnews.com/news/latino/gop-2016-winwill-need-more-40-percent-latino-vote-n394006
Help Promote Analytics Magazine It’s fast and it’s easy! Visit: http://analytics.informs.org/button.html
J U LY / A U G U S T 2 016
|
37
MIS S IN G ME T R I C
The human side of sales analytics Guided by people KPIs alongside activity metrics, analytics are transforming sales management.
BY LISA CLARK he days of thinking about sales professionals as gladhanders who achieve results through client lunches and wheedling are over. As data-driven sales and marketing approaches take off, a new breed of sales staff is helping category-leading organizations achieve their audacious revenue targets via data science, predictive analytics and automation. The trend is giving rise to an exciting field of new sales tech that already is showing positive returns.
T
38
|
A N A LY T I C S - M A G A Z I N E . O R G
The shift has arrived not a minute too soon. Data from the Corporate Executive Board suggests that 57 percent of the purchase decision is made before the buyer even engages a sales rep. Ready online access means buyers do their homework on product features and benefits well in advance. Markets are burgeoning with choices, and new competitors are cropping up all the time. Differentiation is increasingly more difficult to achieve – and maintain. By the time the sales rep calls, the buyer expects a
W W W. I N F O R M S . O R G
Photo Courtesy of 123rf.com | Dzianis Apolka
Missing from CRM’s powerful set of metrics is the one intangible that can make or break pipeline and revenue goals: the capabilities of sales people. consultative discussion with thoughtful context on how the company’s solution uniquely solves their pain. Customer relationship management (CRM) software is the most common way to report on sales process data, particularly in B2B companies. Despite widespread use of CRM, sales activity level reporting and other process automation is only the starting point for monitoring what’s important in driving revenue. The “rearview” results they deliver are, by definition, lagging indicators of sales performance. In the past 18 months or so, a new set of sales analytics technologies have moved to the top of the “sales stack.” They are a group of interoperable
A NA L Y T I C S
technology applications, often integrated with CRM that can support sales enablement functions ranging from content delivery and account planning to sales capabilities measurement and management. They are driving a renaissance in sales technology that has been likened to the early days of the marketing technology boom that gave rise to powerful automation platforms, such as Eloqua and Marketo. Industry analysts have identified more than 60 sales stack solutions in nine sales categories [1] from which organizations can choose to address potential gaps that might exist in their sales process, or simply get ideas based on specific sales process needs.
J U LY / A U G U S T 2 016
|
39
SAL ES AN A LY T I C S
THE MISSING METRIC: PEOPLE Missing from CRM’s powerful set of engagement, process and activity metrics is the one intangible that can make – or break – pipeline and revenue goals: the capabilities of sales people. B2B selling is still reliant on human interaction somewhere along the way to inform, differentiate, clarify and ultimately close the deal. While most sales analytics solutions in the stack can tell managers and executives what and when their sales reps are doing various activities, and even the next, best prospect to call on, they do not reveal how well-suited the sales people are to even win a deal. Among the most interesting new categories in the “stack” are sales capabilities platforms, which offer both prescriptive and predictive metrics on a company’s human capacity to win deals, including real-time market and industry knowledge levels, sales confidence or proficiency with the unique steps of an organization’s sales process. These analytics quantify the human factor – as in, the capabilities of the people who are the sales reps interacting with your customers – that is often the missing link in the analytics view of the sales profession. Sales capabilities platforms recognize that there’s an intangible human experience that takes priority over a gazillion data points and often unites the various layers of the “stack” so they work as a cohesive whole. 40
|
A N A LY T I C S - M A G A Z I N E . O R G
THE POWER OF ‘PEOPLE KPIs’ WITH CRM DATA In early 2015, a $3 billion global Internet company headquartered in San Francisco’s Bay Area, undertook a sweeping global initiative to improve sales performance through a data-driven coaching program. Data from the range of sales acceleration solutions employed by the company became the basis for synthesizing individualized coaching actions. Missing from their strategy to arm front-line sales managers with real-time insights was some kind of metric on the skill set of the sales reps involved. To address this, the company established a coaching dashboard that tracks key performance indicators (KPIs) for each sales rep. The framework takes data from three sources: 1) sales-related attainment metrics from its CRM system, such as individual quota attainment and win rate; 2) metrics for selling skill sets as observed and documented by the sales coach; and 3) sales rep proficiency and engagement metrics from a sales capabilities platform such as Qstream that analyzes data from thousands of sales rep responses to scenario-based challenges presented to sales reps. These scenario challenge responses reflect sales rep proficiency at critical selling capabilities from prospecting, W W W. I N F O R M S . O R G
countering objections, addressing questions while following regulatory guidelines of various industries, and closing. Using the dashboard, sales coaches were able to quickly target and prioritize their 1:1 coaching plans. The dashboard also rolled up overall team skills, summarizing group performance so team-wide gaps in required knowledge or behavior can be quickly identified and addressed. At the sales rep level, the dashboard presents a personalized “rubric” specific to their performance and other qualitative
measures based on 1:1 input from their coaches. Now 1:1 meetings run more efficiently and help surface potential issues faster. The use of a sales capability metrics sends a subtle message that 1:1s are not simply pipeline reviews. It also documents accountability for both managers and reps. NAVIGATING DYNAMIC MARKET CHANGE A provider of automated solutions for securing open source software, Black Duck Software’s sales reps need to have
http://meetings.informs.org/analytics2017 A NA L Y T I C S
J U LY / A U G U S T 2 016
|
41
SAL ES AN A LY T I C S
a broad understanding of the open source movement, legal compliance, security management and software development markets to adequately address the needs of its C-suite customers around the globe. In addition, reps must be able to communicate clearly with executives in functions outside of IT and across a range of industries. In addition to the many sales enablement tools Black Duck employs in its sales stack, including an internal knowledgebase and dialogs where various open source topics are discussed, its sales executives sought a more engaging, scalable and continuous way to ensure to manage and measure the core capabilities of its sales reps as market shifts occurred. Black Duck Software also employed a sales capabilities platform. The system gathered hundreds of thousands of data points through simple, scenario-based challenges delivered to sales reps’ mobile devices on a range of security issues and core messaging they needed to be aware of when speaking with customers. System-generated dashboards provided sales management with a snapshot of rep competencies and individualized coaching actions synthesized from response data. A 98 percent engagement rate – a metric on how often and quickly the reps responded to the challenges – showed that Black Duck sales reps were tuned 42
|
A N A LY T I C S - M A G A Z I N E . O R G
in to the challenges daily and provided insights into rep confidence levels. As a result, the overall proficiency score on security topics, for example, rose from a baseline of 76 percent to 91 percent within weeks. Sales managers were enthusiastic about being able to see how their teams were performing using sales capabilities data, and expressed satisfaction that the teams were getting the reinforcement they needed to successfully manage customer questions. Through data-driven insights, sales capabilities platforms take a quantitative approach to a process most often qualitatively measured. The opportunity to quantify the human side of selling and better manage via analytics in this way is transformative, exciting and a dramatic change to the way that sales management is conducted. It’s just one of the examples of ways that analytics are rewriting entire business functions to make them smarter and more efficient. ❙ Lisa Clark is vice president of marketing at Qstream, a provider of a mobile sales capabilities platform used by leading brands in technology, financial services and life sciences, including 14 of the world’s top 15 pharmaceutical companies, to manage the effectiveness of their sales channels and front-line managers.
REFERENCE 1.
https://smartsellingtools.com/build-a-kick-asssales-technology-stack-step-one/
W W W. I N F O R M S . O R G
S S E N I BUS TICS Y L A N A
Online
MBA
BECOME MORE AT THE
Beacom School of Business Best Value MBA Ranked Top
Top Rated College by Forbes & Princeton Review
10
AFFORDABILITY & ACCREDITATION by Best Master’s Degree
Online MBA Ranked Top
25
IN THE WORLD
by Princeton Review
MBA – General MBA – Business Analytics MBA – Health Services Administration
Get started at
www.usd.edu/onlinemba cde@usd.edu • 800-233-7937
ANALY TICA L J O U R NE Y
Navigating the big data analytics SaaS terrain Focus on data, not infrastructure: Three things to look for, three things to avoid.
BY BRAD KOLAROV
W
ith the continued hype from so many big data companies, it is hard to understand the best way to start down the big data analytics path. After all, the reason we use big data software tools is to improve our data analytics, not to see if we can get the latest and greatest big data tool to work. We have seen too many “square-peg” solutions pounded into “round-hole” problems. This article should educate those data consumers focused on solving analytic challenges, those who have yet to start down the path of big data analytics, those who are stuck in the middle of that journey 44
|
A N A LY T I C S - M A G A Z I N E . O R G
– or the ones who have made it through, but are ready to move to more advanced analytic frameworks. NAVIGATING THE TRADITIONAL TERRAIN Traditional big data service offerings typically cover a single capability in a range of business needs. Some companies simply make it easier to spin up a Hadoop cluster. Others offer proprietary algorithms to track or uncover patterns in data. Still others provide an aggregation platform for these services and more, all under one roof. Regardless of your need, one of these types of Software as a Service (SaaS) W W W. I N F O R M S . O R G
A NA L Y T I C S
J U LY / A U G U S T 2 016
|
45
Photo Courtesy of 123rf.com | Dirk Ercken
offerings can help your business get started when it comes to standing up an enterprise-level, cloud-based big data analytics capability, but they consistently fall short in solving your analytic needs. These kinds of distributed processing systems are notoriously hard to system-engineer. They require continual interaction between the IT department, software developers and internal end-user data analysts. These systems could easily add weeks The path toward big data analytics can take many twists and turns. or months to the time it takes for developers to gain access to a Hadoop infrastructure or analytics models to be cluster. (And the larger the cluster, the built in a self-service environment. longer it may take to get from IT to the Developers can now go to a website developers.) that provides a click-through portal for The next generation of big data access to resources they need, based on analytics tools automates these hardcustomized patterns they define. With a to-system-engineer steps. Through aufew clicks of a mouse, they can have a tomation, developers can gain access dedicated space in their cloud, and one to a Hadoop cluster almost immediateor many big data stacks provisioned for ly, as opposed to the unwieldy lengths them. A few clicks more and they can auof time it might take through conventomatically ingest data and information tional channels. into the data stacks they’ve created, all with the confidence of cloud-based security to protect sensitive enterprise data. NEXT GENERATION OF BIG DATA ANALYTICS This way of working clearly facilitates These new SaaS services have better, faster interaction between develmade automating processes almost opers and IT. That improved interaction push-button easy, allowing quintessential in turn makes it easier and faster for data
B IG DATA A NA LY T I C S
analysts to ingest data and begin gaining critical business insights. Better still, automation offers a level of resilience and creates more robust big data systems, which are traditionally the most fragile part of an IT environment. Of course, not all of these new SaaS products are created equal, and it is very difficult to cut through the marketing façade. Users need to be sure that they’ve chosen the right one for their purposes. Below are a few things to keep in mind and a few to avoid when deciding on which platform is right for you. Three things to look for in big data SaaS: 1. A platform with comprehensive offerings. Most companies need more than just Hadoop and Spark, even if the system can spin up these services in minutes. You should find a SaaS provider that gives you a choice of a broad range of tools with different functions (Kafka, Elasticsearch, Zeppelin, etc.) – but doesn’t make you use them all. This will allow you to fully customize the way your company interacts with data, without having to take on a full load of unnecessary tools. The more options, the more you can do with your data, which is, after all, the point. 2. Systems that automatically ingest data. Provisioning clusters is a relatively easy process and not particularly new to 46
|
A N A LY T I C S - M A G A Z I N E . O R G
the industry – especially for applications like Hadoop. Once you’ve spun up that cluster, though, you need an equally effective system to bring in your enterprise data and start doing the real analytics work. 3. Transparency in security. The data clusters you build should be in your own environment. Consequently, your infrastructure should be securely hosted in your cloud accounts where you have full access and full accountability for your infrastructure and data. THREE THINGS TO AVOID: 1. Proprietary, “black box” software. One key advantage to open source is the vast choice and transparency involved in analyzing your data. Some companies may require you to download a full suite of open source or proprietary software to work with your enterprise data. If that works for your enterprise, great. Most companies, however, find that this approach undermines the entire reason of working with open source software in the first place. In general, it’s better to find a provider that allows you to sidestep proprietary software or distributions and launch right away in your cloud. 2. Big data solutions that consume your data. Data is the most sensitive part of the equation when it comes to using a SaaS system with confidence. Make sure that your provider does not W W W. I N F O R M S . O R G
consume or escrow your data. And avoid any solutions that may host your data in their own cloud. 3. Bleeding edge. Know the difference between cutting edge and bleeding edge. Some online applications are simply not ready yet for the enterprise, so building a Hadoop cluster on one of these systems may cause more problems than it solves, despite the cool factor of saying you use the technology. Make sure the provider you choose gives you access to open source tools that are widely adopted and well understood by enterprise
customers from a security, performance and cost perspective. Automation holds the key to fast development of enterprise big data capability, and today’s SaaS offerings have many levels of automation. Make sure you pick the system that’s right for your current needs – and can grow with your enterprise as those needs change. ❙ Brad Kolarov is managing partner of Stackspace, a big data technology company that simplifies data analysis for faster business decisions. He can be reached at brad@stackspace.io.
https://www.pathlms.com/informs/events
video learning center analyze | organize | optimize VISIT THE UPDATED INFORMS VIDEO LEARNING CENTER AND WATCH CASE STUDIES OF AWARD-WINNING ANALYTICS PROJECTS, SUCH AS: • WINNER: UPS ORION Project • 360i's Digital Nervous System Generates $250MM in Cost Savings and $1B in Revenue Creation for Clients • O.R. Transforms Scheduling of Chilean Soccer Leagues and South American World Cup Qualifiers • Transition State and End State Optimization Used in the BNY Mellon U.S. Tri-Party Repo Infrastructure Reform Program • The New York City Police Department's Domain Awareness System • Bayesian Networks for U.S. Army Electronics Equipment Diagnostic Applications: CECOM Equipent Diagnostic Analysis Tool, Virtual Logistics Assistance Representative • 2016 UPS Smith Prize Winner Reprise: Carnegie Mellon University, Heinz College • And many more!
https://www.pathlms.com/informs/events
A NA L Y T I C S
J U LY / A U G U S T 2 016
|
47
HEALT H CARE A N A LY T I C S
Analytics and machine learning Two essentials for success in value-based medicine
BY STEVE CURD
T
he U.S. healthcare system is well on its way in the transition to value-based payment models that reward providers for delivering quality outcomes and keeping patients healthy. In fact, as of March 2016, the Department of Health and Human Services reported that an estimated 30 percent of Medicare payments were already tied to these new alternative payment systems. Value-based programs are replacing traditional fee-for-service models that pay providers based on the number of services delivered. The newer models are designed to encourage care that is 48
|
A N A LY T I C S - M A G A Z I N E . O R G
well-coordinated, cost-effective and lead to quality patient outcomes. In order to achieve new payment objectives, providers are seeking opportunities to engage patients in their own care, improve patient satisfaction and keep patients healthier. Keeping patients healthy is critical not only for individual providers, but the U.S. economy as a whole. In 2012, about half of all adults suffered from at least one chronic disease such as diabetes, cancer, congestive heart failure (CHF) or chronic obstructive pulmonary disease (COPD). One in four adults had two or more chronic conditions. Treatment of individuals with chronic disease is expensive and accounts for 86 W W W. I N F O R M S . O R G
Photo Courtesy of 123rf.com | everythingpossible
percent of the nation’s total healthcare costs. Under value-based care models, providers must proactively manage the health of individuals with chronic illness to curtail costly complications that can lead to hospitalization, hospital readmission and/or early death. Many chronic diseases are linked to unhealthy Under value-based care models, providers must proactively manage the behaviors, such as lack health of individuals with chronic illness to curtail costly complications. of physical activity, tobacco use and events, which in turn facilitates custompoor nutrition. Providers are motivated to ized care processes, earlier interventions closely monitor these behaviors and take and fewer complications. action to keep patients healthy. TRADITIONAL CARE MODELS The proliferation of smart, wearable health devices and remote monitoring Traditional fee-for-service (FFS) paysystems afford providers more opporment systems reimburse providers based tunities to assess their patients’ health on the delivery of services. For example, outside of traditional care settings. If unwhen a CHF patient has an office visit, healthy behaviors are identified, providthe physician is reimbursed for treating ers can attempt to engage patients in the patient. If for some reason the pathe care process and help them make tient has complications and has to seek healthier choices. While changing patient treatment in the ER, the hospital and the behavior is not an easy task, new techtreating physician are reimbursed for that nologies are helping providers motivate episode of care. their patients to be more engaged in their Providers are not paid to monitor the own care. Other emerging technologies patient’s health between visits and have are leveraging sophisticated algorithms no financial incentive to make sure the and machine-learning technology to help individual is living a healthy lifestyle and providers identify and predict adverse following the recommended treatment A NA L Y T I C S
J U LY / A U G U S T 2 016
|
49
HEALT H CARE A N A LY T I C S
plan. In a FFS world, providers have little need for advanced analytics to predict outcomes or customize treatments, or for remote monitoring technologies to track patients’ health and behaviors. NEW MODELS DRIVE NEED FOR NEW TECHNOLOGIES The healthcare industry often lags behind other sectors when it comes to technology adoption. However, the shift to value-based medicine is fueling the need for innovations that help providers achieve payment objectives. Remote health monitoring systems, for example, have become increasingly sophisticated in recent years. These systems give providers the ability to monitor vitals, symptoms and other health data between office visits. A variety of mobile apps allow users to record their health information manually or via Bluetoothenabled devices, and then transmit details to their providers. If the collected data suggests a decline in health, providers can take corrective action early, before complications ensue. Ultimately these technologies help reduce the care costs associated with unhealthy lifestyles and chronic conditions, and enhance the quality of life for individuals. In the past, providers have rarely been paid for remote care services or for monitoring patient health between visits. 50
|
A N A LY T I C S - M A G A Z I N E . O R G
Today, however, Medicare and most Medicaid programs cover some telehealth services, as well as certain non-face-toface services under Medicare’s Chronic Care Management Services CPT billing code. As these technologies continue to mature, look for expanded coverage for remote patient monitoring services. Another innovation gaining wider acceptance is technology that enhances patient compliance of prescribed medical regimens. For example, by using personality assessment tools that are customized for healthcare, providers can evaluate an individual’s risk of non-compliance and then tailor treatment plans and communication styles based on each patient’s personality. Certain patients, for instance, may respond better to caregivers who communicate in a direct and matter-of-fact style, while others are more motivated by caregivers who adopt a more personal and sympathetic approach. When a patient’s unique personality is taken into account, communication is enhanced, and the patient is more motivated to remain engaged in the prescribed care plan. Outcomes are ultimately enhanced, as is patient satisfaction. MACHINE LEARNING AND PREDICTIVE ANALYTICS Machine-learning technologies and predictive analytics have been utilized for W W W. I N F O R M S . O R G
decades across a number of industries. In recent years, the healthcare sector has begun adopting these technologies for a variety of applications, including chronic disease management, staffing predictions and population health risk assessment. Analytics provide valuable insights into the health of an individual based on collected data and contextual information. Over time, as additional data is captured and available for analysis, these insights become more precise. In the world of value-based medicine, this type of data
is critical for predicting the likelihood of adverse events so that caregivers have adequate time to enact proactive measures that enhance outcomes. Furthermore, the utilization of machine learning allows providers to gain insight into the effectiveness of existing programs and protocols and identify the treatments and interaction styles that yield the best results for specific patients with specific conditions. This customized approach to care is at the cornerstone of the Precision Medicine Initiative, which seeks to tailor medical decisions,
SAVE THE DATE
HEALTHCARE 20 7
OPTIMIZING OPERATIONS & OUTCOMES July 26–28, 2017 Rotterdam, The Netherlands
HTTP://meetings.informs.org/healthcare2017
A NA L Y T I C S
J U LY / A U G U S T 2 016
|
51
HEALT H CARE A N A LY T I C S
practices, and/or products to the individual patient based on each person’s genetic makeup, environment and lifestyle. A VALUE-BASED MEDICINE EXAMPLE Consider the care process for an individual with CHF when provider payments are dependent on the delivery of quality outcomes and care that is cost-effective and well-coordinated. Regardless of whether the person was first seen in the physician’s office or the hospital, the provider’s primary goal would be to stabilize the patient’s health and minimize the risk of high-cost care in the hospital. Traditionally a CHF patient may have office-based visits with physician exams two to 12 times a year and lab tests and echocardiograms as needed. Between visits, the individual is advised to manually track his weight on a daily basis, and keep a record of symptoms, physical activity and diet. If any significant changes are noted, the patient is told to contact his physician. If the patient alternatively has access to remote monitoring devices, much of the tracking and communication can be automated. Sensor technologies and wireless communications can capture patient health data in real time. Then, cloud-based analytics can be deployed to evaluate current and historical vitals 52
|
A N A LY T I C S - M A G A Z I N E . O R G
and symptom data alongside contextual data. Caregivers can extract insights to predict adverse events and assign risk scores that reflect the individual’s condition. Providers can tailor future care based on the patient’s specific results, and/or take proactive measures as necessary. Because the information is processed in real time, providers are able to take action earlier and prevent emergency situations. CONCLUSION As healthcare continues its shift to value-based medicine, providers will need new technologies to manage the health of their patient population, improve outcomes, and keep costs under control. This is especially true when caring for individuals with chronic conditions. Analytics and machine-learning technologies are two innovations that will be essential for providers seeking to improve the quality of care. By incorporating these technologies into traditional care processes, providers will be better equipped to address the unique needs of individual patients and proactively manage their care. ❙ Steve Curd is CEO of Wanda, a company dedicated to advancing the effectiveness and efficiency of medicine by using machine learning in place of conventional technologies and by enabling clinicians to make more informed care decisions.
W W W. I N F O R M S . O R G
To-Do-List Go to the gym Begin to eat healthy
EXPAND YOUR NETWORK!
N I O J S M R O F IN Is the largest association for analytics in the center of your professional network? It should be. • INFORMS Allows You to Network With Your Professional Peers and Others Who Share Your Interests • INFORMS Connect, the New Member-only, Online Community Lets You Network with your colleagues • INFORMS Provides Unsurpassed Networking Opportunities Available in INFORMS Communities and at Meetings • INFORMS Offers Certification for Analytics Professionals • INFORMS Helps You Take Leadership Roles to Help Build your Professional Profile • INFORMS Career Center Provides You with the Industry's Leading Job Board
Join Online Today! http://join.informs.org
HEALT H CARE F RAU D
Analytics: The radiation badge for healthcare FWA BY RODGER SMITH
W
hile it has been a problem since the beginning of modern healthcare, in the past few years public focus on fraud, waste and abuse (FWA) in healthcare has grown substantially, as has the amount of money lost in the U.S. healthcare system as a result. Recent testimony to Congress by Ann Maxwell, assistant inspector general for evaluation and inspections at the Office of Inspector General (OIG), U.S. Department of Health and Human Services (HHS), reported “the estimated improper payments for Medicare and Medicaid to be approximately $88.8 billion” in FY2015. Think of all the healthcare that could be purchased for $88 billion. Then consider 54
|
A N A LY T I C S - M A G A Z I N E . O R G
that while the Centers for Medicare and Medicaid Services (CMS) is the country’s single largest health payer, it’s still just one. That $88.8 billion figure doesn’t take into account all of the dollars lost by private payers and their clients. What makes it particularly frustrating (and attractive to those who do it intentionally) is that FWA can be very difficult to detect, but very impactful. Think of it like a small radiation leak. With radiation there are no obvious signs; often there is nothing to see, hear, taste, touch or smell. But if it’s not found and stopped, over time the effects can be deadly. That’s why employees in facilities that handle radioactive material wear badges that constantly check and display W W W. I N F O R M S . O R G
A NA L Y T I C S
J U LY / A U G U S T 2 016
|
55
Photo Courtesy of 123rf.com | James Steidl
cumulative radiation exposure. These badges can quickly uncover the presence of excessive radiation, either now or in the recent past, to determine whether there is an issue with a nuclear reactor or if there is a problem with equipment using radiation (such as X-ray machines). The badges help people avoid the harmful effects of The healthcare FWA equivalent to the radioactive detecting badge is analytics. excessive radiation. The healthcare FWA equivalent to the that incorporate diverse data sources are radioactive detecting badge is analytics. a key component of modern approaches An advanced data analytics package can and give payers a needed edge. constantly monitor expenditures for healthTypically, payers hire experts and care services, using large volumes of data build rules and processes around known drawn from a multitude of sources, and deand suspected issues. There are several tect subtle patterns or anomalies indicating problems with this approach. possible FWA that would not be obvious to First, the schemes and errors change a human using manual means of data infrequently. Physicians, practice managspection. At the same time, advanced data ers and medical billers do not receive analytics can also recognize certain mitibilling training in school. Similarly, codes gating events and reduce the frequency of and coding rules change as new services false positives so health payers can deploy come to market and treatment protocols their resources most effectively. change. The result is varied and changing errors. Meanwhile, sophisticated players THE MANY FACES OF FWA know that the industry is watching and go FWA runs the gamut, from honest to great lengths to obscure what they’re mistakes that result in one-off overpaydoing. The system’s inherent structure of ments to highly sophisticated criminal trust enables both simple billing errors enterprises stealing millions of dollars. and illicit actors to hide in the shadows of Payers must have strategies to address the murky deep as overpayments quietly each of these. Advanced data analytics siphon money away from legitimate care.
HEALT H CARE F RAU D Of course, fraud involves a misrepresentation of some key fact or event. When repeated misrepresentations are made, they create patterns that can be detected when compared to legitimate claims. Similarly, erroneous claims do not look exactly like valid ones, even within legitimate clinical variations. Advanced analytics that digest many different data sources give payers the means to look at benchmark patterns and results, and identify claims and patterns of billing sufficiently different to merit review. The best analytics are also adept at eliminating false positives so provider audit groups and special investigation units can focus their efforts where they are most likely to yield results. EXPANDING DATA SOURCES Traditional analysis has primarily employed claims data, since it already includes substantial detail. However, claims data can be incorrect or paint a misleading picture because it is incomplete. For example, suppose claims data shows a general surgeon in upstate New York is filing claims for more complex procedures than other surgeons in her region. A simple analysis might flag that provider. A different conclusion is reached, however, if by incorporating credentialing and geographical data into the analytics, the payer discovers she is the only hand surgeon in a 100-mile radius. Since she receives
56
|
A N A LY T I C S - M A G A Z I N E . O R G
cases that are typically more complex than other surgeons, the higher intensity of claims makes sense. Similarly, if a provider wants to prescribe a high-cost drug to a patient in a low-income area to make a substantially higher-than-appropriate profit, he or she may waive the member’s cost share. The provider’s actions undermine the incentive for using the alternative drug and lead to a substantial inappropriate payment. By only using claims data, this subterfuge may go undiscovered. But by analyzing demographic, geographic and other data about the member, the payer will realize it is virtually impossible for any member treated by that provider to pay the cost share, highlighting possible improper activity. DISCOVERING LINKS Another way payers can use advanced analytics to uncover FWA is by analyzing links between multiple providers. After all, if one nuclear reactor is having a radiation issue, it’s prudent to check all the others that were manufactured at that time by the same company. If Provider A is involved in improper billing, it is not uncommon for other providers with which they associate to also be engaged in bad behavior. Thus, many payers work to analyze connected providers. Information on corporate ownership, billing and management companies, social
W W W. I N F O R M S . O R G
media interactions of physicians and staff can reveal whether other physicians, pharmacies, radiology centers, home infusion agencies, etc., are engaged in a broader pattern of referral and collusion. Rather than relying on the current or known state, advanced analytics can look at patterns and behaviors that vary from industry benchmarks, or Office of the Inspector General standards, or even what other providers in the payer’s network are doing. The key is to build innovative algorithms and data models around known issues, using as many data sources as possible, and train them with known patterns and issues. HUMAN EXPERTISE STILL REQUIRED No matter how advanced analytics are, however, FWA detection also requires experts who understand how to work through the analytic output. Vital to this process are nurses who can see what others miss in medical records, former law enforcement officers who understand criminal behavior, claims adjustors who can see how bills twist CPT and HCPCs codes to their advantage, and actuaries and others who can look at mountains of statistics and see things that don’t look right. The insight offered by these individuals must then be fed back into the analytics to reduce false positives further. One other factor is worth mentioning. Given the sums and effort involved
A NA L Y T I C S
for post-payment audits and reviews, it is critical to detect and address as many improper claims as possible before payment. Going after an erroneous payment due to a coding error months after reimbursement is expensive for payers, and creates a great deal of abrasion with providers. The more payers can avoid “pay and chase” scenarios, the better it is for all involved. REDUCING THE COST OF FWA As the OIG testimony points out, healthcare FWA is a large problem in the United States. Ensuring that the trillions of dollars being invested in healthcare are spent properly is a critical step toward making healthcare truly affordable for all. To keep up with the challenges of detecting (and stopping) FWA, payers need to use all the data and analytics tools at their disposal to meet the threat. By thinking creatively, mining all available data sources with advanced analytic tools, and involving experts with specialized knowledge who can recognize even the subtlest anomalies, they can significantly reduce the impact of FWA on the healthcare industry. And keep us all safe from the lifethreatening effect of radiation exposure. ❙ Rodger Smith is the senior vice president for payment integrity at SCIO Health Analytics, an organization dedicated to using healthcare analytics to improve clinical outcomes, operational performance and business results. He can be reached at rsmith@sciohealthanalytics.com.
J U LY / A U G U S T 2 016
|
57
SO FTWARE S U R VE Y
Forecasting 2016 New tools, new capabilities, new trends: Survey of 26 software packages from 19 vendors.
BY CHRIS FRY (LEFT) AND VIJAY MEHROTRA elcome to the biennial forecasting software survey, where we take stock of the latest technologies and trends in forecasting affecting both vendors and users. The data used to develop this survey includes responses from 19 vendors for 26 software packages that span a range of capabilities and price points, with the results from this survey summarized below. As part of the rapid proliferation of business analytics, forecasting models are an increasingly important part of the management landscape. More and
W
58
|
A N A LY T I C S - M A G A Z I N E . O R G
more managers and executives rely on sophisticated forecasting methods, not only for planning purposes but also as a foundation for performance analysis, process improvement and optimization. Open source software continues to be a major factor driving the growth of analytics, offering a unique combination of flexibility, power and low cost. In particular, many leading firms are using tools like R and Python to develop solutions for forecasting and predictive analytics that are customized for their business problems, tightly coupled with their data architectures, and integrated directly into other existing systems.
W W W. I N F O R M S . O R G
Photo Courtesy of 123rf.com | seewhatmitchsee
Corporations are seeking not only to generate forecasts, but also to integrate those forecasts into planning, optimization and reporting systems. Given all of this, the market for forecasting software vendors is dynamic, featuring many different types of innovations. Here are some of the major trends that we are seeing in the marketplace. TOP FORECASTING TRENDS FOR 2016 No. 1: integration. Corporations are seeking not only to generate forecasts, but also to integrate those forecasts into planning, optimization and reporting systems. As one survey respondent, ForecastPro, stated, “More and more business users are calling for comprehensive forecasting solutions that, in
A NA L Y T I C S
addition to simply generating statistical forecasts, can be used as the backbone of ongoing corporate processes such as sales and operations planning (S&OP), demand planning and supply chain optimization.� Many vendors appear to be addressing this demand. No. 2: automation. Forecasters are seeking features such as automated model selection, automated alerts, and automated graphics and reporting. As data velocity and complexity are growing, there appears to be an increasing willingness to entrust model design to the software, especially when many
J U LY / A U G U S T 2 016
|
59
FO RE CAST ING SO F T WA R E S U R VE Y
forecasts are being generated simultaneously. In response, nearly all of the vendors surveyed report having some type of automatic or semi-automatic forecasting capabilities. No. 3: visualization. High-quality visualizations are fast becoming part of the “table stakes” for forecasting software packages. In addition to standard statistical output, many of today’s tools offer a range of visuals, including such features as box plots, normal probability plots, histograms, ANOVA Pareto charts, decomposition charts and automated statistically generated range forecast plots. No. 4: virtualization. Several vendors have begun to move their offerings to the cloud, offering virtual hosted forecasting tools. Statgraphics, XLMiner.com, Roadmap GPS and PEERForecaster/ PEERPlanner all mentioned their cloud offerings specifically in their survey responses. As cloud computing continues to grow, we foresee this trend continuing. No. 5: forecast quality measurement. Forecasters are continually pressured to improve the quality of their forecasts, or defend why their forecasts are as good as they can get. Vendors are addressing this need with such solutions as automated ANOVA analysis, automated naïve forecast generation and automated forecast value added analysis. 60
|
A N A LY T I C S - M A G A Z I N E . O R G
No. 6: capabilities enhancement. Forecasting software vendors are giving increased attention to “hard” forecasting problems such as new product forecasting and forecasting of intermittent demand, while also providing ways to integrate additional machine-learning techniques into their forecasting suites. Some vendors are offering the ability to create automated ensemble forecast models, built by combining multiple forecasts generated using different techniques in order to improve overall forecast accuracy. CASE STUDY: FORECASTING RAINFALL IN CALIFORNIA To illustrate some of these trends, we attempted to forecast monthly rainfall levels in our home state of California, which has been suffering from a drought for the past few years. We utilized publicly available data from the National Oceanic and Atmospheric Administration (NOAA) [1], collecting monthly precipitation data (in inches) from January 2006 through March 2016 for 194 observation stations in California. From this, we used the first 10 years of data as a training set and then attempted to forecast the monthly rainfall at each station for the final three months (i.e., January-March 2016). We also included one exogenous variable, NOAA’s El Niño Index [2] (the El Niño climate W W W. I N F O R M S . O R G
oscillation is a well-known factor in predicting California rainfall, as changing temperatures in the Pacific Ocean affect rain patterns on the West Coast). Figure 1 shows the rainfall history for a sample collection station, along with the El Niño Index history. The rainfall follows a somewhat seasonal pattern at this station, and appears to show some (slightly lagged) relationship with the El Niño index. Note the high El Niño activity toward the end of 2015 and Figure 1: Monthly rainfall for a sample collection station in corresponding heavy rainfall California, along with the corresponding El Niño indices. level in January 2016. We selected IBM’s SPSS Statistics which optimizes model and parameter forecasting package to conduct our selection across a suite of exponential analysis. We had not used SPSS prevismoothing and ARIMA models. ously, so we felt that this choice allowed After collecting and prepping the us to work without any preconceived data, we utilized the Expert Modeler to expectations on how the software generate an initial set of forecasts. SPSS would perform, and would also enable created a separate independent model us to convey a realistic experience in for each weather station, with total run terms of the learning curve required to time less than one minute. SPSS moduse the product. IBM offers a generous, eled the majority of stations (177 out of fully functional two-week free trial for 194) using simple seasonal exponential the SPSS product, which made it easy smoothing. For the remainder of the stafor us to download and start working tions, it used a combination of ARIMA, with the product right away. The packseasonal ARIMA and ARIMA with transage includes a fully automatic forecastfer function models (to incorporate the El ing module, called the Expert Modeler, Niño index). The model structures for the A NA L Y T I C S
J U LY / A U G U S T 2 016
|
61
FO RE CAST ING SO F T WA R E S U R VE Y
remaining models varied widely, from simple “flat line” models (when the software could not find a suitable pattern) to complex multi-term ARIMA models such as ARIMA (0,0,3) (1,1,0) with 13-month delayed seasonally differenced external regressor effect or ARIMA (0,0,11) (0,0,0). In addition to the forecast values, the software also produced goodness Figure 2: Plot showing fitted time series and forecasted rainfall of fit statistics for individual compared to actual rainfall for a sample collection station. and aggregate models, along with parameter estimates and plots. and non-seasonal transfer effects for Visually, the fitted time series looked the external regressor. The new models reasonable when compared to the actuseemed to fit the seasonal peaks better als. Figure 2 shows an example fit, along than the smoothing models did, and also with a three-month projection, for a samprojected higher 2016 rainfall in response ple collection station. to the high El Niño Index. Figure 3 shows We then spent additional time to reexample model fit and forecast plots for view and refine the models. We noted two stations, comparing the SPSS Exthat the software’s initial fitted model outpert Forecast output with that of alternate put included negative rainfall estimates models that we selected instead. in some months, so we adjusted these In all, we felt that despite our limited to zero in our analysis. We also rejected data set, the automated procedure gave some of the forecasts in favor of models us a very quick first-pass at the analysis that we felt were either more intuitive or that seemed quite reasonable, and may would better capture the seasonal belikely be satisfactory in many contexts. havior and El Niño effect. Shown in FigThe experience reconfirmed to us the ure 3 are two “before and after” examples power of modern automatic forecasting in which we replaced an automatically tools, and also reminded us of the valgenerated model with a simple seaue that comes from coupling that power sonal ARIMA model including seasonal with our time and attention as analysts to 62
|
A N A LY T I C S - M A G A Z I N E . O R G
W W W. I N F O R M S . O R G
OH THE THINGS THEY CAN DO! Left to Right: Yash Sharma, Geoffrey Burns, and Anuja Sharma. Project: General Motors in Russelsheim, Germany.
Isn’t it time for your Tauber Team? Learn how our graduate-level engineering and business team projects can benefit your organization with high-impact, high ROI projects. • Lean process design and implementation
• Supply chain implementation plan
• Manufacturing rationalization plan
• New product/process development strategy
• Strategic site assessment
• Product complexity analysis
• Strategic capacity analysis
• Plant floor layout
Submission Deadline Project proposals are due December 1, 2016 for projects starting in summer 2017. Contact Jon Grice at gricej@umich.edu or (734) 647-2220.
Learn more at
tauber.umich.edu
FO RE CAST ING SO F T WA R E S U R VE Y
continually seek out improvements in our models. Happy forecasting to you in 2016! ABOUT THE SURVEY For this year’s forecasting software survey, as in the past, we attempted to include as many forecasting products as possible. We contacted all prior survey participants, as well as any new vendors
that we were able to identify. We asked each respondent to complete an online questionnaire covering a comprehensive list of questions spanning features and capabilities, recent enhancements, licensing and fees, technical support and other areas. We followed up with each vendor to help ensure that we obtained as many responses as possible. Vendors not included in this issue are invited to
Figure 3: Sample model fit and forecast plots comparing SPSS’ automatic forecast output with potential alternate model structures. For the first example (San Jose), the auto-generated model predicted 6.97 inches of rainfall for Q1 2016, while the alternate model predicted 7.86 inches. Actual rainfall was 8.36 inches. For the second example, projected Q1 2016 rainfall was 5.61 inches for the automatic model and 7.39 inches for the alternate model. Actual rainfall was 7.22 inches.
64
|
A N A LY T I C S - M A G A Z I N E . O R G
W W W. I N F O R M S . O R G
submit a completed online questionnaire (h t t p : / / w w w. l i o n h r t p u b . c o m / a n c i l l / fssurvey.shtml), and their product will be added to the online version of the forecasting survey. The purpose of the survey is simply to inform the reader of what is available. The information in the survey comes directly from the vendors, and no attempt was made to verify or validate the information they gave us. Automation levels: Forecasting software varies when it comes to the degree to which the software can find
A NA L Y T I C S
the appropriate model and the optimal parameters of that model. For example, Winters’ method requires values for three smoothing constants, and BoxJenkins models have to be specified with various parameters, such as ARIMA (1,0,1) (0,1,2). For the purposes of this and previous surveys, the ability of the software to find the optimal model and parameters for the data is characterized as follows: • Automatic forecasting software: Software is labeled as automatic if it recommends both the appropriate
J U LY / A U G U S T 2 016
|
65
FO RE CAST ING SO F T WA R E S U R VE Y
model to use on a particular data set and finds the optimal parameters for that model. Automatic software typically searches through multiple potential models to minimize a specific fit metric, such as Akaike Information Criterion (AIC), Normailzed Bayesian Information Criterion (BIC) or RMSE; it then recommends a forecast model for the data, gives the model’s optimal parameters, calculates forecasts for a user-specified number of future periods, and gives various summary statistics and graphs. • Semi-automatic forecasting software: The second automation level is called semi-automatic. Such software asks the user to pick a forecasting model from a menu and some statistic to minimize, and the program then finds the optimal parameters for that model, the forecasts, and various graphs and statistics. • Manual forecasting software: We refer to the third level of automation as manual. Here the user must specify both the model that should be used and the corresponding parameters. The software then finds the forecasts, summary statistics and charts. Note that some products fall into 66
|
A N A LY T I C S - M A G A Z I N E . O R G
more than one category. For example, if you choose a Box-Jenkins model, the software may find the optimal parameters for that model, but if you specify that Winters’ method be used, the product may require that you manually enter the three smoothing constants. Of the software tools included in the survey, 23 (88 percent) offer semi-automatic forecasting, and 18 (69 percent) offer automatic forecasting capabilities. ❙ Chris Fry (chris@strategicgmgmtsolutions.com) is the founder and managing director of Strategic Management Solutions, an analytics consulting and services firm. Vijay Mehrotra (vmehrotra@ usfca.edu) is a professor of business analytics and information systems at the University of San Francisco. Both authors are members of INFORMS. The authors thank Gavin Leeper and Craig Volonoski for their contributions to the case study research. Editor’s note: A version of this article appeared in the June 2016 issue of OR/MS Today.
REFERENCES 1. This data was compiled from requests from NOAA’s database at https://www.ncdc.noaa. gov/cdo-web/search. 2. The El Niño Index tracks temperature changes in the Pacific Ocean. The data we used can be downloaded at https://catalog.data.gov/ dataset/climate-prediction-center-cpcoceanicnino-index.
SURVEY DIRECTORY & DATA To view the forecasting software survey products and responses, along with a directory of forecasting software vendors, click here.
W W W. I N F O R M S . O R G
CO N FERE N C E P R E V I E W
Courtesy of Nashville Convention & Visitors Corporation
INFORMS Annual Meeting set for Nashville
Downtown Nashville, site of the 2016 INFORMS Annual Meeting, serves as a dramatic backdrop for the General Jackson showboat.
BY CHANAKA EDIRISINGHE
68
|
Nashville, Tenn., the Music City, will host the 2016 INFORMS Annual Meeting on Nov. 13-16. While the conference will feature the latest advances in operations research, management sciences and analytics, the host city will offer a sizzling combination of American music, southern hospitality, unbelievable cuisine and a boundless spectrum of enjoyment. The technical program includes an exciting array of academic and practitioner invited presentations
A N A LY T I C S - M A G A Z I N E . O R G
W W W. I N F O R M S . O R G
A NA L Y T I C S
Courtesy of Nashville Convention & Visitors Corporation
highlighting several grand challenges facing the world: • Advanced computing as the driver of technological transformation of our society through human-machine interface design and in dealing with climate change, fusion energy, nanotechnology and biotechnology. Topic-related presentations include an opening plenary lecture on cognitive computing by Guruduth Banavar, VP and chief science officer at IBM; a keynote lecture by Jeff Nichols, director (NCCS) at Oak Ridge National Laboratory; and an invited cluster on High Performance Computing organized by Deepak Rajan of the Lawrence Livermore National Laboratory, • Precision agriculture that aims to leverage predictive analytics using realtime data on weather, soil and air quality, crop maturity, etc., to meet the challenges in increasing the global food production in the face of rising population, expected to grow up to 9.2 billion by 2050. Topicrelated presentations include an invited cluster organized by Robin Lougee of IBM and Joseph Byrum of Syngenta to bring OR/MS to the forefront of this challenge. • Healthcare issues that have captivated attention, especially during this U.S. election year, with discussions from both policy-making and informatics perspectives. Topic-related presentations
Solid gold records: The Country Music Hall of Fame and Museum is one of many attractions to be found in Nashville. include a keynote lecture by Edmund Jackson, chief data scientist and VP of Healthcare Corporation of America; and an invited cluster on O.R.-informed Healthcare Policies, organized by Diwakar Gupta, University of Minnesota. • Exploration of big data and big decisions facing OR/MS researchers and practitioners. Topic-related presentations include a joint plenary lecture by Suvrajeet Sen and Gareth James, University of Southern California; and an invited J U LY / A U G U S T 2 016
|
69
CO N FERE N C E P R E V I E W
cluster on Modeling and Methodologies in Big Data, organized by Jiming Peng, University of Houston. • Challenges in identifying and mitigating risk in the financial industry under changing regulations, emerging technologies and heightened corporate responsibility. Topic-related presentations include an invited cluster on Risk and Compliance organized by Akhtarur Siddique, deputy director (Enterprise Risk and Analysis), Office of the Comptroller of the Currency, U.S. Dept. of the Treasury. • The future of global supply chains and the issue of optimizing their efficiencies. Topic-related presentations include a keynote lecture by Jason Murray, VP of World Wide Retail Systems, Amazon; and the IFORS distinguished keynote lecture on hard practical optimization problems that deal with routing bidirectional traffic by Ralph Möhring, Berlin University of Technology. BUT WAIT, THERE’S MORE … Other invited clusters will focus on such themes as additive/advanced manufacturing, physical Internet, energy systems management and entertainment analytics. In addition, the conference will offer a unique opportunity to celebrate the Omega-Rho 40th Year Anniversary through a plenary lecture given by an expert panel consisting of Alfred Blumstein 70
|
A N A LY T I C S - M A G A Z I N E . O R G
(CMU), John Birge (Chicago), Ralph Keeney (Duke) and John Little (MIT). Throughout the conference, a group of eight academic and six practitioner speakers will give a series of 90-minute tutorial lectures that will be collected as a written volume. The tutorials, a mustmake for beginning and advanced researchers alike, will cover the following high-impact research themes: • Optimization frontiers: reviews of stochastic optimization as the science of sequential decisionmaking under uncertainty, with applications to asset-liability management and Markov decision processes; optimal learning when information is expensive; robust multi-objective optimization theory and applications in engineering, business and management. • Network modeling: systemic risk due to complex dependence structure of interactions among individual components, including banks, financial services providers and regulators – with tutorials in network sampling, resilience under contagion and analysis under behavior of multiple autonomous agents for online social networks and economic and financial markets. • Risk modeling and decisions: review of mathematical finance with W W W. I N F O R M S . O R G
emphasis on the need to avoid and rescind destructive deployment of financial risk models; dealing with unstructured data from corporate filings, expert reports and news headlines for financial text mining for risk factors; valuation and hedging of risk in energy portfolios. • Big data: reviews of dimension reduction techniques, as well as data access methods for efficient analytics. • Healthcare & big projects: review of clinical and health sciences
research, rooted in empirical evidence, and the role of analytics in addressing health risks of populations; and an expanded review of research and teaching opportunities in project management, a global economic activity valued at $12 trillion annually. Last but not least, the conference will include many clusters and tracks organized by the sponsored societies within INFORMS, as well as many contributed presentations. The INFORMS Roundtable, ®
CAREER CENTER
Job Seekers: Find Your Next Career Move INFORMS Career Center contains the largest source of O.R. and analytics jobs in the world. It’s where
careercenter.informs.org JOB SEEKER BENEFITS • POST employers to you. • SEARCH • PERSONALIZED job alerts notify you of relevant job opportunities right to your in-box. • ASK the experts advice, resume critique and writing, career assessment test services and more!
www.informs.org | 800.446.3676
A NA L Y T I C S
powered by
J U LY / A U G U S T 2 016
|
71
Analytics Society, CPMS, Railway Applications Section and other practice-related INFORMS societies will collaborate on a special practice track, while presentations of the finalists for the Daniel H. Wagner Prize for Excellence in Operations Research Practice will serve as a Tuesday afternoon keynote. Nashville, Tenn., home to more If you have missed the than a hundred music venues abstract submission dead- such as the legendary Ryman line for oral presentations, Auditorium (inset), will host the 2016 INFORMS Annual the window is still open to Meeting on Nov. 13-16. submit poster presentations – either to participate in the poster competition or just to distance to the convention center, the atpresent in one of the poster sessions. tendees will find suitable accommodation These run on Monday and Tuesday to fit all budgets. from 12:30-2:30 p.m. New this year is Nashville offers everything from an the “E-Poster Walk,” an electronic postelectrifying multi-genre music scene, er displayed digitally. A panel of judges award-winning cuisine, historic homes, will review the posters entered in the world-class art, a myriad of attractions, competition and prizes will be awarded unique shopping, college and profesthe winners. sional sports and more. There are more than 120 live music venues across the SOMETHING FOR EVERYONE city; you’ll catch pickers and songwriters The annual meeting will take place in all over town, in places such as the bluethe Music City Center, the recently built grass venue Station Inn, the rock venue state-of-the-art convention center, and Exit/In, the honky-tonks on Broadway, the adjacent Nashville Omni Hotel. With the song-centered Bluebird Cafe or the many downtown hotels within walking legendary Ryman Auditorium. 72
|
A N A LY T I C S - M A G A Z I N E . O R G
W W W. I N F O R M S . O R G
Courtesy of Nashville Convention & Visitors Corporation
CO N FERE N C E P R E V I E W
Nashville is also home to many attractions, from the Grand Ole Opry to the worldrenowned Country Music Hall of Fame and Museum. Opened in May 2013, the Johnny Cash Museum and the Musicians Hall of Fame and Museum add to the lineup. Nashville is also rich in visual and fine arts. The Frist Center for the Visual Arts is housed in an exquisitely converted Art Deco post office, the Cheekwood Botanical Garden and Museum of Art houses an extensive art gallery, and art crawls, studios and art galleries are all around town. Dubbed “The Coolest, Tastiest City in the South” by Bon Appétit magazine. Nashville’s creative spirit has infiltrated into its kitchens, turning them into chef’s studios. From southern fare to haute cuisine to quite literally everything in between, Nashville’s palate offers it all. For more information, click here. ❙ Chanaka Edirisinghe is a professor at the Lally School of Management, Rensselaer Polytechnic Institute and the general chair of the 2016 INFORMS Annual Meeting in Nashville, Tenn.
A NA L Y T I C S
2016 Winter Simulation Conference The 2016 Winter Simulation Conference (WSC 2016) will be held Dec. 11-14 in Arlington, Va., at the Crystal Gateway Marriott. The hotel is located minutes from downtown Washington, D.C., and less than a mile from Reagan National Airport. The METRO is connected to the hotel, which makes exploring the area’s monuments, museums, shopping and restaurants very convenient. The WSC is the premier international forum for disseminating recent advances in the field of dynamic systems modeling and simulation. In addition to a technical program of unsurpassed scope and quality, WSC provides the central meeting place for simulation practitioners, researchers and vendors. Research in modeling and simulation is propelled by fostering crossfertilization between various disciplines. The theme for WSC 2016 is “Simulating Complex Service Systems.” The theme emphasizes the increasingly complex engineered and human systems in highly connected environments, the availability of data to help model such systems, technological advances which continue to push the limits of computation, and conceptual and mathematical advances that help make sense of complex systems. These forces enable more informed decisions. WSC 2016 is sponsored by ACM/SIGSIM, IIE, INFORMS-SIM and SCS, along with technical cosponsors ASA, ASIM, EEE/SMC and NIST. For more information or to register for the conference, visit http://www.wintersim.org. J U LY / A U G U S T 2 016
|
73
FIVE- M IN U T E A N A LYST
Photo Courtesy of 123rf.com | Sergey Kohl
Dark side envelopment analysis
BY HARRISON SCHRAMM
74
|
A repeated theme in “The Force Awakens,” the latest blockbuster movie in the “Star Wars” saga, is Kylo Ren’s secret fear of not being as “evil” as his hero, Darth Vader. We, as analysts, have the mythical ability to see if this is actually true. We’re going to use the Force – and the standard Simplex LP solver in MS Excel – to see if this is true. I’ve collected some data on the “achievements” and “failures” that each Force practitioner has seen in his career. (This list was chosen by me and is not by any means allinclusive.) Next, we’re going to apply DEA, which in the literature means “data envelopment analysis,” but we understand to be “dark side envelopment analysis.” Conceptually, we’re going to pretend that each candidate is coming in for a performance review, and we are going to put his accomplishments – and failings – in the best light for him (Table 1). Everything is considered from the dark side’s point of view. Just for fun, we’re including Luke Skywalker too, because I had friends on that Death Star. Now pretend that we are actually filling out their annual evaluations. If we knew how to weight each person’s achievements it would be easy, but we don’t know how to weight them, and there’s no good way
A N A LY T I C S - M A G A Z I N E . O R G
W W W. I N F O R M S . O R G
Achievements
Vader
Ren
Luke
Palpatine
Planet-sized objects destroyed
1
4
1
0
Force Lightning Choking Lifting
4
2
1
2
Aerial Victories
3
0
4
0
Planets Conquered
2 Hoth, Cloud City
0
1
10 (Chancellor)
Failures
Vader
Ren
Luke
Palpatine
Major Stations Lost
2
1
1
1
Temper-tantrums
1
2
1
0
Computer Drives Unrecovered
2
1
0
0
Table 1: Comparison of the careers of Darth Vader and Kylo Ren, Luke Skywalker and Emperor Palpatine. We must remember that while it is tempting to declare Ren the winner because he destroyed four planets at the cost of one base, “The power to destroy planets is insignificant compared to the power of the Force.” to find out. A priori, we might presume (positive score), while normalizing his that destroying a planet (if you’re a Sith) “failures” to be equal to 1. (See technical or Death Star (for Jedi) is really good note later in this article.) and that losing a Death Star is really We apply this scheme to each of our bad regardless of what side you are on. candidates, producing (see Table 2): However, we don’t know that. So what To answer our question: Based on I propose to do is to let each character this example, Kylo Ren assesses himself pick his own weights for Score Pick -> Vader Ren Luke Palpatine both achievements and Vader 100% 50% 100% 100% failures and compute the Ren ratio (achievement/failure), 100% 40% 100% 29% with the constraint that their Luke 38% 0% 100% 0% weights cannot make anyone have a score greater than Palpatine 10% 0% 10% 100% 100 percent efficiency. Each character chooses Table 2: Force user’s efficiency. Each Force user is able to achieve 100 percent efficiency by their own scheme, but they don’t his weights by maximizing their necessarily feel the same way about their peers. own perceived “achievements” A NA L Y T I C S
J U LY / A U G U S T 2 016
|
75
FIVE- M IN U T E A N A LYST
as being “as good (bad) as Darth Vader”; by his weighting scheme (column 2), he and Vader are perfect, and considers Luke and the Emperor to be no-talent Padawans. Conversely, if you asked Vader about Kylo Ren, he would say that Kylo’s journey to the dark side was not yet complete, as he is only about 30 percent of a dark lord. The efficiency metric is not symmetric. Interestingly, Emperor Palpatine thinks highly of everyone except Luke, but nobody thinks very highly of the Emperor. The weighting vectors are what you might expect. For Vader it’s all about
force choking; for Kylo it’s all about blowing up planets; Luke focuses on being an ace star pilot; and Palpatine recalls his nearly bloodless rise to chancellor. When grading “failures,” we all have to talk about the uncomfortable memory of the Death Star, Starkiller Base or Echo Base on Hoth. And Kylo really needs to work on his temper tantrums. Finally, just like our world, nobody wants to talk about the computer drives. ❙ Harrison Schramm (Harrison.schramm@gmail.com), CAP, PStat, is a principal operations research analyst at CANA Advisors, LLC.
TECHNICAL NOTES
Technical note No. 1: DEA can be formulated as an LP with the form:
Where V is the Weight Vector of “achievements,” U is the weight vector of “failures” and j is the index of the particular member under test. The “efficiency” is given by The second constraint is isomorphic to requiring that the efficiency is less than 100 percent, but expressed as a linear relationship. Technical note No. 2: These results are from using the standard Simplex LP solver in Excel on a run-of-themill desktop. It is interesting that for this type of problem, GRG Nonlinear – which my machine defaults to for some unknown reason – performed very badly on this particular problem.
76
|
A N A LY T I C S - M A G A Z I N E . O R G
W W W. I N F O R M S . O R G
DATA ANALYTICA CEREBRUM understanding the underlying methodology and mindset of how to approach and handle data is one of the most important things analytics professionals need to know. informS intensive classroom courses will help enhance the skills, tools and methods you need to make your projects a success.
SAMPLING BIAS PRESCRIPTIVE PREDICTIVE STOCHASTIC MODELS NON-TECHNICAL DECISION MAKERS
UNSTRUCTURED PROBLEMS
REGRESSION OPTIMIZATION vs. SIMULATION DISPARATE INFORMATION
UPCOMING CLASS:
essential practice skills for high-impact analytics projects september 28–29, 2016 | 8:30am – 4:30pm executive conference center 2345 crystal drive arlington, va 22202
limited seating available. Register at www.informs.org/continuinged
CHART1
THIN K IN G A N A LY T I CA LLY
Elevators Figure 1: Race to the bottom: Which elevator will get there first?
A. DISPLAY
B. DISPLAY
1
8–
56
2 3 4 5 6 7 8 9 10 11
66 95 8– 11– 76 75 56 25 75 25
9– 66 116 55 45 55 85 12– 55 4–
Table 1.
The building I work in has three elevators and 12 floors. Like some elevators, these have displays above them that indicate the current floor that the elevator is on and also the travel direction of the elevator. The travel direction display can show Up, Down or Stopped indicated by an up arrow, a down arrow or a dash character, respectively. As an example, Elevator A in the image is on the 2nd floor and is traveling in the up direction. Table 1 displays 10 historic data points C. FIRST TO DISPLAY GROUND FLOOR showing the floor and direction of the B 35 elevators from times that I have used the A 35 B 25 elevator in the past. The table also shows B 85 A 85 which elevator showed up first on the 5– A 12– C ground floor. A 126 B 55 Question: For scenario No. 11, which 12– C ?? 66 elevator is most likely to be first to the ground floor? Send your answer to puzzlor@gmail.com by Sept. 15. The winner, chosen randomly from correct answers, will receive a $25 Amazon Gift Card. Past questions and answers can be found at puzzlor.com. ❙
BY JOHN TOCZEK
78
|
John Toczek is the assistant vice president of Predictive Modeling at Chubb in the Decision Analytics and Predictive Modeling department. He earned his BSc. in chemical engineering at Drexel University (1996) and his MSc. in operations research from Virginia Commonwealth University (2005).
A N A LY T I C S - M A G A Z I N E . O R G
W W W. I N F O R M S . O R G
OPTIMIZATION GENERAL ALGEBRAIC MODELING SYSTEM High-Level Modeling The General Algebraic Modeling System (GAMS) is a high-level modeling system for mathematical programming problems. GAMS is tailored for complex, large-scale modeling applications, and allows you to build large maintainable models that can be adapted quickly to new situations. Models are fully portable from one computer platform to another.
State-of-the-Art Solvers GAMS incorporates all major commercial and academic state-of-the-art solution technologies for a broad range of problem types.
Optimizing Carbon Capture Technologies: The CCSI Optimization Toolset Carbon Capture technologies can significantly reduce atmospheric emissions of CO2 from fossil fuel power plants. A widespread deployment of these technologies is necessary to significantly reduce greenhouse gas emissions and contribute to a clean energy portfolio. But the deployment is both expensive and time-consuming: bringing such technologies online can take industries between 20 and 30 years. Speeding up this process is the express goal of the Carbon Capture Simulation Initiative (CCSI). Founded by the U.S. Department of Energy in 2011, CCSI is a partnership among national laboratories, industry and academic institutions. High-Level Modeling for the Success of Future Technologies CCSI provides an optimization toolset that helps industry to rapidly assess and utilize these new technologies. The optimization tools identify optimal equipment configurations and operating conditions for potential CO2 capture processes, thereby significantly reducing cost, time and risk involved in the implementation. The CCSI research group has developed two advanced optimization capabilities as part of its Framework for Optimization and Quantification of Uncertainty and Surrogates (FOQUS) tool. Both utilize GAMS as an essential element. The first tool performs simultaneous process optimization and heat integration based on rigorous models. The heat integration subproblem is modeled in GAMS as LPs and MIPs and solved by CPLEX. The other tool optimizes the design and operation of a CO2 capture system. The carbon capture system is represented as a MINLP model, which is implemented in GAMS and solved by DICOPT or BARON. GAMS is proud to be a part of this optimization toolset designed to make carbon capture a success.
www.gams.com
GAMS Integrated Developer Environment for editing, debugging, solving models, and viewing data.