Big_Data_Now_2012_Edition.pdf

Sep 25 – 27, 2013Boston, MA

Oct 28 – 30, 2013New York, NY

Nov 11 –13, 2013London, England

Change the world with data. We’ll show you how.strataconf.com

O’Reilly Media, Inc.

Big Data Now: 2012 Edition

ISBN: 978-1-449-35671-2

Big Data Now: 2012 Editionby O’Reilly Media, Inc.

Published by O’Reilly Media, Inc., 1005 Gravenstein Highway North, Sebastopol, CA95472.

O’Reilly books may be purchased for educational, business, or sales promotional use.Online editions are also available for most titles (http://my.safaribooksonline.com). Formore information, contact our corporate/institutional sales department: (800)998-9938 or [email protected].

Cover Designer: Karen Montgomery Interior Designer: David Futato

October 2012: First Edition

Revision History for the First Edition:

2012-10-24 First release

See http://oreilly.com/catalog/errata.csp?isbn=9781449356712 for release details.

Nutshell Handbook, the Nutshell Handbook logo, and the O’Reilly logo are registeredtrademarks of O’Reilly Media, Inc.

Many of the designations used by manufacturers and sellers to distinguish their prod‐ucts are claimed as trademarks. Where those designations appear in this book, andO’Reilly Media, Inc. was aware of a trademark claim, the designations have been printedin caps or initial caps.

While every precaution has been taken in the preparation of this book, the publisherand authors assume no responsibility for errors or omissions, or for damages resultingfrom the use of the information contained herein.

Table of Contents

1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2. Getting Up to Speed with Big Data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3What Is Big Data? 3

What Does Big Data Look Like? 4In Practice 8

What Is Apache Hadoop? 10The Core of Hadoop: MapReduce 11Hadoop’s Lower Levels: HDFS and MapReduce 11Improving Programmability: Pig and Hive 12Improving Data Access: HBase, Sqoop, and Flume 12Coordination and Workflow: Zookeeper and Oozie 14Management and Deployment: Ambari and Whirr 14Machine Learning: Mahout 14Using Hadoop 15

Why Big Data Is Big: The Digital Nervous System 15From Exoskeleton to Nervous System 15Charting the Transition 16Coming, Ready or Not 17

3. Big Data Tools, Techniques, and Strategies. . . . . . . . . . . . . . . . . . . . . 19Designing Great Data Products 19

Objective-based Data Products 20The Model Assembly Line: A Case Study of Optimal

Decisions Group 21Drivetrain Approach to Recommender Systems 25Optimizing Lifetime Customer Value 28Best Practices from Physical Data Products 31The Future for Data Products 35

iii

What It Takes to Build Great Machine Learning Products 35Progress in Machine Learning 36Interesting Problems Are Never Off the Shelf 37Defining the Problem 39

4. The Application of Big Data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41Stories over Spreadsheets 41

A Thought on Dashboards 43Full Interview 43

Mining the Astronomical Literature 43Interview with Robert Simpson: Behind the Project and

What Lies Ahead 48Science between the Cracks 51

The Dark Side of Data 51The Digital Publishing Landscape 52Privacy by Design 53

5. What to Watch for in Big Data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55Big Data Is Our Generation’s Civil Rights Issue, and We

Don’t Know It 55Three Kinds of Big Data 60

Enterprise BI 2.0 60Civil Engineering 62Customer Relationship Optimization 63Headlong into the Trough 64

Automated Science, Deep Data, and the Paradox ofInformation 64(Semi)Automated Science 65Deep Data 67The Paradox of Information 69

The Chicken and Egg of Big Data Solutions 71Walking the Tightrope of Visualization Criticism 73

The Visualization Ecosystem 74The Irrationality of Needs: Fast Food to Fine Dining 76Grown-up Criticism 78Final Thoughts 80

6. Big Data and Health Care. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83Solving the Wanamaker Problem for Health Care 83

Making Health Care More Effective 85More Data, More Sources 89

iv | Table of Contents

Paying for Results 90Enabling Data 91Building the Health Care System We Want 94Recommended Reading 95

Dr. Farzad Mostashari on Building the Health InformationInfrastructure for the Modern ePatient 96

John Wilbanks Discusses the Risks and Rewards of a HealthData Commons 100

Esther Dyson on Health Data, “Preemptive Healthcare,” andthe Next Big Thing 106

A Marriage of Data and Caregivers Gives Dr. Atul GawandeHope for Health Care 112

Five Elements of Reform that Health Providers WouldRather Not Hear About 119

Table of Contents | v

CHAPTER 1

Introduction

In the first edition of Big Data Now, the O’Reilly team tracked the birthand early development of data tools and data science. Now, with thissecond edition, we’re seeing what happens when big data grows up:how it’s being applied, where it’s playing a role, and the conse‐quences — good and bad alike — of data’s ascendance.

We’ve organized the 2012 edition of Big Data Now into five areas:

Getting Up to Speed With Big Data — Essential information on thestructures and definitions of big data.

Big Data Tools, Techniques, and Strategies — Expert guidance forturning big data theories into big data products.

The Application of Big Data — Examples of big data in action, in‐cluding a look at the downside of data.

What to Watch for in Big Data — Thoughts on how big data willevolve and the role it will play across industries and domains.

Big Data and Health Care — A special section exploring the possi‐bilities that arise when data and health care come together.

In addition to Big Data Now, you can stay on top of the latest datadevelopments with our ongoing analysis on O’Reilly Radar andthrough our Strata coverage and events series.

CHAPTER 2

Getting Up to Speed with Big Data

What Is Big Data?By Edd Dumbill

Big data is data that exceeds the processing capacity of conventionaldatabase systems. The data is too big, moves too fast, or doesn’t fit thestrictures of your database architectures. To gain value from this data,you must choose an alternative way to process it.

The hot IT buzzword of 2012, big data has become viable as cost-effective approaches have emerged to tame the volume, velocity, andvariability of massive data. Within this data lie valuable patterns andinformation, previously hidden because of the amount of work re‐quired to extract them. To leading corporations, such as Walmart orGoogle, this power has been in reach for some time, but at fantasticcost. Today’s commodity hardware, cloud architectures and opensource software bring big data processing into the reach of the lesswell-resourced. Big data processing is eminently feasible for even thesmall garage startups, who can cheaply rent server time in the cloud.

The value of big data to an organization falls into two categories: an‐alytical use and enabling new products. Big data analytics can revealinsights hidden previously by data too costly to process, such as peerinfluence among customers, revealed by analyzing shoppers’ transac‐tions and social and geographical data. Being able to process everyitem of data in reasonable time removes the troublesome need forsampling and promotes an investigative approach to data, in contrastto the somewhat static nature of running predetermined reports.

The past decade’s successful web startups are prime examples of bigdata used as an enabler of new products and services. For example, bycombining a large number of signals from a user’s actions and thoseof their friends, Facebook has been able to craft a highly personalizeduser experience and create a new kind of advertising business. It’s nocoincidence that the lion’s share of ideas and tools underpinning bigdata have emerged from Google, Yahoo, Amazon, and Facebook.

The emergence of big data into the enterprise brings with it a necessarycounterpart: agility. Successfully exploiting the value in big data re‐quires experimentation and exploration. Whether creating new prod‐ucts or looking for ways to gain competitive advantage, the job callsfor curiosity and an entrepreneurial outlook.

What Does Big Data Look Like?As a catch-all term, “big data” can be pretty nebulous, in the same waythat the term “cloud” covers diverse technologies. Input data to bigdata systems could be chatter from social networks, web server logs,traffic flow sensors, satellite imagery, broadcast audio streams, bank‐ing transactions, MP3s of rock music, the content of web pages, scansof government documents, GPS trails, telemetry from automobiles,financial market data, the list goes on. Are these all really the samething?

To clarify matters, the three Vs of volume, velocity, and variety arecommonly used to characterize different aspects of big data. They’rea helpful lens through which to view and understand the nature of thedata and the software platforms available to exploit them. Most prob‐ably you will contend with each of the Vs to one degree or another.

Volume

The benefit gained from the ability to process large amounts of infor‐mation is the main attraction of big data analytics. Having more databeats out having better models: simple bits of math can be unreason‐ably effective given large amounts of data. If you could run that forecasttaking into account 300 factors rather than 6, could you predict de‐mand better? This volume presents the most immediate challenge toconventional IT structures. It calls for scalable storage, and a distribut‐ed approach to querying. Many companies already have large amountsof archived data, perhaps in the form of logs, but not the capacity toprocess it.

4 | Chapter 2: Getting Up to Speed with Big Data

Assuming that the volumes of data are larger than those conventionalrelational database infrastructures can cope with, processing optionsbreak down broadly into a choice between massively parallel process‐ing architectures — data warehouses or databases such as Green‐plum — and Apache Hadoop-based solutions. This choice is often in‐formed by the degree to which one of the other “Vs” — variety —comes into play. Typically, data warehousing approaches involve pre‐determined schemas, suiting a regular and slowly evolving dataset.Apache Hadoop, on the other hand, places no conditions on the struc‐ture of the data it can process.

At its core, Hadoop is a platform for distributing computing problemsacross a number of servers. First developed and released as open sourceby Yahoo, it implements the MapReduce approach pioneered by Goo‐gle in compiling its search indexes. Hadoop’s MapReduce involvesdistributing a dataset among multiple servers and operating on thedata: the “map” stage. The partial results are then recombined: the“reduce” stage.

To store data, Hadoop utilizes its own distributed filesystem, HDFS,which makes data available to multiple computing nodes. A typicalHadoop usage pattern involves three stages:

• loading data into HDFS,• MapReduce operations, and• retrieving results from HDFS.

This process is by nature a batch operation, suited for analytical ornon-interactive computing tasks. Because of this, Hadoop is not itselfa database or data warehouse solution, but can act as an analyticaladjunct to one.

One of the most well-known Hadoop users is Facebook, whose modelfollows this pattern. A MySQL database stores the core data. This isthen reflected into Hadoop, where computations occur, such as cre‐ating recommendations for you based on your friends’ interests. Face‐book then transfers the results back into MySQL, for use in pagesserved to users.

Velocity

The importance of data’s velocity — the increasing rate at which dataflows into an organization — has followed a similar pattern to that of

What Is Big Data? | 5

volume. Problems previously restricted to segments of industry arenow presenting themselves in a much broader setting. Specializedcompanies such as financial traders have long turned systems that copewith fast moving data to their advantage. Now it’s our turn.

Why is that so? The Internet and mobile era means that the way wedeliver and consume products and services is increasingly instrumen‐ted, generating a data flow back to the provider. Online retailers areable to compile large histories of customers’ every click and interaction:not just the final sales. Those who are able to quickly utilize that in‐formation, by recommending additional purchases, for instance, gaincompetitive advantage. The smartphone era increases again the rateof data inflow, as consumers carry with them a streaming source ofgeolocated imagery and audio data.

It’s not just the velocity of the incoming data that’s the issue: it’s possibleto stream fast-moving data into bulk storage for later batch processing,for example. The importance lies in the speed of the feedback loop,taking data from input through to decision. A commercial fromIBM makes the point that you wouldn’t cross the road if all you hadwas a five-minute old snapshot of traffic location. There are timeswhen you simply won’t be able to wait for a report to run or a Hadoopjob to complete.

Industry terminology for such fast-moving data tends to be either“streaming data” or “complex event processing.” This latter term wasmore established in product categories before streaming processingdata gained more widespread relevance, and seems likely to diminishin favor of streaming.

There are two main reasons to consider streaming processing. The firstis when the input data are too fast to store in their entirety: in order tokeep storage requirements practical, some level of analysis must occuras the data streams in. At the extreme end of the scale, the Large Ha‐dron Collider at CERN generates so much data that scientists mustdiscard the overwhelming majority of it — hoping hard they’ve notthrown away anything useful. The second reason to consider stream‐ing is where the application mandates immediate response to the data.Thanks to the rise of mobile applications and online gaming this is anincreasingly common situation.

6 | Chapter 2: Getting Up to Speed with Big Data

Product categories for handling streaming data divide into establishedproprietary products such as IBM’s InfoSphere Streams and the less-polished and still emergent open source frameworks originating in theweb industry: Twitter’s Storm and Yahoo S4.

As mentioned above, it’s not just about input data. The velocity of asystem’s outputs can matter too. The tighter the feedback loop, thegreater the competitive advantage. The results might go directly intoa product, such as Facebook’s recommendations, or into dashboardsused to drive decision-making. It’s this need for speed, particularly onthe Web, that has driven the development of key-value stores and col‐umnar databases, optimized for the fast retrieval of precomputed in‐formation. These databases form part of an umbrella category knownas NoSQL, used when relational models aren’t the right fit.

Variety

Rarely does data present itself in a form perfectly ordered and readyfor processing. A common theme in big data systems is that the sourcedata is diverse, and doesn’t fall into neat relational structures. It couldbe text from social networks, image data, a raw feed directly from asensor source. None of these things come ready for integration into anapplication.

Even on the Web, where computer-to-computer communicationought to bring some guarantees, the reality of data is messy. Differentbrowsers send different data, users withhold information, they may beusing differing software versions or vendors to communicate with you.And you can bet that if part of the process involves a human, there willbe error and inconsistency.

A common use of big data processing is to take unstructured data andextract ordered meaning, for consumption either by humans or as astructured input to an application. One such example is entity reso‐lution, the process of determining exactly what a name refers to. Is thiscity London, England, or London, Texas? By the time your businesslogic gets to it, you don’t want to be guessing.

The process of moving from source data to processed application datainvolves the loss of information. When you tidy up, you end up throw‐ing stuff away. This underlines a principle of big data: when you can,keep everything. There may well be useful signals in the bits you throwaway. If you lose the source data, there’s no going back.

What Is Big Data? | 7

Despite the popularity and well understood nature of relational data‐bases, it is not the case that they should always be the destination fordata, even when tidied up. Certain data types suit certain classes ofdatabase better. For instance, documents encoded as XML are mostversatile when stored in a dedicated XML store such as MarkLogic.Social network relations are graphs by nature, and graph databasessuch as Neo4J make operations on them simpler and more efficient.

Even where there’s not a radical data type mismatch, a disadvantageof the relational database is the static nature of its schemas. In an agile,exploratory environment, the results of computations will evolve withthe detection and extraction of more signals. Semi-structured NoSQLdatabases meet this need for flexibility: they provide enough structureto organize data, but do not require the exact schema of the data beforestoring it.

In PracticeWe have explored the nature of big data and surveyed the landscapeof big data from a high level. As usual, when it comes to deploymentthere are dimensions to consider over and above tool selection.

Cloud or in-house?

The majority of big data solutions are now provided in three forms:software-only, as an appliance or cloud-based. Decisions betweenwhich route to take will depend, among other things, on issues of datalocality, privacy and regulation, human resources and project require‐ments. Many organizations opt for a hybrid solution: using on-demand cloud resources to supplement in-house deployments.

Big data is big

It is a fundamental fact that data that is too big to process conven‐tionally is also too big to transport anywhere. IT is undergoing aninversion of priorities: it’s the program that needs to move, not thedata. If you want to analyze data from the U.S. Census, it’s a lot easierto run your code on Amazon’s web services platform, which hosts suchdata locally, and won’t cost you time or money to transfer it.

Even if the data isn’t too big to move, locality can still be an issue,especially with rapidly updating data. Financial trading systems crowdinto data centers to get the fastest connection to source data, becausethat millisecond difference in processing time equates to competitiveadvantage.

8 | Chapter 2: Getting Up to Speed with Big Data

Big data is messy

It’s not all about infrastructure. Big data practitioners consistently re‐port that 80% of the effort involved in dealing with data is cleaning itup in the first place, as Pete Warden observes in his Big Data Glossa‐ry: “I probably spend more time turning messy source data into some‐thing usable than I do on the rest of the data analysis process com‐bined.”

Because of the high cost of data acquisition and cleaning, it’s worthconsidering what you actually need to source yourself. Data market‐places are a means of obtaining common data, and you are often ableto contribute improvements back. Quality can of course be variable,but will increasingly be a benchmark on which data marketplacescompete.

Culture

The phenomenon of big data is closely tied to the emergence of datascience, a discipline that combines math, programming, and scientificinstinct. Benefiting from big data means investing in teams with thisskillset, and surrounding them with an organizational willingness tounderstand and use data for advantage.

In his report, “Building Data Science Teams,” D.J. Patil characterizesdata scientists as having the following qualities:

• Technical expertise: the best data scientists typically have deepexpertise in some scientific discipline.

• Curiosity: a desire to go beneath the surface and discover anddistill a problem down into a very clear set of hypotheses that canbe tested.

• Storytelling: the ability to use data to tell a story and to be able tocommunicate it effectively.

• Cleverness: the ability to look at a problem in different, creativeways.

The far-reaching nature of big data analytics projects can have un‐comfortable aspects: data must be broken out of silos in order to bemined, and the organization must learn how to communicate and in‐terpet the results of analysis.

What Is Big Data? | 9

Those skills of storytelling and cleverness are the gateway factors thatultimately dictate whether the benefits of analytical labors are absor‐bed by an organization. The art and practice of visualizing data is be‐coming ever more important in bridging the human-computer gap tomediate analytical insight in a meaningful way.

Know where you want to go

Finally, remember that big data is no panacea. You can find patternsand clues in your data, but then what? Christer Johnson, IBM’s leaderfor advanced analytics in North America, gives this advice to busi‐nesses starting out with big data: first, decide what problem you wantto solve.

If you pick a real business problem, such as how you can change youradvertising strategy to increase spend per customer, it will guide yourimplementation. While big data work benefits from an enterprisingspirit, it also benefits strongly from a concrete goal.

What Is Apache Hadoop?By Edd Dumbill

Apache Hadoop has been the driving force behind the growth of thebig data industry. You’ll hear it mentioned often, along with associatedtechnologies such as Hive and Pig. But what does it do, and why doyou need all its strangely named friends, such as Oozie, Zookeeper,and Flume?

Hadoop brings the ability to cheaply process large amounts of data,regardless of its structure. By large, we mean from 10-100 gigabytesand above. How is this different from what went before?

Existing enterprise data warehouses and relational databases excel atprocessing structured data and can store massive amounts of data,though at a cost: This requirement for structure restricts the kinds ofdata that can be processed, and it imposes an inertia that makes datawarehouses unsuited for agile exploration of massive heterogenousdata. The amount of effort required to warehouse data often meansthat valuable data sources in organizations are never mined. This iswhere Hadoop can make a big difference.

This article examines the components of the Hadoop ecosystem andexplains the functions of each.

10 | Chapter 2: Getting Up to Speed with Big Data

The Core of Hadoop: MapReduceCreated at Google in response to the problem of creating web searchindexes, the MapReduce framework is the powerhouse behind mostof today’s big data processing. In addition to Hadoop, you’ll find Map‐Reduce inside MPP and NoSQL databases, such as Vertica or Mon‐goDB.

The important innovation of MapReduce is the ability to take a queryover a dataset, divide it, and run it in parallel over multiple nodes.Distributing the computation solves the issue of data too large to fitonto a single machine. Combine this technique with commodity Linuxservers and you have a cost-effective alternative to massive computingarrays.

At its core, Hadoop is an open source MapReduce implementation.Funded by Yahoo, it emerged in 2006 and, according to its creatorDoug Cutting, reached “web scale” capability in early 2008.

As the Hadoop project matured, it acquired further components toenhance its usability and functionality. The name “Hadoop” has cometo represent this entire ecosystem. There are parallels with the emer‐gence of Linux: The name refers strictly to the Linux kernel, but it hasgained acceptance as referring to a complete operating system.

Hadoop’s Lower Levels: HDFS and MapReduceAbove, we discussed the ability of MapReduce to distribute computa‐tion over multiple servers. For that computation to take place, eachserver must have access to the data. This is the role of HDFS, the Ha‐doop Distributed File System.

HDFS and MapReduce are robust. Servers in a Hadoop cluster can failand not abort the computation process. HDFS ensures data is repli‐cated with redundancy across the cluster. On completion of a calcu‐lation, a node will write its results back into HDFS.

There are no restrictions on the data that HDFS stores. Data may beunstructured and schemaless. By contrast, relational databases requirethat data be structured and schemas be defined before storing the data.With HDFS, making sense of the data is the responsibility of the de‐veloper’s code.

Programming Hadoop at the MapReduce level is a case of workingwith the Java APIs, and manually loading data files into HDFS.

What Is Apache Hadoop? | 11

Improving Programmability: Pig and HiveWorking directly with Java APIs can be tedious and error prone. It alsorestricts usage of Hadoop to Java programmers. Hadoop offers twosolutions for making Hadoop programming easier.

• Pig is a programming language that simplifies the common tasksof working with Hadoop: loading data, expressing transforma‐tions on the data, and storing the final results. Pig’s built-in oper‐ations can make sense of semi-structured data, such as log files,and the language is extensible using Java to add support for customdata types and transformations.

• Hive enables Hadoop to operate as a data warehouse. It superim‐poses structure on data in HDFS and then permits queries overthe data using a familiar SQL-like syntax. As with Pig, Hive’s corecapabilities are extensible.

Choosing between Hive and Pig can be confusing. Hive is more suit‐able for data warehousing tasks, with predominantly static structureand the need for frequent analysis. Hive’s closeness to SQL makes it anideal point of integration between Hadoop and other business intelli‐gence tools.

Pig gives the developer more agility for the exploration of large data‐sets, allowing the development of succinct scripts for transformingdata flows for incorporation into larger applications. Pig is a thinnerlayer over Hadoop than Hive, and its main advantage is to drasticallycut the amount of code needed compared to direct use of Hadoop’sJava APIs. As such, Pig’s intended audience remains primarily thesoftware developer.

Improving Data Access: HBase, Sqoop, and FlumeAt its heart, Hadoop is a batch-oriented system. Data are loaded intoHDFS, processed, and then retrieved. This is somewhat of a computingthrowback, and often, interactive and random access to data is re‐quired.

Enter HBase, a column-oriented database that runs on top of HDFS.Modeled after Google’s BigTable, the project’s goal is to host billionsof rows of data for rapid access. MapReduce can use HBase as both asource and a destination for its computations, and Hive and Pig canbe used in combination with HBase.

12 | Chapter 2: Getting Up to Speed with Big Data

In order to grant random access to the data, HBase does impose a fewrestrictions: Hive performance with HBase is 4-5 times slower thanwith plain HDFS, and the maximum amount of data you can store inHBase is approximately a petabyte, versus HDFS’ limit of over 30PB.

HBase is ill-suited to ad-hoc analytics and more appropriate for inte‐grating big data as part of a larger application. Use cases include log‐ging, counting, and storing time-series data.

The Hadoop Bestiary

Ambari Deployment, configuration and monitoring

Flume Collection and import of log and event data

HBase Column-oriented database scaling to billions of rows

HCatalog Schema and data type sharing over Pig, Hive and MapReduce

HDFS Distributed redundant file system for Hadoop

Hive Data warehouse with SQL-like access

Mahout Library of machine learning and data mining algorithms

MapReduce Parallel computation on server clusters

Pig High-level programming language for Hadoop computations

Oozie Orchestration and workflow management

Sqoop Imports data from relational databases

Whirr Cloud-agnostic deployment of clusters

Zookeeper Configuration management and coordination

Getting data in and out

Improved interoperability with the rest of the data world is providedby Sqoop and Flume. Sqoop is a tool designed to import data fromrelational databases into Hadoop, either directly into HDFS or intoHive. Flume is designed to import streaming flows of log data directlyinto HDFS.

Hive’s SQL friendliness means that it can be used as a point of inte‐gration with the vast universe of database tools capable of makingconnections via JBDC or ODBC database drivers.

What Is Apache Hadoop? | 13

Coordination and Workflow: Zookeeper and OozieWith a growing family of services running as part of a Hadoop cluster,there’s a need for coordination and naming services. As computingnodes can come and go, members of the cluster need to synchronizewith each other, know where to access services, and know how theyshould be configured. This is the purpose of Zookeeper.

Production systems utilizing Hadoop can often contain complex pipe‐lines of transformations, each with dependencies on each other. Forexample, the arrival of a new batch of data will trigger an import, whichmust then trigger recalculations in dependent datasets. The Ooziecomponent provides features to manage the workflow and dependen‐cies, removing the need for developers to code custom solutions.

Management and Deployment: Ambari and WhirrOne of the commonly added features incorporated into Hadoop bydistributors such as IBM and Microsoft is monitoring and adminis‐tration. Though in an early stage, Ambari aims to add these featuresto the core Hadoop project. Ambari is intended to help system ad‐ministrators deploy and configure Hadoop, upgrade clusters, andmonitor services. Through an API, it may be integrated with othersystem management tools.

Though not strictly part of Hadoop, Whirr is a highly complementarycomponent. It offers a way of running services, including Hadoop, oncloud platforms. Whirr is cloud neutral and currently supports theAmazon EC2 and Rackspace services.

Machine Learning: MahoutEvery organization’s data are diverse and particular to their needs.However, there is much less diversity in the kinds of analyses per‐formed on that data. The Mahout project is a library of Hadoop im‐plementations of common analytical computations. Use cases includeuser collaborative filtering, user recommendations, clustering, andclassification.

14 | Chapter 2: Getting Up to Speed with Big Data

Using HadoopNormally, you will use Hadoop in the form of a distribution. Much aswith Linux before it, vendors integrate and test the components of theApache Hadoop ecosystem and add in tools and administrative fea‐tures of their own.

Though not per se a distribution, a managed cloud installation of Ha‐doop’s MapReduce is also available through Amazon’s Elastic MapRe‐duce service.

Why Big Data Is Big: The Digital NervousSystemBy Edd Dumbill

Where does all the data in “big data” come from? And why isn’t bigdata just a concern for companies such as Facebook and Google? Theanswer is that the web companies are the forerunners. Driven by social,mobile, and cloud technology, there is an important transition takingplace, leading us all to the data-enabled world that those companiesinhabit today.

From Exoskeleton to Nervous SystemUntil a few years ago, the main function of computer systems in society,and business in particular, was as a digital support system. Applica‐tions digitized existing real-world processes, such as word-processing,payroll, and inventory. These systems had interfaces back out to thereal world through stores, people, telephone, shipping, and so on. Thenow-quaint phrase “paperless office” alludes to this transfer of pre-existing paper processes into the computer. These computer systemsformed a digital exoskeleton, supporting a business in the real world.

The arrival of the Internet and the Web has added a new dimension,bringing in an era of entirely digital business. Customer interaction,payments, and often product delivery can exist entirely within com‐puter systems. Data doesn’t just stay inside the exoskeleton any more,but is a key element in the operation. We’re in an era where businessand society are acquiring a digital nervous system.

Why Big Data Is Big: The Digital Nervous System | 15

As my sketch below shows, an organization with a digital nervous sys‐tem is characterized by a large number of inflows and outflows of data,a high level of networking, both internally and externally, increaseddata flow, and consequent complexity.

This transition is why big data is important. Techniques developed todeal with interlinked, heterogenous data acquired by massive webcompanies will be our main tools as the rest of us transition to digital-native operation. We see early examples of this, from catching fraudin financial transactions to debugging and improving the hiring pro‐cess in HR: and almost everybody already pays attention to the massiveflow of social network information concerning them.

Charting the TransitionAs technology has progressed within business, each step taken hasresulted in a leap in data volume. To people looking at big data now, areasonable question is to ask why, when their business isn’t Google orFacebook, does big data apply to them?

The answer lies in the ability of web businesses to conduct 100% oftheir activities online. Their digital nervous system easily stretchesfrom the beginning to the end of their operations. If you have factories,shops, and other parts of the real world within your business, you’vefurther to go in incorporating them into the digital nervous system.

But “further to go” doesn’t mean it won’t happen. The drive of the Web,social media, mobile, and the cloud is bringing more of each business

16 | Chapter 2: Getting Up to Speed with Big Data

into a data-driven world. In the UK, the Government Digital Serviceis unifying the delivery of services to citizens. The results are a radicalimprovement of citizen experience, and for the first time many de‐partments are able to get a real picture of how they’re doing. For anyretailer, companies such as Square, American Express, and Four‐square are bringing payments into a social, responsive data ecosystem,liberating that information from the silos of corporate accounting.

What does it mean to have a digital nervous system? The key trait isto make an organization’s feedback loop entirely digital. That is, a di‐rect connection from sensing and monitoring inputs through to prod‐uct outputs. That’s straightforward on the Web. It’s getting increasinglyeasier in retail. Perhaps the biggest shifts in our world will come assensors and robotics bring the advantages web companies have nowto domains such as industry, transport, and the military.

The reach of the digital nervous system has grown steadily over thepast 30 years, and each step brings gains in agility and flexibility, alongwith an order of magnitude more data. First, from specific applicationprograms to general business use with the PC. Then, direct interactionover the Web. Mobile adds awareness of time and place, along withinstant notification. The next step, to cloud, breaks down data silosand adds storage and compute elasticity through cloud computing.Now, we’re integrating smart agents, able to act on our behalf, andconnections to the real world through sensors and automation.

Coming, Ready or NotIf you’re not contemplating the advantages of taking more of your op‐eration digital, you can bet your competitors are. As Marc Andreessenwrote last year, “software is eating the world.” Everything is becomingprogrammable.

It’s this growth of the digital nervous system that makes the techniquesand tools of big data relevant to us today. The challenges of massivedata flows, and the erosion of hierarchy and boundaries, will lead usto the statistical approaches, systems thinking, and machine learningwe need to cope with the future we’re inventing.

Why Big Data Is Big: The Digital Nervous System | 17

CHAPTER 3

Big Data Tools, Techniques,and Strategies

Designing Great Data ProductsBy Jeremy Howard, Margit Zwemer, and Mike Loukides

In the past few years, we’ve seen many data products based on predic‐tive modeling. These products range from weather forecasting to rec‐ommendation engines to services that predict airline flight times moreaccurately than the airlines themselves. But these products are still justmaking predictions, rather than asking what action they want some‐one to take as a result of a prediction. Prediction technology can beinteresting and mathematically elegant, but we need to take the nextstep. The technology exists to build data products that can revolu‐tionize entire industries. So, why aren’t we building them?

To jump-start this process, we suggest a four-step approach that hasalready transformed the insurance industry. We call it the DrivetrainApproach, inspired by the emerging field of self-driving vehicles. En‐gineers start by defining a clear objective: They want a car to drive safelyfrom point A to point B without human intervention. Great predictivemodeling is an important part of the solution, but it no longer standson its own; as products become more sophisticated, it disappears intothe plumbing. Someone using Google’s self-driving car is completelyunaware of the hundreds (if not thousands) of models and the peta‐bytes of data that make it work. But as data scientists build increasingly

sophisticated products, they need a systematic design approach. Wedon’t claim that the Drivetrain Approach is the best or only method;our goal is to start a dialog within the data science and business com‐munities to advance our collective vision.

Objective-based Data ProductsWe are entering the era of data as drivetrain, where we use data notjust to generate more data (in the form of predictions), but use data toproduce actionable outcomes. That is the goal of the Drivetrain Ap‐proach. The best way to illustrate this process is with a familiar dataproduct: search engines. Back in 1997, AltaVista was king of the algo‐rithmic search world. While their models were good at finding relevantwebsites, the answer the user was most interested in was often buriedon page 100 of the search results. Then, Google came along and trans‐formed online search by beginning with a simple question: What isthe user’s main objective in typing in a search query?

The four steps in the Drivetrain Approach.

Google realized that the objective was to show the most relevant searchresult; for other companies, it might be increasing profit, improvingthe customer experience, finding the best path for a robot, or balancingthe load in a data center. Once we have specified the goal, the secondstep is to specify what inputs of the system we can control, the leverswe can pull to influence the final outcome. In Google’s case, they couldcontrol the ranking of the search results. The third step was to considerwhat new data they would need to produce such a ranking; they real‐ized that the implicit information regarding which pages linked towhich other pages could be used for this purpose. Only after these firstthree steps do we begin thinking about building the predictive mod‐els. Our objective and available levers, what data we already have andwhat additional data we will need to collect, determine the models wecan build. The models will take both the levers and any uncontrollablevariables as their inputs; the outputs from the models can be combinedto predict the final state for our objective.

20 | Chapter 3: Big Data Tools, Techniques, and Strategies

Step 4 of the Drivetrain Approach for Google is now part of tech his‐tory: Larry Page and Sergey Brin invented the graph traversal algo‐rithm PageRank and built an engine on top of it that revolutionizedsearch. But you don’t have to invent the next PageRank to build a greatdata product. We will show a systematic approach to step 4 that doesn’trequire a PhD in computer science.

The Model Assembly Line: A Case Study of OptimalDecisions GroupOptimizing for an actionable outcome over the right predictive modelscan be a company’s most important strategic decision. For an insur‐ance company, policy price is the product, so an optimal pricing modelis to them what the assembly line is to automobile manufacturing.Insurers have centuries of experience in prediction, but as recently as10 years ago, the insurance companies often failed to make optimalbusiness decisions about what price to charge each new customer.Their actuaries could build models to predict a customer’s likelihoodof being in an accident and the expected value of claims. But thosemodels did not solve the pricing problem, so the insurance companieswould set a price based on a combination of guesswork and marketstudies.

This situation changed in 1999 with a company called Optimal Deci‐sions Group (ODG). ODG approached this problem with an early useof the Drivetrain Approach and a practical take on step 4 that can beapplied to a wide range of problems. They began by defining the ob‐jective that the insurance company was trying to achieve: setting a pricethat maximizes the net-present value of the profit from a new customerover a multi-year time horizon, subject to certain constraints such asmaintaining market share. From there, they developed an optimizedpricing process that added hundreds of millions of dollars to the in‐surers’ bottom lines. [Note: Co-author Jeremy Howard founded ODG.]

ODG identified which levers the insurance company could control:what price to charge each customer, what types of accidents to cover,how much to spend on marketing and customer service, and how toreact to their competitors’ pricing decisions. They also considered in‐puts outside of their control, like competitors’ strategies, macroeco‐nomic conditions, natural disasters, and customer “stickiness.” Theyconsidered what additional data they would need to predict a cus‐tomer’s reaction to changes in price. It was necessary to build this da‐

Designing Great Data Products | 21

taset by randomly changing the prices of hundreds of thousands ofpolicies over many months. While the insurers were reluctant to con‐duct these experiments on real customers, as they’d certainly lose somecustomers as a result, they were swayed by the huge gains that opti‐mized policy pricing might deliver. Finally, ODG started to design themodels that could be used to optimize the insurer’s profit.

Drivetrain Step 4: The Model Assembly Line. Picture a Model AssemblyLine for data products that transforms the raw data into an actionableoutcome. The Modeler takes the raw data and converts it into slightlymore refined predicted data.

The first component of ODG’s Modeler was a model of price elasticity(the probability that a customer will accept a given price) for new pol‐icies and for renewals. The price elasticity model is a curve of priceversus the probability of the customer accepting the policy conditionalon that price. This curve moves from almost certain acceptance at verylow prices to almost never at high prices.

The second component of ODG’s Modeler related price to the insur‐ance company’s profit, conditional on the customer accepting thisprice. The profit for a very low price will be in the red by the value ofexpected claims in the first year, plus any overhead for acquiring andservicing the new customer. Multiplying these two curves creates afinal curve that shows price versus expected profit (see Expected Profitfigure, below). The final curve has a clearly identifiable local maximumthat represents the best price to charge a customer for the first year.

22 | Chapter 3: Big Data Tools, Techniques, and Strategies

Expected profit.

ODG also built models for customer retention. These models predic‐ted whether customers would renew their policies in one year, allowingfor changes in price and willingness to jump to a competitor. Theseadditional models allow the annual models to be combined to predictprofit from a new customer over the next five years.

This new suite of models is not a final answer because it only identifiesthe outcome for a given set of inputs. The next machine on the as‐sembly line is a Simulator, which lets ODG ask the “what if ” questionsto see how the levers affect the distribution of the final outcome. Theexpected profit curve is just a slice of the surface of possible outcomes.To build that entire surface, the Simulator runs the models over a widerange of inputs. The operator can adjust the input levers to answerspecific questions like, “What will happen if our company offers thecustomer a low teaser price in year one but then raises the premiumsin year two?” They can also explore how the distribution of profit isshaped by the inputs outside of the insurer’s control: “What if theeconomy crashes and the customer loses his job? What if a 100-yearflood hits his home? If a new competitor enters the market and our

Designing Great Data Products | 23

company does not react, what will be the impact on our bottom line?”Because the simulation is at a per-policy level, the insurer can view theimpact of a given set of price changes on revenue, market share, andother metrics over time.

The Simulator’s result is fed to an Optimizer, which takes the surfaceof possible outcomes and identifies the highest point. The Optimizernot only finds the best outcomes, it can also identify catastrophic out‐comes and show how to avoid them. There are many different opti‐mization techniques to choose from (see “Optimization in the RealWorld” (page 24)), but it is a well-understood field with robust andaccessible solutions. ODG’s competitors use different techniques tofind an optimal price, but they are shipping the same over-all dataproduct. What matters is that using a Drivetrain Approach combinedwith a Model Assembly Line bridges the gap between predictive mod‐els and actionable outcomes. Irfan Ahmed of CloudPhysics providesa good taxonomy of predictive modeling that describes this entire as‐sembly line process:

When dealing with hundreds or thousands of individual componentsmodels to understand the behavior of the full-system, a search has tobe done. I think of this as a complicated machine (full-system) wherethe curtain is withdrawn and you get to model each significant partof the machine under controlled experiments and then simulate theinteractions. Note here the different levels: models of individual com‐ponents, tied together in a simulation given a set of inputs, iteratedthrough over different input sets in a search optimizer.

Optimization in the Real WorldOptimization is a classic problem that has been studied by Newton andGauss all the way up to mathematicians and engineers in the presentday. Many optimization procedures are iterative; they can be thoughtof as taking a small step, checking our elevation and then taking anothersmall uphill step until we reach a point from which there is no directionin which we can climb any higher. The danger in this hill-climbingapproach is that if the steps are too small, we may get stuck at one ofthe many local maxima in the foothills, which will not tell us the bestset of controllable inputs. There are many techniques to avoid thisproblem, some based on statistics and spreading our bets widely, andothers based on systems seen in nature, like biological evolution or thecooling of atoms in glass.

24 | Chapter 3: Big Data Tools, Techniques, and Strategies

Optimization is a process we are all familiar with in our daily lives, evenif we have never used algorithms like gradient descent or simulatedannealing. A great image for optimization in the real world comes upin a recent TechZing podcast with the co-founders of data-miningcompetition platform Kaggle. One of the authors of this paper wasexplaining an iterative optimization technique, and the host says, “So,in a sense Jeremy, your approach was like that of doing a startup, whichis just get something out there and iterate and iterate and iterate.” Thetakeaway, whether you are a tiny startup or a giant insurance company,is that we unconsciously use optimization whenever we decide how toget to where we want to go.

Drivetrain Approach to Recommender SystemsLet’s look at how we could apply this process to another industry:marketing. We begin by applying the Drivetrain Approach to a familiarexample, recommendation engines, and then building this up into anentire optimized marketing strategy.

Recommendation engines are a familiar example of a data productbased on well-built predictive models that do not achieve an optimalobjective. The current algorithms predict what products a customerwill like, based on purchase history and the histories of similar cus‐tomers. A company like Amazon represents every purchase that hasever been made as a giant sparse matrix, with customers as the rowsand products as the columns. Once they have the data in this format,data scientists apply some form of collaborative filtering to “fill in thematrix.” For example, if customer A buys products 1 and 10, and cus‐tomer B buys products 1, 2, 4, and 10, the engine will recommend thatA buy 2 and 4. These models are good at predicting whether a customerwill like a given product, but they often suggest products that the cus‐

Designing Great Data Products | 25

tomer already knows about or has already decided not to buy. Amazon’srecommendation engine is probably the best one out there, but it’s easyto get it to show its warts. Here is a screenshot of the “Customers WhoBought This Item Also Bought” feed on Amazon from a search for thelatest book in Terry Pratchett’s “Discworld series:”

All of the recommendations are for other books in the same series, butit’s a good assumption that a customer who searched for “Terry Pratch‐ett” is already aware of these books. There may be some unexpectedrecommendations on pages 2 through 14 of the feed, but how manycustomers are going to bother clicking through?

Instead, let’s design an improved recommendation engine using theDrivetrain Approach, starting by reconsidering our objective. The ob‐jective of a recommendation engine is to drive additional sales by sur‐prising and delighting the customer with books he or she would nothave purchased without the recommendation. What we would reallylike to do is emulate the experience of Mark Johnson, CEO of Zite,who gave a perfect example of what a customer’s recommendationexperience should be like in a recent TOC talk. He went into Strandbookstore in New York City and asked for a book similar to Toni Mor‐rison’s Beloved. The girl behind the counter recommended WilliamFaulkner’s Absolom Absolom. On Amazon, the top results for a similarquery leads to another book by Toni Morrison and several books bywell-known female authors of color. The Strand bookseller made abrilliant but far-fetched recommendation probably based more on thecharacter of Morrison’s writing than superficial similarities betweenMorrison and other authors. She cut through the chaff of the obviousto make a recommendation that will send the customer home with anew book, and returning to Strand again and again in the future.

This is not to say that Amazon’s recommendation engine could nothave made the same connection; the problem is that this helpful rec‐ommendation will be buried far down in the recommendation feed,beneath books that have more obvious similarities to Beloved. The

26 | Chapter 3: Big Data Tools, Techniques, and Strategies

objective is to escape a recommendation filter bubble, a term whichwas originally coined by Eli Pariser to describe the tendency of per‐sonalized news feeds to only display articles that are blandly popularor further confirm the readers’ existing biases.

As with the AltaVista-Google example, the lever a bookseller can con‐trol is the ranking of the recommendations. New data must also becollected to generate recommendations that will cause new sales. Thiswill require conducting many randomized experiments in order tocollect data about a wide range of recommendations for a wide rangeof customers.

The final step in the drivetrain process is to build the Model AssemblyLine. One way to escape the recommendation bubble would be to builda Modeler containing two models for purchase probabilities, condi‐tional on seeing or not seeing a recommendation. The difference be‐tween these two probabilities is a utility function for a given recom‐mendation to a customer (see Recommendation Engine figure, be‐low). It will be low in cases where the algorithm recommends a familiarbook that the customer has already rejected (both components aresmall) or a book that he or she would have bought even without therecommendation (both components are large and cancel each otherout). We can build a Simulator to test the utility of each of the manypossible books we have in stock, or perhaps just over all the outputs ofa collaborative filtering model of similar customer purchases, and thenbuild a simple Optimizer that ranks and displays the recommendedbooks based on their simulated utility. In general, when choosing anobjective function to optimize, we need less emphasis on the “function”and more on the “objective.” What is the objective of the person usingour data product? What choice are we actually helping him or hermake?

Designing Great Data Products | 27

Recommendation Engine.

Optimizing Lifetime Customer ValueThis same systematic approach can be used to optimize the entiremarketing strategy. This encompasses all the interactions that a retailerhas with its customers outside of the actual buy-sell transaction,whether making a product recommendation, encouraging the cus‐tomer to check out a new feature of the online store, or sending salespromotions. Making the wrong choices comes at a cost to the retailerin the form of reduced margins (discounts that do not drive extrasales), opportunity costs for the scarce real-estate on their homepage(taking up space in the recommendation feed with products the cus‐tomer doesn’t like or would have bought without a recommendation)or the customer tuning out (sending so many unhelpful email pro‐motions that the customer filters all future communications as spam).We will show how to go about building an optimized marketing strat‐egy that mitigates these effects.

28 | Chapter 3: Big Data Tools, Techniques, and Strategies

As in each of the previous examples, we begin by asking: “What ob‐jective is the marketing strategy trying to achieve?” Simple: we wantto optimize the lifetime value from each customer. Second question:“What levers do we have at our disposal to achieve this objective?”Quite a few. For example:

1. We can make product recommendations that surprise and delight(using the optimized recommendation outlined in the previoussection).

2. We could offer tailored discounts or special offers on products thecustomer was not quite ready to buy or would have bought else‐where.

3. We can even make customer-care calls just to see how the user isenjoying our site and make them feel that their feedback is valued.

What new data do we need to collect? This can vary case by case, buta few online retailers are taking creative approaches to this step. Onlinefashion retailer Zafu shows how to encourage the customer to partic‐ipate in this collection process. Plenty of websites sell designer denim,but for many women, high-end jeans are the one item of clothing theynever buy online because it’s hard to find the right pair without tryingthem on. Zafu’s approach is not to send their customers directly to theclothes, but to begin by asking a series of simple questions about thecustomers’ body type, how well their other jeans fit, and their fashionpreferences. Only then does the customer get to browse a recom‐mended selection of Zafu’s inventory. The data collection and recom‐mendation steps are not an add-on; they are Zafu’s entire businessmodel — women’s jeans are now a data product. Zafu can tailor theirrecommendations to fit as well as their jeans because their system isasking the right questions.

Designing Great Data Products | 29

Starting with the objective forces data scientists to consider what ad‐ditional models they need to build for the Modeler. We can keep the“like” model that we have already built as well as the causality modelfor purchases with and without recommendations, and then take astaged approach to adding additional models that we think will im‐prove the marketing effectiveness. We could add a price elasticitymodel to test how offering a discount might change the probabilitythat the customer will buy the item. We could construct a patiencemodel for the customers’ tolerance for poorly targeted communica‐tions: When do they tune them out and filter our messages straight tospam? (“If Hulu shows me that same dog food ad one more time, I’mgonna stop watching!”) A purchase sequence causality model can beused to identify key “entry products.” For example, a pair of jeans thatis often paired with a particular top, or the first part of a series of novelsthat often leads to a sale of the whole set.

Once we have these models, we construct a Simulator and an Opti‐mizer and run them over the combined models to find out what rec‐ommendations will achieve our objectives: driving sales and improv‐ing the customer experience.

30 | Chapter 3: Big Data Tools, Techniques, and Strategies

A look inside the Modeler.

Best Practices from Physical Data ProductsIt is easy to stumble into the trap of thinking that since data existssomewhere abstract, on a spreadsheet or in the cloud, that data prod‐ucts are just abstract algorithms. So, we would like to conclude byshowing you how objective-based data products are already a part ofthe tangible world. What is most important about these examples isthat the engineers who designed these data products didn’t start bybuilding a neato robot and then looking for something to do with it.They started with an objective like, “I want my car to drive me places,”and then designed a covert data product to accomplish that task. En‐gineers are often quietly on the leading edge of algorithmic applica‐tions because they have long been thinking about their own modelingchallenges in an objective-based way. Industrial engineers were amongthe first to begin using neural networks, applying them to problemslike the optimal design of assembly lines and quality control. BrianRipley’s seminal book on pattern recognition gives credit for manyideas and techniques to largely forgotten engineering papers from the1970s.

When designing a product or manufacturing process, a drivetrain-likeprocess followed by model integration, simulation and optimizationis a familiar part of the toolkit of systems engineers. In engineering, it

Designing Great Data Products | 31

is often necessary to link many component models together so thatthey can be simulated and optimized in tandem. These firms haveplenty of experience building models of each of the components andsystems in their final product, whether they’re building a server farmor a fighter jet. There may be one detailed model for mechanical sys‐tems, a separate model for thermal systems, and yet another for elec‐trical systems, etc. All of these systems have critical interactions. Forexample, resistance in the electrical system produces heat, which needsto be included as an input for the thermal diffusion and cooling model.That excess heat could cause mechanical components to warp, pro‐ducing stresses that should be inputs to the mechanical models.

The screenshot below is taken from a model integration tool designedby Phoenix Integration. Although it’s from a completely different en‐gineering discipline, this diagram is very similar to the Drivetrain Ap‐proach we’ve recommended for data products. The objective is clearlydefined: build an airplane wing. The wing box includes the designlevers like span, taper ratio, and sweep. The data is in the wing mate‐rials’ physical properties; costs are listed in another tab of the appli‐cation. There is a Modeler for aerodynamics and mechanical structurethat can then be fed to a Simulator to produce the Key Wing Outputsof cost, weight, lift coefficient, and induced drag. These outcomes canbe fed to an Optimizer to build a functioning and cost-effective air‐plane wing.

32 | Chapter 3: Big Data Tools, Techniques, and Strategies

Screenshot from a model integration tool designed by Phoenix Integra‐tion.

As predictive modeling and optimization become more vital to a widevariety of activities, look out for the engineers to disrupt industriesthat wouldn’t immediately appear to be in the data business. The in‐spiration for the phrase “Drivetrain Approach,” for example, is alreadyon the streets of Mountain View. Instead of being data driven, we cannow let the data drive us.

Suppose we wanted to get from San Francisco to the Strata 2012 Con‐ference in Santa Clara. We could just build a simple model of distance /speed-limit to predict arrival time with little more than a ruler and aroad map. If we want a more sophisticated system, we can build an‐other model for traffic congestion and yet another model to forecastweather conditions and their effect on the safest maximum speed.There are plenty of cool challenges in building these models, but bythemselves, they do not take us to our destination. These days, it istrivial to use some type of heuristic search algorithm to predict thedrive times along various routes (a Simulator) and then pick the short‐est one (an Optimizer) subject to constraints like avoiding bridge tollsor maximizing gas mileage. But why not think bigger? Instead of thefemme-bot voice of the GPS unit telling us which route to take andwhere to turn, what would it take to build a car that would make thosedecisions by itself? Why not bundle simulation and optimization en‐gines with a physical engine, all inside the black box of a car?

Let’s consider how this is an application of the Drivetrain Approach.We have already defined our objective: building a car that drives itself.The levers are the vehicle controls we are all familiar with: steeringwheel, accelerator, brakes, etc. Next, we consider what data the carneeds to collect; it needs sensors that gather data about the road as wellas cameras that can detect road signs, red or green lights, and unex‐pected obstacles (including pedestrians). We need to define the mod‐els we will need, such as physics models to predict the effects of steer‐ing, braking and acceleration, and pattern recognition algorithms tointerpret data from the road signs.

As one engineer on the Google self-driving car project put it in a recentWired article, “We’re analyzing and predicting the world 20 times asecond.” What gets lost in the quote is what happens as a result of thatprediction. The vehicle needs to use a simulator to examine the resultsof the possible actions it could take. If it turns left now, will it hit that

Designing Great Data Products | 33

pedestrian? If it makes a right turn at 55 mph in these weather condi‐tions, will it skid off the road? Merely predicting what will happen isn’tgood enough. The self-driving car needs to take the next step: aftersimulating all the possibilities, it must optimize the results of the sim‐ulation to pick the best combination of acceleration and braking,steering and signaling, to get us safely to Santa Clara. Prediction onlytells us that there is going to be an accident. An optimizer tells us howto avoid accidents.

Improving the data collection and predictive models is very important,but we want to emphasize the importance of beginning by defining aclear objective with levers that produce actionable outcomes. Datascience is beginning to pervade even the most bricks-and-mortar el‐ements of our lives. As scientists and engineers become more adept atapplying prediction and optimization to everyday problems, they areexpanding the art of the possible, optimizing everything from ourpersonal health to the houses and cities we live in. Models developedto simulate fluid dynamics and turbulence have been applied to im‐proving traffic and pedestrian flows by using the placement of exitsand crowd control barriers as levers. This has improved emergencyevacuation procedures for subway stations and reduced the danger ofcrowd stampedes and trampling during sporting events. Nest is de‐signing smart thermostats that learn the home-owner’s temperaturepreferences and then optimizes their energy consumption. For motorvehicle traffic, IBM performed a project with the city of Stockholm tooptimize traffic flows that reduced congestion by nearly a quarter, andincreased the air quality in the inner city by 25%. What is particularlyinteresting is that there was no need to build an elaborate new datacollection system. Any city with metered stoplights already has all thenecessary information; they just haven’t found a way to suck themeaning out of it.

In another area where objective-based data products have the powerto change lives, the CMU extension in Silicon Valley has an activeproject for building data products to help first responders after naturalor man-made disasters. Jeannie Stamberger of Carnegie Mellon Uni‐versity Silicon Valley explained to us many of the possible applicationsof predictive algorithms to disaster response, from text-mining andsentiment analysis of tweets to determine the extent of the damage, toswarms of autonomous robots for reconnaissance and rescue, to lo‐gistic optimization tools that help multiple jurisdictions coordinatetheir responses. These disaster applications are a particularly good

34 | Chapter 3: Big Data Tools, Techniques, and Strategies

example of why data products need simple, well-designed interfacesthat produce concrete recommendations. In an emergency, a dataproduct that just produces more data is of little use. Data scientistsnow have the predictive tools to build products that increase the com‐mon good, but they need to be aware that building the models is notenough if they do not also produce optimized, implementable out‐comes.

The Future for Data ProductsWe introduced the Drivetrain Approach to provide a framework fordesigning the next generation of great data products and describedhow it relies at its heart on optimization. In the future, we hope to seeoptimization taught in business schools as well as in statistics depart‐ments. We hope to see data scientists ship products that are designedto produce desirable business outcomes. This is still the dawn of datascience. We don’t know what design approaches will be developed inthe future, but right now, there is a need for the data science communityto coalesce around a shared vocabulary and product design processthat can be used to educate others on how to derive value from theirpredictive models. If we do not do this, we will find that our modelsonly use data to create more data, rather than using data to createactions, disrupt industries, and transform lives.

Do we want products that deliver data, or do we want products thatdeliver results based on data? Jeremy Howard examined these questionsin his Strata California 12 session, “From Predictive Modelling to Opti‐mization: The Next Frontier.” Full video from that session is availablehere.

What It Takes to Build Great Machine LearningProductsBy Aria Haghighi

Machine learning (ML) is all the rage, riding tight on the coattails ofthe “big data” wave. Like most technology hype, the enthusiasm farexceeds the realization of actual products. Arguably, not since Google’stremendous innovations in the late ’90s/early 2000s has algorithmictechnology led to a product that has permeated the popular culture.That’s not to say there haven’t been great ML wins since, but none haveas been as impactful or had computational algorithms at their core.

What It Takes to Build Great Machine Learning Products | 35

1. Although MCMC is a much older statistical technique, its broad use in large-scalemachine learning applications is relatively recent.

Netflix may use recommendation technology, but Netflix is still Netflixwithout it. There would be no Google if Page, Brin, et al., hadn’t ex‐ploited the graph structure of the Web and anchor text to improvesearch.

So why is this? It’s not for lack of trying. How many startups have aimedto bring natural language processing (NLP) technology to the masses,only to fade into oblivion after people actually try their products? Thechallenge in building great products with ML lies not in just under‐standing basic ML theory, but in understanding the domain and prob‐lem sufficiently to operationalize intuitions into model design. Inter‐esting problems don’t have simple off-the-shelf ML solutions. Progressin important ML application areas, like NLP, come from insights spe‐cific to these problems, rather than generic ML machinery. Often,specific insights into a problem and careful model design make thedifference between a system that doesn’t work at all and one that peoplewill actually use.

The goal of this essay is not to discourage people from building amaz‐ing products with ML at their cores, but to be clear about where I thinkthe difficulty lies.

Progress in Machine LearningMachine learning has come a long way over the last decade. Before Istarted grad school, training a large-margin classifier (e.g., SVM) wasdone via John Platt’s batch SMO algorithm. In that case, training timescaled poorly with the amount of training data. Writing the algorithmitself required understanding quadratic programming and was riddledwith heuristics for selecting active constraints and black-art parametertuning. Now, we know how to train a nearly performance-equivalentlarge-margin classifier in linear time using a (relatively) simple onlinealgorithm (PDF). Similar strides have been made in (probabilistic)graphical models: Markov-chain Monte Carlo (MCMC) and varia‐tional methods have facilitated inference for arbitrarily complexgraphical models.1 Anecdotally, take at look at papers over the last

36 | Chapter 3: Big Data Tools, Techniques, and Strategies

eight years in the proceedings of the Association for ComputationalLinguistics (ACL), the premiere natural language processing publica‐tion. A top paper from 2011 has orders of magnitude more technicalML sophistication than one from 2003.

On the education front, we’ve come a long way as well. As an undergradat Stanford in the early-to-mid 2000s, I took Andrew Ng’s MLcourse and Daphne Koller’s probabilistic graphical model course. Bothof these classes were among the best I took at Stanford and were onlyavailable to about 100 students a year. Koller’s course in particular wasnot only the best course I took at Stanford, but the one that taught methe most about teaching. Now, anyone can take these courses online.

As an applied ML person — specifically, natural language processing— much of this progress has made aspects of research significantlyeasier. However, the core decisions I make are not which abstract MLalgorithm, loss-function, or objective to use, but what features andstructure are relevant to solving my problem. This skill only comeswith practice. So, while it’s great that a much wider audience will havean understanding of basic ML, it’s not the most difficult part of build‐ing intelligent systems.

Interesting Problems Are Never Off the ShelfThe interesting problems that you’d actually want to solve are farmessier than the abstractions used to describe standard ML problems.Take machine translation (MT), for example. Naively, MT looks like astatistical classification problem: You get an input foreign sentence andhave to predict a target English sentence. Unfortunately, because thespace of possible English is combinatorially large, you can’t treat MTas a black-box classification problem. Instead, like most interestingML applications, MT problems have a lot of structure and part of thejob of a good researcher is decomposing the problem into smallerpieces that can be learned or encoded deterministically. My claim isthat progress in complex problems like MT comes mostly from howwe decompose and structure the solution space, rather than ML tech‐niques used to learn within this space.

Machine translation has improved by leaps and bounds throughoutthe last decade. I think this progress has largely, but not entirely, comefrom keen insights into the specific problem, rather than generic MLimprovements. Modern statistical MT originates from an amazingpaper, “The mathematics of statistical machine translation” (PDF),

What It Takes to Build Great Machine Learning Products | 37

2. The model is generative, so what’s being described here is from the point-of-view ofinference; the model’s generative story works in reverse.

3. IBM model 3 introduced the concept of fertility to allow a given word to generatemultiple independent target translation words. While this could generate the requiredtranslation, the probability of the model doing so is relatively low.

which introduced the noisy-channel architecture on which future MTsystems would be based. At a very simplistic level, this is how the modelworks:2 For each foreign word, there are potential English translations(including the null word for foreign words that have no English equiv‐alent). Think of this as a probabilistic dictionary. These candidatetranslation words are then re-ordered to create a plausible Englishtranslation. There are many intricacies being glossed over: how to ef‐ficiently consider candidate English sentences and their permutations,what model is used to learn the systematic ways in which reorderingoccurs between languages, and the details about how to score theplausibility of the English candidate (the language model).

The core improvement in MT came from changing this model. So,rather than learning translation probabilities of individual words, toinstead learn models of how to translate foreign phrases to Englishphrases. For instance, the German word “abends” translates roughlyto the English prepositional phrase “in the evening.” Before phrase-based translation (PDF), a word-based model would only get to trans‐late to a single English word, making it unlikely to arrive at the correctEnglish translation.3 Phrase-based translation generally results inmore accurate translations with fluid, idiomatic English output. Ofcourse, adding phrased-based emissions introduces several additionalcomplexities, including how to how to estimate phrase-emissions giv‐en that we never observe phrase segmentation; no one tells us that “inthe evening” is a phrase that should match up to some foreign phrase.What’s surprising here is that there aren’t general ML improvementsthat are making this difference, but problem-specific model design.People can and have implemented more sophisticated ML techniquesfor various pieces of an MT system. And these do yield improvements,but typically far smaller than good problem-specific research insights.

Franz Och, one of the authors of the original Phrase-based papers,went on to Google and became the principle person behind the searchcompany’s translation efforts. While the intellectual underpinnings ofGoogle’s system go back to Och’s days as a research scientist at theInformation Sciences Institute (and earlier as a graduate student),

38 | Chapter 3: Big Data Tools, Techniques, and Strategies

much of the gains beyond the insights underlying phrase-based trans‐lation (and minimum-error rate training, another of Och’s innova‐tions) came from a massive software engineering effort to scale theseideas to the Web. That effort itself yielded impressive research intolarge-scale language models and other areas of NLP. It’s important tonote that Och, in addition to being a world-class researcher, is also, byall accounts, an incredibly impressive hacker and builder. It’s this rarecombination of skill that can bring ideas all the way from a researchproject to where Google Translate is today.

Defining the ProblemBut I think there’s an even bigger barrier beyond ingenious modeldesign and engineering skills. In the case of machine translation andspeech recognition, the problem being solved is straightforward tounderstand and well-specified. Many of the NLP technologies that Ithink will revolutionize consumer products over the next decade aremuch more vague. How, exactly, can we take the excellent research instructured topic models, discourse processing, or sentiment analysisand make a mass-appeal consumer product?

Consider summarization. We all know that in some way, we’ll wantproducts that summarize and structure content. However, for com‐putational and research reasons, you need to restrict the scope of thisproblem to something for which you can build a model, an algorithm,and ultimately evaluate. For instance, in the summarization literature,the problem of multi-document summarization is typically formula‐ted as selecting a subset of sentences from the document collectionand ordering them. Is this the right problem to be solving? Is the bestway to summarize a piece of text a handful of full-length sentences?Even if a summarization is accurate, does the Franken-sentence struc‐ture yield summaries that feel inorganic to users?

Or, consider sentiment analysis. Do people really just want a coarse-grained thumbs-up or thumbs-down on a product or event? Or dothey want a richer picture of sentiments toward individual aspects ofan item (e.g., loved the food, hated the decor)? Do people care aboutdetermining sentiment attitudes of individual reviewers/utterances, orproducing an accurate assessment of aggregate sentiment?

Typically, these decisions are made by a product person and are passedoff to researchers and engineers to implement. The problem with thisapproach is that ML-core products are intimately constrained by what

What It Takes to Build Great Machine Learning Products | 39

is technically and algorithmically feasible. In my experience, having atechnical understanding of the range of related ML problems can in‐spire product ideas that might not occur to someone without this un‐derstanding. To draw a loose analogy, it’s like architecture. So much ofthe construction of a bridge is constrained by material resources andphysics that it doesn’t make sense to have people without that technicalbackground design a bridge.

The goal of all this is to say that if you want to build a rich ML product,you need to have a rich product/design/research/engineering team.All the way from the nitty gritty of how ML theory works to buildingsystems to domain knowledge to higher-level product thinking totechnical interaction and graphic design; preferably people who areworld-class in one of these areas but also good in several. Small talentedteams with all of these skills are better equipped to navigate the jointuncertainty with respect to product vision as well as model design.Large companies that have research and product people in entirelydifferent buildings are ill-equipped to tackle these kinds of problems.The ML products of the future will come from startups with smallfounding teams that have this full context and can all fit in the prov‐erbial garage.

40 | Chapter 3: Big Data Tools, Techniques, and Strategies

CHAPTER 4

The Application of Big Data

Stories over SpreadsheetsBy Mac Slocum

I didn’t realize how much I dislike spreadsheets until I was presentedwith a vision of the future where their dominance isn’t guaranteed.

That eye-opening was offered by Narrative Science CTO Kris Ham‐mond (@whisperspace) during a recent interview. Hammond’s com‐pany turns data into stories: They provide sentences and paragraphsinstead of rows and columns. To date, much of the attention NarrativeScience has received has focused on the media applications. That’s anatural starting point. Heck, I asked him about those very same thingswhen I first met Hammond at Strata in New York last fall. But duringour most recent chat, Hammond explored the other applications ofnarrative-driven data analysis.

“Companies, God bless them, had a great insight: They wanted to makedecisions based upon the data that’s out there and the evidence in frontof them,” Hammond said. “So they started gathering that data up. Itquickly exploded. And they ended up with huge data repositories theyhad to manage. A lot of their effort ended up being focused on gath‐ering that data, managing that data, doing analytics across that data,and then the question was: What do we do with it?”

Hammond sees an opportunity to extract and communicate the in‐sights locked within company data. “We’ll be the bridge between thedata you have, the insights that are in there, or insights we can gather,

and communicating that information to your clients, to your man‐agement, and to your different product teams. We’ll turn it into some‐thing that’s intelligible instead of a list of numbers, a spreadsheet, or agraph or two. You get a real narrative; a real story in that data.”

My takeaway: The journalism applications of this are intriguing, butthese other use cases are empowering.

Why? Because most people don’t speak fluent “spreadsheet.” They seeall those neat rows and columns and charts, and they know somethingimportant is tucked in there, but what that something is and how toextract it aren’t immediately clear. Spreadsheets require effort. That’sdoubly true if you don’t know what you’re looking for. And if dataanalysis is an adjacent part of a person’s job, more effort means thosespreadsheets will always be pushed to the side. “I’ll get to those nextweek when I’ve got more time…”

We all know how that plays out.

But what if the spreadsheet wasn’t our default output anymore? Whatif we could take things most of us are hard-wired to understand —stories, sentences, clear guidance — and layer it over all that vital data?Hammond touched on that:

For some people, a spreadsheet is a great device. For most people, notso much so. The story. The paragraph. The report. The prediction.The advisory. Those are much more powerful objects in our world,and they’re what we’re used to.

He’s right. Spreadsheets push us (well, most of us) into a cognitivecorner. Open a spreadsheet and you’re forced to recalibrate your focusto see the data. Then you have to work even harder to extract meaning.This is the best we can do?

With that in mind, I asked Hammond if the spreadsheet’s days arenumbered.

“There will always be someone who uses a spreadsheet,” Hammondsaid. “But, I think what we’re finding is that the story is really going tobe the endpoint. If you think about it, the spreadsheet is for somebodywho really embraces the data. And usually what that person does isthey reduce that data down to something that they’re going to use tocommunicate with someone else.”

42 | Chapter 4: The Application of Big Data

A Thought on DashboardsI used to view dashboards as the logical step beyond raw data andspreadsheets. I’m not so sure about that anymore, at least in terms ofbroad adoption. Dashboards are good tools, and I anticipate we’ll havethem from now until the end of time, but they’re still weighed downby a complexity that makes them inaccessible.

It’s not that people can’t master the buttons and custom reports indashboards; they simply don’t have time. These people — and I includemyself among them — need something faster and knob-free. Simplic‐ity is the thing that will ultimately democratize data reporting and datainsights. That’s why the expansion of data analysis requires a refine‐ment beyond our current dashboards. There’s a next step that hasn’tbeen addressed.

Does the answer lie in narrative? Will visualizations lead the way? Willa hybrid format take root? I don’t know what the final outputs will looklike, but the importance of data reporting means someone will even‐tually crack the problem.

Full InterviewYou can see the entire discussion with Hammond in this interview.

Mining the Astronomical LiteratureBy Alasdair Allan

There is a huge debate right now about making academic literaturefreely accessible and moving toward open access. But what would bepossible if people stopped talking about it and just dug in and got onwith it?

NASA’s Astrophysics Data System (ADS), hosted by the SmithsonianAstrophysical Observatory (SAO), has quietly been working awaysince the mid-’90s. Without much, if any, fanfare amongst the otherdisciplines, it has moved astronomers into a world where access to theliterature is just a given. It’s something they don’t have to think aboutall that much.

The ADS service provides access to abstracts for virtually all of theastronomical literature. But it also provides access to the full text of

Mining the Astronomical Literature | 43

more than half a million papers, going right back to the start of peer-reviewed journals in the 1800s. The service has links to online dataarchives, along with reference and citation information for each of thepapers, and it’s all searchable and downloadable.

Number of papers published in the three main astronomy journals eachyear. Credit: Robert Simpson

The existence of the ADS, along with the arXiv pre-print server, hasmeant that most astronomers haven’t seen the inside of a brick-builtlibrary since the late 1990s.

It also makes astronomy almost uniquely well placed for interestingdata mining experiments, experiments that hint at what the rest ofacademia could do if they followed astronomy’s lead. The fact that thediscipline’s literature has been scanned, archived, indexed and cata‐logued, and placed behind a RESTful API makes it a treasure trove,both for hypothesis generation and sociological research.

For example, the .Astronomy series of conferences is a small workshopthat brings together the best and brightest of the technical community:researchers, developers, educators, and communicators. Billed as“20% time for astronomers,” it gives these people space to think abouthow the new technologies affect both how research and communicat‐ing research to their peers and to the public is done.

[Disclosure: I’m a member of the advisory board to the .Astronomy con‐ference, and I previously served as a member of the programme organ‐ising committee for the conference series.]

44 | Chapter 4: The Application of Big Data

It should perhaps come as little surprise that one of the more inter‐esting projects to come out of a hack day held as part of this year’s .As‐tronomy meeting in Heidelberg was work by Robert Simpson, KarenMasters and Sarah Kendrew that focused on data mining the astro‐nomical literature.

The team grabbed and processed the titles and abstracts of all the pa‐pers from the Astrophysical Journal (ApJ), Astronomy & Astrophy‐sics (A&A), and the Monthly Notices of the Royal Astronomical So‐ciety (MNRAS) since each of those journals started publication — andthat’s 1827 in the case of MNRAS.

By the end of the day, they’d found some interesting results showinghow various terms have trended over time. The results were similar towhat’s found in Google Books’ Ngram Viewer.

The relative popularity of the names of telescopes in the literature. Hub‐ble, Chandra, and Spitzer seem to have taken turns in hogging the lime‐light, much as COBE, WMAP, and Planck have each contributed to ourknowledge of the cosmic microwave background in successive decades.References to Planck are still on the rise. Credit: Robert Simpson.

After the meeting, however, Robert took his initial results and exploredthe astronomical literature and his new corpus of data on the literature.He has explored various visualisations of the data, including wordmatrixes for related terms and for various astro-chemistry.

Mining the Astronomical Literature | 45

Correlation between terms related to Active Galactic Nuclei (AGN). Theopacity of each square represents the strength of the correlation betweenthe terms. Credit: Robert Simpson.

He has also taken a look at authorship in astronomy and is starting tofind some interesting trends.

46 | Chapter 4: The Application of Big Data

Fraction of astronomical papers published with one, two, three, four, ormore authors. Credit: Robert Simpson

You can see that single-author papers dominated for most of the 20thcentury. Around 1960, we see the decline begin, as two- and three-author papers begin to become a significant chunk of the whole. In1978, author papers become more prevalent than single-author pa‐pers.

Compare the number of “active” research astronomers to the number ofpapers published each year (across all the major journals). Credit: RobertSimpson.

Mining the Astronomical Literature | 47

Here we see that people begin to outpace papers in the 1960s. This mayreflect the fact that as we get more technical as a field, and more spe‐cialised, it takes more people to write the same number of papers,which is a sort of interesting result all by itself.

Interview with Robert Simpson: Behind the Project andWhat Lies AheadI recently talked with Rob about the work he, Karen Masters, and SarahKendrew did at the meeting, and the work he has been doing sincewith the newly gathered data.

What made you think about data mining the ADS?

Robert Simpson: At the .Astronomy 4 Hack Day in July, Sarah Ken‐drew had the idea to try to do an astronomy version of BrainSCANr,a project that generates new hypotheses in the neuroscience literature.I’ve had a go at mining ADS and arXiv before, so it seemed like a greatexcuse to dive back in.

Do you think there might be actual science that could be done here?

Robert Simpson: Yes, in the form of finding questions that were un‐expected. With such large volumes of peer-reviewed papers being pro‐duced daily in astronomy, there is a lot being said. Most researcherscan only try to keep up with it all — my daily RSS feed from arXiv isnext to useless, it’s so bloated. In amongst all that text, there must beconnections and relationships that are being missed by the communityat large, hidden in the chatter. Maybe we can develop simple techni‐ques to highlight potential missed links, i.e., generate new hypothesesfrom the mass of words and data.

Are the results coming out of the work useful for auditing academics?

Robert Simpson: Well, perhaps, but that would be tricky territory inmy opinion. I’ve only just begun to explore the data around authorshipin astronomy. One thing that is clear is that we can see a big trendtoward collaborative work. In 2012, only 6% of papers were single-author efforts, compared with 70+% in the 1950s.

48 | Chapter 4: The Application of Big Data

The above plot shows the average number of authors, per paper since1827. Credit: Robert Simpson.

We can measure how large groups are becoming, and who is part ofwhich groups. In that sense, we can audit research groups, and maybeindividual people. The big issue is keeping track of people throughvariations in their names and affiliations. Identifying authors is prob‐ably a solved problem if we look at ORCID.

What about citations? Can you draw any comparisons with h-indexdata?

Robert Simpson: I haven’t looked at h-index stuff specifically, at leastnot yet, but citations are fun. I looked at the trends surrounding theterm dark matter and saw something interesting. Mentions of darkmatter rise steadily after it first appears in the late ’70s.

Mining the Astronomical Literature | 49

Compare the term “dark matter” with a few other related terms: “cos‐mology,” “big bang,” “dark energy,” and “wmap.” You can see cosmologyhas been getting more popular since the 1990s, and dark energy is a recentaddition. Credit: Robert Simpson.

In the data, astronomy becomes more and more obsessed with darkmatter — the term appears in 1% of all papers by the end of the ’80sand 6% today.

Looking at citations changes the picture. The community is writingpapers about dark matter more and more each year, but they are gettingfewer citations than they used to (the peak for this was in the late ’90s).These trends are normalised, so the only regency effect I can think ofis that dark matter papers take more than 10 years to become citable.Either that or dark matter studies are currently in a trough for impact.

Can you see where work is dropped by parts of the community andpicked up again?

Robert Simpson: Not yet, but I see what you mean. I need to build abetter picture of the community and its components.

Can you build a social graph of astronomers out of this data? Whatabout (academic) family trees?

Robert Simpson: Identifying unique authors is my next step, followedby creating fingerprints of individuals at a given point in time. Whendo people create their first-author papers, when do they have the mostimpact in their careers, stuff like that.

What tools did you use? In hindsight, would you do it differently?

50 | Chapter 4: The Application of Big Data

I’m using Ruby and Perl to grab the data, MySQL to store and queryit, JavaScript to display it (Google Charts and D3.js). I may still movethe database part to MongoDB because it was designed to store docu‐ments. Similarly, I may switch from ADS to arXiv as the data source.Using arXiv would allow me to grab the full text in many cases, evenif it does introduce a peer-review issue.

What’s next?

Robert Simpson: My aim is still to attempt real hypothesis generation.I’ve begun the process by investigating correlations between terms inthe literature, but I think the power will be in being able to compareall terms with all terms and looking for the unexpected. Terms maycorrelate indirectly (via a third term, for example), so the entire corpusneeds to be processed and optimised to make it work comprehensively.

Science between the CracksI’m really looking forward to seeing more results coming out of Rob‐ert’s work. This sort of analysis hasn’t really been possible before. It’sshowing a lot of promise both from a sociological angle, with the abilityto do research into how science is done and how that has changed, butalso ultimately as a hypothesis engine — something that can generatenew science in and of itself. This is just a hack day experiment. Imaginewhat could be done if the literature were more open and this sort ofanalysis could be done across fields?

Right now, a lot of the most interesting science is being done in thecracks between disciplines, but the hardest part of that sort of work isoften trying to understand the literature of the discipline that isn’t yourown. Robert’s project offers a lot of hope that this may soon becomeeasier.

The Dark Side of DataBy Mike Loukides

Tom Slee’s “Seeing Like a Geek” is a thoughtful article on the dark sideof open data. He starts with the story of a Dalit community in India,whose land was transferred to a group of higher cast Mudaliarsthrough bureaucratic manipulation under the guise of standardizingand digitizing property records. While this sounds like a good idea, itgave a wealthier, more powerful group a chance to erase older, tradi‐

The Dark Side of Data | 51

tional records that hadn’t been properly codified. One effect of passinglaws requiring standardized, digital data is to marginalize all data thatcan’t be standardized or digitized, and to marginalize the people whodon’t control the process of standardization.

That’s a serious problem. It’s sad to see oppression and property theftriding in under the guise of transparency and openness. But the issueisn’t open data, but how data is used.

Jesus said “the poor are with you always” not because the poor aren’ta legitimate area of concern (only an American fundamentalist wouldsay that), but because they’re an intractable problem that won’t goaway. The poor are going to be the victims of any changes in technol‐ogy; it isn’t surprisingly that the wealthy in India used data to mar‐ginalize the land holdings of the poor. In a similar vein, when Euro‐peans came to North America, I imagine they asked the natives “So,you got a deed to all this land?,” a narrative that’s still being playedout with indigenous people around the world.

The issue is how data is used. If the wealthy can manipulate legislatorsto wipe out generations of records and folk knowledge as “inaccurate,”then there’s a problem. A group like DataKind could go in and figureout a way to codify that older generation of knowledge. Then at least,if that isn’t acceptable to the government, it would be clear that theproblem lies in political manipulation, not in the data itself. And notethat a government could wipe out generations of “inaccurate records”without any requirement that the new records be open. In years pastthe monied classes would have just taken what they wanted, with thegovernment’s support. The availability of open data gives a plausiblepretext, but it’s certainly not a prerequisite (nor should it be blamed)for manipulation by the 0.1%.

One can see the opposite happening, too: the recent legislation inNorth Carolina that you can’t use data that shows sea level rise. Opendata may be the only possible resource against forces that are interestedin suppressing science. What we’re seeing here is a full-scale retreatfrom data and what it can teach us: an attempt to push the furnitureagainst the door to prevent the data from getting in and changing theway we act.

The Digital Publishing LandscapeSlee is on shakier ground when he claims that the digitization of bookshas allowed Amazon to undermine publishers and booksellers. Yes,

52 | Chapter 4: The Application of Big Data

there’s technological upheaval, and that necessarily drives changes inbusiness models. Business models change; if they didn’t, we’d still havethe Pony Express and stagecoaches. O’Reilly Media is thriving, in partbecause we have a viable digital publishing strategy; publishers withouta viable digital strategy are failing.

But what about booksellers? The demise of the local bookstore has, inmy observation, as much to do with Barnes & Noble superstores (andthe now-defunct Borders), as with Amazon, and it played out longbefore the rise of ebooks.

I live in a town in southern Connecticut, roughly a half-hour’s drivefrom the two nearest B&N outlets. Guilford and Madison, the townimmediately to the east, both have thriving independent bookstores.One has a coffeeshop, stages many, many author events (roughly onea day), and runs many other innovative programs (birthday parties,book-of-the-month services, even ebook sales). The other is just asmall local bookstore with a good collection and knowledgeable staff.The town to the west lost its bookstore several years ago, possibly be‐fore Amazon even existed. Long before the Internet became a factor,it had reduced itself to cheap gift items and soft porn magazines. So:data may threaten middlemen, though it’s not at all clear to me thatmiddlemen can’t respond competitively. Or that they are really threat‐ened by “data,” as opposed to large centralized competitors.

There are also countervailing benefits. With ebooks, access is demo‐cratized. Anyone, anywhere has access to what used to be availableonly in limited, mostly privileged locations. At O’Reilly, we now sellebooks in countries we were never able to reach in print. Our printsales overseas never exceeded 30% of our sales; for ebooks, overseasrepresents more than half the total, with customers as far away asAzerbaijan.

Slee also points to the music labels as an industry that has been margi‐nalized by open data. I really refuse to listen to whining about all themoney that the music labels are losing. We’ve had too many years ofcrap product generated by marketing people who only care aboutfinding the next Justin Bieber to take the “creative industry” and itssycophants seriously.

Privacy by DesignData inevitably brings privacy issues into play. As Slee points out (andas Jeff Jonas has before him), apparently insignificant pieces of data

The Dark Side of Data | 53

can be put together to form a surprisingly accurate picture of who youare, a picture that can be sold. It’s useless to pretend that there won’tbe increased surveillance in any forseeable future, or that there won’tbe an increase in targeted advertising (which is, technically, much thesame thing).

We can bemoan that shift, celebrate it, or try to subvert it, but we can’tpretend that it isn’t happening. We shouldn’t even pretend that it’s new,or that it has anything to do with openness. What is a credit bureau ifnot an organization that buys and sells data about your financial his‐tory, with no pretense of openness?

Jonas’s concept of “privacy by design” is an important attempt to ad‐dress privacy issues in big data. Jonas envisions a day when “I havemore privacy features than you” is a marketing advantage. It’s certainlya claim I’d like to see Facebook make.

Absent a solution like Jonas’s, data is going to be collected, bought,sold, and used for marketing and other purposes, whether it is “open”or not. I do not think we can get to Jonas’s world, where privacy issomething consumers demand, without going through a stage wheredata is open and public. It’s too easy to live with the illusion of privacythat thrives in a closed world.

I agree that the notion that “open data” is an unalloyed public good ismistaken, and Tom Slee has done a good job of pointing that out. Itunderscores the importance of a still-nascent ethical consensus abouthow to use data, along with the importance of data watchdogs, Data‐Kind, and other organizations devoted to the public good. (I don’tunderstand why he argues that Apple and Amazon “undermine com‐munity activism”; that seems wrong, particularly in the light of Apple’sre-joining the EPEAT green certification system for their productsafter a net-driven consumer protest.) Data collection is going to hap‐pen whether we like it or not, and whether it’s open or not. I am con‐vinced that private data is a public bad, and I’m less afraid of data that’sopen. That doesn’t make it necessarily a good; that depends on howthe data is used, and the people who are using it.

54 | Chapter 4: The Application of Big Data

CHAPTER 5

What to Watch for in Big Data

Big Data Is Our Generation’s Civil Rights Issue,and We Don’t Know ItBy Alistair Croll

Data doesn’t invade people’s lives. Lack of control over how it’s used does.

What’s really driving so-called big data isn’t the volume of information.It turns out big data doesn’t have to be all that big. Rather, it’s about areconsideration of the fundamental economics of analyzing data.

For decades, there’s been a fundamental tension between three at‐tributes of databases. You can have the data fast; you can have it big;or you can have it varied. The catch is, you can’t have all three at once.

I first heard this as the “three V’s of data”: Volume, Variety, and Ve‐locity. Traditionally, getting two was easy but getting three was very,very, very expensive.

The advent of clouds, platforms like Hadoop, and the inexorable marchof Moore’s Law means that now, analyzing data is trivially inexpensive.And when things become so cheap that they’re practically free, bigchanges happen — just look at the advent of steam power, or the copy‐ing of digital music, or the rise of home printing. Abundance replacesscarcity, and we invent new business models.

In the old, data-is-scarce model, companies had to decide what to col‐lect first, and then collect it. A traditional enterprise data warehousemight have tracked sales of widgets by color, region, and size. This actof deciding what to store and how to store it is called designing theschema, and in many ways, it’s the moment where someone decideswhat the data is about. It’s the instant of context.

That needs repeating:

You decide what data is about the moment you define its schema.

With the new, data-is-abundant model, we collect first and ask ques‐tions later. The schema comes after the collection. Indeed, big datasuccess stories like Splunk, Palantir, and others are prized because oftheir ability to make sense of content well after it has been collected —sometimes called a schema-less query. This means we collect infor‐mation long before we decide what it’s for.

And this is a dangerous thing.

When bank managers tried to restrict loans to residents of certain areas(known as redlining), Congress stepped in to stop it (with the FairHousing Act of 1968). They were able to legislate against discrimina‐tion, making it illegal to change loan policy based on someone’s race.

56 | Chapter 5: What to Watch for in Big Data

Home Owners’ Loan Corporation map showing redlining of “hazardous”districts in 1936. Credit: Wikipedia

“Personalization” is another word for discrimination. We’re not dis‐criminating if we tailor things to you based on what we know aboutyou — right? That’s just better service.

In one case, American Express used purchase history to adjust creditlimits based on where a customer shopped, despite his excellent creditlimit:

Johnson says his jaw dropped when he read one of the reasons Amer‐ican Express gave for lowering his credit limit: “Other customers whohave used their card at establishments where you recently shoppedhave a poor repayment history with American Express.”

We’re seeing the start of this slippery slope everywhere from tailoredcredit-card limits like this one to car insurance based on driver pro‐files. In this regard, big data is a civil rights issue, but it’s one that societyin general is ill-equipped to deal with.

We’re great at using taste to predict things about people. OKcupid’s2010 blog post “The Real Stuff White People Like” showed just howeasily we can use information to guess at race. It’s a real eye-opener

Big Data Is Our Generation’s Civil Rights Issue, and We Don’t Know It | 57

(and the guys who wrote it didn’t include everything they learned —some of it was a bit too controversial). They simply looked at the wordsone group used which others didn’t often use. The result was a list of“trigger” words for a particular race or gender.

Now run this backwards. If I know you like these things, or see youmention them in blog posts, on Facebook, or in tweets, then there’s agood chance I know your gender and your race, and maybe even yourreligion and your sexual orientation. And that I can personalize mymarketing efforts towards you.

That makes it a civil rights issue.

If I collect information on the music you listen to, you might assumeI will use that data in order to suggest new songs, or share it with yourfriends. But instead, I could use it to guess at your racial background.And then I could use that data to deny you a loan.

Want another example? Check out Private Data In Public Ways, some‐thing I wrote a few months ago after seeing a talk at Big Data London,which discusses how publicly available last name information can beused to generate racial boundary maps:

Screen from the Mapping London project.

This TED talk by Malte Spitz does a great job of explaining the chal‐lenges of tracking citizens today, and he speculates about whether theBerlin Wall would ever have come down if the Stasi had access to phonerecords in the way today’s governments do.

58 | Chapter 5: What to Watch for in Big Data

So how do we regulate the way data is used?

The only way to deal with this properly is to somehow link what thedata is with how it can be used. I might, for example, say that my musicaltastes should be used for song recommendation, but not for bankingdecisions.

Tying data to permissions can be done through encryption, which isslow, riddled with DRM, burdensome, hard to implement, and bad forinnovation. Or it can be done through legislation, which has about asmuch chance of success as regulating spam: it feels great, but it’sdamned hard to enforce.

There are brilliant examples of how a quantified society can improvethe way we live, love, work, and play. Big data helps detect diseaseoutbreaks, improve how students learn, reveal political partisanship,and save hundreds of millions of dollars for commuters — to pick justfour examples. These are benefits we simply can’t ignore as we try tosurvive on a planet bursting with people and shaken by climate andenergy crises.

But governments need to balance reliance on data with checks andbalances about how this reliance erodes privacy and creates civil andmoral issues we haven’t thought through. It’s something that most ofthe electorate isn’t thinking about, and yet it affects every purchasethey make.

This should be fun.

Big Data Is Our Generation’s Civil Rights Issue, and We Don’t Know It | 59

Three Kinds of Big DataBy Alistair Croll

In the past couple of years, marketers and pundits have spent a lot oftime labeling everything “big data.” The reasoning goes something likethis:

• Everything is on the Internet.• The Internet has a lot of data.• Therefore, everything is big data.

When you have a hammer, everything looks like a nail. When you havea Hadoop deployment, everything looks like big data. And if you’retrying to cloak your company in the mantle of a burgeoning industry,big data will do just fine. But seeing big data everywhere is a sure wayto hasten the inevitable fall from the peak of high expectations to thetrough of disillusionment.

We saw this with cloud computing. From early idealists saying every‐thing would live in a magical, limitless, free data center to today’spragmatism about virtualization and infrastructure, we soon took offour rose-colored glasses and put on welding goggles so we could ac‐tually build stuff.

So where will big data go to grow up?

Once we get over ourselves and start rolling up our sleeves, I think bigdata will fall into three major buckets: Enterprise BI, Civil Engineering,and Customer Relationship Optimization. This is where we’ll see mostIT spending, most government oversight, and most early adoption inthe next few years.

Enterprise BI 2.0For decades, analysts have relied on business intelligence (BI) productslike Hyperion, Microstrategy and Cognos to crunch large amounts ofinformation and generate reports. Data warehouses and BI tools aregreat at answering the same question — such as “what were Mary’ssales this quarter?” — over and over again. But they’ve been less good

60 | Chapter 5: What to Watch for in Big Data

at the exploratory, what-if, unpredictable questions that matter forplanning and decision-making because that kind of fast explorationof unstructured data is traditionally hard to do and therefore expen‐sive.

Most “legacy” BI tools are constrained in two ways:

• First, they’ve been schema-then-capture tools in which the analystdecides what to collect, then later captures that data for analysis.

• Second, they’ve typically focused on reporting what AvinashKaushik (channeling Donald Rumsfeld) refers to as “known un‐knowns” — things we know we don’t know, and generate reportsfor.

These tools are used for reporting and operational purposes, and areusually focused on controlling costs, executing against an existingplan, and reporting on how things are going.

As my Strata co-chair Edd Dumbill pointed out when I asked forthoughts on this piece:

The predominant functional application of big data technologies to‐day is in ETL (Extract, Transform, and Load). I’ve heard the figurethat it’s about 80% of Hadoop applications. Just the real grunt workof log file or sensor processing before loading into an analytic databaselike Vertica.

The availability of cheap, fast computers and storage, as well as opensource tools, have made it okay to capture first and ask questions later.That changes how we use data because it lets analysts speculate beyondthe initial question that triggered the collection of data.

What’s more, the speed with which we can get results — sometimes asfast as a human can ask them — makes data easier to explore interac‐tively. This combination of interactivity and speculation takes BI intothe realm of “unknown unknowns,” the insights that can produce acompetitive advantage or an out-of-the-box differentiator.

Cloud computing underwent a transition from promise to compro‐mise. First, big, public clouds wooed green-field startups. Then, a fewyears later, incumbent IT vendors introduced private cloud offerings.These private clouds included only a fraction of the benefits of theirpublic cousins — but were nevertheless a sufficient blend of smoke,

Three Kinds of Big Data | 61

mirrors, and features to delay the inevitable move to public resourcesby a few years and appease the business. For better or worse, that’swhere most IT cloud budgets are being spent today according to IDC,Gartner, and others. Big data adoption will undergo a similar cycle.

In the next few years, then, look for acquisitions and product intro‐ductions — and not a little vaporware — as BI vendors that enterprisestrust bring them “big data lite”: enough innovation and disruption tosatisfy the CEO’s golf buddies, but not so much that enterprise IT’sjobs are threatened. This, after all, is how change comes to big organ‐izations.

Ultimately, we’ll see traditional “known unknowns” BI reporting livingalongside big-data-powered data import and cleanup, and fast, ex‐ploratory data “unknown unknown” interactivity.

Civil EngineeringThe second use of big data is in society and government. Already, datamining can be used to predict disease outbreaks, understand trafficpatterns, and improve education.

Cities are facing budget crunches, infrastructure problems, and crowd‐ing from rural citizens. Solving these problems is urgent, and cities areperfect labs for big data initiatives. Take a metropolis like New York:hackathons; open feeds of public data; and a population that generatesa flood of information as it shops, commutes, gets sick, eats, and justgoes about its daily life.

I think municipal data is one of the big three for several reasons: it’s agood tie breaker for partisanship, we have new interfaces everyonecan understand, and we finally have a mostly-connected citizenry.

In an era of partisan bickering, hard numbers can settle the debate. So,they’re not just good government; they’re good politics. Expect to seebig data applied to social issues, helping us to make funding moreeffective and scarce government resources more efficient (perhaps tothe chagrin of some public servants and lobbyists). As this works inthe world’s biggest cities, it’ll spread to smaller ones, to states, and tomunicipalities.

62 | Chapter 5: What to Watch for in Big Data

Making data accessible to citizens is possible, too: Siri and Google Nowshow the potential for personalized agents; Narrative Science takescomplex data and turns it into words the masses can consume easily;Watson and Wolfram Alpha can give smart answers, either throughcurated reasoning or making smart guesses.

For the first time, we have a connected citizenry armed (for the mostpart) with smartphones. Nielsen estimated that smartphones wouldovertake feature phones in 2011, and that concentration is high inurban cores. The App Store is full of apps for bus schedules, commut‐ers, local events, and other tools that can quickly become how gov‐ernments connect with their citizens and manage their bureaucracies.

The consequence of all this, of course, is more data. Once governmentsgo digital, their interactions with citizens can be easily instrumentedand analyzed for waste or efficiency. That’s sure to provoke resistancefrom those who don’t like the scrutiny or accountability, but it’s a sideeffect of digitization: every industry that goes digital gets analyzed andoptimized, whether it likes it or not.

Customer Relationship OptimizationThe final home of applied big data is marketing. More specifically, it’simproving the relationship with consumers so companies can, as Ser‐gio Zyman once said, sell them more stuff, more often, for more money,more efficiently.

The biggest data systems today are focused on web analytics, ad opti‐mization, and the like. Many of today’s most popular architectureswere weaned on ads and marketing, and have their ancestry in directmarketing plans. They’re just more focused than the comparativelyblunt instruments with which direct marketers used to work.

The number of contact points in a company has multiplied signifi‐cantly. Where once there was a phone number and a mailing address,today there are web pages, social media accounts, and more. Trackingusers across all these channels — and turning every click, like, share,friend, or retweet into the start of a long funnel that leads, inexorably,to revenue is a big challenge. It’s also one that companies like Salesforceunderstand, with its investments in chat, social media monitoring, co-browsing, and more.

Three Kinds of Big Data | 63

This is what’s lately been referred to as the “360-degree customer view”(though it’s not clear that companies will actually act on customerdata if they have it, or whether doing so will become a complianceminefield). Big data is already intricately linked to online marketing,but it will branch out in two ways.

First, it’ll go from online to offline. Near-field-equipped smartphoneswith ambient check-in are a marketer’s wet dream, and they’re comingto pockets everywhere. It’ll be possible to track queue lengths, storetraffic, and more, giving retailers fresh insights into their brick-and-mortar sales. Ultimately, companies will bring the optimization thatonline retail has enjoyed to an offline world as consumers becometrackable.

Second, it’ll go from Wall Street (or maybe that’s Madison Avenue andMiddlefield Road) to Main Street. Tools will get easier to use, and whilesmall businesses might not have a BI platform, they’ll have a tablet ora smartphone that they can bring to their places of business. Mobilepayment players like Square are already making them reconsider thecheckout process. Adding portable customer intelligence to the toolsuite of local companies will broaden how we use marketing tools.

Headlong into the TroughThat’s my bet for the next three years, given the molasses of marketconfusion, vendor promises, and unrealistic expectations we’re aboutto contend with. Will big data change the world? Absolutely. Will it beable to defy the usual cycle of earnest adoption, crushing disappoint‐ment, and eventual rebirth all technologies must travel? Certainly not.

Automated Science, Deep Data, and theParadox of InformationBy Bradley Voytek

A lot of great pieces have been written about the relatively recent surgein interest in big data and data science, but in this piece I want toaddress the importance of deep data analysis: what we can learn fromthe statistical outliers by drilling down and asking, “What’s differenthere? What’s special about these outliers, and what do they tell us aboutour models and assumptions?”

64 | Chapter 5: What to Watch for in Big Data

The reason that big data proponents are so excited about the bur‐geoning data revolution isn’t just because of the math. Don’t get mewrong, the math is fun, but we’re excited because we can begin to distillpatterns that were previously invisible to us due to a lack of information.

That’s big data.

Of course, data are just a collection of facts; bits of information thatare only given context — assigned meaning and importance — byhuman minds. It’s not until we do something with the data that any ofit matters. You can have the best machine learning algorithms, thetightest statistics, and the smartest people working on them, but noneof that means anything until someone makes a story out of the results.

And therein lies the rub.

Do all these data tell us a story about ourselves and the universe inwhich we live, or are we simply hallucinating patterns that we want tosee?

(Semi)Automated ScienceIn 2010, Cornell researchers Michael Schmidt and Hod Lipson pub‐lished a groundbreaking paper in Science titled “Distilling Free-FormNatural Laws from Experimental Data.” The premise was simple, andit essentially boiled down to the question, “can we algorithmically ex‐tract models to fit our data?”

So they hooked up a double pendulum — a seemingly chaotic systemwhose movements are governed by classical mechanics — and traineda machine learning algorithm on the motion data.

Their results were astounding. (See them here.)

In a matter of minutes the algorithm converged on Newton’s secondlaw of motion: f = ma. What took humanity tens of thousands of yearsto accomplish was completed on 32-cores in essentially no time at all.

In 2011, some neuroscience colleagues of mine, lead by Tal Yarkoni,published a paper in Nature Methods titled “Large-scale automatedsynthesis of human functional neuroimaging data.” In this paper, theauthors sought to extract patterns from the overwhelming flood ofbrain imaging research.

To do this, they algorithmically extracted the 3D coordinates of sig‐nificant brain activations from thousands of neuroimaging studies,

Automated Science, Deep Data, and the Paradox of Information | 65

along with words that frequently appeared in each study. Using thesetwo pieces of data along with some simple (but clever) mathematicaltools, they were able to create probabilistic maps of brain activationfor any given term.

In other words, you type in a word such as “learning” on their websitesearch and visualization tool, NeuroSynth, and they give you back apattern of brain activity that you should expect to see during a learningtask.

But that’s not all. Given a pattern of brain activation, the system canperform a reverse inference, asking, “given the data that I’m observing,what is the most probable behavioral state that this brain is in?”

Similarly, in late 2010, my wife (Jessica Voytek) and I undertook aproject to algorithmically discover associations between concepts inthe peer-reviewed neuroscience literature. As a neuroscientist, the goalof my research is to understand relationships between the humanbrain, behavior, physiology, and disease. Unfortunately, the facts thattie all that information together are locked away in more than 21 mil‐lion static peer-reviewed scientific publications.

How many undergrads would I need to hire to read through that manypapers? Any volunteers?

Even more mind-boggling, each year more than 30,000 neuroscientistsattend the annual Society for Neuroscience conference. If we assumethat only two-thirds of those people actually do research, and if weassume that they only work a meager (for the sciences) 40 hours aweek, that’s around 40 million person-hours dedicated to but onebranch of the sciences.

Annually.

This means that in the 10 years I’ve been attending that conference,more than 400 million person-hours have gone toward the pursuit ofunderstanding the brain. Humanity built the pyramids in 30 years. TheApollo Project got us to the moon in about eight.

So my wife and I said to ourselves, “there has to be a better way.”

Which lead us to create brainSCANr, a simple (simplistic?) tool (cur‐rently itself under peer review) that makes the assumption that themore often two concepts appear together in the titles or abstracts ofpublished papers, the more likely they are to be associated with oneanother.

66 | Chapter 5: What to Watch for in Big Data

For example, if 10,000 papers mention “Alzheimer’s disease” that al‐so mention “dementia,” then Alzheimer’s disease is probably related todementia. In fact, there are 17,087 papers that mention Alzheimer’sand dementia, whereas there are only 14 papers that mention Alz‐heimer’s and, for example, creativity.

From this, we built what we’re calling the “cognome,” a mapping be‐tween brain structure, function, and disease.

Big data, data mining, and machine learning are becoming criticaltools in the modern scientific arsenal. Examples abound: text miningrecipes to find cultural food taste preferences, analyzing cultural trendsvia word use in books (“culturomics”), identifying seasonality of moodfrom tweets, and so on.

But so what?

Deep DataWhat those three studies show us is that it’s possible to automate, orat least semi-automate, critical aspects of the scientific method itself.Schmidt and Lipson show that it is possible to extract equations thatperfectly model even seemingly chaotic systems. Yarkoni and collea‐gues show that it is possible to infer a complex behavioral state giveninput brian data.

My wife and I wanted to show that brainSCANr could be put to workfor something more useful than just quantifying relationships betweenterms. So we created a simple algorithm to perform what we’re calling“semi-automated hypothesis generation,” which is predicated on a ba‐sic “the friend of a friend should be a friend” concept.

In the example below, the neurotransmitter “serotonin” has thousandsof shared publications with “migraine,” as well as with the brain region“striatum.” However, migraine and striatum only share 16 publica‐tions.

Automated Science, Deep Data, and the Paradox of Information | 67

That’s very odd. Because in medicine there is a serotonin hypothesisfor the root cause of migraines. And we (neuroscientists) know thatserotonin is released in the striatum to modulate brain activity in thatregion. Given that those two things are true, why is there so little re‐search regarding the role of the striatum in migraines?

Perhaps there’s a missing connection?

Such missing links and other outliers in our models are the essence ofdeep data analytics. Sure, any data scientist worth their salt can take amountain of data and reduce it down to a few simple plots. And suchplots are important because they tell a story. But those aren’t the onlystories that our data can tell us.

For example, in my geoanalytics work as the data evangelist for Uber,I put some of my (definitely rudimentary) neuroscience network an‐alytic skills to work to figure out how people move from neighborhoodto neighborhood in San Francisco.

At one point, I checked to see if men and women moved around thecity differently. A very simple regression model showed that the num‐ber of men who go to any given neighborhood significantly predictsthe number of women who go to that same neighborhood.

No big deal.

68 | Chapter 5: What to Watch for in Big Data

But what’s cool was seeing where the outliers were. When I looked atthe models’ residuals, that’s where I found the far more interestingstory. While it’s good to have a model that fits your data, knowingwhere the model breaks down is not only important for internal met‐rics, but it also makes for a more interesting story: What’s happeningin the Marina district that so many more women want to go there?And why are there so many more men in SoMa?

The Paradox of InformationThe interpretation of big data analytics can be a messy game. Maybethere are more men in SoMa because that’s where AT&T Park is. Butmaybe there are just five guys who live in SoMa who happen to takeUber 100 times more often than average.

While data-driven posts make for fun reading (and writing), in thesciences we need to be more careful that we don’t fall prey to ad hoc,just-so stories that sound perfectly reasonable and plausible, but whichwe cannot conclusively prove.

In 2008, psychologists David McCabe and Alan Castel published apaper in the journal Cognition, titled “Seeing is believing: The effectof brain images on judgments of scientific reasoning.” In that paper,they showed that summaries of cognitive neuroscience findings thatare accompanied by an image of a brain scan were rated as more credi‐ble by the readers.

This should cause any data scientist serious concern. In fact, I’ve for‐mulated three laws of statistical analyses:

1. The more advanced the statistical methods used, the fewer criticsare available to be properly skeptical.

2. The more advanced the statistical methods used, the more likelythe data analyst will be to use math as a shield.

3. Any sufficiently advanced statistics can trick people into believingthe results reflect truth.

The first law is closely related to the “bike shed effect” (also known asParkinson’s Law of Triviality) which states that, “the time spent on anyitem of the agenda will be in inverse proportion to the sum involved.”

Automated Science, Deep Data, and the Paradox of Information | 69

In other words, if you try to build a simple thing such as a public bikeshed, there will be endless town hall discussions wherein people argueover trivial details such as the color of the door. But if you want to builda nuclear power plant — a project so vast and complicated that mostpeople can’t understand it — people will defer to expert opinion.

Such is the case with statistics.

If you make the mistake of going into the comments section of anynews piece discussing a scientific finding, invariably someone willleave the comment, “correlation does not equal causation.”

We’ll go ahead and call that truism Voytek’s fourth law.

But people rarely have the capacity to argue against the methods andmodels used by, say, neuroscientists or cosmologists.

But sometimes we get perfect models without any understanding ofthe underlying processes. What do we learn from that?

The always fantastic Radiolab did a follow-up story on the Schmidtand Lipson “automated science” research in an episode titled “Limitsof Science.” It turns out, a biologist contacted Schmidt and Lipson andgave them data to run their algorithm on. They wanted to figure outthe principles governing the dynamics of a single-celled bacterium.Their result?

Well sometimes the stories we tell with data…they just don’t makesense to us.

They found “two equations that describe the data.”

But they didn’t know what the equations meant. They had no context.Their variables had no meaning. Or, as Radiolab co-host Jad Abum‐rad put it, “the more we turn to computers with these big questions,the more they’ll give us answers that we just don’t understand.”

So while big data projects are creating ridiculously exciting new vistasfor scientific exploration and collaboration, we have to take care toavoid the Paradox of Information wherein we can know too manythings without knowing what those “things” are.

Because at some point, we’ll have so much data that we’ll stop beingable to discern the map from the territory. Our goal as (data) scientistsshould be to distill the essence of the data into something that tells as

70 | Chapter 5: What to Watch for in Big Data

true a story as possible while being as simple as possible to understand.Or, to operationalize that sentence better, we should aim to find bal‐ance between minimizing the residuals of our models and maximizingour ability to make sense of those models.

Recently, Stephen Wolfram released the results of a 20-year long ex‐periment in personal data collection, including every keystroke he’styped and every email he’s sent. In response, Robert Krulwich, theother co-host of Radiolab, concludes by saying “I’m looking at yourdata [Dr. Wolfram], and you know what’s amazing to me? How muchof you is missing.”

Personally, I disagree; I believe that there’s a humanity in those num‐bers and that Mr. Krulwich is falling prey to the idea that sciencesomehow ruins the magic of the universe. Quoth Dr. Sagan:

It is sometimes said that scientists are unromantic, that their passionto figure out robs the world of beauty and mystery. But is it not stirringto understand how the world actually works — that white light is madeof colors, that color is the way we perceive the wavelengths of light,that transparent air reflects light, that in so doing it discriminatesamong the waves, and that the sky is blue for the same reason that thesunset is red? It does no harm to the romance of the sunset to knowa little bit about it.

So go forth and create beautiful stories, my statistical friends. See youafter peer-review.

The Chicken and Egg of Big Data SolutionsBy Jim Stogdill

Before I came to O’Reilly I was building the “big data and disruptiveanalytics practice” at a major systems integrator. It was a blast to spendevery week talking to customers in different industries who were wak‐ing up to the possibilities that technologies like Hadoop offered theirbusinesses. Many of these businesses are going to fundamentallychange as they embrace this stuff (or be replaced by those that do). Butthere’s a catch.

Twenty years or so ago large integrators made big business buildingapplications on the then-new relational paradigm. They put in Oracle

The Chicken and Egg of Big Data Solutions | 71

databases with custom code, wrote PowerBuilder apps on Sybase, andof course lots of businesses rolled their own with VB and SQL Server.It was an era of custom coding where Oracle, Sybase, SQL Server, In‐formix and etc. were thought of as platforms to build stuff on.

Then the market matured and shifted to package solution implemen‐tation. ERP, CRM,…, etc. The big guys focused on integrating againand told their clients there was no ROI in building custom stuff. ROIwould come from integrating best-of-breed solutions. Databases be‐came commodity back ends to the applications that were always thereal focus.

Now along comes big data, NoSQL, data science, and all that stuff andit seems like we’re starting the cycle over again. But this time clients,having been well trained over the last decade or so, aren’t having anyof that “build it from scratch” mentality. They know that Hadoop andother new technologies can be transformative to their business, butthey want it packaged up and solution’ified like they are used to. I hearda lot of “let us know when you have a solution already built or availableto buy that does X” in the last year.

Also, lots of the shops that do this stuff at scale are built and staffedaround the package implementation model and have shed many of theskills they used to have for custom work. Everything from staffingmodels to methodologies are oriented toward package installation.

So, here we are with all of this disruptive technology, but we seem tohave lost the institutional wherewithal to do anything with it in a lotof large companies. Of course that fact was hard on my numbers. I hada great pipeline of companies with pain to solve, and great technologiesto solve it, but too much of the time it was hard to close it withoutreadymade solutions.

Every week I talked to the companies building these new platforms toshare leads and talk about their direction. After a while I started cuttingthem off when they wanted to talk about the features of their nextrelease. I just got to the point where I didn’t really care, it just wasn’tall that relevant to my customers. I mean, it’s important that they aremaking the platforms more manageable and building bridges to tra‐ditional BI, ETL, RDBMS, and the like. But the focus was too much onplatforms and tools.

I wanted to know “What are you doing to encourage solution devel‐opment? Are you staffing a support system for ISVs? What startups

72 | Chapter 5: What to Watch for in Big Data

and/or established players are you aware of that are building solutionson this platform?” So when I saw this tweet I let out a little yelp. Awe‐some! The lack of ready-to-install solutions was getting attention, andfrom Mike Olsen.

You can watch the rest of what Mike Olson said here and you’ll findhe tells a similar story about the RDBMS historical parallel.

I talked to Mike a few weeks ago to find out what was behind hiscomment and explore what else they are doing to support solutiondevelopment. It boils down to what he said — he will help connect youwith money — plus a newly launched partner program designed toprovide better support to ISVs among others. Also, the continued at‐tention to APIs and tools like Pig and Hive should make it easier forthe solution ecosystem to develop. It can only be good for his businessto have lots of other companies directly solving business problems,and simply pulling in his platform.

Hortonworks also started a partner program in the fall, and I thinkwe’ll see a lot more emphasis on this across the space this year. How‐ever, at the moment wherever I look (Hortonworks Partners, ClouderaPartners, Accel big data portfolio) the focus today remains firmly onplatform and tools or partnering with integrators. Tresata, a startupfocused on financial risk management, pops up in a lot of lists as theobvious odd one out — an actual domain-specific solution.

What about other people that could be building solutions? Is it thematurity level of the technology, the lack of penetration of Hadoop,etc., into your customer’s data centers, or some combination of otherfactors that is slowing things down?

Of course, during the RDBMS adoption it took a lot of years beforethe custom era was over and thoroughly replaced by the era of packageimplementation. The question I’m pondering is whether customer ex‐pectations and the pace of technology will make it happen faster thistime? Or is the disruptive value of big data going to continue to accrueonly to risk-taking early adopters for the foreseeable future?

Walking the Tightrope of VisualizationCriticismBy Andy Kirk

Walking the Tightrope of Visualization Criticism | 73

In a talk at the SEE conference in Germany, data illustrator StefaniePosavec opened her talk with a sobering observation about how shehad found the data visualization field really intimidating.

Her experience was that many visualization bloggers and active par‐ticipants seem to believe in one right way and lots of wrong ways tocreate a visualization. To those entering the field, these types of viewswill create a fair amount of confusion, inconsistency, and contradic‐tion. It demonstrates our current glass-is-never-full tendency towardcritical evaluation.

This should act as an important wake-up call to all of us who care aboutmaintaining an accessible and supportive community around datavisualization and infographic design, particularly as these disciplinescontinue to penetrate the mainstream consciousness.

The fear is that Posavec’s expressing of this view is just the tip of aniceberg. Who knows how many designers outside of the spotlight holda similar perception and are reluctant to share their work and engagewith the field?

In this article, I seek to take a detached view of the visualization fieldand weave in my experiences from delivering introductory data visu‐alization training courses over the past year. I want to take a look atthe constituency of this discipline and the role of critique to see howPosavec’s experiences could have materialised and contrast them withthe people I meet in my classes.

The Visualization EcosystemOne of the most rewarding personal experiences from my trainingcourses has been getting the chance to mix with a variety of delegatesfrom different countries, cultures, occupations, and industries. Spend‐ing time with essentially everyday people, learning as much from themas they do from me, has been hugely refreshing. The term “everydaypeople” could be perceived as condescending, but far from it. Allowme to elaborate.

When you are an active participant in a field like data visualizationyou spend most of your time consuming and digesting informationfrom your peers. This can create a bubble of exposure to just thosehardcore connoisseurs — the academics, the authors, the designers,and the bloggers — who have spent years refining their knowledge andperfecting their craft.

74 | Chapter 5: What to Watch for in Big Data

This is the sharp-end of the field where the intensity of debate, knowl‐edge exchange, and opinion expression is high. The observationsemerging at this level represent the most perceptive, creative, andcomprehensive insights into the design techniques on show today. Theattention to detail, the care for quality, and the commitment to eval‐uation and feedback is significant. However, it can inadvertently createa certain suffocating or perhaps inhibiting barrier for many lookingto learn and develop their capabilities.

A key observation from my training courses has been the sense thatwe in the field could be accused, at times, of a certain amount of designsnobbery. We criticise and lambast many of the popular but “trashy”infographics, and believe them to be an inferior practice. However,during training sessions, I invite delegates to assess a variety of differ‐ent types of visualization design, including such infographic pieces. Ioften hear comments that express and reason a preference or even a“like” for pieces that I would not. This has proven to be a highly illu‐minating experience.

A consequence of associating with or belonging to this top-tier “bub‐ble” is that you can become somewhat detached and even oblivious tothe opinions of those who might be considered to exist in the realworld. These are the casual enthusiasts, the everyday people I men‐tioned early. They are likely to be beginning their journey into the fieldor have been nibbling around the edges for a while, but probably nevertoo seriously until now. In contrast to the hardcore connoisseurs, thislower-expertise but more highly populated tier of the field’s pyramidof participants makes up a totally different demographic and psycho‐graphic.

These people provide a great tapestry of different opinions, back‐grounds, and capabilities and, generally, they offer a more sympathetic,fresh, and open-minded view on visualization design. Without theburden of knowledge, theories, and principles that the rest of us carryaround with us all the time, and by not living and breathing the subjectacross every waking hour, their appreciation of visualization is morerooted in taste and instinct and fueled by a fresh enthusiasm to con‐sume information in visual form.

Beyond and beneath this middle-tier sits, well, everybody else. Theseare the purely occasional consumers and nothing more. Their dailyroles may not have anything to do with data, they possibly don’t evenknow or probably care what visualization is. Yet, they belong to the

Walking the Tightrope of Visualization Criticism | 75

almost silent but abundant cohort of people who are occasionally cu‐rious enough to look at an attractive visualization or light-weight in‐fographic. They don’t want or need to learn about the field, they justfind enough interest in having a look at some of its output.

This is the true make-up of the visualization and infographic field, andwe need to appreciate its relevance.

The Irrationality of Needs: Fast Food to Fine DiningThere is a prominent, long-established film critic in the U.K. who isgenerally considered a fair and sound judge of movies. He has deepsubject expertise and is capable of fully reasoning all his reviews withthorough analysis. Despite this, he does occasionally resort to the ri‐poste “other opinions are available, but they’re the wrong ones” whenchallenged by readers or viewers.

As with any subject’s “expert” tier, we in data visualization can findourselves being a little too closed off, perhaps believing the merit ofour views hold greater weight than other, contrary opinions from out‐side. But this is largely because we don’t always entirely appreciate thevariety of intentions and needs behind visualization designs. Further‐more, there are so many different contexts, target audiences, and for‐mats through which visual communication of data can exist.

Sometimes we’re looking to impart a data-driven communicationwhere the absolute accuracy of interpretation is vital. On other occa‐sions it might be about creating a visual representation of data to im‐pact more on an emotional level, trying to change behaviour and con‐nect with people through non-standard methods. Sometimes we areworking on subjects that are important, complex, and deep, and re‐quire a more engaging and prolonged interactive exploratory experi‐ence. By contrast, we might just be presenting some rather lightweightfacts or stats that relate to a harmless, maybe even “fun,” subject matter.

This is where a comparison with other creative territories like music,TV, movies, and food is appropriate to help illustrate how fundamen‐tally impulsive, inconsistent, and irrational our tastes can be. Ofcourse, the intention is very different with these channels of expres‐sion, but still we can relate to experiences when we sometimes prefera fast food meal or to feast on junk food snacks as opposed to sittingdown to a wholesome, home-cooked meal. We know it’s probably badfor us, we’ll probably spend more money on it, and we know we’re likelyto be hungry again in an hour, but we still do it.

76 | Chapter 5: What to Watch for in Big Data

You will typically never be too far away from running across intelligent,well-written movies or TV programs, but sometimes a trashy, loud,special-effects-laden blockbuster just does the trick. The critics mighthave told us how much we should hate them and how we should havespent our time with a more critically acclaimed work, but we don’t care;we just want some mindless escapism. You can extend this to writing.Maybe we should all be sitting down in our spare time reading Shake‐speare or Keats, enriching our minds. But most of us aren’t. I know I’mnot.

You can extend this to music, art, or really to any other creative chan‐nel. Of course, there are many other factors at play (access, time, re‐sources, peer influences, etc.), but we still instinctively seek to mixthings up on occasion and go against the grain. Being told what weshould and shouldn’t do can create as many followers as it does op‐ponents.

It’s the same with visualization. For many people, sometimes a harm‐less infographic showing some throw-away facts or stats about socialmedia, or demonstrating how to avoid getting bitten by a shark is justwhat people fancy viewing at that point in time. This explains the vastsuccess of works presented on gallery sites like visual.ly, the growth ofdesign agencies like Column Five, and the general phenomenon ofmodern-day tower infographics.

Whilst more important subjects and works from leading organisationslike the New York Times are arguably where we should be paying ourattention to learn and respond to critical issues, occasionally we justneed a release. We just want a blend of different visuals. This is thevisualization ecosystem, and we need to appreciate its value. NathanYau recently wrote an insightful comment piece about this pattern.

Extend this discussion further and consider the appeal of fun and ofaesthetic attraction to help stimulate the brain into engaging andlearning with representations of information. This has been proposedas an important attribute of design for a long time but still exists assuch a divisive issue within the data visualization field.

Whilst I recently remarked that there might be a sense that the tradi‐tional factions in the field were starting to better appreciate each other,I feel there is still more visible polarity than harmony. Indeed, arguablymore polarity than even co-existence. This is an indication that the

Walking the Tightrope of Visualization Criticism | 77

field is still evolving but needs to mature, and it is through our critiquewhere these fault-lines and opinion clashes manifest themselves. Mostof it is valuable and healthy debate, but equally, we need to make sureit remains reasoned and accessible.

Grown-up CriticismA key part of the training sessions I deliver is focused on trying toequip delegates with a more informed sense of how to evaluate a vis‐ualization piece. It urges them to attempt to understand the process,the purpose, and the parameters that have surrounded a project. Rath‐er than drawing conclusions from a superficial “taste” reaction, theyare asked to take a forensic approach to assessing the quality and ef‐fectiveness of a visualization, peeling through the layers of a visuali‐zation’s anatomy and putting themselves into the mind of the designer.

This is something we should all try to do before publishing our knee-jerk conclusions to the world. To empathise with the constraints thatmight have existed within the project, the limitations of the data, tryto imagine the brief and the influencing factors the designer had tocontend with. When we view and evaluate a piece, we are looking atsomething that has not benefited from infinite time, endless resources,and limitless capability. Could we have done better ourselves given thesame context?

On a perverse level, I feel this part of the training risks eroding the rawinnocence (without being disrespectful) that enables more casual ob‐servers to take visualizations and infographics on face value. They arenot cursed by the depth of analysis and variety of lenses through whichthey should evaluate a piece.

However, I shouldn’t worry because what always comes across fromthe delegates when we do this exercise is the very grounded, realistic,and practical appreciation of what works and doesn’t work in differentcontexts. There is a mature and pragmatic acceptance and appreciationof the type of limitations, pressures, constraints, and interferences thatmight have shaped the resulting design.

Such experiences in my training course have made me think that thoseof us in the connoisseur’s cohort are occasionally guilty of assessingvisualization pieces too harshly, too readily, and too rapidly. This wasthe essence of Stefanie Posavec’s observation. It’s not so much lookingat the glass being half empty; it’s more akin to seeing the slightestshortcoming and amplifying the importance of this perceived flaw.

78 | Chapter 5: What to Watch for in Big Data

A recent observation on Twitter from Santiago Ortiz highlights thisidea, characterising the type of critique that often exists about differentvisualization methods and approaches.

And here’s the vehicle Ortiz was referring to:

Via the Visual Dictionary.

This observation resonates with a question I have been asked on sev‐eral occasions by training course delegates. Many express a frustrationin their struggle to understand and identify what makes a perfect vis‐ualization. By extension, they admit to a difficulty in establishingclarity in their own convictions about judging what is a right way anda wrong way to approach a visualization design.

Entering the field, you begin with fundamentally no informed rea‐soning for appreciation of quality; it is a gut instinct based on the effectit has on you. Yet, through the influence of reading key articles andexposure to social media, when you see others expressing a conviction,you feel obliged to jump off the fence and hurriedly wave a flag, anyflag, of your judgment. It’s not so much a case of following the crowd,rather more about feeling a need to express an opinion as quickly andas clearly as everybody else seems to.

Here’s the truth: Developing clarity of your design conviction is diffi‐cult. If it were purely about taste, it would be easier. That’s why you canbe much more affirmative about your tastes in things like music, art,or movies. “Did it connect with you?” is a very open but fitting questionthat easily allows you to arrive at a Boolean type of response and theclarity of your judgment.

Walking the Tightrope of Visualization Criticism | 79

I recently wrote an article to discuss the visualizations I like. In thispiece, I talked much less about style, approach, subjects, technique, orprinciples, but instead focused on those visualizations that give backmore in return than you put in. That is my conviction, but it has takena long while to arrive at that level of clarity. As many others will, I’vebeen through a full discovery cycle of liking things that I now don’tlike and disliking things that I now do.

This conviction is informed by knowledge, by exposure to other dis‐ciplines and methods, and also through greater appreciation of whatit takes to craft an effective visualization solution that works for theproblem context it is responding to. Fundamentally, this is a hard dis‐cipline to do well.

Final ThoughtsThe balance, fairness, and realism of our criticism needs to improve.

The desire of those active “experts” in the field to influence widespreadeffective practice needs to be matched by a greater maturity and sen‐sitivity in the way we also evaluate the output of this creativity. More‐over, commentators and critics, myself included, need to develop asmarter appreciation of the different contexts in which these works arecreated.

A creative field, by its very nature, will have many different interpre‐tations and perspectives, and the resolution and richness of this opin‐ion is important to safeguard. Of course, promoting a more open-minded approach to evaluation doesn’t mean to say there should beno critical analysis. We also need to ensure there isn’t too much dem‐onstration of the emperor’s new clothes attitude, especially when awork looks cool or demonstrates impressive technical competence.

There is great importance in having the conviction and confidence toask the question “so what?,” to engage in constructive and mature cri‐tique (for example), and to exhibit a desire to understand and probethe intention behind all visualisation work. From this, we will all learnso much more and help create an environment that facilitates encour‐agement rather than discouragement.

This article likely contains some sweeping generalisations that manageto over-simplify things, but hopefully they help illustrate the impor‐

80 | Chapter 5: What to Watch for in Big Data

tance of lifting our heads above the noise and seeing what’s actuallygoing on, who is active in this field, what roles they are taking on, andthe value they are bringing to the whole visualization ecosystem, notjust to the top table.

Fundamentally, what we need to avoid is inadvertently creating bar‐riers to people trying to enter and develop in this field by creating theimpression that a 1% missed opportunity is more important than the99% of a design’s features that were a nailed-on success.

I know I will be making a concerted effort to achieve this balance andfairness in my own analyses.

Walking the Tightrope of Visualization Criticism | 81

CHAPTER 6

Big Data and Health Care

Solving the Wanamaker Problemfor Health CareBy Tim O’Reilly, Julie Steele, Mike Loukides and Colin Hill

The best minds of my generation are thinking about how to makepeople click ads.

— Jeff Hammerbacher early Facebook employee

Work on stuff that matters.— Tim O’Reilly

In the early days of the 20th century, department store magnate JohnWanamaker famously said, “I know that half of my advertising doesn’twork. The problem is that I don’t know which half.”

The consumer Internet revolution was fueled by a search for the an‐swer to Wanamaker’s question. Google AdWords and the pay-per-click model began the transformation of a business in which adver‐tisers paid for ad impressions into one in which they pay for results.“Cost per thousand impressions” (CPM) was outperformed by “costper click” (CPC), and a new industry was born. It’s important to un‐derstand why CPC outperformed CPM, though. Superficially, it’s be‐cause Google was able to track when a user clicked on a link, and wastherefore able to bill based on success. But billing based on successdoesn’t fundamentally change anything unless you can also change thesuccess rate, and that’s what Google was able to do. By using data tounderstand each user’s behavior, Google was able to place advertise‐ments that an individual was likely to click. They knew “which half ”of their advertising was more likely to be effective, and didn’t botherwith the rest.

Since then, data and predictive analytics have driven ever deeper in‐sight into user behavior such that companies like Google, Facebook,Twitter, and LinkedIn are fundamentally data companies. And dataisn’t just transforming the consumer Internet. It is transforming fi‐nance, design, and manufacturing — and perhaps most importantly,health care.

How is data science transforming health care? There are many waysin which health care is changing, and needs to change. We’re focusingon one particular issue: the problem Wanamaker described whentalking about his advertising. How do you make sure you’re spendingmoney effectively? Is it possible to know what will work in advance?

Too often, when doctors order a treatment, whether it’s surgery or anover-the-counter medication, they are applying a “standard of care”treatment or some variation that is based on their own intuition, ef‐fectively hoping for the best. The sad truth of medicine is that we don’talways understand the relationship between treatments and outcomes.We have studies to show that various treatments will work more oftenthan placebos; but, like Wanamaker, we know that much of our med‐icine doesn’t work for half or our patients, we just don’t know whichhalf. At least, not in advance. One of data science’s many promises isthat, if we can collect enough data about medical treatments and usethat data effectively, we’ll be able to predict more accurately whichtreatments will be effective for which patient, and which treatmentswon’t.

84 | Chapter 6: Big Data and Health Care

A better understanding of the relationship between treatments, out‐comes, and patients will have a huge impact on the practice of medicinein the United States. Health care is expensive. The U.S. spends over$2.6 trillion on health care every year, an amount that constitutes aserious fiscal burden for government, businesses, and our society as awhole. These costs include over $600 billion of unexplained variationsin treatments: treatments that cause no differences in outcomes, oreven make the patient’s condition worse. We have reached a point atwhich our need to understand treatment effectiveness has become vi‐tal — to the health care system and to the health and sustainability ofthe economy overall.

Why do we believe that data science has the potential to revolutionizehealth care? After all, the medical industry has had data for genera‐tions: clinical studies, insurance data, hospital records. But the healthcare industry is now awash in data in a way that it has never beenbefore: from biological data such as gene expression, next-generationDNA sequence data, proteomics, and metabolomics, to clinical dataand health outcomes data contained in ever more prevalent electronichealth records (EHRs) and longitudinal drug and medical claims. Wehave entered a new era in which we can work on massive datasetseffectively, combining data from clinical trials and direct observationby practicing physicians (the records generated by our $2.6 trillion ofmedical expense). When we combine data with the resources neededto work on the data, we can start asking the important questions, theWanamaker questions, about what treatments work and for whom.

The opportunities are huge: for entrepreneurs and data scientists look‐ing to put their skills to work disrupting a large market, for researcherstrying to make sense out of the flood of data they are now generating,and for existing companies (including health insurance companies,biotech, pharmaceutical, and medical device companies, hospitals andother care providers) that are looking to remake their businesses forthe coming world of outcome-based payment models.

Making Health Care More EffectiveWhat, specifically, does data allow us to do that we couldn’t do before?For the past 60 or so years of medical history, we’ve treated patients assome sort of an average. A doctor would diagnose a condition andrecommend a treatment based on what worked for most people, asreflected in large clinical studies. Over the years, we’ve become moresophisticated about what that average patient means, but that same

Solving the Wanamaker Problem for Health Care | 85

statistical approach didn’t allow for differences between patients. Atreatment was deemed effective or ineffective, safe or unsafe, based ondouble-blind studies that rarely took into account the differences be‐tween patients. With the data that’s now available, we can go muchfurther. The exceptions to this are relatively recent and have been do‐minated by cancer treatments, the first being Herceptin for breastcancer in women who over-express the Her2 receptor. With the datathat’s now available, we can go much further for a broad range of dis‐eases and interventions that are not just drugs but include surgery,disease management programs, medical devices, patient adherence,and care delivery.

For a long time, we thought that Tamoxifen was roughly 80% effectivefor breast cancer patients. But now we know much more: we know thatit’s 100% effective in 70 to 80% of the patients, and ineffective in therest. That’s not word games, because we can now use genetic markersto tell whether it’s likely to be effective or ineffective for any givenpatient, and we can tell in advance whether to treat with Tamoxifen orto try something else.

Two factors lie behind this new approach to medicine: a different wayof using data, and the availability of new kinds of data. It’s not juststating that the drug is effective on most patients, based on trials (in‐deed, 80% is an enviable success rate); it’s using artificial intelligencetechniques to divide the patients into groups and then determine thedifference between those groups. We’re not asking whether the drugis effective; we’re asking a fundamentally different question: “for whichpatients is this drug effective?” We’re asking about the patients, notjust the treatments. A drug that’s only effective on 1% of patients mightbe very valuable if we can tell who that 1% is, though it would certainlybe rejected by any traditional clinical trial.

More than that, asking questions about patients is only possible be‐cause we’re using data that wasn’t available until recently: DNA se‐quencing was only invented in the mid-1970s, and is only now cominginto its own as a medical tool. What we’ve seen with Tamoxifen is asclear a solution to the Wanamaker problem as you could ask for: wenow know when that treatment will be effective. If you can do the samething with millions of cancer patients, you will both improve outcomesand save money.

Dr. Lukas Wartman, a cancer researcher who was himself diagnosedwith terminal leukemia, was successfully treated with sunitinib, a drug

86 | Chapter 6: Big Data and Health Care

that was only approved for kidney cancer. Sequencing the genes ofboth the patient’s healthy cells and cancerous cells led to the discoveryof a protein that was out of control and encouraging the spread of thecancer. The gene responsible for manufacturing this protein could po‐tentially be inhibited by the kidney drug, although it had never beentested for this application. This unorthodox treatment was surpris‐ingly effective: Wartman is now in remission.

While this treatment was exotic and expensive, what’s important isn’tthe expense but the potential for new kinds of diagnosis. The price ofgene sequencing has been plummeting; it will be a common doctor’soffice procedure in a few years. And through Amazon and Google, youcan now “rent” a cloud-based supercomputing cluster that can solvehuge analytic problems for a few hundred dollars per hour. What isnow exotic inevitably becomes routine.

But even more important: we’re looking at a completely different ap‐proach to treatment. Rather than a treatment that works 80% of thetime, or even 100% of the time for 80% of the patients, a treatmentmight be effective for a small group. It might be entirely specific to theindividual; the next cancer patient may have a different protein that’sout of control, an entirely different genetic cause for the disease. Treat‐ments that are specific to one patient don’t exist in medicine as it’scurrently practiced; how could you ever do an FDA trial for a medi‐cation that’s only going to be used once to treat a certain kind of cancer?

Foundation Medicine is at the forefront of this new era in cancer treat‐ment. They use next-generation DNA sequencing to discover DNAsequence mutations and deletions that are currently used in standardof care treatments, as well as many other actionable mutations that aretied to drugs for other types of cancer. They are creating a patient-outcomes repository that will be the fuel for discovering the relationbetween mutations and drugs. Foundation has identified DNA muta‐tions in 50% of cancer cases for which drugs exist (information via aprivate communication), but are not currently used in the standard ofcare for the patient’s particular cancer.

The ability to do large-scale computing on genetic data gives us theability to understand the origins of disease. If we can understand whyan anti-cancer drug is effective (what specific proteins it affects), andif we can understand what genetic factors are causing the cancer tospread, then we’re able to use the tools at our disposal much moreeffectively. Rather than using imprecise treatments organized around

Solving the Wanamaker Problem for Health Care | 87

symptoms, we’ll be able to target the actual causes of disease, and de‐sign treatments tuned to the biology of the specific patient. Eventually,we’ll be able to treat 100% of the patients 100% of the time, preciselybecause we realize that each patient presents a unique problem.

Personalized treatment is just one area in which we can solve the Wa‐namaker problem with data. Hospital admissions are extremely ex‐pensive. Data can make hospital systems more efficient, and to avoidpreventable complications such as blood clots and hospital re-admissions. It can also help address the challenge of health care hot-spotting (a term coined by Atul Gawande): finding people who use aninordinate amount of health care resources. By looking at data fromhospital visits, Dr. Jeffrey Brenner of Camden, NJ, was able to deter‐mine that “just one per cent of the hundred thousand people who madeuse of Camden’s medical facilities accounted for thirty per cent of itscosts.” Furthermore, many of these people came from only two apart‐ment buildings. Designing more effective medical care for these pa‐tients was difficult; it doesn’t fit our health insurance system, the pa‐tients are often dealing with many serious medical issues (addictionand obesity are frequent complications), and have trouble trustingdoctors and social workers. It’s counter-intuitive, but spending moreon some patients now results in spending less on them when theybecome really sick. While it’s a work in progress, it looks like buildingappropriate systems to target these high-risk patients and treat thembefore they’re hospitalized will bring significant savings.

Many poor health outcomes are attributable to patients who don’t taketheir medications. Eliza, a Boston-based company started by Alexan‐dra Drane, has pioneered approaches to improve compliance throughinteractive communication with patients. Eliza improves patient drugcompliance by tracking which types of reminders work on which typesof people; it’s similar to the way companies like Google target adver‐tisements to individual consumers. By using data to analyze each pa‐tient’s behavior, Eliza can generate reminders that are more likely tobe effective. The results aren’t surprising: if patients take their medicineas prescribed, they are more likely to get better. And if they get better,they are less likely to require further, more expensive treatment. Again,we’re using data to solve Wanamaker’s problem in medicine: we’respending our resources on what’s effective, on appropriate remindersthat are mostly to get patients to take their medications.

88 | Chapter 6: Big Data and Health Care

More Data, More SourcesThe examples we’ve looked at so far have been limited to traditionalsources of medical data: hospitals, research centers, doctor’s offices,insurers. The Internet has enabled the formation of patient networksaimed at sharing data. Health social networks now are some of thelargest patient communities. As of November 2011, PatientsLikeMehas over 120,000 patients in 500 different condition groups; ACOR hasover 100,000 patients in 127 cancer support groups; 23andMe has over100,000 members in their genomic database; and diabetes health socialnetwork SugarStats has over 10,000 members. These are just the largercommunities, thousands of small communities are created around rarediseases, or even uncommon experiences with common diseases. Allof these communities are generating data that they voluntarily sharewith each other and the world.

Increasingly, what they share is not just anecdotal, but includes anarray of clinical data. For this reason, these groups are being recruitedfor large-scale crowdsourced clinical outcomes research.

Thanks to ubiquitous data networking through the mobile network,we can take several steps further. In the past two or three years, there’sbeen a flood of personal fitness devices (such as the Fitbit) for moni‐toring your personal activity. There are mobile apps for taking yourpulse, and an iPhone attachment for measuring your glucose. Therehas been talk of mobile applications that would constantly listen to apatient’s speech and detect changes that might be the precursor for astroke, or would use the accelerometer to report falls. Tanzeem Choud‐hury has developed an app called Be Well that is intended primarilyfor victims of depression, though it can be used by anyone. Be Wellmonitors the user’s sleep cycles, the amount of time they spend talking,and the amount of time they spend walking. The data is scored, andthe app makes appropriate recommendations, based both on the in‐dividual patient and data collected across all the app’s users.

Continuous monitoring of critical patients in hospitals has been nor‐mal for years; but we now have the tools to monitor patients constantly,in their home, at work, wherever they happen to be. And if this soundslike big brother, at this point most of the patients are willing. We don’twant to transform our lives into hospital experiences; far from it! Butwe can collect and use the data we constantly emit, our “data exhaust,”

Solving the Wanamaker Problem for Health Care | 89

to maintain our health, to become conscious of our behavior, and todetect oncoming conditions before they become serious. The mosteffective medical care is the medical care you avoid because you don’tneed it.

Paying for ResultsOnce we’re on the road toward more effective health care, we can lookat other ways in which Wanamaker’s problem shows up in the medicalindustry. It’s clear that we don’t want to pay for treatments that areineffective. Wanamaker wanted to know which part of his advertisingwas effective, not just to make better ads, but also so that he wouldn’thave to buy the advertisements that wouldn’t work. He wanted to payfor results, not for ad placements. Now that we’re starting to under‐stand how to make treatment effective, now that we understand thatit’s more than rolling the dice and hoping that a treatment that worksfor a typical patient will be effective for you, we can take the next step:Can we change the underlying incentives in the medical system? Canwe make the system better by paying for results, rather than paying forprocedures?

It’s shocking just how badly the incentives in our current medical sys‐tem are aligned with outcomes. If you see an orthopedist, you’re likelyto get an MRI, most likely at a facility owned by the orthopedist’spractice. On one hand, it’s good medicine to know what you’re doingbefore you operate. But how often does that MRI result in a differenttreatment? How often is the MRI required just because it’s part of theprotocol, when it’s perfectly obvious what the doctor needs to do?Many men have had PSA tests for prostate cancer; but in most cases,aggressive treatment of prostate cancer is a bigger risk than the diseaseitself. Yet the test itself is a significant profit center. Think again aboutTamoxifen, and about the pharmaceutical company that makes it. Inour current system, what does “100% effective in 80% of the patients”mean, except for a 20% loss in sales? That’s because the drug companyis paid for the treatment, not for the result; it has no financial interestin whether any individual patient gets better. (Whether a statisticallysignificant number of patients has side-effects is a different issue.) Andat the same time, bringing a new drug to market is very expensive, andmight not be worthwhile if it will only be used on the remaining 20%of the patients. And that’s assuming that one drug, not two, or 20, or200 will be required to treat the unlucky 20% effectively.

It doesn’t have to be this way.

90 | Chapter 6: Big Data and Health Care

In the U.K., Johnson & Johnson, faced with the possibility of losingreimbursements for their multiple myeloma drug Velcade, agreed torefund the money for patients who did not respond to the drug. Severalother pay-for-performance drug deals have followed since, paving theway for the ultimate transition in pharmaceutical company businessmodels in which their product is health outcomes instead of pills. Sucha transition would rely more heavily on real-world outcome data (arepatients actually getting better?), rather than controlled clinical trials,and would use molecular diagnostics to create personalized “treatmentalgorithms.” Pharmaceutical companies would also focus more ondrug compliance to ensure health outcomes were being achieved. Thiswould ultimately align the interests of drug makers with patients, theirproviders, and payors.

Similarly, rather than paying for treatments and procedures, can wepay hospitals and doctors for results? That’s what Accountable CareOrganizations (ACOs) are about. ACOs are a leap forward in businessmodel design, where the provider shoulders any financial risk. ACOsrepresent a new framing of the much maligned HMO approaches fromthe ’90s, which did not work. HMOs tried to use statistics to predictand prevent unneeded care. The ACO model, rather than controllingdoctors with what the data says they “should” do, uses data to measurehow each doctor performs. Doctors are paid for successes, not for theprocedures they administer. The main advantage that the ACO modelhas over the HMO model is how good the data is, and how that datais leveraged. The ACO model aligns incentives with outcomes: a prac‐tice that owns an MRI facility isn’t incentivized to order MRIs whenthey’re not necessary. It is incentivized to use all the data at its disposalto determine the most effective treatment for the patient, and to followthrough on that treatment with a minimum of unnecessary testing.

When we know which procedures are likely to be successful, we’ll bein a position where we can pay only for the health care that works.When we can do that, we’ve solved Wanamaker’s problem for healthcare.

Enabling DataData science is not optional in health care reform; it is the linchpin ofthe whole process. All of the examples we’ve seen, ranging from cancer

Solving the Wanamaker Problem for Health Care | 91

treatment to detecting hot spots where additional intervention willmake hospital admission unnecessary, depend on using data effec‐tively: taking advantage of new data sources and new analytics tech‐niques, in addition to the data the medical profession has had all along.

But it’s too simple just to say “we need data.” We’ve had data all along:handwritten records in manila folders on acres and acres of shelving.Insurance company records. But it’s all been locked up in silos: insur‐ance silos, hospital silos, and many, many doctor’s office silos. Datadoesn’t help if it can’t be moved, if data sources can’t be combined.

There are two big issues here. First, a surprising number of medicalrecords are still either hand-written, or in digital formats that arescarcely better than hand-written (for example, scanned images ofhand-written records). Getting medical records into a format that’scomputable is a prerequisite for almost any kind of progress. Second,we need to break down those silos.

Anyone who has worked with data knows that, in any problem, 90%of the work is getting the data in a form in which it can be used; theanalysis itself is often simple. We need electronic health records: pa‐tient data in a more-or-less standard form that can be shared effi‐ciently, data that can be moved from one location to another at thespeed of the Internet. Not all data formats are created equal, and someare certainly better than others: but at this point, any machine-readableformat, even simple text files, is better than nothing. While there arecurrently hundreds of different formats for electronic health records,the fact that they’re electronic means that they can be converted fromone form into another. Standardizing on a single format would makethings much easier, but just getting the data into some electronic form,any, is the first step.

Once we have electronic health records, we can link doctor’s offices,labs, hospitals, and insurers into a data network, so that all patient datais immediately stored in a data center: every prescription, every pro‐cedure, and whether that treatment was effective or not. This isn’t somefuturistic dream; it’s technology we have now. Building this networkwould be substantially simpler and cheaper than building the networksand data centers now operated by Google, Facebook, Amazon, Apple,and many other large technology companies. It’s not even close topushing the limits.

Electronic health records enable us to go far beyond the current mech‐anism of clinical trials. In the past, once a drug has been approved in

92 | Chapter 6: Big Data and Health Care

trials, that’s effectively the end of the story: running more tests to de‐termine whether it’s effective in practice would be a huge expense. Aphysician might get a sense for whether any treatment worked, butthat evidence is essentially anecdotal: it’s easy to believe that somethingis effective because that’s what you want to see. And if it’s shared withother doctors, it’s shared while chatting at a medical convention. Butwith electronic health records, it’s possible (and not even terribly ex‐pensive) to collect documentation from thousands of physicians treat‐ing millions of patients. We can find out when and where a drug wasprescribed, why, and whether there was a good outcome. We can askquestions that are never part of clinical trials: is the medication usedin combination with anything else? What other conditions is the pa‐tient being treated for? We can use machine learning techniques todiscover unexpected combinations of drugs that work well together,or to predict adverse reactions. We’re no longer limited by clinical tri‐als; every patient can be part of an ongoing evaluation of whether histreatment is effective, and under what conditions. Technically, this isn’thard. The only difficult part is getting the data to move, getting datain a form where it’s easily transferred from the doctor’s office to ana‐lytics centers.

To solve problems of hot-spotting (individual patients or groups ofpatients consuming inordinate medical resources) requires a differentcombination of information. You can’t locate hot spots if you don’thave physical addresses. Physical addresses can be geocoded (con‐verted from addresses to longitude and latitude, which is more usefulfor mapping problems) easily enough, once you have them, but youneed access to patient records from all the hospitals operating in thearea under study. And you need access to insurance records to deter‐mine how much health care patients are requiring, and to evaluatewhether special interventions for these patients are effective. Not onlydoes this require electronic records, it requires cooperation across dif‐ferent organizations (breaking down silos), and assurance that the datawon’t be misused (patient privacy). Again, the enabling factor is ourability to combine data from different sources; once you have the data,the solutions come easily.

Breaking down silos has a lot to do with aligning incentives. Currently,hospitals are trying to optimize their income from medical treatments,while insurance companies are trying to optimize their income byminimizing payments, and doctors are just trying to keep their headsabove water. There’s little incentive to cooperate. But as financial pres‐

Solving the Wanamaker Problem for Health Care | 93

sures rise, it will become critically important for everyone in the healthcare system, from the patient to the insurance executive, to assumethat they are getting the most for their money. While there’s intensecultural resistance to be overcome (through our experience in datascience, we’ve learned that it’s often difficult to break down silos withinan organization, let alone between organizations), the pressure of de‐livering more effective health care for less money will eventually breakthe silos down. The old zero-sum game of winners and losers mustend if we’re going to have a medical system that’s effective over thecoming decades.

Data becomes infinitely more powerful when you can mix data fromdifferent sources: many doctor’s offices, hospital admission records,address databases, and even the rapidly increasing stream of datacoming from personal fitness devices. The challenge isn’t employingour statistics more carefully, precisely, or guardedly. It’s about lettinggo of an old paradigm that starts by assuming only certain variablesare key and ends by correlating only these variables. This paradigmworked well when data was scarce, but if you think about, these as‐sumptions arise precisely because data is scarce. We didn’t study therelationship between leukemia and kidney cancers because that wouldrequire asking a huge set of questions that would require collecting alot of data; and a connection between leukemia and kidney cancer isno more likely than a connection between leukemia and flu. But theexistence of data is no longer a problem: we’re collecting the data allthe time. Electronic health records let us move the data around so thatwe can assemble a collection of cases that goes far beyond a particularpractice, a particular hospital, a particular study. So now, we can usemachine learning techniques to identify and test all possible hypoth‐eses, rather than just the small set that intuition might suggest. Andfinally, with enough data, we can get beyond correlation to causation:rather than saying “A and B are correlated,” we’ll be able to say “A causesB,” and know what to do about it.

Building the Health Care System We WantThe U.S. ranks 37th out of developed economies in life expectancy andother measures of health, while by far outspending other countries onper-capita health care costs. We spend 18% of GDP on health care,while other countries on average spend on the order of 10% of GDP.We spend a lot of money on treatments that don’t work, because wehave a poor understanding at best of what will and won’t work.

94 | Chapter 6: Big Data and Health Care

Part of the problem is cultural. In a country where even pets can havehip replacement surgery, it’s hard to imagine not spending every pennyyou have to prolong Grandma’s life — or your own. The U.S. is a weal‐thy nation, and health care is something we choose to spend our moneyon. But wealthy or not, nobody wants ineffective treatments. Nobodywants to roll the dice and hope that their biology is similar enough toa hypothetical “average” patient. No one wants a “winner take all”payment system in which the patient is always the loser, paying forprocedures whether or not they are helpful or necessary. Like Wana‐maker with his advertisements, we want to know what works, and wewant to pay for what works. We want a smarter system where treat‐ments are designed to be effective on our individual biologies; wheretreatments are administered effectively; where our hospitals our usedeffectively; and where we pay for outcomes, not for procedures.

We’re on the verge of that new system now. We don’t have it yet, butwe can see it around the corner. Ultra-cheap DNA sequencing in thedoctor’s office, massive inexpensive computing power, the availabilityof EHRs to study whether treatments are effective even after the FDAtrials are over, and improved techniques for analyzing data are the toolsthat will bring this new system about. The tools are here now; it’s upto us to put them into use.

Recommended ReadingWe recommend the following articles and books regarding technology,data, and health care reform:

• Ahier, Brian. “Big data is the next big thing in health IT,” O’ReillyRadar. February 27, 2012.

• Bigelow, Bruce. “Big Data, Big Biology, and the ‘Tipping Point’ inQuantified Health,” Xconomy. April 26, 2012.

• Brawley, Otis Webb. How We Do Harm: A Doctor Breaks RanksAbout Being Sick in America. St. Marten’s Press, 2012.

• Christensen, Clayton M. et al. The Innovator’s Prescription: A Dis‐ruptive Solution for Health Care. McGraw Hill, 2008.

• Howard, Alex. “Data for the Public Good,” O’Reilly Radar. Feb‐ruary 22, 2012.

• Manyika, James et al. “Big data: The next frontier for innovation,competition, and productivity,” McKinsey Global Institute. May,2011.

Solving the Wanamaker Problem for Health Care | 95

• Oram, Andy. “Five tough lessons I had to learn about healthcare,” O’Reilly Radar. March 26, 2012.

• Shah, Nigam H and Jessica D Tenenbaum. “The coming age ofdata-driven medicine: translational bioinformatics’ next frontier,”Journal of the American Medical Informatics Association (JAMIA).March 26, 2012.

• Trotter, Fred and David Uhlman. Meaningful Use and Beyond.O’Reilly Media, 2011.

• Wilbanks, John. “Valuing Health Care: Improving Productivityand Quality” [PDF], Ewing Marion Kauffman Foundation. April,2012.

Dr. Farzad Mostashari on Building the HealthInformation Infrastructure for the ModernePatientBy Alex Howard

To learn more about what levers the government is pulling to catalyzeinnovation in the healthcare system, I turned to Dr. Farzad Mostashari(@Farzad_ONC). As the National Coordinator for Health IT, Mosta‐shari is one of the most important public officials entrusted with im‐proving the nation’s healthcare system through smarter use of tech‐nology.

Mostashari, a public-health informatics specialist, was named ONCchief in April 2011, replacing Dr. David Blumenthal. Mostashari’s fullbiography, available at HHS.gov, notes that he “was one of the leadinvestigators in the outbreaks of West Nile Virus and anthrax in NewYork City, and was among the first developers of real-time electronicdisease surveillance systems nationwide.”

I talked to Mostashari on the same day that he published a look backover 2011, which he hailed as a year of momentous progress in healthinformation technology. Our interview follows.

What excites you about your work? What trends matter here?

96 | Chapter 6: Big Data and Health Care

Farzad Mostashari: Well, it’s a really fun job. It feels like this is the idealtime for this health IT revolution to tie into other massive megatrendsthat are happening around consumer and patient empowerment, pay‐ment and delivery reform, as I talked about in my TED Med Talk withAneesh Chopra.

These three streams [how patients are cared for, how care is paid for,and how people take care of their own health] coming together feelsgreat. And it really feels like we’re making amazing progress.

How does what’s happening today grow out of the passage of theHealth Information Technology for Economic and Clinical HealthAct (HITECH) Act in 2009?

Farzad Mostashari: HITECH was a key part of ARRA, the AmericanRecovery and Reinvestment Act. This is the reinvestment part. Peoplethink of roadways and runways and railways. This is the informationinfrastructure for healthcare.

In the past two years, we made as much progress on adoption as wehad made in the past 20 years before that. We doubled the adoption ofelectronic health records in physician offices between the time thestimulus passed and now. What that says is that a large number ofbarriers have been addressed, including the financial barriers that areaddressed by the health IT incentive payments.

It also, I think, points to the innovation that’s happening in the healthIT marketplace, with more products that people want to buy and wantto use, and an explosion in the number of options people have.

The programs we put in place, like the Regional Health IT ExtensionCenters modeled after the Agriculture Extension program, give ahelping hand. There are local nonprofits throughout the country thatare working with one-third of all primary care providers in this coun‐try to help them adopt electronic health records, particularly smallerpractices and maybe health centers, critical access hospitals, and soforth.

This is obviously a big lift and a big change for medicine. It moves atwhat Jay Walker called “med speed,” not tech speed. The pace of trans‐formation in medicine that’s happening right now may be unparal‐leled. It’s a good thing.

Dr. Farzad Mostashari on Building the Health Information Infrastructure for the Modern ePatient| 97

Healthcare providers have a number of options as they adopt elec‐tronic health records. How do you think about the choice betweenopen source versus proprietary options?

Farzad Mostashari: We’re pretty agnostic in terms of the technologyand the business model. What matters are the outcomes. We’ve reallyleft the decisions about what technology to use to the people who haveto live with it, like the doctors and hospitals who make the purchases.

There are definitely some very successful models, not only on the EHRside, but also on the health information exchange side.

(Note: For more on this subject, read Brian Ahier’s Radar post on theHealth Internet.)

What role do open standards play in the future of healthcare?

Farzad Mostashari: We are passionate believers in open standards. Wethink that everybody should be using them. We’ve gotten really greatparticipation by vendors of open source and proprietary software, interms of participating in an open standards development process.

I think what we’ve enabled, through things like modular certification,is a lot more innovation. Different pieces of the entire ecosystem couldbe done through reducing the barrier to entry, enabling a variety ofdifferent innovative startups to come to the field. What we’re seeing is,a lot of the time, this is migrating from installed software to web serv‐ices.

If we’re setting up a reference implementation of the standards, likethe Connect software or popHealth, we do it through a process wherethe result is open source. I think the government as a platform ap‐proach at the Veterans Affairs department, DoD, and so forth is tre‐mendously important.

How is the mobile revolution changing healthcare?

We had Jay Walker talking about big change [at a recent ONC GranteeMeeting]. I just have this indelible image of him waving in his left handa clay cone with cuneiform on it that is from 2,000 B.C. — 4,000 yearsago — and in his right hand he held his iPhone.

He was saying both of them represented the cutting edge of technologythat evolved to meet consumer need. His strong assertion was that thisis absolutely going to revolutionize what happens in medicine at techspeed. Again, not “med speed.”

98 | Chapter 6: Big Data and Health Care

I had the experience of being at my clinic, where I get care, and thepharmacist sitting in the starched, white coat behind the counter tell‐ing me that I should take this medicine at night.

And I said, “Well, it’s easier for me to take it in the morning.” And hesaid, “Well, it works better at night.”

And I asked, acting as an empowered patient, “Well, what’s the halflife?” And he answered, “Okay. Let me look it up.”

He started clacking away at his pharmacy information system; clickityclack, clickity clack. I can’t see what he’s doing. And then he says, “Ahhell,” and he pulls out his smartphone and Googles it.

There’s now a democratization of information and information tools,where we’re pushing the analytics to the cloud. Being able to put thatin the hand of not just every doctor or every healthcare provider butevery patient is absolutely going to be that third strand of the DNA,putting us on the right path for getting healthcare that results in health.

We’re making sure that people know they have a right to get their owndata, making sure that the policies are aligned with that. We’re makingsure that we make it easy for doctors to give patients their own infor‐mation through things like the Direct Project, the Blue Button, mean‐ingful use requirements, or the Consumer E-Health Pledge.

We have more than 250 organizations that collectively hold data for100 million Americans that pledge to make it easy for people to getelectronic copies of their own data.

Do you think people will take ownership of their personal healthdata and engage in what Susannah Fox has described as “peer-to-peer healthcare”?

Farzad Mostashari: I think that it will be not just possible, not evenjust okay, but actually encouraged for patients to be engaged in theircare as partners. Let the epatient help. I think we’re going to see thatemerging as there’s more access and more tools for people to do stuffwith their data once they get it through things like the health datainitiative. We’re also beginning to work with stakeholder groups, likeConsumer’s Union, the American Nurses Association, and some of thedisease groups, to change attitudes around it being okay to ask for yourown records.

This interview was edited and condensed.

Dr. Farzad Mostashari on Building the Health Information Infrastructure for the Modern ePatient| 99

John Wilbanks Discusses the Risks andRewards of a Health Data CommonsBy Alex Howard

As I wrote earlier this year in an ebook on data for the public good,while the idea of data as a currency is still in its infancy, it’s importantto think about where the future is taking us and our personal data.

If the Obama administration’s smart disclosure initiatives gathersteam, more citizens will be able to do more than think about personaldata: they’ll be able to access their financial, health, education, or en‐ergy data. In the U.S. federal government, the Blue Button initiative,which initially enabled veterans to download personal health data, isnow spreading to all federal employees, and it also earned adoption atprivate institutions like Aetna and Kaiser Permanente. Putting healthdata to work stands to benefit hundreds of millions of people. TheLocker Project, which provides people with the ability to move andstore personal data, is another approach to watch.

The promise of more access to personal data, however, is balanced byaccompanying risks. Smartphones, tablets, and flash drives, after all,are lost or stolen every day. Given the potential of mhealth, and bigdata and health care information technology, researchers and policymakers alike are moving forward with their applications. As they doso, conversations and rulemaking about health care privacy will needto take into account not just data collection or retention but contextand use.

Put simply, businesses must confront the ethical issues tied to massiveaggregation and data analysis. Given that context, Fred Trotter’s poston who owns health data is a crucial read. As Fred highlights, the realissue is not ownership, per se, but “What rights do patients have re‐garding health care data that refers to them?”

Would, for instance, those rights include the ability to donate personaldata to a data commons, much in the same way organs are donatednow for research? That question isn’t exactly hypothetical, as the fol‐lowing interview with John Wilbanks highlights.

Wilbanks, a senior fellow at the Kauffman Foundation and director ofthe Consent to Research Project, has been an advocate for open data

100 | Chapter 6: Big Data and Health Care

and open access for years, including a stint at Creative Commons; afellowship at the World Wide Web Consortium; and experience in theacademic, business, and legislative worlds. Wilbanks will be speakingat the Strata Rx Conference in October.

Our interview, lightly edited for content and clarity, follows.

Where did you start your career? Where has it taken you?

John Wilbanks: I got into all of this, in many ways, because I studiedphilosophy 20 years ago. What I studied inside of philosophy was se‐mantics. In the ’90s, that was actually sort of pointless because therewasn’t much semantic stuff happening computationally.

In the late ’90s, I started playing around with biotech data, mainlybecause I was dating a biologist. I was sort of shocked at how the datawas being represented. It wasn’t being represented in a way that wasvery semantic, in my opinion. I started a software company and weran that for a while, [and then] sold it during the crash.

I went to the World Wide Web Consortium, where I spent a year help‐ing start their Semantic Web for Life Sciences project. While I wasthere, Creative Commons (CC) asked me to come and start their sci‐ence project because I had known a lot of those guys. When I startedmy company, I was at the Berkman Center at Harvard Law School,and that’s where Creative Commons emerged from, so I knew thepeople. I knew the policy and I had gone off and had this bioinfor‐matics software adventure.

I spent most of the last eight years at CC working on trying to builddifferent commons in science. We looked at open access to scientificliterature, which is probably where we had the most success becausethat’s copyright-centric. We looked at patents. We looked at physicallaboratory materials, like stem cells in mice. We looked at differentlegal regimes to share those things. And we looked at data. We lookedat both the technology aspects and legal aspects of sharing data andmaking it useful.

A couple of times over those years, we almost pivoted from science tohealth because science is so institutional that it’s really hard for any ofthe individual players to create sharing systems. It’s not like software,where anyone with a PC and an Internet connection can contribute tofree software, or Flickr, where anybody with a digital camera can li‐cense something under CC. Most scientists are actually restricted bytheir institutions. They can’t share, even if they want to.

John Wilbanks Discusses the Risks and Rewards of a Health Data Commons | 101

Health kept being interesting because it was the individual patientswho had a motivation to actually create something different than thesystem did. At the same time, we were watching and seeing the capacityof individuals to capture data about themselves exploding. So, at thesame time that the capacity of the system to capture data about youexploded, your own capacity to capture data exploded.

That, to me, started taking on some of the interesting contours thatmake Creative Commons successful, which was that you didn’t needa large number of people. You didn’t need a very large percentage ofWikipedia users to create Wikipedia. You didn’t need a large percentageof free software users to create free software. If this capacity to generatedata about your health was exploding, you didn’t need a very largepercentage of those people to create an awesome data resource: youneeded to create the legal and technical systems for the people whodid choose to share to make that sharing useful.

Since Creative Commons is really a copyright-centric organization, Ileft because the power on which you’re going to build a commons ofhealth data is going to be privacy power, not copyright power. What Ido now is work on informed consent, which is the legal system youneed to work with instead of copyright licenses, as well as the tech‐nologies that then store, clean, and forward user-generated data tocomputational health and computational disease research.

What are the major barriers to people being able to donate their datain the same way they might donate their organs?

John Wilbanks: Right now, it looks an awful lot like getting onto theInternet before there was the Web. The big ISPs kind of dominated theearly adopters of computer technologies. You had AOL. You had Com‐puServe. You had Prodigy. And they didn’t communicate with eachother. You couldn’t send email from AOL to CompuServe.

What you have now depends on the kind of data. If the data that in‐terests you is your genotype, you’re probably a 23andMe customer andyou’ve got a bunch of your data at 23andMe. If you are the kind ofperson who has a chronic illness and likes to share information aboutthat illness, you’re probably a customer at PatientsLikeMe. But thosetwo systems don’t interoperate. You can’t send data from one to theother very effectively or really at all.

102 | Chapter 6: Big Data and Health Care

On top of that, the system has data about you. Your insurance companyhas your billing records. Your physician has your medical records. Yourpharmacy has your pharmacy records. And if you do quantified self,you’ve got your own set of data streams. You’ve got your Fitbit, the datacoming off of your smartphone, and your meal data.

Almost all of these are basically populating different silos. In somecases, you have the right to download certain pieces of the data. Forthe most part, you don’t. It’s really hard for you, as an individual, tobuild your own, multidimensional picture of your data, whereas it’sactually fairly easy for all of those companies to sell your data to oneanother. There’s not a lot of technology that lets you share.

What are some of the early signals we’re seeing about data usagemoving into actual regulatory language?

John Wilbanks: The regulatory language actually makes it fairly hardto do contextual privacy waiving, in a Creative Commons sense. It’shard to do granular permissions around privacy in the way you cando granular conditional copyright grants because you don’t have in‐tellectual property. The only legal tool you have is a contract, and thecontracts don’t have a lot of teeth.

It’s pretty hard to do anything beyond a gift. It’s more like organ don‐ation, where you don’t get to decide where the organs go. What I’mworking on is basically a donation, not a conditional gift. The regula‐tory environment makes it quite hard to do anything besides that.

There was a public comment period that just finished. It’s an an‐nouncement of proposed rulemaking on what’s called the CommonRule, which is the Department of Health and Human Services privacylanguage. It was looking to re-examine the rules around letting de-identified data or anonymized data out for widespread use. They gota bunch of comments.

There’s controversy as to how de-identified data can actually be andstill be useful. There is going to be, probably, a three-to-five year pro‐cess where they rewrite the Common Rule and it’ll be more modern.No one knows how modern, but it will be at least more modern whenthat finishes.

Then there’s another piece in the U.S. — HIPAA — which creates atotally separate regime. In some ways, it is the same as the Common

John Wilbanks Discusses the Risks and Rewards of a Health Data Commons | 103

Rule, but not always. I don’t think that’s going to get opened up. Theway HIPAA works is that they have 17 direct identifiers that are labeledas identifying information. If you strip those out, it’s considered de-identified.

There’s an 18th bucket, which is anything else that can reasonablyidentify people. It’s really hard to hit. Right now, your genome is notconsidered to fall under that. I would be willing to bet within a yearor two, it will be.

From a regulatory perspective, you’ve got these overlapping regimesthat don’t quite fit and both of them are moving targets. That createsa lot of uncertainty from an investment perspective or from an ana‐lytics perspective.

How are you thinking about a “health data commons,” in terms ofweighing potential risks against potential social good?

John Wilbanks: I think that that’s a personal judgment as to the risk-benefit decision. Part of the difficulty is that the regulations are verysyntactic — “This is what re-identification is” — whereas the conceptof harm, benefit, or risk is actually something that’s deeply personal.If you are sick, if you have cancer or a rare disease, you have a verydifferent idea of what risk is compared to somebody who thinks of himor herself as healthy.

What we see — and this is born out in the Framingham Heart Studyand all sorts of other longitudinal surveys — is that people’s attitudestoward risk and benefit change depending on their circumstances.Their own context really affects what they think is risky and what theythink isn’t risky.

I believe that the early data donors are likely to be people for whomthere isn’t a lot of risk perceived because the health system alreadyknows that they’re sick. The health system is already denying themcoverage, denying their requests for PET scans, denying their requestsfor access to care. That’s based on actuarial tables, not on their personaldata. It’s based on their medical history.

If you’re in that group of people, then the perceived risk is actuallypretty low compared to the idea that your data might actually get usedor to the idea that you’re no longer passive. Even if it’s just a donation,you’re doing something outside of the system that’s accelerating theodds of getting something discovered. I think that’s the natural group.

104 | Chapter 6: Big Data and Health Care

If you think back to the numbers of users who are required to createfree software or Wikipedia, to create a cultural commons, a very lowpercentage is needed to create a useful resource.

Depending on who you talk to, somewhere between 5-10% of allAmericans either have a rare disease, have it in their first order family,or have a friend with a rare disease. Each individual disease might nothave very many people suffering from it, but if you net them all up, it’sa lot of people. Getting several hundred thousand to a few millionpeople enrolled is not an outrageous idea.

When you look at the existing examples of where such commonshave come together, what have been the most important concretepositive outcomes for society?

John Wilbanks: I don’t think we have really even started to see thembecause most people don’t have computable data about themselves.Most people, if they have any data about themselves, have scans oftheir medical records.

What we really know is that there’s an opportunity cost to not trying,which is that the existing system is really inefficient, very bad at dis‐covering drugs, and very bad at getting those drugs to market in atimely basis.

That’s one of the reasons we’re doing this is as an experiment. We wouldlike to see exactly how effective big computational approaches are onhealth data. The problem is that there are two ways to get there.

One is through a set of monopoly companies coming together andworking together. That’s how semiconductors work. The other isthrough an open network approach. There’s not a lot of evidence thatthings besides these two approaches work. Government interventionis probably not going to work.

Obviously, I come down on the open network side. But there’s an im‐plicit belief, I think, both in the people who are pushing the cooper‐ating monopolies approach and the people who are pushing the opennetworks approach, that there’s enormous power in the big-data-driven approach. We’re just leaving that on the table right now by nothaving enough data aggregated.

The benefits to health that will come out will be the ability to increas‐ingly, by looking at a multidimensional picture of a person, predict

John Wilbanks Discusses the Risks and Rewards of a Health Data Commons | 105

with some confidence whether or not a drug will work, or whetherthey’re going to get sick, or how sick they’re going to get, or what life‐style changes they can make to mitigate an illness. Right now, basically,we really don’t know very much.

Esther Dyson on Health Data, “PreemptiveHealthcare,” and the Next Big ThingBy Alex Howard

If we look ahead to the next decade, it’s worth wondering whether theway we think about health and health care will have shifted. Will healthcare technology be a panacea? Will it drive even higher costs, creatinga broader divide between digital haves and have-nots? Will openinghealth data empower patients or empower companies?

As ever, there will be good outcomes and bad outcomes, and not justin the medical sense. There’s a great deal of thought around the po‐tential for mobile applications right now, from the FDA’s potential de‐cision to regulate them to a reported high abandonment rate. Thereare also significant questions about privacy, patient empowerment,and meaningful use of electronic health care records.

When I’ve talked to US CTO Todd Park or Dr. Farzad Mostasharithey’ve been excited about the prospect for health data to fuel betterdashboards and algorithms to give frontline caregivers access to crit‐ical information about people they’re looking after, providing criticalinsight at the point of contact.

Kathleen Sebelius, the U.S. Secretary for Health and Human Services,said at this year’s Health Datapalooza that venture capital investmentin the health care IT area is up 60% since 2009.

Given that context, I was more than a little curious to hear what EstherDyson (@edyson) is thinking about when she looks at the intersectionof health care, data, and information technology.

Dyson, who started her career as a journalist, is now an angel investorand philanthropist. Dyson is a strong supporter of “preemptive healthcare” — and she’s putting her money where her interest lies, with herinvestments.

Our interview, which was lightly edited for content and clarity, follows.

How do you see health care changing?

106 | Chapter 6: Big Data and Health Care

Dyson: There are multiple perspectives. The one I have does not in‐validate others, nor it is intended to trump the others, but it’s the onethat I focus on — and that’s “health” as opposed to “health care.”

If you maintain good health, you can avoid health care. That’s one ofthose great and unrealizable goals, but it’s realizable in part. Any healthcare you can avoid because you’re healthy is valuable.

What I’m mostly focused on is trying to change people’s behavior.You’ll get agreement from almost everybody that eating right, notsmoking, getting exercise, avoiding too much stress, and sleeping a lotare good for your health.

The challenge is what makes people do those things, and that’s wherethere’s a real lack of data. So a lot of what I’m doing is investing in thatspace. There’s evidence-based medicine. There’s also evidence-basedprevention, and that’s even harder to validate.

Right now, a lot of people are doing a lot of different things. Many ofthem are collecting data, which over time, with luck, will prove thatsome of these things I’m going to talk about are valuable.

What does the landscape for health care products and services looklike to you today?

Dyson: I see three markets.

There’s the traditional health care market, which is what people usu‐ally talk about. It’s drugs, clinics, hospitals, doctors, therapies, devices,insurance companies, data processors, or electronic health records.

Then there’s the market for bad health, which people don’t talk abouta lot, at least not in those terms, but it’s huge. It’s the products and allof the advertising around everything from sugared soft drinks to cig‐arettes to recreational drugs to things that keep you from going to bed,going to sleep, keep you on the couch, and keep you immobile. Lookat cigarettes and alcohol: That’s a huge market. People are being en‐couraged to engage in unhealthy behaviors, whether it’s stuff thatmight be healthy in moderation or stuff that just isn’t healthy at all.

The new [third] market for health existed already as health clubs.What’s exciting is that there’s now an explicit market for things thatare designed to change your behavior. Usually, they’re information-and social-based. These are the quantified self — analytical tools, toolsfor sharing, tools for fostering collaboration or competition with peo‐ple that behave in a healthy way. Most of those have very little data to

Esther Dyson on Health Data, “Preemptive Healthcare,” and the Next Big Thing | 107

back them up. The business models are still not too clear, because ifI’m healthy, who’s going to pay for that? The chances are that if I’ll payfor it, I’m already kind of a health nut and don’t need it as much assomeone who isn’t.

Pharma companies will pay for some such things, especially if theythink they can sell people drugs in conjunction with them. I’ll sell youa cholesterol-lowering drug through a service that encourages you toexercise, for example. That’s a nice market. You go to the pre-diabeticsand you sell them your statin. Various vendors of sports clubs and soforth will fund this. But over time, I expect you’re going to see em‐ployers realize the value of this, then finally, long-term insurance com‐panies and perhaps government. But it’s a market that operates mostlyon faith at this point.

Speaking of faith, Rock Health shared data that around 80% of mo‐bile health apps are being abandoned by consumers after two weeks.Thoughts?

Dyson: To me, that’s infant mortality. The challenge is to take the 20%and then make those persist. But you’re right, people try a lot of stuffand it turns out to be confusing and not well-designed, et cetera.

If you look ahead a decade, what are the big barriers for health dataand mobile technology playing a beneficial role, as opposed to a moredystopian one?

Dyson: Well, the benign version is we’ve done a lot of experimentation.We’ve discovered that most apps have an 80% abandon rate, but the20% that are persisting get better and better and better. So the 80% thatare abandoned vanish and the marketplace and the vendors focus onthe 20%. And we get broad adoption. You get onto the subway in NewYork and everybody’s thin and healthy.

Yeah, that’s not going to happen. But there’s some impact. Employersunderstand the value of this. There’s a lot more to do than just these[mobile] apps. The employers start serving only healthy food in thecafeteria. Actually, one big sign is going to be what they serve forbreakfast at Strata RX. I was at the Kauffman Life Sciences Entrepre‐neur Conference and they had muffins, bagels, and cream cheese.

Carbohydrates and fat, in other words.

108 | Chapter 6: Big Data and Health Care

Dyson: And sugar-filled yogurts. That was the first day. They respon‐ded to somebody’s tweet [the second day] and it was better. But it’s notjust the advertising. It’s the selection of stuff that you get when you goto these events or when you go to a hotel or you go to school or yougo to your cafeteria at your office.

Defaults are tremendously important. That’s why I’m a big fan of what[Michael] Bloomberg is trying to do in New York. If you really wantto buy two servings of soda, that’s fine, but the default serving shouldbe one. All of this stuff really does have an impact.

Ten years from now, evidence has shown what works. What works isworking because people are doing it. A lot of this is that social normshave changed. The early adopters have adopted, the late adopters arebeing carried along in the wake — just like there are still people whosmoke, but it’s no longer the norm.

Do you have concerns or hopes for the risks and rewards of openhealth data releases?

Dyson: If we have a sensible health care system, the data will be helpful.Hospitals will say, “Oh my God, this guy’s at-risk, let’s prevent himfrom getting sick.” Hospitals and the payers will know, “If we let thisguy get sick, it’s going to cost us a lot more in the long run. And weactually have a business model that operates long-term rather thansimply tries to minimize cost in the short-term.”

And insurance companies will say, “I’m paying for this guy. I betterkeep him healthy.” So the most important thing is for us to have asystem that works long-term like that.

What role will personal data ownership play in the health care systemof the future?

Dyson: Well, first we have to define what it is. From my point-of-view,you own your own data. On the other hand, if you want care, you’vegot to share it.

I think people are way too paranoid about their data. There will, in‐evitably, be data spills. We should try to avoid them, but we should alsonot encourage paranoia. If you have a rational economic system, pri‐vacy will be an issue, but financial security will not. Those two havegotten mingled in people’s minds.

Yes, I may just want to keep it quiet that I have a sexually transmitteddisease, but it’s not going to affect my ability to get treatment or to get

Esther Dyson on Health Data, “Preemptive Healthcare,” and the Next Big Thing | 109

1. Dyson is an investor in 23andMe.

insurance if I’ve got it. On the other hand, if I have to pay a little morefor my diet soda or my hamburger because it’s being taxed, I don’t thinkthat’s such a bad idea. Not that I want somebody recording how manyhamburgers I eat, just tax them — but you don’t need to tax me per‐sonally: tax the hamburger.

What about the potential for the quantified self-movement to some‐day reveal that hamburger consumption to insurers?

Dyson: People are paranoid about insurers, but they’re too busy.They’re not tracking the hamburgers you eat. They’re insuring popu‐lations. I went to get insurance and I told Aetna, “You can have mygenetic profile.” And they said, “We wouldn’t know what to do with it.”I’m not saying that [tracking is] entirely impossible, but I really thinkpeople obsess too much about this kind of stuff.

How should — or could — startups in health care be differentiatingthemselves? What are the big problems they could be working onsolving?

Dyson: There’s the whole social aspect. How do you design a game, asocial interaction, that encourages people to react the way you wantthem to react? It’s like the difference between Facebook and Friendster.They both had the same potential user base. One was successful; onewasn’t. It’s the quality of the analytics you show individuals about theirbehavior. It’s the narratives, the tools and the affordances that you givethem for interacting with their friends.

For what it’s worth, of the hundreds of companies that Rock Health oranybody else will tell you about, probably a third of them will disap‐pear. One tenth will be highly successful and will acquire the remaining57%.

What are the health care startup models that interest you? Why?

Dyson: I don’t think there’s a single one. There’s bunches of them oc‐cupying different places.

One area I really like is user-generated research and experiments. Ob‐viously, there’s 23andMe.1 Deep analysis of your own data and theoption to share it with other people and with researchers. User-generated data science research is really fascinating.

110 | Chapter 6: Big Data and Health Care

And then social affordance, like HealthRally, where people interactwith each other. Omada Health — which I’m an investor in — is a RockHealth company that says we can’t do it all ourselves — there’s a des‐ignated counselor for a group. Right now it’s focused on pre-diabetics.

I love that, partly because I think it’s going to be effective, and partlybecause I really like it as an employment model. I think our countryis too focused on manufacturing and there’s a way to turn more peopleinto health counselors. I’d take all of the laid off auto workers and turnthem into gym teachers, and all the laid off engineers and turn theminto data scientists or people developing health apps. Or somethinglike that.

What’s the biggest myth in the health data world? What’s the thingthat drives you up the wall, so to speak?

Dyson: The biggest myth is that any single thing is the solution. Thebiggest need is for long-term thinking, which is everything from anindividual thinking long-term about the impact of behavior to a fi‐nancial institution thinking long-term and having the incentive tothink long-term.

Individuals need to be influenced by psychology. Institutions, and theindividuals in them, are employees that can be motivated or not. Asan institution, they need financial incentives that are aligned with thelong-term rather than the short-term.

That, again, goes back to having a vested interest in the health of peoplerather than in the cost of care.

Employers, to some extent, have that already. Your employer wantsyou to be healthy. They want you to show up for work, be cheerful,motivated and well rested. They get a benefit from you being healthy,far beyond simply avoiding the cost of your care.

Whereas the insurance companies, at this point, simply pass itthrough. If the insurance company is too effective, they actually haveto lower their premiums, which is crazy. It’s really not insurance: it’s acost-sharing and administration role that the insurance companiesplay. That’s something a lot of people don’t get. That needs to be fixed,one way or another.

Esther Dyson on Health Data, “Preemptive Healthcare,” and the Next Big Thing | 111

A Marriage of Data and Caregivers Gives Dr.Atul Gawande Hope for Health CareBy Alex Howard

Dr. Atul Gawande (@Atul_Gawande) has been a bard in the healthcare world, straddling medicine, academia and the humanities as apracticing surgeon, medical school professor, best-selling author, andstaff writer at the New Yorker magazine. His long-form narratives andbooks have helped illuminate complex systems and wicked problemsto a broad audience.

One recent feature that continues to resonate for those who wish toapply data to the public good is Gawande’s New Yorker piece “The HotSpotters,” where Gawande considered whether health data could helplower medical costs by giving the neediest patients better care. Thatstory brings home the challenges of providing health care in a city,from cultural change to gathering data to applying it.

This summer, after meeting Gawande at the 2012 Health DataPaloo‐za, I interviewed him about hot spotting, predictive analytics, net‐worked transparency, health data, feedback loops, and the problemsthat technology won’t solve. Our interview, lightly edited for contentand clarity, follows.

Given what you’ve learned in Camden, N.J. — the backdrop for yourpiece on hot spotting — do you feel hot spotting is an effective wayfor cities and people involved in public health to proceed?

Gawande: The short answer, I think, is “yes.”

Here we have this major problem of both cost and quality — and wehave signs that some of the best places that seem to do the best jobscan be among the least expensive. How you become one of those placesis a kind of mystery.

It really parallels what happened in the police world. Here is somethingthat we thought was an impossible problem: crime. Who could pos‐sibly lower crime? One of the ways we got a handle on it was by di‐recting policing to the places where there was the most crime. It soundskind of obvious, but it was not apparent that crime is concentrated andthat medical costs are concentrated.

112 | Chapter 6: Big Data and Health Care

The second thing I knew but hadn’t put two and two together about isthat the sickest people get the worst care in the system. People withcomplex illness just don’t fit into 20-minute office visits.

The work in Camden was emblematic of work happening in pocketsall around the country where you prioritize. As soon as you look at thesystem, you see hundreds, thousands of things that don’t work properlyin medicine. But when you prioritize by saying, “For the sickest people— the 5% who account for half of the spending — let’s look at whattheir $100,000 moments are,” you then understand it’s strengtheningprimary care and it’s the ability to manage chronic illness.

It’s looking at a few acute high-cost, high-failure areas of care, such ashow heart attacks and congestive heart failure are managed in the sys‐tem; looking at how renal disease patients are cared for; or looking ata few things in the commercial population, like back pain, being a hugesource of expense. And then also end-of-life care.

With a few projects, it became more apparent to me that you genuinelycould transform the system. You could begin to move people fromdepending on the most expensive places where they get the least careto places where you actually are helping people achieve goals of carein the most humane and least wasteful ways possible.

The data analytics office in New York City is doing fascinating pre‐dictive analytics. That approach could have transformative applica‐tions in health care, but it’s notable how careful city officials havebeen about publishing certain aspects of the data. How do you thinkabout the relative risks and rewards here, including balancing socialgood with the need to protect people’s personal health data?

Gawande: Privacy concerns can sometimes be a barrier, but I haven’tseen it be the major barrier here. There are privacy concerns in thedata about households as well in the police data.

The reason it works well for the police is not just because you have abunch of data geeks who are poking at the data and finding interestingthings. It’s because they’re paired with people who are responsible forresponding to crime, and above all, reducing crime. The commanderswho have the responsibility have a relationship with the people whohave the data. They’re looking at their population saying, “What arewe doing to make the system better?”

That’s what’s been missing in health care. We have not married thepeople who have the data with people who feel responsible for ach‐

A Marriage of Data and Caregivers Gives Dr. Atul Gawande Hope for Health Care | 113

ieving better results at lower costs. When you put those people to‐gether, they’re usually within a system, and within a system, there isno privacy barrier to being able to look and say, “Here’s what we canbe doing in this health system,” because it’s often that particular.

The beautiful aspect of the work in New York is that it’s not at a terriblyabstract level. Yes, they’re abstracting the data, but they’re also helpingthe police understand: “It’s this block that’s the problem. It’s shifted inthe last month into this new sector. The pattern of the crime is that itlooks more like we have a problem with domestic violence. Here are afew more patterns that might give you a clue about what you can goin and do.” There’s this give and take about what can be produced andachieved.

That, to me, is the gold in the health care world — the ability to peerin and say: “Here are your most expensive patients and your sickestpatients. You didn’t know it, but here, there’s an alcohol and drug ad‐diction issue. These folks are having car accidents and major traumaand turning up in the emergency rooms and then being admitted with$12,000 injuries.”

That’s a system that could be improved and, lo and behold, there’s anintervention here that’s worked before to slot these folks into treatmentprograms, which by and large, we don’t do at all.

That sense of using the data to help you solve problems requires twothings. It requires data geeks and it requires the people in a system whofeel responsible, the way that Bill Bratton made commanders feel re‐sponsible in the New York police system for the rate of crime. Wehaven’t had physicians who felt that they were responsible for 10,000ICU patients and how well they do on everything from the cost to howlong they spend in the ICU.

Health data is creating opportunities for more transparency intooutcomes, treatments, and performance. As a practicing physician,do you welcome the additional scrutiny that such collective intelli‐gence provides, or does it concern you?

Gawande: I think that transparency of our data is crucial. I’m not surethat I’m with the majority of my colleagues on this. The concerns arethat the data can be inaccurate, that you can overestimate or under‐estimate the sickness of the people coming in to see you, and that mypatients aren’t like your patients.

114 | Chapter 6: Big Data and Health Care

That said, I have no idea who gets better results at the kinds of oper‐ations I do and who doesn’t. I do know who has high reputations andwho has low reputations, but it doesn’t necessarily correspond to thekinds of results they get. As long as we are not willing to open up datato let people see what the results are, we will never actually learn.

The experience of what happens in fields where the data is open is thatit’s the practitioners themselves that use it. I’ll give a couple of exam‐ples. Mortality for childbirth in hospitals has been available for a cen‐tury. It’s been public information, and the practitioners in that fieldhave used that data to drive the death rates for infants and mothersdown from the biggest killer in people’s lives for women of childbearingage and for newborns into a rarity.

Another field that has been able to do this is cystic fibrosis. They haddata for 40 years on the performance of the centers around the countrythat take care of kids with cystic fibrosis. They shared the data privately.They did not tell centers how the other centers were doing. They justtold you where you stood relative to everybody else and they didn’tmake that information public. About four or five years ago, they beganmaking that information public. It’s now available on the Internet. Youcan see the rating of every center in the country for cystic fibrosis.

Several of the centers had said, “We’re going to pull out because thisisn’t fair.” Nobody ended up pulling out. They did not lose patients inhoards and go bankrupt unfairly. They were able to see from one an‐other who was doing well and then go visit and learn from one andother.

I can’t tell you how fundamental this is. There needs to be transparencyabout our costs and transparency about the kinds of results. It’s murkydata. It’s full of lots of caveats. And yes, there will be the occasionaljournalist who will use it incorrectly. People will misinterpret the data.But the broad result, the net result of having it out there, is so muchbetter for everybody involved that it far outweighs the value of closingit up.

U.S. officials are trying to apply health data to improve outcomes,reduce costs and stimulate economic activity. As you look at the suc‐cesses and failures of these sorts of health data initiatives, what doyou think is working and why?

A Marriage of Data and Caregivers Gives Dr. Atul Gawande Hope for Health Care | 115

Gawande: I get to watch from the sidelines, and I was lucky to partic‐ipate in Datapalooza this year. I mostly see that it seems to be followinga mode that’s worked in many other fields, which is that there’s a fun‐damental role for government to be able to make data available.

When you work in complex systems that involve multiple people whohave to, in health care, deal with patients at different points in time,no one sees the net result. So, no one has any idea of what the actualexperience is for patients. The open data initiative, I think, has inno‐vative people grabbing the data and showing what you can do with it.

Connecting the data to the physical world is where the cool stuff startsto happen. What are the kinds of costs to run the system? How do Iget people to the right place at the right time? I think we’re still inprimitive days, but we’re only two or three years into starting to makesomething more than just data on bills available in the system. Eventhat wasn’t widely available — and it usually was old data and not veryrelevant to this moment in time.

My concern all along is that data needs to be meaningful to both thepatient and the clinician. It needs to be able to connect the abstractworld of data to the physical world of what really happens, whichmeans it has to be timely data. A six-month turnaround on data is notgreat. Part of what has made Wal-Mart powerful, for example, is theytook retail operations from checking their inventory once a month tochecking it once a week and then once a day and then in real-time,knowing exactly what’s on the shelves and what’s not.

That equivalent is what we’ll have to arrive at if we’re to make oursystems work. Timeliness, I think, is one of the under-recognized butfundamentally powerful aspects because we sometimes over prioritizethe comprehensiveness of data and then it’s a year old, which doesn’tmake it all that useful. Having data that tells you something that hap‐pened this week, that’s transformative.

Are you using an iPad at work?

Gawande: I do use the iPad here and there, but it’s not readily part ofthe way I can manage the clinic. I would have to put in a lot of effortfor me to make it actually useful in my clinic.

For example, I need to be able to switch between radiology scans andpast records. I predominantly see cancer patients, so they’ll have 40pages of records that I need to have in front of me, from scans to labtests to previous notes by other folks.

116 | Chapter 6: Big Data and Health Care

I haven’t found a better way than paper, honestly. I can flip betweenscreens on my iPad, but it’s too slow and distracting, and it doesn’t letme talk to the patient. It’s fun if I can pull up a screen image of this orthat and show it to the patient, but it just isn’t that integrated intopractice.

What problems are immune to technological innovation? What willneed to be changed by behavior?

Gawande: At some level, we’re trying to define what great care is. Greatcare means being able to provide optimally knowledgeable care in theright time and the right way for people and not wasting resources.

Some of it’s crucially aided by information technology that connectsinformation to where it needs to be so that good decision-makinghappens, both by patients and by the clinicians who work with them.

If you’re going to be able to make health care work better, you’ve gotto be able to make that system work better for people, more efficientlyand less wastefully, less harmfully and with much better teamwork. Ithink that information technology is a tool in that, but fundamentallyyou’re talking about making teams that can go from being disconnec‐ted cowboys in care to pit crews that actually work together towardsolving a problem.

In a football team or a pit crew, technology is really helpful, but it’s onlya tiny part of what makes that team great. What makes the team greatis that they know what they’re aiming to do, they’re very clear abouttheir goals, and they are able to make sure they execute every basicthing that’s crucial for that success.

What do you worry about in this surge of interest in more data-driven approaches to medicine?

Gawande: I worry the most about a disconnect between the peoplewho have to use the information and technology and tools, and thepeople who make them. We see this in the consumer world. Funda‐mentally, there is not a single [health] application that is remotely likemy iPod, which is instantly usable. There are a gazillion number ofways in which information would make a huge amount of difference.

That sense of being able to understand the world of the user, the taskthat’s accomplished and the complexity of what they have to do, and

A Marriage of Data and Caregivers Gives Dr. Atul Gawande Hope for Health Care | 117

connecting that to the people making the technology — there just aren’tthat many lines of marriage. In many of the companies that have someof the dominant systems out there, I don’t see signs that that’s neces‐sarily going to get any better.

If people gain access to better information about the consequencesof various choices, will that lead to improved outcomes and qualityof life?

Gawande: That’s where the art comes in. There are problems becauseyou lack information, but when you have information like “youshouldn’t drink three cans of Coke a day — you’re going to put onweight,” then having that information is not sufficient for most people.

Understanding what is sufficient to be able to either change the careor change the behaviors that we’re concerned about is the crux of whatwe’re trying to figure out and discover.

When the information is presented in a really interesting way, peoplehave gradually discovered — for example, having a little ball on yourdashboard that tells you when you’re accelerating too fast and burningoff extra fuel — how that begins to change the actual behavior of theperson in the car.

No amount of presenting the information that you ought to be drivingin a more environmentally friendly way ends up changing anything.It turns out that change requires the psychological nuance of present‐ing the information in a way that provokes the desire to actually do it.

We’re at the very beginning of understanding these things. There’s alsothe same sorts of issues with clinician behavior — not just information,but how you are able to foster clinicians to actually talk to one anotherand coordinate when five different people are involved in the care ofa patient and they need to get on the same page.

That’s why I’m fascinated by the police work, because you have thedata people, but they’re married to commanders who have responsi‐bility and feel responsibility for looking out on their populations andsaying, “What do we do to reduce the crime here? Here’s the kind ofinformation that would really help me.” And the data people comeback to them and say, “Why don’t you try this? I’ll bet this will helpyou.”

It’s that give and take that ends up being very powerful.

118 | Chapter 6: Big Data and Health Care

Five Elements of Reform that Health ProvidersWould Rather Not Hear AboutBy Andy Oram

The quantum leap we need in patient care requires a complete overhaulof record-keeping and health IT. Leaders of the health care field knowthis and have been urging the changes on health care providers foryears, but the providers are having trouble accepting the changes forseveral reasons.

What’s holding them back? Change certainly costs money, but the in‐dustry is already groaning its way through enormous paradigm shiftsto meet current financial and regulatory climates, so the money mightas well be directed toward things that work. Training staff to handlepatients differently is also difficult, but the staff on the floor of theseinstitutions are experiencing burn-out and can be inspired by a newdirection. The fundamental resistance seems to be expectations byhealth providers and their vendors about the control they need toconduct their business profitably.

A few months ago I wrote an article titled “Five Tough Lessons I Hadto Learn About Health Care.” Here I’ll delineate some elements of anew health care system that are promoted by thought leaders, that echothe evolution of other industries, that will seem utterly natural in acouple decades — but that providers are loathe to consider. I feel thatleaders in the field are not confronting that resistance with an equiv‐alent sense of conviction that these changes are crucial.

1. Reform Will Not Succeed Unless Electronic Records Standardizeon a Common, Robust Format

Records are not static. They must be combined, parsed, and analyzedto be useful. In the health care field, records must travel with the pa‐tient. Furthermore, we need an explosion of data analysis applicationsin order to drive diagnosis, public health planning, and research intonew treatments.

Interoperability is a common mantra these days in talking about elec‐tronic health records, but I don’t think the power and urgency of recordformats can be conveyed in eight-syllable words. It can be conveyedbetter by a site that uses data about hospital procedures, costs, andpatient satisfaction to help consumers choose a desirable hospital. Oran app that might prevent a million heart attacks and strokes.

Five Elements of Reform that Health Providers Would Rather Not Hear About | 119

Data-wise (or data-ignorant), doctors are stuck in the 1980s, buyingproprietary record systems that don’t work together even between dif‐ferent departments in a hospital, or between outpatient clinics andtheir affiliated hospitals. Now the vendors are responding to pressuresfrom both government and the market by promising interoperability.The federal government has taken this promise as good coin, hopingthat vendors will provide windows onto their data. It never really hap‐pens. Every baby step toward opening up one field or another requiresadditional payments to vendors or consultants.

That’s why exchanging patient data (health information exchange —HIE) requires a multi-million-dollar investment, year after year, andwhy most HIEs go under. And that’s why the HL7 committee, puta‐tively responsible for defining standards for electronic health records(EHR), keeps on putting out new, complicated variations on a longhistory of formats that were not well-enough defined to ensure com‐patibility among vendors.

The Direct Project and perhaps the nascent RHEx RESTful exchangestandard will let hospitals exchange the limited types of informationthat the government forces them to exchange. But it won’t create aplatform (as suggested in this PDF slideshow) for the hundreds of ap‐plications we need to extract useful data from records. Nor will it openthe records to the masses of data we need to start collecting. It remainsto be seen whether Accountable Care Organizations (ACO), which arethe latest reform in U.S. health care and are described in this video,will be able to use current standards to exchange the data that eachmember institution needs to coordinate care. Shahid Shaw has laid outin glorious detail the elements of open data exchange in health care.

2. Reform Will Not Succeed Unless Massive Amounts of Patient DataAre Collected

We aren’t giving patients the most effective treatments because we justdon’t know enough about what works. This extends throughout thehealth care system:

• We can’t prescribe a drug tailored to the patient because we don’tcollect enough data about patients and their reactions to the drug.

• We can’t be sure drugs are safe and effective because we don’t col‐lect data about how patients fare on those drugs.

• We don’t see a heart attack or other crisis coming because we don’ttrack the vital signs of at-risk populations on a daily basis.

120 | Chapter 6: Big Data and Health Care

• We don’t make sure patients follow through on treatment plansbecause we don’t track whether they take their medications andperform their exercises.

• We don’t target people who need treatment because we don’t keeptrack of their risk factors.

Some institutions have adopted a holistic approach to health, but as asociety there’s a huge amount more that we could do in this area.

Leaders in the field know what health care providers could accomplishwith data. A recent article even advises policy makers to focus on thedata instead of the electronic records. The question is whether pro‐viders are technically and organizationally prepped to accept it in suchquantities and variety. When doctors and hospitals think they own thepatients’ records, they resist putting in anything but their own notesand observations, along with lab results they order. We’ve got to changethe concept of ownership, which strikes deep into their culture.

3. Reform Will Not Succeed Unless Patients Are in Charge of TheirRecords

Doctors are currently acting in isolation, occasionally consulting withthe other providers seen by their patients but rarely sharing detailedinformation. It falls on the patient, or a family advocate, to rememberthat one drug or treatment interferes with another or to remind treat‐ment centers of follow-up plans. And any data collected by the patientremains confined to scribbled notes or (in the modern Quantified Selfequivalent) a website that’s disconnected from the official records.

Doctors don’t trust patients. They have some good reasons for this:medical records are complicated documents in which a slight reword‐ing or typographical error can change the meaning enough to risk alife. But walling off patients from records doesn’t insulate them againsterrors: on the contrary, patients catch errors entered by staff all thetime. So ultimately it’s better to bring the patient onto the team andeducate her. If a problem with records altered by patients — deliber‐ately or through accidental misuse — turns up down the line, digitalcertificates can be deployed to sign doctor records and output fromdevices.

The amounts of data we’re talking about get really big fast. Genomicinformation and radiological images, in particular, can occupy dozensof gigabytes of space. But hospitals are moving to the cloud anyway.

Five Elements of Reform that Health Providers Would Rather Not Hear About | 121

Practice Fusion just announced that they serve 150,000 medical prac‐titioners and that “One in four doctors selecting an EHR today choosesPractice Fusion.” So we can just hand over the keys to the patients andstorage will grow along with need.

The movement for patient empowerment will take off, as experts inhealth reform told U.S. government representatives, when patients arein charge of their records. To treat people, doctors will have to ask forthe records, and the patients can offer the full range of treatment his‐tories, vital signs, and observations of daily living they’ve collected.Applications will arise that can search the data for patterns and rele‐vant facts.

Once again, the U.S. government is trying to stimulate patient em‐powerment by requiring doctors to open their records to patients. Butmost institutions meet the formal requirements by providing portalsthat patients can log into, the way we can view flight reservations onairlines. We need the patients to become the pilots. We also need togive them the information they need to navigate.

4. Reform Will Not Succeed Unless Providers Conform to PracticeGuidelines

Now that the government is forcing doctors to release informationabout outcomes, patients can start to choose doctors and hospitals thatoffer the best chances of success. The providers will have to apply morerigor to their activities, using checklists and more, to bring up thescores of the less successful providers. Medicine is both a science andan art, but many lag on the science — that is, doing what has beenstatistically proven to produce the best likely outcome — even at pres‐tigious institutions.

Patient choice is restricted by arbitrary insurance rules, unfortunately.These also contribute to the utterly crazy difficulty determining whata medical procedure will cost as reported by e-Patient Dave and WBURradio. Straightening out this problem goes way beyond the doctors andhospitals, and settling on a fair, predictable cost structure will benefitthem almost as much as patients and taxpayers. Even some insurershave started to see that the system is reaching a dead-end and they areerecting new payment mechanisms.

5. Reform Will Not Succeed Unless Providers and Patients Can FormPartnerships

122 | Chapter 6: Big Data and Health Care

I’m always talking about technologies and data in my articles, but noneof that constitutes health. Just as student testing is a poor model foreducation, data collection is a poor model for medical care. What pa‐tients want is time to talk intensively with their providers about theirneeds, and providers voice the same desires.

Data and good record keeping can help us use our resources moreefficiently and deal with the physician shortage, partly by spreadingout jobs among other clinical staff. Computer systems can’t deal withcomplex and overlapping syndromes, or persuade patients to adoptpractices that are good for them. Relationships will always have to bein the forefront. Health IT expert Fred Trotter says, “Time is the gasthat makes the relationship go, but the technology should be focusedon fuel efficiency.”

Arien Malec, former contractor for the Office of the National Coor‐dinator, used to give a speech about the evolution of medical care.Before the revolution in antibiotics, doctors had few tools to actuallycure patients, but they live with the patients in the same communityand know their needs through and through. As we’ve improved thescience of medicine, we’ve lost that personal connection. Malec arguedthat better records could help doctors really know their patients again.But conversations are necessary too.

Five Elements of Reform that Health Providers Would Rather Not Hear About | 123

Copyright
Table of Contents
Chapter 1. Introduction
Chapter 2. Getting Up to Speed with Big Data

What Is Big Data?

What Does Big Data Look Like?
In Practice

What Is Apache Hadoop?

The Core of Hadoop: MapReduce
Hadoop’s Lower Levels: HDFS and MapReduce
Improving Programmability: Pig and Hive
Improving Data Access: HBase, Sqoop, and Flume
Coordination and Workflow: Zookeeper and Oozie
Management and Deployment: Ambari and Whirr
Machine Learning: Mahout
Using Hadoop

Why Big Data Is Big: The Digital Nervous System

From Exoskeleton to Nervous System
Charting the Transition
Coming, Ready or Not

Chapter 3. Big Data Tools, Techniques, and Strategies

Designing Great Data Products

Objective-based Data Products
The Model Assembly Line: A Case Study of Optimal Decisions Group
Drivetrain Approach to Recommender Systems
Optimizing Lifetime Customer Value
Best Practices from Physical Data Products
The Future for Data Products

What It Takes to Build Great Machine Learning Products

Progress in Machine Learning
Interesting Problems Are Never Off the Shelf
Defining the Problem

Chapter 4. The Application of Big Data

Stories over Spreadsheets

A Thought on Dashboards
Full Interview

Mining the Astronomical Literature

Interview with Robert Simpson: Behind the Project and What Lies Ahead
Science between the Cracks

The Dark Side of Data

The Digital Publishing Landscape
Privacy by Design

Chapter 5. What to Watch for in Big Data

Big Data Is Our Generation’s Civil Rights Issue, and We Don’t Know It
Three Kinds of Big Data

Enterprise BI 2.0
Civil Engineering
Customer Relationship Optimization
Headlong into the Trough

Automated Science, Deep Data, and the Paradox of Information

(Semi)Automated Science
Deep Data
The Paradox of Information

The Chicken and Egg of Big Data Solutions
Walking the Tightrope of Visualization Criticism

The Visualization Ecosystem
The Irrationality of Needs: Fast Food to Fine Dining
Grown-up Criticism
Final Thoughts

Chapter 6. Big Data and Health Care

Solving the Wanamaker Problem for Health Care

Making Health Care More Effective
More Data, More Sources
Paying for Results
Enabling Data
Building the Health Care System We Want
Recommended Reading

Dr. Farzad Mostashari on Building the Health Information Infrastructure for the Modern ePatient
John Wilbanks Discusses the Risks and Rewards of a Health Data Commons
Esther Dyson on Health Data, “Preemptive Healthcare,” and the Next Big Thing
A Marriage of Data and Caregivers Gives Dr. Atul Gawande Hope for Health Care
Five Elements of Reform that Health Providers Would Rather Not Hear About

Get Your Paper Before the Deadline. Our Services are 100% private and Confidential

Useful Links That Will Help You Around

Link to new order https://studentsolutionsusa.com/orders/stud/new

link to login page https://studentsolutionsusa.com/orders/login

New user registration link https://studentsolutionsusa.com/orders/register

Forgot Password https://studentsolutionsusa.com/orders/forgot/password

FREQUENTLY ASKED QUESTIONS

Question: How does this work?

Answer: Good Question. We are a group of a freelance board of students and professional essay writers. At our website, you may get help with any type of academic assignments: essay, coursework, term paper, business plan, case study, article review, research paper, presentation, and speech. Top writers can help with complex assignments such as dissertations, thesis papers, etc. All of them are professionals possessing excellent knowledge in their field of expertise, perfect writing skills, quality, and speed. When you place an order on our website, we assign it to the best writer. Once the writer finishes the work, the paper is submitted to our quality assurance desk who go through it and ensure it is unique and plagiarism free and that the instructions were followed to the detail. After this step we upload the paper in your account, we also send a copy to the email that you used to register the account with. we can guarantee you that the paper will be 100% plagiarism free. Besides, our services are 100% private and confidential

Question: How do I place an Order after getting to the order page

Answer: There are three major steps in the ordering process

Step 1 ....................................................paper details In this step, you will fill in the instructions of your paper; you can upload any materials that you feel will make your assignment a success. Besides, you can also email us at [email protected] Remember to specify the correct academic level. Please note that sources mean the number of references.

Step 2...................................................... Price calculation Kindly specify the number of pages, type of spacing and the correct deadline. This step will give you the estimated cost minus discount -- you may add the extra features if you wish.

Step 3 ....................................................discount and payment Use the discount code HAPPY2018 to enjoy up to 30% discount of your total cost After this step, proceed to safe payment; you can checkout using your card or PayPal Please note we will send the complete paper to the email you will provide while registering. A copy will also be uploaded to your account

Question: How will I know when my paper is complete? or How will I get the complete Paper?

Answer: Once we are done with the paper, we will be uploaded to your account. A copy will also be sent to the email you registered with. We can guarantee you the following:- 1. Our service is private and confidential; we don't spam or share your contacts with anyone 2. The final paper will be plagiarism free. We will send a Turnitin Report to the email you registered with 3. At our company, willing to do free unlimited revisions until you are satisfied with your paper

Question:- Am a new client, How can I get the guarantee that the paper will be completed and sent to me before my deadline?

Answer: Thank you for expressing your concerns. We would love to have you as our loyal customer. We are certain if we do good work, you will come back for me. Besides, you will give us referrals to your friends and family. For that reason, we can’t fail to deliver your paper within your specified time frame. We will ensure we submit the paper on time so that you can have enough time to go through it, if you have problems with the paper delivered, you can request a free revision. You are entitled to as many revisions as you would wish until you get a paper that satisfies you

Big_Data_Now_2012_Edition.pdf

Useful Links That Will Help You Around

FREQUENTLY ASKED QUESTIONS

Question: How does this work?

Question: How do I place an Order after getting to the order page

Question: How will I know when my paper is complete? or How will I get the complete Paper?

Question:- Am a new client, How can I get the guarantee that the paper will be completed and sent to me before my deadline?

Useful Links That Will Help You Around

Contact Us