Cloudcroft-S Persuasion and the Presidency PREDICT402-56 W2015

Running head: PERSUASION AND THE PRESIDENCY 1
Persuasion and the Presidency: Models to Influence Outcome
Sheila M Cloudcroft
Northwestern University

PERSUASION AND THE PRESIDENCY 2
Persuasion and the Presidency: Models to Influence Outcome
In November 2012, Barak Obama battled for and won a second term as President using
custom marketing personalization developed from big data insights and a predictive model-based
approach. Online journal InfoWorld revealed the key performance indicator (KPI) around which
all campaign analysis pivoted: the number who planned to vote for Obama, divided by those who
planned to vote overall (Lampitt, 2013). A team of leading data scientists maximized the KPI by
aggregating intelligence on over 190 million citizens in an optimized Amazon Web Service high-
speed Elastic Compute Cloud (EC2) IT infrastructure then using it to identify, understand,
effectively engage with, and persuade enough voters in key states to ensure victory (Gallagher,
2012; Pilkington & Michel, 2012).
2012 Obama for America: Calculated Actions
No activity engaged in by the Obama for America (OFA) campaign was coincidental. In
every case data was used to determine whether and how to contact voters (Siegal, 2013). Each
action was carefully planned to generate maximum impact to the KPI. One staff member
revealed, “Call lists in field offices, for instance, didn’t just list names and numbers; they also
ranked names in order of their persuadability, with the campaign’s most important priorities first
(Scherer, 2012).” A senior official said, “We ran the election 66,000 times a night [to predict]
chances of winning these states…and that is how we allocated resources (qtd in Scherer 2).”
Staff formulated, constructed, and then relied heavily on uplift models to deliver personalized,
persuasive marketing for ongoing shifts in targeted demographic voters (Preslar, 2013).
Personalized, persuasive marketing is a fairly new, methodical approach to accurately
determine persuasive communication styles, content, and delivery outlets to use on a targeted

subject to get them to buy a product. In this case, the product being sold was the incumbent for
President. The method relies heavily on harvested demographic and personal data.
The OFA team verified efficacy of planned actions to raise the KPI by conducting
randomized controlled experiments on samples of targeted populations before launching them to
the full population ("What We Do: Experimentally Informed Programs", n.d., para. 2). For
example, a comprehensive report “Inside the Cave” (DC, 2012) reports that 240 A/B tests were
conducted with visitors to the Obama team donation page to optimize campaign donations.
Microtargeting Voters: The Personal is the Political
Highly accurate knowledge of specific voters was mined from terabytes of data generated
from demographic and regional analysis, social media trawling, research, experimentation, and
voter characterization resources (Goldberg, 2013; Mitchell R., 2012; Mitchell R.L., 2012,
Scherer, 2012). The campaign contracted with the company Catalist to obtain an additional “500
terabytes of data detailing…demographic (race, gender, age, income) and psychographic
(interests, hobbies, lifestyles) information on individuals and households (qtd in Mitchell, 2).”
Elaborate knowledge of social media networks such as Facebook was compiled.
Persuadable target voters who engaged in social networking were linked to influencers
(Pilkington & Michel, 2012), those individuals whose posts frequently elicit some reaction from
others. Once that network was known, a subtle campaign began. With the goal of triggering a
viral cascade of comments, likes, and shares from a single post, strategists encouraged loyal
supporters to “tell your own story.” These word-of-mouth posts were distributed across friend
networks and provided a personal touch to get voters aligned with Obama. Influencers were
identified using a method called influence maximization, where a small set of seed nodes capable
of triggering response in the maximal range of nodes connected in a network is determined

(Huang, Cheng, Shen, Zhou, & Jin, 2012). The OFA team stimulated the seed nodes to elicit the
desired cascade. Huang et al. (2012) exposit that “greedy algorithm[s]” such as that developed by
Kempe, Kleinberg, & Tardos (2003) and Leskovec et al. (2007), have long been known to
guarantee a near-optimal response in networks.
The team was able to generate fairly accurate predictions whether a topic was likely to
gain media attention within a few days of releasing ads or news briefs based on whether it
generated Twitter chatter (Issenberg, 2012). Research performed by Hong & Nadler (2012)
develops mathematical formulas purposed to confirm linkages between popularity of news and
Twitter tweets. They found that there is “strong evidence positively linking the number of times a
politician is mentioned by popular traditional media with the number of mentions he gets on
Twitter. On average, a 10% increase in the number of media mentions for a politician is
associated with a 4 to 6% increase in the number of Twitter mentions” (Hong & Nadler, 2012).
Issenberg (2012) said 500,000 conversations were initiated by Obama volunteers.
Randomly selected voters identified as persuadable who had received no prior contact from the
campaign were surveyed after a controlled phone conversation with a volunteer (Issenberg,
2012). Issenberg (2012) quotes National campaign director Jeremy Bird, “We definitely [found]
certain people moved more than other people” (Issenberg, 2012). Bird states that these attributes
became the core of a persuasion model predicting the likelihood that a voter could be pulled in
Obama’s direction after a single volunteer interaction (Issenberg, 2012).
Apps were written for volunteers going door-to-door that delivered and transmit back to
headquarters intelligence on voters (Goldberg, 2013). Specific individuals were contacted during
foot surveys because they were identified as persuadable (Siegal, 2013). In a webcast sponsored
by HP Vertica, applications developer and Democratic National Committee director of data

architecture Chris Wegrzyn (qtd in Goldberg, 1) said: “When a volunteer knocked on the door,
and [the voter] said, ‘I am not sure I support the president. I would like to hear about health care
law,’ they would enter that into contact database.” The campaign created a means for the
volunteer to follow up by email with the undecided target voter, delivering personalized
messages with information such as that about health care policies, along with an invitation to
discuss the issue further.
Did it Work?
History said: Yes. Barak Obama was re-elected. “Results [of personalized, persuasive
modeling] were that significant lift was shown over just targeting voters who were undecided or
had registered as non-partisan (Siegal, 2013).”
Online report “Inside the Cave” (2012) states that OFA campaign donations rose
by 49% after utilizing A/B tests for website design. Fundraising efforts exceeded $1
billion. Simulation models used to predict the vote were accurate within 0.2% in Ohio and
0.4% in Florida, but were 1% too cautious in Colorado. One model predicted 57.68% of
the OH Early Vote would go to Obama; actuals were 57.16%. 600,000 purposely chosen
influencers reached 5 million voters of whom 20% took some action such as registering to
vote.

References
What We Do: Experimentally Formed Programs (n.d.). In bluelabs. Retrieved from
Experimentally Formed Programs: http://www.bluelabs.com/
DC, E. (2012, December). Inside the Cave. In enga.ge. Retrieved from
http://enga.ge/projects/inside-the-cave/
Gallagher, S. (2012). How Team Obama's tech efficiency left Romney IT in dust: Obama
campaign’s tech team beat Romney by using opposite strategy – “insourcing.” In
Technology Lab: Information Technology. Retrieved from arstechnica
http://arstechnica.com/information-technology/2012/11/how-team-obamas-tech-
efficiency-left-romney-it-in-dust/
Goldberg, M. (2013). Inside the Obama Campaign's Big Data Analytics Culture. Big Data and
Analytics in the Enterprise. DataInformed. Retrieved from http://data-
informed.com/inside-the-obama-campaigns-big-data-analytics-culture/
Hong, S., and Nadler, D. (2012). Which candidates do the public discuss online in an election
campaign?: The use of social media by 2012 presidential candidates and its impact on
candidate salience. Government Information Quarterly 29(4), pp. 455-461.
Huang, J., Cheng, X., Shen, H., Zhou, T., and Jin, X. (2012). Exploring Social Influence via
Posterior Effect of Word-of-Mouth Recommendations. Proceedings of the fifth ACM
international conference on Web search and data mining. WSDM '12, pp. 573-582. New
York, NY, USA. ACM.
Issenberg, S. (2012). How President Obama’s campaign used big data to rally individual voters.
In MIT Technology Review. Retrieved from

http://www.technologyreview.com/featuredstory/509026/how-obamas-team-used-big-
data-to-rally-voters/
Kempe, D., Kleinberg, J., & Tardos, E. (2003). Maximizing the spread of inﬂuence through a
social network. In Proceedings of the ninth ACM SIGKDD international conference on
Knowledge discovery and data mining, KDD ’03, pages 137–146. New York, NY, USA.
ACM.
Lampitt, A. (2013). Think Big Data: The real story of how big data analytics helped Obama win |
HP Vertica played a major role, as did an org structure that centralized analytics and
lowered barriers between teams. InfoWorld. Retrieved from
http://www.infoworld.com/article/2613587/big-data/the-real-story-of-how-big-data-
analytics-helped-obama-win.html
Leskovec, J., Krause, A., Guestrin, C., Faloutsos, C., VanBriesen, J., and Glance, N. (2007).
Cost-eﬀective outbreak detection in networks. In Proceedings of the 13th ACM SIGKDD
international conference on Knowledge discovery and data mining, KDD ’07, pages 420–
429, New York, NY, USA. ACM.
Mitchell, R. (2012). Big Data, Campaign 2012: Mining for voters | Data-driven campaigning
goes mainstream. ComputerWorld. Retrieved from
http://www.computerworld.com/article/2492578/big-data/campaign-2012--mining-for-
voters.html
Mitchell, R. L. (2012). Reality Check, Election 2012: Obama for America's "Moneyball"
moment. Computer World. Retrieved from
http://www.computerworld.com/article/2473398/business-intelligence/election-2012--
obama-for-america-s--moneyball--moment.html

Pilkington, E., & Michel, A. (2012). Obama, Facebook and the power of friendship; the 2012
data election. The Guardian. Retrieved from US News
http://www.theguardian.com/world/2012/feb/17/obamadigitaldatamachinefacebookelecti
on
Preslar, E. (2013). How uplift modeling helped Obama's campaign -- and can aid marketers.
Predictive analytics. In SearchBusinessAnalytics. Retrieved from
http://searchbusinessanalytics.techtarget.com/video/How-uplift-modeling-helped-
Obamas-campaign-and-can-aid-marketers
Scherer, M. (2012). Inside the Secret World of the Data Crunchers Who Helped Obama Win.
2012 Election. Retrieved from TIME http://swampland.time.com/2012/11/07/inside-the-
secret-world-of-quants-and-data-crunchers-who-helped-obama-win/
Siegal, E. (2013). Persuasion by Numbers| Beyond Swing Voters: How Persuasion Modeling
Helped Obama Win His Second Term (Nov 2012). In E. Siegal, Predictive Analytics: The
Power to Predict Who Will Click, Buy, Lie, or Die (pp. 213-218). Hoboken, New Jersey,
USA: John Wiley & Sons, Inc.
Siegal, E. (n.d.). Team Obama Mastered the Science of Mass Persuasion. In big think. Retrieved
from http://www.bigthink.com/experts-corner/team-obama-mastered-the-science-of-
mass-persuasion-and-won
Stinebring, J. (2014). How Forecasting Revolutionized Election Resource Allocation. Blog Data
Science. Retrieved from Civis Analytics https://civisanalytics.com/blog/data-
science/2014/08/04/How-Forecasting-Revolutionized-Election-Resource-Allocation/

Tables
Table 1
Online Article Cross Reference for Table 2.
Cross Reference
# Article
1 Scherer, M. (2012). Inside the Secret World of the Data Crunchers Who Helped Obama
Win.
2 What We Do: Experimentally Formed Programs (n.d.).
3 Mitchell, R. L. (2012). Reality Check, Election 2012: Obama for America's "Moneyball"
moment.
4 Mitchell, R. (2012). Big Data, Campaign 2012: Mining for voters | Data-driven
campaigning goes mainstream.
5 Goldberg, M. (2013). Inside the Obama Campaign's Big Data Analytics Culture.
6 Stinebring, J. (2014). How Forecasting Revolutionized Election Resource Allocation.
7 Gallagher, S. (2012). How Team Obama's tech efficiency left Romney IT in dust: Obama
campaign’s tech team beat Romney by using opposite strategy – “insourcing.”
8 Preslar, E. (2013). How uplift modeling helped Obama's campaign -- and can aid
marketers.
9 Lampitt, A. (2013). Think Big Data: The real story of how big data analytics helped Obama
win | HP Vertica played a major role, as did an org structure that centralized
analytics and lowered barriers between teams.
10 DC, E. (2012, December). Inside the Cave.
11 Pilkington, E., & Michel, A. (2012). Obama, Facebook and the power of friendship; the
2012 data election.
12 Siegal, E. (n.d.). Team Obama Mastered the Science of Mass Persuasion.
13 Issenberg, S. (2012). How President Obama’s campaign used big data to rally individual
voters.

Table 2
Summary of Notes from Various Online Article.
Methods
Article Method Purpose Quotation
1 Persuasion Donations, vote George Clooney had an almost gravitational tug on West Coast females ages 40 to 49. The
women were far and away the single demographic group most likely to hand over cash, for a
chance to dine in Hollywood with Clooney — and Obama.
1 Persuasion Donations, vote They sought out an East Coast celebrity who had similar appeal among the same demographic,
aiming to replicate the millions of dollars produced by the Clooney contest. “We were blessed
with an overflowing menu of options, but we chose Sarah Jessica Parker,”
1 Core
Reqmnt
Central Data
Repository
It was like the FBI and the CIA before 9/11: the two camps never shared data. “We analyzed
very early that the problem in Democratic politics was you had databases all over the place,”
said one of the officials. “None of them talked to each other.” So over the first 18 months, the
campaign started over, creating a single massive system that could merge the information
collected from pollsters, fundraisers, field workers and consumer databases as well as social-
media and mobile contacts with the main Democratic voter files in the swing states.
1 Persuasion Donations, vote The new megafile didn’t just tell the campaign how to find voters and get their attention; it also
allowed the number crunchers to run tests predicting which types of people would be
persuaded by certain kinds of appeals.
1 Persuasion Vote Call lists in field offices, for instance, didn’t just list names and numbers; they also ranked names
in order of their persuadability, with the campaign’s most important priorities first. About 75%
of the determining factors were basics like age, sex, race, neighborhood and voting record.
Consumer data about voters helped round out the picture.
1 Persuasion Donations, vote “We could [predict] people who were going to give online. “
1 Persuasion Donations, vote We could model people who were going to give through mail.
1 Persuasion Service We could model volunteers,” said one of the senior advisers about the predictive profiles built
by the data.
1 Core
Reqmnt
All “In the end, modeling became something way bigger for us in ’12 than in ’08 because it made
our time more efficient.”
1 Persuasion Donations, vote Early on, for example, the campaign discovered that people who had unsubscribed from the
2008 campaign e-mail lists were top targets, among the easiest to pull back into the fold with
some personal attention.
1 Persuasion Donations, vote The strategists fashioned tests for specific demographic groups, trying out message scripts that
they could then apply.
1 Persuasion Donations, vote They tested how much better a call from a local volunteer would do than a call from a volunteer
from a non–swing state like California.
1 Persuasion All As Messina had promised, assumptions were rarely left in place without numbers to back them
up.
1 Persuasion Donations, vote A large portion of the cash raised online came through an intricate, metric-driven e-mail
campaign in which dozens of fundraising appeals went out each day. Here again, data collection
and analysis were paramount. Many of the e-mails sent to supporters were just tests, with
different subject lines, senders and messages.
1 Persuasion Donations, vote Chicago discovered that people who signed up for the campaign’s Quick Donate program, which
allowed repeat giving online or via text message without having to re-enter credit-card
information, gave about four times as much as other donors.
1 Persuasion Donations, vote By the end of October, Quick Donate had become a big part of the campaign’s messaging to
supporters, and first-time donors were offered a free bumper sticker to sign up.
1 Persuasion Donations, vote The magic tricks that opened wallets were then repurposed to turn out votes. The analytics
team used four streams of polling data to build a detailed picture of voters in key states. In the
past month, said one official, the analytics team had polling data from about 29,000 people in
Ohio alone — a whopping sample that composed nearly half of 1% of all voters there —
allowing for deep dives into exactly where each demographic and regional group was trending
at any given moment.

Methods
1 Persuasion Vote “We ran the election 66,000 times every night,” said a senior official, describing the computer
simulations the campaign ran to figure out Obama’s odds of winning each swing state. “And
every morning we got the spit-out — here are your chances of winning these states. And that is
how we allocated resources.”
1 Persuasion Friends, Vote Online, the get-out-the-vote effort continued with a first-ever attempt at using Facebook on a
mass scale to replicate the door-knocking efforts of field organizers. In the final weeks of the
campaign, people who had downloaded an app were sent messages with pictures of their
friends in swing states. They were told to click a button to automatically urge those targeted
voters to take certain actions, such as registering to vote, voting early or getting to the polls.
The campaign found that roughly 1 in 5 people contacted by a Facebook pal acted on the
request, in large part because the message came from someone they knew.
1 Persuasion Media, Vote Rather than rely on outside media consultants to decide where ads should run, Messina based
his purchases on the massive internal data sets. “We were able to put our target voters through
some really complicated modeling, to say, O.K., if Miami- Dade women under 35 are the targets,
[here is] how to reach them,” said one official. How much more efficient was the Obama
campaign of 2012 than 2008 at ad buying? Chicago has a number for that: “On TV we were able
to buy 14% more efficiently … to make sure we were talking to our persuadable voters,” the
same official said.
1 Persuasion Media, Vote The numbers also led the campaign to escort their man down roads not usually taken in the late
stages of a presidential campaign. In August, Obama ...answer questions on the social news
website Reddit...Because a whole bunch of our turnout targets were on Reddit.”
2 Core Reqnt Modelling,
Media
persuasion modeling, which identifies individuals who are most likely to change their opinions
or behaviors in response to a particular message. Our modeling team is led by Dan Porter and
Matt Holleque, respectively the Obama campaign’s director and deputy director of predictive
modeling. In 2012, they developed the Obama campaign’s groundbreaking voter persuadability
model, which informed nearly all of the campaign’s programs, including paid media, volunteer
outreach, and digital strategies.
2 Core
Reqmnt
Modelling,
Media
Our team of experienced scientists provides immediate actionable insight by designing,
implementing, and quickly analyzing results from randomized controlled experiments. For
instance, a campaign or organization can test its direct mail on a small scale, learn from the
test’s outcome, then launch its broader mail program to the best possible universe.
2 Core
Reqmnt
Modelling,
Media
revolutionary media optimization technology helps clients identify the best media platforms
and TV shows for reaching target audiences in the most cost-effective way. BlueLabs’ media
optimization technology moves beyond traditional demographic based approaches to
advertising, enabling our clients to target their content to particular individuals. By combining
predictive modeling with cable set-top box data, we pinpoint your target voters, customers, and
so on, identify the shows that they are most likely to watch, and ultimately develop the most
cost-effective buying strategy. Chris Wegrzyn, a BlueLabs co-founder, led the engineering effort
behind the original media optimizer used by the Obama campaign in 2012.
3 Persuasion Donations, vote You'll see more canvassers knocking at your door with devices like Square to take credit card
donations on the spot, along with mobile apps that tell the volunteer how much to ask for
based on previous donation history.
3 Persuasion Donations, vote "Facebook and to some extent Twitter have become a repository for consumer data that's
probably unequaled in history in the sense of consumer intent and preferences and political
affiliations," says Ruffini. All of that is now feeding into the hopper to drive both tactics and
strategy.
3 Persuasion Donations, vote auto-dialer get-out-the-vote programs ...decide who to call and give volunteers a script with a
message designed to have the highest probability of moving each micro-targeted segment of
voters to action.
4 Persuasion Donations, vote As campaign volunteers go door to door, they might rely on mobile apps for customized
messages about specific households. They could look at profiles that not only indicate whether
an individual is a Republican or a Democrat...canvassers can use apps to capture details of
interactions with voters and upload that information to the campaign database, thereby
providing continuous, real-time feedback.
4 Persuasion Modelling, Vote It's a two-step process, Duvall explains. Analysts apply data mining techniques against a massive
database that provides very detailed profiles of its own members as well as "look-alikes" who fit
the profile of swing voters. From there they develop models that predict which voter profiles
will be most likely to respond positively to a campaign message and which type of issue will be
most likely to move them to action.

Methods
4 Persuasion Modelling, Vote "In some instances, we can take this research a level deeper through real-world
experimentation," Duvall says. To accomplish this, Sierra staffers try out a range of specific
messages on test groups to determine which will be the most effective before launching the
campaign to the target audience. "We can see which messages are moving the voters. Before
we could do cross-tabs and see the broad categories of people who might be moving, but with
data mining we can go much deeper."
4 Persuasion Central Data
Repository,
Modelling, Vote
Duvall works with Catalist, a consortium of progressive organizations that maintains a 500-
terabyte database of information describing both registered and unregistered U.S. voters. Laura
Quinn "Our database is about civic behavior and transactions, what issues you care about, what
causes you support, whether you tend to vote or not, and so on," says Catalist CEO Laura Quinn.
Catalist matches up the Sierra Club's member database with its own data and provides access to
the full database, which combines state voter registration lists with commercial consumer data
that includes demographic (race, gender, age, income) and psychographic (interests, hobbies,
lifestyles) information on individuals and households. Catalist buys commercial consumer data
from traditional data aggregators and reporting agencies such as Acxiom and Equifax. Voter lists
come from the states. For those states that don't release voter registration data, Catalist has
developed models that predict, at the household level, who is likely to be Republican or
Democrat and how they're likely to vote -- something it couldn't do in 2008. "Our database is
about civic behavior and transactions, what issues you care about, what causes you support,
whether you tend to vote or not, and so on," says Catalist CEO Laura Quinn. Yair Ghitza, senior
scientist at Catalist, explains further: "Our clients determine the likelihood that someone is
going to vote, care about certain issues or has leanings on certain issues, their partisanship and
ideologies, and the actions they're most likely to engage in when they take civic action," he says.
4 Persuasion Central Data
Repository,
Modelling, Vote
Aristotle Inc. offers a similar service to trade associations and campaigns, including both
presidential campaigns, according to CEO John Aristotle Phillips. Its database of more than 700
data fields, which describe the traits of more than 85 million registered voters, is used for both
fundraising and get-out-the-vote initiatives. John Aristotle Phillips "What we're seeing in 2012 is
much more effective use of real-time access" to databases about voters, says John Aristotle
Phillips, CEO of Aristotle Inc. Clients use it to create models that find people who are similarly
minded or likely to contribute. Aristotle then helps them deliver a targeted message to
individuals who match the criteria through various channels, including TV, direct mail, email and
social media.
5 Persuasion Vote At the start of 2011, Wegrzyn said, his team had to figure out how to inject this kind of analysis
into all activities of the campaign, not just getting out the vote, but to create systems that
enabled “voter to voter” interactions.
5 Persuasion Vote A presidential campaign has the luxury of focusing on one goal: getting a majority of votes on
Election Day. Wegrzyn said the campaign set up five groups—the field organization along with
teams for digital, communications, media relations and finance—to focus on three activities:
registration to increase the pool of eligible voters, persuasion to win support, and voter turnout.
5 Method Modelling While the campaign had less than 10 terabytes of raw data, the analysts would end up
generating many times that amount in their experiments. There were many datasets to
manage, including new sources—and the pace of the work meant they did not have time to do
ETL processes.
5 Method Modelling In six months of evaluations, Wegrzyn said his team chose HP Vertica as its analytics platform
after rejecting Hadoop based systems for most use cases (high learning curve and long
development lead time) and appliances (the team needed performance before it needed
massive storage). Vertica provided a massively parallel processing database with familiar SQL
queries and a path to scalability as analysts continued to pound on the system, he said.
5 Method Modelling Users manipulated and analyzed data using the R and Stata statistical programming languages,
as well as SQL queries and third party analytics programs, Wegrzyn said. (Analysts also used
data mining tools from KXEN (http://www.kxen.com/).)
5 Method Modelling AirWolf, which helped the campaign create a connection between volunteer field workers and
digital marketing efforts. By correlating email addresses with voter information, the campaign
could connect volunteers in the field with voters. The application used SQL queries against the
Vertica database, a data reporting tool the campaign developed called Stork to send custom
email via Amazon’s Simple Email Service. Analysts also could analyze results. The effect was a
more personalized politics. “When a volunteer knocked on the door, and [the voter] said, ‘I am
not sure I support the president. I would like to hear about health care law.’ They would enter
that into contact database,” Wegrzyn said. The campaign created a means for the volunteer to
follow up by email with the undecided voter, to send a personalized message with information
about health care policies, and an invitation to discuss the issue further. “Those were brand new
for the campaign and exciting for us,” he said.

Methods
5 Method Modelling ...an application that optimized advertising purchases. ...employed predictive models to identify
target voters. ...received anonymized data from media ratings companies. Combining the two
datasets and then adding pricing data for various media outlets allowed the campaign to pick
the best programs during which to advertise targeted messages. Analysts ran queries to explore
the data to confirm the recommendations made sense. “We were able from experiments to get
a sense at the individual level, of what we termed persuadability” of certain precise voting
groups, he said. Correlating these results with ratings of those groups as television audiences
directed the team’s attention to optimize certain media channels for advertising. The result was
a precision targeted advertising onslaught.
6 Method Modelling (code named “Golden”) while working for Obama for America (OFA) during the 2012 election
cycle. The algorithm ingests internal predictive modeling, consultant polling, and public pollster
toplines, and in turn generates state-level estimates of candidate support, orderings of relative
competitiveness, and an overall likelihood of victory on Election Day. ...deciding which states in
which to invest and the proper mix of resource allocation between them is often the difference
between victory and defeat. Rigorous election forecasting helps make sense of divergent data
points and leverages uncertainty for more informed decision making.
6 Method Modelling The “Golden” model had two major facets. First, for each pollster and poll, estimates of partisan
bias and expected error were generated. When new polling results were released, the model
could efficiently situate the new piece of information in a larger political context.
6 Method Modelling Second, the “Golden” algorithm would use this base of reconciled public opinion to simulate the
Election Day results 62,000 times each evening. These simulations were then interpreted to
generate state-by-state estimates of support, rank competitiveness of states, and give an
overall likelihood of victory.
6 Method Modelling As opposed to seeing each new piece of polling information as disjointed and discrete, this
information could then contribute to building a holistic image of the race. Second, the “Golden”
algorithm would use this base of reconciled public opinion to simulate the Election Day results
62,000 times each evening.
6 Method Modelling the “Golden” algorithm showed stable levels of candidate support and accurately predicted the
ultimate outcome of all 50 states. Those forecasts, refreshed and refined daily throughout the
rest of the cycle, never presented an alternate view of the electoral landscape. The final
“Golden” estimates in the days before the election were within 1.1% of the actual results in
every battleground state.
7 Method Central Data
Repository
maximizing the value of the Obama campaign's IT spending was its use of open source tools and
open architectures. Linux—particularly Ubuntu—was used as the server operating system of
choice. "We were technology agnostic, and used the right technology for the right purpose,"
VanDenPlas said. "Someone counted nearly 10 distinct DBMS/NoSQL systems, and we wrote
something like 200 apps in Python, Ruby, PHP, Java, and Node.js."
Repository
for internally developed applications, relied almost exclusively on Amazon Web Service for its
infrastructure, eliminating a lot of the financial burden of infrastructure management. "For the
applications built by the OFA [Obama for America] technology team, 99.999 percent were AWS
hosted,"
Repository
The system configurations for the campaign's Elastic Compute Cloud (EC2) instances were
created using the Puppet configuration management tool and were built as Debian packages
kept in the campaign's own Advanced Packaging Tool (apt) repository—both for internally
developed and third party applications. As the number of applications and the scale of the
campaign's AWS infrastructure use climbed, the DevOps team shifted to using Asgard—an open
source tool developed by Netflix to manage cloud deployments.
Repository
To help optimize applications, the OFA technology team used New Relic, a tool also used by the
Romney campaign. "It is really a fantastic tool that increases your visibility into where your
applications are spending time," VanDenPlas said. "They support the major languages we used
(Python, Ruby, PHP) as well as the frameworks (Flask, Rails, Kohana)." While AWS's tools were
used for performance monitoring and to trigger automatic scaling up of capacity, VanDenPlas
said, much of the monitoring was handled by a suite of commercial and open source tools and
homegrown code, "consisting of Cacti, Opsview, StatsD, Graphite, and Seyren, and a number of
custom applications that continued to evolve right up until Election Day," VanDenPlas said.
Repository
To get better aggregated alerting and metric data, the team built a lightweight plugin for Nagios
(the open source basis of Opsview) in Python based on boto (the Python programming interface
to AWS's services) and dotCloud's ZeroRPC messaging interface. "Using this," VanDenPlas
explained, "we could constantly query thousands of nodes for near realtime statistics and feed
them right back into the same alerting and monitoring system (Nagios) we used elsewhere."

Methods
Repository
Other performance monitoring and user experience data was collected using Chartbeat and
Google Analytics. "Akamai also provided very useful statistics and logging," VanDenPlas said,
"but these were mostly contextual rather than actionable." But, he added, the most heavily
used monitoring system was "our community of internal and external supporters. The human
factor in monitoring is huge. There are countless incidents where (OFA User Support Director)
Brady Kriss notified us of pending problems derived from community help tickets."
Repository
As the election approached and the infrastructure demands surged, the engineering team took
advantage of Amazon's multiple availability zones within its Virgina data center. "We built out a
triply redundant, encrypted, and compressed WAN optimized tunnel between AWS regions,"
VanDenPlas said, "using a combination of OpenVPN, CloudOptimizer, and some DNS trickery."
Repository
The team shifted its domain name service to Amazon's Route 53 service, which uses latency
based routing to direct users to the host running in the AWS availability region with the shortest
network trip time.
Repository
The Obama campaign's websites were also hosted on Amazon and hardened. The campaign's
engineers built an application that created static HTML snapshots of the sites stored in
Amazon's Simple Storage Service (S3) ; in the event of a Web server failure, requests would be
instantly directed to the latest snapshot.
Repository,
Investment
Strategy / Costs
cost/expenditure breakout of the Obama campaign
8 Method Modelling The campaign needed a way to increase support for the president while making the most
effective use of its available resources. Enter uplift modeling, a form of predictive analytics that
aims to identify individuals who are likely to be positively influenced by ads, mailings, phone
calls and other outreach efforts.
8 Method Modelling uplift modeling, also known as persuasion modeling. Its goal is to pinpoint the "persuadables" --
in this case, voters who were leaning toward Republican nominee Mitt Romney but might
decide to vote for Obama if they were contacted. With that information in hand, the Obama
campaign could avoid spending money and volunteer time contacting people who the analytical
models showed were already committed to voting for the president or for Romney.
8 Method Modelling "The old adage in advertising is, 'I know half of my advertising isn't working; I just don't know
which half,'" Porter said. "With uplift modeling, you can identify which half is working and which
half isn't -- or more specifically, what customers are most receptive to advertising and what
customers aren't."
8 Method Modelling To accomplish that for the Obama campaign, Porter's team first set up an experiment in which
some people were called and others weren't; in a presentation at the conference, Porter said
the two groups were then polled, and support for Obama was five percentage points higher in
the "treatment group" than among the voters who weren't contacted. From there, he said, the
analysts used a matrix of political, demographic and household data to develop a set of
predictive models that "applied a score to every voter in all the battleground states."
8 Method Modelling After field testing and revisions, the final versions of the models were put into use in October
2012. Porter said they "guided every door knock and every phone call" in the final weeks of the
campaign. The Obama campaign also used the models to target persuadable voters through
direct mail, social media advertising, Facebook messages and TV ads; for the latter, he said, an
optimizer was used to develop a list of "good buys" based on what channels and programs large
numbers of the desired voters typically watched. The models enabled the campaign to reach
out to voters at an individual level, based on what message they were most likely to be
receptive to and what form of contact they were most likely to be persuaded by, he added.
Repository,
Investment
Strategy / Costs
the Obama campaign built a 100-strong analytics staff to churn through dozens of terabytes of
data with a combination of the HP Vertica MPP (massively parallel processing) analytic database
and predictive models with R and Stata to gain a competitive edge.
9 Method Modelling evaluating the results of experiments helped the campaign learn how its actions actually
influenced people.
9 Core
Reqmnt
Metric the key performance indicator for the campaign was the number who planned to vote for
Obama, divided by those who planned to vote overall.
9 Core
Reqmnt
Strategy The campaign understood there were three levers to maximize that number: registration,
persuasion, and turnout.

Methods
Repository
while Hadoop was an important complementary technology, it required highly technical skills
and was not designed for the real-time queries the team needed. They also realized that a large
analytic appliance, which they used in previous campaigns, would not scale out sufficiently.
Ultimately, the team settled on HP Vertica. It was SQL-based, affordable, and scalable, as well as
a strong performer in proof-of concept tests. On the statistical analytics side, the team used R
and Stata.
Repository
AirWolf was built to integrate the field and digital teams' efforts. A common problem in prior
campaigns was that the field teams' actions, such as recording a person's particular interest in
voting issues, could not be easily followed up by the Digital team (for example, with email
correspondences). With AirWolf, when a voter was contacted by the field team in a door-to-
door campaign, that voter's particular interests were recorded and fed back to Vertica. Then the
digital team ran email blasts from the local organizer to voters, each corresponding to a voter's
favorite campaign issues. This greatly enhanced the ability to pinpoint messaging and make it
more feasible to sway voters.
9 Method Modelling,
Strategy
With the overall picture combining likely voters for Obama, the shows they watch, and the
prices of the ads -- as well as the analysis feedback loop -- it was much easier to determine the
most efficient ad buys.
9 Method Modelling,
Strategy
All the analytic solutions shared a number of attributes: They were a combined effort of both
analysts and engineers. They were time sensitive, implemented in weeks rather than months.
They were built around an unconstrained, yet centralized environment with Vertica.
10 Method Modelling,
Strategy
In Ohio, OFA had ballot test data on 29,000 voters, more than 1% of the electorate, allowing for
deep demographic analysis.
10 Method Battleground
States Analytics,
Modelling,
Strategy
The final simulations were accurate to within 0.2% in Ohio and 0.4% in Florida, but were 1% too
cautious in Colorado.
10 Method Battleground
States Analytics,
Modelling,
Strategy
Battleground States Survey
• Single poll taken across battlegrounds: CO, FL, IA, MI, NV, NH, NC, OH, PA, VA, WI
• Initially once every three weeks; 2 per week in final 2 months
• Used to measure broad public opinion, not individual states; no national polling done
State Tracking Polls
• Three-day rolling sample in each state
• 500-900 voters
Analytics
• Run by Analytics Department
• Live callers, large sample sizes, short questionnaire
• 8,000 - 9,000 calls per night
Strategy
call centers that completed these analytics surveys typically specialize in "voter identification,"
the process of contacting most or all individual voters in a state to identify supporters who can
then be targeted in subsequent "get out the vote" efforts. But the Obama campaign's approach
to voter targeting was different. It called very large random samples of voters to develop
statistical models that generated scores applied to all voters, which were then used for get-out-
the-vote and persuasion targeting."
- Mark Blumenthal, Huffington Post http://www.huffingtonpost.com/2012/11/21/obama-
campaign-polls-2012_n_2171242.html
Strategy
Hamilton County, OH Early Vote
57.68% Model Prediction
57.16% Actual Results
10 Method Communications
Analytics,
Modelling,
Strategy
Matthew Rattigan, an analyst on the team, built a tool for looking at the coverage of speeches
in local newspapers so it could break down by geographic region how people reacted and which
parts were quoted most. Speechwriters were therefore able to see how the messages they
wanted to convey were actually the ones that were covered."

Methods
Analytics,
Modelling,
Strategy
Hey
(The most successful subject line of the campaign, based on e-mail opens.)
Analytics,
Modelling,
Strategy
Driven by Romney's newfound fundraising advantage, the campaign's "I will be outspent" email
raised $2,673,278. This was the strongest of 13 test emails that day. If they had gone with the
lowest performing, they would have raised $2.2 million less.
Analytics,
Modelling,
Strategy
see image -->>
Source: “The Science Behind Those Obama Campaign E-Mails,” Bloomberg Businessweek
http://www.businessweek.com/articles/2012-11-29/the-science-behind-those-obama-
campaign-e-mails
Analytics,
Modelling,
Strategy
how OFA turned around initially weak fundraising and raised $1 billion.
1. Send A LOT more email than 2008 (At least 404 national fundraising e-mails in 2012)
2. Test everything.
3. Make people think they were going to lose.
Analytics,
Modelling,
Strategy
• 10,000 segments tested during the campaign
• Email Team with 18 staff
• Regularly tested as many as 18 variations on subject line and email copy
• Tip: Avoid “Frankenstein” emails that mix and match best subject with best copy
without testing the combination first
• Could see up to an 80% difference between versions
• Developed a “special sauce” for asking the optimal amount from
previous donors
• Could test with segments as small as 18,000 people
• They tested sending less email. It got “mixed results.”
• AND... little to no interference from campaign management on content
(!!!)
Analytics,
Modelling,
Strategy
• (Quik Donate) Raised $115 million - $75 million of which would not have been raised without
the program
Source: "Corporations Want Obama's Winning Formula,"
Bloomberg Businessweek
http://www.businessweek.com/articles/2012-11-21/corporations-want-obamas-winning-
formula
Analytics,
Modelling,
Strategy
• In-house system raised $250 million from 4,276,463 donations
Analytics,
Modelling,
Strategy
The campaign conducted 240 A/B tests on their donation page.
• This resulted in a 49% increase in their conversion rate.
• By making the platform 60% faster, they saw a 14% increase in donations. (Speed matters.)
Analytics,
Modelling,
Strategy
In June 2012, the campaign switched to the 4 step donation process and saw a 5% increase in
conversions (donations).
Analytics,
Modelling,
Strategy
see image -->>
Source: KyleRush.net http://kylerush.net/blog/optimization-at-the-obama-campaign-ab-
testing/
“Turns out you can get more users to the top of the mountain if you show them a gradual
incline instead of a steep slope.”
- Kyle Rush, Deputy Director of Frontend Web Development

Methods
10 Method Tools Built by the
Technology
Division,
Modelling,
Strategy
Narwhal: Synchronized data from multiple sources to build complete profiles of supporters
• Dashboard: Enabled supporters to connect with supporters near them and take action from
home
• Call Tool: Allowed supporters in nonbattleground states to use their home phones to call
voters in battleground states
• Stork: Transferred data from vendors to databases for querying 54
Analytics,
Strategy, Vote
• Dashboard was the campaign's grassroots organizing platform. Unlike 2008, it mapped directly
to how the campaign was structured in the Field
• Anyone who signed up for Dashboard was reached out to within 72 hours by field staff. After
72 hours, the likelihood that they would take action would drastically diminish
• Each of approximately 8,000 neighborhood teams had a presence on Dashboard
• Incorporated online data and offline, "hard" numbers from VAN
• Dashboard was the frontend for Narwhal
see images ->>
Analytics,
Strategy, Friends,
Vote
98% of the U.S. Facebook population was friends with someone who liked Barack Obama
Analytics,
Strategy, Friends,
Vote
• When it came to reaching the crucial 18-29 year old demographic, the Obama campaign came
to a startling realization
• 50% of their targets in this demographic were unreachable by phone
• But 85% of them were friends with an Obama 2012 Facebook app user
• OFA launched “targeted sharing” to Facebook friends who were voters in swing states
• Like Quick Donate, integration with the rest of the technology stack was key. Users received
an email requesting that they contact six specific friends, with their names and photos
• 600,000 people reached 5 million voters
• 20% of those 5 million took some action, such as registering to vote
Analytics,
Strategy, Friends,
Vote
Every time an individual volunteers to help out – for instance by offering to host a fundraising
party for the president – he or she will be asked to log onto the re-election website with their
Facebook credentials. That in turn will engage Facebook Connect, the digital interface that
shares a user's personal information with a third party. Consciously or otherwise, the individual
volunteer will be injecting ...crucially, network of friends – directly into the central Obama
database. "If you log in with Facebook, now the campaign has connected you with all your
relationships," a digital campaign organizer who has worked on behalf of Obama says.
Analytics,
Strategy, Friends,
Vote
The Obama database incorporates Vote Builder, a store of essential information such as age,
postal address, occupation and voting history drawn from the voter files of 190 million active
voters. It lines up and matches those voter files with data gathered from online interactions
with the president's supporters – notably the millions of pieces of information its army of
canvassers collected across the nation during the 2008 race, a list of email addresses of
supporters that it has amassed and that now stands at about 23 million, as well as the contact
information of Obama's 25 million Facebook fans.
Analytics,
Strategy, Friends,
Vote
"Influencers" – those people who tend to act as thought leaders among their friends on
Facebook – can be identified and prioritized. Teddy Goff, the digital director of the re-election
team, told Social Media Week that as the year progresses there would be more and more
"persuasion through interaction". Individual voters would be given access to digital platforms
from which they will be able to tell their own stories "and that's far more powerful than
anything we can say", Goff said. "That will be the story of this election. People's own stories
really moves votes."
Analytics,
Strategy, Friends,
Vote
if you are sent a message from your Facebook friend encouraging you to turn up to an event or
donate to Obama, you are vastly more likely to respond than if the request comes from an
anonymous campaign staffer.

Methods
12 Method Modelling Instead, they predicted persuasion:
• Who would be convinced to vote Obama if (and only if) contacted
This is the new microcosmic battleground of political campaigns - significantly more refined
than the ill-defined concept of "swing voter." Put another way, they predicted: For which voters
campaign contact would make a difference. Who is influenceable, susceptible to appeal? If a
constituent were already destined to vote for Obama, contact would be a waste. If an individual
was predicted as more likely swayed toward Obama by contact than not swayed at all, they
were added to the "to-contact" list. Finally, to top it off, if the voter was predicted to be
negatively influenced by a knock on the door - a backfired attempt to convince - he or she was
removed from the campaign volunteers' contact list and labeled: "Do-not-disturb!"
12 Method Metric they used the collected data not just to measure the overall effectiveness of campaigning, but
to predict the persuadability of individual swing state constituents. Each person got a score, and
the scores drove the army of volunteers' every move.
13 Method Persuasion
Modelling,
Strategy
A similar strategy of targeting an unexpected population emerged from a July EIP testing
Obama’s messages aimed at women. The voters most responsive to the campaign’s arguments
about equal-pay measures and women’s health, it found, were those whose likelihood of
supporting the president was scored at merely 20 and 40 percent. Those scores suggested that
they probably shared Republican attitudes; but here was one thing that could pull them to
Obama. As a result, when Obama unveiled a direct-mail track addressing only women’s issues, it
wasn’t to shore up interest among core parts of the Democratic coalition, but to reach over for
conservatives who were at odds with their party on gender concerns. “The whole goal of the
women’s track was to pick off votes for Romney,” says Walsh. “We were able to persuade
people who fell low on candidate support scores if we gave them a specific message.”
13 Method Persuasion
Modelling,
Strategy
Obama volunteers attempted 500,000 conversations with the goal of winning new supporters.
Voters who’d been randomly selected from a group identified as persuadable were polled after
a phone conversation that began with a volunteer reading from a script. “We definitely find
certain people moved more than other people,” says Bird. Analysts identified their attributes
and made them the core of a persuasion model that predicted, on a scale of 0 to 10, the
likelihood that a voter could be pulled in Obama’s direction after a single volunteer interaction.
The experiment also taught Obama’s field department about its volunteers. Those in California,
which had always had an exceptionally mature volunteer organization for a non-battleground
state, turned out to be especially persuasive: voters called by Californians, no matter what state
they were in themselves, were more likely to become Obama supporters.
13 Method Metric Lundry decided to focus on more manageable ways of measuring what he called the
information flow. His team converted topics of political communication into discrete units they
called “entities.” They initially classified 200 of them, including issues like the auto industry
bailout, controversies like the one surrounding federal funding for the solar power company
Solyndra, and catchphrases like “the war on women.” When a new concept (such as Obama’s
offhand remark, during a speech about our common dependence on infrastructure that “you
didn’t build that”) emerged as part of the election-year lexicon, the analysts added it to the list.
They tracked each entity on the National Dialogue Monitor, TargetPoint’s system for measuring
the frequency and tone with which certain topics are mentioned across all media. TargetPoint
also integrated content collected from newspaper websites and closed-caption transcripts of
broadcast programs. Lundry’s team aimed to examine how every entity fared over time in each
of two categories: the informal sphere of social media, especially Twitter, and the journalistic
product that campaigns call earned press coverage. Ultimately, Lundry wanted to assess the
impact that each type of public attention had on what mattered most to them: Romney’s
position in the horse race. He turned to vector autoregression models, which equities traders
use to isolate the influence of single variables on market movements. In this case, Lundry’s
team looked for patterns in the relationship between the National Dialogue Monitor’s data and
Romney’s numbers in Gallup’s daily tracking
13 Method Within three or four days of a new entity’s entry into the conversation, either through paid ads
or through the news cycle, it was possible to make a well-informed hypothesis about whether
the topic was likely to win media attention by tracking whether it generated Twitter chatter.
That informal conversation among political-class elites typically led to traditional print or
broadcast press coverage one to two days later, and that, in turn, might have an impact on the
horse race. “We saw this process over and over again,” says Lundry.

Methods
13 Method e-mail blasts asking people to volunteer could take their past donation history into
consideration, and the algorithms determining how much a supporter would be asked to
contribute could be shaped by knowledge about his or her reaction to previous solicitations.
This integration enriched a technique, common in website development that Obama’s online
fund-raising efforts had used to good effect in 2008: the A/B test, in which users are randomly
directed to different versions of a thing and their responses are compared. Now analysts could
leverage personal data to identify the attributes of those who responded, and use that
knowledge to refine subsequent appeals.
13 Method Under a $350,000 deal she worked out with one company, Rentrak, the campaign provided a
list of persuadable voters and their addresses, derived from its microtargeting models, and the
company looked for them in the cable providers’ billing files. When a record matched, Rentrak
would issue it a unique household ID that identified viewing data from a single set-top box but
masked any personally identifiable information.
13 The Obama campaign had created its own television ratings system, a kind of Nielsen in which
the only viewers who mattered were those not yet fully committed to a presidential candidate.
But Davidsen had to get the information into a practical form by early May, when Obama
strategists planned to start running their anti-Romney ads. She oversaw the development of a
software platform the Obama staff called the Optimizer, which broke the day into 96 quarter-
hour segments and assessed which time slots across 60 channels offered the greatest number
of persuadable targets per dollar. (By September, she had unlocked an even richer trove of
data: a cable system in Toledo, Ohio, that tracked viewers’ tuner histories by the second.)

Cloudcroft-S Persuasion and the Presidency PREDICT402-56 W2015

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Similar to Cloudcroft-S Persuasion and the Presidency PREDICT402-56 W2015

Similar to Cloudcroft-S Persuasion and the Presidency PREDICT402-56 W2015 (20)

Cloudcroft-S Persuasion and the Presidency PREDICT402-56 W2015