Driving Campaigns with Data
Tufts alumnus Daniel Scarvalone credits his experience in a Tisch College course taught by Alan Solomont with setting him on a path to working for the Obama campaign.
Daniel Scarvalone, A08, is working at the cutting-edge of big data and political campaigns. National Reporting Director for the Obama 2012 campaign, Scarvalone helped manage and analyze the masses of information the campaign used to target voters and evaluate the campaign’s state programs. Now the Director of Data and Modeling for the Democratic Congressional Campaign Committee, he is bringing that experience and expertise to upcoming races for the House.
Scarvalone sees this work as deeply connected to active citizenship.
“The most basic, fundamental thing about a campaign is having conversations with human beings about the issues that matter to their lives,” he said. “This is how I use my voice, and can spend my energy, time, and effort improving the lives of people and advocating on behalf of the policies I care about. Politics takes a long time, but I believe in doing whatever I can in the time I have.”
Scarvalone arrived at Tufts with interests in both politics and medicine. He credits a course sponsored by Tisch College as having a decisive influence.
“Decision 2008 was a senior seminar led by Alan Solomont,” said Scarvalone. Solomont, A70, was Tisch College’s founding board chair and currently serves as the Unites States Ambassador to Spain and Andorra; he recently spoke on-campus about his experiences. “He had some of his friends in as guest speakers, Howard Dean, Richard Gephardt, Sandy Berger, people who had spent their lives in politics. The way they talked about democracy and getting involved in the political process, for me that was the transition from learning about it in class to really wanting to go out and do it.”
Scarvalone began spending weekends canvassing for the Obama 2008 campaign in New Hampshire. After graduating, he moved to Colorado as a “summer fellow,” a volunteer field organizer in and around Denver. Six weeks in he was hired full-time, and spent the rest of the campaign in the state which Obama would go on to win by nearly 9 points, the first Democrat to carry Colorado since 1992. Scarvalone returned there in 2010, working on the successful re-election of Senator Michael Bennet in what would prove to be one of the closest and costliest senate races in the country.
“We used a lot of data and a lot of micro-targeting to precisely identify our potential voters and get them to turn out,” said Scarvalone. “Colorado is special in that 85% of voters vote by mail before election day. We developed measures that told us more or less in real-time whether we were winning or not. We also know from past elections, many Republicans vote early while more Democrats vote on election day. So we knew that as long as the gap in turnout between voters who were likely to support us and those who were likely to support the other guy was under a certain threshold ahead of election day, we were going to win. In the end it was slim but enough.”
To judge whether a voter is likely to support a candidate, campaigns use computer models to score each person in their database on different dimensions. To create the scores, modelers pull in data from publicly available sources, such as the Secretary of State’s office in each state. Depending on the state, this can include not only the voter’s name and registration status, but also their party affiliation, history of elections they’ve voted in, gender, date of birth, and other characteristics. This information is combined with consumer data purchased from private vendors, which can span everything from phone number, magazine subscriptions, favorite kinds of yogurt, and pet ownership, to reported income, race and ethnicity, and religious affiliation.
Combined in massive databases, detailed algorithms compare the characteristics of supporters and non-supporters to create predictions on the individual level. Campaigns can then use those predictions to make lists that determine their tactics and strategy to getting to 51% (or more) on election night.
“Campaigns are all about lists,” Scarvalone said. “From a campaign point of view, most voters can be sorted into one of five buckets.”
Those five buckets are based on the voters’ inclinations and their likelihood to turn out. The first two are the habitual voters, people who are basically certain to vote no matter what, either for or against you. The next two are the marginal voters, who may or may not vote at all, but if they do vote will lean one way or another. Last are the persuadables or swing voters, who are likely to vote, but could go either way.
The key to winning, says Scarvalone, is identifying which marginal voters can be most influenced by your campaign’s mail, TV, and field programs – and then getting them to vote for your candidate through targeted outreach. Each contact with a voter, be it a phone call or an in-person conversation, is also entered into the database and affects that voter’s predicted behavior. And the data doesn’t stop there.
“The Obama 2012 campaign evaluated every part of its program on commonly standardized goals,” said Scarvalone. “Every day for the entire election, we generated a top-line report for senior staff in Chicago. It showed for each state, county, and local organizing turf, how many attempts had been made to reach people at their doors and on their phones, how many of those yielded actual conversations, and how that compared to the goals we’d set. Every state’s program was evaluated based on a strategic analysis of how many attempts it would take to get the number of conversations needed to get the turnout to win.”