Obama campaign’s investment in data crunching paid off
CHICAGO — Early on election day, in two tight, tucked-away rooms at Obama headquarters known as the Cave and the Alley, the campaign’s data-crunching team awaited the nation’s first results, from Dixville Notch, a New Hampshire hamlet that traditionally votes at midnight.
Dixville Notch split 5-5. It did not seem an auspicious outcome for the president.
But for the math geeks and data wizards who spent more than a year devising sophisticated models to predict which voters would back the president, Dixville Notch was a victory. Their model had gotten it right, predicting that about 50% of the village’s voters were likely to support President Obama.
Daniel Wagner, the 29-year-old chief analytics officer, erupted in joy. The model was also projecting that Obama would be reelected. And as the night wore on, swing state after swing state came in with results that were very close to the model’s prediction.
For the nation, last Tuesday was election day. For Wagner’s crew, it’s now known as Model Validation Day.
“We’re kind of a weird bunch of kids,” he said, standing near the Cave, where one wall was covered with a large canvas of a Martian landscape. “I haven’t seen the sun in a while. We worked brutally inhuman hours this cycle. Twenty-hour days, often. But they bet a lot on us being right. And it was good to be right.”
For years, campaigns have used reams of information to predict voter behavior, relying on a science known as analytics. But Obama’s advisors elevated the practice to new heights, very likely changing the way presidential campaigns will be conducted in the future.
No other presidential campaign has so completely embraced this science. The campaign hired a team that topped out at 54 people and invested undisclosed millions in the effort. Analytics helped the campaign efficiently recruit volunteers, buy ads, tailor emails and mailers, raise money, dispatch surrogates — and, most importantly, scour the swing states for hard-to-find voters most likely to support the president.
Political guru David Plouffe and campaign manager Jim Messina made key decisions based on real-time reports from the geek squad, according to many people on the campaign’s staff.
“Our entire goal is to make the maximum use of our time and our volunteers’ time. And that means using analytics across the campaign spectrum,” Messina said after the election. “We invested unprecedented resources to do this because our entire theory was to get as micro-targeted — to get as close to the ground — as we could.”
Another campaign official, who asked not to be named because he was not authorized to speak for the campaign, put it this way: “It’s about turning over control to some nerds. And more than any other year, campaign leadership really took that leap of faith.”
For campaign professionals, that is a major leap. Politics long has been ruled by truisms, conventional wisdom and intuition, with millions spent based on a murky mix of polling and focus groups. The shift to data-driven decision-making has been gradual and steady — becoming increasingly sophisticated as political parties amass more information about individual voters through traditional means, such as polls, and new ones, such as data mining.
The Obama campaign has made the transition over two elections. In this one, it employed analytics in a far more systematic and thorough way, officials said. But the work was a closely guarded secret. Officials denied requests for interviews with the analytics experts, and when journalists visited Obama headquarters, the team was ordered to shut the Cave door.
Victory opened that door — a crack.
At its most basic, Messina, Wagner and others explained, the goal was to rank individual voters in the swing states based on their likelihood of voting for the president or of being persuaded to vote for him, to volunteer for his campaign and to vote early. The Obama campaign developed separate models for each.
To build the “support model,” the campaign in 2011 made thousands of calls to voters — 5,000 to 10,000 in individual states, tens of thousands nationwide — to find out whether they backed the president. Then it analyzed what those voters had in common. More than 80 different pieces of information were factored in — including age, gender, voting history, home ownership and magazine subscriptions.
Much has been made of the last type of data — the tastes or trends that divide Democrats and Republicans. But Obama officials downplayed the impact of consumer habits, claiming the more important information is the most basic and publicly available.
It’s not about subscriptions to “Cat Fancy” magazine, Wagner joked as he scribbled on a white board, blinking through his rectangular glasses. The most important data, he said, was the Democratic National Committee’s database, containing voting history and demographic information, as well as feedback from contacts with individual voters going back to 1992, such as whether a voter warmly received a door-to-door canvasser or shut the door curtly.
The campaign compared the data on individual voters with the “support model” and ranked voters from 1 to 100, which told the campaign where to focus its turnout efforts.
The notion of a campaign looking for groups such as “soccer moms” or “waitress moms” to convert is outdated. Campaigns can now pinpoint individual swing voters. “White suburban women? They’re not all the same. The Latino community is very diverse with very different interests,” Wagner said. “What the data permits you to do is to figure out that diversity.”
One of the campaign’s most powerful tools was its “persuasion model,” which sifted through millions of voters that the DNC had labeled as not very partisan to find those most likely to be won over to Obama’s side.
In Obama’s 2008 campaign, as in most campaigns, these “middle partisans” were assumed to be swing voters and were heavily targeted. But in reality, many are partisan conservatives, and many more don’t vote.
So early this year, the Obama campaign called more than 10,000 voters in the category and talked to them about the president’s views on healthcare and taxes, the official said. Then it called those voters back a few days later to find out if their opinions had shifted. The campaign analyzed those who had moved toward Obama to find out what they had in common and, from those results, created a separate model of persuadable voters for each swing state.
A data director applied the model to the voter databases and generated lists of voters to be contacted. Those were put in the hands of canvassers who were also armed with a script tailored to an individual voter’s pet issues.
The Obama campaign even found voters to target in ruby-red precincts, a break from earlier campaigns when solidly partisan precincts were simply written off.
Among its many decisions driven by data, the campaign chose to stick it out in Florida, even though polls and conventional wisdom raised doubts about Obama’s odds in the GOP-tilted battleground.
Just weeks before the election, the analytics team’s assessment suggested a 30% to 40% chance of winning the state, Wagner said. But after the team added information about roughly 250,000 new voter registrations, the projection shifted, showing that 80% of the new registrants would vote and they would heavily support Obama.
When the computers spat out this data, indicating that Obama was likely to win in Florida, a howl went up from the Cave. A mathematician from the University of Alabama started it off with the ‘Bama fighting words, “Roll Tide!”
Two days before the election, the president went to Florida to bolster that turnout — and ended up narrowly winning the state.
Days after election day, Obama campaign workers were packing up, pulling flags off the ceiling and pictures off the wall.
But the Cave still hummed with computers. The decor was still up, including the landscape of Mars, its presence explained by a printout taped next to it. In a July 2011 Wall Street Journal column, Peggy Noonan, a conservative author who wrote speeches for President Reagan, mocked an Obama campaign hiring notice for specialists in “predictive modeling/data mining.”
“It read like politics as done by Martians,” she wrote.
With the Martian scene behind him, Andrew Claster, an economist and deputy chief, pored over exit polls and early vote data. Nearby, Rayid Ghani, the team’s chief data scientist, tried to discern more about how to motivate people online. It’s part of an ongoing effort to mine the data for future Democratic campaigns and causes.
“What we’ve had since ’06 and especially since ’08,” Wagner said, “is a change in how people think about this information and how they use it.”
In other words, the Martians have landed.