Going Beyond the Boundaries of the Obama Digital Model in Virginia
BY Miranda Neubauer | Monday, November 11 2013
A key element in the Democratic win in Virginia this week was the ability to adapt Obama campaign capabilities to integrate historic voter file data and recent data from field organizers and partner groups with analytics capabilities to develop targeting models, representatives from NGP VAN, Blue Labs and the Virginia Democratic Party said in a press call Friday.
techPresident reported earlier this week how NGP VAN and Bully Pulpit Interactive helped shape the Democratic victories in Virginia, New York City and Boston.
In June, Tom Edsall had written for the New York Times how the Virginia campaign would be a "unique proving ground for two of the most significant developments in contemporary politics — Republican intra-party schism and the critical role of campaign technology ... On the Democratic side, the Virginia election will be the first off-year test of Obama campaign techniques without Obama actually on the ballot." He specifically highlighted the McAuliffe campaign's choice of Michael Halle, the 2012 North Carolina election director for Obama, to help coordinate election activities with the Virginia Democratic Party and other local Democratic and progressive groups.
Using the NGP VAN tool to coordinate with multiple groups was one of the "most impactful" factors, said Halle, who was the Democratic Party's coordinated campaign director in Virginia in 2013. In the first meetings with partner groups like Planned Parenthood and the League of Conservation Voters, "we decided to make sure that everyone was using the same data and working out of the same database," he said, allowing the groups to feed the supporter data they were gathering into one file. "Part of the appeal for groups was that not only did they have access to their own information, but they also had access to the information from all of the other groups that were participating as well."
According to Halle, that was a capability the Republicans "weren't even close" to having. He suggested that even if they had "trusted each other enough" to have that cooperation they would not have access to the same kind of platform.
He also emphasized the flexibility of the tools, allowing some groups to use the iPhone app MiniVAN, while others could still participate in the effort using paper.
The joint database effort enabled the groups to coordinate direct mail messages to voters, he stressed, using "experiment informed programs," a tactic of in effect A/B testing campaign messages that was also a key element of the Obama campaign. The effort involved testing three different direct mail messages, and then building models based on how voters responded to those messages, he explained. "As we moved into the fall, each group knew when they were dropping their pieces of mail," he said. "When we talked with people who were receiving Republican mail, there were several days in the final two weeks of the campaign where they received three, four, five pieces of mail on the same day. Obviously that becomes really ineffective ... Our same voters that we were targeting would receive a new piece of mail every two days with absolutely no overlap."
Many of the Republican efforts were counterproductive and harmful, he said. "We knew that the mail worked with the people we were mailing," he said, as the groups made sure not to talk to voters who might shift in the other direction because mail or ads "turned them off." He characterized several Cuccinelli ads as "largely unhelpful," coupled with several ads from independent group with messages such as "Don't let THEM Detroit Virginia," and suggested they didn't have access to the data that would indicate to them that those ads "were a silly thing to air."
Elan Kriegel, co-founder of Blue Labs, said working with NGP VAN helped McAuliffe "run the most innovative statewide campaign that there has ever been."
As Blue Labs looked to build on the analytical work done for the Obama campaign with campaigns like the one in Virginia, "we were going beyond the boundaries of what we did with Obama," Kriegel said. A main focus was on building models of voters' likelihood of support, likelihood to vote, likelihood to care about certain issues and likelihood of voters to change their mind or to be persuaded on the phone, he said. The models were based on not just the historic data in the VAN database from previous elections, but also data constantly coming in during the campaign, he explained, creating a "constant feedback loop" between the field and the central campaign.
"People in the field are interacting with voters and recording their interactions in VAN, what people care about, what they're feeling about the race," he said. "That data then feeds from VAN into our models...we put those models back in the VAN, and now those same volunteers have a new list of people to talk to and are targeting people differently and then after those interactions they are going to put that data back in the VAN and then we're building new models." The "huge advantage", Kriegel said, is that the extensive data helps volunteers be more efficient by being able to focus on contacting voters who might be home, who intend to vote and are interested in having a conversation. "Rather than knocking on every single door, they can knock on two or three doors and have real conversations," he noted.
"The same door knocks that were happening in 2008 and 2012, and that voter information, all feed into building these larger models," Halle added. Often, he noted, the assumption is that turnout efforts should focus on people who voted in federal elections but might not turn out in off-year elections, he said. "The reality is that in the 2009 election, 32 percent of the people who voted in '05, did not vote in '09," he said. "So there is a huge group of people that do vote in off-year elections that you still need to make sure you are turning out and having a conversation with." In past elections, many Republicans and Democrats have only focused on 2008 or 2012 voters that did not vote in off-year races, "and there's no precision beyond that," he said.
"What the voter file and what Elan's turn-out model allow us to do is look at more variables than just past vote history. We were able to look at the voting history of other people in the household," he said. "We're able to look at income of the household and income in the precinct that they live in, college education. ... which allows us to determine who are those '09 people that we need to make sure we're going back to. No one else was talking to them about voting because the assumption is they are going to vote because they voted before which is clearly a false assumption. And then there's a group of 2012 people who didn't vote in '09 that are far more likely to turn out with a contact than another group that are never going to come out in the off year."
Through a state party voter file swap, the data used in Virginia is also part of the larger effort by the Democratic National Committee to build a national voter file, said NGP VAN COO Brian Whitaker.
Halle said the file was "the core dataset that drove communication across the campaign," including online advertising and targeting.
Kriegel pointed out that it was especially helpful to have Harrison Kreisberg, a Blue Labs data and analytics director, embedded with the McAuliffe campaign, allowing for a smoother, more effective collaboration.
Blue Labs also used its models to run an effort called Airwolf that originated with the Obama campaign to help bring together offline and online campaign interactions, Kriegel and Chris Wegrzyn, the other Blue Labs co-founder, noted. The program helped connect people who expressed interest in volunteering over phone calls or who signed up after being driven to the website through online advertising with the relevant field organizer, Wegrzyn explained. He called it a "data integration project" that took "new volunteers coming into the system," found "which field organizer is assigned to that person," and then sent a customized e-mail on behalf of that field organizer to the volunteers. "We can facilitate a fast response, take some of the load off of the field organizers but still create that personal connection."