Buzzword: Modeling

Let’s say you’re developing your campaign’s get out the vote strategy. Your team breaks down the voter file by election history: Voters who have cast ballots in each of the last four elections; voters who have cast absentee ballots; and voters who have turned out for similar elections to the one you’re in.

But that only yields more questions. What, for example, about voters with no voter history? Are there groups within that block who are more likely to vote than others? Would some of those be more receptive to an absentee ballot program? Are there individuals who voted in two of the last four elections but are more likely to vote this time than those who voted in three of the last four?

The answer to all of the above may be voter modeling—a term that came into vogue with President Barack Obama’s 2008 campaign and has dominated strategy discussions since. Think of it as the next phase of microtargeting.

While modeling has been around for years—in the commercial and political worlds—new technologies are allowing campaigns to analyze the electorate with stunning accuracy. “Now we’re at the point where people are saying, ‘What kind of habits does a person have, who fits a statistical modeling profile and what are they most likely to be most responsive to?’” says Democratic consultant Michael Bronstein of Bronstein and Weaver. “Those are very different questions.”

The modeling of old focused on the macro level, such as developing win scenarios and turnout goals. New modeling operates on the micro level, developing ways to predict attitudes, behaviors and characteristics of individuals far in advance of Election Day.

“Integrating data and the tools to understand what the data is telling you and to then take action on that information is where campaigns are heading for 2010 and 2012,” says John Phillips, CEO of Aristotle, the preeminent data collection firm. “Despite all the things that campaigns did right in 2008, it is hard to find a campaign that won’t tell you it was harder to model than it should have been.”

Joel Rivlin, the director of analytics at MSHC Partners, says the old axioms, such as basing efforts on how many previous elections in which a voter participated, are quickly becoming things of the past. By combining countless sets of information such as census data, voter files, demographics and survey results, Rivlin creates detailed models that try to predict the attitudes and behavior of individual voters.
Rivlin believes these models are the future of political campaigning.

“You can find those people who may, on the surface, look less likely to vote but are more likely to vote,” Rivlin says in praise of modeling. “And you can look at people who don’t have a vote history and work out which of those folks are more likely to vote than others.”

Rivlin is only one of several consultants developing technologies that are the next wave in how campaigns analyze and interact with existing data. In its early days, modeling used data to give campaigns a sense of who is in the electorate. The new technologies are allowing campaigns to develop a much more vivid picture and interact with that data to target voters for messaging, fundraising and, perhaps most effectively, get out the vote efforts.

Public information is combined with other databases and surveys that can be used to extrapolate how similar voters will behave. For example, to predict how receptive a voter would be to an absentee ballot program if voter history isn’t available, Alex Lundry, a Republican consultant at TargetPoint, looks at several data points.

“Things like age make a difference,” he says. “Also whether someone is homebound. We’re now seeing a rise in convenience absentee voting, so you look to see if the voter lives a more frenetic lifestyle. For example, how many kids do they have?”

That data is gathered, layered together and the individual is given a score of 1 to 100 for how receptive he or she would be to absentee balloting. Then analysts like Rivlin and Lundry extrapolate from that one voter to predict the behavior of others who share similar characteristics. “If you give someone a score of 70,” says Rivlin, “that means that if you find 100 people with those characteristics, 70 will have that attribute.”

These attitudinal models apply to other areas of the campaign, as well. Rivlin, for example, says he is developing models that predict how receptive individuals will be to advocacy solicitation. Others have models to predict whether an individual is inclined to take action on specific issues such as healthcare.

Still others are developing even more personalized ways to use existing data. Phillips says having the data is “just one leg on the stool,” and his company, Aristotle, has created the “Relationship Viewer” to “take a deep dive” into the data. The Relationship Viewer draws connections between individuals and whom they interact with, such as coworkers, neighbors, fellow board members and others. Much like the famous “Six Degrees of Kevin Bacon” game, the Relationship Viewer has the potential to allow campaigns to contact a targeted individual via a volunteer who best knows him or her. Models and technologies like the Relationship Viewer are only as good at the data they use and there have been developments on that front as well.

Catalist, a Democratic data firm led by Harold Ickes, has developed a data consortium that is updated by liberal groups with information from their voter contacts. The result is a model that gives every person over 18 in the country a 1 to 100 score on the likelihood he or she will turnout in an election.

“We built probably the first ever national turnout model that was used heavily by organizations working with us in the 2008 cycle,” says Laura Quinn, Catalist’s CEO.
Catalist has also developed models on what kind of campaigning a voter may be receptive to. Quinn calls these “channel responses,” and they predict whether a voter is more likely to be influenced by a phone call, an Internet ad or a television commercial. Catalist has also modeled whether an individual is likely to be an activist on issues such as healthcare or the environment.

When asked about the influence of their work, Quinn says that these models, with their 1 to 100 scores, are changing all aspects of campaigns. “It creates a big shift,” she says. “Campaigns were generalizing the message for large swaths of people. Now campaigns are specializing the message for smaller swaths of people.”

Jeremy P. Jacobs is the staff writer for Politics magazine.