Analyzing voter data has become a flourishing business in Washington. Democratic campaigns across the country access a voter file in the VoteBuilder software provided by NGP VAN. Progressive organizations use another database, called Catalist, for general campaigning. The Obama campaign’s top analysts have gone on to form new ventures like BlueLabs, which uses data to find voters and donors, and Civis Analytics, which helps organizations use big data to make decisions. TargetPoint Consulting has helped conservative campaigns target voters at the individual level for more than a decade, and the Republican National Committee recently formed its own data analysis startup, Para Bellum Labs.
Data analysis is valuable for mobilizing likely voters with reminders to go to the polls. But it is harder to use the data that campaigns gather to accurately target persuasive messages to people who are undecided or might change their minds, says Eitan Hersh, an assistant professor of political science at Yale and author of the upcoming book Hacking the Electorate: How Campaigns Perceive Voters. In a conversation with MIT Technology Review special projects editor Kristin Majcher, he explains the challenges of accurately predicting which voters will be persuadable—something that can be important in close elections.
You argue that it’s not changes in technology alone that are affecting how campaigns interact with voters, but technology combined with state public-records laws that determine what kind of voter information they have access to.
The law is generating all sorts of different data rules and data access points, and when you change them, you change how campaigns behave. So things like states’ Freedom of Information Acts or open-records laws, or state voter registration [data] policies—we don’t think of them as being so integral to campaign strategy or grassroots mobilization. But these data policies that nobody’s paying attention to have a huge impact on how campaigns decide how they’re going to engage with voters.
Overall, have campaigns become more sophisticated?
On the one hand, campaigns now have made big advances over campaigns, let’s say, 20 years ago. On mobilization, campaigns know a lot more. There’s just so much research about how to do that right. That’s how we know, for example, that neighbors are better at mobilizing people than non-neighbors. On persuasion, I would say that campaigns know very little. All of the research seems to be really context-dependent. Like this little novel trick worked right here at this moment in this kind of campaign, but then it was tried again in a different environment and it didn’t work at all.
Why is it so difficult to figure out what makes voters persuadable?
It’s funny—sometimes before an election, pundits say, “Who can possibly be undecided about this?” But, you know, who is? It’s hard to find [undecided people], and then if campaigns find them it’s hard to convince them, and once they try to convince them it’s hard to measure whether they’ve succeeded.
Modeling persuasion is just really hard. Even the fanciest microtargeting model of persuasion can’t really do a good job … If you think about what kind of data campaigns have about voters—whether it’s their party affiliation, their age, their gender, the kind of neighborhood they live in—none of those variables are really predictive of persuadability, and it’s not because the campaigns are doing anything wrong. It’s because persuadability is a psychological disposition that is really hard to predict.
The Obama campaign famously tried to use Facebook to win over voters. Is persuading friends and family effective?
The problem with persuading your friends is—no one wants to do it! The kinds of people who want to volunteer for campaigns, they like having the camaraderie of meeting some other activists that want to do this, and then going to strangers’ houses and talking to strangers. That is much more appealing than calling their uncle and trying to convince him to vote for someone that he doesn’t want to vote for.
Predicting persuasion is hard, but campaigns are still using data-mining techniques to try to figure it out. Are they just guessing?
Campaigns gather as much data as they can about voters—from neighborhood statistics, individual-level data coming from the commercial world, and from governmental sources, data from prior campaigns. These data are like little hints about a voter’s persuadability. But the hints do not collectively add up to a very accurate picture of persuadable voters.
So does that mean all of this data analysis we hear about is overhyped?
The Obama campaign was employing social scientists and data scientists and doing lots of experiments, and that is all new and deserving of attention because it’s really interesting, and it’s very different from what campaigns have been doing in the past. Where I think the hype comes in is in overpromising what this kind of technology can do. It’s clearly not a secret recipe to figure out who’s persuadable.