With campaign season underway, we’re inundated with stories about how the candidates are using data to target voters. Just this week, Politico reported on how the Ted Cruz campaign is relying on a donor’s firm to develop “psychographic” profiles of voters.
Yale political scientist Eitan Hersh has a new book, “Hacking the Electorate,” that delves into the real world of campaign targeting. It is itself deeply informed by data, but arrives at some conclusions starkly at odds with of the conventional wisdom about campaign targeting. Hersh answered some questions via e-mail. Below is our lightly edited exchange.
Q: We’re used to hearing how political campaigns have this amazing data on everyone — right down to the cars you drive or the magazines you read. CNN once said that “campaigns know you better than you know yourself.” Your book pushes back against this notion. Why is it wrong?
Campaign spokespeople are not generally in the business of being modest about their technological capabilities, and reporters usually want to tell a story of cutting-edge campaign sophistication. So the image we get in the news is that campaigns “know you better than you know yourself.”
But even with all the technological advancements of the last decade, it’s still difficult for campaigns to predict the preferences of individual voters. Can the presidential primary contenders accurately tell which voters are in their camp? In the general election, will candidates be able to predict which voters are persuadable? The answer to both questions is basically no.
This is not because the campaigns are unsophisticated. It’s because these are really hard questions to answer. Voters are complicated, and the data that campaigns have at their disposal allow them to make only rough approximations of voters’ attitudes.
We don’t like the idea that marketing gurus or statisticians are working behind the scenes to anticipate our political opinions and and move us according to their wishes. Thankfully, that image is off base. The wizard behind the curtain just isn’t that powerful.
Q: So for the sake of predicting people’s political opinions, there’s not much value in knowing what cars we drive?
Knowing what car you drive might have value, but it probably doesn’t have much added value. Here’s why. For many voters, it’s straightforward to determine whether they are likely to be Democrats or Republicans, or voters or likely abstainers — and without any help from consumer information. Before knowing what car you drive, campaigns already have obtained data from the Census Bureau about all sorts of characteristics of the neighborhood and town that you live in. From your voter registration, they also know your gender, age, and maybe race and party affiliation.
On top of this information, the car you drive doesn’t provide much new insight. If a campaign knows that you live in a highly educated neighborhood in the center of a college town and that you voted in the last five Democratic primaries, it doesn’t need to know that you own a Prius to peg you as a Democrat.
Moreover, for the more complicated dispositions that campaigns care about — like how persuadable you are or who you will support in a primary — consumer data are generally irrelevant. Americans are politically divided, but those divisions are much more apparent when voters are segmented by geography, race, age or gender than when they are segmented by the cars they drive or the food they eat.
Q: So what is the really valuable information that campaigns use in targeting, and how does it matter?
By far, the most valuable information that campaigns use in targeting is not data about consumer habits, but data that comes from public records, like the voter registration system, state licensing agencies, and from the Census. Almost everything campaigns know about voters comes from data produced by states and the federal government.
What this means is that campaign databases are a direct consequence of public policies. Laws about data are what create these databases and make them useful to campaigns. Especially interesting is that states differ in the data they produce about the public. In some states, registered voter must declare their racial identity, which becomes a public record. But most states don’t do that. In some states, voters register with a political party, and this becomes a public record. But other states don’t require that.
In my book, I show how campaigns perceive voters differently depending on whether they can obtain key public records from the state. For example, in states that collect partisanship data from voters and distribute that data to campaigns, campaigns focus more on mobilizing partisan supporters and less on persuading undecided voters than in states that don’t collect the data.
Furthermore, in states that don’t collect partisanship data, campaigns have a lot more accidental contact with voters of the opposite party. That’s because they don’t have as good of a sense of which voters are with them or against them. These are consequences of government policy about the collection and distribution of personal data.
Q: You also talk about the value of knowing voters’ racial identity. What impact does this have on targeting?
In a state like North Carolina that collects racial data from voters, campaigns are much more likely to mobilize voters on the basis of race. In a state like Virginia that doesn’t collect racial information, but that’s otherwise similar to North Carolina, campaigns focus less on race. A Democratic campaign looking to mobilize African Americans in Virginia must focus on overwhelmingly black neighborhoods, not on individual black voters. It will largely ignore black voters who live in mixed-race neighborhoods.
Q: While commentators often complain about how campaigns violate privacy by collecting all of this data on voters, you raise a different concern: politicians’ conflict of interest. What is the issue here?
Once we see the close connection between data policies and campaign strategy, it is easy to understand why we ought to be nervous about conflicts of interest. Politicians are responsible for deciding what kinds of personal data should be collected by the government for administrative purposes and the terms under which that information should be transmitted to third parties like campaigns. These politicians are obviously aware that how administrative data can help them and their allies.
So the conflict of interest takes two forms. First, politicians, ostensibly acting as policymakers, may decide to collect personal data from the public that has little or no administrative value, but that is mainly useful to political campaigns. In the book, I discuss a number of state policymakers who were clearly thinking about campaigning when they introduced legislation calling for the collection of personal data.
Second, official government offices sometimes use campaign-style microtargeting strategies in their interactions with constituents.
That’s something I’ll be writing more about in the coming weeks. Stay tuned.