Data-Mining Your Psyche
When a state representative named Thom Tillis ran for U.S. Senate in North Carolina in 2014, his campaign followed the now-standard practice of sending voters online and direct-mail advertisements referring to particular issues. Which issues mattered to which people, from ISIS to the Affordable Care Act, could be gleaned from the voters’ memberships and donations or inferred from demographic information and databases of everything from their purchases to their Web history.
- Behavioral Microtargeting System
But some of Tillis’s advertisements tried something new. One, which showed Tillis smiling broadly with a soft-focus background of green foliage, promised he’d “restore common sense in Washington.” Another, featuring a man wearing a hard hat while poring over blueprints with his team, asserted that Tillis had “the experience to get the economy working.” A third showed the camouflage-smeared face of a soldier and contained this promise from Tillis: “Your safety is his top priority.”
Which version went to whom depended on how the Tillis campaign assessed the recipient’s personality. The grinning picture? That went to people who tended to be “agreeable.” Hard hat? That one was aimed at people deemed “conscientious.” Soldier? That one was reserved for people thought “neurotic.”
The Tillis campaign made the ads with help from Cambridge Analytica—an offshoot of a British firm called SCL Group, which has long been involved in campaigns (commercial, political, military) that leverage behavioral-science research. Like other big-data analysis companies, it categorizes voters on the basis of demographics and issues, but it appears to be the first to add personality typing to the mix. The company says it has assessed the personalities of all 190 million registered voters in the United States. “We can now start to really look at people based on their psychographics, not just their demographics,” says Alex Tayler, chief data officer at Cambridge Analytica’s headquarters in London.
It’s a fairly new approach, and no published data shows that what Cambridge Analytica is doing adds any real value. “Many basic categories of data tend to be very predictive in terms of voting habits: voting history, party registration, age, gender, race, marital status, presence of children, and some basic census measures,” says Daniel Kreiss, a professor of political communication at the University of North Carolina, Chapel Hill. “Everything else gives you only marginal potential advantages. Is personality the strongest predictor of whether someone is persuadable versus any of these other things? I haven’t seen any data that has suggested that this form of modeling does or doesn’t work.”
The “Big Five” Traits
If you rank high in …
- Openness You are intellectually curious and tend to think and act in individualistic and nonconforming ways.
- Conscientiousness You set clear goals and are reliable, organized, and self-controlled.
- Extroversion You enjoy being with other people, are full of energy, and tend to seek out opportunities for excitement.
- Agreeableness You value getting along with others and believe people are basically honest, decent, and trustworthy.
- Neuroticism You tend to experience negative feelings and to interpret ordinary situations as threatening.
In fact, there are large gaps in our understanding about many kinds of data-driven campaigning, says Gregory Huber, a political scientist at Yale. In 2012, President Obama’s campaign operatives analyzed data about individual voters’ issue preferences and demographics to craft messages tailored to them (see “A More Perfect Union,” January/February 2013). Obama won reëlection, but did the microtargeted persuasion efforts have anything to do with it? “There is almost nothing known about whether that stuff worked or not,” says Huber. “There are books written that are descriptive exercises of what the campaign was doing. But in terms of a valid, unbiased, randomized assessment of its effectiveness, I don’t think we know much at all.”
Nonetheless, academic research has established a relationship between personality and political leanings—and has found that some types of personalities are more persuadable than others. Such basic research doesn’t necessarily translate into a practical campaign strategy, but it does suggest that personality typing might provide something that demographics alone can’t. And it is tantalizing enough to persuade some campaigns to give Cambridge Analytica’s tools a try. The company, which represents Republican and conservative candidates, worked on 44 state and federal races in the United States in 2014. This year it has worked for the presidential campaign of Ted Cruz, the U.S. senator from Texas.
Are you agreeable?
Cambridge Analytica’s technology attempts to define a voter’s predominant personality type—adding this to the list of things campaigns know about you. But while the microtargeting is new, the basic concepts about personality are old. In the past three decades, disparate research groups have identified a set of basic personality traits. Often called the “Big Five,” they can be summarized by the acronym “OCEAN”: openness, conscientiousness, extroversion, agreeableness, and neuroticism. Classification is typically done through questionnaires. Among the typical questions: Are you someone who worries about things? Makes friends easily? Has a vivid imagination? Trusts others? Completes tasks successfully? And so on.
Here’s how the company does it: Cambridge Analytica administers such questionnaires online, promoting them using ads that promise to tell you the relative weight of your personality traits. The company says it has used these tests to “harvest” the personalities of several hundred thousand Americans. Even if you haven’t taken one of its tests, the company categorizes you by extrapolating. It concludes that you tend to be, say, agreeable or neurotic by matching statistical profiles made up of as many as 5,000 commercially or publicly available data points about you to the statistical profiles of people who actually took the personality tests and came out as agreeable or neurotic and so on. (It will not discuss the particulars of these statistical matches but says the data come from consumer database companies including Acxiom, Experian, Infogroup, and Aristotle, as well as the Republican Party’s voter file.)
By this means, Cambridge Analytica claims to have acquired a rough sense of the personalities of much of the U.S. population. Assuming the method is accurate, your personality type then provides an extra layer of information about you, atop the demographics and other information campaigns and their consultants already use. The result could be an ad geared toward your agreeable personality, even if you are demographically quite similar to the neurotic guy next door.
When this information is shared with political candidates, their campaigns can devise and test ads in small groups to see which ones are effective. To serve individual ads online, Cambridge Analytica uses the standard marketing practice of matching its data to cookies—small pieces of data that track your computer’s visits to websites—using third-party services like Acxiom that merge online and offline data about you. In addition, Facebook serves as a major platform for targeted ads. While Facebook says it does not provide data that could reveal your personality, it will display customized ads to lists of individuals provided by a political campaign. Finally, not all of the targeted messaging occurs online; some campaigns mail customized ads to voters’ homes.
Too much information
A large and growing body of research shows that personality traits predict a wide array of behaviors, financial decisions, and happiness levels in jobs and relationships. When it comes to politics, research at the University of Toronto and the University of Minnesota has found that people with “openness” as a major trait tend to favor liberal candidates, while people who scored relatively high on “conscientiousness” tend to be more conservative.
Your personality also probably plays at least some role in your overall susceptibility to persuasion—a valuable thing for a political campaign to know. Several years ago a group of political scientists at Yale University and other institutions found that social pressure could encourage people to vote. If voters were presented with information detailing which of their neighbors had voted—and were told that their own record of showing up at the polls might be similarly publicized—they were far more likely to go to the polls. (The fact that you voted—but not for whom you voted—is public information.) People who scored high on “emotional stability” and “openness” responded more strongly to this social pressure.
The ad voters received from Thom Tillis’s 2014 Senate campaign in North Carolina depended on whether the campaign thought the recipient was conscientious, neurotic, or agreeable.
So if you wanted to devise a campaign that used social pressure to get your supporters out to vote but had finite resources to do it, it would make sense to focus on reaching people you suspected had personalities characterized by emotional stability and openness, says Huber, who cowrote the study.
At the same time, there’s reason to wonder if we’ve reached the limit of what data analytics can actually achieve by microtargeting individual voters. Campaigns have plenty of methods to experiment with: as of 2012, by Kreiss’s count, technology and data-science staffers on presidential campaigns had founded 67 companies to work for Democrats and another 13 for Republicans. But the campaigns may be getting overconfident in what algorithms can really do.
“The problem with data analytics and all these models is that they all look backwards, not forward,” says Micah Sifry, cofounder of Personal Democracy Media, which produces the annual Personal Democracy Forum conference on how technology is changing politics. “The best thing they can do is offer you a prediction on what people are likely to do based on what they did in response to stimuli in the past.”
In this election year, filled with fast-changing trends and unpredictable candidates, that dependence on past behaviors and basic personality types could be an unreliable way to gauge a person’s political preferences. Maybe we need to add another distinguishing trait to match with voters: anger.