We reveal our lives through our online affiliations: bands we like, brands we follow, political causes we support. But for the past few years, University of Cambridge researchers have been finding more obscure connections — and by some measures, it turns out, Facebook may judge your character more accurately than your friends or family.
In a recent PNAS paper, researchers Wu Youyou, Michal Kosinski, and David Stillwell concluded that Facebook "likes" could predict someone's self-described personality better than an evaluation from a friend, family member, roommate, or coworker. They started with a discovery that Kosinski and others published in 2013: that likes were solid indicators of everything from sexual orientation to religious views to drug use. This wasn't just a matter of expressing support for the Republican Party or an activist group; it was hidden in preferences as innocuous as being fond of "The Joy of Painting with Bob Ross."
Marilyn Manson, The Oatmeal, and 'war' are all competitive interests
Then, they took it a step further. Instead of looking for hard behaviors, they applied the principle to general traits, looking for links between likes and what's known as the "big five" model of personality elements: openness, agreeableness, extraversion, conscientiousness, and neuroticism. What they found was sometimes downright stereotypical: liking "Snooki" or "BEER PONG!" is linked with high extraversion, Prozac Nation correlates with neuroticism, and liking "hugs" or "war" could put you on, respectively, the "cooperative" and "competitive" side of agreeableness. Others are less obvious: League of Legends fans tend to be spontaneous, Bourne Identity fans trend cooperative, and Wikipedians may be reserved.
Some 80,000 volunteers profiled themselves the traditional way, filling out a 100-part "big five" questionnaire that asked them to agree or disagree with statements like "I am the life of the party," "I make plans and stick to them," and "I insult people." Around 17,000 of those people successfully invited one Facebook friend to fill out a much smaller 10-part test, and 14,000 got two people to profile them. Using an online tool, researchers then evaluated participants' Facebook likes and compared the resulting scores.
The idea of quantifying a personality raises its own set of questions. When you're describing abstract concepts like "openness" or "agreeableness," should your own feelings determine the "real" you? Or the way others see you? Or the long record of interests you've left behind? For most of the study, the authors used what people said about themselves as the gold standard. But even that isn't totally consistent. If you take a questionnaire twice with a gap of weeks or even days, according to Kosinski, you'll end up with a correlation of somewhere around 0.7, not a perfect match of 1. If you managed to get the same scores, "it would be perfectly accurate," he says. "But no measures in life are perfectly accurate."
I don't like enough things
That includes Facebook profile measurements, which get more accurate as likes increase. Using data sampled last year for another project, the authors estimate an average Facebook user has around 227 of them. That's a count of every page, brand, band, baby photo, and birthday announcement you've ever clicked your favor to, some of which will end up discarded by the model. "Obviously, not all of the likes are equally predictive," says Kosinski. Nonetheless, with enough data, you can get impressively close to your own score. With an average number of likes, the model did far better than the scores of friends, family members, and roommates. It came in slightly under spouses, a difference of 0.58 to 0.56. With more than 500 likes, the correlation jumped to 0.66. Correspondingly, it's steadily less accurate with fewer likes. You can test Cambridge's prediction system online, but I actually had too few likes for it to work. (According to my profile, I've got about 30.)
There was another interesting wrinkle: the authors report that in some cases, their Facebook evaluation could predict behaviors better than someone's own personality questionnaire. Traits like extraversion or neuroticism can be linked to more observable things, like large social circles or depression. And when the authors tested 15 of these behaviors, four or five of them ended up correlating more strongly than friends, family, or even the participant. "It turns out that computer prediction of your extraversion is a better predictor of number of friends that you have than your own score," for example, says Kosinski.
Kosinski tentatively attributes this to people's unconscious bias while filling out the questionnaire. "People like to feel good about themselves," he says. "Maybe they're not so social, but they believe that being social is really important — they sit at home a lot, but they believe 'I sit at home, but I really like to go out!'" It's an obvious idea, but an interesting one in light of some common wisdom on Facebook: that you use it to make yourself look better.
"People like to feel good about themselves."
In some ways, the idea of Facebook understanding you better than friends or family is fantastical or frightening. In others, it's obvious. Facebook likes share a major similarity with personal questionnaires: they're both someone's own carefully curated self-presentation. While the correlations themselves — and the sheer quantity of them — are impressive and sometimes inexplicable, the practical implication is that answering questions about what you like and what you feel like may get you closer answers than asking someone else to describe you. Is it any surprise that saying you like "partying" is a good proxy for saying you enjoy the company of other people?
But one of those things is usually private, and the other is often available to a wide network of friends, colleagues, and companies. The study's authors suggest that recruiters could use it to help match candidates to jobs, or that Facebook members themselves could take their personality scores into account when making decisions about where to work or who to take on a date, using it to "augment their own intuitions and judgments."
The deeper question is how much weight we should give these tests, no matter how they're administered. Is it better to assume that you're putting on a social media act when you click "I hate it when all other schools near you have a snow day and you don't" (correlated with spontaneity), or that it's an expression of your most fundamental self? Kosinski thinks that whatever happens, it's up to us. "Technology is neither good nor bad, but it can be used for both, right? It can be used to improve people's experience, help them to apply for jobs or find a match on a dating website," he says. "But obviously it can also be used to figure out their personality and craft marketing messages to sell them something they don't really need."