That Facebook can’t keep a secret is the secret to its profitability. Where you live, what you eat, who you date, how you spend money—what more could an advertiser ask for? As the Canadian developer Andrew Lewis has observed, “If you’re not paying for it, you’re not the customer; you’re the product.”
But what if your most naked data aren’t the details you furnish directly—favorite film, religious views, alma mater—but the little slices of self you reveal every time you click the “Like” button?
As it happens, what you “Like” on Facebook says a lot about who you really are—whether or not you openly acknowledge it. In a paper published online by the Proceedings of the National Academy of Sciences, three Cambridge researchers report that a user’s lifetime of “Likes” can predict everything from her religion and political persuasion to sexual orientation and drug habits.
To tease out the correlation between online “Likes” and real-world characteristics, Michal Kosinski and David Stillwell of Cambridge University and Thore Graepel of Microsoft’s machine learning group collected personality and intelligence tests from some 58,000 users, and cross-referenced the results with personal details scraped (with permission) from the users’ profiles.
“Essentially, we had a big database of people’s personality and the things that they ‘Liked’ on Facebook,” Stillwell told me. “The first thing we did was say, ‘Okay, what do extroverts like? What do introverts like?’ Sometime later, we realized that we could twist the game. Instead of saying, ‘Extroverts like theater,’ we can say, ‘If someone likes theater, they’re probably an extrovert.’ ”
With one piece of information, Stillwell continued, you can’t make a very good prediction. “But once you’ve put all their ‘Likes’ together, you can start predicting some very personal things.”
In fact, the authors write, how they parsed “Likes” predicted gender with 93 percent accuracy, and race with 95 percent. The same “Likes” predicted religion, sexual orientation, and political persuasion correctly more than 80 percent of the time, and drug and alcohol use at a slightly lower rate.
Among the most revealing “Likes” for men were Modern Warfare 2, Dos Equis, and Bruce Lee. Women? Gillette Venus, Wet Seal, and Bebe. African-American users “Liked” Erykah Badu and “Next Friday,” while white users gave the thumbs-up to Harley-Davidson, Bret Michaels, and bonfires.
Extraversion correlated to “Liking” beer pong and cheerleading, introversion to anime and Terry Pratchett (if you have to Google it, you’re not an introvert). Fans of Plath, Wilde, and Bukowski typically possessed liberal and artistic personalities, while supporters of NASCAR, “Teen Mom 2,” and the group “I don’t read” possessed conservative ones.
When asked about life satisfaction, users who “Liked” Sarah Palin, Glenn Beck, and Jesus Christ were among the happiest; fans of science, iPods, and the cartoon character Stewie Griffin were just the opposite. Gay men “Liked” Kathy Griffin and the musical “Wicked,” while their straight counterparts flocked to World Wrestling Entertainment and the Wu-Tang Clan. Lesbians preferred singing sisters Tegan and Sara and Showtime’s “The L Word” to straight women’s Lipton Brisk and Foot Locker.
Some predictors were obvious, says Stillwell. Fans of Mozart, for example, were likely to be intelligent. But high IQ was also predicted by “Liking” thunderstorms, curly fries, and Morgan Freeman’s voice.
“To be honest, we don’t know why many of these things are related,” he told me, and acknowledged that, in some cases, statistical noise could be muddying the results. “What is it about Morgan Freeman’s voice that intelligent people like? I don’t know.”
It’s not difficult to imagine how advertisers will use “Likes” to micro-target specific demographics. This may be welcome in some cases—after all, I enjoy ads for running shoes far more than, say, mascara—but invasive, or downright creepy, in others. In their paper, Stillwell and his colleagues recall the now-apocryphal story of Target sending maternity-specific mailers to moms-to-be, after the company designed an algorithm to predict pregnancy based on past sales (i.e. skin cream, home tests, etc.) The only problem? Some of these women hadn’t yet told their partners, or in some cases their parents, about the baby on the way.
Privacy concerns or no, Facebook continues to process 2.7 billion “Likes” every day. Because while most users are careful to curate their interests, photos, and privacy settings—no Cabo photos for mom, or Office Space GIFs for the boss—“Likes” hardly seem so serious. After all, what’s one tidbit of data?
“It’s so easy to click a ‘Like,’ ” Stillwell says. “They’re so seductive. It’s an impulsive thing to do. But eventually, over time, those things add up. Four years later, those ‘Likes’ can say quite a lot about you.”