Be Careful What You “Like”
Facebook Likes may reveal more about yourself than you intend.
Your sparse Facebook profile isn’t protecting your privacy as much as you think. Those innocuous “Likes”—your favorite TV show, athlete, or potato chip—may open a bigger window on your personal life than you probably intended. Liking Harley-Davidson motorcycles pegs you as a white American of below-average intelligence. Your obsession with Hello Kitty may give away your political leanings: Fans of the cat are more likely to vote Democratic.
In a study published in the Proceedings of the National Academy of Sciences of the United States of America, psychologists Michal Kosinski and David Stillwell of Cambridge University and Microsoft computer scientist Thore Graepel recruited more than 58,000 volunteers from Facebook and compiled background information on each. Then they set out to find connections among various Likes—a median of 68 per user—and particular traits.
Distinguishing between white and black and male and female users was easy work for the researchers’ model. It correctly determined users’ race in 95 percent of cases and their gender in 93 percent. More than eight out of every 10 users’ political party and religious affiliation could be forecast. About three-quarters of the time, the model correctly predicted if the user smoked cigarettes or drank alcohol.
People seldom realize what their Likes reveal. The researchers found that less than five percent of gay users liked a group or statement with an obvious connection to homosexuality, such as “Gay Marriage.” But liking pages such as Desperate Housewives or the musical Wicked correctly tipped off the researchers to male users’ homosexuality 88 percent of the time. Only gender and ethnicity were better predicted. The model also found a strong correlation between high intelligence and liking a handful of seemingly random things such as “Curly Fries,” but it would probably take one of those frites-loving geniuses to unravel the connection.
Kosinski, Stillwell, and Graepel were least successful in forecasting whether users’ parents had split up during their childhood, failing 40 percent of the time. The Likes that were strong predictors—statements such as “When Ur Single, All U See Is Happy Couples N Wen Ur in A Relationship All U See Is Happy Singles”—showed that these users exhibited an unusually strong interest in relationships.
Kosinski and his colleagues used voluntarily submitted data, but (depending on a user’s privacy settings) Likes are freely available to others on Facebook, including people whose intentions may not always be as benign as those of these researchers. And much more data can be mined from Facebook and other sources, from browsing history to purchase records. There’s a potential for good here, the authors say, in improving such things as customer service and users’ online experience. But the new prediction techniques could also be used in ways that threatened individuals’ “well-being, freedom, or even life.”
If people feel exposed and endangered, caution the authors, they may back away from technology. Balancing “the promises and perils of the Digital Age” requires transparency and giving individuals control of their own information.
THE SOURCE: “Private Traits and Attributes Are Predictable From Digital Records of Human Behavior” by Michal Kosinski, David Stillwell, and Thore Graepel, in Proceedings of the National Academy of Sciences of the United States of America, March 11, 2013.