Facebook Privacy - An Oxymoron?

What your Facebook likes say about you — including the things you never posted

You never told Facebook your politics, your religion, or your sexual orientation. According to a Cambridge University study, it could probably guess anyway.

Researchers from Cambridge's psychometrics team, working with a Microsoft research centre, analyzed data from 58,000 Facebook users and built models to predict traits users had never disclosed. The reported accuracy was striking: 88 percent for male sexual orientation, 95 percent for race, around 80 percent for religion and political leanings. Personality traits and emotional stability came in lower, between 62 and 75 percent — better than a guess, worse than a confession.

The point the researchers wanted to make was not that liking a page is destiny. It was that the pattern of likes is. People who liked "curly fries," the study found, tended to score higher on IQ measures; fans of a page called "Sliding on Floors With Your Socks On" were less likely to use drugs. No single like means anything. Thousands of them, fed to an algorithm, start to mean a lot.

Facebook declined to comment at the time. But one of the study's authors, Michal Kosinski, told the Financial Times the quiet part out loud: the team used simple, generic methods. "Marketing companies and internet companies could spend much more time and resources, and hence get much higher accuracy than we did."

That is the part worth sitting with. Less than 5 percent of users the model flagged as gay were connected to explicitly gay groups — meaning the signal came from everything else. The data trail you leave while trying to keep something private is, in the aggregate, louder than the thing you are hiding.

The timing was awkward for the industry. The same week the study made headlines, the EU watered down a planned overhaul of data privacy rules after heavy lobbying from tech companies. Personal data was already big business: UK lender Wonga was making credit decisions in seconds from thousands of data points including Facebook profiles, and Tesco had begun mining shopping histories for targeted ads.

Kosinski's own advice was not to log off. "The milk is to some extent already spilt," he said — there is too much about all of us online already. His suggestion: raise your privacy settings, and put consumer pressure on the services that protect privacy best.

Thirteen years later, that advice reads less like a warning and more like a weather report. The study was early proof of something we now take for granted: on the internet, you are not what you say about yourself. You are what your clicks say about you.