ADVERTISEMENT

The Shaky Science Behind Cambridge Analytica's Psychographics

The Shaky Science Behind Cambridge Analytica's Psychographics

(Bloomberg View) -- It’s possible that despite capturing Facebook data on 50 million people, Cambridge Analytica didn’t actually accomplish anything noteworthy in the realm of politics. At the center of the company’s claim to significance is a technique for turning people’s Facebook “likes” into psychological profiles, and a further claim that these profiles rendered victims vulnerable to manipulation.

But scientific papers used to back those claims leave doubt as to whether the company’s technique was likely to work as advertised.

It sounds chilling. Quoted in The Guardian, ex-employee Christopher Wyle called the company's technique “psychological warfare.” The idea, as Wyle described it, was to target people with political ads “designed to work on their particular psychological makeup.”

On a British television expose, company officials boasted that their profiling enabled them to market Donald Trump to the American people. Wyle, who had studied fashion, told the Guardian they convinced Americans to vote for Trump the same way companies convinced people to buy Crocs and other ugly shoes.

But the science behind the claims suggests that what Cambridge Analytica was really marketing was the idea that it could create a kind of psychological alchemy.

As described in the New York Times, the psychological targeting concept rests in part on a 2015 scientific paper, published in the Proceedings of the National Academy of Sciences. In it, researchers showed that when given access to Facebook likes, an algorithm could do better than a subject’s friends at guessing how they would score on a personality test measuring the so-called big five traits: openness, conscientiousness, extroversion, agreeableness and neuroticism. The items liked could seem weirdly trivial -- hobbies, favorite celebrities, bands, candy bars.

That sounds scary, but is it really that surprising? As several social scientists pointed out to me, the friends didn’t do a very good job of predicting personality traits. There was about a 50 percent correlation between the way people scored themselves and the way their friends scored them: The friends’ guesses were 49.9 percent accurate, and the algorithm was marginally better at 56 percent accurate.

It doesn’t mean people don’t know their own friends. I know a lot of things about mine. I know who wouldn’t mind picking me up at the airport, who shares my taste in literature, who’s a vegetarian, who likes craft beer and gourmet cheese, who would prefer carrot cake for a birthday surprise, who would prefer chocolate, and who is avoiding sugar.

I’m less sure how they would score on tests of neuroticism, openness or extroversion. Most people I know are extroverted some of the time but can also be self-contained and introspective. I’m not sure I could predict my own results very well, as I tend to score close to the middle range on those sorts of tests.

Personality scores have some predictive power when applied to large groups of people, but they don’t define individuals any more than SAT or IQ scores. As I learned in reporting this column about research on personality and success, nearly all of the interesting predictive power researchers found was concentrated in only one of the big five traits: conscientiousness. And here, they were looking at the possibility that it’s more of a skill than a trait -- something that can be improved to set young people up for success.

But even if its algorithm had some power to guess how Facebook users would score on personality tests, could Cambridge Analytica use this to change enough voting behavior to matter?

That part of the claim rests largely on the work of a team of researchers studying effects of targeting advertising based on personality scores. After the election, they laid out the evidence for personality-based targeting in the Proceedings of the National Academy of Sciences. They reported that people scoring high on extroversion were a little more likely to click on an extrovert-targeted ad than were others, and people high in openness were much more likely to click ads aimed at them.

But there’s a caveat, said sociologist Duncan Watts of Microsoft Research, who has studied the effectiveness of advertising. When the researchers showed the ads that were targeting people high in openness, those generated the most clicks for everyone. When shown to the extroverts, that is, the openness-targeted ads did at least 30 times better than the ads that were supposed to be extrovert-targeted. What this means is that the openness-targeted ads were probably more compelling in ways that swamped any effect of targeting.

Even if psychometric targeting had only a small influence, Trump only needed a tiny edge to win. But why focus on it when there are so many other factors that are much likely to have made a difference? A big one is the power of Facebook ads without any psychological profiling, said David Rand, a professor of psychology, economics and management at Yale. Facebook is an advertiser’s dream, he said, because it provides a completely different, fine-grained way to target ads based on all kinds of demographic data.

Even if Facebook improves its security and protects information on which movies and bands people like, there’s plenty of publicly available information to target us for the marketing of cars, candy and presidents.

This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.

Faye Flam is a Bloomberg View columnist. She has written for the Economist, the New York Times, the Washington Post, Psychology Today, Science, New Scientist and other publications. She has a degree in geophysics from the California Institute of Technology, and has been a Knight-Wallace fellow at the University of Michigan.

To contact the author of this story: Faye Flam at fflam1@bloomberg.net.

To contact the editor responsible for this story: Tracy Walsh at twalsh67@bloomberg.net.

For more columns from Bloomberg View, visit http://www.bloomberg.com/view.

©2018 Bloomberg L.P.