3 March 2017

Did the dark lords of data really win it for Trump?

By Sophie Warnes

Cambridge Analytica is the company on everybody’s lips right now, and for all the wrong reasons.

The data analytics company has been credited with helping Donald Trump win the White House, as well as contributing significantly to the efforts of the Leave.EU campaign. And it’s done it all using your Facebook data.

According to its website, Cambridge Analytica “use big data and advanced psychographics to grow audiences, identify key influencers, and move people to action”. What this means is that it claims to have found a way to leverage social media to pinpoint the “personality type” of millions of people – in fact, of all 220 million adults of voting age in the United States.

At the heart of Cambridge’s methodology is the OCEAN score – the standard personality test based on Openness, Conscientiousness, Extroversion, Agreeableness, and Neuroticism.

Your “agreeableness” and “openness” can determine whether or not you’re likely to be persuaded by advertising – and Cambridge can use the other factors, plus all the other data it has on you (what magazines you read, what churches you attend, what you buy on your credit card), to pick advertising messages that will specifically influence you.

Say you are trying to sell firearms – or campaigning for Second Amendment rights. “For a highly neurotic and conscientious audience,” Cambridge’s chief executive, Alexander Nix, told a conference last year, you would use social media adverts based on “the threat of a burglary – and the insurance policy of a gun”. For “a closed and agreeable audience, people who care about tradition, and habits, and family” you would instead show a father and son shooting ducks together.

This, some people have suggested, is how Cambridge – which is backed by the hedge fund billionaire Robert Mercer, Trump’s biggest single donor – helped swing both the UK referendum and the US election.

The idea that anyone can perform such data wizardry on large swathes of the population, influencing them to change their minds, is certainly pretty scary.

But it’s just not really true.

First, Cambridge are by no means the first people to use these techniques. Second, there’s no real evidence that they work. And frankly, the notion that social media is the single biggest influence on our lives goes against everything we know about how humans work.

On the first point, it’s important to note that this sort of strategy – utilising Facebook to target people with specific adverts – is nothing new. Marketers have been doing it ever since Facebook launched tools like Facebook Ads and Facebook Insights.

I’ve used these tools myself, and it’s incredibly easy to find the right people to target. Facebook Likes are the most powerful tool at advertisers’ disposal – it tells them exactly what you enjoy, and a combination of Likes builds up a really good picture of the kind of person you are.

You can pick out audiences based on demographics, the music they like, the films they like, the people they interact with, their location, their humour, and their favourite pub. You can even use that data to predict everything from people’s political leanings to their sexuality.

Cambridge Analytica’s contribution was to create personality quizzes for people to fill out, which not only told you what personality type you were, but also told them. Using a combination of other data, such as Likes, they then extrapolated this everyone else.

To simplify this dramatically, the idea was that if lots of people turn out to be agreeable and also like fish and chips, then the assumption is that people who like fish and chips are agreeable.

But there’s a problem here – and it’s not just that people are more complicated than these relatively simplistic groupings allow for. It’s that it’s hard to see such data wizardry having had a transformative effect.

The EU referendum, for example, was won by a relatively slim margin, in a country whose media and political leaders have been criticising the EU’s every action for decades. Equally, in the US, Trump lost the popular vote to Hillary Clinton by three million votes.

If Trump, or Leave.EU, had all this data behind them, shouldn’t they have won by more? Or could it be that harvesting data from unsuspecting internet users only gets you so far?

The explanation, of course, is that the internet is not the world. Although it is of increasing importance, we don’t just conduct our lives online. We interact with friends and family, we see newspapers, we notice billboards in the street. To lay the success of any political campaign at the feet of a purely data-driven strategy is to ignore everything else that helps form our personalities and opinions.

Friends and family, in particular, are a huge influence on our lives. The more time you spend with someone, the more they influence you and vice versa. In fact, research shows that having friends who vote for a certain party makes you more likely to vote for that party – even if you generally vote for a different one.

Another survey from 2015 on how voters make decisions shows that they use a mix of sources depending on what they want to know. People wanting to find out about the values and the policies of the party they want to vote for will look at online news or watch TV. Social media is mostly useful, people say, as a source of other people’s opinions.

So, just using targeted advertising on social media isn’t the be-all and end-all. In fact, BuzzFeed managed to speak to over a dozen people who had seen Cambridge Analytica’s work – and all of them said that the claims about its electoral impact had been wildly exaggerated.

Gary Coby, the man behind Trump’s digital campaigning, repudiates the claim that Trump’s team “tested 175,000 different ad variations” aimed at different personalities. He says it is a “100 per cent lie” and that there were no “psychographs” involved in the testing.

Daniel Kreiss, professor of political communication at the University of North Carolina, told MIT Technological Review: “Is personality the strongest predictor of whether someone is persuadable versus any of these other things? I haven’t seen any data that has suggested that this form of modelling does or doesn’t work.”

That’s partly because this kind of model is predictive – that is to say it is based on past actions and experience, which is not always the best way to know how someone will behave in future.

To see how far the data-crunchers still have to go, it’s worth taking a look at the insights that have been garnered about you personally.

DataSelfie is a Chrome extension which tracks you in the same way Facebook does, and can tell you about your shopping preferences, your political views, your personality, and even about your health. Facebook has its own way of showing you what it knows about you. And you can also see what Google knows by looking at your Google Ads Settings.

Here’s how accurate this online targeting is for me: YouTube thinks I want pregnancy tests (no thanks); Google thinks I’m into folk music, winter sports, and “men’s interests” (no to all three); and Facebook thinks I like “unboxing”, mirrors, and online shopping (couldn’t care less about any of them). I’m pretty certain your supposed “interests” won’t wow you either.

I’m sure these techniques do something to help people pinpoint persuadable voters, or customers. But ultimately, only you can decide what to be influenced by, and only you can decide who and what to vote for.

We are not puppets on a string, waiting to be manipulated by our masters – we are critical, complex creatures – and thankfully, we can exercise our free will to say “no” to advertising too.

Sophie Warnes is a freelance data-driven journalist based in London