If you live in Leeds Central, a constituency that has been a Labour stronghold since its creation in 1983, you might have seen an advert on Facebook in the run-up to the 2017 UK general election that was created by the Conservative Party, but didn’t mention Conservative policies or politicians. Instead it featured former Labour voters explaining why they were stepping away from their party to vote Tory. One woman said: ‘Jeremy Corbyn would be disastrous as a leader and I cannot see him around the negotiating table.’
Meanwhile, immediately to the south in the constituency of Morley and Outwood, voters were treated to a very different advert. In the 2015 election in that constituency, Conservative Andrea Jenkyns had narrowly beaten Labour’s former Shadow Chancellor Ed Balls, with UKIP candidate David Dews in third place. This time round voters reported seeing a Conservative video in which Boris Johnson urged UKIP supporters to back the Tories. He said: ‘If you vote UKIP, I’m afraid that will be doing exactly what Jeremy Corbyn and Diane Abbott want to happen because it will make it more likely they are in charge.’
Why did these two neighbouring constituencies see such different messages? Because political campaigns are increasingly using social media to reach voters in particular locations, customising messages to appeal directly to those people. And it’s not just geography they can target – it’s also your age, sex, tastes, preferences and personality.
Across the country, voters reported seeing targeted ads on social media for political parties, including Labour, the Liberal Democrats, the Green Party and the Scottish National Party. Welcome to the brave new world of ‘dark ads’, where parties can pay Facebook to make sure certain ads are only seen by certain people. A party could design one advert for people under thirty-five, for example, and another for those over thirty-five, and each age group would see only the one that the party has decided best aligns with their political viewpoints.
The idea that parties use tailored messages and other tools to manipulate us is nothing new.
Political insiders have long used so-called ‘dog whistle’ messages, intended to say different things to different voters. But modern technology is allowing campaigners to take this to a completely different level. The internet, and social media in particular, has opened up what can be a hugely powerful channel not just to communicate with us, but to use our online footprint to analyse and target us – it could be a game-changer for those who can afford to exploit it.
Most of us seem to have accepted the idea that big companies are sharing vast amounts of personal data in order to target us more effectively with advertising. Like me, I’m sure you’ve browsed items of, say, furniture online only to see adverts for the same items pop up on other sites later. But people are often less aware that political campaigns can also use our online activity to target us with their messages – something that is potentially much more unsettling.
Following the 2016 Brexit referendum and US presidential election, rumours started to swirl about a little-known company named Cambridge Analytica. At the time of writing, this company offers what they call ‘data-driven campaigns’. ‘By knowing your electorate better,’ their marketing materials boast, ‘we achieve greater influence while lowering overall costs.’ Today, however, the company has become more widely known for supposedly having used psychological profiling based on social media activity to help target campaign material for the 2016 Trump and Brexit campaigns. But what exactly the company did and how successful it was remains the subject of speculation.
CEO Alexander Nix has boasted of the company’s ‘secret sauce’ algorithms, which could predict voters’ psychological profiles, but others (including current and former employees) have dismissed such claims as nothing new: crafting different messages to voters depending on their particular profiles is something politicians have been doing for a long time, after all.
As a scientist, but also as a concerned citizen, what is extremely troubling for me is that it is hard to judge the effectiveness of the most recent campaigns as we simply don’t know what they are doing. British election campaigns are strictly regulated; unlike America, Canada and Australia, parties and politicians are not allowed to advertise on TV and radio, only on billboards, in newspapers and via election pamphlets. Messages on TV are only allowed through national PPBs – party political broadcasts – TV slots (usually around five minutes long) allocated free of charge using a formula set by Parliament. But those regulations have not yet caught up with – indeed they don’t even mention – social media, which is consequently much murkier territory. We also don’t know to what extent these companies are trying to oversell their abilities or past successes in order to promote their brand and products.
Well, whether or not companies like Cambridge Analytica are exaggerating their achievements, their claims are certainly based on some valid scientific foundations.
We know that a data scientist can use your activity on social media sites such as Facebook and Twitter to make accurate guesses about your personality. We also know (as we discussed in Chapter 2) that there are surprising links between our personalities and how we vote.
So yes, in theory, by analysing our activity on social media, campaigns might be able to identify who is most likely to vote for them, and then target campaign material at those people.
We also know from earlier chapters that ‘openness to new experiences’ is a particularly important dimension to personality when it comes to predicting how we vote. Research by Michal Kosinski and colleagues at the Psychometrics Centre at the University of Cambridge demonstrated that this was something you could accurately guess from people’s social media activity.
The researchers created software called ApplyMagicSauce. This software attempts to predict a lot about you, including your age, gender, sexual orientation, personality profile and, importantly, your political leaning. If you look up the website (applymagicsauce. com) and connect it to your Facebook or Twitter account you can see what it says about you.
It may not get everything spot on (it got my age, gender and politics right, but my sexual orientation wrong) but on average, testing hundreds of people, it is impressively accurate. In one of the 100 most influential papers published in 2015, the software proved more accurate in describing someone’s personality profile than a close friend.1
The software requires sophisticated algorithms that detect patterns in data using millions of observations that would be too subtle for humans ever to detect themselves. It can only make an accurate guess about you based on your online activity, however, so the more you do online, the more accurate its prediction of your personality. If you’ve only ever liked a few things on Facebook, for example, the algorithms won’t have a lot to go on and are about as accurate in guessing your personality as a work colleague would be. After around 70 ‘likes’, however, they perform (on average) better than a friend would. After 100 likes, the software can outperform a family member. And, rather incredibly, after 300 likes it can describe your personality profile more accurately than your partner can!
Running and interpreting these algorithms means paying for experts in data processing and analysis. This in turn means that political campaigns with more cash to spend on data science might have an advantage in identifying and reaching receptive voters. There is also some speculation that campaigns are using the same techniques to identify voters who are likely to support the other side, then targeting them with campaign material intended to deter them from voting, such as by attempting to discredit particular politicians, or creating cynicism in the political process.
Overall Donald Trump spent far less on his campaign than Hillary Clinton did on hers, but he spent a lot more than she did on social media. By the end of the campaign, Trump’s digital team was estimated to have spent around $85 million on digital and online advertising. It is also estimated that Trump’s campaign made hundreds of thousands of different pieces of creative content, testing and refining their messages.
Wired journalist Issie Lapowsky spoke to Gary Coby, a member of Trump’s digital team and director of advertising at the Republican National Committee: ‘On any given day, Coby says, the campaign was running 40,000 to 50,000 variants of its ads . . . On the day of the third presidential debate in October, the team ran 175,000 variations.’
We don’t know exactly what they were doing with all of those ad variants, but it is possible that they were targeting voters based on their psychological profile. And there are some areas where using psychological profiling clearly might have helped. For example, people who score lower on openness to new experiences are more likely to be receptive to messages about controlling immigration – so the best way to engage them might therefore be to target them on that particular subject.
Writing about Cambridge Analytica and the Trump campaign, New York Times reporters Nicholas Confessore and Danny Hakim explained how they thought voters might be targeted: ‘A voter deemed neurotic might be shown a gun-rights commercial featuring burglars breaking into a home, rather than a defense of the Second Amendment; political ads warning of the dangers posed by the Islamic State could be targeted directly at voters prone to anxiety.’
Trump also made a clear appeal to a sense of national pride, with his call to ‘make America great again’. Had I been working for a data firm for Trump’s campaign, I would have been trying to target voters with the strongest sense of in-group loyalty, who would be the most receptive to that message.
Another core group that Trump appealed to was white working-class voters who saw global trade agreements and outsourcing as a threat to their jobs. If Trump’s campaign had also used demographic data (such as age, gender, occupation) to identify and target such voters with his messages about bringing jobs back to America and protecting US industries, you can start to see how an effective modern digital campaign might work.
Again, however, the big problem is that while there is clear evidence behind the idea of psychological profiling, there is still a big gap between that knowledge and what campaigns are doing in real life.
During the 2017 UK election, a group of ‘political technologists’ in London set out to develop software that would reveal what political campaigns are doing on social media. They created a plug-in called Who Targets Me – citizens in the UK could install this plug-in, and it would automatically send the developers information about any targeted political adverts directed to those voters. This platform is therefore designed to help bring ‘dark ads’ into the light, so that we can compare what messages are being targeted to different users in different parts of the country.
The developers are themselves calling for clearer regulation of online political campaigning, but in the meantime this ‘hack’ will hopefully offer a little bit of insight into what political parties are doing online. At the time of writing these results are still being analysed, but one thing that seems clear already is that some parties, particularly the Liberal Democrats, are doing a lot of A/B testing with their messages. This is a classic technique in behavioural research on marketing; two (or more) similar messages are developed, both are then released and assessed on a limited basis, and then the more effective of the two versions is rolled out.
But what also seems to be clear is that these British efforts are amateurish by comparison with what the Americans were doing the year before: Gary Coby referred to his team’s approach for the Trump campaign as ‘A/B testing on steroids’.
On one level this seems like a fairly harmless way of increasing the efficiency of political campaigns, helping parties develop effective messaging as quickly as possible. Indeed, you could argue that this strategy could help campaigns engage more voters on the issues that matter most to them. Given all that we have learned about confirmation biases, however, it also means campaigns could end up feeding us the messages we want to hear, rather than enabling us to make a genuinely informed vote.
There are also some reports suggesting that the UK Labour Party (after rather neglecting the use of social media in 2015), has started integrating its ‘on the ground’ and ‘online’ campaigns. Apparently, if you mention concerns about the NHS to a Labour canvasser on the doorstep, you might find adverts about Labour’s stance on the NHS appearing in your Facebook newsfeed later that day.
But does that really matter? After all, we’ve seen that it’s not that easy to change someone’s mind. Indeed, the conventional wisdom in modern election campaigns is that you shouldn’t focus on trying to convince people on the other side. You should instead aim for the political centre where the so-called ‘swing voters’ are likely to be found – and most importantly ensure that your base actually turns out to vote. Most of the ‘door-knocking’ by party canvassers in the run-up to an election is about identifying people who already support their party to ensure they come out to vote on election day.
As it turns out, however, with a little bit of psychological insight, campaigns might be able to achieve a lot more than that. In the previous chapter, I mentioned a study that suggested it was possible to change people’s views on a polarising topic (the kind of topic that can shape how people vote) just by having a canvasser talk to them for a few minutes. Two scientists exposed this study for being based on false data, but when they finished running their own study, on transgender discrimination, their results showed that the canvassing strategy really can change people’s biases on a particular topic.2 The data used in the first study may have been false buts its findings were not.
There was no dispute about the second study’s validity: the researchers found that sending canvassers to talk to people for ten minutes about how transgender people suffered from prejudice substantially reduced transphobia, and the effects were still evident three months later.
The two researchers cautioned against the received wisdom of only focusing on voters who already agree with you, concluding instead that it may be in campaigns’ best interests to engage people across partisan lines, even on controversial topics. In this instance the researchers encouraged voters to ‘take the perspective’ of transgender individuals, but it is plausible that on other key topics there are also messages or techniques that might be particularly effective in changing people’s minds. Indeed, in the previous chapter we saw how the message that ‘97% of climate scientists agree that man-made climate change is happening’ has been identified as a key ‘gateway’ belief in changing people’s minds on this hot topic.
Perhaps political parties have in fact been aiming too low. Perhaps they should be designing campaigns to try to change beliefs and attitudes on key issues rather than simply getting their base to turn out. What is certainly clear is that as the science of why we vote the way we do advances, political parties are likely to turn to increasingly more sophisticated ways of winning our vote.
The use of psychological insights to target voters on social media is a relatively recent step, but attempts to use psychology to influence citizens’ behaviour in other ways stretch back over at least the last fifteen years – and have developed rapidly in that time. Government departments and policy-makers are increasingly aware that a little bit of psychological insight can go a long way in changing what people do. So much so that the approach has a name: nudge (from the book of the same name by Cass R. Sunstein and Richard H. Thaler).
So, what does ‘nudge’ mean in this context?
The underlying logic of the nudge agenda (also referred to as ‘behavioural insights’) is that by understanding why people behave the way they do, we may be able to subtly ‘nudge’ them, using very small adjustments, to make particular choices. Famously, the urinals at Schiphol Airport in Amsterdam painted a fly next to the drain holes and reduced spillage by 80 per cent. ‘It turns out that, if you give men a target, they can’t help but aim at it,’ wrote Sunstein and Thaler.
The idea is not to try to force people into doing things they might not otherwise have done, but rather to encourage choices that people themselves would agree are beneficial. A lot of nudge projects look at the ways in which governments communicate with people, to make those systems as efficient as possible. For example, in the UK the NHS has tested various ways of sending people different kinds of letters, emails and even timely and personalised texts to make sure they turn up to their hospital appointments on time. And governments in the USA and UK have both used the ‘nudge’ approach to get people to save for their pensions and pay their taxes on time.
Not surprisingly, it isn’t just governments that are making use of such psychological insights. Technology companies are also trying to understand how to change our behaviour. Most of the time these changes are simply aimed at improving the company’s profits. Some companies, however, are also exploring how they could help promote civic engagement.
In 2012, researchers from Facebook published a paper in collaboration with a team from the University of California that demonstrated the power the social network has both to experiment on its users and to influence elections.3 On the day of the 2010 US congressional elections, Facebook presented millions of users with a clickable ‘I voted’ button at the top of their newsfeed (you may well have seen a similar button if you used Facebook on the day of the 2017 general election in the UK).
However, in that election not everybody saw the same thing. One group of people were just given the option of clicking on the button to show friends that they had voted, while other users also saw the profile pictures of some friends who had already clicked the ‘I voted’ button themselves.
The study demonstrated that voters who saw the ‘I voted’ button along with an image of any friends who had pressed this button, were more likely to click on it themselves than those who simply saw the button. And the closer the friends, the more likely people were to follow suit. Even more importantly, using public voter records (in the USA there is a public record of who voted, although not who they voted for) the researchers argued that the people who saw that their close friends had pressed the button were also more likely to actually vote.
The researchers estimated that by placing this simple button on people’s news feeds, Facebook may have increased turnout rates by something in the order of 60,000 votes. In a closer contest, such as the famously tight race between George W. Bush and Al Gore back in 2000, this small social nudge could have been enough to change the result of an election. And given that Facebook use is still higher among younger generations, who are more likely to vote to the left, this ‘politically neutral’ attempt by a social media channel to increase voter participation could have inadvertently nudged the vote in favour of the Democrats.
This result demonstrates an important principle in the science of behaviour change. Namely, if you want something to be effective in changing people’s behaviour, make it social. It might seem crude or gimmicky to promote voting using online ‘peer pressure’ but such action actually has roots in the past: in Italy there used to be a tradition of pinning the names of all voters to the front door of the town hall (to show who had lived up to their civic duty). As we will see, there is much more that exponents of psychology could do to develop civic rituals like this to promote voting.
Many people criticised Facebook for the experiment described above and for other such large-scale trials (which were all carried out without informing users – who automatically give their consent to be experimented on when they agree to the terms and conditions of service), arguing that they are manipulating users and potentially influencing election outcomes.
The reality is that large tech companies such as Facebook, Amazon and Google are constantly trying things out on us on us, working out what combinations of adverts or posts are more likely to keep us using their sites, and engaging with things that make them money – and of course how best to use the huge amounts of data that they have about each of us. It’s just that we normally don’t see the results. In this instance, the fact that Facebook published the results of their experiment so openly could be said to count in their favour, and it could also be argued that they were attempting to encourage something that almost everyone would agree is a positive outcome: higher voter turnout.
In fact there is so much consensus on the importance of increasing turnout that there has been a lot of nonpartisan academic research aimed at trying to understand what does and doesn’t work.
One of the most famous studies in this area highlighted something very similar to Facebook’s experiment, namely that social pressure matters. In 2006, leading political scientists Alan Gerber and colleagues at Yale University set out to test whether they could increase voter participation in a Michigan state election by writing to potential voters with different messages, applying varying degrees of pressure to vote.4 Some citizens received a standard letter encouraging them to do their civic duty; others were informed that their voting records were being studied; and one group received a letter explaining that after the election, another letter would be sent with a complete list of everybody who had and hadn’t voted on that street so that ‘you and your neighbors will all know who voted and who did not’.
The results were both dramatic and depressing. The usual level of voting was 29.7 per cent, but for those in the ‘neighbours’ group this shot up to 37.8 per cent – a proportional increase of almost 30 per cent. Quite a dramatic effect for a simple letter – and a vastly larger one than was achieved in the Facebook study. The cost of sending everyone a letter was certainly higher than putting a button on voters’ Facebook feeds, but it was much cheaper and less labour-intensive than traditional door-knocking or phone-bank approaches.
At the same time the study also highlights just how low levels of voter participation can be, and how sad it is that we might regard 37.8 per cent as a successful outcome. Perhaps it only seems successful compared to other attempts to get people to vote: the letter reminding people of their civic duty to vote only increased voting levels to 31.5 per cent. But this does demonstrate the importance of testing the effectiveness of different ways of campaigning. All too often well-meaning campaigners will try to promote some agenda (like voting, recycling, taking public transport), but make no attempt to test whether their campaign will actually make any difference.
So, knowing your neighbours will find out whether you have voted seems to be a much greater motivation than being reminded of your civic duty. Humans are very susceptible to what other humans are doing, and to what other humans will think they are doing.
However, although it is interesting that social pressure can influence our behaviour, the psychological elephant in the room is really why so many people don’t vote at all. Other experiments have demonstrated ways to encourage us into the ballot booth, from voter registration lotteries5 to canvassing campaigns, phone banks and leaflets,6 but most of the effects are rather small. From a psychological perspective the more important question is why any of this is even necessary in the first place. If humans are such inherently social animals, why do so many of us not participate in what is probably one of the most influential collective acts of communication we get to make?
