We now know what Cambridge Analytica was up to – and how they did it. They linked personality analysis (psychographics) with Facebook data. But they weren’t even particularly good at what they did.
Cambridge Analytica – remember them?
Last week the BBC R4’s Jolyon Jenkins explored digital psychographic analysis in a series of radio broadcasts entitled Slice: Politics and Personality[I] which shed light on a “scandal” that never was.
The Cambridge Analytica story began at Cambridge University where a group of academic psychologists determined two extraordinary things. The first was that it was possible to model people’s personalities from their Facebook “likes”. The second was that it was possible to change people’s behaviour by targeting people with particular personality types with particular types of message. Cambridge Analytica (CA) picked up on this original academic research and sought to turn it into a tool that could be used in political campaigns.
What was the CA scandal really all about? CA used a personality model to profile all potential voters in the USA. Now there is nothing new about personality testing – many readers of this piece will have been asked to sit a personality test by their employers at some point in their career. It is how they applied it that has caused concern.
The inexact science of psychographics
Personality tests of one kind or another have been around for a long time. Going back to the 1930s the Humm-Wadsworth Temperament scale was developed following a workplace violence incident in which an employee killed his supervisor. The test was basically designed to weed out unfortunates who were deemed “odd”. Employers would prefer people who came out of the test as “normal”. By the time of WWII the military in both the USA and the UK began to use psychological tests to determine who would make good officers. The Nazis were also keen on personality tests – especially those which apparently confirmed their ludicrous racial ideology: Aryans were considered more reliable and so forth.
The man credited with bringing intellectual rigour to the strange world of personality testing was Raymond Cattell, a British psychologist who moved to America just before the outbreak of WWII. Cattell, while at Harvard University, developed a model of human personality which is the basis for modern personality testing based on the Sixteen Personality Factor Questionnaire (16PF). The most popular personality test used today in the workplace, the Myers-Briggs test, is a variant of this.
Cattell started by determining how many words in the English language referred to characteristics of personality. He found about 10,000 in the dictionary which he was able to whittle down to about two hundred key words. Through extensive field work using questionnaires he reduced these to just 16 “personality factors”.
Subsequently, psychologists refined the 16 factor model down to just five key personality traits: openness, conscientiousness, extroversion, agreeableness and neuroticism. The last factor, neuroticism– sometimes called emotional stability – is the least generally understood. It tests the resilience with which we cope with negative emotions such as fear.
Individuals can be ranked on a scale for each of these factors and this approach has become entrenched in the psychological establishment in recent times. There is, however, a keen debate around whether the “Big Five” are really sufficient determinants of the very complex and amorphous thing that is a human being’s personality. How can the Big Five capture, for example, stubbornness or irritability? In any case, psychologists have known for a long time that there are significant correlations between each of these five factors (for example, between extroversion and conscientiousness), so they are not truly independent variables.
How CA applied their personality model
In America, it turns out that Democrats “like” Harry Potter on Facebook while Republicans “like” camping and the Bible. That does not mean, of course, that all campers are Republicans. But if you could measure personality by analysing Facebook likes and interests then you would have a very powerful tool. That is what Alexander Nix, the founder of CA, set out to do.
A young psychologist at Cambridge University by the name of David Stillwell had devised a personality questionnaire which was made available to users of social media as a fun quiz app called My Personality Test. The data created was then donated to Dr Stillwell’s academic research programme.
Once you have profiled the personality type of each Facebook user you can target that user with a message that plays on his or her personality type. This is called micro-targeting. For example, for a user who is high in conscientiousness and neuroticism, you send them a message that is both rational and fear-based.
In 2016 Alexander Nix attended a conference of the National Rifle Association (NRA) – often referred to as America’s gun lobby. He explained how they could influence opinion on gun ownership in America by micro-targeting their messages according to recipients’ personality types. The NRA could get a pro-gun rights message through most effectively to neurotic people by targeting them with a fear-based message for example about the prevalence of burglary; whereas conscientious people would be better influenced by a message that focused on tradition – we’ve alwatys been free to bear arms in this state.
CA worked out that you didn’t even need to get Facebook users to fill in the questionnaire. They could just download users’ Facebook profiles (digital footprint) and map them to their “Big Five” personality model. David Stillwell declined the offer of a job at CA, so instead Alexander Nix recruited Dr Alexandr Kogan, a Cambridge academic who was already a consultant for Facebook. Dr Kogan was tasked to create an algorithm that could map Facebook likes to the personality model.
It has to be said that when Dr Kogan was grilled by the Digital, Culture, Media and Sport Committee at the House of Commons on 24 April this year he played down his algorithm’s effectiveness. Other academic observers have also expressed doubt about the validity of Dr Kogan’s work.
What came out of the Parliamentary enquiry was that Facebook uses its own systems of modelling personalities based not on the Big Five factors but on hundreds of different dimensions. CA was actually behind the curve, applying late 20th-century psychographics to the digital age.
In any case, our Facebook “likes” are often misleading. On Facebook people tend to project an idealised image of themselves, so they tend to be as it were on their best behaviour. It turns out that our web browsing history and our search logs are much more revealing about us than our Facebook “likes” because these are not so easily self-moderated.
Increasingly, the fall of Cambridge Analytica looks overdone. The real scandal here – if indeed there was one – was the fact that CA acquired the Facebook data in the first place.
In 2009 the BBC launched the Big Personality Test which was promoted by such luminaries as Professor (Lord) Winston. Nearly 400,000 people responded to the BBC’s questionnaire. Some of the results of this experiment were extremely difficult to interpret but one outcome was clear. We tend to share many personality characteristics with our physical neighbours: personality traits are clustered together
Never miss an issue of Master Investor Magazine – sign-up now for free!
. Of course, those of us who always believed in the primacy of national culture instinctively knew this to be true.
But do people choose to live in places which are more congenial to their personality types, or does the local environment mould their personality? Needless to say, that is a tricky question. What we can say, however, is that there are correlations between certain personality traits and political affiliations. Openness (the preparedness to try new experiences) is naturally associated with social liberalism. The areas of the UK which voted more strongly in favour of Brexit are apparently lower on openness.
Political conservatives tend to score lower on openness and higher on conscientiousness. Unsurprisingly, residents of the Eastern coastal states of the USA and of California score higher on openness than people who live in the states-between. There is even academic research in the US that attempts to show that conservatives have tidier and cleaner bedrooms than liberals!
The other tricky question is whether your personality moulds your political values or whether it is the other way round. Or is it our genes that cause both? All this is quite unclear. So CA’s idea that you could target people based on their personality profile with tendentious messages and thus change their opinions is highly questionable.
Moreover, there are many people who are politically and economically conservative and socially liberal (I am one of them); just as there are traditional working class communities – and even more frequently now, Muslim communities – who are politically socialist but socially conservative. So Cambridge Analytica’s technology seems increasingly dubious.
Personality is not the key to understanding political affiliation. Adherence to a political party is much more about identity – which community you identify with – than about personality. But there is one “personality factor” in particular which I find intriguing.
In practise few campaigners and pollsters think that you can change a voter’s point of view completely during an election, though that sometimes happens. Their main task is to get apathetic voters down to the polling station at all – or to encourage potential opponents to stay at home.
One Cambridge University study formulated one ad for extroverts and one ad for introverts in an ad campaign for beauty products. Reportedly, the researchers were able to increase sales as a result. But in terms of political advertising the best response seems to occur when people who score highly in neuroticism are sent ads that play on their fears.
Neuroticism may be defined as one’s sensitivity to what might go wrong. This is particularly prevalent in debates around, for example, mass immigration – the fear that society will become less law-abiding, that we shall become more vulnerable to terrorism and that the indigenous community will lose its distinct identity. None of these fears is (in my view) irrational – but some people are more susceptible to them than others.
And – you’ve guessed it – Trump-supporting regions in the USA and Brexit-supporting regions within the UK tend to score higher on neuroticism. Although much of the Remain campaign was also based on promoting fear of the terrible consequences of voting Leave – hence the soubriquet Project Fear. But Project Fear –led by the current editor of the Evening Standard – did not succeed in translating dry economic prognoses of negative import into Remain votes.
Politics has always been about hope and fear. The internet and social media are just modern weapons in a very old war.
During the 2016 US presidential election the Trump campaign ran an estimated 50,000 Facebook ads each day targeting different elements within the US electorate. Mr Trump’s team did not need help from any data analytics firms such as Cambridge Analytica – they just availed themselves of the Facebook tools which any advertiser can use.
Apparently, the Hillary campaign spent even more than the Trump campaign on Facebook. That is why its key figures have been so loath to participate in the Mueller investigation into the (entirely unproven) Russian interference in the campaign.
In the UK general election of 2017, the Labour Party team wanted to run Facebook ads from the outset but, as Steve Howell, Deputy Director of Strategy and Communications revealed, at the beginning of the campaign Labour simply did not have the money.
As the campaign continued, however, fundraising improved and Labour spent £400,000 on Facebook ads. The data team at Labour Head Office had matched up canvassing returns (obtained by activists knocking on peoples’ doors) with Facebook profiles, and targeted messages to individual voters. One purpose of this was to reassure Labour Brexiteers in northern England that Labour was not pro-Remain; while metropolitan southerners were given alternative messages about the National Health Service, social care and so forth. There is absolutely nothing unethical about this: Labour just learnt more quickly how to use the tools available than their less digitally savvy Tory counterparts.
Facebook can classify people’s interests according to thousands of categories based on what kind of pages they like (or what makes them angry or laugh). As a result, Facebook knows what are our passions and our prejudices.
Any political party or organisation which has an email list can share that to Facebook who will provide a “look-alike” list of similar people with similar interest profiles. So if you can build a website that captures the email addresses of, say, one thousand people, you can then take those names to Facebook and they will sell you the names and addresses of 10,000 similar people based on their interest profiles. Reportedly, this was a tool that the Trump campaign used extensively.
Facebook has huge teams of data analysts who spend their professional lives trying to get the right ad in front of the right people, in the expectation that they will respond to it. But it’s not just Facebook. The Labour Party bought the services of Google during the UK election campaign too. By buying search terms they could get access to everyone who searched Google for a particular topic. As a result anyone who googled the terms “dementia tax” or “Brexit” during the campaign were automatically provided with a link to related content on the Labour Party website.
According to the analysts Adthena, the Labour Party bought links from Google to 539 search terms while the Tories bought links to about one tenth of that. Adthena claim to have the ability to map out the entire relevant landscape of key words for any given search activity. They provide their clients with insights about what their competitors are doing in order to help them to acquire new customers (or, in the case of political parties, voters).
Of course, many people do not know that the first few outcomes of each Google search are actually the result of paid advertising; just as when you go “Home” on Facebook the first posts on your wall are paid for by people who are after your business!
Which links pop up on Google and which ads you see on Facebook are tailored to your personal profile. But there is something else going on. There is a bidding war taking place in the background between advertisers whereby Google and Facebook give priority to the advertiser who pays most. That said, content that is known to be effective in a social media context – for example because many users have shared it – is preferred. Supposedly, the Trump campaign was able to hit more voters per dollar spent than the Hillary campaign because their content was more engaging.
The advent of micro-targeting has two important consequences. Firstly, the cost of a new political party or campaign group spreading their core message is considerably reduced. Secondly, campaigners can customise their message in such a way as to maximise its appeal to potentially each individual voter according to their interests and their personality. So issues become more important than catch-all slogans (Make America great again – For the many, not the few…).
If you don’t like the idea that political activists of all kinds are now in the game of manipulating you, there are things you can do. You can confuse Facebook and the advertisers who use its data by liking things you don’t really like or friending people you despise. You could visit websites which bore you rigid. But you’ll never be able to outwit Google long-term unless you become a Buddhist monk or nun and stop using the internet altogether.
An uncertain conclusion
Alexander Nix pulled the plug on CA on 01 May and it became defunct, having reportedly worked for various parties on over 200 elections around the world. The reputational damage of the media onslaught led by Channel 4 News proved fatal. But that is most certainly not the end of the data analytics industry. Indeed, Mr Nix and Robert Mercer’s daughters have already founded a new entity called Emerdata[ii].
The pseudo-scandal of Cambridge Analytica’s supposedly illicit involvement in the Brexit referendum and the US presidential election turns out itself to have been fake news. Doubly faked, in fact, since Cambridge Analytica vaunted its own importance even though there is no evidence that it had any impact at all on voter intentions in any election.
And the hypocrisy of the Labour Party back in March, when Mr Corbyn made the pseudo-scandal the centrepiece at PMQs (Mr Nix was a Tory donor – so what?) now seems astounding given the revelations of the extent to which they deployed social media in the 2017 general election campaign.
One thing we can be sure of is that data analytics as a field is here to stay. In the July edition of the MI magazine I’ll explain how data manipulation and cyber-security are two sides of the same coin. There are three types of cyber-attack: distributed denial of service attacks; hacking; and information wars.
By far the most intractable of these is the unceasing information war in which we all now live.