“It’s like a boomerang. You send your data out, it gets analysed, and it comes back at you as targeted messaging to change your behaviour.”
Brittany Kaiser’s words feature in The Great Hack, a documentary chronicling one of the most explosive scandals to erupt in the past five years: Cambridge Analytica (CA).
CA was alleged to have mined Facebook data from millions of people worldwide. The data was detailed enough for CA to create complex psychographic profiles of its subjects, to deliver pinpointed adverts to them and propel them into new behaviour patterns. The CA whistleblower Christopher Wylie described it as “Steve Bannon’s psychological warfare mind-fuck tool”.
“If we targeted enough persuadable people in the right precincts, then those states would turn red instead of blue,” said Kaiser, the former business development director for CA. “We bombarded them through blogs, websites, articles, videos on every platform you can imagine until they saw the world the way we wanted them to – until they voted for our candidate.”
The story had a seismic effect on political discourse. Two of the most unpredictable events in the recent political past – Donald Trump winning the US presidency and Brexit – were pinned on the company. The idea that the electoral system was undermined by CA’s underhand tactics was trumpeted by many factions of the media and political establishment. The argument that the operation amounted to a rupture in the fabric of democracy proliferated.
Three years since the scandal began to emerge, such ideas endure. CA is still thought by many to have played a key role in influencing both the Trump and Brexit votes. However, that position has become harder to maintain.
On 2 October a three-year investigation into Cambridge Analytica by the UK Information Commissioner’s Office (ICO) concluded with findings that were underwhelming to many, and devastating to some. After trawling through information including more than 700 terabytes of data seized at Cambridge Analytica’s London offices, the data regulator found no evidence that Cambridge Analytica had misused data to influence Brexit or aid Russian intervention in elections (the ICO had previously passed evidence of a possible Russian IP address to the National Crime Authority).
More damning was the finding that Cambridge Analytica wasn’t doing anything particularly unique. The information commissioner, Elizabeth Denham, told parliament that “on examination, the methods that SCL (a company that is corporately interlinked with Cambridge Analytica) was using were, in the main, well-recognised processes using commonly available technology”. This assessment jarred with reporting at the time that had imbued CA with Derren Brown-like abilities to tinker with perceptions and sway credulous masses.
Although the ICO investigation didn’t proffer a smoking gun, the elucidation of the CA’s dealings wasn’t without consequence. Facebook was fined £500,000 by the ICO as well as the maximum levy of $5bn (£4bn) in the US for its leaky data practices that led to 50 million users’ data being misused by CA. The Vote Leave and Leave.EU campaigns were also fined for breaking rules on spending limits. CA itself was fined £18,000 by the ICO for failure to comply with an enforcement notice to hand over data.
But for one of the central claims made about the company – that it had a decisive effect on the Trump and/or the Brexit vote – more evidence has accrued in the negative. Many were sceptical of the true power of Cambridge Analytica at the time, but in parts of the press and political establishment, the company’s supposed democracy-destabilising powers were amplified. Why, in the end, was the truth far more prosaic?
In many cases, the protagonists of the CA scandal seemed to have bought into the company’s own marketing spin. In The Great Hack, Kaiser’s remarks (quoted at the beginning of this article) were uncritically relayed, seemingly without any independent analysis of whether what she was saying was true or not. The claim that through targeted ads, CA successfully turned states from “blue to red” is an enormous one, but was there ever any evidence to back it up? A review of the documentary in the Economist at the time read: “So credulous is The Great Hack that if Cambridge Analytica had not shut down, its bosses would be using the movie as a testimonial”.
Wylie, the fluorescent-haired whistleblower, also appeared to have absorbed the hype. At a series of UK parliamentary committee meetings, he proselytised about CA’s ability to change hearts and minds. When asked at the Department for Digital, Culture, Media and Sport’s disinformation inquiry whether there was any evidence that the company’s work had affected the outcome of the EU referendum, Wylie answered floridly yet evasively: “When you’re caught in the Olympics doping, there’s not a debate about how much illegal drug you took, or ‘well he probably would’ve come in first anyway’, or ‘he only took half of the amount’. It doesn’t matter; if you’re caught cheating, you lose your medal […] You shouldn’t win by cheating.”
The ICO investigation highlighted another instance of where journalists had uncritically absorbed the talking points of CA’s salespeople. The organisation’s investigation found that the CA’s much-aired claim that it held more than “5,000 data points per individual on 230 million adult Americans”, appeared to be an exaggeration.
While perhaps shocking to some, the ICO’s findings were in line with what many experts in political science had suspected. “Many social scientists, at least in my sphere, have been saying for a long time that Cambridge Analytica was snake oil,” says assistant professor of political science and social data analytics at Pennsylvania State University, Kevin Munger.
Cambridge Analytica’s selling point was psychometric testing data that was layered on top of more commonplace commercial data sources. Cambridge University researcher Aleksandr Kogan developed Facebook personality tests and fed the harvested data to SCL, Cambridge Analytica’s parent company, through his business, Global Science Research. Kogan has denied that he knew the data use was illegal, saying he was assured by CA that it was a “normal use case of Facebook data”. Facebook, however, claimed that Kogan said the data would not be used for commercial purposes.
SCL’s interest in personality modelling was fuelled by a couple of high-profile scientific papers by two Cambridge University psychologists, Michal Kosinski and David Stillwell, that claimed to show a link between Facebook likes and sexual orientation and personality traits. As Carole Cadwalladr noted in the Guardian, the research attracted the support of the US military’s Defense Advanced Research Projects Agency and military contractor Boeing.
But the psychographic marketing technique turned out not to be the revolutionary tool it was envisaged as – in the context of political persuasion anyway. As the ICO found in its examination of internal emails, CA wasn’t able to accurately predict the personalities of people based on the information it possessed.
Even if it had, there is not an established scientific literature on the link between personality and how best to target political advertising for persuasion. There is some evidence that psychographic marketing can work in the context of consumer goods. For example, a 2017 study by Stillwell, Kosinski and others found that social media users were more likely to buy a beauty product if the advertising was tailored to match an individual’s extroverted or introverted qualities.
However, the research of Brendan Nyhan, a political researcher at Dartmouth College, New Hampshire, has pointed towards this not being currently effective for political advertising. Personality profiling and algorithmic targeting sound sexy, but there are two, far more mundane, characteristics that beat all others when predicting voter affiliation: party registration and voting history. Doctoral student in information, communication and the social sciences at the Oxford Internet Institute, Felix Simon, called the CA’s psychographic prong “mainly branding”.
This element was bolted on to more routine uses of commercially available data such as nationwide voter files and consumer data, along with other types of legally accessed Facebook data including a user’s likes and social network. These methods were not unusual at the time. Micro-targeting – using Facebook data to segment audiences – was actually pioneered by successive Obama campaigns.
It’s a technique that presidential campaigns are using this time around too. New York University’s Ad Observatory is tracking Democratic candidate Joe Biden and Trump’s political advertising on social media. Their data shows that Biden is using Facebook to target voters based on certain age brackets and geographical locations, combined with interests such as NPR, MSNBC, The Rachel Maddow Show, Democracy Now and Occupy Wall Street. (There is less data for the Trump campaign because voters have to be persuaded to install a browser extension to track the ads, which the researchers have found Trump supporters are less willing to do.)
There are disputes over whether the illegally harvested Facebook data was actually used in the targeting of US voters by the Trump campaign. But even if it was used in the US, as the ICO suspects, it’s unlikely that the work of CA clinched the election – because, personality profiling aside, research shows that political micro-targeting is simply not very effective at persuasion.
Studies examining political advertising tend to find small effects on behaviour. A recent study published in Science Advances, with an enviable sample size of 34,000, found that the effects of political advertising on behavioural outcomes including candidate favourability and vote were small “regardless of sender, receiver, content and context”. This challenges the typical assertion in this field that small average effects mask greater heterogeneity associated with certain factors.
“Advertising generally has small effects and while micro-targeted advertising can have some limited effects, research suggests that they are often tiny,” says Simon. “The best evidence we have for the existence of these small effects is around product purchases, but that’s a different ballgame than political opinions where it is much, much harder to persuade people.”
Kevin Munger says that social media advertising is most effective at fundraising or encouraging supporters to take action for a particular candidate or campaign. This is because there is a clear link between someone who has “liked” a Bernie Sanders Facebook page and their willingness to donate to his campaign. The connection between someone’s personality and how best to persuade them to vote Republican, for example, is far murkier.
This does not have relevance just for the Cambridge Analytica scandal. It has relevance to the evaluation of any attempts to sway voter behaviour with micro-targeting. There have been many claims made that the Russian government had a hand in helping to elect Donald Trump, based in part on an advertising campaign run by the Internet Research Agency on Facebook.
Aside from the fact that only 11 per cent of the ads in this comparatively tiny $100,000 Facebook campaign (Hillary Clinton and Trump spent a combined $81m on the platform) were related to the election, for the reasons stated above it’s very unlikely that Russian ads alone helped Trump get his foot in the White House. “Their ads aren’t any better than anyone else’s ads,” says Munger. “And no one’s ads are very good.”
A recent investigation by Channel 4 claims evidence that the Trump campaign tried to “suppress” some voters, who were disproportionately likely to be black, through Facebook micro-targeting. This suppression technique was purportedly delivered through negative advertising that attacked Clinton. Channel 4’s reporting on the topic stresses the fact that black voter turnout was lower in 2016 than it was 2012 in some of the areas targeted by Trump’s campaign. However, this risks confusing correlation for causation.
“That it actually worked (and on such a scale as suggested by C4) is highly doubtful and – based on everything we know about (targeted) advertising and attempts at persuasion – most likely pales in comparison to very real voter suppression efforts, which include removing polling stations, gerrymandering, or restrictive voting laws,” says Simon. An application of Occam’s Razor elevates explanations such as Clinton’s unpopularity compared to Barack Obama, and genuine voter suppression tactics above the persuasive power of targeted Facebook ads.
Theories about Cambridge Analytica’s role in the outcome of the 2016 US election or the Brexit vote quickly rose to prominence as a means of explaining two events that seemed shocking to many. “Both challenged the old hegemony which had seemed so stable,” says Simon. “Unfortunately, we are all somewhat prone to mono-causal explanations (‘the big data magic did it’) – it’s simply a bit easier to blame new technologies than to go for the longue durée approach where you accurately weigh the importance of other, long-term, structural factors in these events.”
All of this does not diminish the nefariousness of Cambridge Analytica and SCL. In fact, arguably the most egregious aspect of the company was their work in countries other than the US and UK. In an undercover sting, Channel 4 filmed CA CEO Alexander Nix boasting of the firm’s willingness to engage in unethical tactics including honeypots, bribes, blackmail and entrapment to swing an election. (For these reasons, he’s been barred from directing a UK company for seven years.) CA worked in at least 68 countries, but its activities in places other than the UK and US have received far less media coverage.
In one of her first articles on Cambridge Analytica – headlined “The great British Brexit robbery: how our democracy was hijacked” – Carole Cadwalladr draws a picture of an organisation that was “effectively part of” the UK and US military defence establishments and which one source referred to as “MI6 for hire”. The same source described wielding military-style psychological operations “to win elections in the kind of developing countries that don’t have many rules”. While it was far less discussed than the company’s influence on the outcome of Brexit and Trump, perhaps the biggest story about CA was the one that received the least attention.
[see also: Could the polls be underestimating Donald Trump?]