The Big Data Panic

by Felix Simon (feat. Heidi Tworek)

Medium
March 25, 2018

Even the juicy details from Channel4’s undercover investigation raise doubts about the success of the allegedly superior data analysis and targeting. In the secretly recorded meetings, the company’s top brass presented CA above all as a company that investigates opponents and tries to defame them with dubious material — dirty propaganda work, rather than data magic. CEO Alexander Nix and managing director of CA Political Global, Mark Turnbull, boasted of injecting propaganda “into the bloodstream to the internet” and described how their services could include bribing politicians by recording them accepting bribes or sending “very beautiful” Ukrainian “girls” to entrap them. A company that really believes in its ability to persuade voters with “Big Data Analytics” one should assume would not have to deal with such dubious tricks.

What The Evidence Says

Statements by political experts are one thing, but they are by far not the only ones doubting the supposed effectiveness of CA. One of them is Daniel Kreiss. He is a professor at the University of North Carolina at Chapel Hill and considered one of the leading experts on data-driven campaigning. “There are many reasons to be skeptical,” Kreiss says when I approach him on Twitter about Cambridge Analytica. “There is little research evidence that psychometric targeting is effective in politics and lots of theoretical expectations that it would not be.”

To understand this statement, it is worth taking a quick look at the exact method Cambridge Analytica claims to use. According to Cambridge Analytica, they were able to build psychographic profiles for millions of Americans: profiles that are a combination of demographic factors such as age, gender and location and character traits such as openness, conscientiousness, extraversion, agreeableness and neuroticism — the characteristics of the classic “Big Five” model. Political inclinations were then assigned to different varieties of the characteristics.

The character profiles of the allegedly 320 000 (the numbers vary between 200 000, 270 000 and 320 000) original users of the personality test app developed by Cambridge scientist Alexander Kogan — through which Cambridge Analytica had gotten the Facebook data — may be reasonably accurate. This, however, can no longer be said for the remaining 50 million profiles (Kogan himself speaks of 30 million) which CA amassed by hoovering up data from the friends of app users.

For them, character traits were merely modelled, reverse-engineered so to speak, using Facebook likes, true to the motto: “Johnny Walker, a personality test app user, has profile A and likes X and Y. Accordingly, his Facebook friend Mary Mueller, who also likes X and Y, probably also has profile A.”

As studies show, even this first step is based on shaky assumptions and of limited value over time as preferences change. The subsequent linking of these profiles with political tendencies is, however, even less precise — from what we know the “Big Five” can only predict about 5 percent of the variation in individuals’ political orientations. Or as Antonio García Martínez writes in WIRED: “It’s making two predictive leaps to arrive at a voter target: guessing about individual political inclinations based on rather metaphysical properties like “conscientiousness;“ and producing what sort of Facebook user behaviours are also common among people with that same psychological quality. It’s two noisy predictors chained together.”

“There is little research evidence that psychometric targeting is effective in politics and lots of theoretical expectations that it would not be.”

For Jessica Baldwin-Philippi, a professor at Fordham University, who has studied data-driven campaigning in recent years, it’s a reason to remain sceptical. “No one really knows how effective the vast majority of campaign tactics are, and these specific tactics seem significantly untested.” According to Baldwin-Philippi, one could model a ton of ‘personality traits,’ and “maybe they are generally correct,” but it was unclear how, for instance, Facebook “Likes” are reliably associated with character traits, and how these variables, in turn, can be associated with certain messaging strategies and political attitudes. “All of those steps multiply the difficulty to know. It’s broadly been the case that the most productive things to target on are publicly available prior data — voting data and census.” Psychographics? Nothing more than a marketing term.

And Kreis and Baldwin-Philippi are by no means the only ones who are not convinced by CA’s claims. Rasmus Kleis Nielsen, Professor of Political Communication and Director of Research at the Reuters Institute for the Study of Journalism at Oxford University, finds even clearer words: “Cambridge Analytica is a private, for-profit company selling consultancy services, and it is absurd to accept their self-interested claims as evidence of their efficiency,” he explains by email. According to Kleis Nielsen, various forms of microtargeting are useful for campaigns, but the effect should not be overestimated — a finding supported by independent research. “It is not a silver bullet and not a decisive factor in electoral outcomes.”

Apart from the scientific questionability of psychographic targeting, it is generally extremely difficult to influence people and change their political opinions, especially with microtargeting — a fact that researchers have stressed repeatedly in recent years. “Political persuasion is really difficult because so much of the [US] electorate is already sorted into partisan camps,” explains Kreiss. It’s a separation that runs so deep that it is very difficult to influence their opinion — no matter how much effort you put into it. And: How people cast their votes also depends, for example, on how they see the current economic situation, their educational background and much more. Political advertising has comparatively little influence in this context, as various studies have shown (example one and example two).

“Cambridge Analytica is a private, for-profit company selling consultancy services, and it is absurd to accept their self-interested claims as evidence of their efficiency.”

Finally, in what should put the last nail into Cambridge Analytica’s coffin, we should remind ourselves of how difficult it is to convince people of anything, in particular with advertising. The old hypodermic needle model — a model of communication suggesting that an intended message is directly received and wholly accepted by the receiver — has long been debunked. Instead, as communication researcher W. Russell Neuman argues in his seminal book “The Digital Difference”, communication is fundamentally polysemic: We do not do not necessarily understand messages, adverts or stories — even if they are tailored to our worldview or personalities — in the way the sender wants us to understand them.

On the contrary, any “media message, intended to be persuasive or otherwise, is not likely to stimulate a singular response, but rather a distribution of responses across a population of those who have encountered the message.” Or to put it more clearly: Just because a micro-targeted message wants voter Molly Average to stay away from the ballot box, it does not mean that Molly Average will understand this intention of the message (or understand and recognise it at all — as Neuman finds, we are surprisingly good at ignoring advertising and propaganda). Even if Cambridge Analytica was as effective as it claimed in targeting individuals with tailor-made content that doesn’t mean that it had any significant effect.

Ultimately, Neuman’s arguments also hint at a more fundamental problem in the entire debate around CA’s methods: We all too quickly portray humans as “sheeple” which are easily misled, especially when they’ve made decisions that markedly differ from our own or if we didn’t like the outcome of an election (the famous third-person effect). But most audience members are far from being the gullible and isolated pawns we claim them to be. To claim the contrary would be to deny them any agency of their own.

Why We Like To Blame Technology

If there is so little evidence that Cambridge Analytica’s methods have had any actual effect, why then are we still having this discussion? Just as everything else in this complex story, the answer boils down to a range of reasons.

First of all, we seem to have a tendency to panic over the effects of technology on us. As Heidi Tworek, a professor of history at the University of British Columbia has argued from a historical perspective, the fear of mass manipulation by new media and technologies is almost as old as humanity itself. Similarly, the fear of an “influence machine” that will corrupt us is hardly new. For some reason, panic about the ramifications of technology seems to be hard-wired into our systems.

And there are even more factors that the media hype around Cambridge Analytica ignores. Income inequality in the US, for instance, has grown steadily in the past and Trump, more than anyone else, presented himself as the “friend” of the “common man” — for many a reason to elect the populist to the White House. And it doesn’t stop here: There is the flawed campaign by Trump’s opponent Hillary Clinton which even some of her own staffers reportedly called weak; the split within the Democratic Party between Bernie Sanders and Hillary Clinton supporters, an organised anti-Clinton disinformation campaign, coordinated by Russian intelligence services and Wikileaks, and the decision by James Comey, the FBI director at the time, to notify Congress in an October 28 letter that he was reopening the inquiry into Clinton’s private emails.

Finally, if the US election has shown us one thing, it is that the US is still deeply marked by racism. For Baldwin-Philippi, this is a crucial point in understanding Trump’s victory: “Focusing on the question if Cambridge Analytica made Trump (or Brexit) win is the wrong question, and it distracts people from other important questions. The idea that CA somehow tricked people into voting for a candidate who has said and done incredibly racist and sexist things allows people to avoid grappling with the fact that people willingly voted for that candidate, and that those feelings are systemic in the US. While CA certainly acted improperly, in this case, they are also a convenient way to avoid this hard conversation.”

The Aftermath: What Should We Do With Facebook?

The blowback from the Cambridge Analytica scandal has damaged the firm, perhaps beyond repair. However, it has also alerted the public to the dubious, if not illegal, practices of many technology companies. What remains to be seen is what lessons can be learnt from this.

For Kleis Nielsen the focus should now shift to the greater environment in which the abuses of users’ privacy took place. “The real issue in the Cambridge Analytica revelations is not the probably limited effects their work based on Facebook data had on individual voters,” he explains, “but about how technology companies, political consultants, and political campaigns work together, and what kind of rules and norms we need to make sure that they work together in ways that enhance our democracy rather than undermine it.”

For others, the attention should first and foremost be on the company that many see as the enabler of Cambridge Analytica’s dark practices: Facebook. After the revelations, the company faces an onslaught of criticism for its careless handling of users’ data. Lawsuits and regulations loom.

For Kreiss, this isn’t too surprising. Speaking about the company’s reaction to the scandal he remarked that it demonstrated “a massive failure of leadership and organisation and especially a failure to think through why commercial advertising is different from political advertising and a failure to safeguard their platform against manipulation and their data against being compromised.” 
 
He’s by far not the only one to argue that now is the time to address the problems at the company that gathered all the user data in the first place. Baldwin-Philippi, too, opines that the discussion shouldn’t just focus on Cambridge Analytica’s dubious practices. “We need to discuss how technology firms collect, manage, use, and sell data,” she says. “What happened here is clearly a case of misuse, but it was also incredibly easy to do that because there was little oversight [from Facebook], and no punishment until this story was made public. If we want better data practices, we have to set regulations and policies that will do that.”

It’s a view echoed by others. One of them is Ifeoma Ajunwa, a professor at Cornell University and a faculty associate at the Berkman Klein Center at Harvard University. “We cannot merely rely on Facebook’s statements about how it plans to fix the problems inherent to its business model,” she says. Ajunwa argues that instead governmental agencies should step in to ensure that Facebook’s business operations conform to existing laws and business ethics. And, she says, that just regulating Facebook cannot be the solution.“This type of regulation is needed not just for Facebook, but for all platforms to which consumers entrust their personal data.”

Asked about the steps that should be taken to reign in Facebook’s and other tech companies, Ajunwa contends that it is paramount to strengthen the rights of consumers. “One first step is a legal mandate that data collection should be opt-in, rather than opt-out. As it stands, current opt-out policies mean that hapless consumers, who neither have the time nor the legal knowledge to properly understand the intentionally inscrutable terms of service that masquerade as informed consent, are blindly agreeing to the collection of very personal data.”

“We cannot merely rely on Facebook’s statements about how it plans to fix the problems inherent to its business model.”

At the present moment, there is no doubt that the Guardian’s revelations have set a story in motion that will keep academics, lawmakers and the public busy for a long time to come. Ultimately, we will probably never have any final certainty as to what exactly — and against all expectations — brought Trump to power and how important Cambridge Analytica was in all this.

“Without data from Facebook or CA about the outcomes of this targeting, it is impossible to know for sure whether the strategic communications CA said it engaged in was effective,” cautions Kreiss, an assessment that Ralph Schroeder shares: “It is likely that we will never definitely know the effect that Cambridge Analytica’s efforts had on the American election in 2016.” Yet, just as his colleagues around the world, Schroeder has no illusions that Facebook and others will be able to continue as if nothing had happened.

Even so, according to him, one problem will remain: “No regulation will address the more fundamental problem that populists can use social media to spread their ideas by circumventing traditional media” Cambridge Analytica may not be as dangerous as many assume. A reason to sit back this is not.