Microtargeting of low-information voters
A very seductive hypothesis: did Cambridge Analytica use its psychographics profiles to find the audience for our post-truth world? Did they target gullible voters on behalf of Trump’s campaign? Why is a Cambridge Analytica employee reading about the Need for Cognition Scale?
In a remarkable paper (unpublished as of March 1st 2017, summarised here), Fording and Schram analysed some survey results from January 2016 and highlighted the disproportionate support for Trump among “low-information voters”. They define those as voters who either score low in political knowledge or low on the so-called Need for Cognition Scale. The latter criterion means that they tend to make decisions based on heuristics rather than by thinking through the issue methodically. Of course, the two factors correlate: if you don’t think much about politics, you are unlikely to know a lot about it and vice-versa.[1]
Fording and Schram went on to assess the preference for Trump or Clinton segmented through those criteria. Because the data was collected in January, this is not expressed as a vote preference, but as a “warmth index” (or a difference of warmth), in the Y-axis.
Through comparison with a similar survey from 2012, they also showed that this phenomenon is not a permanent Republican advantage: it didn’t really benefit Romney, for instance. They also compared the effect to other effects that readers should be more familiar with.
Fording and Schram also explain in their paper how emotions create the heuristics needed to substitute for reasoned decision processes, which ends up emphasizing inherent anxieties in the individual.
Psychographics
Cambridge Analytica is a firm claiming to do psychographic targeting for political campaigns. Usually they hype up the OCEAN scale, which assesses individuals along five dimensions: Openness, Conscientiousness, Extraversion, Agreeableness and Neuroticism. They don’t actually need to directly interact with the individual they are profiling: instead, they acquire data about the individual from data brokers (demographics, social media, consumer data, lifestyle data) and run models calibrated against other individuals who have also completed actual psychological tests.
Despite all claims to the contrary by Cambridge Analytica, those individuals often don’t understand how their data will be used. Cambridge Analytica claims publicly to have 5000 data points on each of the 200 million voters in the US elections, enough to build an accurate psychological profile and address ads most effectively.
Cambridge Analytica worked for Ted Cruz until he dropped out [2, see correction], and they attribute Cruz’ success at the Iowa caucus to their psychological profiling. At around the same time, Cruz was already highlighting the appeal of Trump to low-information voters.
Note that Cruz rebounds on two aspects: uninformed and angry voters. The first one can be assessed fairly easily using readily available information. A good proxy would be education levels, which all campaigns are using in their targeting. It gets harder to target voters, say on Facebook, based on their anger level or how susceptible they would be to an emotional reaction. Could this be done?
Trump campaign
A Bloomberg article (written before election day) describes the flow of data between the RNC, Cambridge Analytica and the Trump’s campaign digital strategist, Brad Parscale:
Cambridge Analytica’s statistical models isolated likely supporters whom Parscale bombarded with ads on Facebook
Parscale has repeatedly said he was trying to get away from facts [3], and instead to provoke emotions in potential voters.
The big question then becomes: did Cambridge Analytica use the Need for Cognition Scale to prune the RNC list? This is a very seductive hypothesis, as it would explain many facts of this election, among them: the emotional grip Trump had on his electorate, his electorate’s weakness for fake news [4], and Trump’s immunity to any form of fact-checking.[5]
Could this hypothesis be construed as a left-wing conspiracy theory? Maybe, but I have one small but precise piece of evidence. It’s hidden in the video below, can you catch it? [video since removed from YouTube, but available here].
Supporting evidence, no proof
SkyNews ran a report on Cambridge Analytica a couple weeks before the election. That report includes one shot, at 0:59, that is particularly interesting. [6]
Through some online sleuthing, one can figure out that this employee is reading page 24 of the academic paper Method Effects and the Need for Cognition Scale [7]. I am not a psychologist, but I understand this paper to be comparing methods for refining raw survey data into a precise measurement on the Need for Cognition Scale. In other words, it’s the paper you would read if you were conducting Need for Cognition Scale assessments, and wanted to go beyond a rough, standard measurement. It’s really only justified if you have conducted lots of such surveys and are looking for nuances.
I find this compelling evidence that Cambridge Analytica had interest in assessing many individuals on that scale. What is lacking for a proof of our hypothesis is whether this was used for the Trump campaign. There might be a match between the reader of that article and the woman appearing a few seconds later in the video, with a Trump sticker, but that’s very weak.
What can be done?
I believe that American citizens have the right to ask from Cambridge Analytica or its parent company SCL for any of their personal data (including whether they were assessed on the Need for Cognition Scale), as these two companies are established in the U.K. for data protection purposes: here and here (on the same day). When I tried to encourage Americans to ask for their data in this way, just as I did, Cambridge Analytica changed its privacy policy to — illegally and retroactively — limit access rights of U.S. citizens (compare: old, new).[8]
I am looking for U.S. journalists who might want to do some digging here.
Update
Here is part of the conversation that led to their change in privacy policy (the rest of the exchange took place through PersonalData.IO or public tweets).
Update 2
Of course it didn’t fail: NakedCapitalism has referenced this piece saying: “This is a fancified reworking of the liberal trope that non-liberals are stupid”.
In simple terms, Fording and Schram said that there is a disproportionate support for Trump among people who either don’t use cognitive processes to make decisions, or don’t have political knowledge to start these processes. Even if you shortcut that to “stupid” (see footnote [4] why you shouldn’t do that), the parts of Fording and Schram’s paper that I rely on say “Stupid people disproportionately support Trump”, not “Trump supporters are stupid”. I myself suggested that Trump’s campaign might have intentionally targeted those voters as a consequence.
Thanks for reading! My name is Paul-Olivier Dehaye, I am a mathematician at the University of Zurich, and the co-founder of PersonalData.IO. I have also written about the geopolitical implications of data collection by Cambridge Analytica and its parent group, SCL, and contributed background research to an article about the transfer of data and techniques from Cambridge University to the Trump campaign, through Cambridge Analytica.
References
[1] Fording and Schram used data from the January 2016 American National Election Study Pilot Survey. To be clear, I am not citing here the political analysis in the Fording and Schram paper, which is indeed way too partisan for my taste. Any criticism of that part should be taken up with them. Instead, I am using their (very straightforward) analysis of the raw data. This is more objective, but not immune to criticism. They could, for instance, have selectively picked questions to construct proxy indicators and help their case. Or the data collection methodology could be flawed at the outset. I have not tried to question this, and would welcome any criticism of those aspects.
This being said, I am reusing the term “low-information voters”, which is loaded. I am using the same definition as Fording and Schram, because it is very objective, can be assessed from the data collected and matches the literal interpretation (“low on information” means either “low on relevant facts” or “low on processing of those facts”). I am fully aware that the term has been used and abused by both sides before, for political purposes (Fording and Schram detail this history, in a partisan way). I do hope the sensitivity of the term will itself attract more eyeballs.
[2] This followed the pattern of Robert Mercer, who is reported to be their part-owner and to have funnelled money to Cambridge Analytica through his SuperPAC, which also went from supporting Cruz to supportingTrump.
EDIT: I was told, but didn’t check, that Cambridge Analytica and Cruz parted ways a bit before him dropping out of the race.
[3] That strategy was also used for Brexit. Cambridge Analytica provided polling services to the Leave.EU campaign in the very early stages of the Brexit campaign: “Facts don’t work”, as the strategist Gerry Gunster put it to the man who funded the initial Brexit campaign, Aaron Banks. You can see Britany Kaiser of Cambridge Analytica sitting next to Banks and Gunster in the initial press conference of Leave.EU. Another organisation campaigning for Brexit became the reference organisation for that side of the campaign, and Cambridge Analytica was dropped from the team. The campaign kept the same strategy though. The full quote:
Banks engaged the Washington campaign strategy firm Goddard Gunster to advise Leave.EU; it swiftly identified immigration as the critical issue and told him how to exploit it. As Banks recalled in the week after the referendum victory: “What they said early on was, ‘Facts don’t work,’ and that’s it. The Remain campaign featured fact, fact, fact, fact, fact. It just doesn’t work. You have got to connect with people emotionally. It’s the Trump success.”
[4] In a context of active disinformation (for many reasons), if you are not critical of what you read and only react emotionally, this makes you very gullible. To be 100% clear, I am against disinformation, but not emotional reactions, which have a place in politics. It is also easy to see the highest reaches of the Need for Cognition Scale as purely positive and rational, but that is not the case. Humans have developed emotions and cognition for different circumstances, and both have their place. In addition, even in situation s where cognition is needed, high Need for Cognition individuals are often vulnerable to their own biases through “ad hoc hypothesis” or more simply over-rationalization.
[5] There could even be, but that is even more speculative, a feedback loop with Facebook’s algorithms, since these tend to favour content producing emotional reactions.
[6] To avoid any ambiguity, this report was filmed at the headquarters of SCL, the parent company of Cambridge Analytica. Based on a lot of additional evidence not presented here, there is little doubt in my mind that Cambridge Analytica is just SCL with a Delaware registration, and a few staff members embedded into the U.S. campaigns they were working for. A lot of the heavy technical lifting still happened in the UK at SCL headquarters.
[7] The International Journal of Educational and Psychological Assessment, December 2012, Vol. 12(1)
[8] Cambridge Analytica’s CEO and Data Protection Officer, Alex Tayler, have consistently shown a very weak understanding of EU data protection laws and how they work. In particular, Cambridge Analytica’s privacy policy has to grant a baseline right of access regardless of the nationality or country of residence of the data subject (admittedly, this is super complicated, but I have a lot of relevant evidence going in this direction).
