Fundraising was certainly important to Trump. Despite the fact that he announced on several occasions during the campaign that he was planning to fund his own race, once the primaries were over, he sought other sources of income. According to Federal Election Commission filings, by election day Trump had raised a total of $957.6 million. This came from his personal campaign, party and joint fundraising committees, and political action committees (PACs). All but $65.8 million of this money came from sources beyond Trump (Washington Post Staff, 2017).
Clearly, then, Trump did not neglect fundraising in any absolute sense. He did, however, neglect it in a comparative sense. Judged against Hillary Clinton’s in 2016 (and, indeed, both Barack Obama’s and Republican Mitt Romney’s in 2012), Trump’s fundraising totals were timid. Clinton raised $1.4 billion from all sources in 2016. Her personal campaign total of $623 million was nearly double Trump’s $335 million. In 2012, Obama raised a staggering $731 million from his personal campaign, but even the lackluster Romney managed to raise $474 million in that race (Washington Post Staff, 2017). Trump came nowhere near these totals, reversing the post-1990s trend toward ever-increasing spending in U.S. presidential elections. This might lead us to the conclusion that one of the chief goals of a data-driven campaign—acquiring large numbers of donations—was not a priority. But if we consider the type of donations Trump received, the precise nature of the Republicans’ approach to data starts to unfold.
It is now well established that if a campaign prioritizes online fundraising it can pursue, and is likely to receive, large quantities of small donations. This is because the marginal costs of raising money online are low when compared with other methods, such as receiving checks through the mail (Anstead, 2008; Bimber, 2003). A campaign can also use its email list to reach beyond the highly engaged activists who are more likely to make larger donations. Since 2004, the Democrats have consistently outperformed the Republicans on small-dollar fundraising. In chapter 6, I showed in some detail how this worked. Small-dollar online fundraising is based on three basic ingredients: an accurate email list generated from sign-ups at rallies and online after media events, particularly televised debates; careful use of email to target likely donors; and a donation “subscription” model that encourages repeat giving of modest amounts. Success in small-dollar fundraising is therefore a rough but reasonably good measure of whether a data-driven campaign is working.
In 2012, 32 percent of Obama’s money was from donations of $200 or less, dwarfing Mitt Romney’s 5 percent. In 2016, Clinton could not match Obama’s record; only 16 percent of her money came from small donations. But the most impressive story on fundraising in 2016 was how Trump turned around the Republican Party’s poor record on small donations. Trump received 26 percent of his funds from small donors, more than five times the percentage Romney received in 2012. Trump’s small-dollar haul of more than $100 million is the most any Republican candidate has received in sub-$200 donations (Washington Post Staff, 2017). This is all the more impressive given that Trump did not even begin his email fundraising drive until June 2016, after the primary and well into the presidential campaign (Goldmacher, 2016). Trump also received far less money from PACs than Clinton. While we need to bear in mind that Trump also received huge amounts of money from large donations, it is clear that he also managed to reach beyond the usual suspects that had fueled the fires of Republican campaigns before 2016. Clearly, Trump’s digital campaign worked in this regard. But what shape did it take more generally?
THE EVOLUTION OF TRUMP’S DIGITAL STRATEGY
The world of digital campaigning is notorious for its toxic blend of hype and secrecy, not least because the firms that sell their services to candidates often try to protect their commercial reputations. This is especially true of the closing stages of a campaign, when firms set out their stalls to attract business at the next election cycle. We need to bear this in mind when trying to establish what we know about Trump’s digital strategy (Karpf, 2016b). It is also worth noting that, throughout the campaign, the vast majority of opinion polls showed that Trump was going to lose (RealClearPolitics, 2016). There was little in-depth reporting from inside Trump’s war room because journalists have had a bias toward success stories in which digital methods help winners. Trump also kept tight control over access to his digital campaign team (Green & Issenberg, 2016). It is easy to see how the narrative caught hold that Trump’s campaign lacked sophistication and was stuck in a pre-digital time warp (Marshall, 2016; Sifry, 2016).
A few weeks after the election, an extended article about Trump’s supposed use of psychometric targeting appeared on the news website Vice. The piece, which had originally appeared in a German publication, Das Magazin, featured Cambridge Analytica (CA), a data marketing firm owned in part by billionaire Trump-backer Robert Mercer (Grassegger & Krogerus, 2017). CA was first hired during the primaries by Republican candidate Ted Cruz. When Trump won the primaries, he brought CA on board.
The Das Magazin article claimed that CA had used information from a large database of online personality questionnaires to predict and model the dominant personality profile for what it described as “220 million” people—effectively every adult American. The article also alleged that CA may have unethically integrated data from the myPersonality Project, a spin-out business from an academic study based at the University of Cambridge between 2007 and 2012.
While this might sound low key, myPersonality is, in fact, a very large repository of personal information about Facebook users. The project’s website states that, between 2007 and 2012, with users’ consent, it gathered some 7.5 million psychometric test results and downloaded 4 million Facebook profiles from a Facebook app that enabled people to find out their “personality types.” Users could take a test and give consent to myPersonality to harvest their Facebook likes (Grassegger & Krogerus, 2017; myPersonality, 2016). myPersonality was able to use test results, Facebook likes, and other data sources, such as profile pictures and descriptions, to establish correlations between user behaviors and personality types in the so-called “big five” traits model common in psychological research: openness, conscientiousness, extraversion, agreeableness, and neuroticism. In common with many academic projects, these data had been made available to more than 200 registered “collaborators” on the project’s website (myPersonality, 2016).
The Das Magazin article gave the impression that the psychometric method was used by Trump’s campaign to target individual voters with online messaging geared to voters’ personality types. For example, voters who scored high on neuroticism might be targeted with messages that asked them to consider what they would do if a burglar broke into their home. This caused a predictable whirlwind of commentary about the supposed previously unknown dark arts of Trump’s online campaign.
Yet the article gave no real evidence of precisely how psychometric targeting was used to mobilize voters, and the methods it described were publicly denied by CA itself (Confessore & Hakim, 2017). Instead, the practices the piece described, such as door knocking to gather data and scoring voters based on their preferred issues and whether they were likely to vote for Trump, were, by 2016, the stock-in-trade of all presidential campaigns (Karpf, 2017).
In February 2017, as Republican campaign workers began to do post-mortem interviews, it transpired that psychometric data did not play a major role in Trump’s digital campaign, and what role it did play simply derived from CA’s presence in the campaign war room, rather than a direct application of psychometric modeling (Teggart, 2017). As I show in the following, the company participated in informing the development of specific messaging for Facebook advertising and for doorstep canvassing. But the staples of this approach were demographic data, voter files, market research data, and the databases of voter preferences gathered from door-to-door and phone canvassing (Confessore & Hakim, 2017; Grassegger & Krogerus, 2017; Green & Issenberg, 2016). And in any case, the story about CA masks the broader point that Trump’s real innovation was h
is huge investment in Facebook advertising, irrespective of whether the ad campaign was at all informed by psychometrics.
In May 2016, as the primaries were drawing to a close, Trump went on the record in an interview with the Associated Press as saying that he believed data operations were “overrated” (Pace & Colvin, 2016). Trump said he preferred “big rallies” to generate word-of-mouth approval and free media coverage. As I showed earlier in this chapter, this was a key part of his campaign. Yet, much less noticed in Trump’s May interview were two points that revealed a subtler approach. First, Trump indicated that his campaign would be spending some “limited money” on data analysis to model voter turnout and the 270 electoral college votes he needed to beat Clinton. Second, he was putting together a deal with the Republican National Committee (RNC) to harness their voter targeting database and their volunteers on the ground (Pace & Colvin, 2016).
As the campaign went into the summer, both of these goals had been achieved. The RNC, which, between 2012 and 2014, had spent $100 million on improving its digital infrastructure (Kreiss, 2016: 168–203; Vogel & Samuelsohn, 2016), merged with Trump’s list its email operation of more than 6 million supporters and twelve dedicated support staff. The RNC agreed to raise funds using Trump’s name, but Trump agreed to allow the RNC to keep 80 cents of every dollar. Trump also integrated other Republican email lists, from Tea Party supporter networks and the list built by Newt Gingrich during his 2012 bid for the Republican nomination (Green & Issenberg, 2016).
Trump’s field office count (207) never came close to rivaling Clinton’s (489) but, then again, Clinton’s total was itself well short of the 790 offices that Obama had established in 2012 (Darr, 2016). But there is little doubt that during the closing four months of the race, Trump significantly intensified his digital campaign. He started to use rallies to grow his email list and raise funds, not by insisting on sign-ups at the entrance door (an Obama tactic) but by insisting that attendees register in advance on his website and then confirm their attendance via mobile phone. This handed the campaign hundreds of thousands of phone numbers (Green & Issenberg, 2016). According to email marketing company Return Path, which mines the content of 2.5 million consumer inboxes, a couple of weeks before election day, Trump’s email list had grown approximately 9 percent larger than Clinton’s. At one point it had been about 20 percent larger (BusinessWire, 2016).
The target of dismissive accounts of Trump’s digital operation was Brad Parscale, a Trump family friend whose San Antonio, Texas, company worked on building websites for firms, including Trump’s own. Parscale had no experience whatsoever in political campaigns, yet Trump made him his director of digital campaigning, partly on the basis that Parscale offered to build a campaign website for just $1,500 (Green & Issenberg, 2016).
Less noticed was that from July 2016 onward, Parscale was given significant amounts of money by the Trump campaign and was asked to recruit up to a hundred new employees. Parscale also started spending significant sums—over $7 million to begin with—on Facebook ads. These ads helped boost Trump’s fundraising haul of $80 million during July alone (Lapowsky, 2016b). While it is doubtful that Parscale recruited a hundred new staff so late in the race, the campaign’s investment helped him build a digital team for Trump. And, in any case, Parscale already had sixty employees in his business (Marshall, 2016).
During August 2016, with over 100 digital staff in place, Trump further strengthened his digital expertise when he appointed as his campaign chief executive Steve Bannon, director of the right-wing news site Breitbart, an outlet that had learned how to grow its readership through Facebook shares (Green & Issenberg, 2016).
By the close of the campaign, Parscale and his team had been given $90 million. By previous campaigns’ standards, this was a substantial sum. In fact, Trump spent a greater proportion of his campaign budget on digital than Clinton (Lapowsky, 2016a).
“LOOKALIKES” AND “DARK POSTS”
Most significant here is Trump’s innovation with Facebook advertising. On August 31, to coincide with Trump’s visit to Mexico, his campaign reportedly ran 107,000 different ads on Facebook, generating a haul that day of $5 million. The peak, however, came on October 19, the day of the third and final televised presidential debate, when the team reportedly ran 175,000 variations of their ads. The goal was to integrate Trump’s televised debate messages with a social media offensive (Lapowsky, 2016a; Mims, 2016). Lindsay Walters of the RNC stated that an average campaign day involved 40,000 to 50,000 Facebook ads (Goldmacher, 2016).
In an excellent example of how seemingly obscure technical changes in social media platforms can send ricochets through political communication practices, the new emphasis on social media ads was made possible by a significant development at Facebook. In the 2008 and 2012 campaigns, a major challenge for both parties was how to match their gathered email addresses with Facebook accounts. In 2014, however, Facebook opened up its API to other companies, such as consumer data brokers Datalogix, Experian, and Axciom. These organizations could now match Facebook accounts to full names, addresses, and phone numbers. This greatly improved the email-to-Facebook matching rate, from around 30 percent to 65–85 percent (Kreiss, 2016: 177). This enabled the RNC to start targeting Facebook advertising to members of their email list. It also helped them to use Facebook’s “Lookalike Audiences” feature. Lookalike Audiences is a platform tool that allows advertisers to use what they know about their existing audience to find other audiences that Facebook data reveal have similar preferences and attributes (Mims, 2016). The advantage to campaigns is that they can quickly grow by reaching people who are more likely to support their candidate, because campaigns know that these “lookalikes” share similar demographics and expressed interests to those who signed up for the campaign email list.
Trump also commissioned his own opinion polls from Trump Tower in New York, and, in the final month of the campaign, his digital war room in San Antonio was spending an additional $100,000 a week on their own voter surveys. These sources helped the campaign’s data scientists, including CA, build a model from email lists, voter files, and Facebook ad tests that identified 13.5 million voters in sixteen swing states whom they believed could be persuaded to support Trump. These were the so-called “shy” Trump supporters and younger, rural Republicans that CA argued traditional opinion polls and forecasts had failed to identify. CA also developed a forecasting model to help choose the best locations for rallies. These were based on the likely numbers of voters in specific locations whose data profiles indicated that they might be persuaded to vote for Trump (Green & Issenberg, 2016).
Some of this hinged on deterring likely Clinton voters from turning out. In 2008, a major objective for Obama had been expanding the electorate by mobilizing previously non-voting African Americans (see chapter 6). Trump’s campaign tried to put this in reverse through a concerted campaign of what one of its staff openly termed “voter suppression.” To illustrate this point, it is worth quoting at length from journalists Joshua Green and Sasha Issenberg’s account of the Trump war room:
“We have three major voter suppression operations under way,” says a senior official. They’re aimed at three groups Clinton needs to win overwhelmingly: idealistic white liberals, young women, and African Americans. [During the televised debate of October 19] Trump’s invocation at the debate of Clinton’s WikiLeaks e-mails and support for the Trans-Pacific Partnership was designed to turn off Sanders supporters. The parade of women who say they were sexually assaulted by Bill Clinton and harassed or threatened by Hillary is meant to undermine her appeal to young women. And her 1996 suggestion that some African American males are “super predators” is the basis of a below-the-radar effort to discourage infrequent black voters from showing up at the polls—particularly in Florida. (Green & Issenberg, 2016)
The aim here was to integrate the themes of Trump’s media appearances with a targeted online and broadcast advertising campaign. These themes, particularly Clinton’s 1
996 “super predator” remark about African American men, were reinforced by targeted radio advertising on African American stations, but more importantly by Facebook’s so-called “dark posts.”
A dark post, which Facebook also calls an “unpublished page post,” has become a common feature of advertising campaigns on the platform, though it did not attract much attention before 2016. These posts allow those with Facebook advertising accounts to use the platform’s “power editor” software to promote content by creating multiple ads targeted to the news feeds of specific groups of Facebook users. These might be based on simple demographics, for example men between the age of forty-five and sixty-five in Columbia County, Wisconsin. However, the platform also allows advertisers to search for keywords related to users’ profiles and likes. Crucially, these ads do not appear on a campaign’s Facebook page (Facebook, 2017). This feature is particularly important if a campaign is segmenting its target users into different groups and testing multiple, slightly different versions of the same ad to identify which ads produce the highest engagement rates—the so-called A/B testing that served Obama’s email campaign so well in 2008 (see chapter 6). Trump’s team experimented with different formats—video and still images, different subtitles, and text overlays. Facebook’s ad platform also now enables rapid testing of the engagement rates of dark posts and allows advertisers to run quick opinion polls to gauge user reactions. Just as important is that dark post ads have higher click-through rates than both Facebook banner ads and so-called “boosted” and “promoted” posts because users are less likely to see dark posts as spam.
The Hybrid Media System Page 44