Donald Trump’s team knew they couldn’t win the 2016 election simply by persuading people to vote for Trump.
They also had to make sure Hillary Clinton supporters didn’t come out to the polls.
So the campaign and its allies used big data to target Black communities along Miami-Dade County’s historically disenfranchised Interstate 95 corridor. There, residents became some of the 12.3 million unwitting subjects of a groundbreaking nationwide experiment: A computer algorithm that analyzed huge sums of potential voters’ personal data — things they’d said and done on Facebook, credit card purchases, charities they supported, and even personality traits — decided they could be manipulated into not voting. They probably wouldn’t even know it was happening.
Internally, Trump’s staff described this part of their operation with a term that went beyond the usual strategy of negative campaigning.
They called it “deterrence.”
The campaign blasted these voters selected for deterrence — usually wavering Hillary Clinton supporters — with advertisements, disinformation and misleading messaging designed to convince them to lose faith in Clinton and not show up to the polls, according to an investigation by the Miami Herald and the U.K.’s Channel 4 News, which exclusively obtained a massive cache of internal Trump campaign data from 2016.
What exactly went into the selection algorithm isn’t known –– the Trump campaign’s machine-learning model remains a black box. But however sophisticated the model, it produced clear results: In Miami-Dade, more than 116,000 Black people identified by the campaign as potential voters were selected for deterrence, roughly half of all Black voters in the county.
That was almost twice the rate of deterrence for non-Black voters, a Herald analysis shows.
Not only were Blacks far more likely to be selected for deterrence, but even non-Black voters were more likely to be on the deterrence list if they lived in Black communities like those along Interstate 95 heading north to Broward County.
“The laser-like focus on suppressing Black turnout is clear,” said Dan Smith, a political scientist at the University of Florida who reviewed the Herald’s analysis of the campaign data. “It’s very striking.”
The data show that while women were always more likely than men to be labeled deterrence, nearly half of all women living in traditionally Black communities fell into that category, as opposed to one-third living elsewhere.
Even Hispanic voters — who were otherwise more likely to be selected for persuasive ad campaigns designed to get them to vote for Trump — were lumped into the campaign’s turnout-depression strategy in greater numbers along the I-95 corridor.
As Election Day approaches on Nov. 3, the data provide a window into how the Trump campaign undercut Clinton’s Florida campaign in 2016, and how it might be able to do the same in 2020, as Democratic nominee Joe Biden relies heavily on high Black voter turnout to win the nation’s most prized battleground state. Several key staffers who worked on Trump’s 2016 data operation have returned for his 2020 re-election bid. Polls show a tight race in Florida. Millions of dollars are being spent in ads by both sides. The Trump campaign in particular is attacking Biden’s 1994 crime bill, which contributed to the increased mass incarceration of Blacks, and his running mate Kamala Harris’ record as a prosecutor.
Trump’s approach in 2016 seemed designed to peel off Black voters and especially Black women, key voting blocs for Clinton.
Behind the 2016 deterrence campaign was Cambridge Analytica, a British data firm and political consultancy that collapsed two years after helping Trump to a shock victory when the company was revealed to have improperly obtained data on 87 million Facebook users.
Before going under, Cambridge Analytica quietly touted its ability to suppress the Black vote at a time when Stephen Bannon, eventually hired to lead the Trump campaign, was the firm’s vice president, a former employee said.
“[Cambridge Analytica] offered ‘voter disengagement’ as a service in the United States and there are internal documents that I have seen that make reference to this tactic,” whistle-blower Christopher Wylie told the Senate Judiciary Committee in 2018. “The firm would target African American voters and discourage them from participating in elections.”
But little concrete evidence had been offered to back up those claims until Channel 4 News obtained the campaign’s internal data, sharing it with the Herald.
While all modern campaigns rely on big data to find and turn out supporters, critics of the Trump campaign, including former employees of Cambridge Analytica, believe its focus on deterrence — rather than persuasion — amounted to a form of digital voter suppression. They say the data strategy mirrored efforts by Republican-controlled state legislatures to close polling places in urban centers, limit early-voting days or prevent felons who served their time from casting ballots.
The Trump campaign now says it didn’t seek to deter anyone from voting — contradicting what campaign officials have acknowledged, even boasted about, in the past.
“This is nonsense,” Tim Murtaugh, the Trump campaign’s spokesman, told the Herald Wednesday. “Every ad is meant to draw more supporters to President Trump. He has a far better record for the Black community than Joe Biden and it’s not even close.”
Murtaugh did not answer questions about how the campaign defined deterrence, what messages it targeted to those voters, or why the deterrence category even existed in its internal data if those voters were shown the same ads as persuasion voters.
The I-95 corridor heavily targeted by the campaign in 2016 includes the deeply impoverished municipality of Opa-locka, Miami’s Little Haiti neighborhood, and the more prosperous city of Miami Gardens, where the Miami Dolphins play. Heavily Black areas like Florida City and Richmond Heights were also marked for deterrence.
All are Democratic strongholds.
Disenfranchisement is nothing new along the interstate highway that city planners dropped into Black communities in the 1960s, fracturing thriving neighborhoods. At a Miami political rally in May of 1939, the Ku Klux Klan appeared with burning crosses and hung a dummy from a power pole with the words “THIS N***** VOTED” written on its clothes. Black voters were banned from participating in Florida Democratic Party primaries until 1946 and intimidated at the polls for far longer. In today’s Florida, more than 338,000 Black people cannot vote because of a felony conviction, about 15 percent of the state’s Black voting-age population, according to the Sentencing Project, a nonprofit.
“We’ve been fighting this forever and it’s depressing and it’s tiresome,” said Patricia Johnson, a 63-year-old Democrat who lived in Opa-locka during the 2016 election.
Johnson was selected for deterrence. She doesn’t remember any negative ads on Facebook — the social media platform is a favored vehicle for political campaigns to “micro-target” individual voters with ads specially designed to sway them — and cast her ballot for Clinton.
But nearly 60,000 other voters marked for deterrence in the area around I-95 did not show up to the polls — a drop of six percentage points among those same voters compared to 2012, according to a Herald analysis.
That decrease in turnout happened even as voting levels stayed flat in Miami-Dade, thanks to voters selected for persuasion and get-out-the-vote campaigns going to the polls at a higher rate than they did in 2012.
Overall, voters of all races and ethnicities who were marked for deterrence in 2016 saw their turnout drop 1.4 percentage points from 2012. The effects were more stark for Black voters marked for deterrence: Those voters saw their turnout fall by 8.2 points, compared to a 6.9 point turnout drop for Black voters overall.
Trump won Florida by more than 112,000 votes in 2016, benefiting from increased support among White voters in Southwest and Central Florida and lower Black voter turnout across the state.
It’s not clear how much of the result can be attributed to his campaign’s data operation, which also targeted other swing states where he scored upset victories, including Wisconsin, Pennsylvania and Michigan.
Without President Barack Obama on the ticket, Black turnout was certain to fall no matter what. And scholars disagree on how effective advertising can be at changing people’s long-held political opinions.
But the Herald’s analysis shows “there is definitely some evidence this micro-targeting strategy worked in suppressing the Black vote,” according to Smith, the UF professor.
“The high-deterrence folks have the greatest drop-off in turnout,” he said.
Republicans got walloped by President Obama’s data effort during his 2012 re-election.
They knew they had to respond. And so the Republican National Committee got to work creating a database that they say contains more than 3,000 data points on every voter.
Working out of an office in San Antonio, Texas, the Trump campaign and Cambridge Analytica pushed the data even further by moving suppression efforts into the digital sphere.
They called it Project Alamo.
In total, the Project Alamo archive obtained by Channel 4 News contains dozens of databases with thousands of tables on nearly 200 million American voters. It includes voter data that was gathered by the RNC, personal information purchased from commercial providers, and political donor lists.
The firm was bankrolled by hedge fund tycoon Robert Mercer and his family, who were major Trump donors in 2016 and closely linked to Bannon.
Brad Parscale, who was replaced as Trump’s campaign manager earlier this year, led the joint effort.
Survey and polling data were fed into the system during the 2016 race and then analyzed to inform the campaign’s messaging, events, door-to-door canvassing and get-out-the-vote operations.
“You put that [data] into a machine and then you start to learn,” Parscale told PBS’ “Frontline” in 2018.
The campaign used its algorithm to predict voters’ political beliefs, as well as their likelihood to cast a ballot. The profiles the algorithm developed also include detailed information on their income, race, ethnicity, place of origin, religion, language, marital status, gun ownership and more.
Then it segmented voters into eight categories for targeting:
▪ Core Clinton (committed Clinton supporters)
▪ Core Trump (committed Trump supporters)
▪ Get Out the Vote (Trump supporters who needed to be rallied to the polls)
▪ Persuasion (swing voters who could be convinced to vote for Trump)
▪ Deterrence (Clinton supporters who could be demotivated from voting)
▪ Disengaged Clinton (Clinton supporters unlikely to vote)
▪ Disengaged Trump (Trump supporters unlikely to vote)
▪ Deadbeats (apathetic voters with no clear candidate preference)
The designations helped the campaign figure out which kinds of ads would appeal to certain voters — and convince them to go to the polls or stay home.
Among the most prominent ads were a video clip of Michelle Obama that made it appear she was criticizing Clinton — she wasn’t — as well as messages attacking the Clinton Foundation’s work in Haiti. (Miami-Dade has the highest population of Haitian immigrants in the country.)
Facebook’s advertising platform played a key role in Trump’s success. Andrew Bosworth, a Facebook executive, said that Trump “got elected because he ran the single best digital campaign I’ve ever seen from any advertiser. Period. … Parscale and Trump just did unbelievable work.” A Facebook employee was even assigned to work with Project Alamo. (He now works for a progressive nonprofit.)
Parscale did not respond to requests for comment this week.
Neither did Facebook, which has reported earning $625 million from selling political advertisements in the last three months — equivalent to 7% of its second-quarter U.S. revenue, according to a review of the company’s Ad Library and financial reports.
Project Alamo focused on 16 battleground states, including Florida. Nationwide, Trump selected 3.5 million Black voters for deterrence, with nearly 658,000 in Florida.
One of them was Oliver Gilbert, the bow-tie-wearing, 48-year-old mayor of Miami Gardens — one of the country’s largest majority-Black cities and home to a growing Black middle class.
Nearly half of the Miami Gardens voters were on the deterrence list, or more than 30,000 people, according to a Herald analysis. Seventy percent were Black. To Gilbert, it’s simple, cynical math on the part of the Trump campaign.
“If Black people don’t vote, Hillary Clinton doesn’t become president,” Gilbert said.“Look, we can say what we want to about the modern nature of politics and what they’re doing on Facebook, but basically it comes down to the same old Southern strategy of convincing poor White people that they’re different than newly freed slaves.”
Overall turnout among Miami Gardens voters in the Trump data fell five percentage points, but it dropped almost seven percentage points for those labeled deterrence.
Clinton’s campaign in Florida was not aware of Trump’s deterrence strategy, said Reggie Cardozo, who served as deputy director for the state in 2016.
“We always knew that we needed to turn out the African-American vote,” Cardozo said. “But I can’t tell you that it was something that we knew was happening this blatantly and with so much detail, and that we did this and that to work around it. I don’t remember anyone saying, ‘Hey … they’re targeting this community and we need to inoculate ourselves.’ ”
Statewide, Project Alamo also disproportionately classified Black voters for deterrence.
Black voters accounted for 25 percent of Floridians marked for deterrence, even though they made up 13 percent of registered voters at the end of 2016.
“This is the new era of voter suppression,” said Oscar Braynon, a Democratic state senator whose South Florida district includes Miami Gardens.“It used to be reading tests. It used to be cops at the polls. …In the digital age [it is] these ads of disinformation to make you question the candidate who is naturally for you and the things you believe in.”
‘A VOTER SUPPRESSION OPERATION’
The messages the Trump campaign and allied super PACs used to deter Black voters were designed to lessen their faith in Clinton, not increase their support for Trump.
Some contained falsehoods and disinformation. Some did not.
What mattered is that they were designed to speak to Black audiences — and shrink the size of the Democratic electorate.
Brittany Kaiser, who worked as the head of business development for Cambridge Analytica, told the Herald the strategy was simple: “If you send them enough negative messages, then they might not go out to vote.”
Kaiser said she learned after the election that Black people and minorities were specifically targeted in what she called a “blatant, data-driven, voter suppression campaign,” something that she said was only talked about “in private.”
One ad, which was produced by a pro-Trump super PAC working with Cambridge Analytica, shows a young Black actress apparently making a campaign commercial for Clinton. “Hillary Clinton is honest and trustworthy,” the woman says before trailing off and asking for a cut.
“What’s the problem?” the voice of an imaginary director off-screen says.
“I can’t say these words,” the actress replies. “I don’t believe what I’m saying. … I’m not that good of an actress.”
Other ads were outright false.
In a video by the same super PAC, Make America Number 1, Michelle Obama is shown saying: “If you can’t run your own house, you certainly can’t run the White House.” The ad claimed Obama was talking about Hillary Clinton, and presumably making a reference to the Monica Lewinsky scandal. But a review of her full comments from 2007 showed the future first lady was discussing the challenges of raising her own family.
The spot aired for two weeks on Florida television stations and was also posted on Facebook. Obama has high approval ratings, especially among Black women.
In internal documents released last week by Kaiser, the super PAC wrote that the ad was “very effective in persuading women in our principal audience not to vote for Hillary Clinton.”
Messages attacking the Clinton Foundation with unproven allegations of corruption were specifically shown to Black voters in Little Haiti, Trump campaign sources told Bloomberg in 2016. In Little Haiti, more than 28,000 residents were marked for deterrence, the leaked campaign data shows. Half were Black.
The anti-Clinton messaging, which included a campaign stop by Trump, was “really hard for folks from the Haitian community to overlook in order to vote for her,” said Vanessa Joseph, chairperson of the Haitian American Voter Empowerment Coalition. “That’s not to say the entire Haitian community felt that way, but certainly enough of them did feel that way to either vote for another candidate, or just abstain from voting.”
Clinton received roughly 6,000 fewer votes than Obama in the state House district that includes Little Haiti, according to Steve Schale, a Democratic operative in Florida who analyzes election data.
Trump, however, didn’t pick up more support than GOP nominee Mitt Romney — meaning that he made gains because Haitian Americans didn’t turn out, not because they flipped to his side.
That matches what the Trump campaign was saying at the time.
One unnamed senior Trump campaign official called the data-and-messaging work part of a “voter suppression operation” shortly before the election, Bloomberg reported, saying the goal was to stop Clinton supporters from showing up to the polls.
“We know because we’ve modeled this,” the official said. “It will dramatically affect her ability to turn these people out.”
And the next year, Matthew Oczkowski, who worked at Cambridge Analytica and served as the Trump campaign’s chief data scientist in 2016, described deterrence voters as “folks that we hope don’t show up to vote.”
Oczkowski is now overseeing the Trump campaign’s 2020 data operation.
Many of the 2016 ads were so-called “dark posts” on Facebook, designed to be seen only by small groups of voters and then disappear forever, making it almost impossible to know what ads might have influenced voter behavior and how.
But Frederica Wilson, a Democratic congresswoman from South Florida, vividly remembers her constituents complaining about one Facebook ad in particular.
“Single mothers who were Black, who had children in voucher schools … they actually got personalized Facebook messages with a Black boy with gold teeth, gold frontal grille and his hairs standing up all over his head, walking down a school hallway with an AK-47 in his hands shooting at the children,” Wilson said. “And you could see the lockers and the bullets ricocheting from the lockers as the children were screaming. [The ad said] if you vote for Democrats, your children will be relegated to low-performing schools. Vouchers will be gone — so vote Republican.”
The Herald could not verify the existence of that particular ad. But Wilson said it was one of many examples of disinformation meant to suppress the Black vote in 2016.
No group of Americans was targeted more than Blacks by Russian “information operatives” seeking to sow discord in the election, according to a 2019 U.S. Senate intelligence committee report.
Sean Pierre Jackson, a Black Republican activist in South Florida, said he did not believe the Trump campaign relied solely on deterring Black voters and sought to persuade them as well. He pointed to figures showing that Trump received 8% of Florida’s Black vote in 2016, compared to 4% for Romney in 2012, a year the nation’s first Black president was up for re-election
“Hillary Clinton was the one that went out and out of her own mouth called Black men ‘superpredators,’ ” Jackson said, referring to comments Clinton made about criminals in 1996 and often used in anti-Clinton messaging in 2016, although she was not specifically referencing Blacks. “We wanted to make sure that we deterred every Black voter humanly possible from voting for Hillary Clinton … whether that meant them coming to vote for us or whether it meant them staying home.”
‘THEY MISJUDGED ME’
Even though there is some evidence that Cambridge Analytica’s approach worked in Miami-Dade in 2016, and perhaps elsewhere, it’s still not clear how well or why.
Donald Green, a political scientist at Columbia University whose research has found that online political advertising has negligible effects on voters, said the firm’s claims were overblown.
“The notion that you’re going to have this fantastically accurate targeting of individuals is based on the hype of those selling the databases,” Green said.
An investigation by the United Kingdom’s Information Commissioner found that the company’s claims about its predictive model were exaggerated and ineffective, according to The Associated Press.
Several voters in the Trump database told the Herald the information about them was incorrect.
“I have no idea where they could have come up with this data,” said Kent Pollock, who lived in Opa-locka in 2015 and said the campaign was wrong about what it believed his top issues were, although it got most of his personal information right. “They misjudged me.”
But there is some academic research that shows targeted online advertising can sway voter behavior.
People exposed to “voter suppression” ads in 2016 were 3% less likely to vote than those who hadn’t seen them, according to preliminary research by Young Mie Kim, a professor at the University of Wisconsin-Madison’s School of Journalism and Mass Communication.
The drop in voter participation was even greater for non-White voters living in majority-minority counties in swing states, who were the most likely group to see such ads.
“This shows clear racial targeting when it comes to digital voter suppression ads,” Kim said.
The ads, all from untraceable groups, included messages that encouraged voters to boycott the election; gave incorrect information about when, where and how to vote; promoted third-party candidates to voters with similar ideologies (for instance, promoting Green Party candidate Jill Stein to left-leaning Clinton supporters); or contained attacks on a candidate targeting the candidate’s likely supporters.
Whatever the results, campaigns are still spending millions of dollars on Facebook ads.
Four years later, Facebook ads tracked by the nonprofit group WhoTargetsMe shows the Trump campaign is again engaged in microtargeting using some of the same terminology from 2016, including the word “persuasion.” (No ads were outwardly labeled “deterrence,” although Sam Jeffers, the group’s co-founder, said he believes deterrence ads may be a subcategory of the persuasion grouping.)
During the 2020 cycle, disinformation targeting Black voters has been spreading through informal channels on social media, according to a report by NBC News.
One viral meme shared by Black conservative social media influencers claimed to show a portrait of Harris made up of “all the Black men she locked up and kept in prison past their release date for jail labor” during her time as a prosecutor. It’s been shared 23,000 times on Facebook, NBC found, despite the fact that it is actually composed of just a few faces repeated over and over again.
And earlier this month Twitter banned a network of accounts pretending to be Black Trump supporters.
“This is the evolution of digital media,” said Ashley Bryant, one of the leaders of Win Black / Pa’lante, a nonprofit created after the 2016 election to combat disinformation targeted toward Black and Latino voters, “and how it’s been weaponized against Black and brown voters.”