Why Am I Seeing That Political Ad? Check Your ‘Trump Resistance’ Score.

In the lead-up to the 2020 presidential election, a voter analytics firm called PredictWise came up with a novel approach to help Democratic campaigns target persuadable Republicans: “Covid concern” scores.

To create the scores, the company first analyzed an immense data set showing the cellphone locations of tens of millions of Americans during the initial lockdown months of the pandemic. Then it ranked people based on their travel patterns.

Republicans whose phone locations showed that they left home a lot received high “Covid-19 decree violation” scores, while those who mainly stayed home received low scores, according to a PredictWise report. In follow-up surveys of some voters, researchers found that stay-at-home Republicans were almost as concerned about the pandemic as Democrats.

The firm said it had used the data to help Democrats in several swing states target more than 350,000 “Covid-concerned” Republicans with Covid-related campaign ads. In Arizona, PredictWise reported, the scores helped Democrats “open up just over 40,000 persuasion targets” for Mark Kelly, who was running for Senate. (Senator Kelly’s office did not respond to emails and calls requesting comment.)

Voter-profiling systems like the Covid-19 scores may be invisible to most people. But they provide a glimpse into a vast voter data-mining ecosystem in the United States involving dozens of political consulting, analytics, media, marketing and advertising software companies.

In the run-up to the midterm elections next month, campaigns are tapping a host of different scores and using them to create castes of their most desirable voters. There are “gun owner,” “pro-choice” and “Trump 2024” scores, which cover everyday politics. There are also voter rankings on hot-button issues — a “racial resentment” score, for example, and a “trans athletes should not participate” score. There’s even a “U.F.O.s distrust government” score.

Campaign and media consultants say such political-issue scores make it easier for candidates to surgically target messages to, and mobilize, the most receptive voters.

“We’re seeing not only U.S. congressional races, but State Senate races that are diving into this, and consultants using it to help them find those perfect targets,” said Paul Westcott, a marketing executive at L2, a leading voter database firm. He added that even some county campaigns were using scoring models to target voters on local ballot measures.

But the same nano-targeting that may help mobilize some people to vote could also disenfranchise others as well as exacerbate political polarization, political researchers say.

What is voter scoring?

Consumers are subject to a host of predictive scoring systems — hidden rankings based on factors like their demographic profile, socioeconomic status, online activities and offline interests.

Retailers and other services often use “customer lifetime value” scores to try to predict how much money individual clients might spend over time. Universities use “retention” scores to identify students at risk of dropping out.

Voter scores work similarly. They are intended to predict the likelihood that an individual agrees or disagrees with a particular party or political stance, like a belief in gun control. They are also used to predict a person’s likelihood of voting.

Ad tech firms often use the scores to help political campaigns narrowly target audiences on streaming video services, podcasts, websites and apps. Candidates, political party committees and advocacy groups also use the scores to help create lists of specific voters to call, text or canvas in person.

But researchers and privacy experts say that the scores are speculative and invasive, and that they could cause harm if they leaked to hackers or employers.

The process can involve classifying more than 150 million voters — using ratings like “gay marriage” scores or “non-Christian” scores — on personal beliefs they might have assumed were private. The scoring systems can also enable campaigns to quietly aim different, and perhaps contradictory, messages at different voters with little public accountability or oversight.

“In a democracy, we would like to know what promises are being made so that candidates can be held to account,” says Erika Franklin Fowler, a government professor at Wesleyan University who studies political advertising. “That’s harder to do if they’re saying different things to different people.”

How are voter scores calculated?

To calculate the scores, voter-profiling firms typically use commercially available dossiers thick with data on the election participation, demographics and consumer habits of millions of adults in the United States.

The files contain public information, obtained from state voter registration databases, like a person’s name, date of birth and address, as well as the election years in which the person has voted. They may also include a phone number, political party registration and race or ethnicity.

The voter profiles are often enhanced with commercially available details on consumers like: net worth, education level, occupation, home value, number of children in one’s household, gun ownership, pet ownership, political donations and hobbies or habits such as cooking, woodworking, gambling or smoking. Such details can be purchased from data aggregators that acquire information from customers’ loyalty-card records and other sources.

Next, profiling firms survey a representative sample of voters, scoring respondents according to their stances on issues like marijuana legalization. Firms then use machine learning to identity common characteristics across the dossiers — like low-income households, say, or a preference for low-fat foods — that correlate with voters’ stances.

The characteristics enable profiling firms to find “look-alike” voters in their files. Then they often calculate scores on issues like climate change for all the voters in their files.

What do voter scores look like?

Voter-profiling companies each have their own proprietary ranking systems. But they typically do not allow voters to see their scores.

One prominent conservative firm, i360, offers a number of scores, including a “Marriage Model” that ranks voters on a scale of 0.0 to 1.0. Scores near a full point indicate voters with a high likelihood of supporting “laws that preserve traditional marriage.”

HaystaqDNA, a predictive analytics firm that worked with Barack Obama’s 2008 presidential campaign, has posted an extensive catalog with dozens of proprietary scores on taxes, Covid and other issues. These include a “QAnon Believer” score, ranking people based on whether they believe a “deep state” within the federal government operated child-trafficking rings.

TargetSmart, a prominent progressive firm, developed a “Trump Resistance” model, which gives voters scores between 0 and 100 based on their likelihood of opposing Donald Trump. In a statement Tom Bonier, TargetSmart’s chief executive, said the company used public and commercially available data to help campaigns reach voters on “the issues they care most about.” The firm did not respond to questions about its voter scores.

Despite the marketing of these scores, people’s voting histories and political party affiliations remain the best predictors of their voter behavior, political researchers say.

“There’s a lot of hype in this space,” says Katherine Haenschen, an assistant professor of political science at Northeastern University who studies how digital communications affect voter turnout. “The most likely predictor is what the person has done in the past.”

Where did voter targeting start?

Trying to target and sway voters is an electioneering practice that dates at least as far back as 1840. That was the year Abraham Lincoln helped write a campaign circular for the Whig Party that laid out a plan for identifying and mobilizing individual voters.

The Lincoln directive, which ran in newspapers, instructed local party committees to “make a perfect list of all the voters” in their districts and ascertain “with certainty for whom they will vote.” It also treated undecided voters differently, instructing party committees to “keep a CONSTANT WATCH on the DOUBTFUL VOTERS” and try to “enlighten and influence them.”

The advent of computer modeling helped automate voter targeting, making it more efficient.

In the 1960s, a market researcher in Los Angeles, Vincent Barabba, developed a computer program to help political campaigns decide which neighborhoods to target. The system overlaid voting precinct maps with details on individuals’ voting histories along with U.S. census data on household economics, ethnic makeup and family composition.

In 1966, political consultants used the system to help Ronald Reagan’s campaign for governor of California identify neighborhoods with potential swing voters, like middle-aged, white, male union members, and target them with ads.

Critics worried about the technology’s potential to influence voters, deriding it as a “sinister new development dreamt up by manipulative social scientists,” according to “Selling Ronald Reagan,” a book on the Hollywood actor’s political transformation.

By the early 2000s, campaigns had moved on to more advanced targeting methods.

For the re-election campaign of President George W. Bush in 2004, Republican consultants classified American voters into discrete buckets, like “Flag and Family Republicans” and “Religious Democrats.” Then they used the segmentation to target Republicans and swing voters living in towns that typically voted Democrat, said Michael Meyers, the president of TargetPoint Consulting, who worked on the Bush campaign.

In 2008, the Obama presidential campaign widely used individualized voter scores. Republicans soon beefed up their own voter-profiling and targeting operations.

A decade later, when Cambridge Analytica — a voter-profiling firm that covertly data-mined and scored millions of Facebook users — became front-page news, many national political campaigns were already using voter scores. Now, even local candidates use them.

This spring, the Government Accountability Office issued a report warning that the practice of consumer scoring lacked transparency and could cause harm. Although the report did not specifically examine voter scores, it urged Congress to consider enacting consumer protections around scoring.