This Election Violates Everything We Thought We Knew About Data

Donald Trump famously thumbed his nose at data-driven campaigning, and it worked. Is our system broken?

David Karpf
Published in
6 min readNov 7, 2016


The lesson of the 2012 election can be summed up like this: Campaigning is a science, not an art.

Over the past few election cycles, campaigns have developed a laser-like precision. Hypotheses are formulated and tested; Voters are rated and targeted. The logic of “moneyball” has been applied to elections. Digital tools and data analytics help campaigns listen to, observe, and engage with their voters in deep and specific ways.

This 2016 election was supposed to be an important step in the forward march of progress, featuring campaigns that made significant advances in data and experimentation. Both Hillary Clinton and Bernie Sanders poured campaign resources into tracking supporter engagement and improving technology—building on the foundations Obama laid in 2012. We expected to argue over who was using communications technology best, and to keep debating whether the rise of micro-targeting was making campaign propaganda too dangerously effective.

In the aftermath of an eight-year Democratic hold on the oval office, Republicans concluded that they needed to build a better data operation. This year the parties would equalize. The Republicans would begin to catch up to the Democrats in modern campaign science.

Instead, we got Donald Trump.

It wasn’t too long ago that campaign strategy consisted, almost exclusively, of an endless wave of commercials—with unclear impact. But then campaigns began to create better databases, built mostly from the public voter file. Depending on which state you live in, these dockets include your voter registration, voting history, and some basic demographic information. By using this data to get a sense of the shape of the electorate, campaigns can better target their communication and resources. When Clinton campaigners go door-to-door, reminding people to vote on Tuesday, they have a pretty good sense of whose door they’re knocking on, and whether that person is a likely supporter.

A sophisticated, professional campaign operation won’t single-handedly win an election. Things like the quality of a candidate, the mood of the electorate, and the state of the economy matter far more. But running a campaign that uses data to drive decision making certainly helps. The general rule of thumb is that excellent campaign strategy can move the outcome of the election between 2 to 5 percent. That’s enough to win a close race, but not enough to prevent a landslide.

Until recently, the debate over data has focused on asking if it works too well. The trend toward better data and more precise scientific testing has been clear, and the question is, “what should we make of it?” Some academics, such as the techno-sociologist Zeynep Tufekci, worry that the coming age of “big data” would provide campaigns with such comprehensive information on voters that they would be able to surgically microtarget the electorate, arming themselves with near-perfect propaganda techniques. Others, such as Eitan Hersh, argue that the new reliance on data and experimentation just produces marginally more efficient political campaigns, and remains harmless so long as the data stays confined to the electoral arena. We have all been convinced that American political campaigns are trending toward better data, so we’ve busied ourselves with the “so what” question.

While we were busy debating, another option emerged. Donald Trump is a performance artist. He fine-tunes his message like a standup comedian: by taking his act on the road, trying it out in front of his audience, and seeing what draws a reaction. He has spoken openly about his contempt for data. Instead, he puts his confidence in rallies. His data operation is run by a guy who used to build discount real estate websites for Trump’s business.

The differences between the two campaign strategies are extreme. Trump’s campaign has spent more on hats than on polling. Trump has barely set up a field operation, even in critical swing states. Hillary Clinton’s digital team communicates a disciplined message through her social media channels. Trump’s social media managers seem focused on curtailing their candidate’s 3 am Twitter rampages. While the Clinton campaign has been running experiments to optimize its door-knocking and produce messages that persuade key demographics, the Trump campaign put one of its field offices under the direction of a 12-year-old.

Trump’s campaign isn’t a throwback to older campaigns, and it’s not an innovative approach to the traditional challenges of campaigning. It’s a bit like if an American football team showed up at the Cricket World Cup and, knowing none of the rules, jumped in and started whacking things.

None of this was supposed to work in modern political campaigning. And yet, here we are. Trump’s “rev-’em-up” style worked in the primaries because, as it turns out, his rally audience is very similar to the Republican primary electorate. And, in the general election, his partisan base gives him a natural floor of 45 percent of the electorate, so it is hard to tell whether it is working or failing now.

If Donald Trump manages to win on November 8, after running a campaign entirely free of the last decade of data-driven know-how, it raises questions about the efficacy of the entire system. A Trump win means, in effect, that decades of research designed to organize and influence voters can be overpowered by a chaotic, from-the-gut performance—if the performance is riveting enough.

Recently, the Trump data team has taken to arguing that it isn’t far behind the Clinton camp. In an interview with Bloomberg Businessweek, Brad Parscale, Trump’s digital director, pointed to the massive email list that the Trump campaign has amassed over several months. According to operatives, the list is roughly 9 percent larger than Hillary Clinton’s email list. “We knew how valuable this [email list] would be from the outset,” Parscale told Businessweek. “We own the future of the Republican Party.”

The size of the list is indeed impressive, but there is zero evidence that the campaign is using the addresses for anything more innovative than email blasts. The Trump campaign isn’t running experiments on its base, or using the voter file to inform how it treats its data. (As I noted last week, Parscale’s operation has all the hallmarks of “campaign tech bullshit” that tend to surface from trailing campaigns when election day draws near.)

Instead Trump uses the email list as an extension of the rally phenomenon. Trump knows how to read, ramp up, and react to an adoring crowd. He is perhaps the most talented individual performer we’ve seen run for the presidency since the 1980s. And the material that wins at Trump rallies and in email threads is different than what we’ve come to expect from a presidential campaign. Among converts, “lock her up” and “build that wall” are crowd-pleasers. That same crowd will listen in rapt attention while Trump attacks judge Gonzalo Curiel, in what has been labeled a “textbook definition of a racist comment.” They’ll forgive misogynistic comments as “locker room talk,” and they’ll retweet, forward, and donate to their favored candidate with wild abandon.

Do these performances persuade undecided voters? Do they mobilize uncommitted voters to turn out on election day? We can’t yet know, because we’ve never seen a presidential campaign like this before.

This election day is more than a referendum on two candidates. It’s bigger than a referendum on the first female president, or on the performance of the Obama administration, or on Hillary’s email server management. It will also serve as a referendum on what we’ve come to view as the modern campaign.

Hillary Clinton has a professionally run operation from top to bottom. From field operations to communications to advertising, her campaign is using state-of-the-art data and social science research to persuade and mobilize supporters on election day. Donald Trump has been doing something else entirely.

How much does running a data-driven campaign matter when it comes to luring voters to the polls? On November 8th, it seems, we’ll find out.



David Karpf

-Associate Prof, George Washington University -Author of “The MoveOn Effect” and “Analytic Activism”