The United States has low voter turnout, by international standards. In 2016, over 40 percent of people who were eligible to vote did not, including over 2.5 million people in Michigan, 3.5 million people in Pennsylvania and 1 million people in Wisconsin, the three states where Trump defeated Clinton by a collective 77,774-vote margin to win the presidency.

As part of its “100 Million Project,” the Knight Foundation surveyed thousands of nonvoters (defined as those who are eligible but not registered, or registered but have cast no more than one ballot in the last six national elections) in 10 of the most competitive states to better understand what could be the most pivotal portion of the electorate, especially as 2020 may be on track to have a record-shattering turnout.

Whereas nationally more nonvoters say they would rather vote this November for a generic Democratic nominee than vote for Trump, in Arizona, Florida, Pennsylvania and Virginia, more nonvoters say they would vote for Trump than a Democrat. In Georgia, which Trump won in 2016, more nonvoters say they would support the Democratic nominee than Trump, and in the other battleground states polled, neither Trump nor the Democratic nominee had an advantage—or the advantage was below the margin of error.

“It is an irony” says Daniel Smith, a professor of political science at the University of Florida who has done research on how Republican lawmakers have attempted to restrict the vote in electorally competitive states, when asked about the results. “But I’m not that surprised,” he added. “I worked as a poll worker in Florida in 2016, and anecdotally, I can tell you that the people who were not on the voter rolls in my precinct … were disproportionately more likely to be Republicans.”

Smith says that the notion that Republicans are concerned with voter fraud while Democrats care more about democratic principles may be based on recent evidence but should never have been taken as an indication of the parties’ basic values. “Both sides are doing this with respect to strategy as much as some inherent principle of the sanctity of the vote or the accessibility of the vote,” Smith said.

Andy Bernstein, the executive director of HeadCount, a nonpartisan get-out-the-vote organization, told POLITICO: “It’s a very sad thing that voting itself has been politicized.” “There’s a perception that any get-out-the-vote effort is somehow a partisan effort,” Bernstein said. “Voter registration has become perceived as in and of itself a means to a political end.” Any organization like his, which partners with cultural figures and events to register voters, Bernstein says, “is going to be perceived as having left leanings.”

On the other hand, it’s not hard to associate voter restrictions with the right. In 2018, the Supreme Court ruled by a 5-4 vote, with all the Republican-appointed justices in the majority, in favor of the Republican secretary of state of Ohio in a case over a practice that’s colloquially described as “use it or lose it”—when a state strikes inactive voters who don’t respond to a change-of-residence notice from its voter rolls.

“Is it a principle that we want to have clean voter rolls?” Smith, the University of Florida professor, who gave expert testimony in the 2018 Supreme Court case, asked rhetorically in an interview with POLITICO. “Or is it a strategy of trying to limit the size of the electorate for partisan advantages?”

Ironically, the strategy may be misguided for Republicans. At least with regard to Trump.

In Pennsylvania, one of several states where the conservative group Judicial Watch is pressuring officials to “clean” their voter rolls, the Knight survey shows far more nonvoters say they would vote for Trump than for a Democratic nominee in 2020. These very people could find themselves disenfranchised by the efforts of an organization that the president has praised and amplified.

Similarly, in Arizona last year, Republican state lawmakers attempted (but didn’t succeed) to purge some 200,000 voters from the Permanent Early Voting List for missing two election cycles. Joel Edman, the executive director of the Arizona Advocacy Network, told NBC News: “Being on the PEVL has proven to be a strong driver of turnout among those least likely to vote. Removing those marginal voters from the PEVL would only further depress their turnout rate.” According to the new data from Knight, Arizona’s nonvoter population is overwhelmingly more likely to support Trump than the Democratic nominee in 2020.

Tom Fitton, the president of Judicial Watch denies that their work is motivated by partisan goals. “People mistake ideology for partisanship. In the case of the left, you’ve got this ideology that’s against enforcing the rule of law. Our ideology is for the rule of law,” he told POLITICO. “Dirty voting rolls can lead to dirty elections. It provides additional opportunities for potential fraud.” A now-disbanded commission by the Trump administration found no evidence to support claims of widespread voter fraud. But an analysis by the liberal-leaning Brennan Center found that voter purges are prone to errors and can end up making it more difficult for eligible people to vote.

“It’s not just Democrats who don’t turn out to vote and then therefore might be taken off the rolls from these more restrictive states,” says Smith. “I suspect a story like POLITICO’s might have some of these officials start questioning what they’re doing.”

Trump’s re-election campaign has already hinted at reconsidering the usual strategy. The Associated Press reported last year that the Trump campaign views nonvoters as “an untapped stash of Republican support that can help him overcome stubbornly low poll numbers and his difficulties in winning over voters in the shrinking political center.” Bill Stepien, a senior political adviser to the Trump campaign, told reporters: “There’s a new math spurred by a new candidate at the top of his ticket. And I think we need to throw out the old way we look at how elections are won and lost.”

Furthermore, last November, Justin Clark, a senior political adviser and senior counsel to Trump’s re-election campaign, reportedly said privately: “Traditionally, it’s always been Republicans suppressing votes in places. Let’s start protecting our voters.” He clarified his remarks to the Associated Press, saying that he was referring to false accusations of the GOP engaging in voter suppression. “Neither I nor anyone I know or work with would condone anyone’s vote being threatened or diluted, and our efforts will be focused on preventing just that.”

One possible theory for why a generic Democratic nominee might be less popular than Trump to nonvoters in some battleground states is that nonvoters in those states are significantly whiter than nonvoters nationwide. According to the Knight data, in the 10 battleground states, there are 10 percent fewer black or Latino nonvoters and 10 percent more white nonvoters than there are nationwide.

Another theory for why Trump may be more popular than a generic Democratic nominee among nonvoters is that the political parties themselves are very unpopular, and nonvoters are excited by non-traditional candidates, like Trump.

The Knight survey results show that the vast majority—about three-fourths—of nonvoters are registered to vote. The most common reason to not be registered is lack of interest. The No. 1 reason voters think other people don’t vote is because they think nonvoters don’t believe their vote matters. In reality, the No. 1 reason nonvoters cite for not voting is that they don’t like the candidates. This is especially true in the battleground states.

“It makes sense to me the idea that nonvoters or unreliable voters don’t have a strong attachment to either the Republican Party or the Democratic Party but may have an attachment to Donald Trump,” says Meagan Day, a staff writer at the democratic socialist Jacobin magazine. “This population has no love for the two major parties, but it’s possible for a charismatic change-oriented candidate whose campaign has a lot of vigor, to break through and win their affections where the sort of generic partisan affiliation wouldn’t necessarily.”

In fact, according to the Knight data, nonvoters do view both parties unfavorably, as do voters. But fewer nonvoters view Trump unfavorably than voters do, and more nonvoters have a favorable opinion of the president than have a favorable opinion of the Republican Party itself. Fewer nonvoters consider themselves liberal, moderate or conservative than voters do, with more nonvoters volunteering that they “don’t think in those terms.” Nonvoters are also less likely than voters to identify with either of the major political parties. And that’s even more true in the battleground states, where more nonvoters identify as independent or say they’re unsure.

In Day’s estimation, Bernie Sanders, like Trump, is appealing to people who don’t typically vote because he has managed to put some distance between himself and the Democratic Party establishment. “There are a lot of people who like Donald Trump who don’t necessarily love the Republicans,” she notes. “When we look at who doesn’t vote, we’re looking at people who are feeling alienated from the political process because they feel like there’s nothing on offer from them from either party. And they don’t get the sense that anybody wants to fight for them.”

Fewer nonvoters than voters, according to the Knight survey, are confident that the results of an election represent the will of the people, and many nonvoters say that’s because they think the system is rigged or corrupt. Voters who are not confident that the results of an election represent the will of the people cite campaign financing or the electoral college for their lack of confidence more than a rigged or corrupt system.

“It’s not left-right,” Day adds. “While Elizabeth Warren has some very progressive policy proposals,” she admits, Sanders’s campaign is the only one she feels is capable of reaching the nonvoting population in a significant way. And she may be right.

In a New York Times survey of swing state nonvoters, Sanders was the most popular Democrat, besting Trump by 4 percentage points, while Warren was actually 1 point less popular than Trump among nonvoters. The same survey showed that between Democratic voters and nonvoters who favor Democrats, fewer nonvoters consider themselves “very liberal” or say they want a candidate who “promises to fight for a bold progressive agenda,” but more nonvoters than voting Democrats want a candidate who “will fundamentally change America.” The Times survey also showed that more Democratic-leaning nonvoters than Democratic voters in swing states support single-payer health care.

In the Knight survey, far fewer nonvoters than voters both in battleground states and nationwide who say they think they’ll vote in 2020 cite “civic duty” as their top reason for doing so. More nonvoters than voters cite health care policy. Similarly, more nonvoters than voters point to health care as well as jobs and the economy as the most important issues to them; whereas more voters point to issues like immigration, gun control, climate change and racism. Knight asked nonvoters: What, if anything, could motivate you to vote in more elections? A plurality responded with “a candidate I believe in.”

“We can’t rely on high turnout alone,” Day said. Democrats could still lose in November, especially if many people who have been nonvoters up to now in key battleground states show up at the polls for Trump. Democrats, she believes, would have the best chance of motivating nonvoters if they nominate Sanders. “The ratio of nonvoters to voters is so high in the United States,” Day emphasized, “that just a small fluctuation in it can have serious consequences for a general election.