Iowa Democrats had gotten their hopes up, and honestly, how could they not? On Saturday night, J. Ann Selzer—the most renowned pollster in Iowa, if not the entire country—released her final pre-election survey, finding that Kamala Harris was leading Donald Trump by three points in a state the former president had carried by eight in 2020.

The poll seemed to portend a big night for Harris not only in Iowa but across the Midwest, suggesting a surge of support from women that would virtually assure her election. It also found a pair of Democratic House candidates in Iowa leading Republican incumbents, pointing to a Democratic majority in the chamber.

On Monday night, as Democrats packed inside a gymnasium in Des Moines for a rally, Selzer’s survey was all anyone could talk about. “I know that was exciting,” Lanon Baccam, the Democrat running for the local congressional seat, told the crowd, which erupted in cheers at the mere mention of the poll, “but I don’t think anyone in this room is surprised.”

[Read: How to understand the election returns so far]

The following night, many of the same Democrats gathered for a watch party inside a hotel ballroom downtown, their hopes turning to nerves and finally to resignation as a far bleaker picture emerged. The Selzer poll was way off, and Trump was poised to win Iowa by his largest margin ever. Iowa Democrats haven’t had much to celebrate since Barack Obama’s victory in 2012, and last night wasn’t any different.

“Iowa has changed dramatically over the past 20 years. Republicans are in the advantage right now,” Bill Brauch, the Democratic Party chair in Polk County, which includes Des Moines, told me. “We hoped that would change someday, but it isn’t today.”

Democrats had been optimistic about Iowa for the same reasons they were optimistic across the country. After foregoing most door-knocking due to the pandemic in 2020, they had built a robust turnout operation that dwarfed the GOP’s organizing efforts, which Democrats saw little evidence of as they canvassed neighborhoods. Enthusiasm, Brauch told me, was “through the roof.” And indeed, he said turnout was high in Des Moines. But more voters went Republican than Democrats expected, cutting into the margins that Democrats needed to offset the GOP’s strength in rural counties, where Republican turnout was also high.

The dynamic was the same across the country as returns came in: Despite strong turnout in many areas, Harris could not match Joe Biden’s 2020 performance in the counties that powered his victory over Trump. As of early Wednesday morning, the GOP had flipped at least two Senate seats, in West Virginia and Ohio, giving Republicans an all-but-certain majority, and they had a chance of ousting Democratic incumbents in several other battlegrounds that were too close to call. The House landscape was less certain, as Democrats still had a chance to flip enough GOP districts to recapture control.

They needed a net gain of four House seats for a majority, and although some of the party’s best pickup opportunities were in blue states such as New York and California, Democrats began seeing races in the Midwest trend in their direction in the closing weeks, opening up the possibility of more paths to the majority and larger gains nationally. But the Midwest surge did not materialize.

Democrats had poured late money into the two most competitive House races in Iowa, where they saw evidence that voters wanted to punish Republicans for enacting a state abortion ban—one of the strictest in the country—that took effect this summer following months of legal battles. In 2022, low Democratic turnout in places like Polk County helped Republicans flip a House seat, giving them all four in the state. The abortion ban, however, sparked hope among Democrats that Iowa would see the same blue shift that other states saw in 2022 after the Supreme Court overruled Roe—a belief that the Selzer poll reinforced.

Selzer has achieved a near-mythical status among political insiders. On Monday night, when I asked Tom Vilsack, the secretary of agriculture and former two-term Iowa governor, whether he believed her latest findings, he replied with a detailed history of Selzer’s past predictive successes. In 2008, her polling correctly forecast that Obama would defeat Hillary Clinton in the Iowa caucuses, and in both of the past two presidential elections, it came close to nailing Trump’s margin of victory when most other polls underestimated his support. “Anyone who doubts Ann Selzer when it comes to Iowa does so at their own risk,” Vilsack told me. “So do I believe it?” he added, referring to her Saturday poll. “Absolutely.”

On Tuesday night, the Democrats who showed up to rejoice instead realized that Selzer’s survey was just another poll—one of many that appeared to once again underestimate Trump’s support. As the night wore on, they held out hope that Baccam would defeat Representative Sam Nunn, a first-term Republican. (As of this writing, the Democrat in Iowa’s other competitive House race is narrowly trailing with nearly all precincts reporting.) But a podium set up for victory speeches stayed empty, and when, at around 11:20 p.m. local time the Associated Press called the race for Nunn, only a smattering of Democrats were there to see the news.

Brauch, the county Democratic chair, was at a loss to explain how his party fell so far short once more. “I don’t think any of us knows what the answer is,” he told me. “If we did, we’d be doing better tonight.”

For the third consecutive election, the nation remains divided almost exactly in half around the polarizing presence of Donald Trump.

Early this morning, the race between Trump and Vice President Kamala Harris appears likely to again come down to Michigan, Pennsylvania and Wisconsin, the same states that decided Trump’s 2016 and 2020 races by razor-thin margins. Trump held a narrow but clear advantage in all of them as of midnight.

In 2016, those three Rust Belt battlegrounds made Trump president when he dislodged them by a combined margin of about 80,000 votes from the “Blue Wall” of states Democrats had won in all six presidential races from 1992 to 2012; four years later, they made Joe Biden president when he wrested them back from Trump by a combined margin of nearly 260,000 votes. Now, with Trump regaining an upper hand across Sun Belt battlegrounds where Biden made inroads in 2020, the three Rust Belt behemoths appeared likely to decide the winner once more.

The results as of midnight suggested that those three states were tipping slightly to Trump; the patterns of returns looked more like 2016, when Trump beat Hillary Clinton in them, than 2020, when Biden beat Trump. Given that Trump appears highly likely to also win the Southeast battlegrounds of North Carolina and Georgia, and has a strong hand in Arizona, Trump will likely win the presidency again if he captures any of the three Blue Wall states. He would become only the second man, after Grover Cleveland in the late 1800s, to win the presidency, lose it, and then regain it again on a third try.

Not only are the same industrial-state battlegrounds at the fulcrum of Trump’s third race, but they remain mostly divided along very familiar lines. As he did in both 2016 and 2020, Trump is running up big margins in exurbs, small towns, and rural communities where most voters are white, culturally conservative people without a college degree. Harris is amassing big—though, in some cases, diminished—margins in the populous, well-educated suburbs around the major cities of Philadelphia, Pittsburgh, Detroit, and Milwaukee. The one potentially crucial shift from 2020: The exit polls conducted by Edison Research for a consortium of media organizations showed Trump making gains among Black and Latino voters, and especially men, not only in the pivotal former Blue Wall states but also elsewhere.

In many respects, the results available as of midnight were a reminder that even in a race involving a figure as unique as Donald Trump, in politics (as in Casablanca), the fundamental things apply. Since World War II, it has been extremely difficult for parties to hold the White House when an outgoing president was unpopular: The White House flipped partisan control when Harry Truman left office in 1952, Lyndon Johnson in 1968, and George W. Bush in 2008. Popular presidents haven’t always been able to guarantee victory for their party when they leave (the White House changed hands when relatively popular chief executives stepped down in 1960, 2000, and 2016), but unpopular outgoing presidents have usually presented an insurmountable obstacle.

If Harris ultimately falls short, that pattern would represent a big part of the reason. Biden’s deep unpopularity at the end of his term operated as a huge headwind for her. In the national exit poll, only 40 percent of voters said they approved of Biden’s job performance as president. In the battlegrounds, Biden’s approval rating ranged from a low of only 39 percent (in Wisconsin) to a high of 43 percent (Pennsylvania). Harris ran better than usual for a nominee from the same party among voters who disapproved of the outgoing president’s performance. But even so, the large majority of discontented voters in all of these states provided a huge base of support for Trump. In the national exit poll, fully two-thirds of voters described the economy in negative terms. Only one in four said they had suffered no hardship from inflation over the past year.

A lot has changed for Trump since the 2020 election. He launched a sustained campaign to overturn the results of that election, which culminated in the January 6 insurrection; Supreme Court justices he’d appointed helped overturn the constitutional right to abortion; he was indicted on multiple felony counts in four separate cases, and convicted on 34 of them; and he was hit with civil judgments for financial fraud and sexual abuse.

Yet the exit polls, at least, found remarkably little change in his support levels from 2020 among white voters across the battlegrounds. In Michigan, Wisconsin, and Georgia, his white support was virtually unchanged from 2020; he suffered a small decline in Pennsylvania, and a slightly larger one in North Carolina.

Compared with 2020, white voters with at least a four-year college degree moved slightly, but not dramatically, away from Trump in those five big battlegrounds. Harris won about three in five white women with a college degree, a big improvement from what the exit polls recorded in 2020. But Trump offset that by improving at least slightly since 2020 among white voters without a college education, who tended to give Biden especially low marks for his performance. Crucially for Trump, he retained overwhelming support among white women without a college degree everywhere except Wisconsin, where he split them evenly. Democrats had hoped those women might abandon him over abortion rights and a general revulsion to his demeaning language about women. Because those blue-collar white women appeared on track to provide Trump as big a margin as they did in 2016 and 2020, the national exit polls showed Trump winning most white women against Harris—just as he did against Biden and Clinton. That will likely be a subject of intense frustration and debate among Democrats in the weeks ahead, whether or not Trump wins the race.

Overall, the abortion issue benefited Harris substantially, but not as much as it did the Democratic gubernatorial candidates who swept Michigan, Pennsylvania, and Wisconsin in 2022, the first election after the Supreme Court decision overturning Roe. In that election, the exit polls found that Democrats Gretchen Whitmer in Michigan and Josh Shapiro in Pennsylvania won more than four-fifths of voters who said abortion should remain legal in all or most circumstances; in Wisconsin, Democrat Tony Evers won three-fourths of them. But this time—with the economy weighing on those voters—Harris won only about two-thirds of those pro-choice voters in Michigan and Wisconsin, and about seven in 10 in Pennsylvania. That slight shift might prove decisive. (In the national exit poll, Trump won almost three in 10  voters who said abortion should be legal all or most of the time; one-fourth of women who supported legal abortion backed Trump.)

Because abortion rights did not give her as much of a lift as it did the Democratic gubernatorial candidates in 2022, Harris did not appear on track to expand on Biden’s margins in many of the big suburban counties key to the modern Democratic coalition. She looked to be roughly matching Biden’s huge advantages in the big four suburban counties outside Philadelphia. But she did not narrow the roughly 3–2 deficit Biden faced in Waukesha County, outside Milwaukee, perhaps the biggest Republican-leaning white-collar suburb north of the Mason-Dixon line, as of midnight. In Oakland County, outside Detroit, Trump appeared on track to slightly narrow her margin, perhaps dealing a fatal blow to her chances.

In the well-educated county centered on Ann Arbor, Harris’s margin of victory seemed on track to decline from 2020, in what might be a reflection of youthful discontent over the support she and Biden have provided for Israel’s war in Gaza. In Dane County, Wisconsin, centered on Madison, she appeared in line to match only Biden’s 2020 share and not the even higher number Evers reached in 2022. Overall, in several of the suburban counties across the Blue Wall states, Harris appeared on track to finish closer to Hillary Clinton’s margins in 2016, when she lost these states, than Biden’s in 2020, when he won them.

The failure to expand on Biden’s performance in suburban areas left Harris vulnerable to what I’ve called Trump’s pincer movement against her.

As in both of his earlier races, he posted towering numbers in rural areas and small towns. Trump posted his usual imposing advantages in the blue-collar suburbs around Pittsburgh, and appeared to gain dramatically in the mostly blue-collar counties including and around Green Bay.

From the other direction, he appeared to further narrow the traditional Democratic margins in heavily minority central cities. That was particularly evident in Philadelphia. Exit polls showed Trump slightly improving among Black voters in North Carolina, Michigan, and Pennsylvania; that contributed to his win in North Carolina and gave him gains that placed him on the brink of flipping Wisconsin and Michigan as of midnight. In the national exit poll, Harris basically matched Biden’s vote share among white voters overall—but she fell slightly among Black voters and more substantially among Hispanic voters.

Almost lost in the ominous news for Democrats from the battleground states was the possibility that Harris would win the national popular vote, even if Trump also appeared likely to improve on his showings on that front from 2016 and 2020. If Harris did win the national popular vote, it would mark the eighth time in the past nine presidential elections that Democrats have done so—something no party has done since the formation of the modern party system, in 1828.

Yet even if Democrats achieved that historic feat, they faced the bracing prospect that Republicans could win unified control of the House, the Senate, and the White House while losing the national popular vote. Until the 21st century, that had happened only once in American history, in 1888; if it happens again this year, it would mark the third time in this century that Republicans will have won complete control of Washington while losing the popular vote.

Trump isn’t likely to view losing the national popular vote, if he does, for a third time (something only William Jennings Bryan had previously done) as a caution light. If anything, he will likely view the prospect that he could win the decisive battleground states by bigger margins than he did in 2016 and gain among voters of color as a signal to aggressively pursue the combative agenda he laid out this year. That includes plans for massive new tariffs, the largest deportation program in U.S. history, a purge of the civil service, and the use of the military against what he calls “the enemy from within.” Unless something changes dramatically in the final counts from the decisive states, American voters will have chosen, once again, to leap into that murky unknown.