LatestInvestigationsShowsCritical ThinkingResourcesSubscribeDemocrats won the midterms on fewer votes than Kamala Harris lost with. Republicans came up more than 20 million short of Trump. I went looking for where everybody went.
Democrats won big on November 3rd. They've flipped 43 seats in the House so far, with 18 races the AP hadn't called by Monday. NPR titled its page "A blue tsunami led by Black voters washes over the South." The Washington Post went with "Historic voter turnout powered Democrats' blue wave."
From the headlines, it sounded like a sea change had happened in the US. And a lot of statistics started being cherry picked and shared via social media. I started sharing the same numbers - because it felt good. But a certain set of stats caught my eye and they didn't seem to line up. One was how different demographics voted and the other was how many people in those demographics voted.
So, I started researching all of the numbers and found that these midterms told a story more complex than the one being presented. This is that story.
Kamala Harris got 75.0 million votes in 2024 and lost. House Democrats have 65.0 million so far this year. Donald Trump got 77.3 million. House Republicans have 49.4 million.
In terms of votes - Dems are down about 10 million votes from 2024, and the GOP are down about 28 million. Notice that I'm talking about votes and not parties. This is not about the number of Democrats versus the number of Republicans - this is who got the votes.
Now that's settled, first let's talk about exit polls - cause every story you've read about who showed up last week came out of one, and this year's isn't the poll you think it is.
So how does a party win a wave with fewer votes than it lost with?
On election night the TV tells you who won. Then it tells you who voted - how many were women, how many were Latino, how many were under 30.
That second part all comes from one place, and it's called the exit poll.
For about twenty years it worked the same way. A company called Edison Research ran it for the big TV networks.
Pew Research Center, a research group that doesn't take sides, says that in 2016 Edison sent people to stand outside nearly 1,000 polling places and hand voters a paper form on their way out. About 85,000 people filled one out.
You can't catch an early voter or a mail voter at the door. So Edison called them - about 16,000 of them that year.
Most people don't take the form. The pollsters' own trade group says fewer than half of the voters who get asked say yes, and the ones who say no aren't always like the ones who say yes.
So the pollster has a stack of forms from the people who voted and also felt like filling out a form. That's a smaller crowd, and a pickier one.
So the pollster adjusts. Some answers get counted a little more and some a little less, till the stack looks more like the voters. That's called weighting.
The person at the door helps. Joe Lenski, who co-founded Edison, told Pew how it works. His people jot down a guess at the age, sex and race of everybody who walks past without stopping, and the poll adjusts for that.
They can't see the rest. "People don't have a D or an R stamped on their foreheads," he said, "and you don't know as they walk past you whether or not they have a college degree."
Then the real votes come in. The pollster goes back and adjusts the whole poll again till it matches the actual count.
So by the next morning an exit poll can't be wrong about who won. It's been set to the result.
What's left is its best guess at who those voters were.
A company called SSRS bought Edison in October 2025.
This year ABC, CBS, CNN, NBC, Fox, MS NOW and the Associated Press all used one poll, The Voter Poll by SSRS. NBC and CBS still call theirs an exit poll.
But most of it didn't happen at an exit. SSRS surveyed 81,290 registered voters online and by phone, starting two weeks before Election Day.
It found them on the voter rolls and in two survey panels. One of those panels signs up some of its people through internet ads.
People did stand outside polling places, in 15 states. They did 21,995 interviews, and those went into the state polls.
In SSRS's own table, the national sample is 81,290 people from the survey and zero from a polling place. The national numbers had no exit in them at all.
A survey taken before Election Day doesn't know yet who's really gonna vote. So the poll works it out. It checks people against the voter file, it calls them back, and it uses a model.
CNN's own fine print says the survey answers were "weighted by voters' probability of turning out."
Then, same as always, the whole thing gets matched to the real result. So it still can't miss the winner.
The new way might even be better. It's still a different tool, and a number from the old one next to a number from the new one isn't a trend.
On election night in 2000 the networks called Florida for Al Gore a little before 8 o'clock, off the exit poll. Around 10 they took it back.
After 2 in the morning they called it for George W. Bush - off bad vote counts that time - and around 4 they took that back too. The company that ran that poll, Voter News Service, was shut down in 2003, and Edison got the job.
In 2004 the early numbers leaked, and they had John Kerry winning. The national exit poll showed Kerry 51, Bush 48. Bush won by 2.5 points.
The pollsters' own review said the poll overstated Kerry in 26 states, most likely cause Kerry voters said yes to the form more than Bush voters did. Nobody called the wrong winner that year. The final poll got adjusted to match the count.
In 2016 the poll got matched to the result like always, and it still had the crowd wrong.
It said half of all voters had a college degree. Two years later Pew checked its own survey against actual voting records and got 37 percent.
The exit poll said 16 percent of voters were 65 or older, and Pew found 27. It said 19 percent were under 30, and Pew found 13. It had Black voters at 12 percent, and Pew had them at 10.
So the old exit poll ran young, and it ran a little heavy on Black voters. Edison changed its math in 2018 to fix some of that.
An exit poll is good for who won, cause it's set to the real count. It's pretty good at how a group voted, in broad strokes. The pollsters' trade group, the American Association for Public Opinion Research, says that's the main job - explaining why an election came out the way it did.
It's shakiest on how many of each group showed up. Pew says it's been knocked for years for overestimating how big a share certain groups make up, and it names nonwhite voters, young people and college grads. TIME said in 2018 that the thing exit polls are worst at - the breakdown of who voted - is the thing they get quoted for most.
And how many of each group showed up is the only thing I used it for.
Right after the election I made graphics out of the NBC exit poll. I took each group's share of the voters and multiplied it by the total vote - 155.2 million votes for president in 2024 and about 124 million ballots this year.
But the more I looked at the numbers, the more I found discrepancies.
Look at the New York Times. It makes its own estimate from voter files, precinct results and its own polling, and it has Black voters at 11 percent of the electorate this year. The networks' poll says 9.
The Times counts about 120 million voters. Two points of that is about 2.4 million people.
The two polls don't match on older voters either. The old exit poll had people 65 and up at 28 percent of voters in 2022 and 28 percent in 2024. The new poll has them at 36, and the Times has 34.
Did older voters jump from 28 percent to 36 percent of the voters in two years? Maybe. But we probably won't know how much any of these numbers are off before next summer.
What we do know is that older voters still make up a much bigger part of this election than their share of the country.
When I posted the demographics of voters a lot of people were upset that they showed the number of black voters went down since 2024 and were basically flat since 2022. So, I decided to dig deeper into those numbers. A poll is just an estimate, after all. But a voter file is a count - it's the state's own list of who voted. Only a few places have theirs out this early, so I started with those.
Georgia does. The Brennan Center, a law and policy institute at NYU, went through the state's early records and found about 24 percent of the Black Georgians who voted in 2024 skipped this one. For white Georgians it was over 29 percent.
Philadelphia posts turnout by ward. The city's 26 majority-Black wards cast 69.5 percent as many ballots as they did in 2024. The other 40 wards cast 68.2 percent, so that's a tie.
Chicago posts it by congressional district. The two majority-Black pieces of the city cast 27 percent fewer ballots than in 2024. The mostly white pieces cast 33 percent fewer. In Detroit, 209,924 people voted - that's 85 percent of what the city cast in 2024, when the whole country cast 79. Chicago is still counting mail, and in 2024 more of it came from the mostly white areas, so that gap may shrink.
Then I noticed something about how the exit polls could end up showing two things.
The exit poll shows that voting by Black women dropped the most between 2024 and 2026. It also shows that the percentage of Black women voting for Dems went up from 92 to 93%. They also show that they were a key factor in swinging elections in southern states when Black candidates were on the ticket. So, while fewer Black women voted, they showed up for the elections with candidates that they related to and they helped those candidates win.
Young voters are still the smallest slice of the electorate. In the Times's estimate, people under 30 were 12 percent of the voters this year, even though they're 20 percent of the people who are allowed to vote.
But more of them signed up. Tufts University's CIRCLE, which studies young voters, found 3 percent more 18- to 29-year-olds registered than at the same point before the 2022 midterm. For 18- and 19-year-olds it was up 8 percent.
One group went the other way. Young Black registrations were down 8 percent.
And the young voters who did show up swung harder than almost anybody. The Times has them going from Dems by 9 in 2024 to Dems by 42 this year.
Who voted in 2026, by age, against who is allowed to vote
Under 30
30 to 44
45 to 64
65 and up
Under 30: 12 percent of the voters, 20 percent of the people allowed to vote. 65 and up: 34 percent of the voters, 24 percent of the people allowed to vote.
The rolls tell their own story. In North Carolina, the state's own count shows 105,214 fewer registered Democrats on Election Day this year than on Election Day 2024. Registered Republicans were down 15,556. Unaffiliated voters were up 212,529.
So a lot of people didn't switch to the other party. They dropped the party label altogether.
Pennsylvania went the same direction, which I get to below. And in both states, Dems won some of it back during 2026 - but not all of it.
The biggest swing came from white voters - and they're about seven in ten of the people who voted, so a swing there moves the whole country.
White men voted Republican by 22 points for president in 2024, 60 to 38. This year they went Republican by 8, 53 to 45. That's a 14-point swing toward Dems.
White women went for Trump 53 to 46 in 2024. This year they went for Dems, 52 to 46. That's a 13-point swing - and it flipped which side white women picked.
For comparison, Black women barely moved, from 92 percent for Dems to 93, cause they were already there.
On its face that looks like a lot of people swung towards Democratic candidates - but it only looks that way because Republicans stayed home.
The exit poll tells you who it thinks voted. The vote count just counts. So I lined up two numbers for each party - what its candidate for president got in 2024, and what its House candidates got this year.
I picked 2024 on purpose. Trump was on the ballot and he won. So the question is whether the side against him has gotten any bigger since.
Harris got 75.0 million. House Democrats have 65.0 million so far. That's 10.0 million fewer.
Trump got 77.3 million. House Republicans have 49.4 million. That's 27.9 million fewer.
The count isn't done. The AP figures about 4.9 million ballots are still out, and almost half of them are in California.
So give every one of them to the Democrats. They still finish 5.1 million short of Harris. Give every one to the Republicans instead, and they're still 23.0 million short of Trump.
A vote for president and a vote for Congress aren't the same thing, so I ran it House against House too.
In 2024 House Democrats got 70.6 million votes and House Republicans got 74.4 million. Democrats are down 5.6 million from that so far. Republicans are down 25.0 million. Some districts had no Republican on the ballot, mostly in California, and that trims the Republican total a little.
Put this year's Democratic vote next to Trump's 2024 vote and it loses by more than 7 million, even with every uncounted ballot thrown in.
Every midterm draws fewer voters than a presidential year, so both of those numbers were always gonna drop. More Democrats showed up than in a normal midterm, too. I'll get to that.
But look at what this one was about. National Guard troops were still deployed in Washington and still deployed in Memphis when we voted. ProPublica has counted more than 500 citizens held by immigration agents since January 2025. This year, detention camps spread across the South - I call them concentration camps, and the Japanese American National Museum does too.
Thirteen states ban abortion, and ProPublica has named seven women who died between 2021 and 2024 after care came too late or never came. Flock cameras log every car that drives past them, and Memphis police say there are 200 here. "Some pardons came after well-timed donations or a business deal," CBS's 60 Minutes reported. And the president's own ethics filing shows more than a billion dollars from crypto in his first year back, plus more than 21,000 trades in his accounts.
This was the midterm to push back. Democrats still got 5 to 10 million fewer votes than Harris did. Against the 2024 House vote, it's 1 to 6 million fewer.
So I asked it a different way. Out of every 100 votes a party got for president, how many did it still have at the next midterm?
In 2018, Trump's first midterm, Democrats kept 92 out of 100. Republicans kept 80. Everybody called that one a blue wave too.
In 2022 it went the other way. Democrats kept 63 and Republicans kept 73, and Republicans took the House.
This year Democrats have kept 87 so far. If every last uncounted ballot is theirs, it's 93. That's about what they kept in 2018.
Republicans have kept 64. With every uncounted ballot it's 70. That's the lowest of the three.
So the Democratic number barely moved between the last blue wave and this one. The Republican number is down somewhere between 10 and 16.
How much of its vote for president each party kept in the next midterm
2018 midterm, against the 2016 vote for president
2022 midterm, against the 2020 vote for president
2026 midterm, against the 2024 vote for president
Then I went seat by seat. A flipped seat is one that went from Republican to Democrat. Democrats have flipped 43 seats so far. California's four are still being counted, so I set those aside. That leaves 39.
In at least 36 of 39, the Democrat who won got fewer votes than Kamala Harris got in that same district in 2024. One more, in North Carolina, is within about 700 votes of her with ballots still being counted.
And Harris lost most of those districts. Trump carried 35 of them. The two winners who beat her number are both in South Texas.
Nine states drew new maps for this election, and new lines can change who lives in a district, so I checked the states that didn't. Of the 28 flips called by Monday in those states, every winner got fewer votes than Harris. That's 28 of 28.
It mostly holds candidate against candidate, too. In the states that kept their maps, 27 of those seats had a Democrat on the ballot in 2024.
In 22 of those 27, the Democrat who won this year got fewer votes than the Democrat who lost that seat two years ago.
Janelle Stelson ran against Scott Perry both times. In 2024 she got 200,434 votes and lost by about 5,000. This year she has 173,906 and she's ahead by 32,000, cause Perry went from 205,567 to 141,802.
Christina Bohannan won her rematch in Iowa the same way. She got 206,156 votes in 2024 and 184,126 this year.
The 39 seats Democrats flipped outside California: each winner and loser against 2024
Each dot is one seat. Left of the dashed line means fewer votes than the party's candidate for president got there. 37 of the 39 winners are left of it on Cook's count, which runs a little behind the AP. One of them, in North Carolina, is within 700 votes of the line.
A few states can tell you how many registered Republicans and registered Democrats actually voted. That's a count too.
Florida is one. In 2024, 4,697,079 registered Republicans voted there. This year it was 3,561,765.
That's 24 percent fewer. Democrats were down 19 percent.
The biggest drop there was people registered with neither party, down 31 percent.
Republicans were 42.7 percent of Florida's voters in 2024 and 42.8 percent this year. And Byron Donalds still won the governor's race, by 106,240 votes.
Colorado mails every voter a ballot and reports who sent theirs back. Two days after the 2024 election and two days after this one, Democrats were down 12 percent.
Unaffiliated voters were down 14. Republicans were down 21.
And on the rolls, the side against Trump didn't grow. Pennsylvania had 3,998,602 registered Democrats on Election Day 2024 and 3,887,465 this year.
Republicans there lost fewer, about 73,000. Democrats did win some of that back during 2026 - NBC caught the turn in October - but they're still below where they started.
Then I went through Tennessee, county by county.
Trump got 1,966,865 votes here in 2024. Marsha Blackburn just won the governor's race with 957,136. She got less than half of what he got, and it didn't matter. Bill Hagerty, the Republican for Senate on the same ballot, got 1,138,436. So some of those voters showed up and skipped Blackburn.
Harris got 1,056,265. Jerri Green got 902,953. Blackburn won by 54,183.
Green got fewer votes than Harris in all 95 counties. Blackburn got fewer than Trump in all 95 too. An independent named David Hatley took 195,564. That's more than three times Blackburn's margin.
Then there's Shelby County. Harris got 201,759 votes here. Green got 164,812. Of the six biggest counties in Tennessee, Shelby held on to the smallest share of its 2024 Democratic vote.
The county cast 329,855 ballots in 2024 and 257,109 this year. That's 72,746 fewer, before provisional ballots. In my own county, that's more than one voter in five.
In May the legislature split Shelby County across three congressional districts. Republicans won all three. Their three margins add up to 24,745 votes.
Every Tennessee county: the 2026 vote for governor as a share of the 2024 vote for president
Jerri Green, against Harris
Marsha Blackburn, against Trump
One dot is one county. Every dot is left of the line: Green got fewer votes than Harris in 95 of 95 counties, and Blackburn fewer than Trump in 95. The ringed dot is Shelby.
Brookings, a Washington think tank, said it three days after the election. Elaine Kamarck and William Galston wrote that the sweep "was driven less by Republican voters switching sides than by discontented Republicans staying home while motivated Democrats turned out."
They ran the same kind of check on three governor's races. In Texas they have Greg Abbott 30 percent below Trump's vote. In Florida, Byron Donalds is 32 percent below. In Ohio, Vivek Ramaswamy is 36 percent below - and Amy Acton beat him with 5 percent fewer votes than Harris got there. One of their examples cuts the other way. In Texas, the Democrat for governor, Gina Hinojosa, got about 8 percent more votes than Harris did.
CNN's Ronald Brownstein saw it coming in June. His headline was "Who stays home may threaten Republicans this year as much as who votes." The Philadelphia Inquirer said it the day after. Even Axios, under the headline "Black voters deliver a GOP reckoning," wrote that "Election returns don't prove a nationwide surge in Black turnout".
And the networks' poll backs it up. By its numbers, 91 percent of the Trump voters who showed up this year voted Republican again. Seven percent crossed over.
Democrats like to say they outnumber conservatives. In 2024, with him on the ballot, not enough of them showed up to beat him. This year, with his policies in full effect, even fewer did.
Measured against 2024, Democrats won this big mostly cause Republicans stayed home. I think calling it a blue tsunami is misleading. People are gonna think it shows everybody finally came to their senses and joined the right side - and that they've got a good chance of winning again in 2028.
I think that's a false sense of comfort.
Most of those missing Republican voters didn't switch sides. They just didn't vote, and whether they come back in 2028 isn't something Democrats control.
This wasn't really a blue tsunami as much as a parting of the red sea.
But there is one takeaway that might be comforting, and that's that most of Trump's voters might not follow any local races - and couldn't name their current House or Senate leaders, or even know how Congress works. Trump wasn't on the ballot - so they didn't show up. And maybe that means they won't show up in 2028 either.
The question remains - are Dems willing to take that bet? Or is now the time to see where mistakes were made and start rethinking how to approach 2028?
Stats can be read a lot of different ways, and they usually get read based on what you're trying to prove. So here are the other ways to look at these numbers, with the numbers behind them.
House Democrats got 51.3 million votes in 2022. They have 65.0 million so far this year. That's 13.7 million more, and it's up 27 percent. Republicans are down about 4.8 million from 2022.
House Republicans got 50.5 million votes in 2018 and 54.2 million in 2022. They have 49.4 million so far this year, with more to count. With Trump on the ballot it was 63.0, then 74.2, then 77.3 million.
So you can say the missing voters only show up for Trump, and he can't run again. That's a fair point. My answer is that most of them didn't change sides, they stayed home, and nobody knows who they show up for next.
About 123.5 million ballots were cast, which is more than in 2018. It's the second-highest turnout rate for a midterm since 1914. "Historic turnout" is true.
And House Democrats have about 4.7 million more votes than they got in 2018, with more to count.
If Republicans had kept 80 of every 100 Trump votes, the way they did in 2018, Democrats would still win the House vote by somewhere around 2 to 6 points. So Republicans staying home explains how big the win was. It doesn't explain the win.
The president's side always keeps fewer of its voters at the midterm. In 2018 it was Republicans, and in 2022 it was Democrats. This year's gap is the biggest of the three, but the pattern is old.
The polls have independents at 49 to 46 for Harris in 2024 and 58 to 38 for Democrats this year. It's the same warning as before - two different tools, and a race for president next to a race for the House. The Times has Black, white and Hispanic voters all moving toward Democrats from 2024.
The Times has Hispanic voters going from Democrats by 11 in 2024 to Democrats by 41. That's either people changing their minds or Trump's Latino voters staying home, and the numbers can't tell me which yet.
The two flipped seats where the winner beat Harris's number are both in South Texas.
Republicans were 42.7 percent of Florida's voters in 2024 and 42.8 percent this year. The group that fell the most was people in neither party.
Both parties always fall off from a presidential year. The party that wins a midterm always gets fewer votes than its candidate for president got. When I say even fewer Democrats showed up, that's true of every midterm.
My answer is the stakes. That's my judgment as the writer. It isn't a finding. And by the kept numbers, Democrats did about as well this year as they did in 2018. But look what happened in 2020.
NBC found Democrats gaining registrations in five of seven close states by the middle of 2026. In Pennsylvania, their lead over Republicans grew by about 78,000 between the 2025 election and this one.
When I count ballots in a majority-Black ward or district, I'm counting everybody who lives there. Only a voter file can say who actually voted, and most states haven't released theirs.
Every place I checked for Black turnout has a big Black population. Milwaukee, Cleveland, Columbus, Indianapolis, Pittsburgh, Kansas City and St. Louis are still unchecked.
Every 2026 number on this page is unofficial. The biggest share of the ballots still out is in California. One uncalled seat, Colorado 4, could add a flip where the Democrat beats Harris's number. I've given the high end and the low end wherever it matters.
Pick a different baseline and you get a different story. I picked 2024, and I told you why. The numbers are all here, so check me.