Democrats' generic ballot lead is large by midterm standards
Our average puts the Democrats 8.8 points ahead three weeks out; past final polls overstated them, so our forecast works from D+6.7.
Written in house from our poll data and our forecast.
The Democrats lead the generic congressional ballot by 8.8 points on our average of national polls, three weeks before the midterms. That is close to where the Democrats stood at this point in 2018 and well short of 2006, and the record of the final polls suggests the lead on election day will be smaller than the one the polls show now.
Where the lead stands
Our average stood at D+3.9 on 1 January, when we rebuild it from the polls published by then. It reached D+7.0 by early July, peaked at D+10.0 on 23 September and has eased to D+8.8, down 0.4 points in the past 30 days. That last move is within the normal scatter of the polls.
The newest polls span a wide range. Quinnipiac University had the Democrats ahead by 12 points among registered voters at the end of September. Cygnal and YouGov each had them up nine on 2 October and Pew Research Center up eight among registered voters on 4 October. At the other end, HarrisX had a two-point Democratic lead on 28 September, and Rasmussen Reports has had three to five points in its recent releases. Reports of other averages are close to ours: political.org reported that Nate Silver's average stood at D+8.9 on 7 October.
Donald Trump's job approval is 35% on our average of 14 pollsters, down from 42% at the start of the year.
Against past midterms
Our "this day" page compares each cycle using a single pollster asked the same way every time: CNN's generic ballot among registered voters, at the reading nearest to 20 days before the election. At this point in 2006, with George W. Bush in the White House, CNN had the Democrats ahead by 18. In 2018, under Trump, it had them ahead by nine. In 2014 it had the Democrats, then the president's party, up six; in 2010 it was even; in 2022 the Republicans led by one.
CNN's latest reading this cycle dates from January, at D+5.0, so today's figure has to come from our average. That average is on the decided vote and runs a little larger than CNN's measure, which counts people who back neither party. With that caveat, the 2026 lead sits close to 2018's at the same stage and well short of 2006's.
The approval comparison is less ambiguous. Presidents' approval at this point in the five midterms since 2006 ranged from 37%, Bush in 2006, to 46%, Barack Obama in 2010. Trump's 35% is below all of them, though Gallup has not published a reading since December 2025, so ours is an average of every pollster rather than the same measure.
Two patterns pulling apart
History offers two lessons, and they point in opposite directions.
The first is about the last three weeks. In the five midterms since 2006, the national House vote ended up worse for the president's party than CNN's reading at this point, by 2.0 points on average. The pattern is far from uniform. The president's party did 11.7 points worse than CNN's October reading in 2014, 6.8 points worse in 2010 and 1.6 points worse in 2022, but did better than it in 2006, when Republicans finished 9.9 points closer than CNN's 18-point gap, and slightly better in 2018. The typical gap between CNN's reading at this point and the result, in either direction, is 4.8 points. If the average pattern held, it would help the Democrats.
The second is about the final polls. We hold the last generic-ballot polls before every House election since 2000, between one and six each cycle, and compare them with the national House vote. In 11 of those 13 elections the final polls overstated the Democrats. The exceptions were 2008 and 2018. In the midterms the misses were large in 2002 and 2014, when the final polls missed each party's share by an average of 5.3 and 4.9 points, and under one point in 2018 and 2022.
Our method test takes this second pattern seriously. Predicting each election with a correction worked out from the other elections only, the corrected figure came closer than the published polls in 11 of 13 elections, with an average miss per party of 1.92 points against 2.44. So our forecast applies it: the polls' D+8.8 becomes D+6.7 in our national House vote, or 53.1% to 46.4%.
We do not apply the first pattern. Five midterms are too few to test it, so the forecast instead widens its range for the time left before polling day.
What it means in seats
The difference between those readings is large in seats. If the generic-ballot polls are exactly right at D+8.8, our model gives the Democrats 242 House seats and 54 senators. At our corrected D+6.7, it gives them 231 seats and 52 senators. Our forecast currently gives them a 90% chance of a House majority and 83% for the Senate.
Below that, the House holds up better than the Senate. At D+5.7 the Democrats still win 221 seats and a 51-49 Senate. At D+4.7 they keep the House on 221 but lose the Senate 51-49. The House majority goes at about D+4: our what-if page gives the Democrats 218 seats at D+4.0 and the Republicans 218 at D+3.9.
Put against history, a Democratic House needs the final national margin to stay above roughly four points, about five points below today's polls. Of the past misses we replay, only the 2002, 2014, 2016 and 2020 cycles were big enough to do that. The Senate needs the margin to stay above roughly five points, which the 2024 replay already fails.
The latest
The polls arriving this month have not changed the broad picture. NPR reported on 8 October that both parties' spending decisions suggest the question is not whether Democrats take the House but how large the majority may be. KCUR, carrying a Kansas Reflector report, quoted a Senate Majority PAC spokesperson attributing the wider Senate map to Trump's unpopularity, tied to the Iran war and gas prices; that is the group's claim, and our data cannot test it.
What our data can say is that a D+8.8 lead three weeks out is large by the standard of recent midterms, that leads of this kind have usually shrunk in the final polls rather than grown, and that the Democrats' House majority survives every plausible miss except a large one, while their Senate majority survives only a small one.
Polling Forecast reads every poll in ten countries, corrects for how pollsters have missed before, and turns them into seats. Our forecasts, the polls and our methods are free to explore.
Elsewhere on Polling Forecast
Rome prosecutors open a file on Vannacci's Futuro Nazionale
Futuro Nazionale is on 6.5% in our average and would win 23 seats outside both coalitions; the centre-left wins the Camera in 63% of our simulations.
AfD's Siegmund will stand for premier in Saxony-Anhalt on 8 December
The AfD holds 39 of the 83 seats in Magdeburg; nationally it leads our average by 5.7 points and would win 193 Bundestag seats.
Nawrocki promotes Domański reluctantly as the tribunal fight deepens
The government keeps a majority in only 10% of our simulations, while PiS, Konfederacja and KKP together would have 245 seats.