House Effects
Monday, September 21, 202643 days until Election Day, November 3, 2026

The 2026 midterms, as the polls model them

Weighted averages of 2,366 polls, corrected for each pollster's measured lean, blended with partisanship, incumbency, and fundraising, and given the uncertainty that 2,313 past races say they deserve. How it works.

56% Democrats win control
Democrats 56%, Republicans 44%. The party with more seats runs the Senate, so 50 Democrats beat 49 Republicans if an unaligned independent holds the last seat (8% of simulations); a tie goes to Republicans through the vice president. Democrats average 50.3 seats (47 now); nine in ten simulations land between 46 and 54.

If the polls overstate Democrats the way they did in 2016–24 (about 2.4 points here): Democrats 28%, Republicans 72%. If they miss the other way by as much: Democrats 81%, Republicans 19%.

4347515551: outright majority →
Democrats control Republicans controlDemocratic seats after the election, 40,000 simulations
80% Democrats win control
Democrats average 227.0 seats (215 now); nine in ten simulations land between 209 and 248.

If the polls overstate Democrats the way they did in 2016–24 (about 1.8 points here): Democrats 59%, Republicans 41%. If they miss the other way by as much: Democrats 92%, Republicans 8%.

198216228258218 for Democratic control →
Democrats control Republicans controlDemocratic seats after the election, 40,000 simulations

Governors

All races →
26.3 of 50 governorships
Democratic governors expected after the election, against 24 now. Nine in ten simulations land between 21 and 32.

If the polls overstate Democrats the way they did in 2016–24 (about 4.1 points here): 22.8 Democratic governors. If they miss the other way by as much: 29.7 Democratic governors.

1624263624 now
share of simulationsDemocratic governors after the election, 40,000 simulations
Where things stand
Written from the model's output as of Sep 17
The numbers above are from Sep 21; the note is rewritten by hand and may lag them.

Democrats are favored to take the House and are slight favorites in a Senate fight that is close to a coin flip. The model gives them an 80% chance of a House majority against 20% for Republicans, with the simulations averaging 227 seats, 218 needed, and a 90% range of 209 to 247 that still contains defeat. The Senate is close to a toss-up: 54% Democratic control, 46% Republican, with the simulations averaging 50.2 seats from 47 now and a 90% range of 46 to 54. The party with more seats organizes the chamber, and the vice president only breaks ties, so Democrats need 51 seats, or 50 if an unaligned independent holds a seat Republicans would otherwise have. In 8% of simulations Democrats get control that way, most often through Dan Osborn in Nebraska; in 9% Republicans keep it only on the vice president's tie-breaking vote. Democratic governorships average 26.3, up from 24, in a range of 21 to 32.

Six or seven races decide the Senate. North Carolina (Roy Cooper at 90%) and Maine (Troy Jackson at 71%) are the likelier Democratic pickups, and Michigan (Abdul El-Sayed at 69%) is the least secure seat Democrats already hold. The fight sits in Texas, where James Talarico projects at D+1.1 and 58%, and Ohio's special, where Sherrod Brown projects at D+1.0 and 57%. In both, polls are better for Democrats than fundamentals: Texas polls average D+2.6 against fundamentals of R+3.7, Ohio D+2.9 against R+3.2. The model weights polls at 76% and 68% there, so the projection lands closer to the polls without fully believing them. Alaska (Mary Peltola, 42%) and Iowa (39%) lean the other way. Nebraska's non-Republican is independent Dan Osborn, at 44% on six polls.

The scenarios matter more than the headline. If polls overstate Democrats as they did on average from 2016 through 2024, Senate control odds fall to 27% and House odds to 59%. A miss of the same size in the other direction, as in the 2025 governor's races, puts them at 80% and 92%. The headline numbers assume no direction to the error. That assumption is doing most of the work.

National environment

The generic ballot

Asked which party they want in Congress, voters favor Democrats by 7.0 points in our average of 462 polls. Since 1998 that question has overstated Democrats by about 1.8 points at this stage, so the model reads the national environment as D+5.3, give or take 3.0.

R+10R+5EvenD+5D+10D+15D+20Jun ’25SepDecMar ’26JunSepD+7.0Election Day
adjusted weighted average shading: which side the average favors each poll, as published party-aligned pollster or sponsor
Generic ballot average
D+7.0
Latest poll ended Sep 20
Environment the model uses
D+5.3
After the historical 1.8-point correction; ±3.0 typical miss
Trump net approval
−18.1
40% approve, 58% disapprove. Shown for context; the model does not use it. 648 polls; latest ended Sep 20.
Senate

16 Senate races in play

Each row is one race: the dot is the projected margin, the bar holds half of simulated outcomes, and the line holds nine in ten. The hollow diamond is where the polls alone sit, and the heavy rule is the line for control: Democrats need every race above it. Shown: races where each party has at least a 1-in-20 chance; the rest are on the full Senate page.

Race
D+20EvenR+20
Proj.
projected margin (polls + fundamentals) polls alone half of simulated outcomes nine in ten outcomes
Governors

23 governor's races in play

Governor's races lean far less on partisanship than Senate races do, so the fundamentals get little weight here and unpolled races carry wide intervals. All 36 races.

Race
D+20EvenR+20
Proj.
projected margin (polls + fundamentals) polls alone half of simulated outcomes nine in ten outcomes
House

The 39 closest districts

Most districts have no public polls, so House projections lean on district partisanship, incumbency, fundraising, and the national environment, with district polls blended in where they exist. All 435 districts.

Race
D+20EvenR+20
Proj.
projected margin (polls + fundamentals) polls alone half of simulated outcomes nine in ten outcomes
Fresh data

Latest polls

The newest race polls in the database. See all of them, including the generic ballot.

DatesRacePollsterSampleDem.Rep.As publishedAdjustedShare of average
Sep 16–18Iowa Senate
Josh Turek v. Ashley Hinson
Trafalgar GroupREP.-ALIGNED1,089 LV42.144R+2D+1.2
moved 3.1 toward D
9.3%
source
Sep 16–17New Hampshire Senate
Chris Pappas v. John Sununu
InsiderAdvantage1,200 LV47.839.9D+8D+7.8
moved 0.1 toward R
17%
source
Sep 15–17Virginia’s 1st District
Shannon Taylor v. Robert Wittman
DCCC Targeting and Analytics DepartmentDEM.-ALIGNED519 LV4647R+1R+4.7
moved 3.7 toward R
91%
source
Sep 15–17Texas Senate
James Talarico v. Ken Paxton
Trafalgar GroupREP.-ALIGNED1,079 LV46.445.2D+1D+4.3
moved 3.1 toward D
7.6%
source
Sep 16–17Michigan Senate
Abdul El-Sayed v. Mike Rogers
InsiderAdvantage1,200 LV46.644.7D+2D+1.8
moved 0.1 toward R
7.1%
source
Sep 15–17Florida Senate (special)
Angie Nixon v. Ashley Moody
St. Pete Polls
for Florida Politics
913 LV44.845.2EvenR+0.6
moved 0.2 toward R
28%
source
Sep 16–17North Carolina Senate
Roy Cooper v. Michael Whatley
InsiderAdvantage1,200 LV48.442.5D+6D+5.8
moved 0.1 toward R
13%
source
Sep 15–17Florida Governor
David Jolly v. Byron Donalds
St. Pete Polls
for Florida Politics
913 LV43.443.5EvenR+0.3
moved 0.2 toward R
19%
source
Sep 14–17Alaska Senate
Mary Peltola v. Dan Sullivan
co/efficient799 LV4648R+2R+1.1
moved 0.9 toward D
24%
source
Sep 13–16Idaho Senate
Todd Achilles v. Jim Risch
Advanced Targeting Research700 RV4839D+9D+7.9
moved 1.1 toward R
25%
source
Sep 14–16Nebraska Governor
Lynne Walz v. Jim Pillen
Wedgewood Polls500 LV3653R+17R+16.6
moved 0.4 toward D
34%
source
Sep 15–16Kansas Senate
Adam Hamilton v. Roger Marshall
Emerson College
for Nexstar
750 LV44.843.3D+1D+1.4
32%
source

LV: likely voters. RV: registered voters. A: all adults. “Adjusted” is the published margin after removing the sponsor shift and the pollster’s measured house effect; the small print says how far that moved the poll. Dem.-aligned and Rep.-aligned mark polls paid for by a party, a campaign, or an allied group, or taken by a firm whose published work is mostly for one side. They are shifted by the average lean of such polls, then counted at full weight.