PW 2026 Forecast — House & Senate
A fundamentals-first seat model: presidential approval and real special-election results carry 70% of the weight, polling 30% with a disclosed bias correction. Every input, weight, and the model's record against the last two midterms is on this page. Updated 2026-08-16.
The most likely outcome — it happens in 82% of our simulations — is that Democrats take the House while Republicans keep the Senate. That is roughly 3-in-4 or better odds.
House seat distribution (20000 simulations)
House battlegrounds
| VA-01 | Rob Wittman (R) | R 51% | R incumbent |
| TX-28 | Henry Cuellar (D) | D 51% | D incumbent |
| WI-01 | Bryan Steil (R) | D 52% | R incumbent |
| IA-03 | Zach Nunn (R) | D 52% | R incumbent |
| PA-10 | Scott Perry (R) | R 52% | R incumbent |
| OH-10 | Mike Turner (R) | R 53% | R incumbent ⚠ old-lines lean |
| CA-22 | David Valadao (R) | R 53% | R incumbent ⚠ old-lines lean |
| MI-10 | John James (R) | D 54% | open seat |
| TX-34 | Vicente Gonzalez (D) | D 54% | D incumbent |
| MI-04 | Bill Huizenga (R) | R 56% | R incumbent |
| NC-11 | Chuck Edwards (R) | R 58% | open seat ⚠ old-lines lean |
| TX-35 | open seat | R 58% | open seat |
| PA-07 | Ryan Mackenzie (R) | D 59% | R incumbent |
| OH-09 | Marcy Kaptur (D) | D 61% | D incumbent ⚠ old-lines lean |
| MO-02 | Ann Wagner (R) | R 62% | R incumbent ⚠ old-lines lean |
| ME-02 | Jared Golden (D) | R 63% | open seat |
| CA-41 | Linda Sánchez (D) | D 64% | D incumbent ⚠ old-lines lean |
| WI-03 | Derrick Van Orden (R) | R 65% | R incumbent |
| IA-02 | Ashley Hinson (R) | R 65% | open seat |
| CA-13 | Adam Gray (D) | D 66% | D incumbent ⚠ old-lines lean |
| OH-15 | Mike Carey (R) | R 66% | R incumbent ⚠ old-lines lean |
| CO-08 | Gabe Evans (R) | D 67% | R incumbent |
| CA-40 | open seat | D 68% | no incumbency data ⚠ old-lines lean |
| MI-07 | Tom Barrett (R) | D 70% | R incumbent |
| PA-08 | Rob Bresnahan (R) | R 70% | R incumbent |
| NJ-07 | Thomas Kean Jr. (R) | D 70% | R incumbent |
| OH-07 | Max Miller (R) | R 71% | R incumbent ⚠ old-lines lean |
| IA-01 | Mariannette Miller-Meeks (R) | R 71% | R incumbent |
| CA-03 | Ami Bera (D) | D 71% | D incumbent ⚠ old-lines lean |
| AZ-06 | Juan Ciscomani (R) | D 72% | R incumbent |
Senate battlegrounds
| NC | Thom Tillis (R) | D 76% | |
| FL-S special | Ashley Moody | R 85% | |
| MI | Gary Peters (D) | D 85% | |
| IA | Joni Ernst (R) | R 85% | |
| OH-S special | Jon Husted (R) | R 92% | |
| GA | Jon Ossoff (D) | D 93% | |
| NH | Jeanne Shaheen (D) | D 96% | |
| AK | Dan Sullivan (R) | R 96% | |
| ME | Susan Collins (R) | D 97% |
The national environment, shown openly
- Combined environment: D+5.7 (national House two-party margin)
- approval -22 -> R-8.3 (slope 0.115, int -5.8; fit 1978-2014)
- specials overperf -15.8 x k=0.5 vs pres baseline -> R-6.2
- generic ballot R-5.5, bias-corrected +3.0 -> R-2.5
- Weights: approval 0.3, specials 0.4, generic ballot 0.3 — 70% of the model is poll-free.
Track record before you trust it
Before publishing, this exact machinery was required to retrodict the last two midterms using only information available at the time — both years held out of every fit. National margin: 2018 predicted D+6.4 (actual D+7.4); 2022 predicted R+0.8 (actual R+3.2). Seats, on the 2022 maps: predicted 224 Republican seats; the real number was 222, with 95.9% of the 435 district calls correct. If a future update fails that test, we will say so on this page.
Why not just use the polls? Because in three of the last four cycles the final generic-ballot average understated Republicans — so this model leans on what cannot be polled wrong: how districts actually voted, whether the incumbent is on the ballot, the president's approval, and real special-election results, including the ten states that redrew their maps for 2026. Where our data is thinner than we'd like — a redistricted seat still awaiting new-lines numbers, a district with no public polling — the table says so instead of pretending.
A forecast is a probability, not a promise: a 25% chance means that outcome happens one time in four. Model and computation are Patriot Watch's own. Generic-ballot input via PollingSource average (pollingsource.com); approval via our primary-source pollster reads; district data from public election records.