2026 Midterm Election Forecast: Senate, House and Governors

How the forecast works

The model estimates each race's margin (the Democratic or independent candidate minus the Republican) from three inputs, then simulates the election 1,000,000 times to turn margins into probabilities.

1. Polling average. General-election polls compiled by 270toWin, for every Senate and governor race and every House district within 15 points. Each poll is weighted by recency (its weight halves every 14 days), sample size, population (likely-voter polls count more than registered-voter or adult samples) and pollster quality: 538's pollster grades, from 0.5 to 3.0, set a multiplier from about 0.6 to 1.2, and unrated pollsters get 0.75. Each pollster's house effect, meaning how far its results run from other pollsters' in the same races, is estimated across all races, shrunk toward zero for pollsters with few polls, capped at 5 points and removed before averaging. Undecided voters are not allocated; only the gap between the two main candidates is used.
2. Expert ratings. The published ratings of up to 12 forecasters, including the Cook Political Report, Inside Elections and Sabato's Crystal Ball, converted to a margin (Safe 22, Likely 12, Lean 6, Tilt 3, Toss-up 0 points) and averaged. Ratings carry information polls do not, such as candidate quality, fundraising and private polling.
3. Fundamentals. The state's or district's partisan lean (2026 Cook PVI, which accounts for redrawn House maps), plus the national environment from the generic congressional ballot (D+7.8 in late-October polling; the model centers on D+7.5), plus about three points for an incumbent seeking re-election. Governor's races get less partisan weight because voters split tickets for governor more often.

Blending. A race with plenty of recent polling is about 62% polls, 23% ratings and 15% fundamentals. Races with little polling lean on ratings and fundamentals. A House district with polling moves up to halfway toward its adjusted polling average, depending on how much polling it has; other districts are a blend of ratings and fundamentals.

Pollster adjustments

The most active pollsters this cycle, their 538 grade, and the house effect the model removes from their polls (a pollster marked D+2 has run about two points more Democratic than other pollsters in the same races).

Pollster grades: 538 pollster ratings (ABC News, last updated February 2025), used under CC BY 4.0. House effects are PoliGov's own estimates from this cycle's polls.

Simulation. Each simulated election draws one national error (standard deviation 3.3 points) that moves every race in the same direction, as real polling misses do, plus a separate race-level error from a fat-tailed distribution (about 4 points for well-polled races, up to 7 for unpolled ones). A party's win probability is the share of simulations it wins, and chamber control is counted the same way.

Updates. Until Election Day the forecast reran automatically every two hours with newly posted polls. The final forecast was published Nov 3, 2026, 6:12 AM ET and is now locked, so it can be scored against the results on the scorecard.

Ratings on this page come from the projected margin: Toss-up when the projected margin is under 1.5 points, Tilt under 4.5, Lean under 9, Likely under 17, and Safe at 17 or more.

Limitations. Alaska's ranked-choice voting, Louisiana's new closed primaries, candidate scandals and events after the model run are reflected only through the polls and ratings.