2026 Midterm Forecast

National mood (D−R generic ballot, points)

National environment

Generic congressional ballot (Democratic minus Republican)

Trump net job approval (approve minus disapprove)

Line: our trend fitted to every individual poll, with each pollster's house effect and the likely-voter / registered-voter / adult offsets removed. Shaded band: 95% interval for where the polling average stands (sampling error plus uncertainty in the house effects). This is not an Election Day range: how far polls can miss the result is much wider, and the model carries that separately (4.1 points, one standard deviation). Dots: individual polls adjusted to the same scale; hover for the pollster. Diamonds at the right edge of the generic-ballot chart: the published averages. Polls from VoteHub, RealClearPolling, Polling USA and pollster releases; see Sources below.

House — every district, coloured by P(Democratic win)

Each district on the 2026 maps, shaded by the chance a Democrat wins; black outline = open seat. Hover for the member, PVI, model margin and probability; scroll or pinch to zoom, drag to pan. The national mood shifts every simulated margin one-for-one.

District boundaries: U.S. Census Bureau 2020 block groups assigned to the 2026 (120th Congress) districts with the Census block equivalency file; drawn with D3 and TopoJSON.

≥90% D 70 50 30 ≥90% R

Governors — the 36 races

–mean D governors

–10–90%

–P(D ≥ 26)

Senate — the 35 races

P(challenger) from the poll blend and fundamentals; the national mood enters the Senate at slope 0.8 (fitted on 326 races 2006-2024). Seats not up: . † independent challenger: hover for how the prior is set.

House seats

–mean D seats

–10–90%

–P(D majority)

Distribution of Democratic seats from 20,000 correlated simulations at this mood.

Senate seats

–mean D seats

–10–90%

–P(D ≥ 51)

–P(R ≤ 49)

Closest House races at this mood

The polls themselves

House effects, pollsters active in the last 120 days (points the pollster runs above/below the field, Democratic-positive):

Method

Sources and credits

Polls. Wikipedia race and polling tables (CC BY-SA 4.0); the VoteHub open polling API; RealClearPolling poll listings; Polling USA on Bluesky; individual pollster releases entered directly (for example the Marist Poll and Strength in Numbers / Verasight); a leaked Texas internal poll as reported by the Quorum Report. Every poll belongs to its pollster.

Poll history and calibration. 538 / ABC News archived polling averages, rated-poll history, pollster ratings and partisan-lean data (CC BY 4.0), used to measure poll error, house effects, sponsor effects and state polling misses.

Election results and fundamentals. MIT Election Data and Science Lab (Senate and presidential returns); the Federal Election Commission (official House results and candidate fundraising through the OpenFEC API); Wikipedia yearly results for governors and the national House vote; Cook Political Report PVI; race ratings from Cook, Sabato's Crystal Ball, Inside Elections and others, shown for comparison only; county presidential results compiled by Tony McGovern.

Approval and special elections. Gallup presidential approval via The American Presidency Project (UC Santa Barbara); The Downballot special-elections Big Board.

Demographics and maps. U.S. Census Bureau: American Community Survey 2020-2024 5-year estimates (citizen voting-age population, education, age, home heating fuel), 2020 cartographic block groups, and block equivalency files for the 119th and 120th Congresses. This site is not endorsed or certified by the Census Bureau.

Prices. U.S. Energy Information Administration retail gasoline and diesel; Maine Department of Energy Resources heating fuel survey; HomeHeat supplier prices; heating oil futures (ICE / NYMEX via Yahoo Finance).

Software. D3 (ISC), TopoJSON client (ISC), mapshaper (MPL 2.0), pandas, NumPy, SciPy and GeoPandas.

Model, code and any errors: Shawn Corvec. Not affiliated with or endorsed by any of the sources above.