Forecasts of the 2026 midterm elections from polls, prediction markets, professional forecasters, academic models and the PLSC 2219 class model, updated daily.
Forecasts are collected from each source once or twice a day. The raw pages and files are saved as received, and the numbers are then read from those saved copies.
A full description of how each source was collected and processed will be published with the data archive after the election.
Each line on the chart is the average of the forecasts from that type of source on that date. If a source did not update on a given day, its most recent forecast is used.
Some sources enter the archive partway through the series. On the dates before a source's first forecast, its first national margin is used, so the line does not jump on the day the source is added. Seat counts and chances of control for those dates are calculated from that margin using the district maps in force on each date.
Some sources publish only a national House vote margin. For those, this site calculates seat counts and chances of control by applying the national margin to each district's or state's partisan lean. These numbers can differ from a forecaster's own seat count. Market seat counts and probabilities come from market prices, and the class model produces its own.
For a single race, the polling line uses only polling averages of that race, from the table of polling averages in the race's Wikipedia article. It is shown when at least three are available.
Probabilities are the chance that Democrats win control. In the Senate that means 51 seats, since the vice president is a Republican.
| Line | Sources |
|---|---|
| Polling | Polling averages from RealClearPolitics, Silver Bulletin, Decision Desk HQ, FiftyPlusOne, 270toWin, VoteHub and Race to the WH, mostly as listed on Wikipedia, plus a simple average of individual generic ballot polls. |
| Markets | Kalshi, Polymarket, PredictIt. |
| Professional | Race to the WH. More forecasters are being added. This line is hidden until it includes at least three. |
| Academic | Generic ballot (Bafumi, Erikson & Wlezien); Economic pessimism (Lockerbie); Political history (Lewis-Beck & Quinlan); Referendum (approval + income, no exposure); Ray Fair's House vote equation (fairmodel.econ.yale.edu). |
| Class model | PLSC 2219 (see below). |
Other inputs: presidential approval polls (Wikipedia), economic data (FRED), past election returns (MIT Election Data and Science Lab), and district and state partisan lean.
These are published models from political science research, run by this site on current data.
The class model was built in PLSC 2219. It predicts the Democratic share of the two-party vote in each race from the district's or state's partisan lean, the national environment and incumbency. The national environment is predicted from past elections using whether it is a midterm year, the change in real income, the change in gas prices, and presidential approval. Seats and chances of control come from 10,000 simulations.
It is re-run each day with the latest approval average, gas prices and income, and the district maps in force on that day. Dates before it was added to the site were computed afterward from the approval data available on each date and the current economic data. The model stops updating on November 1.