Election forecasts you can inspect.Tested models, shown with their sources and their uncertainty.
Menu · National environment
The national environment
What the country as a whole is telling the 2026 forecast: the generic ballot, the president's approval and the economy, through the cycle, and how each enters.
Series as of October 22, 2026
Generic ballot
D+8.7 from 15 polls, estimated October 18, 2026
The model's own average: recent polls weigh more, large and independent samples weigh more, and its estimation error is about 3.4 points. Explained on Polls.
Presidential approval
37% approve, 60% disapprove; mean of 12 aggregators, October 18, 2026
Each aggregator's latest published figure, averaged. Recorded daily by the site and back-filled from archived copies of the same table; the national model reads this mean moved onto Gallup's scale.
National House vote
D+6.1 the model's estimate of the national House popular vote margin
A blend of the generic ballot corrected for its past overstatement of Democrats (D+5.0) and fundamentals: approval, the midterm, real income growth and the last result (D+8.7). Over the 11 elections since 2004 this estimate missed the result by 2.0 points on average.
Generic ballot, Democrats minus Republicansgeneric ballot D+8.7
Each point is the estimate the model would have made that Sunday from the polls published by then; the band is its estimation error. Polls are read as they were released, so a week with few polls rests on fewer.
Presidential approval, mean of the aggregatorsapprove 36.9%, disapprove 60.4%
approvedisapprove
Each aggregator's latest figure as published by that Sunday, averaged over the aggregators with a figure less than 45 days old. The record begins with the first archived copy of the table that carries it.
Unemployment rateunemployment rate 4.2%Inflation, change over a yearinflation 3.7%Real disposable income, change over a yearreal disposable income 1.1%
Monthly, from the Federal Reserve Bank of St. Louis (FRED), as the national model reads them; it takes each year's August values.
How it enters the forecast
The House seat total. Every district starts from its own history and is moved by the national House vote estimate above, with that estimate's record of error; the few districts with polls of their own weigh them against it. The House page says which.
Statewide races. A Senate or governor race's starting point is its last result with that year's national tide taken out and this year's put in, before its polls are added. Where a race has no usable polls, that is most of the forecast.
What can still move. The forecast widens for the distance between today and Election Day: the generic ballot's own record of how far it drifts over that many days, and how well today's average is measured. A thinly polled week is a wider forecast.
Your own view. The national environment slider on You decide moves every race by the same amount and counts the simulations again, with marks for where past elections landed against their final polls.
The Methodology page gives the terms; the Accuracy page gives the record. Each number here is an input the forecast already uses, shown with the date it carries, not a forecast of its own.
Every midterm since 1946
The president's approval going into the autumn (Gallup, August to September) and how the president's party did in the House vote: its margin over the other party, in points. The 2026 row is this year's approval on the same scale and the model's estimate of the result.