Back to the forecast

How the forecast works

The forecast simulates the 2026 House, Senate and governor elections 20,000 times. Each simulation draws a national polling miss shared by every race, regional and demographic swings shared by similar places, and each race's own uncertainty, so close races tend to break together, as they do in real elections.

The national mood slider

The slider sets how far Democrats lead the national House vote (the generic ballot). Moving it shifts every simulated race at once: House margins move one point for each point on the slider, Senate races by 0.8 points (fitted on 326 Senate races from 2006 to 2024), and governor races by about half a point. The rest of each race, its polls, lean and candidates, stays fixed, so the slider answers "what if the national environment were different?".

It starts at today's polls, adjusted for one known bias: likely-voter screens rely on people saying they will vote, and young voters, the most Democratic group, overstate their turnout. The historical generic-ballot miss of about three points came mostly from registered-voter averages.

How the poll trend is built

The line on the poll charts is fitted to every individual poll, not to other people's averages. Each pollster's house effect (how far it runs from the field) is estimated and removed, as are the offsets between likely voters, registered voters and all adults. Polls fielded over several days are dated at their midpoint, and several versions of one survey (for example likely and registered voters) share one poll's weight.

The shaded band is the 95% range for where the polling average stands today: sampling error plus uncertainty in the house effects. It is not an Election Day range. Polls can miss the result by much more, and the model carries that separately: 4.1 points, one standard deviation, for the national environment.

House

Every district is on its 2026 map, including the mid-decade redraws. A district's expected margin starts from its partisan lean, incumbency and the candidates' fundraising, then blends in district polls in proportion to how informative they are. Polls from campaigns and party committees are corrected for their typical lean, and polls that disagree with each other count for less than their number suggests. On the map, outlined districts are open seats.

Senate

Senate races use the same approach with a prior refitted on every Senate race since 2006. Only polls naming the candidates actually on the ballot are used; three-way races use polls of the full field. Where an independent is the main challenger (for example in Nebraska), the chance shown is the independent's, and an independent win is not counted as a Democratic seat. That is why the forecast shows both "Democrats win the Senate" (51 Democratic seats) and "Republicans lose the majority" (counting independents).

Governors

Fuel prices

Track record

The same code was run as if it were September of past election years, using only the information available then.

Sources and credits