Method

Settings

Weight of own vote vs neighbors

Every respondent gives two reads on each race: who they would vote for (own vote) and how they expect the people around them to vote (neighbors). This sets how much each counts in the forecast.

75% own vote / 25% neighbors

How a state is surveyed

Each state gets its own study: a random draw of adults (18+) from Foresight's census-grounded population of that state, so counties, ages and sexes come in their real proportions. Each respondent's race, education, urban/suburban/rural setting, and opinions are drawn from Foresight's digital twin population and reflect their county's population rather than national averages. Every respondent answers the same questionnaire about the races on their ballot. Only seats on the 2026 ballot are asked about.

Each respondent also knows how they voted in 2024, or that they didn't (those under 20 were too young to): drawn from a turnout and vote model fitted to their county's certified result and adjusted for their race, education, age and sex, so every county's 2024 vote is reproduced on average. They also know whether they voted in the 2022 midterms (those under 22 were too young to): about two in three 2024 voters did, fewer among the young and those without degrees (Census CPS), one in ten 2024 non-voters did (Pew validated voters), and each state's 2022 turnout is reproduced on average.

Three reads per race (two for the House)

1 · Own vote: 'who would you vote for?'.

2 · Neighbors: 'how will the people around you vote?'. Social-circle questions called more swing states than own-intention questions in the 2016 US and 2017 French elections (Galesic et al., Nature Human Behaviour, 2018). The 'neighbor' polls YouGov fielded in 2024 for the French trader who bet on Trump pointed the right way in MI, NC, PA and WI but overshot Trump, and YouGov found them less accurate than the standard question, so this read is blended, never used alone. Its value here: it sidesteps shy-voter effects and the respondent's own lean.

3 · Expectations: 'who will win?'. Expectations called 81% of state races vs 69% for intentions in ANES data, 1952–2008 (Rothschild & Wolfers). Shown for reference, not blended, and asked for Senate and governor races only: the House generic ballot has no named candidates to call.

Calibration against 2024

Respondents also report their 2024 vote and how they think their neighbors voted in 2024. We reweight Trump and Harris voters so the sample's recalled vote matches the certified state result (recalled-vote weighting, as several major pollsters now do), and subtract the gap between perceived and actual 2024 neighborhood votes from the neighbor read. This corrects a sample that leans one way. One caveat from 2026 polling: some real 2024 Trump voters no longer report it, so weighting to his 2024 share can slightly favor Republicans (about 1 point on the generic ballot).

Turnout is calibrated too. Respondents who didn't vote in 2024 (or were too young to) make up about a fifth of a state's likely voters in the simulation and lean heavily Democratic, while validated-voter data puts people who skipped the previous presidential election at roughly 3–8% of a midterm electorate (Pew: 68% of 2020 voters turned out again in 2022, as turnout fell from 66% to 46%). We cap them at 6% of each state's likely voters. One national turnout edge then weighs Harris voters ×1.04 and Trump voters ×0.96 in every state, set so the survey keeps its own national House vote: the cap changes how that vote spreads across states, not its level.

Counties are read from the respondents living there and, when they are few, pulled toward the county's 2024 result moved by the state's swing (8 pseudo-respondents of shrinkage), so a county with 3 respondents keeps its usual lean instead of swinging wildly.

Adjustments

Governors running for re-election get 6 points on the margin: governor races are more candidate-driven than federal ones, and the simulated respondents barely reward incumbents. Senate and House races get no bonus.

Where an independent's simulated support runs far beyond what their state's partisanship suggests (this year, Osborn in Nebraska), they keep 50% of their own-vote lead over the state's 2024 presidential result (Harris's margin, for an independent on the Democratic side), since independents usually fade by Election Day. Other independents are read as respondents answer, and the neighbor read is left as is.

Governor races then keep 80% of their distance from the national environment (the generic ballot read): the simulated respondents split their tickets less than real voters, and governor races, the most candidate-driven, are where real voters split them most. Senate and House reads are left as they are.

House seats

The survey asks the generic ballot (Democrat or Republican for Congress). Each of the 435 districts, as drawn for 2026 (new maps in nine states), starts from its 2024 presidential result within those lines (The Downballot's county-by-county calculations) and moves by its state's swing. A district adds its own deviation only as far as its counties have respondents: it needs about 400 before its own read weighs as much as the state's, so at 1,000 respondents per state the state swing dominates. States not surveyed yet take the surveyed states' average swing, with extra error. Each district also carries 5 points of error for local trends, which its own respondents can narrow, and 3.5 points for candidates and incumbency, which a generic ballot can't see. Seats whose finalists share a party, or with a single candidate, are decided; an independent standing in for a missing major-party candidate (Alaska's Hill, California's Kiley) is read on that side.

Seat odds come from 5,000 simulated elections with errors shared nationally (3 pts), within each state (its survey error) and per district. 218 seats for control.

Questionnaire (example for Georgia)

1. 1The general election is on Tuesday, November 3, 2026. Which of these best describes you?
2. 2If the November 3, 2026 election for U.S. Senate in Georgia were held today, who would you vote for?
3. 3Think about the people who live around you: your neighbors and others in your community. How do you think they will vote in the race for U.S. Senate in Georgia?
4. 4If the November 3, 2026 election for Governor of Georgia were held today, who would you vote for?
5. 5Think about the people who live around you: your neighbors and others in your community. How do you think they will vote in the race for Governor of Georgia?
6. 6If the November 3, 2026 election for the U.S. House of Representatives in your district were held today, would you vote for the Democratic candidate or the Republican candidate?
7. 7Think about the people who live around you: your neighbors and others in your community. How do you think they will vote for the U.S. House of Representatives?
8. 8In the 2024 presidential election, who did you vote for?
9. 9Thinking back to the 2024 presidential election, how do you think the people who live around you voted?