Election Forecasts
Weighted polling averages, seat ranges from thousands of simulations, uncertainty stated plainly.
United Statesmidterms · 3 Nov 2026
United Kingdomnext general election
Brazilrunoff · 25 Oct 2026
New Zealand7 Nov 2026
Spain29 Nov 2026
France2027 · early readUS & UK always on · other races join as votes near, retire once counted · updated 2026-10-30 · refreshes three times a week
A forecast is labelled speculation. Everything else in this atlas records what happened; this page estimates what might. The ranges are wide because they should be: polls taken months or years before an election have missed by six points and more, and every number below carries that history. When the ranges look absurdly wide, they are working correctly.
Scored in public
Each forecast is frozen before its election and graded once the result is known.
How this is measured
Forecast frozen 2026-10-03, 1 day out
Brier — · mean miss 3.12 points of vote share
All 6 of these figures were frozen as bare numbers with no range, so they count toward the mean miss but cannot say whether the range was honest. Every forecast frozen since 9 October 2026 carries an 80% range.
Flávio Bolsonaro led the first round 47.03% to Lula's 45.16%; runoff 25 October.
Forecast frozen 2026-10-24, 1 day out
Provisional count as of 26 October 2026Brier 0.6939 · 0/1 picks · mean miss 4.09 points of vote share · 0% of intervals contained the result
Luiz Inácio Lula da Silva won the runoff 51.79% to Flávio Bolsonaro's 48.21% (valid votes).
Forecast frozen 2026-10-26, 1 day out
Provisional count as of 30 October 2026Brier 0.0002 · 2/2 picks · seat MAE 1.39 · 100% of intervals contained the result
govBloc = Likud + Otzma + Shas + UTJ + RZP-Zehut, the snapshot's gov bloc.
Calibration: 93.8% of 80% ranges held the result (15 of 16). Track record
Frozen and waiting
The full ledger, alongside football, the NFL and college football on one axis, is at /predictions/scoreboard.
United States: the 2026 midterms
3 November 2026, 0.5 months away.
How this is measured
The House 435 seats · 218 to control
Median and 80% range of 20,000 simulations. Democrats take the House in 88.1% of them. A lead, not a lock: the generic ballot at this range still misses by ±2.4 points.
The aggregators
Individual-poll feeds thinned after FiveThirtyEight closed in 2025; the aggregators above are averaged as the national input.
The Senate 35 seats up · Democrats need 51, Republicans 50 with the Vice President
From today's 53–47: 22 Republican and 13 Democratic seats are on the ballot. The map, not the national mood, is the obstacle. Democrats can win the night and still miss the majority.
The races that decide it
Consensus of the eight ratings agencies (Cook, Inside Elections, Sabato and peers), converted to probabilities and simulated with a shared national swing.
The Governors 36 of 50 mansions on the ballot · 26 for a majority
From today's 26–24: 18 Republican and 18 Democratic mansions are on the ballot; the other 14 carry over. Governorships are independent state offices, so "majority" means most of the fifty, not control of anything, and the two figures fall short of 100% by the odds of an exact 25–25 split.
The races that decide it
Consensus of the ratings agencies (Cook, Inside Elections, Sabato, Race to the WH, RCP, Fox and VoteHub), converted to probabilities and simulated with a shared national swing.
Democrats win the House in 88.1% of simulations and control the Senate in 64.7%. The 14 races below are where those numbers are decided.
Battleground board
The Senate and governor races rated competitive, closest first.
How this is measured
The midterms map
Every Senate and governor race, shaded by who the forecast favours.
How this is measured
United Kingdom: the next general election
Due by August 2029 (modelled as 2029-05), 30.1 months away.
How this is measured
The polling average today · 80% range on polling day
Five parties within thirteen points of the lead: the most fragmented polling picture in modern British history, in a system built for two parties.
What the simulations say
Across 4,000 simulated elections: a hung parliament 47.6% of the time · Labour largest 51.9% · Reform UK largest 24.2% · Conservative largest 23.9% · Labour majority 34.1% · Reform UK majority 9.8%.
326 seats for a majority. Northern Ireland's 18 seats are held at their 2024 outcome.
Seat ranges projected seats and 80% range of the simulations · 650 seats
Projected seats are the average across every simulation, so together they fill the Commons: Labour 253, Conservative 140, Reform UK 131, Liberal Democrat 52, SNP 24, Green 15, Plaid Cymru 4, others 13, and Northern Ireland's 18 = 650. If the election were held today on these polls: Labour 298, Conservative 129, Reform UK 102, Liberal Democrat 64, SNP 15, Green 6, Plaid Cymru 4.
Proportional swing from the 2024 result in each of Great Britain's 632 constituencies, with national and per-seat error drawn from the historical record. First past the post makes seat counts hypersensitive: a two-point national move can swing a hundred seats, which is why the bars overlap so heavily.
The tracker fortnightly averages since the 2024 election
Brazil: the 2026 presidential election
Runoff 6 October 2030, 47.2 months away. The first round was voted on 4 October.
How this is measured
First round vote share, % · 80% range
Nobody near the 50% needed to win outright, so the numbers point firmly to a runoff.
The runoff 25 October 2026
F. Bolsonaro 52.2% of the two-way vote (45.9–58.6)
Bar shows the polls. The chance adds 2.6 points to F. Bolsonaro: half the 5.3-point margin the first-round polls missed by.
Head-to-head averages simulated with the historical error of Brazilian polls at this horizon: an 8-point first-round lead is far from safe once the field consolidates.
New Zealand: the 2026 general election
Set for 7 November 2026, 0.5 months away.
How this is measured
The polling average today · 80% range on polling day
MMP makes the blocs the real contest: National + ACT + NZ First against Labour + Greens + Te Pāti Māori, with TOP hovering at the threshold.
What the simulations say
Across 10,000 simulated elections: the governing right bloc holds its majority 75.8% of the time · the left bloc takes one 13% · neither reaches a majority in 11.2%.
A majority is half the House plus one. The House has 122 seats in the median simulation and grows past 120 through overhang in 81% of them.
Seat ranges median and 80% range · 122 seats in the median House
Who forms the government
National-led 77% · Labour-led 8% · another arrangement 2% · no government 14%
Chance each party is in government
Prime minister
- Christopher Luxon · National77%
- Chris Hipkins · Labour9%
- No government forms14%
A National-led government is the most likely result, in 77% of 10,000 simulated parliaments.
NZ First and ACT are the pivotal partners: each is needed for the majority in about 77% of simulations.
No government forms in 14%: in those parliaments no majority combination survives the pledges parties have made, even allowing some of them to break, or talks fail.
In each simulated Parliament, every combination of parties with a majority of the House (overhang included) is scored on how often those parties have governed together, how many seats it has to spare, how many parties it needs, and the pledges parties have made about each other (as of 2026-10-15). Support parties count toward the majority, as confidence-and-supply partners have since 1999. Pledges can break: a ruled-out partnership still happens in 15% of simulations, and in 35% when nothing else adds up. The weights come from 20 government formations in New Zealand and Israel since 1992. Israel's coalition model is not shown: it failed its back-test on the 2019 to 2022 elections.
Spain: the 29 November snap election
Called for 29 November 2026, 1 months away.
How this is measured
The polling average today · 80% range on polling day
Small provinces reward the two big parties, so seats are not proportional to votes. Sumar's successor, FA, and Podemos run separately for the first time.
What the simulations say
Across 10,000 simulated elections: PP and Vox together reach 176 seats 97.3% of the time · PP alone 0.1% · PP is the largest party 84.2% · the parties that elected Sánchez in 2023 hold a majority 2.2%.
350 seats, 176 for a majority. A majority of seats is not an investiture: abstentions decide that, below.
Who can be invested
A Spanish prime minister needs 176 yes votes on the first ballot, or more yes than no votes on the second, so the parties that abstain decide it. Share of 10,000 simulated Parliaments in which each route has the votes:
These are seat counts, not predictions of how parties will vote. Our odds of who is actually invested are withheld: the model has not passed its test on the 15 government formations since 1979 (it now handles the 2016 repeat election, but still misses 2023, when Junts backed Sánchez after an amnesty deal made after the vote), and we publish it only once it does. In that model Vox never votes for a PSOE candidate; every other party's refusal on record can break, and a party whose stated condition is not met votes no.
Seat ranges median and 80% range · 350 seats
France: the 2027 presidential election
Due spring 2027 (modelled as 2027-04), 5.4 months away.
How this is measured
First round scenario average, % · 80% range
33 scenario polls; latest 2026-10-22. The top two advance to the runoff, and second place is wide open.
Tested runoffs head-to-head averages
Le Pen 55.2% of the two-way vote (51.3–59.0)
Le Pen 56.2% of the two-way vote (52.3–60.1)
Le Pen 68.5% of the two-way vote (64.6–72.3)
Le Pen 56.5% of the two-way vote (52.6–60.3)
Le Pen 58.6% of the two-way vote (54.9–62.5)
Le Pen 57.8% of the two-way vote (54.1–61.6)
Nine months out in 2016, the eventual runoff pairing wasn't being polled at all. Treat these as today's temperature, nothing more.
How these forecasts work
TestedThe New Zealand and Israel seat draws fill the chamber and every coalition is a majority under property tests; the US and UK swing models are checked by hand. About the tiers
The averages
Each pollster's latest poll within 45 days, weighted by recency (14-day half-life) and sample size. A simple transparent average, validated on 22 past elections across four countries, where it called 87% of winners within 1.3 points.
Seats from votes
US House: the generic-ballot margin mapped through the 2012–2024 seats-votes relationship. US Senate and governors: the ratings agencies' consensus per race, converted to win probabilities and simulated with a shared national swing, the governors aggregated to mansions held out of 50, starting from today's 26–24 Republican edge. UK: each party's national movement applied proportionally to its 2024 result in all 632 GB constituencies (Commons Library data). New Zealand: Sainte-Laguë over the simulated party vote. Israel: the pollsters' own seat projections, averaged. Brazil and France: round-by-round candidate averages. Spain: each party's 2023 vote in every province swung to today's average, the 3% provincial threshold, then D'Hondt in 50 provinces (Ceuta and Melilla first past the post); published only while it passes a back-test on the 2023 election, where its ranges held all 11 parties' actual seats.
Why the ranges are wide
Polls this far from polling day are weather, not prophecy: three years out, national surveys have missed final results by six points and more (FiveThirtyEight's raw-polls archive; Jennings & Wlezien). The error scales into every simulation, which is why the honest answer is a range.
Sources and refresh
Polling and ratings tables from Wikipedia (CC BY-SA 4.0); GE2024 constituency results from the House of Commons Library (Open Parliament Licence); historical calibration from FiveThirtyEight's open archive (CC BY 4.0). A scheduled job (Mon/Wed/Fri) re-scrapes, re-simulates and republishes.