# How the GB Elects forecast works ## Starting point Each constituency starts from its July 2024 result, plus some of the people who didn't vote then. Polls show those non-voters now lean towards Reform and Restore Britain. ## The polling average Every GB poll from the last eight weeks counts, weighted towards recent fieldwork (a poll's weight halves every two weeks) and larger samples, with no single pollster allowed to dominate. You can switch to a two- or four-week average in Adjust, which reacts faster but is noisier. Polls paid for by a party, such as Restore Britain's own Find Out Now polls, are listed but left out of the average. As a precaution, JL Partners polls are also listed but not counted, following public questions about how some of its polling was paid for; this is not a judgement on the accuracy of any of its polls. Pollsters that publish less often, such as Verian, Techne, Focaldata and Convergent Opinion, count in the weeks after each of their polls and so shape the trend over time. Many pollsters still don't offer Restore Britain as an option, so their figures are adjusted using the pollsters that do. While only a few polls prompted for it, that estimate is held back, so one poll can't set it for everyone. ## Where voters are moving How each party's vote varies with a seat's age, education, ethnicity and religion is checked against the seat-level figures published by YouGov, More in Common and Electoral Calculus MRPs. That check showed the model was too low for Reform in young, diverse seats, so those effects were softened. Polls break down current voting intention by how people voted in 2024. Those crosstabs split each party's 2024 voters between the parties they back now. They are pooled from More in Common (late August to mid October, about 13,000 interviews) and YouGov, with separate figures for Scotland and Wales. ## Scotland Scotland matters more than anywhere else, because the SNP's lead over Labour moves dozens of seats and pollsters disagree about it. The Scottish figure combines the full Scottish polls since July (three quarters of the weight) with the Scottish samples inside every GB poll, about 10,000 interviews (one quarter). Each pollster is corrected for its error at the May 2026 Holyrood election. More in Common had the SNP six points too low there, so its Scottish SNP figure is raised by 3.5 points. Find Out Now was three points too high. The simulations add a separate Labour–SNP swing (about ±2 points) on top of normal polling error. Conservative, Reform and Lib Dem voters strongly prefer a non-SNP winner when voting tactically, as Aberdeen South showed. ## Restore Britain Restore has no 2024 result, so its vote is built from where its supporters come from. That is about 11% of 2024 Reform voters, 2% of Conservatives, 1–2% of Labour and Lib Dem voters and 9% of non-voters. Its gains lean heavily towards places where Reform was already strong, and towards non-graduate, White British areas. That matches its crosstabs (about 2.5 times stronger among non-graduates) and Makerfield, where it took 6.8% while polling about 3% nationally. ## Local patterns and candidates People who switch tend to move to parties already strong in their area, so 2024's tactical patterns carry over. Census 2021 data adjusts the flows further, and sitting MPs, especially the 2024 intake, hold on to more of their vote. Strong independents such as Jeremy Corbyn keep a personal vote that national polls can't measure, so it is set seat by seat and not scaled to the polls. Personal votes are also much less certain than party votes, especially for an MP standing outside their old party for the first time, such as Zarah Sultana in Coventry South. The simulations give them a wider range, and when they rise or fall the difference mostly comes from the party they draw on. By-election results, Rupert Lowe in Great Yarmouth, the four Conservative MPs who joined Reform, and MPs who have left their party are all built in. ## Matching the polls and tactical voting Seat shares are scaled until they add up to the poll figures for England, Scotland and Wales. Within England, each region also takes account of regional crosstabs. Then supporters of parties out of contention can switch to one of the top two. How many switch depends on how close the race is and how far behind their own party is. Which way they switch comes from YouGov's tactical voting surveys of February and September 2026, which asked people how they would vote if only two named parties could win in their seat. For Conservative, Labour, Lib Dem, Reform and Green voters, in contests between those five, the model uses those answers directly, including voters who split both ways. Reform voters lean towards the Conservatives about as readily as Lib Dems lean towards Labour, while Conservatives split between Reform and Labour. One figure is adjusted: where Reform and the Conservatives are the top two, 45% of Labour voters are taken to back the Conservatives, between YouGov's September figure (43%) and Find Out Now's for Electoral Calculus in October (50%). People overstate what they would do in a hypothetical, so the stated shares are halved. That level was checked against the recent MRPs and by-elections. SNP, Plaid and Restore Britain voters, and contests YouGov didn't ask about such as SNP against Labour, use estimated preferences instead. ## Uncertainty Each simulation draws a national polling error for every party, regional and Scottish errors, a random error in every seat and a random strength of tactical voting. Seat shares shown are the average across simulations, so they line up with the win chances. In a close three-way race, the party with a slightly lower average share can still win more often, because tactical voting tends to break its way when the race tightens. ## Majorities A formal majority is 326 of 650. Sinn Féin's MPs don't take their seats, and the Speaker and three deputies don't vote, so about 320 is enough in practice. Every majority figure here uses that working majority. ## Limitations - Northern Ireland uses 2024 results with random swings, since there's no recent Westminster polling there. - Scottish education and age figures are estimates, because the census source covers England and Wales only. - Every party is assumed to stand everywhere, and candidate effects are modelled only in the seats listed above. - Restore Britain has beaten its national polling in real contests (8.7% in the Greater Manchester mayoral by-election), so it may be understated. - The forecast describes an election held now. The next one is due by August 2029.