Beginning this year, &Elections is producing General Election Equivalent Seat Estimates (GEESE) and General Election Equivalent Vote Estimates (GEEVE) for local election results to make it easier to more accurately and wholistically understand the attitudes of the electorate in between general elections than polling or modelling of local election results has hitherto allowed.

View &Elections' LEGEEE Model

&Elections Website

Overview of the Results

This GEESE for this year’s local elections indicates that Reform would be short of a majority by 31 seats if a general election coincided with the local elections. The Liberal Democrats would make modest gains; although, by virtue of being the only establishment party in England to not be eviscerated by Reform, they would form the Official Opposition – with their efficiently concentrated vote share meaning that Davey’s party would achieve this despite being 4.8pp behind the Conservatives’ second-place vote share, and still being fifth in overall vote share. Badenoch’s party, meanwhile, would face major losses overall, but the picture would remain mixed for them – as the unmitigated obliteration they would suffer in their rural heartlands to Reform and the Liberal Democrats would be somewhat (albeit nowhere near wholly) displaced by minor victories in affluent metropolitan areas where they would seek to make gains from Labour.

The governing Labour party, however, would see its vote share more than halve to 17.0% and would be pulverised to only 56 seats if the local election results were borne out in a general election scenario. This would be not only catastrophic, but could have a legitimate implication of being existentially damaging to the Labour Party, as despite the Tories being relegated to an abysmal showing, they would have a comparatively more resilient constituency of voters to build from, as the lack of an absolute electoral vapourisation despite being trounced by Reform on their right would be viewed as a minor victory. Labour, unlike the Conservatives, must contend with losing six-sevenths of their seats to both an insurgent party on the right in Reform, and an insurgent party on the left in the Greens, and would face losses to literally every single one of the other major parties. On this point, the inefficient concentration of Labour votes compared to other parties is also noteworthy because it indicates that their coalition of voters is, at least geographically, increasingly disparate – with heretofore unlosable inner metropolitan seats being swept up by the Greens (who would be four seats behind them despite having a vote share 1.7pp less than Starmer’s party) and their post-industrial heartlands, rife with political alienation and chronic under-investment, being mopped up by Reform.

As for the SNP and Plaid Cymru, the civic nationalist parties of Scotland and Wales would for the first time top the poll in each together, although the attitudes of voters for each of these parties are not always the same. In Scotland, the SNP’s vote share is set to increase only marginally on their performance in 2024, with the relative homogeneity of the political behaviour of most Scottish seats meaning that they would fall like dominoes back to the SNP as 2024 Scottish Labour voters desert Starmer’s party in droves, despite a malaise of political disenchantment for both. As for Plaid Cymru, supporters are much more hopeful about the potential the party has, with them having never previously formed a government in Cardiff Bay, and Labour voters appear to be abandoning their party in droves for them – meaning that, under this estimate, Plaid Cymru would win the vast majority of Welsh seats, with Reform winning the rest, and Labour on zero.

With regard to independents and others, it is worth examining that &Elections’ recent forecasts have not adequately accounted for the disconnect between minor right-wing and left-wing parties and independents: in the last general election, all of the independents that won did so on a left-wing platform; however, with the rise of Rupert Lowe’s Restore Britain to Farage’s right, those opting for others and independents in opinion polls are increasingly right-wing – but until now, no empirical data has been able to establish correctly where this disparate group of voters sit. &Elections’ GEESE would have others on 8 seats in total, with seats like Lowe’s own Great Yarmouth set to be in this column. Despite the increased accuracy at identifying where voters sit ideologically, it is still important to remember that the prominence of independents and others in parliamentary constituencies does not exactly match their prominence in council wards – perhaps best illustrated by Jeremy Corbyn’s Islington North being set to return to Labour under this estimate.

How It’s Calculated

England

The results used are based on the percentage of the absolute total number of votes cast for each party throughout the whole of each local authority. As, in some areas, there are multiple tiers of local government (in county and district councils), whereas in others there is just one – for elections to upper-tier county councils (East Sussex, Essex, Hampshire, Norfolk, Suffolk and each of the new Surrey shadow unitary authorities), the result for each of their component lower-tier districts is used. Where there were concurrent elections to both a district council and its county council (for example, Fareham and Hastings), the cumulative absolute vote of both elections is used.

Plainly, it is not feasible to make calculations based on every individual election in every ward for a multiplicity of reasons, including that there are thousands of wards and no publicly accessible centralised database of ward-level results of local elections; that several authorities’ ward boundaries have changed since the latest boundary review of parliamentary constituencies in 2023; and that county council divisions are not conterminous with the boundaries of their districts’ wards, which define parliamentary constituency boundaries. Because of this, another measure needs to be used to calculate where votes are concentrated – this is where &Elections’ underlying general election forecast, backdated to the day of the local elections, is utilised.

To ensure the results between council elections and the general election projection can be mapped accurately, each constituency is broken down into its individual local authority ward components, as established in the 2023 boundary review of parliamentary constituencies (with partial wards separated). After this, each ward is attributed a number of votes for each party, equal to the number of votes &Elections’ backdated forecast had for its constituency, multiplied by the proportion of registered voters in the whole of the constituency that reside there. For each local authority, the backdated forecast for each ward is accumulated to translate the backdated general election forecast from constituencies onto them. This means that for local authorities that had elections, the magnitudinal difference for each party in the percentage of forecast and actual votes is calculated. Then, the weighted average (by forecasted general election turnout) of the magnitudinal difference for each of the authorities that had elections is applied to the backdated forecast for those that did not to estimate how those local authorities would have voted, and the results and estimations of those that did and did not have elections are aggregated. This then flows back through the ward data, back to the individual constituencies – with the results for each ward multiplied by the magnitudinal difference between the backdated forecast for the ward’s constituency and local authority to ensure the estimation reflects how each individual constituency would have voted as opposed to the aggregate of its component local authorities.

A significant issue with this method of modelling, however, is that local authorities in England elect councillors via a hybrid of first past the post (FPTP) and plurality block voting (PBV), and where an authority exclusively uses one system or another, this theoretically does not cause issues, although, if a local authority has wards with varying numbers of members, this makes the tabulation of absolute votes throughout the authority a less reliable indicator of the public opinion expressed in the election. This is because, under PBV (which is used to elect candidates to multi-member wards) people are entitled to vote as many times as there are candidates to elect, meaning that if a party has 1,000 voters in a ward with one seat, but 3,000 voters in a ward with three seats, assuming the party stands the maximum number of candidates, 9,000 votes would be cast for the party in the three-member ward, and only 1,000 votes would be cast in the single-member ward. Now, usually the discrepancy between the cumulative absolute number of votes, and the number of votes weighed by turnout is not substantial, and this does not even apply to all authorities, but even using turnout-weighted vote could produce misleading results, as, where parties stand one or more, but below the maximum number of candidates they may field to a multi-member ward, their vote is artificially suppressed by prospective voters hypothetically using their other vote or votes for other parties. Without additional cumbersome modelling, it would be impossible to both model for, and assume what the genuine voting intention for all individual wards would be, and both would likely not add any additional substantial accuracy to the model either.

Even though this does not account for individual wards, it is nevertheless a relatively accurate way of estimating the result in each constituency. For example, in Makerfield – the seat on most people’s minds at the time of writing – &Elections’ LEGEEE model is accurate to the local elections results within the constituency for every party within 2.8pp.

Scotland

For the Scottish parliamentary election, the model fundamentally operates on the same principles as England’s – whereby the results of Westminster constituencies are mapped to compare alongside the results in Holyrood constituencies through the council wards that define their boundaries. Where this differs with England’s model, however, is that it does not have to estimate the results of entire constituencies, as every constituency had an election in Scotland. It is worth noting that the results for regional lists were discarded – as they have little bearing on a FPTP contest, even though it is hypothetically possible that the flexibility of the additional member system (AMS) as a hybrid system could mean that tactical voting in constituencies may be more prominent.

In seats that were not contested by all main parties, their vote share, if they had been standing, was estimated and calculated similarly to unknown results in England – where the result of the backdated forecast was applied, modified by the magnitudinal difference between known results and itself. This means that the forecast assumes a general election scenario where all parties stand candidates in all constituencies, which, while it would not be guaranteed to occur, was chosen as it is much more likely that, for example, the Scottish Greens (who only stood six constituency candidates at this election) would be closer to standing a full slate of candidates than only six in a general election. Importantly, though, it should be reconciled: this does have the potential to make the estimate slightly more inaccurate, as even though it attempts to mimic the realism of universal candidacy in a general election, it means that the estimate (particularly for the Scottish Greens) is grounded in comparatively much less empirical data, and that which does exist may still be flawed due to the multitude of variable dynamics that the use of different voting systems entails (such as tactical voting).

Wales

The new electoral system makes Wales much easier to forecast – both because of the conterminousness of constituency boundaries with those for Westminster and because the election has illustrated that constituency-based party list proportional representation (which normally favours large parties, because it can only be proportional to the closest sixth) elicits the dynamic of tactical voting that exists with FPTP. Because the conterminousness of boundaries means that each Senedd constituency is simply two adjacent Westminster constituencies combined – all that this estimate fundamentally has to do is separate the result of each Senedd constituency into two. This is achieved by taking the backdated forecast for each Westminster constituency and aggregating these into their Senedd constituency equivalents, before applying the magnitudinal difference between the Senedd results and Westminster forecast to Senedd constituencies onto the backdated forecast for each Westminster constituency.

Despite this voting system meaning that vote share can be more accurately translated to that for Westminster elections, the large size of constituencies needed to facilitate this system means that parties’ local support becomes diluted as voters choose parties they believe will have a better chance of winning nationally. This estimate, for example, has the Liberal Democrats in fourth in Brecon, Radnor and Cwm Tawe – on 13.7%, down from 29.5% at the general election – and has Plaid Cymru winning Cardiff South and Penarth despite coming fifth with 8.2% at the general election – having performed nearly half as well as the Greens on 14.5% (their best performance in Wales), whose support would only have increased by 0.3pp in this estimate.

Comparisons With Existing Models

Currently, Sky News and BBC News have their own models of estimating what the results of local elections would be as a nationwide poll. These models, however, are fundamentally different from &Elections’ LEGEEE. BBC News’ Projected National Share (PNS) and Sky News’ National Equivalent Vote (NEV) are both exclusively based on trends in England – as these are easier to consistently model year-on-year, as local elections do not happen in Wales or Scotland every year. The replicability of these models in every election is their primary advantage over &Elections’ projection. Furthermore, instead of being based on the results of every council up for election, these models are based on the results and swings of wards that are representative of the national electorate and in which the major parties have all contested.

Ultimately, because these projections are not designed to be representative of an election to the House of Commons, they are more unreliable at estimating what the nationwide vote share for an election to Parliament would be. Therein lies an issue with Sky News’ House of Commons projection based on its National Equivalent Vote estimate – as the NEV was not designed specifically to be representative of a parliamentary election, the figures it produces are less accurate. &Elections’ General Election Equivalent Vote/Seat Estimate (GEEV/SE), however, is specifically designed to capture what the equivalent result of a general election would be using the local election results. As this means many seats are grounded exclusively in the tangible results of the local elections in each local area, the GEESE model is more able to accurately capture the number of seats each party would have won. As Sky’s NEV seat estimate only applies the uniform national swing onto a generic projection, it risks encountering the issue of producing an estimate that would be impossible to replicate with the empirical results of the election – and, in fact, does encounter this issue – with Sky’s NEV estimating that the results of the local elections would yield the Greens only 13 seats, while, under &Elections’ GEESE model, 38 of the 52 seats it projects for the Greens are based on the actual results of the election – and despite it being hypothetically possible that unanticipated intra-authority discrepancies could mean that some seats estimated solely on election results could be inaccurately projected, it would be nigh impossible for that to have occurred in at least 25 of 38 instances.

The underlying methodology of &Elections’ projections is not changing. However, some alterations have been made to how this projection yields the seat totals projected for each party. Most forecasters’ seat totals are derived directly from the number of seats each party is anticipated to win at any given time. This may superficially seem the best way to project a result, although small discrepancies between the forecast and actual result in each constituency can cumulatively make the projected result for larger parties quite inaccurate, especially where there are many marginal constituencies. This means that marginal and safe seats alike are in a binary fashion either wholly or are in no way contributed to the tabulation of any party, hence removing arguably necessary nuance in the confidence of the result for marginal races.

Instead, some forecasters opt to calculate seat tabulations probabilistically, where the total number of seats for each party is derived from the sum of their normalised probability of winning in every constituency. This methodology has the disadvantage that it will almost certainly result in a visible discrepancy between the projected seat totals and the actual number of seats each party is projected to win under the forecast. It, though, is notoriously accurate at forecasting the result of larger parties, and is the methodology the broadcasters’ exit poll uses to calculate seat totals. Unlike the binary model, using probability ensures that the nuances of marginal races can be better accounted for, and the model veers away from forecasting unrealistic landslides. This does not, though, mean that it cannot produce flawed results. In the 2024 election, the methodology’s tendency to project Lib Dem victories at a highly probable but not certain rate in the few dozen constituencies they were challenging, and to forecast their chance of winning at below 0.1% in the hundreds of races they weren’t competitive in meant that this model underestimated the performance of the Liberal Democrats in the 2024 cycle compared to the binary model. As for the glaring overestimation of Reform in 2024, their high and relatively evenly spread vote share meant that there were many dozens of seats they were not forecast to win, but their probability of victory was just high enough that these seats were able to significantly increase their inaccurately forecast seat share despite the application of the binary model to the same projection oftentimes being almost fully accurate.

&Elections’ new combined model aims to take the advantages that both the binary and probabilistic models have, by employing an altered version of the probabilistic model that is more prone to producing nigh-binary results in constituencies where an outcome is more certain, ensuring that seats a party is not realistically competitive in do not contribute to overestimating the final outcome, and parties who are frontrunners in targeted campaigns are not underestimated. Additionally, this model still accounts for the lack of confidence in forecasting actual marginal races for a single party, so that they are not tabulated with the same certainty as safe seats.

Seat Tabulation in Practice

We can see in practice how the different models used impact the result by feeding in forecasts of the 2024 general election into the model. &Elections’ final call overestimated the swing from Conservative to Labour in the election, and employing the probabilistic model dampens the landslide projected, even though it remains inaccurate. Elsewhere, though, it goes from producing a very accurate forecast for the Liberal Democrats and Reform to one that is much more inaccurate. The combined model, whilst still projecting Labour and the Tories inaccurately, does moderate the error somewhat, by ensuring that the larger parties are more accurately forecast without jeopardising the accuracy of smaller parties.

If we use the most accurate MRP poll of the 2024 cycle, YouGov’s final one, the headline figures under the binary model produce a forecast that, albeit very close, is not perfect. When the Probabilistic model is used, the Labour result is forecast almost entirely correctly, and the result for the Conservatives, Liberal Democrats and Reform is still more accurate than the exit poll, but the fact that the Lib Dem and Reform figures produce similar errors to the exit poll illustrates in practice its inadequacy at more accurately forecasting smaller parties in the last election, even with extremely accurate polls.

One of the primary disadvantages with the probabilistic model remains that where a forecast over- or under-estimates the swing between the largest parties, but is relatively accurate at projecting the result for minor parties, then its corrections for larger parties would not yield absolute accuracy are counteracted by the increased inaccuracy of smaller parties, meaning it doesn’t actually make it a much better model for seat tabulation. &Elections’ combined model of tabulation would ensure that the strengths of &Elections 2024 model in projecting the results for minor parties would not be dampened into a forecast that was wholly mediocre if the probabilistic model were employed, even though the model has undergone several updates since the last election to better accuracy.

&Elections Website

Subscribe now

&Elections Methodology

Nationwide Polling Aggregation

&Elections has undertaken several methodological improvements for aggregating opinion polls. Before the last election, most polls were only weighted by sample size and recency, however, with the introduction of some historical polling data, &Elections also now weighs polls in part by the historical accuracy of each pollster, including the result for each party. Additionally, where a party (most often Plaid Cymru) is not prompted by a pollster, the result for them is inferred based on recent polls at that time. To ensure all polls with a UK-wide sample can be included to project the vote share across Great Britain, such polls have the results for “others” adjusted to exclude Northern Ireland. Elsewhere, minor changes to the pre-existing methods of weighting and adjusting polls (primarily based on recency and how much polls are adjusted based on the recent trends) have been undertaken to further better accuracy in projecting nationwide vote share.

Regional Polling

Instead of applying national polling uniformly to each constituency, &Elections’ model now applies polls on a regional basis. For the statistical regions of Great Britain that are frequently polled for WVI, this involves aggregating and collating polls using a similar methodology for how nationwide WVI polls are aggregated. As regional polls are scarcer than nationwide polls, though, alterations to increase the longevity of the total weight of polls for each region by adjusting the result based on both the trend of recently undertaken polls, and marginally by the uniform national swing (UNS) since the weighted average date of conduction of polls for that region. The inferred result for non-polled regions is determined based on normalising the projections for each region to ensure the results for them combined remain consistent with national WVI.

Constituency Polling

Constituency polls are usually much more reliable at producing a snapshot of public opinion in a specific area, although their scarcity means they do not help in illustrating how public sentiment has changed in each area over time, which means they can become inaccurate when their statistical longevity in the weight of a constituency’s total forecast is too long. To increase their longevity, the only major change to constituency polling since the general election has been to alter the result of each poll by the normalised magnitude of the UNS for each party since the conduction of each poll.

Demographic Data and Projections

&Elections now uses data from the 2021 England and Wales and 2022 Scotland censuses; data from various pollsters’ post-election analyses; polls with reliable demographic crosstabs and a sample of over 10,000 (to avoid the unnecessary noise of the more unreliable sub-sectional data of smaller polls); and current WVI figures to produce thorough demographic breakdowns in both population and voting intention for every constituency.

New APSEE and LETA Models

With the inclusion of constituency-level information and forecasts for where demographics reside and how they vote, &Elections now uses an all parties standing every estimate (APSEE) for how each constituency would have voted if all seven major parties stood candidates everywhere (in the regions they do), and if every constituency had at least one minor party candidate standing. &Elections’ new linear elections trend adjustment (LETA) model infers how constituencies would vote today provided the magnitude of the swing between the 2019 notional general election result and the 2024 APSEE remain at a 20% constancy to the present. The LETA model is used as part of the foundational projection of &Elections’ forecast, alongside &Elections’ MRP aggregation, and is designed to be heavily weighted in the interim between a general election and when several pollsters have released MRP polls, although its weight does also increase when MRP polls are more scarce.

MRP Poll Weighting

Since the general election, significant changes have been made to how MRP polls are weighted. MRP polls are now weighted based on recency: both based on the literal recency of the poll; but more prominently based on the recency in the chronology of MRP polls conducted by each pollster - as this allows the model to weigh the methodologies of each pollster more proportionately. Additionally, MRP polls are now weighted based on accuracy - both on a nationwide level, and on a seat-by-seat basis. The current model weighs each candidate based on the accuracy of each pollsters’ projection of them in the last election, and either amplifies or deflates the weight of each constituency as a whole for each pollster based on whether their final MRP poll for the 2024 cycle projected the result in each seat correctly.

Separating “Other” Candidates

In calculating projections, &Elections now consciously separates the most prominent minor party or candidate in a constituency from all others, primarily to avoid a sitaution where “others” cumulatively are projected to win a seat, but the most prominent among them in actuality is not. This also better accounts for MRP polls that do separate between then the most prominent and all other minor candidates, and better allows to gauge how support for minor candidates changes over time in each constituency.

Seat Tabulation

&Elections now uses a probability-inferred formula to tabulate seats, instead of simply producing a total of the number of seats each party is expected to win. Probabilistic models tend to be better for accounting for marginal seats, where a winner is not guaranteed, and on the balance of probability, where a single party may be more likely to win in various single constituencies, it may simultaneously be not likely that that is repeated in every marginal race where a specific party is favoured. This form of modelling is usually incredibly accurate at projecting the final result for larger parties (and is the methodology the exit poll uses) although can be less accurate with smaller parties, so &Elections now uses a slightly modified version of this model of seat tabulation to project the final outcome as accurately as possible.

&Elections website

Subscribe now

Stage One: The Foundational Model

The foundational model of my projection is comprised of a mix of two key elements: the linear elections trends adjustment (LETA) model and the multi-level regression with post-stratification (MRP) aggregation model. Where few MRP polls exist, or where none have been conducted for a while, the foundational model will move toward favouring the LETA model and vice versa.

The LETA model assumes that the swing between the 2^nd most recent general election (2019) and &Elections’ all parties standing everywhere estimate (APSEE) of the most recent general election (2024) for all parties in every constituency maintains a 20% constancy to the present. The APSEE involves inferring the results for parties in constituencies where no candidate stood in the previous election by using demographic data from the 2021 England and Wales and 2022 Scotland censuses, coupled with &Elections’ model for how different demographics voted in 2024.

The MRP aggregation model involves weighing MRP polls based on the accuracy of the pollster’s final MRP poll for the 2024 general election; the sample size of the poll; and the recency of the poll. As some pollsters conduct MRP polls on a much more frequent basis than others, to ensure that the methodologies of each pollster are accounted for proportionately, recency weighting does not significantly account for how recently every poll was conducted, but rather weighing based on the recency of each poll in the chronology of polls by any specific organisation. Furthermore, &Elections’ model goes further by weighing each candidate in each constituency based on the accuracy of each pollster in forecasting their vote share in the 2024 cycle, followed by also marginally adjusting the result based on its inaccuracy too.

Stage Two: The Final Model

The final model involves mapping the current voting intention for each region onto the cumulative results for each region under the foundational model, followed by applying constituency-level opinion polls (CLOPs), if applicable.

The current Westminster voting intention (WVI) for each region is determined by a combination of national and regional opinion polls. As not all regions are polled, it is initially assumed that each region has a uniform swing between the last general election and current WVI, before polling for regions is added (which currently only exists for Scotland, Wales and London) with the results for other regions being adjusted to ensure WVI figures are consistent with the cumulative projection for each region. These results are mapped onto the findings of the foundational model for each region to give an accurate projection for every constituency consistent with current WVI. Finally, where CLOPs and by-elections exist, the results of these are weighted and adjusted, and the findings are then applied to the constituencies where they apply to produce the final projection.

Nationwide polling and regional polling are both weighted by recency and sample size. As regional polls are scarcer, polls are more statistically significant for a longer period. Nationwide polling also accounts for accuracy, based on the accuracy of each pollster’s final WVI poll before an election in all elections since 1970. Both national and regional WVI polls have the trends among recent polls mapped onto the projected result to allow the model to be better at registering when significant changes happen in polling and so that it more reflects voting intention now, rather than simply reflecting the weighted polls over a recent period. Furthermore, this is useful for regional polls, as it allows the model to maintain accuracy when polls are more infrequent by following the general trends of the region, although the total significance of regional polls for each region is weighted down when this is the case. CLOPs are weighted slightly differently – whilst sample size is weighted broadly similarly, for the reason that they do tend to provide a much more accurate snapshot of the political landscape than any MRP poll, to ensure that they can maintain their statistical significance over time, each constituency poll is adjusted by the uniform national swing between when they were conducted and the present.

&Elections website

Subscribe now