United Kingdom General electionby 15 Aug 2029in 999 days Majority for any party 34% Labour largest party 93% Calendar Track record Research
Research · United Kingdom · Start here

Forecasting a British general election when the polls are wrong: an overview

Abstract. Published polls have missed at every British general election since 2015, and in England mostly in the same directions. This paper sets out how Polling Forecast measures that error at each stage of a poll, corrects what respondents actually said, and carries the result through one seat model and one simulation; tested on four elections, the correction was closer than the pollsters' own figures at three and level at the fourth, but it did not foresee the 2024 overstatement of Labour.

Polling Forecast · updated 9 November 2026 · 18 min read · overview, polling error, England, raw answers, uniform swing, simulation, forecasting method

What has changed since this paper was written

Updated 9 November 2026. The rest of the paper is as first published, except where a sentence below says it was corrected.

This paper is the front door to the Polling Forecast research library. It states the problem the site is built around, describes each step of the forecast in order with a short formal statement, summarises the evidence that the method earns its place, and says plainly what the forecast does not claim. The final section is a guide to the other papers, each of which takes one part of the method and treats it in full.

The premise

Opinion polls are the best single guide to a British general election, and they are also reliably imperfect. The final published polls missed the result in 2015, 2017, 2019 and 2024. That alone would be unremarkable: sampling error guarantees some miss, and a comparative study of polling across many countries and decades found no evidence that polls have become worse over time (Jennings and Wlezien, 2018). What matters for a forecaster is the shape of the miss. If polling errors were independent from one election to the next, the best forecast would take the polls at face value and draw a sensible band around them. If they lean the same way repeatedly, a forecast that takes them at face value is leaving information on the table.

In England the misses have leaned. The table below gives the published final polls' error against the England result at each election, party by party. The Conservatives were understated at three of the four elections, and the UKIP, Brexit Party and Reform family (treated as one line throughout) was overstated at every one.

ElectionFinal pollsConLabLDReformGreenLab lead
20152−5.7+1.5+1.8+0.8+0.9+7.1
20174+0.6−3.9+0.7+2.1+0.6−4.5
20195−1.6+0.9−0.6+0.8+0.1+2.5
20246−3.5+4.3−0.9+2.0−0.6+7.8
Mean−2.6+0.7+0.3+1.4+0.2+3.3
Root mean square3.53.01.11.50.6
Live Published final polls minus the result, England, points (positive = the polls overstated the party). UKIP and the Brexit Party count as Reform. Computed from the database on 19 November 2026.

The inquiry into the 2015 miss concluded that the main cause was unrepresentative samples, not a late swing or respondents misreporting their intentions (Sturgis et al., 2016). Mellon and Prosser (2017) showed how the people who answer surveys are more politically engaged than the electorate and that non-voters are missing from samples in ways that weighting does not fully repair. Pollsters have responded with better weighting and with turnout models, and those adjustments have helped at some elections and hurt at others (Prosser and Mellon, 2018). The headline figure a pollster publishes is therefore the product of two things: a sample that leans in a fairly stable direction, and a set of corrections whose effect varies from one election to the next.

Most published forecasts begin from the pollsters' headline figures and treat them, implicitly, as unbiased estimates of the vote. Our starting point is that this assumption has been wrong at every election for a decade, and that the error can be partly anticipated if one looks inside the poll rather than at its headline.

What we do differently

The method rests on five choices. Each is described briefly here and defended at length in its own paper.

The first is to measure polling error at each stage of a poll, not only at its headline. A modern voting-intention poll passes through a sequence of steps: the weighted answers of everyone asked, then the pollster's turnout adjustment, then the handling of don't knows, then the published headline. Most pollsters print the tables for several of these stages. We store every table as printed and measure, for the final polls of 2015–2024 whose tables survive, how far the answers missed the England result at each stage. The finding that drives everything else is that the raw answers, before the pollster's turnout and don't-know steps, have missed in steadier directions than the headlines.

The second is to correct the raw answers, not the headline. We take each current poll's raw answers for England and remove a fixed share of the average past raw-answer miss, party by party. This is a shrinkage estimator in the ordinary sense (Efron and Morris, 1975): the correction is pulled halfway towards zero, because four elections are too few to trust the full historical average.

The third is to run every reading of the polls through the same seat model. Our projection, the published polls as they stand, and a set of alternative "lenses" (the polls as wrong as they were in 2015, in 2017, and so on) all become nation-level vote shares and pass through one uniform-swing projection. Differences between them are therefore differences in the reading of the polls, never artefacts of different models.

The fourth is to simulate uncertainty from real past error patterns. Rather than giving each party an independent error, each simulated election replays the joint error of one actual past election, scaled at random. When the Conservatives have been understated, Labour has usually been overstated at the same time, and the simulation keeps that correlation.

The fifth is to rate pollsters on their record and on how much of their working they publish, with the rating built so that an unproven pollster does not receive the benefit of the doubt.

The forecast, step by step

The forecast is a pipeline of seven steps. We state each formally but briefly; the notation is local to this paper, and the other papers develop it further. Throughout, \(p\) indexes parties, \(j\) indexes polls, \(t\) indexes past general elections (\(t \in \{2015, 2017, 2019, 2024\}\)) and \(n\) indexes the three nations of Great Britain. Vote shares are in percentage points.

Step 1: polls to nation averages

For each nation we keep a separate average, because Great Britain contains three party systems. The England average uses each poll's England figure (Step 2). Scotland and Wales use Scotland-only and Wales-only Westminster polls; where a nation has had none recently, its 2024 result is moved by the change in the GB polls since 2024.

Let \(J_n(d)\) be the set of qualifying polls in nation \(n\) formed by taking the latest poll from each pollster whose fieldwork ended within a window of \(W_n\) days before date \(d\), with \(W_n = 28\) for England and \(W_n = 120\) for Scotland and Wales, where polls are rarer. The published average for party \(p\) is

\[ H_{n,p}(d) = \frac{1}{|J_n(d)|} \sum_{j \in J_n(d)} y_{j,p}, \tag{1} \]

where \(y_{j,p}\) is poll \(j\)'s published share. Every pollster has equal weight, so that a pollster publishing weekly does not outweigh one publishing monthly. Restore Britain, a new party that not every pollster prompts, is averaged only over the polls that report it, and Others is 100 minus the named parties. Stand-alone MRP releases and party-commissioned polls are excluded.

Step 2: the England figure

Polling error is measured in England because the SNP and Plaid Cymru do not stand there and because England holds 543 of Great Britain's 632 seats. An England figure is taken only from the pollster's own tables, in this order of preference: the pollster's England column; otherwise the English regions added up by their printed weighted bases; otherwise, where no weighted bases are printed, the GB total minus Scotland and Wales at their shares of GB adults. Each poll records which route was used. A UK headline is never scaled to produce an England figure. The route matters because the rebuilt figures carry a small approximation error; From answers to headlines: the stage model of a published British poll and The UK data pipeline: collecting, reading and checking British pollsters' tables give the details and the tolerances.

Step 3: our reading of the raw answers

Let \(r_{j,p}\) be poll \(j\)'s raw England share for party \(p\): the weighted answers of everyone asked, before the pollster's turnout and don't-know steps, with don't knows, would-not-votes and refusals set aside and the parties rescaled to 100. For each past election \(t\), let \(\bar{r}_{t,p}\) be the average raw share across that election's final polls with surviving tables and \(v_{t,p}\) the England result. The average raw-answer miss is

\[ \bar{e}_p = \frac{1}{T} \sum_{t=1}^{T} \left( \bar{r}_{t,p} - v_{t,p} \right), \tag{2} \]

positive when the raw answers overstated the party, over the \(T\) = 4 elections 2015, 2017, 2019 and 2024. With shrinkage factor \(\lambda\) = 0.5, each current poll is read as

\[ \tilde{v}_{j,p} = \max\!\left(0,\; r_{j,p} - \lambda\, \bar{e}_p\right), \qquad \hat{v}_{j,p} = 100 \, \frac{\tilde{v}_{j,p}}{\sum_q \tilde{v}_{j,q}}. \tag{3} \]

The correction applies to the Conservatives, Labour, the Liberal Democrats, the Reform family and the Greens. Parties with no raw-answer history (Restore Britain, for one) are not corrected, only rescaled. Our England figure \(\hat{V}_p\) is then the equal-weight average of \(\hat{v}_{j,p}\) over the latest poll from each pollster whose full tables we hold, within the same 28-day window. The correction currently in use is shown below.

ConLabLDReformGreen
Average raw-answer miss−4.0+2.0+0.1+1.3+0.5
Correction (−0.5 × miss)+2.0−1.0−0.1−0.6−0.3
Live The correction applied to today's raw answers, England, points. Computed from the database on 19 November 2026.

Why half, and why the raw answers rather than the headline, are the subjects of Correcting the raw answers: the method behind our UK projection and Why we start from British pollsters' raw answers rather than the published headline.

Step 4: the projection as a lens

Every reading of the polls is expressed as an adjustment to the published England average. For our projection the adjustment is \(a_p = H_{E,p} - \hat{V}_p\), so that applying it returns our figure for the named parties; the result is rescaled to 100. A lens adjusts nation \(n\) by

\[ S_{n,p} = \frac{\max\!\left(0,\, H_{n,p} - a_{n,p}\right)}{\sum_q \max\!\left(0,\, H_{n,q} - a_{n,q}\right)} \times 100. \tag{4} \]

For the projection, \(a_{n,p} = 0\) in Scotland and Wales: the raw-answer correction is measured in England only, and correcting those nations' polls by their own past error made them worse when tested (see Scotland and Wales: sparse polls and no steady lean). Great Britain's shares are the three nations combined, weighted by each nation's 2024 valid vote. The projection becomes the headline only when at least two pollsters' full tables from the last 28 days are loaded; otherwise the site falls back to the published polls. The alternative lenses, and what each assumes, are set out in Lenses: reading the same British polls several ways.

Step 5: votes to seats by uniform swing

Each seat \(i\) in nation \(n(i)\) starts from its 2024 result on the current boundaries, \(b_{i,p}\). With \(B_{n,p}\) the nation's 2024 share, the projected share is

\[ s_{i,p} \propto \max\!\left(0,\; b_{i,p} + S_{n(i),p} - B_{n(i),p}\right), \tag{5} \]

rescaled to 100, and the seat goes to the party with the largest \(s_{i,p}\). This is uniform swing in the tradition of Butler and Stokes (1969), applied separately within each nation. A party that barely stood in 2024 is in effect added evenly across the nation's seats. Northern Ireland's 18 seats are not modelled and are held at their 2024 winners, and the Speaker's seat stays with the Speaker, which leaves 631 seats projected.

We chose uniform swing by a backtest. Starting from the notional 2019 result on the 2024 boundaries and giving each model the true 2024 nation shares, uniform swing called 479 of 631 seats correctly, a blend of uniform and proportional swing 461 and proportional swing 445. All three under-called Labour badly. From votes to Commons seats: UK swing models and their test reports the test.

Step 6: simulation

The central projection says what happens if our projection is exactly right. The simulation asks what happens if it is wrong in the ways polls have actually been wrong. In run \(k\) of 1,000, we draw a past election \(\tau_k\) uniformly from the four, a common scale \(z_k \sim N(0,1)\), and independent noise \(\epsilon_{k,p} \sim N(0,1)\). The national error applied to party \(p\) is

\[ \delta_{k,p} = z_k \, e_{\tau_k,p} + \sigma_N \, \epsilon_{k,p}, \qquad \sigma_N = 1, \tag{6} \]

where \(e_{t,p}\) is the published final polls' England error at election \(t\) (the first table above). Because \(z_k\) is centred on zero, the simulation assumes no direction; it borrows only the size and the shape of past misses, and parties move together as they did. Scotland and Wales use their own measured error from the same election year with the same \(z_k\). Each nation's shares become \(S_{n,p} - \delta_{k,p}\), rescaled; every seat is projected by equation (5); and each party's share in each seat then receives independent noise with standard deviation \(3\sqrt{s_{i,p}/25}\) points, so that local surprises are larger where a party is larger. Seat ranges, win probabilities and the chance of a hung parliament are read off the 1,000 runs. Simulating UK polling error: where the chances come from discusses the design.

Step 7: from today to polling day

Everything above answers the question "if the election were held today". The election-day forecast is a separate model, kept separate by design, because error measured on final polls says nothing about how opinion moves in the months before an election. Using every GB poll since 1983, we found that months or years out, the Conservatives, Labour and the Liberal Democrats finished much closer to their previous result than their polls implied, a pattern consistent with the wider finding that polls become more predictive as the election nears (Wlezien et al., 2013; Erikson and Wlezien, 2012). For those three parties, with \(d\) days to polling day and \(R_{n,p}\) the party's 2024 result in nation \(n\),

\[ S^{\mathrm{ED}}_{n,p} = R_{n,p} + w(d)\,\bigl(S_{n,p} - R_{n,p}\bigr), \qquad w(d) = f + (1-f)\,e^{-d/\kappa}. \tag{7} \]

The fitted floor is currently \(f\) = 0.62 and the time constant \(\kappa\) = 30 days. The challengers (the Reform family and the Greens), new parties, the SNP and Plaid Cymru are carried forward as polled, because pulling the challengers back made past predictions worse. Uncertainty for the time remaining grows with the logarithm of \(d\) and is added to the simulation. From today to polling day: how far British polls move before a general election gives the evidence.

The evidence so far

The central claim is modest and specific: over the four elections for which we hold the final polls' raw tables, reading the raw answers and taking off half their average past miss would have been closer to the England result than the pollsters' own headlines. The test is leave-one-out cross-validation (Stone, 1974): each election is predicted with the correction estimated from the other three only, so that no election helps to predict itself.

ElectionPollstersPollsters minus past missOursLead: pollstersLead: oursOurs vs pollsters
20152.131.751.79+7.1+5.5better
20171.582.541.29−4.5−3.3better
20190.790.620.64+2.7+1.9better
20242.191.822.18+7.8+7.8level
Mean1.671.681.475.54.63 better
Live Leave-one-out test: each election predicted with the correction estimated from the other elections only. Mean absolute error per party, England, points; lead = error in the Labour lead over the Conservatives (mean row: mean absolute). Computed from the database on 19 November 2026.

Averaged across the four elections, the mean absolute error per party was 1.47 points for our reading against 1.67 for the same polls' published headlines. Ours was better by more than 0.05 points at 3 elections and no worse at 4. The comparison is made on the 18 final polls whose raw tables survive, so that both columns describe the same polls. The obvious alternative, taking the published headlines and subtracting their own average past miss, scored 1.68: no better than doing nothing, because the headlines' misses swing from one election to the next.

Leave-one-out uses later elections to predict earlier ones, which no real forecaster could do. The forward test removes that advantage: each election from 2017 onwards is predicted only from the elections before it.

ElectionPollstersOurs: ½ × mean of all earlier½ × previous electionAll of previous election
2017 from 20151.58 (−3.9)1.17 (−2.3)1.17 (−2.3)2.03 (−3.1)
2019 from 2015, 20170.79 (+1.0)0.90 (+1.8)1.15 (+2.6)1.44 (+3.3)
2024 from 2015, 2017, 20192.19 (+4.8)2.18 (+5.7)2.00 (+5.2)1.43 (+4.2)
Mean1.521.421.441.64
Live Forward test: each election predicted from the elections before it only. Mean absolute error per party, England, points; Labour's own error in brackets. Computed from the database on 19 November 2026.

In the forward test the average miss was 1.42 for our method against 1.52 for the pollsters. Two further results matter. Assuming that the next election will repeat the previous election's miss in full did worse than the pollsters themselves, which is why the correction averages over all earlier elections. And, most importantly, no variant we tested, in this table or in the wider comparisons of Why we start from British pollsters' raw answers rather than the published headline, anticipated 2024: every one left Labour several points too high (the bracketed figures), because 2015–2019 gave no warning of a Labour overstatement of that size. What we don't know yet: limitations and open questions in our UK forecast treats this at length.

These are small margins on a small sample. A sample of 18 polls at 4 elections cannot establish that the correction will work next time; they establish that it would have worked, modestly and fairly consistently, over the period for which the evidence exists.

Today's projection

The table below is the headline forecast, as of 19 November 2026: Great Britain vote shares and seats on the published polls and on our projection, with the 80% range of seats from the simulation.

PartyPolls %Our bet %Seats (polls)Seats (our bet)80% range
Lab28.127.2329317237–372
Con21.622.614315488–217
LD10.110.2727262–75
Reform21.821.2373629–81
Green9.59.8444–6
Live Great Britain: vote share and seats on the polls as they stand against our projection, with the 80% range of seats from the simulation. Computed from the database on 19 November 2026.

On this reading the largest party would be Labour, with 317 seats against 326 needed for a majority. The published polls, run through the same seat model, give their largest party 329. The simulation puts the chance of a hung parliament at 66%. The projection is built from 5 pollsters whose full tables we hold (Find Out Now, YouGov, More in Common, Survation and Opinium).

How we get from the published average to our figure is set out as a sequence of steps in England. The first row is the average every other forecast would use; the last is ours. The rows beneath split the difference into three parts: which pollsters we can read, what undoing the pollsters' own adjustments does, and what our correction does.

StepConLabReformLDGreen
1. Average of all pollsters' published figures (5 pollsters)22.929.322.110.69.8
2. Only the 5 pollsters with full tables22.929.322.110.69.8
3. Their raw answers22.129.222.010.810.3
4. Correction+2.0−1.0−0.6−0.1−0.3
5. Our bet24.128.221.410.710.1
Pollster mix (2 − 1)0.00.00.00.00.0
Undoing the pollsters' adjustments (3 − 2)−0.8−0.1−0.1+0.2+0.6
Our correction after rescaling (5 − 3)+2.0−1.0−0.6−0.1−0.3
Total (5 − 1)+1.2−1.1−0.7+0.2+0.3
Live From the pollster average to our projection, England, today. Computed from the database on 19 November 2026.

This decomposition matters because the three parts often point in different directions. Undoing the pollsters' turnout and don't-know steps tends to move the Conservatives down and Labour up, because at every election since 2015 those steps raised the Conservatives and lowered Labour; our correction then pushes the other way. Some of the gap between our figure and the published average is also simply a matter of which pollsters publish their tables. The site states each part openly, so that a reader can see how much of our call is method and how much is sample.

What we do not claim

We do not claim to know why the raw answers lean as they do. The correction is empirical. It rests on the observation that the lean has been fairly stable in direction, and it would fail if the causes changed: if, for instance, the lean reflected something about the Conservatives being in government at all four elections studied, which they were.

We do not claim that the correction would have caught 2024. It did not. On Labour alone, in the forward test, our method was further from the result than the pollsters at two of the three elections it predicts (2019 and 2024), because the pollsters' own adjustments lowered Labour and were right to do so at three of four elections, and our method sets those adjustments aside.

We do not claim to model Scotland and Wales as carefully as England. Their polls are fewer, their error record is shorter, and our correction is not applied there.

We do not claim constituency-level precision. Uniform swing is a deliberately simple model that under-called Labour's seats in 2024 and cannot see tactical voting. We have not built a multilevel regression and post-stratification (MRP) model of the kind described by Hanretty, Lauderdale and Vivyan (2016) and Lauderdale et al. (2020), and we do not use other organisations' MRP estimates.

We do not claim that the "today" forecast predicts polling day. It answers a hypothetical question. The election-day forecast answers the real one, with much wider uncertainty, and is presented separately.

Finally, the site's ranges are calibrated to four past error patterns. If the next election's polling miss is larger than any of those, or differently shaped, the ranges will be too narrow.

A guide to the papers

The library has thirteen papers. This one, Forecasting a British general election when the polls are wrong: an overview, is the introduction; the others are described below in roughly the order in which the forecast uses them.

Measuring British polling error in England, 2015–2024 sets up the benchmark on which everything else rests. It defines a final poll, the rules for which polls enter the primary benchmark and which are kept for sensitivity analysis only, the grading of sources from the pollster's own table down to a press report, and the metrics used (mean absolute error per party in England, the Labour–Conservative lead error and others). It presents the published polls' England error at each election from 2015 to 2024 and discusses what the pattern can and cannot tell us.

From answers to headlines: the stage model of a published British poll describes the anatomy of a poll. It defines the stages we record, from the weighted answers of everyone asked through the pollster's turnout step to the headline, explains how England is extracted from each pollster's banner, and reports what the stages show historically: that the raw answers leaned against the Conservatives at every election, and that the pollsters' own adjustments helped at three elections and went badly wrong in 2017.

Correcting the raw answers: the method behind our UK projection is the core methods paper. It states the correction formally, reports the leave-one-out and forward tests in full, shows every shrinkage factor tried from none to the full historical miss, explains why half was chosen, and records the alternatives that were tested and rejected, including giving don't knows back to the party they voted for last time.

Why we start from British pollsters' raw answers rather than the published headline asks whether starting from the raw answers is actually better than correcting the headline. It compares both starting points, and a half-and-half blend, under several amounts of correction and both tests. It explains why the headlines' swinging misses make them hard to correct, and why none of the variants would have predicted Labour's 2024 overstatement.

Lenses: reading the same British polls several ways explains the idea of reading the same polls in different ways and passing each reading through the same seat model: the published polls, the average past miss, the misses of each past election replayed, each pollster corrected by its own record, and the pollsters most favourable to each party. It shows how far the seat picture depends on which reading one believes.

From votes to Commons seats: UK swing models and their test covers the translation from votes to seats: uniform, proportional and blended swing, the 2019-to-2024 backtest on which uniform swing won with 479 of 631 seats, the systematic under-call of Labour that all three share, and the conditions a demographic and tactical model will have to meet to replace it.

Simulating UK polling error: where the chances come from sets out how uncertainty is generated: replaying real past error patterns rather than independent party errors, the national and seat-level noise, how Scotland and Wales draw their own errors, and what the resulting ranges and probabilities do and do not mean.

From today to polling day: how far British polls move before a general election presents the separate model of how opinion moves between now and polling day, estimated from British polls before the seven elections from 2001 to 2024: the pull of the three older parties back towards their previous result, the absence of that pull for the challengers, the fitted curve, and the way uncertainty grows with the time left.

Scotland and Wales: sparse polls and no steady lean reports the polling record in Scotland and Wales from nation-only final polls, compares it with the Scottish and Welsh columns of GB polls (which miss by roughly twice as much), and explains why correcting those nations' polls by their own past miss was tested and not adopted.

Rating the British pollsters: accuracy, the Labour–Conservative lead and openness describes the rating out of 100: accuracy at checked final polls, the size of each pollster's lead miss, and how much of its working it publishes, with every record pulled towards that of a poor pollster so that a short record cannot earn a high score. It also presents the party-by-party record and the measure of sample skew before weighting.

The UK data pipeline: collecting, reading and checking British pollsters' tables documents how the data are gathered and checked: the readers written for each pollster's file format, the arithmetic checks every table must pass, the automatic daily check of pollsters' own sites under strict politeness rules, the storage of every source file by its hash, and the principle that nothing is overwritten.

What we don't know yet: limitations and open questions in our UK forecast collects every known weakness of the method in one place, from the small number of elections to the Labour weak spot, the England-only correction and the simplicity of the seat model, and states for each what evidence or work would address it.

References

Butler, D. and Stokes, D. (1969). Political Change in Britain. London: Macmillan.

Efron, B. and Morris, C. (1975). Data analysis using Stein's estimator and its generalizations. Journal of the American Statistical Association, 70(350), 311–319.

Erikson, R. S. and Wlezien, C. (2012). The Timeline of Presidential Elections: How Campaigns Do (and Do Not) Matter. Chicago: University of Chicago Press.

Hanretty, C., Lauderdale, B. and Vivyan, N. (2016). Combining national and constituency polling for forecasting. Electoral Studies, 41, 239–243.

Jennings, W. and Wlezien, C. (2018). Election polling errors across time and space. Nature Human Behaviour, 2(4), 276–283.

Lauderdale, B. E., Bailey, D., Blumenau, J. and Rivers, D. (2020). Model-based pre-election polling for national and sub-national outcomes in the US and UK. International Journal of Forecasting, 36(2), 399–413.

Mellon, J. and Prosser, C. (2017). Missing nonvoters and misweighted samples: explaining the 2015 great British polling miss. Public Opinion Quarterly, 81(3), 661–687.

Prosser, C. and Mellon, J. (2018). The twilight of the polls? A review of trends in polling accuracy and the causes of polling misses. Government and Opposition, 53(4), 757–790.

Stone, M. (1974). Cross-validatory choice and assessment of statistical predictions. Journal of the Royal Statistical Society: Series B, 36(2), 111–147.

Sturgis, P., Baker, N., Callegaro, M., Fisher, S., Green, J., Jennings, W., Kuha, J., Lauderdale, B. and Smith, P. (2016). Report of the Inquiry into the 2015 British General Election Opinion Polls. London: Market Research Society and British Polling Council.

Wlezien, C., Jennings, W., Fisher, S., Ford, R. and Pickup, M. (2013). Polls and the vote in Britain. Political Studies, 61(S1), 66–91.

How to cite

Polling Forecast (2026). Forecasting a British general election when the polls are wrong: an overview. Retrieved 20 November 2026 from https://www.pollingforecast.com/research/uk-overview. Live figures change as new polls arrive; the date above is when the text was last revised.

You may also like