First J.L. Partners SRP Model Shows Labour On Course For A Landslide
With less than a week to go our new SRP model shows Labour on course for the biggest majority in modern British political history, taking 450 seats in the House of Commons leading to a majority of 250. Meanwhile, the Conservatives stand to lose around 260 seats - one of the greatest seat losses in over a century. The Liberal Democrats are also predicted to return to a similar position as they held under the New Labour government from 1997 to 2010, increasing their seat number by 44. Whilst Reform are the third biggest party in terms of vote share they stand to only win 2 seats due to the inefficiency of their vote. In Scotland, the Scottish National Party are set to lose around 33 seats, dropping from 48 in 2019 to a projected 15 on July 4th. Plaid Cymru are set to gain 1 seat on their 2019 notional seat count with 3 MPs.
Labour are set for a 250 Seat Majority
Currently, modelling suggests that the Labour Party could be in line for the biggest majority of any single party - on a national vote share that is smaller than their 2017 result of 40%. This outcome is being driven in part by the splintering of the Conservative vote, which is being lost mainly to the insurgent Reform UK Party as well as the Labour Party. Whilst the Labour national vote share has remained relatively constant over the course of the campaign, there have been shifts in the Conservative and Reform UK vote shares. This movement mainly started after Nigel Farage announced his intention to run for the Clacton seat - although a longer term trend had already developed by the start of the campaign.
The seat predictions make grim reading for the Conservative party. Current projections suggest they stand to lose 260 seats in this election, placing them dangerously close to a complete collapse of their parliamentary party. The most successful election-winning machine in European history is now teetering on a knife-edge - a few points movement against the Conservatives could almost wipe them out.
Despite potentially getting a lower share of the vote now than in 2017, Labour stands to take the largest single-party majority in British Political history. Increasing their seat count by 248 from the 2019 notional results. The Labour party would have one of the most efficient vote shares in history, in part due to the ascension of Reform UK.
The Liberal Democrats are on course to recover their seat count to around the same level they won in 2001, winning 55 seats in the House of Commons. This is an increase of 44 on their 2019 notional results, with a vote share increase of just 1 point - a clear demonstration of the efficiency of the Liberal Democrat vote.
The Scottish National Party are also on course to lose 33 seats, falling from a notional 48 to a projection of just 15. This is due to the Labour Party dominating the industrial, central belt of Scotland.
Even the best case scenario for the Conservative Party projects them winning the lowest number of seats they have won post-World War One. The lowest seat estimate for the Labour Party would still have them winning more seats than any party has ever won giving it the largest majority that any single party has even been handed by the electorate.
Whilst our model does not suggest there is a real possibility of the Conservatives being the third party in the House of Commons, the 81 low seat estimate raises to possibility of a Conservative party that may struggle to recover. On the other end of the estimates, the Conservatives could win up to 139 seats - still lower than their 1997 results but no where near the same level of catastrophe as the low estimate of 89.
A worst case outcome for the Liberal Democrats would still land them on around the same number of seats they won under Paddy Ashdown in 1997. The upper end of the estimates would take them back to their position under Nick Clegg in 2010, giving them 63 seats in the House of Commons. Most of these extra seats would be gained at the cost of the Conservative Party.
Reform UK are likely to win five or fewer seats, with a central estimate of two. This is mainly due to the inefficiency of their vote, which has essentially seen a uniform increase from 0 points to around 15 - 25 points across the country. This range is generally not big enough to win many seats, but it is big enough to hand Labour a large number of previous Conservative seats.
The Conservatives Stand To Lose 260 Seats
The electoral map of Britain is likely to see a generational change in the coming election and the SRP projections suggest that Labour is set to dominate British politics for at least the next five years. The insurgency of Reform UK is also set to split the right-wing vote. This will shatter the historic electoral record of the Conservative Party.
Scotland’s electoral landscape is also predicted to drastically change. The Scottish National Party will remain relatively popular - winning around 30% of the Scottish vote - but only win around a quarter of the Scottish Westminster seats. Labour are only 5 points ahead North of the Border, but could take 36 of the Westminster seats. This is over double the Scottish National Party’s projected wins. This is attests to the concentration of Labour Party support in the Industrial Heartlands of Scotland.
Notably, there are 80 seats where the Conservatives are second in marginal seats (< 5 points behind the leader). This means that a small uniform national swing could hand the party many of the seats they stand to lose. Of the 80 marginal seats where the Conservative are second, 60 are projected to be won by the Labour Party whilst 17 are projected to be taken by the Liberal Democrats. Despite this potentially good news for the Conservatives, there are no safe Conservative seats (margin of victory > 15 points). Every seat they hold will have a margin less than 15 points. The safest seats for the Conservatives are projected to be Louth & Horncastle (Victoria Atkins) and Weald of Kent (Katie Lam) with margins of just under 15 points and 13 points respectively.
That being said, a uniform national swing of around 2 points would see the Conservative Party go from 105 seats to 139 seats and a swing of 3 points would see them gain 153 seats. This highlights the knife-edge on which this election rests. A swing of just over 3 points from Reform UK to the Conservatives could see the Tories obtain around the same number of seats they won in the landslide Labour victor of 1997.
By contrast, Labour is currently projected to win 253 seats that would be classed as safe Labour seats. This includes Keir Starmer’s which we project him winning by 50 points with the Liberal Democrats coming second on 12%. There are 53 seats where Labour are second in marginal seats of which, 47 are currently projected to be held by the Conservatives. Clearly a uniform swing of around 5 - 6 points from other parties to Labour could see the party break the 500 barrier and decimate the Conservative Party. A uniform swing of around a single point to Labour would take the them from 450 seats to around 465 seats, which in turn would reduce the Conservative party seat count by 12, taking them below the 100 seat threshold.
Out of the 15 SNP seats projected, 5 of them are marginal with 2 of the seats placing the Conservatives second. One of these seats is Aberdeenshire North & Moray East where the leader of the Scottish Conservatives, Douglas Ross, is standing. A swing of just two points would see Ross take the seat from the SNP. The SNP is second in 42 of the Scottish seats with 9 of these being marginal seats. Hence a small change from Labour to the SNP could see them breaking the 20 seat threshold.
Reform UK are projected to win 2 seats, Clacton (Nigel Farage) and Ashfield (Lee Anderson) - a coup for the leader and the incumbent. They are projected to be runner up in 99 seats with just 2 of these predicted to be marginal seats and 87 either safe or lean Labour seats. The majority of these seats, around 68, are in the North of England (North East, North West and Yorkshire & Humber). Currently we are projecting that Richard Tice will lose the Boston & Skegness seat to the Conservative candidate by 32% to 37%. In 7 seats Reform UK is less than 10 points behind the front runner and given the uniform nature of Reform’s vote increase it will be difficult for them to win many more seats.
The Notable Losses for the Conservatives
A number of senior Conservative MPs and Ministers are set to lose their seats according to our analysis, including leadership hopeful Penny Mordaunt. She is set to lose her seat to Labour by just four points, owing to a high Reform UK vote share in her seat. If just a sixth of Reform UK voters switched to the Conservatives in Mordaunt’s seat then she may hold onto Portsmouth North.
Another notable loser includes the current Chancellor of the Exchequer, Jeremy Hunt, who has held various Cabinet positions over the past 14 years. Currently he stands to lose the Godalming & Ash constituency by 12 points to the Liberal Democrats. It is hard to see a scenario, based off the predictions from our SRP, whereby he could win this seat.
Lucy Frazer, the current Secretary of State for Culture, Media and Sport, is set to also lose her seat to the Liberal Democrats in what is a potentially a three-way race between the Conservatives, Liberal Democrats and Labour. Given the “left” coalition vote share totals 54%, it is again hard to see how she could win this seat.
Other notable losses can be found here.
Changing Tides
The Conservatives are currently on track to lose 11 points of their 2019 vote share to Reform UK, with around 1 in 4 previous Conservative voters switching to the new party. A further 7 points are lost directly to the Labour party as 17% of previous Conservatives defect to the Labour party.
The Labour Party has done a much better job at retaining its previous support with 80% of those who voted for Corbyn in 2019 intending to vote for the Starmer’s party on July 4th. This in turn means that 26 of the 38 points Labour is projected to win comes from these previous Labour voters. Whilst 2019 Conservatives appear happy to shop around for their new party, Labour voters are much less likely to switch.
Meanwhile, the Liberal Democrats retain just under two-thirds of their previous voter base but make gains from defecting Conservative and Labour voters. These previous Conservative voters account for 3 of the points the Liberal Democrats stand to win and the Labour switchers account for 1.
Ultimately, Reform UK’s base is largely made up of previous Conservative voters, with around 65% of their final vote share coming from these disaffected Conservatives. Some previous Labour voters do shift to the Reform UK camp but only a small number of the party’s final headline vote intention is composed of these switchers.
This Election Will See Major Swings to Labour and Liberal Democrats
The swing map indicates a significant projected shift from the Conservatives to Labour across most of England. The exception is the South West, where the Conservatives are primarily losing ground to the Liberal Democrats. The largest predicted swings from Conservative to Labour are occurring in constituencies with the strongest Conservative majorities from the 2019 general election. This trend is particularly noticeable in the East of England, West Midlands, Skipton & Ripon in Yorkshire and the Humber, and Wiltshire in the South West. The efficiency of the Labour swing is such that there are actually small swings against Labour in some of their strongest seats, especially in London. Reform UK are projected to receive the largest swing of all – 39.9% in Clacton.
In Scotland, the Conservatives are projected to lose support in most areas, except the Highlands and Aberdeen North, with the majority of the support shifting to the Scottish National Party (SNP). However, in the central belt, the shift is from the Scottish National party to Labour. In the Highlands, support for the SNP is predicted to decrease, to the main three parties. Aberdeen North and Inverness, Skye & West Ross-shire are the only constituencies to see an increase in predicted Conservative support in Scotland.
Projections in Wales suggest most constituencies will experience a decline in Conservative support, generally to Labour’s benefit. Exceptions are in areas previously won by Plaid Cymru in 2019, where support for Plaid Cymru is anticipated to increase. Seats held by Labour in Wales since the 2019 general election are predicted to show only a moderate increase in Labour support.
The Tightest Margins
Labour are projected to win three of the ten most marginal seats, Liberal Democrats are projected to win four, Conservatives two and SNP one. Labour are set to win North Warwickshire and Bedworth, Waveney Valley, and North Bedfordshire by less than one point. Whilst 140 seats are projected to be won by less than 5 points. Hence a uniform national swing of around 2 points would see the Conservative Party go from 105 seats to 139 seats and a swing of 3 points would see them gain 153 seats.
Of the 30 most marginal seats, 18 of them are direct contests between the Conservatives and Labour. Tiny swings to either of these parties dramactically changing the outcome of the election for the Conservative Party. In 13 of the most marginal Labour/Conservative contests Reform UK achieves a vote share of at least 20%. Clearly in these marginals Reform UK are hurting the Conservatives.
In the border constituency of Berwickshire, Roxburgh & Selkirk the Conservatives are only one point behind the Scottish Nationalists. Small amounts of “unionist” tactical voting could see the constituency fall back in to Conservative hands. This seat has often been a stronghold of the Scottish Conservatives, but a collapse in their vote share to just 11% in Scotland might lead to the loss of one of their historic seats.
Where Could Reform UK Win Next?
In the closest races where Reform UK are in second place, 5 are projected to be won by the Conservatives and 15 are set to return Labour MPs. Reform UK looks set to finish behind Labour in seats such as Doncaster Central, Barnsley North, and Doncaster East and the Isle of Axholme. Richard Tice is forecast to lose the Boston & Skegness seat to the Conservatives by 5 points.
Across the model Reform are projected to be runner up in 99 seats with just 2 of these considered marginal, 87 of those seats are projected to be either safe or lean Labour seats. Notably, 68 of these seats are in the North of England (North East, North West and Yorkshire & Humber), these second place finishes will boost Reform UK’s claim to be the party of opposition to Labour in the North.
Conservative-Lib Dem Marginals
Theresa May’s former seat of Maidenhead is currently projected to be won by the Conservatives with 35% of the vote compared to 34% going to the Liberal Democrats. Of the 20 most marginal seats where the Liberal Democrats are projected to finish second, all currently projected to return Conservative MPs. The two most marginal seats, South Devon and Chelmsford, in our SRP model are both won by the Liberal Democrat currently.
Many of these marginal seats are in the South West and around the Cotswolds highlighting the propensity of a “yellow” wall to build up in those areas. Small amounts of tactical voting in these constituencies could see the Liberal Democrats win a further 5 seats in the South West. This would take them up to 60 MPs and putting the Conservative on the precipice of less than 100 MPs.
The Highest Conservative Vote Share Is Just 42%
The highest projected vote share for the Conservatives in any constituency is in Louth & Horncastle where the Conservative candidate is predicted to win 42% of the vote share. This is in stark contrast to 2019 where the Conservatives won 77% of the vote in the Castle Point constituency. Many of the swings in constituencies are beyond uniform and move into the proportional linear model of swing. To understand why the Conservatives are projected to lose 260 seats, and why Reform UK will only win 2 seats despite potentially winning 17% of the vote, it is useful to look at the distribution of constituency vote shares.
In over 200 constituencies the Conservatives are predicted to win between 25 - 30% of the vote share, with just 98 seats projected to have a Conservative vote share above 30%. The concentration of Conservative vote share around the national value of 22% gives some explanation as to why the Conservatives are on course for a historic defeat. They struggle to push their vote share in constituencies across the country much above this average value.
This concentration is also Reform UK’s issue. Their distribution shows a clustering around the 17% and given the uniform nature of their swing - most constituencies are in the 15 - 25% range - the party struggles to capitalise on their surge of support. Outside of a handful of seats, the party struggle to concentrate their vote share in the same way the Liberal Democrats do.
On the other hand, Labour’s vote share spans a much wider range of values from 10% to 75%. This means the party has 110 seats where they are predicted to win 50% or more of the vote share, and 283 where they are projected to win 40% or more. This spread of seats makes it very easy for the party to potentially pick up almost 300 seats without much effort owing to the spread of their vote. A similar situation is true for the Liberal Democrats, but in fewer seats. The efficiency of the vote of these two parties partly explains why they will see a surge of seats across the country.
Undecided Voters Are Falling To Labour
At JLP we use machine learning to impute the undecided voters. We use this approach as we believe it assumes as little about undecided voters as possible. The model learns the decision-making process voters go through and then applies this to the undecided voters.
Our latest analysis uses 1,026 GB adults who said they were undecided. Over half of these undecided voters were Conservative in 2019 but now just over fifth are projected to vote Tory with just 38% of previous Conservatives predicted to vote for the party again. The bulk of the Conservative to undecided loses are predicted to go to Reform UK, with 32% of previous Conservative abandoning the Tories for Reform UK. These voters form 81% of the undecided to Reform UK vote share.
Currently, the Labour party is predicted to be the biggest winner amongst the undecided voters. Over a quarter of undecided voters are predicted to end up voting for the Labour Party on July 4th with 7 points coming directly from Conservative to Labour switchers. Whilst the percentage of imputed did not votes is only 10%, this should be taken in the context of only 7% of undecideds not voting in 2019. This strongly suggests that there will be a number of undecided voters who will not vote in the coming election - above the 10% predicted here.
That being said, this voting bloc is incredibly volatile. Over the course of the past few months various parties have been projected to be the most popular amongst this group of voters. Typically the bloc has split between the Conservatives, Reform UK and Labour. A Reform UK surge occured in the weeks after Nigel Farage’s annoucenment to run. This was then followed by a drop in undecided Reform UK support before recovering in out latest data. The Liberal Democrat’s fortunes success with this group has waxed and waned over the past few months, ranging from a minimum of 0 points to a high of 17 points.
The Fracturing of Johnson’s Coalition
The vote share of 2019 Conservative voters opting for the Conservatives has continued to decline since the beginning of the campaign. This potentially reflects negative perceptions of the Conservative campaign, such as D-Day, and allegations of insider gambling. It may also reflect a wider disillusionment with the current government, with the increasing number of 2019 Conservatives defecting to Reform UK. Reform UK’s projection as a party that challenges the entrenched systems potentially makes it a suitable protest vote for socially and economically conservative voters.
Since the announcement of Farage as leader of the party, Reform UK’s vote shares have increased dramatically. In turn, the 2019 Conservative to Conservative vote share has significantly decreased from Farage’s announcement, pointing to the fact that some of the loss in Conservative vote share comes from defection to Reform UK.
The proportion of Conservative 2019 voters switching to Labour has remained fairly stable throughout the election campaign. Although it is likely to make a significant dent in the Conservatives vote share. The fact 1 in 5 2019 Conservative voters are defecting straight from Conservative to Labour, along with defections to Reform UK, strongly suggest the Tories are fighting an electoral war on two fronts.
Contributors
Callum Hunter
Joe Alder
Julian Gallie
Rufus Vijayaratnam
Alma Talbot
STACKED REGRESSION & METHODOLOGY
What is SRP and how does SRP differ from MRP?
SRP brings AI and machine learning to MRP models. We are now all familiar with MRP – the models that break the electorate down into small demographic chunks, predict their voting behaviour and then ‘build’ a version of each seat in the country from the right mixture of demographics. The first part of that process – using polling data to predict how different types of voters will vote has, up until now, always used a simple multi-level regression model. But we have better models now from the world of AI and machine learning and SRP takes advantage of the best of them, stacking them all together and utilising the best bits of each.
How does SRP differ from MRP?
Stacked Regression and Poststratification (SRP) and Multilevel Regression and Poststratification (MRP) differ in the way they make predictions. The main difference is that the former uses more than one model, including but not limited to Multilevel Regression, to make predictions whilst the latter relies solely on the Multilevel Regression model. The second major difference is that SRP uses non-parametric Machine Learning Models as part of its architecture which offers advantages over parametric alternatives.
The use of stacking, combining estimates from many models into a single final estimate, underpins algorithms like Random Forests and Neural Networks. The main advantage of this kind of stacking is that we can use different models to probe different parts of the data giving a more holistic set of predictions that consider many different facets of the data. This includes models that are superior at unpicking the constituency level predictors whilst other models can investigate the individual level effects in much more detail. Some models, like Multilevel Regression, can analyse both the individual level and constituency level data simultaneously. The combination of these models then produces estimates that more accurately represent the nuances in the underlying data.
The majority of models used in our stacking procedure are “non-parametric”. One of the major inputs that goes into any MRP is the underlying structure of the model that link the predictors chosen by the modeller to the vote intention of individuals. This underlying structure is ultimately arbitrary and there are theoretically many trillions of possible underlying models that could be used. This is what makes MRP parametric – the modeller decides the interactions and relationship. Furthermore, MRP is inherently linear (unless otherwise specified) which can ignore more complex relationships in the data. A non-parametric model, like a random forest, can approximate the true underlying relationship that links the predictor variables to vote intention without input from the modeller. All the modeller does is select the variables used for prediction.
This offers obvious advantages. It reduces the number of assumptions that the modeller has to make in terms of selecting the parameterisation of the model whilst also allowing the model to find the best possible, arbitrary, combination of parameters.
The inclusion of multiple non-parametric models offers advantages to making seat estimates as it hugely reduces the effect of modeller-based decisions. It also offers improved accuracy in terms of fitting the underlying data and uses the most up-to-date methods for classification problems.
How we handle Don’t Knows and Turnout
Handling undecided voters as well as voter turnout are important parts of the modelling process. Ignoring either of the factors risks missing out on key data and trends that can affect the outcome of a model and hence careful methodology is required to account for these factors.
The central assumption in our methodology for imputing undecided voters’ eventual choice is that those who say they are undecided go through the same decision process as those that have decided. This assumption implies that the political decision-making process is dependent upon previous behaviour as well as current assessments of leadership qualities, policy positions and the environment in which voters live. This assumption is generally quite weak in the sense that it assumes as little about voters as possible. Using this assumption, we develop a Machine Learning Model that learns this decision process which can then be used to infer how undecided voters may vote. This allows us to include the all-important undecided voters in the final model.
Turnout runs along the same lines but uses a slightly different set of predictor variables. Since we ultimately have to post-stratify turnout in order to predict how likely each group is to turnout at the election we only use demographics, previous voting behaviour and constituency level information as predictors for turnout. Whilst this is fewer predictors than the undecided model this does not generally have an adverse effect on the performance of the model. This follows as we only take those who say they are very likely to vote as turning out – giving them a vote score of 1. We can then learn the trends that lead to this score of 1 and take the probability of scoring 1 as the implied turnout. Since we are only predicting a binary outcome (people either turnout or they don’t) the model can be simpler and still perform well.
Both of these models, and most of the models in our SRP, are non-parametric – thus offering the advantages that the SRP models give.
Unwinding the Proportional Swing
The models underpinning all election predictions are inherently linear, meaning they often assume that much of the swing is due to linear changes. For example, if the Conservatives lose go from 40% in 2019 to 20% in 2024, models assume 50% of the overall vote share is lost. Hence a seat that had the Conservatives on 60% in 2019 would only get 30% in 2024.
Whilst all elections have had some linear component in them (see research conducted here) the strength of correlation imputed by many election models is too strong. Hence many pollsters use so-called “unwinding” methods to reduce the strength of this proportionality. JLP uses a similar method based off the research we conducted looking at all general elections since 1966.
Using this historical data we have developed a method that allows some relaxing in this proportionality, shifting the distribution of seat estimates from a highly linear model to a more mixed model that has been the usual in the past.
Tactical Voting
We assume that any tactical voting is already accounted for in the data. Whilst tactical voting may affect our final results, we believe that the variables used in our models are able to pick up on areas where tactical voting is likely to happen. Another reason to avoid making assumptions about tactical voting is due to the fact that boundaries have changed and in many places the party that won last time is not the same as the notional result on new boundaries. Given this asymmetry of information we have chosen to make no post-hoc tactical voting adjustments.
Fieldwork Dates
The fieldwork for this model was carried out between 7th - 25th June. This covers the period during which Reform UK reached their zenith as well as the period since Nigel Farage announced his standing in the election.
Model Developers
This technique was developed by Joseph T. Orstein in this paper and was operationalised for the UK by Callum Hunter (Data Scientist).