Ipsos MRP

Today Ipsos publishes its first MRP model of the 2024 General Election. Ipsos’ MRP model is unique as it uses data collection from our online random probability Ipsos KnowledgePanel, with participants recruited via random unclustered address-based sampling, the gold standard in UK survey research. It is also the first such model to be produced with data collection taken entirely after Nigel Farage announced he was returning to lead Reform UK.

KEY FINDINGS | LABOUR | CONSERVATIVES | LIBERAL DEMOCRATS | SNP | REFORM UK | GREENS | SEATS TOO CLOSE TO CALL | KEY FINDINGS FOR SCOTLAND | INTERPRETING THE DATA

Key findings 

This Ipsos MRP projection uses a large-scale online survey of nearly 20,000 participants on the Ipsos KnowledgePanel, and population data at a constituency level, to project which party will win individual seats at the upcoming General Election. The estimated seat counts, with upper and lower range estimates, are as follows:

• Labour 453 (with a range of 439 to 462 seats)
• Conservatives 115 (with a range of 99 to 123 seats)
• Liberal Democrats 38 (with a range of 35 to 48 seats)
• SNP 15 (with a range of 13 to 23 seats)
• Plaid Cymru 4 (with a range of 2 to 5 seats)
• Reform UK 3 (with range of 3 to 10 seats)
• Green Party 3 (with a range of 0 to 4 seats)

The implied vote shares from this model are Labour 43%, Conservative 25%, Reform UK 12%, Liberal Democrats 10%, Greens 6%, SNP 3%, Plaid Cymru 1% and Other 1%.

117 seats have a winning margin of less than 5 points, and are considered too close to call.  The Conservatives are second in 50 of these, Labour in 43, and the Liberal Democrats 13, demonstrating that small changes in the parties’ performance could still lead to big changes in the final outcome.  

Notable seat projections: Our model suggests Nigel Farage winning in Clacton but Jeremy Corbyn losing in Islington North. High profile Conservatives who are at risk of losing their seats include Grant Shapps, Penny Mordaunt, Gillian Keegan, Johnny Mercer, and Jacob Rees-Mogg. Jeremy Hunt faces a tight race, while James Cleverly and Kemi Badenoch hold on.

Findings in detail 

Labour: 453 seats (range 439 to 462)

Mid-way through the campaign, Labour are likely to win a substantial majority, reaching the 326 seats mark even before considering seats that are too close to call.

Labour’s vote is increasing across the country, especially in Scotland and the North East, with more modest advances or even some drops in highly urban areas where they have been traditionally strong, like London, Liverpool, Manchester, and Birmingham. While Labour are also only picking up a few more votes in Wales, they could still gain 8 or 9 seats due to the decline in the Conservative vote.

Conservatives: 115 seats (range 99 to 123)

The Conservatives fall to a record low of 115 seats on these estimates.  Rishi Sunak’s party is losing votes across the country, with particularly severe drops in the East and South of England,  and across the Midlands.  As was seen in the local elections, the raw survey data, even before modelling, suggests the Conservatives are losing vote share in the areas where they were strongest in 2019.

Liberal Democrats: 38 seats (range 35 to 48)

The Liberal Democrats are likely to hold on to their 8 seats from 2019 as well as gain at least 20 seats (with another seven too close to call) from the Conservatives, mostly in the South East and South West of England.  This is despite their national vote share being marginally down from 2019, most notably in London.

SNP: 15 seats (range 13 to 23)

In Scotland, the fate of the SNP is still very much up in the air, with the implied vote shares in Scotland sitting at Labour 36%, SNP 33%, Conservatives 13%, Liberal Democrats 8%, Reform UK 5%, and Greens 3%. 

Our model shows the SNP expected to win around 15 seats. They are currently at risk of losing 29 of their 2019 notional seats to Labour and 3 to the Liberal Democrats, while 8 are too close to call.  In total, 12 out of the 57 seats in Scotland are too close to call.

Note that Scotland results should be interpreted with particular care – Census results have not yet been published at the constituency level, meaning we have had to rely on 2011 Census data and make adjustments where other data was available.  This means the model is working with less accurate population data at the constituency level and the risk of error is higher.

Reform UK: 3 seats (range 3 to 10)

Reform UK are currently estimated to win relatively comfortably both Ashfield and Clacton, where Nigel Farage could overturn a massive Conservative majority. North West Leicestershire also leans Reform (albeit with only a small lead) due to a collapse in the Conservative vote, leaving Labour in a possible second place. Overall, Reform UK gets its highest share of the vote in the North East, East Midlands, and East of England.

As the Brexit Party did not stand candidates in Conservative seats at the 2019 election, modelling their votes across constituencies is more difficult than for other parties where we have more information about voters’ previous behaviour. With 3 weeks to go from when our data collection ended, things could still change, especially in hotly contested seats. 

Still, Nigel Farage’s party is currently coming second in 30 constituencies, up from only 2 seats for the Brexit Party in the 2019 notional results. While most of these are relatively safe or likely Labour wins, 4 are toss ups where Reform UK (as well as other parties) are challenging the Conservatives.

Greens: 3 seats (range 0 to 4)

The Greens are making headway in Bristol Central and North Herefordshire, both on their list of targets seats, where they could overtake Labour and the Conservatives respectively, while they are also part of the close race in Waveney. Yet they are struggling to hold on to Brighton Pavillion, where Labour currently lead.  Overall, as the Greens stand more candidates than before, they are looking to improve their vote share by around 3 points relatively consistently across most regions, and especially so in safer Labour seats in some city and university constituencies.

Explaining the seats that are ‘too close to call’ 

Yet the final seat outcome at this stage of the campaign is not guaranteed – these figures are an estimate of people’s current voting intentions, not a prediction of what will happen at the election itself. At the time of fieldwork, the deadline for candidate submissions had only just passed, meaning the campaign is still likely to intensify in some areas. Additionally, as with all MRP models, ours attempts to model constituency-level outcomes from a national survey, meaning there are a number of uncertainties to take into account (more detail on interpreting MRP results and the potential limitations is given in the technical note, and include sensitivities within the model to small changes in vote shares and assumptions, wider margins of error for individual constituency estimates, and the potential to produce a high level of proportionate swing).   Taking into account unique local dynamics is also difficult. In 20 constituencies with high profile candidates outside the main parties, we prompted candidate names (where these were known) in order to capture some of the local context. We still urge caution when interpreting constituency-level outcomes.

Our model identifies 117 seats that are currently too close to call, with a lead of less than 5 ppts.  All of these are currently held by the Conservatives or SNP according to the 2019 notional results. Most are seats where the notional results show the Conservatives comfortably winning in 2019, and where Labour, even where they are achieving large swings, are not in a position to guarantee a victory. Another 19 are Conservative – Liberal Democrat contests that could also go either way.

With so many seats still very close, it means if the Conservatives improve their vote share by even just a couple of points it could help them pick up more seats. They are currently just behind in second place in 50 of these toss up seats. But equally, they are only just ahead in 56 of them, which means the result could turn out worse.

Commenting on the findings, Kelly Beaver MBE, Chief Executive, Ipsos UK and Ireland, said:

"Ipsos’ new MRP, based on data collected from our online random probability KnowledgePanel, supports the trends that suggest Labour is on course to win the 2024 election with a very healthy majority, while the Conservatives are facing the potential for record losses.  Labour is increasing its 2019 vote share across the country, especially in Scotland and the North East, while the Conservatives are losing votes in all regions – especially in the East and South of England, and across the Midlands.  What is perhaps most concerning for them are signs in the data that they are particularly losing vote share in the areas where they were strongest in 2019.

There is also movement for the other parties, with the Conservatives facing challenges from both sides. The Liberal Democrats are estimated to pick up seats from them in South East and South West, and Reform are estimated to win their first parliamentary seats too - including Nigel Farage in Clacton and Lee Anderson in Ashfield.  Meanwhile, the Greens are making headway in some of their key target seats, while in Scotland the SNP is likely to lose many of their seats to Labour – making Keir Starmer’s task to achieve a majority that much easier.

Nevertheless, this is just a snapshot of people’s current voting intentions, and there is still time for things to change.  As with any survey and any model, there are uncertainties to take into account, such as margins of error, the impact of unique local dynamics, and sensitivities to the data that goes in.  There are also 117 seats that are currently too close to call, all of which means that small changes to the parties’ performance could still make significant changes to the number of seats they get.

But this data, in line with most of the evidence that we have seen both in the run-up to this election and since the campaign started, in terms of the mood of the nation and real election results in local elections and byelections, suggests that the British political scene could be heading for yet another significant shift.”

Key findings for Scotland

In Scotland the estimated seat counts, with upper and lower range estimates, are as follows:

• Labour 34 (with a range of 28 to 36 seats)
• SNP 15 (with a range of 13 to 23 seats) 
• Conservatives 3 (with a range of 0 to 4 seats)
• Lib Dems 5 (with a range of 4 to 5 seats)

The implied vote shares from this model are: Labour 36%, SNP 33%, Conservative 13%, Liberal Democrats 8%, Reform UK 5%, Greens 3%, Other 1%.

While the SNP are set to face significant losses at this General Election, just how heavy those losses will be is uncertain at this point. Our model suggests the party are currently at risk of losing 29 of their 2019 notional seats to Labour and 3 to the Liberal Democrats, while 8 are too close to call.  

12 of Scotland’s 57 seats have a winning margin of less than 5 points, and are considered too close to call. The SNP are second in 7 of these, Labour in 4 and the Conservatives in 1, demonstrating that small changes in the parties’ performance could still lead to big changes in the final outcome.

The model suggests Labour taking seats from the SNP across the central belt, including all six seats in Glasgow. It indicates Douglas Ross losing Aberdeenshire North and Moray East and the Conservatives potentially holding on in just three seats - Gordon and Buchan, Berwickshire, Roxburgh and Selkirk, and Dumfriesshire, Clydesdale and Tweeddale. However, all three contests are too close to call between the Conservatives and the SNP in our model, with a winning margin of less than 2 points. The Conservative-SNP contests in the North East and South of Scotland will be ones to watch closely on election night.

Commenting on the findings, Emily Gray, Managing Director of Ipsos in Scotland, said:

“We’re excited to publish Ipsos’ new MRP model, uniquely based on data collected from our online random probability KnowledgePanel. These results underline that there’s still a huge amount for Scotland’s political parties to play for in the remaining two weeks before election day. In Ipsos’ model one in five Scottish seats are too close to call, with a margin of less than 5 points, so even small shifts in vote share could make big differences to the number of seats each party wins. Our model suggests Labour are on course to win more seats in Scotland than any other party, but the scale of their gains over the SNP is still uncertain.  For the SNP, even small improvements in their share of the vote could reduce their likely losses on election night. Meanwhile for the Conservatives, small changes in vote share could make the difference between winning 4 seats or electoral wipeout in Scotland. Remember that this is just a snapshot of people’s current vote intentions, and there is still time for things to change.”

Technical note and how to interpret MRP Data

Multi-level regression and post stratification (MRP) is an advanced modelling technique that estimates the likely vote share for main parties in each constituency by using a national survey with a large sample size. Data on how people say they will vote (and if they will vote) is analysed by a wide range of factors to see how different types of people, in different areas, are likely to act. For example, it estimates the probability that a woman, aged 25-34, with a degree, living in a Lab/Con marginal, who voted Labour in 2019 will vote for each party running in that constituency. These estimates are then applied to the differing profiles of each constituency to estimate vote counts for each party. 

While a lot has been written about MRP and how they have been used to provide seat estimates for elections in the UK and elsewhere, we would like to point out some caveats.

The quality of the estimates will depend on:

• The quality of the survey data collected
• The quality of the population data used in the post-stratification frame
• The overall model, including which variables are or are not included

The survey data Ipsos has used for this model was collected via the Ipsos UK KnowledgePanel. Panellists are recruited using random probability unclustered address-based sampling, the gold-standard in UK survey research, meaning that every household in the UK has a known chance of being selected to join the panel. Crucially, members of the public who are digitally excluded are given a tablet and provided with an email address. Ipsos interviewed 19,689 adults aged 18+, residing in Great Britain. Data collection took place between 7-12 June 2024, using the standard Ipsos voting intention and likelihood to vote question wording. This will be the first UK election when any voting intention data collected via a probability panel has been published, and this MRP is the first model of this type using probability data. 

Very importantly, the MRP estimates individual seat outcomes based on respondents’ current voting intention. The estimates should not be taken as a prediction of what will happen at the general election. With weeks still to go, voter attitudes and voting intentions could still change.  The MRP results are also sensitive to the exact vote shares that go into the model – relatively small changes in these can make a significant impact on the final seat estimates. 

Two models underpin the estimates: the first estimates turnout rates, and the second estimates party choice. They include various factors about each panellist – such as their age and gender – and the constituency in which they live.  MRP models are very sensitive to the assumptions that go into each model, and changing the assumptions can have a significant impact on the final seat estimates.  For example, past vote is a strong predictor of future vote and we include it in our party choice model. However, we know that recall of past vote can be inaccurate due to false recall. We account for this by adjusting the overall past vote targets so that they match what we find in the survey data. Our models rely on constituency-based information on previous vote. We have used the notional 2019 results on the revised boundaries produced by Colin Rallings and Michael Thrasher.

As the modelling makes use of a national survey, caution should be taken when looking at individual seats. While MRP is good at taking into account the different demographic profiles of each constituency, with relatively few respondents per constituency, it is unlikely to be able to capture the full local context, especially where there are unique political dynamics. In order to aid this, we have prompted with all candidate names, where known, in 20 constituencies, mainly where there are high profile candidates outside the main traditional parties.  These were Aberdeenshire North and Moray East, Alloa and Grangemouth, Ashfield, Boston and Skegness, Brighton Pavilion, Bristol Central, Chingford and Woodford Green, Clacton, Clwyd East, Cowdenbeath and Kirkcaldy, Great Yarmouth, Islington North, Leicester East, Lothian East, Na h-Eileanan an Iar, North Herefordshire, North West Leicestershire, Rochdale, Solihull West and Shirley, Waveney Valley.

All constituency vote shares are accompanied by confidence intervals which we have published, which are wider than a normal survey. We have called individual seats as strong or likely wins, leaning, or toss ups (broadly based on size of leads, with some adjustment for incumbency at the margins) and would encourage readers to not place too much certainty into specific point estimates.  

The 2021 Census in England and Wales provides robust, and recent, population statistics. Unfortunately, the equivalent data is not yet available for Scotland at a constituency level. We have relied on a mix of 2011 Census data and trends in population statistics to impute estimates for constituencies in Scotland. This means any interpretation of outcomes in Scotland should be interpreted with particular care.

A note on proportional swing

As many have commented, MRP models published recently have tended to show a proportional swing, such as the Conservative party losing more votes in seats where they currently enjoy larger leads. Historically, elections in Britain have tended to follow a pattern closer to a Uniform National Swing (although not exactly).  This MRP model, like several others, does broadly follow a proportionate pattern, which means it does produce a larger Labour majority than a simple UNS model.  This is likely to be partly a function of the model itself bringing some proportionality into the vote estimates, which is another reason to be cautious when interpreting the results, however it should be pointed out that we also observe a proportional swing in the underlying survey data (which reflects, for example, the pattern seen in the local election results).

Data for all constituencies can be found here.