Our latest UK general election 2024 MRP projects a 154 seat majority for Labour
New MRP figures released today from YouGov project that, if the country were voting in a general election tomorrow, Labour would win 403 seats nationwide. Crossing the 400-seat line is a significant milestone for Keir Starmer’s party in what is our second MRP projection this year.
Their Conservative rivals would win just 155 seats according to our model, down from 169 in our January projection, suggesting that the electoral situation is getting worse, rather than better, for Rishi Sunak and the Conservatives as we draw closer to the election.
The model is based on vote intention data collected and analysed by YouGov from 18,761 British adults interviewed from 7-27 March.
These latest results push Keir Starmer closer toward repeating a Blair-level result for Labour, a full 27 years since Labour’s longest-serving prime minister first took office. In that election, Blair won 418 out of the available 659 House of Commons seats.
By contrast, Rishi Sunak is now heading for a worse result than John Major’s 1997 total of 165 seats.
The coming tidal wave projected by this model would sweep away several major Conservative figures. The most prominent casualty could be chancellor Jeremy Hunt, who is currently fractionally behind the Lib Dems in his Godalming and Ash seat. Science secretary Michelle Donelan is also currently trailing the Lib Dems in her Melksham and Devizes seat, and Michael Gove is just one point ahead in his Surrey Heath seat.
And with recent weeks seeing rumours that Penny Mordaunt could issue a leadership challenge against Rishi Sunak, we find that the Commons leader is four points behind Labour in her Portsmouth North seat. Cabinet colleague Welsh secretary David TC Davies is also trailing Labour in our latest model, as are former leader Iain Duncan Smith and former Leader of the House of Commons, Jacob Rees-Mogg.
The Liberal Democrats are up by one seat based on our January model, to 49, continuing to set the path for a significant parliamentary comeback without any significant changes to their national vote share. The increasing efficiency of the Liberal Democrat vote, coupled with large drops in Conservative support across their Southern target seats, is a story we have seen replicated at successive local authority elections in recent years.
In Scotland YouGov now projects Labour to comfortably be the largest party in terms of seats won north of the border. Based on this data and this model, we would expect Labour to win 28 Scottish seats to the SNP’s 19. The Conservatives and the Liberal Democrats would win five each.
The Greens would continue to hold Brighton Pavilion, and are also still pushing Labour hard in the newly created Bristol Central seat (though we would expect Labour to hold that, based on this data).
Plaid Cymru are up by one versus our January model, to four seats total, with the Welsh nationalists now projected to be ahead in Caerfyrddin.
Despite their growing voting intention share in recent weeks, Reform UK are not projected to win any seats, although they do place second in 36. However, they are not close to winning any of them – in the seats where they have their highest vote share (Barnsley North, as well as Hartlepool) their 27% of the vote is fully twenty points behind that of projected winners Labour, and in no constituency in the country are they within ten points of the winning party.
The headline results based from this MRP model would be Labour on 41% of the vote, the Conservatives on 24%, the Liberal Democrats on 12%, the Greens on 7%, Reform UK on 12%, and others on 1%.
Vote intention derived from our MRP will look different to regular headline vote intention figures published by YouGov because the MRP model probabilistically matches 'missing voters' (i.e., those indicating they do not currently have a vote intention) to similar respondents who do express a vote preference, rather than excluding them from the estimations.
Methodological changes
There have been three important changes to the methodology of the YouGov British general election MRP for the release of this model. One concerns a brand new innovation for MRP models, while the other two concern updating of important data which is fed into the model.
The ‘unwinding’ algorithm
Firstly, recent commentary has highlighted well the tendency of MRP models to produce ‘proportional swing’ when estimating constituency-level shares. That is to say, MRP models tend to end up with an assumption that if a party is dropping by a certain percentage of their vote nationally then they will also be dropping by a similar amount proportionally within constituencies. The proportional model starts with the observation that if Party A is dropping by 10 points nationally from a 40 point starting position, then this constitutes a drop of 25% of their overall vote. Proportional swing would suggest that Party A vote shares will decline by 25% of their vote share from constituency to constituency. So, for example, if Party A was on 50% in a constituency at the previous election, the proportional swing model would have them on 37.5% in this election.
This goes against the conventional way in which swing tends to operate in British elections, in the sense that it tends to be ‘uniform’ across the country. That is to say, if a party is down by five points nationally, they will tend to drop by five points in each constituency (not by whatever a five-point drop implies as a proportion of their overall vote share).
Rather, though, than this being an assumption of the MRP model, it is a symptom of the fact that it is a regression-driven process. Regression models tend toward the mean, and so the eventual estimated shares will ‘bunch up’ according to correlations with other data fed into the model.
To address this, we have developed a new technique called ‘unwinding’. The unwinding algorithm looks at historical results and learns from them what the typical distribution of party vote shares tends to look like (for each party), and re-fits constituency-level shares in the posterior distribution to better reflect this variance. This has the effect of ‘unwinding’ the posterior distributions to better reflect the spread of constituency-level results at British general elections, and in turn reduces the proportionality of the swing.
That said, the drop in Conservative party support estimated by this model is still proportional. They are for instance dropping much more in places like South Devon (-24 points) than they are in Liverpool Riverside (-3 points). This is however a mathematical necessity – if a party is dropping by a large amount nationally (as the Conservatives currently are), there are constituencies (such as Liverpool Riverside) where their total vote share is smaller than their national decline. It would be impossible for the Conservatives to be on minus vote share in Liverpool Riverside. To make up for this, it must be the case that Conservative party shares are dropping faster where its shares are larger. What the unwinding algorithm does is create a more realistic picture of where, and the extent to which, this is happening.
Updated background data
This is the first MRP model published by YouGov since the BBC released the ‘Rallings and Thrasher’ 2019 notional results for the new House of Commons parliamentary boundaries. We have updated our model with this new data.
The British Election Study team have also updated their 2019 random probability post-election survey with validated turnout measures. We have adapted our turnout model to bring in this new data.
Technical notes
YouGov interviewed 18,761 British adults interviewed from 7-27 March. Constituency-level projections were estimated using the same statistical method which correctly predicted the 2017 and 2019 UK general elections – multi-level modelling and post-stratification (MRP). MRP constituency projection models first estimate the relationship between a wide variety of characteristics about prospective voters and their opinions – in this case, ‘which party would you vote for if a general election were being held tomorrow' – in a ‘multilevel model’. It then uses data at the constituency level to predict the outcomes of seats based on the concentration of various different types of voters who live there, according to what the multilevel model says about their probability of voting for various parties (‘post-stratification’). In this instance, 500 draws from the posterior distribution of the multilevel model were used to predict the constituency probabilities. The multilevel model was estimated through 20,000 iterations. Turnout likelihood for each voter group was estimated using a multilevel model on British Election Study data from 2017 and 2019. The precise multilevel model equation has been benchmarked to correctly estimate the 2019 General Election to within a couple of seats of each party’s actual performance in that election, and the overall approach to MRP by YouGov has been used to successfully predict elections as recently as Spain in July 2023.
What do you think about British politics, the forthcoming general election, and everything else? Have your say, join the YouGov panel, and get paid to share your thoughts. Sign up here.
Photo: Getty
