What do the latest MRP polls tell us about the UK election?
A look at the recently published MRP polls that gave constituency level predictions for the July election. This post is best viewed using the browser or Substack app.
Introduction.
This article examines the results of four recent 'MRP' polls that provided constituency-level forecasts for the forthcoming UK general election. Since each polling firm employs distinct methodologies and gathers their own data, it is interesting to analyse the outcomes to discern common trends and explore any notable differences.
A reminder that polls can only tell us about the state of opinion when the polling data is collected – voters can, and often do, change their minds as the campaign progresses.
The article starts with a brief introduction to MRP models. This is followed by looking at the main differences in the predictions from the four MRP polls recently published. Finally, the MRP predicted outcomes are analysed to show the changes from the 2019 General Election and how they map at the constituency level.
What is an MRP model?
MRP stands for Multilevel Regression and Post-stratification. This statistical modelling approach involves pollsters combining extensive polling data with additional sources, such as census information, to estimate party support levels across different constituencies.
MRP is a relatively new technique that has been successfully used to predict the outcomes of the last two UK elections at the constituency level. Traditional polling methods often use a 'Uniform National Swing' (UNS) to forecast the number of seats each party will win in an election. However, this does not always reflect reality. For instance, in 2019, the Conservative vote share increased by 5% in constituencies that predominantly voted Leave in the EU referendum, but decreased by 3% in those that voted Remain. Regionally, Labour's vote share dropped by 13% in the North East, compared to a 6% drop in the South of England. Such local variations are not accounted for by UNS, leading to potential overestimations or underestimations of party performance in certain areas.
The MRP method tries to solve this problem by working out the relationship between peoples' voting intention and their individual demographic characteristics like age, income, educational background and a variety of other data sources, including past voting behaviour. Then, you can count how many of these different types of voters there are in each constituency, to make a prediction of how parties will perform on a seat-by-seat basis.
Overview of the latest MRP predictions
Four polling companies have published the results of their MRP models in the past two weeks. These are More in Common, YouGov, Electoral Calculus (with Find Out Now) and Survation (on behalf of Best for Britain) and their high level findings together with details of the poll are summarised in the following table.
It is important to note that the polling data was gathered before Nigel Farage assumed leadership of Reform and prior to Rishi Sunak's D-Day mishap. The table also highlights considerable variations in constituency predictions across each poll, which is likely due to the distinct survey collection methods and MRP methodologies used.
The table shows that More in Common has the lowest Labour majority but is based on the oldest survey data which was collected between Apr 9 to May 15. In addition, the survey used for the More in Common MRP had the lowest national vote share lead for Labour over the Conservative at 14%. Whereas the Survation MRP, which gave Labour the largest seat majority, had the highest vote share lead for Labour over the Conservatives at 27%. Finally, YouGov’s prediction of a 213 seat Labour majority is in the middle of the predictions and is close to how the current national vote share would likely translate into seats.
While the predicted range for seat totals varies widely, there is consensus among the models on a clear Labour majority. They also concur on the vast majority of individual constituency outcomes. The table below displays the predicted seat count by party, where all four polling companies agree on the result or at least three of the four do. In cases without a general agreement, the table indicates whether the prediction is split between two or three parties.
The table indicates that all four polling companies concur in predicting that Labour will safely secure 364 seats, while the Conservatives will safely secure only 25. When three of the MRPs are in agreement, they forecast that Labour will likely win 53 seats and the Conservatives will likely win 46 seats. Finally, there are 92 seats without a clear consensus among the MRPs, featuring a two-party split in 82 seats and a three-party split in the remaining 10.
What is predicted to change from the 2019 Election.
When examining electoral changes it is useful to understand the underlying details behind the changes. Sankey ‘flow’ diagrams based on polling data are a good way of showing the transition from the previous election to the predicted election outcomes.
The following Sankey diagram shows the predicted flow of seats by party starting from the 2019 general election. The bars on the left side show the number of seats each party won at the general election in 2019 adjusted for the boundary changes. The bars at the centre show the number of predicted seats that are both ‘Safe’ where all four MRPs agree and ‘Likely’ where three of the MRPs agree. The bar marked ‘None’ is where there was no overall consensus amongst the pollsters. The bars on the right side show how each party’s predicted seats split by ‘Safe’ and ‘Likely’.
The diagram clearly shows how much the Conservative vote is fragmenting and where they are going. Whilst most Conservative seats are moving to Labour a large number are also moving to ‘None’ where there is no consensus that they will be retained. In contrast nearly all Labour seats are predicted to be retained with only Bristol Central moving to ‘None’ where the polls are split evenly between predicting a Labour or Green Party win.
Most seats where there is no consensus are two way splits. Of the 82 seats with a two way split there are 59 which are split Conservative - Labour and 12 which are split Conservative - Liberal Democrat.
The diagram also shows the Scottish Nationalist vote is fragmenting with 23 seats predicted to move to Labour by all or most of the MRPs.
Mapping the MRP predictions at the constituency level
In the final section, the MRP predictions are presented at the constituency level. The following interactive hexagon map displays the predicted outcome for each constituency where all or three of the MRPs agree on the same party. If there is no consensus then the hexagon is coloured purple. Moving the cursor over a hexagon reveals more details about the constituency, including whether the prediction is unanimous or by a majority of the polling companies. In cases where there is no consensus then the hexagon show the parties which split the MRP predictions.
The map shows that Labour are predicted to dominate in constituencies in London and the North. Seats predicted to be Conservative are widely distributed predominately in rural areas. Interestingly, the seats where there is no consensus amongst the pollsters are also in rural areas with many in the South West and East of England.
In conclusion
MRPs serve as a valuable tool for forecasting election outcomes at the constituency level. Nevertheless, they can yield varying predictions owing to the different methods used.
Aggregating various MRP prediction results can enhance our understanding of their uncertainties by examining the points of agreement and difference.
Of particular interest there are 92 constituencies where there is no consensus on who will win. Having said that, all of the recent MRPs are projecting a large Labour majority.
Finally, a reminder that polls can only tell us about the state of opinion when the polling data is collected – voters can, and often do, change their minds as the campaign progresses.
As always, if you have any comments on this report then please post a message below.
Bring it on. And don't forget to vote tactically folks, where necessary. Beat FPTP at its own nineteenth century game.