Nigel Marriott's Blog

An independent statistician using data to understand our world and to predict the future

You are here: Home / Elections / UK General Election 2024 #5 – My Final Forecast as of 3rd July 2024

UK General Election 2024 #5 – My Final Forecast as of 3rd July 2024

By

My third and final forecast of the number of seats to be won by each party in the 2024 UK General Election is –

  • CON 130 (+17)
  • LAB 420 (-17)
  • LD 50
  • SNP 25
  • GRN 3
  • PC 2
  • REF 2
  • OTH NI 18

The numbers in brackets are the changes from my second forecast of 30th June 2024.

This article is based on polling data as of 2000 on 3rd July 2024.  If more polling data comes in and causes me to change my forescript, the changes will be detailed in a postscript at the end of the article.

My Articles on the 2024 UK General Election
  1. The most accurate forecaster of the 2019 general election – I was independently assessed as the most accurate forecaster beating even Sir John Curtice’s exit poll.
  2. Keir Starmer’s Train to Downing Street – My assessment in 2021 of what Labour needed to do to win the next election.
  3. My election forecasting Track Record 2010 to 2024 – A list of all election forecasts I have made for General, European and Local elections, how I made them, how they turned out and what lessons I learned.
  4. How Accurate are Voting Intention Polls (Revised) – A recent article which explains why I now think the polls are accurate when before they would often underestimate the Conservative lead over Labour.
  5. Going Beyond the Swing in 2024 – A preliminary look at the 2024 general election at the start of the year.  I give probabilities for 10 specific outcomes.
  6. My 2024 UK General Election Forecasting Model – I use a top down forecasting approach based on the sum and difference in Conservative and Labour vote share to predict share of seats won.
  7. My 2024 UK General Election Forecast as of 23rd June 2024 – This explains how I use simulation to arrive at my estimate of CON 108, LAB 442, OTH 100.
  8. My 2024 UK General Election Forecast as of 30th June 2024 – This is an update which explains how I arrived at my estimate of CON 113 LAB 437 OTH 100

 

Data used in this article

All polling data prior to April 2024 comes from Mark Pack’s invaluable Pollbase.  For data since April, I am using the BBC poll tracker supplemented by the Wikipedia poll tracker.  For a summary of what the polls are saying as of 1600 3rd July 2024, see this X/Twitter thread.

All electoral data I display and use in this article comes from the House of Commons Research Library.  The PDF file I use most of the time is this one UK Election Trends 1918-2019.  I’ve created a spreadsheet GB General Election Data 1918-2019 – votes and seats which contains all the data used to build my models.  For clarity –

  • Seat Share is the percentage of seats in Great Britain won by each party.
  • National Vote Share is the percentage of all votes cast in Great Britain for each party.
  • Average Vote Share (per Seat) is the average of the vote share in each British constituency the party stands in.

I first explained the national and average vote share concepts in my article “How accurate are voting intention polls? – Revised“.

 

Reminder of my forecasting model for 2024

I am using a top down model to predict total number of seats won by the major parties based on the sum and difference in the Conservative and Labour party average vote share per seat.

Once I’ve estimated sum seat share CON+LAB and seat share difference CON–LAB, I will have two simultaneous equations which can be solved to estimate the percentage of the 632 seats in Great Britain that will be won by the Conservative and Labour parties.  I explain my model in full here.  For now, my model does not yet split seats won by other parties into the separate parties e.g. Lib Dems, SNP, etc.  I am working on this and will include this feature when ready.

One of the questions I have to resolve is which of the fitted lines in the above charts should I use.  In fact I will be using simulation where each line has a certain probability of being chosen.

For CON+LAB, the probabilities I’ve decided to use are –

  • Black line 1/13 – it’s been over 30 years since this was the fit hence the low probability
  • Green line 9/13 – the last 7 elections have used this fit hence why I say it is the most likely
  • Brown line 3/13 – given the high probability of a new low for CON+LAB vote share, I am not convinced the green line will hold hence why I’ve used a weight of 3 to 1 between the green and brown lines.
  • Overall the three lines have probabilities in a 1 to 3 to 9 ratio.  I like to use fixed ratios to avoid tweaking weights to get what I want!

For CON–LAB, the probabilities I’ve decided to use are –

  • Black line 1/4 – the last 3 elections have sat on this fit so it can’t be ruled out.
  • Green line 3/4 – the last 3 times where Labour kicked out a Conservative government with a big swing (1929, 1945, 1997) all sat on the green line.  Polls are pointing to something similar in 2024.
  • Again I am using a 1 to 3 ratio between the two lines.

The most likely outcome is that both models use their green fitted lines.  This is what I will use to illustrate my forecast.

 

Reminder of how accurate the polls are

I recently realised voting intention polls on average do predict the vote share difference CON–LAB provided one compares polls with average vote share per seat rather than national vote share.

When it comes to the sum vote share CON+LAB, polls are accurate on average over the whole timescale 1950 to 2019 but that hides notable differences over time.  The last four elections, coinciding with a rise in number of pollsters and more web polling, underestimated the total vote share of the two main parties by 3% on average.  The step change in 2010 is statistically significant.

 

What do the latest polls say?

For the week ending 3rd July 2024, using published data from 16 pollsters as of 2000 hours, Labour has a 17.8% lead over the Conservatives.  This would be a new record for Labour if this occurs on 4th July.

At the same time, the combined CON+LAB vote share of 60.6% would be the lowest on record since the CON/LAB duopoly began in 1922, beating the 66.6% in 2010.

Reform are in 3rd place at 16% and are holding steady.  For reference, UKIP’s national vote share in 2015 was 12.9%.

The Lib Dems are in 4th place on 11% and are holding steady.  For reference, the Lib Dems national vote share in 2019 was 11.8%.

The Greens are in 5th place on 6.5%.  This would easily be their best ever general election beating their previous highest national vote share of 3.8% in 2015.

 

Will the polls get it wrong in 2024?

First we need to define what we mean by “wrong“.

I have usually used a difference between actual and polled vote share of 4 percentage points as the definition of a “major error“.  That is because errors on this scale can cause 30 seats or more to change hands, often changing the narrative of the election.  Conversely, any error less than 2 percentage points is not an error in my view.  So I arrive at the following definition of “wrong“.

  • No Error – Errors less than 2 percentage points
  • Minor Error – Errors between 2 & 4 percentage points
  • Major Error – Errors greater than 4 percentage points

For the Conservative lead over Labour (CON–LAB vote share) statistic as estimated by the polls and compared to actual CON–LAB using average vote share per seat, there have been 5 major errors in the 20 elections since 1950.  They are 1950, 1983 & 2017 where Labour did better than expected and 1992 & 2015 when the Conservatives did better than expected.  Overall, 25% of elections have experienced a major error for the Conservative lead over Labour.

For the 2024 general election, I put the probability of a major error for the Conservative lead over Labour at greater than 50%.   There are two reasons why I put it this high.

First is looking back at elections with a similar narrative to 2024.  Throughout this election, the narrative has been “How large will Labour’s majority be?”  The very first election I remember following was 1983 and that had an identical narrative except this time it was “How large will the Tories majority be?”  I would identify 1966, 1997, 2001 and 2017 as similar elections.  Those 5 elections are highlighted in the graphic here.

In all 5 elections, the party that was expected to get hammered did better than the polls.  The final lead for the party expected to get a large majority ended up on average being 3 percentage points smaller than expected.  Currently Labour have a 18% lead in the polls over the Conservatives but if my analysis of history is right, it may end up at 15% instead.

The second reason is the pollsters can be divided into two camps based on whether they attempt to squeeze or model those saying don’t know to the pollsters.  A  squeeze question is something like “Suppose you were required to vote by law (as in Australia), which party would you vote for?”  Based on this list from X/Twitter, it appears half of the pollsters are Squeezers and half just ignore those saying Don’t Know.

On average the Squeezers have CON–LAB leads 1% smaller than non-Squeezers.  That combined with my first observation from elections with similar narratives to 2024 is why I think a major error on the CON–LAB lead is more likely than not.

 

How do I use simulation to make my forecast?

My official forecast is an average of 10,000 simulations of possible scenarios.  The process works like this.

My first simulation assumes the polls are correct and that both the sum seat share and seat share difference models use the green fit lines.  I call this the Illustrative Example and the calculation works like this.

  • The estimated sum seat share CON+LAB is 87.6% using the green fit line i.e. Other parties will win 12.4% of the 632 seats in Great Britain.
  • The estimated seat share difference CON–LAB is -48.0% using the green fit line.
  • The estimated CON seat share is 19.8% = (sum+diff)/2 = ( 87.6% + -48% )/2
  • The estimated LAB seat share is 67.8% = (sum-diff)/2 = ( 87.6% – -48% )/2

With 632 seats in all of Great Britain, the illustrative example results in 125 CON, 429 LAB and 78 OTH seats.

Thereafter, the other 9,999 simulations are processed as follows –

  1. A random number is added to the CON+LAB vote share sum indicated by the polls.  Given the notable step change in 2010, I am currently drawing a random number from a normal distribution with mean +2.0% and standard deviation 1.0%.  I am keeping this element under review.
  2. A random number between 0 and 1 is created and this decides whether the fitted model for sum seat share CON+LAB uses the black (8%), green (69%) or brown(23%) fitted lines.
  3. The chosen fitted line is used to estimate the sum seat share CON+LAB.
  4. To the estimated sum seat share CON+LAB, a random number drawn from a normal distribution with mean 0% and standard deviation 0.5% is added to allow for residual error.
  5. A random number is added to the CON–LAB vote share difference indicated by the polls.  This is more straightforward as it is a random draw from a normal distribution with mean 0.0% and standard deviation of 2.5%.
    • Important – this step does not include the expected major error I refer to above.  If I were, I would change the mean to +2% or +3%.
  6. A random number between 0 and 1 is created and this decides whether the fitted model for seat share difference CON–LAB uses the black (25%) or green (75%) fitted lines.
  7. The chosen fitted line is used to estimate the seat share difference CON–LAB.
  8. To the estimated seat share difference CON–LAB, a random number drawn from a normal distribution with mean 0% and standard deviation 2.5% is added to allow for residual error.

On completion of all simulations, the average of the estimated sum seat share CON+LAB after step 4 and the estimated seat share difference CON–LAB after step 8 are calculated.  These are then solved to get the estimated number of seats to be won by the Conservatives, Labour and Others.  The 10,000 simulations provide data on possible margins of error and are summarised in the table below.  An explanation of the table is given in this X/Twitter thread as well.

 

How do I forecast the smaller parties?

The table at the end of the previous section summarises my simulations and only gives estimates for the Conservatives, Labour and Others across Great Britain.  In my article explaining my model in detail, I stated I would run a similar forecasting exercise for England only for the following reasons –

  1. The expected number of seats for Other parties in England will be mostly Liberal Democrat wins.
  2. The difference between expected number of seats for Others in Great Britain and in England will capture mostly SNP and Plaid Cymru wins.

Any further splits into wins for Greens, Reform and Liberal Democrats in Scotland & Wales will be a finger in the air job by me based mostly on what I have read from other election analysts on X/Twitter.  I have not had the time to do detailed seat modelling for this election so as far as I am concerned, I should be judged on how many seats are won by the Conservatives and Labour.

The forecasting process is exactly the same as for Great Britain, the only differences are the polls are based on England only crossbreaks of the national polls and the models for CON+LAB and CON–LAB are the ones shown in the chart below.

After running 10,000 simulations for England, the outcome was the forecast 82 seats for Others in Great Britain split into 53 for England and 29 for Scotland and Wales.

Now comes the finger in the air bit!  I’ve decided to go with the following round numbers –

  • LD – 50 seats
  • SNP – 25 seats
  • PC – 3 seats
  • GRN – 2 seats
  • REF – 3 seats

It’s worth making the following points about the Liberal Democrats because I see a lot of excitement on social media.

  • Their national vote share in 2019 was 11.8% & latest polls suggest they will drop to 11% in 2024.
  • In 1997, the LD national vote share fell to 17.2% from 18.3% in 1992 but the number of seats they won rose from 20 to 46.
  • For 2024, I am predicting they will go from 11 to 50 seats but starting at a lower base and falling back a bit in terms of votes.
  • The main reason is the Conservatives are expected to do worse than 1997.

A final point to note is anti-Conservative tactical voting.  This was a recognised factor in 1997 and some people think we will see the same effect in 2024.  The problem with that argument is the Lib Dem vote of 2019 already includes a lot of tactical voting built up over 2017 & 2019 due to remain voters seeking to frustrate Brexit.  Therefore I ask how much more tactical voting can we expect in 2024?

Putting all of this together, I would be surprised if the Lib Dems got more than 50 seats.  I seriously thought about reducing the forecast to 45 seats but I wasn’t sure where to put the other 5 seats.  That is why I have stuck with 50.

 

My Official Forecast for the 2024 UK General Election

As of 3rd July 2024, my simulations gave me my final official forecast as follows –

  • CON 130 seats (113 to 150)
  • LAB 420 seats (436 to 408)
  • LD 50
  • SNP 25
  • GRN 3
  • PC 2
  • REF 2
  • OTH NI 18

The ranges in brackets represent the interquartile range of my 10,000 scenarios i.e. from the 25th to the 75th percentile.  Note the error in CON and LAB are inversely correlated hence why Labour’s range is from high to low i.e. if the Conservatives do end up with 113 seats, Labour are more likely to get around 436 seats.

Note this forecast does not include a major polling error in the Conservative lead over Labour as discussed earlier.  If that were to happen, the Conservatives would be 10 seats higher and Labour 10 seats lower.

Compared to my forecast of 30th June, there has been a shift of 17 expected seats from Labour to the Conservatives.  This is because the polls show Labour’s lead over the Conservatives has narrowed.

 

Postscript –

Should there be any significant updates to the polls, I will state my revised forecast here.

 

— Would you like to comment on this article? —-

Please do leave your comments on this X/Twitter thread.

— Subscribe to my newsletter to receive more articles like this one! —-

If you would like to receive notifications from me of news, articles and offers relating to Elections & Polling, please click here to go to my Newsletter Subscription page and tick the Elections and/or Surveys category and other categories that may be of interest to you.  You will be able to unsubscribe at anytime.

More articles on elections can be found by clicking on the Elections tab at the top of your screen.