Pollsters Electoral Calculus and Find Out Now have conducted a new large-scale MRP poll for the Daily Mirror. Polling using MRP, which stands for Multi-level Regression and Poststratification, is a relatively recent innovation in polling science which have been used to successfully predict the last two British general elections, and gives the estimated result in each (new) Westminster constituency.
Fieldwork was conducted from 24 January to 12 February 2024, with a sample size of over 18,000 people.
The headline voting intention is:
The headline number of seats predicted to be won by each party is:
The table below gives the prediction in numerical form:
Party | GE 2019 | Predicted Election Result | ||
Votes | Seats | Votes | Seats | |
CON | 45% | 376 | 22% | 80 |
LAB | 33% | 197 | 42% | 452 |
LIB | 12% | 8 | 11% | 53 |
Reform | 2% | 0 | 10% | 0 |
Green | 3% | 1 | 7% | 2 |
SNP | 4% | 48 | 4% | 40 |
Plaid | 0.5% | 2 | 1% | 4 |
OTH | 1% | 0 | 2% | 1 |
Our prediction is for Labour landslide with a Labour majority of 254 seats. The Conservatives are set to lose most of their seats, and could have below 100 seats after the election.
This prediction has been cross-checked with alternative (non-MRP) models using the same polling data and we have confidence in it (see 3.3 below).
Our figures indicate a substantial Labour landslide, with Keir Starmer gaining a majority of over 250 seats at Westminster. The Conservatives would have fewer than 100 seats. They would be the official opposition, but they would have less than half of the opposition MPs (80 out of 198).
The forecast Labour victory would be larger than Tony Blair's landslide in 1997 (419 seats) and be the largest win by any party in modern parliamentary history, except for 1931. The 80 seats for the Conservatives would be their worst result since at least 1900.
The Liberal Democrats tend to do better when the Conservatives do badly, and are predicted to win 53 seats. This parliamentary comeback would return them to similar seat totals that they enjoyed from 1997 to 2015.
Nineteen Conservative Cabinet Members are likely to lose their seats.
Conservative cabinet members who are likely to keep their seats and might contest the next Conservative leadership contest:
The poll also asked voters for their top three political priorities, and also for their top three policies.
For priorities, the table shows how many people chose each issue as one of their top three political priorities.
Issue | All voters | CON voters | LAB voters |
---|---|---|---|
Health | 46% | 39% | 61% |
Economy | 39% | 39% | 41% |
Immigration and asylum | 35% | 60% | 14% |
Environment and climate | 23% | 13% | 37% |
Inflation | 26% | 23% | 29% |
Housing | 24% | 15% | 34% |
Crime | 19% | 27% | 11% |
Defence and foreign affairs | 12% | 20% | 8% |
Education | 11% | 7% | 17% |
Tax | 10% | 11% | 9% |
Pensions | 8% | 11% | 5% |
Welfare benefits | 9% | 7% | 10% |
The top issue overall is health, followed by the economy and then immigration and asylum.
For those who voted Labour in 2019, the top issues are health, the economy and the environment.
For policies, the next table shows how many people chose each policy as one of their top three policies.
Policy | All Voters | CON voters | LAB voters |
---|---|---|---|
Nationalise utility companies | 44% | 35% | 65% |
Build more houses | 33% | 27% | 46% |
Send illegal migrants to Rwanda | 28% | 53% | 7% |
Legalise euthanasia | 23% | 22% | 22% |
Increase public spending and raise taxes | 18% | 11% | 33% |
Cut taxes and public spending | 23% | 26% | 18% |
Elect the House of Lords | 16% | 17% | 22% |
Bring back the death penalty | 18% | 28% | 7% |
Reduce regulation on businesses | 9% | 14% | 5% |
Restrict car use | 6% | 3% | 5% |
Don't know | 10% | 6% | 6% |
None of these | 4% | 3% | 3% |
The most popular policy is to nationalise the utility companies, followed by house building and then the government's Rwanda policy. Three times as many people wanted to bring back the death penalty (18%) as wanted more restrictions on car use (6%).
Top three policies for those who voted Labour in 2019 are utility nationalisation, house building and increasing public spending through higher taxes.
In mid-January, YouGov saw the following national vote shares: Con 26%, Lab 40%, Lib 12%, Reform 9%, Green 8%, which implies a Labour lead of only 14%. YouGov estimated that Labour would win 386 seats and the Conservatives would win 169 seats.
Our MRP poll has a Labour lead of 20% which is more consistent with other recent published national polling, which also shows a Labour lead of 20% (average of last six national public polls). YouGov's own classic polling from 10-11 January showed a Labour lead of 23%.
The difference in seat totals is mostly due to the difference in estimated vote share. Each 1% increase in Labour's poll lead costs the Conservatives about nine more seats.
Because YouGov estimated a relatively low Labour lead, they predicted the Conservatives would win many more seats.
Our larger estimate of the Labour lead, which we think reflects current voter intention, naturally converts into fewer Conservative seats.
Distinguished pollster Peter Keller wrote an article last year detailing how MRP has a tendency to model parties losing more votes in seats where they are the strongest, and lose fewer votes in seats in which they are weakest.
One of his main points is the distinction, for a declining party such as the Conservatives, between additive and multiplicative models. In an additive model, the Conservative vote share would decline by a constant additive amount in each seat, and in a multiplicative model, the decline would be by a constant multiplicative fraction. For example, the Conservatives decline from 45% to 22% could be modelled either by subtracting 23% from the Conservative vote share in each seat, or by multiplying it by 0.49 (which is 22%/45%).
Kellner points out that historically, party decline has been additive, but the MRP model is essentially multiplicative. The multiplicative model would predict greater seat losses, particularly in strong Tory seats, which might be incorrect.
To investigate this, we looked at our poll in more detail. We divided the country up into eight groups of seats, according to the strength of the Conservative vote share in the 2019 election from the 75 strongest Conservative seats (Group 1 seats), followed by the next 75 strongest (Group 2 seats), and so down to the weakest Conservative seats (Group 8 seats).
GE 2019 | VI from Poll | Con change | ||||
---|---|---|---|---|---|---|
CON | LAB | CON | LAB | Additive | Multiplicative | |
Group 1 | 45% | 13% | 21% | 19% | -23% | -53% |
Group 2 | 42% | 15% | 17% | 22% | -25% | -59% |
Group 3 | 40% | 15% | 19% | 24% | -21% | -53% |
Group 4 | 36% | 20% | 16% | 27% | -20% | -55% |
Group 5 | 32% | 22% | 16% | 27% | -16% | -50% |
Group 6 | 25% | 26% | 13% | 31% | -12% | -48% |
Group 7 | 19% | 29% | 9% | 28% | -10% | -52% |
Group 8 | 11% | 33% | 5% | 37% | -6% | -54% |
For each group of seats, we measured how respondents in that group were planning to vote, using standard quota-weighted polling. The MRP model was not used to make these figures.
We see that the additive change in Conservative vote share varies strongly by group, but the multiplicative change is relative constant at around -53%. This suggests that the multiplicative model is a fair model of current voter behaviour, though in strong Conservative seats the additive model works well too.
We used an alternative non-MRP model to double-check the MRP results. We took a simple uniform national swing (UNS) model, but applied it separately to each of the five Conservative seat groups described above. We also used two styles of the UNS model - one additive, and one multiplicative.
The additive model applies, in each seat, a constant swing to each party equal to the national swing. It also ensures that no party's vote share in a single seat can be negative.
The multiplicative model scales the vote shares of each party according to the ratio of its current vote share divided by its vote share in the previous election. This is based on a Markovian model with a constant transition matrix.
Party | MRP | UNS Additive | UNS Multiplicative |
---|---|---|---|
CON | 80 | 79 | 75 |
LAB | 452 | 455 | 472 |
LIB | 53 | 44 | 35 |
Reform | 0 | 0 | 0 |
Green | 2 | 1 | 1 |
SNP | 40 | 47 | 47 |
Plaid | 4 | 3 | 2 |
OTH | 1 | 3 | 0 |
The table shows that there is very little variation between the MRP model and each of the UNS models.
This gives reassurances on two points. Firstly, it shows, once seats are grouped by Conservative strength, then the difference between the additive and multiplicative approaches mostly disappears. Secondly, it shows the MRP seat prediction is close to the UNS seat prediction, which provides a double-check on the MRP results.
Finally, we can look at the MRP predictions for vote shares in the eight seat groups. These are:
VI from Poll | MRP | Difference | ||||
---|---|---|---|---|---|---|
CON | LAB | CON | LAB | CON | LAB | |
Group 1 | 21% | 19% | 21% | 19% | 0% | -1% |
Group 2 | 17% | 22% | 18% | 21% | 1% | -1% |
Group 3 | 19% | 24% | 18% | 23% | -1% | -1% |
Group 4 | 16% | 27% | 15% | 27% | -1% | 0% |
Group 5 | 16% | 27% | 16% | 26% | 0% | -1% |
Group 6 | 13% | 31% | 12% | 29% | -1% | -2% |
Group 7 | 9% | 28% | 9% | 29% | -1% | 1% |
Group 8 | 5% | 37% | 5% | 33% | -1% | -3% |
This shows a good correspondence between the standard quota-weighted polling analysis and MRP for the sub-samples represented by these eight seat groups.
That is not entirely surprising, since the seat group of each respondent was a regression variate for the MRP analysis, but provides additional reassurance nonetheless.
Ahead of the 2019 general election, we were the most accurate pre-poll predictors.
Martin Baxter, founder of Electoral Calculus, said "The public seem even more disenchanted with the Conservatives under Rishi Sunak than they were with John Major in 1997. A Labour landslide looks increasingly likely, and Labour voters want nationalisation, increased public spending and higher taxes. The next election could have a seismic impact on British politics as the recent Conservative era crashes to a close."
Find Out Now polled 18,151 GB adults online between 24 January – 12 February 2024. The sample was weighted to be representative by gender, age, social grade, other demographics and past voting patterns. Regression techniques were used to infer projected seat results.
Find Out Now and Electoral Calculus are both members of the British Polling Council and abide by its rules.
Data tables are available here, including full list of seats.
Electoral Calculus is a political consultancy specialising in quantitative analysis and modelling for electoral and other market research projects. Its pre-poll prediction for the 2019 general election was the most accurate published forecast. It was founded by Martin Baxter, its CEO.
Electoral Calculus is a member of the British Polling Council.
Find Out Now gathers poll responses from Pick My Postcode, a daily panel from 2.8 million members. Highly profiled respondents can be targeted instantly, and at scale to deliver reliable results fast. More than 124 million responses have been received to Find Out Now's polls since it launched in November 2018. Find Out Now are Market Research Society Company Partners and a member of the British Polling Council.
Modern polling analysis often uses statistical regression techniques to get more accurate and geographically detailed results. Also called MRP (multi-level regression and post stratification) they have been used successfully by Electoral Calculus and other pollsters to predict general elections, local elections and the 2019 European elections.