The pollsters’ conundrum: how big or small a victory could Labour win?
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
This UK general election presents an unusual headache for pollsters: while everyone agrees Sir Keir Starmer’s Labour party is on course for victory on July 4, there is huge disagreement about how big that win could be.
The FT’s polling average shows the opposition party with 43 per cent support, 21 points above the Conservatives. But this headline figure disguises a wide variation between pollsters.
How much do the polls differ by?
In the 55 polls released since the start of the campaign on May 22, Labour’s lead over Prime Minister Rishi Sunak’s Tories has varied from 12 percentage points at its lowest to 27 at its highest.
Support for smaller parties is similarly uncertain, with Nigel Farage’s Reform UK party ranging from eight to 19 points and the Liberal Democrats varying between six and 15.
Projections of the number of seats each party will win differ even more.
In the four multilevel regression and post-stratification (MRP) polls released since the election was called, the Tories are projected to win anywhere from 66 to 180 seats. MRP polls use far larger samples than conventional polls to model behaviour among specific groups of voters.
That range is the difference between a near-extinction level event for the Tories, or a defeat on the scale of the Conservative loss to Labour in 1997.
This level of discrepancy was not seen at the last election. In 2019 the four MRPs released during the campaign predicted the Tories would win between 337 and 359 seats. On the night, they won slightly more, 365.
Why is this election particularly hard to forecast?
A large expected national swing and high numbers of former Tory supporters who are currently undecided are putting polling models under stress.
Based on the average headline polling, the expected swing of 16 percentage points from the Tories to Labour would be the biggest in living memory, dwarfing the 10-point swing in 1997.
The four MRPs all have similar national vote shares for the Conservatives, but the differences in how they expect Tory votes to be distributed across the country are driving wide disparities in seat numbers.
If the Tories hold on to support where they have no prospect of winning, but lose it in their heartlands, it could be devastating to their seat total. The reverse effect would minimise losses.
MRPs may also fare poorly when confronted with the big national polling swing expected this year.
A large swing makes predicting specific constituency results particularly tough, as the current electoral landscape has shifted so much that using historical results as a baseline is likely to be less accurate than at previous elections, when fewer voters switched between parties.
This challenge is exacerbated by the UK’s first-past-the-post electoral system, where parties win seats if they get a plurality of the vote in a constituency, no matter how big or small the margin of victory.
If the Tory share of the national vote falls far enough, a tipping point would be reached where they go from winning many seats narrowly to losing many seats narrowly.
Another wrinkle is that many people who voted Tory in 2019 are now undecided.
A Redfield & Wilton poll of 10,000 respondents found that 10 per cent of those who voted Conservative in 2019 answered “don’t know”, compared to just 2 per cent of 2019 Labour voters.
Pollsters would traditionally ignore “don’t know” voters, but some have begun to guess how they will vote by factoring in past behaviour and the preferences of similar voters.
JL Partners has found their method of reassigning undecided voters narrowed Labour’s lead by 3 percentage points.
What impact do pollsters’ different approaches have?
Rival pollsters’ have different solutions to the particular challenges of this election and to more common problems, which contributes to differences in their projections.
One issue is tactical voting, which is hard to predict. When voters respond to polls, their response may reflect who they really support or a tactical view of how to defeat a particular party in their constituency.
Pollsters can’t easily differentiate the two, and attempts to do so risk overestimating the number of tactical voters as such voters typically do not fully make up their mind until very close to the election.
To weed out tactical voters and reduce uncertainty, some pollsters have experimented with so-called “squeeze” questions.
In addition to the industry standard “If there were a UK General Election tomorrow, which party would you vote for?”, YouGov also asks: “Now, thinking specifically about your own constituency, and imagining that these were the political parties standing, which party do you intend to vote for in the July 4th 2024 UK general election?”
“Imagine yourself in the shoes of a strong Labour supporter living in a Lib Dem-Tory marginal, lots of those people answer Labour initially, but the squeeze question makes them say Liberal Democrats,” said Anthony Wells, director of YouGov’s political and social opinion polling.
Another major difference between pollsters is how they treat smaller parties.
Some, such as Opinium and Savanta, do not list the Reform party in the initial options they present to respondents.
Instead, Reform is among a list of parties that are only offered if “Other” is chosen first. This tends to reduce Reform’s vote share, while increasing that of the Conservatives.
The different ways pollsters adjust their raw data to make their samples representative of the population can lead to significant differences in headline results.
More in Common, another pollster, has found that treating older voters in their sample as one group of over-65s shows Labour’s lead to be six percentage points higher than if they split them into 65-74 and 75+.
This is because people aged over 75 are significantly more likely to vote Conservative but far less likely to respond to online polls than even people only slightly younger.
Comments