Byline Times is an independent, reader-funded investigative newspaper, outside of the system of the established press, reporting on ‘what the papers don’t say’ – without fear or favour.
To support its work, subscribe to the monthly Byline Times print edition, packed with exclusive investigations, news, and analysis.
Polling company YouGov has set alarm bells ringing with its projection that Nigel Farage will win Clacton with over 40% of the vote, prompting Alastair Campbell to call for Labour and the Conservatives to combine to block him – while others called on the Lib Dems and Greens to back Labour in the seat.
A panic had already been triggered by a Survation projection that Reform UK could win seven seats. And then Farage’s collaborator Matthew Goodwin popped up with a ‘poll’ that showed Reform nationally nine points ahead of the Conservatives.
It’s right to take the projections from independent polling companies seriously, if not Goodwin’s. Clearly Farage could win his seat, he should be blocked and tactical voting for a real alternative (not the Conservatives who are largely indistinguishable from Reform) is urgently needed. Yet we need to be clear that these are not conventional polls: they are projections, which although they use polling data, are based on complex “MRP” (multi-level regression and post-stratification) statistical models, and it is very difficult to evaluate them sensibly.
As polling expert Matt Singh points out, there is an almost complete “lack of transparency”; none of the pollsters has “yet even listed their variables either for turnout or vote choice”, let alone explained fully how they combine the granular demographic data which is MRPs’ unique selling point with assumptions about political behaviour.
One of the more transparent pollsters, Ipsos Mori, specifically warns: “we would encourage readers to not place too much certainty into specific point estimates.” Yet this is what even widely publicised and reputable tactical voting sites are doing.
The Exmouth case
My own alarm bells were triggered by Survation’s call, since their ‘Reform’ seats included – of all places – the new constituency of Exmouth & Exeter East, which is mostly the old East Devon seat, next door to where I live. While Reform will undoubtedly garner some support, locals reported that it had hardly been seen in the constituency, and no one credited its projected first place.
Reform’s candidate Garry Sutherland appears even by its standards to be distinctly lacking in charm – he has a conviction for kicking a dog and has shared David Icke videos – and his behaviour at a recent hustings confirmed the impression this information gives.
Sceptics like British Future director, Sunder Katwala, were quick to suggest on X, formerly Twitter, that the source of Survation’s apparently rogue prediction for Exmouth was that its MRP model was finding it difficult to cope with the very unusual result in East Devon in 2019, when a left-leaning Independent, Claire Wright, was the main opposition to the Conservatives, winning 40% of the vote.
This difficulty had already caused one important tactical site, StoptheTories.vote, to pull a Labour recommendation for Exmouth based partly on MRP projections.
This might not have mattered too much – none of the other projections agree that Reform is ahead in Exmouth and some like Electoral Calculus suggest that the Lib Dems, not Labour, are the challenger.
However, Best for Britain’s Get Voting, the most heavily promoted tactical voting site, is using Survation’s projection and using it to urge Labour voting in Exmouth – which may have the effect of helping the Conservatives cling on. This case therefore brings to the fore some major concerns about how MRPs, which are widely publicised both by their producers and the TV sites, could perversely skew tactical voting.
MRPs and tactical voting advice
The problem with using MRPs for tactical voting is not only that they are routinely described as ‘polling’ on Get Voting and other tactical sites, when they are not polls but statistical projections. It is also that their primary purpose is not to guide tactical voting, but to provide more accurate overall projections of the overall arithmetic in the next parliament.
MRPs have gained prominence because of the increasing fragmentation of the British political scene, with greater regional and local variation in how swings in public opinion affect constituency results and hence the parties’ national tallies.
The various MRPs use different models, and although their claims for greater precision than traditional polling might seem suspect – they currently offer a huge range of possible Labour majorities – they can claim modest successes in their first major outings, the 2017 and 2019 elections, and they are at least an attempt to deal with the reality that British elections are decided by 650 separate constituency contests.
The problem with MRPs is that while in the aggregate, they could improve predictions of the electoral outcome – although by how much is debatable – they may not be especially reliable in predicting individual constituencies, especially where something that doesn’t figure in their models has happened. They use impressively large national samples: Survation’s most recent had 42,000 respondents, but that works out at only 65 people per seat, and the raw local data is not published, but processed via the model.
A senior pollster at one of the major firms pointed out to me that their MRPs aren’t designed to provide constituency-level advice and are published with health warnings. Another quickly admitted that his MRP could produce “surprising” results in what he called “idiosyncratic” seats.
Yet while tactical sites are ultimately responsible for how they use the data and voters for how they interpret it, pollsters can’t shrug off their responsibilities here – they produce projections in the knowledge of how tactical voting sites and voters, who will mostly understand little about them, could use them. Indeed some polls, like Survation’s for Best for Britain, are even commissioned with this in mind.
When local knowledge is an ‘anomaly’
The irony is that while MRPs work by producing local projections, many don’t appear to use much local political knowledge apart from the result of the last general election. Indeed many MRPs are even allergic to local knowledge, since it complicates their models.
In the current campaign, many have failed to adequately incorporate – or at all – obviously relevant political data which is far more recent than the 2019 election, like the results of the by-elections which have upended politics in dozens of local areas, and of council elections. Rather, they seem bent on forcing tactical sites and voters alike to somehow compute the significance of such new information for themselves.
I know this because the constituency I live in, Honiton & Sidmouth, is mostly the old Tiverton & Honiton seat where the Lib Dems famously overturned a huge Conservative majority in 2022’s ‘porngate’ by-election.
Labour had been a poor second in the 2019 election, but were squeezed to a tiny 3.7% in the by-election, while the winning Lib Dem came from third to get 53%. Yet many MRPs and TV sites, basing their projections on the 2019 result despite the startling more recent data from 2022, were still calling Honiton and Sidmouth for Labour at the start of the general election campaign.
Indeed even on 15 June, Survation projected Labour to get more votes than the Lib Dems in Honiton & Sidmouth, which flies in the face of what everyone on the ground knows – even Labour themselves, who are not campaigning in the seat and have sent their activists off to Plymouth.
When I talked to a pollster about MRPs not taking account of by-election results, the response was that it would introduce an “anomaly” into the model because not all seats had by-elections. Yet this “anomaly” is a crucial political reality in so many seats – how can the data it has created not even be acknowledged?
Survation’s Honiton & Sidmouth projection obviously posed a problem for Best for Britain’s Get Voting, its tactical voting partner site, which they solved by manually overriding the projection and recommending a Lib Dem vote on the ground of the new MP’s ‘incumbency’.
Yet Get Voting still publishes the Survation projection with its frankly absurd figures, alongside this recommendation, potentially confusing any voter who looks at it.
The reductio ad absurdum of this approach is how many MRPs and tactical sites are treating Exmouth and Exeter East. In this case, they are not merely disregarding the fact that a party got a deposit-losing vote, as Labour did in the Tiverton and Honiton by-election.
Rather, since they have difficulty in factoring the Independent’s 40% vote in 2019 into their predictions, many have actually used the tiny Labour vote of 4.5% in 2019 to help project Labour as ahead of the Lib Dems (who had an even tinier 2.8%) in 2024.
Both parties had been almost completely squeezed, but these miserable results are still steering flawed MRP projections and tactical advice almost five years later.
Get Voting actually has a ‘local factors’ option for override Survation’s projection, but they haven’t used it for Exmouth & Exeter East despite its obvious idiosyncrasy.
What should be done?
Tactical voting is an essential way for voters to ensure, under Britain’s flawed electoral system, that they get the result they want – in this election, so many people in every seat want to take their once-in-a-lifetime opportunity to ensure that they no longer have a Conservative MP.
Tactical voting sites rightly put a lot of money and effort into producing advice, but they need to get it right, since if they don’t, they could help split the anti-Conservative vote and help some undeserving Conservatives cling on.
In the end, it is difficult to know how far the problem of skewed advice extends, although it probably affects a large swathe of the South where, even without by-elections, the Lib Dems have re-emerged in the last three years, making weak Labour second places in 2019 doubtful guides to 2024 voting.
It will also be a particular concern in the seats in which significant Independents are standing, such as Jeremy Corbyn in Islington North, Faiza Shaheen in Chingford and Woodford Green – who is given a notional 0.06% by one polling site – and pro-Palestinian candidates.
At this late stage of the game, when postal votes are already being cast on the basis of inadequate information and advice, tactical voting sites should go into damage limitation mode – and the pollsters should help them to do this.
ENJOYING THIS ARTICLE? HELP US TO PRODUCE MORE
Receive the monthly Byline Times newspaper and help to support fearless, independent journalism that breaks stories, shapes the agenda and holds power to account.
We’re not funded by a billionaire oligarch or an offshore hedge-fund. We rely on our readers to fund our journalism. If you like what we do, please subscribe.
The tactical sites should acknowledge the limitations of MRP projections for their advice, and look hard at how – and in some cases whether – they should publish the MRP data, and be more willing to press their manual override buttons.
For the future, pollsters should reconsider the good old constituency poll in the less predictable cases – in this election, only The Economist has commissioned these, and they have shown significantly different results from most of the projections for the seats in question.
And tactical voting sites should invest much more in local knowledge. It really shouldn’t be beyond them to build up qualitative political profiles of the 650 constituencies instead of relying too uncritically on statistical models, designed for other purposes, which are often insufficient for voting purposes.