When pollsters asked voters in Finchley and Golders Green in early October how they would vote in a possible general election, 41 per cent backed the Liberal Democrats ahead of 29 per cent for the ruling Conservative party. Labour trailed in a distant third.
The constituency — a Conservative-held marginal seat in north London — has been shared in recent elections by the two main UK parties. But the internal Lib Dem polling appeared to vindicate the decision to parachute in Luciana Berger as the party’s candidate. The former Labour MP quit the party in February over its failure to address anti-Semitism within its ranks. Finchley and Golders Green, which voted overwhelmingly to remain in the EU at the Brexit referendum, has the largest Jewish population of any seat in the UK.
Yet five weeks later, by which time the election had been called, a second survey, conducted by a different pollster, Deltapoll for The Observer newspaper, asked 500 constituents the same question. This time the Conservatives led on 46 per cent, the Berger campaign had slipped to 32 per cent and Labour was even further behind.
The two polls graphically illustrate the huge uncertainty surrounding next week’s British general election. They highlight the volatility of the electorate — the 17-point gain for the Conservatives between two surveys just five weeks apart is unprecedented — in a campaign which has been dominated by Brexit and the breakdown in traditional party loyalties it has created.
They also help explain why tactical voting is set to be one of the decisive factors in the election. The latest polls indicate the Conservatives could be on track for a majority because the party has managed to consolidate a large section of voters who supported leaving the EU in 2016 behind its “Get Brexit Done” sloganeering. The final result, however, could depend on whether Remain supporters, armed with opinion poll data, will vote tactically for the party most likely to beat the Conservative party led by Prime Minister Boris Johnson — which in most constituencies is either Labour or the Lib Dems.
Yet the push for more tactical voting will only work if the polls can be trusted — which is especially hard in an environment where voting intentions are so fluid and where every constituency presents its own story.
“It’s hugely difficult to know what’s happening on the ground. It’s much more acute this year,” says Shaun Roberts, director of campaigns and elections for the Lib Dems. “[There] is a realignment of voters around Brexit as opposed to traditional party factors.”
For a political system once famed for its stability, Brexit is forcing a dramatic sorting out of voters. More than 3m people are predicted to change party allegiance at this election.
Between 1964 and 2017, the number of people swapping their allegiance between parties at UK general elections rose from 13 per cent to almost a third — in 2015 it was 42 per cent. Yet in the 2019 European election that figure jumped massively: 74 per cent switched away from the party they had backed just two years earlier at the 2017 poll.
A combination of Brexit — both main parties have fought their own civil wars over which position they should take since 2016 — and unpopular leaders in Mr Johnson and Labour’s Jeremy Corbyn are expected to produce enormous voter churn. Large numbers of Conservatives who voted to remain in the EU and Labour leave voters have found themselves looking for new homes — providing an opening for much greater use of tactical voting.
Most public polling is done on a national basis and surveying the same seat twice, as happened in north London, is virtually unheard of. But from Great Grimsby in the north of England to North East Somerset in the west, the parties are doing more of this granular work than ever.
These constituency-level polls have become integral to campaigns. By making the information public — something that was very rare in previous elections — the parties hope to shift behaviour and convince voters that they are the best tactical choice.
Yet the recent record of the pollsters is not strong. Blindsided by the Conservative majority in 2015, some polling companies overcorrected in 2017, resulting in an underestimation of the Labour vote share by five points. Add to that the mixed bag of results on the 2016 EU referendum, and it means that the stakes are also high for the polling industry.
Making sense of this blurred, uneven landscape is not just an academic pursuit. Parties use this data to determine where to deploy resources; where to invest the time and energy of volunteers and identify changes that give them an opening. For smaller groups like the Lib Dems, that won just 7.4 per cent of the vote in 2017, such information is critical.
“People look at what the Lib Dems did in 2017 and say ‘there’s no chance’,” says Damian Lyons-Lowe, chief executive at pollster Survation, which conducted the first Finchley survey. “So the party is asking itself what is a good way to convince people that the facts are as they are. The constituency polls are an effective strategy.”
The volatility in the electorate is playing out differently across the country. As a result, parties cannot easily apply the lessons learnt in one constituency to another, creating a messier map than at any time in election history.
“A lot of people don’t know [how they might vote],” says Sally, who lives outside Portsmouth on England’s south coast, and asked not to give her surname. “It’s like the thing we want isn’t on the ballot paper.”
That sense of a lack of a choice is widely felt across the UK. Alan Dollery, a former trucker who lives in Birkenhead, near Liverpool, explains his dilemma: “At the last election I struggled but voted for Corbyn. [But] I’m a leaver — like quite a few of us here. And since the referendum it’s been very confusing.
“The NHS and everything else I wouldn’t put it in the hands of Johnson,” he adds. “But I don’t like Jeremy Corbyn either.”
In constituencies such as Birkenhead independents could split the vote even further. Frank Field, the MP for the area since 1979 who had a majority of 33,000 at the last election, fell out with the Labour leadership and is now running as an independent.
“I am not voting Labour because of Corbyn. But I won’t vote Liberal because of the coalition,” says Bill Nock, a former Labour councillor in Birkenhead, referring to the austerity policies introduced by the coalition government that took power in 2010. He is backing Mr Field.
In an effort to address this voter volatility, polling companies are using more methodologies at the 2019 election than ever before in the UK.
Since the 1950s the method adopted to project national vote shares at general elections on to constituencies has been known as uniform national swing — which takes each party’s change in national vote share from the previous election, compares it with the latest polls, and applies that adjustment up or down to its share in every seat.
Based on that the Conservatives would win 47 per cent of the vote in Finchley and Golders Green, Labour 34 per cent and the Lib Dems 14 per cent — a Tory hold. But the swing calculation here is the same in every constituency, ignoring that the factors at play in Finchley are different to those in Ashfield in Nottinghamshire or Stirling in Scotland.
To address that, an alternative method has risen to prominence in the past five years, multilevel regression and post-stratification, or MRP, which uses one mega-poll of tens of thousands of people across the country to model how every constituency is likely to vote based on the demographic and political profile of its electorate, and any other local factors such as Ms Berger’s standing with Jewish voters.
Hailed as the future of election forecasting in 2017 MRP was used by YouGov, working with political scientists Ben Lauderdale and Jack Blumenau, to correctly predict a hung parliament while other forecasts suggested a Conservative majority. It also, on average, came closer than rival methodologies to predicting the results of every individual seat. YouGov surveyed just over 100,000 people for its 2019 MRP poll published last week that predicted a 68-seat Conservative victory.
Yet its 2017 success has given a sense of infallibility to MRP that its record elsewhere doesn’t necessarily warrant. YouGov was not the only pollster using MRP in 2017 — the other, Michael Ashcroft’s company, forecast a hefty 64-seat Tory majority.
Outside the UK, MRP polling has been most accurate in strict two-party systems where the link between politics and demographics tends to be neater, but unlike the UK where several parties from the Lib Dems to the Brexit party who could all have an outsized impact on the outcome.
“MRP won’t work as well for smaller parties,” says Kevin Cunningham, an independent political statistician who built one of the MRP models being used in the 2019 election. “A party on 15 per cent may not even appear [among the respondents] in many constituencies.”
This imprecision is clear in the MRP models that are circulating. YouGov’s model estimates that the Lib Dems are running in second place in 117 constituencies, but another model, carried out by data analytics company FocalData on behalf of pro-Remain tactical voting site Best for Britain, gives them only 96 second-places.
Yet, there is consensus among pollsters, political scientists and statisticians that MRP is the best available election-forecasting tool for estimating at scale what is happening in individual seats.
The core challenge of modelling how different people will vote, comes down to which factors — from education to ethnicity and past voting records — the model-builders allow into their calculations. Another issue — one that thwarted some pollsters in 2015 and 2017 — was modelling turnout. “The people who are very hard to poll are also very unlikely to vote,” says Mr Lauderdale, “so you end up missing out on parts of the population that don’t vote and don’t show up in the poll.”
These decisions on factors can influence the shape of the electoral results map that each model generates, and there are early signs that some of the models circulating in public are falling into the trap of producing unnatural swings in vote share between parties.
“This can happen if you have a model that is failing to capture the ways that Labour voters in really Labour places are different from Labour voters in places that were less Labour,” says Mr Lauderdale. “That’s not to say it’s not happening, but it’s probably not happening to that extent.”
Before polling day on December 12 results from several MRP-based polls models are expected to be circulated, fuelling anti-Brexit tactical voting campaigns. But it is not certain that they will give people a clearer view of how their vote might be most effectively used.
Even though three of the models — those of YouGov, DataPraxis and FocalData — all belong to the same core MRP family, their results for each seat differ. In most cases these discrepancies are small enough that they matter little to party strategists. But this time round variations as small as a single percentage point could make all the difference to tactical voting operations — if they can accurately interpret the moving picture.
In Kensington, Sam Gyimah, until September a Conservative MP is standing for the Lib Dems against incumbent Emma Dent Coad, who won the London seat for Labour in 2017. At the end of October, Focal Data’s MRP model estimated that the Lib Dems had leapfrogged Labour and were set for a straight fight with the Conservatives.
Two weeks later Deltapoll ran a constituency poll telling a similar story: the Tories were on 36 per cent, narrowly ahead of the Lib Dems on 33, with Labour third on 27 per cent. Last weekend the DataPraxis model emerged, and put Labour back ahead of the Lib Dems by 30 to 28, identifying Ms Dent Coad as perhaps the best vessel for tactical voters opposed to leaving the EU. Three days later the YouGov model gave the Lib Dems the lead by 29 points to 26.
These are small differences, but for any pro-Remain tactical voting exercise to succeed, they are critical.