Opinion pollsters can sometimes seem like horse racing tipsters – often right, sometimes spectacularly wrong. Particularly interesting is when they are in a minority of one when the others are unanimous for the most fancied runner. Usually an outlier is such because it is wrong – but occasionally it is because it is right.
That was the position of Survation in the final polls for the 2017 UK general election, They put the Conservatives one point ahead of Labour (41-40), close to the actual result (43-41). Eight other companies doing regular party share polling estimated the Conservative lead as between 5 and 13 points. Survation’s success (almost emulated in 2015 when they chose not to publish a poll closer to the result than any of the other 11 published) arose from not adjusting their raw data for what had been presumed to be a general tendency for Labour poll identifiers not to turn out on the day.
The failures discredited pollsters and this has chimed in with the ebbing away of polling as a lead story. Newspapers won’t publicise their rivals and will sometimes present their own polls misleadingly – the Evening Standard headlined ‘Boris soars into 17-point opinion poll lead’ on 1 November when their fieldwork had concluded four days earlier, and more recent polls told a different story.
The BBC tracks poll trends on its news website but will not use them as the basis of a lead story on TV or radio. Sky has taken a different approach by deciding to part-commission polls from YouGov and publicise them quickly. The first of these, on 5-6 November, also asked a Brexit question which, surprisingly rarely, replicated that of the 2016 referendum and found a 50-50 split (excluding 10% don’t knows).
The pattern that is emerging from polls in recent days is a clear order of parties – Conservatives about 40%, Labour close to 30%, LibDems in the mid-teens and Brexit Party slipping back into single figures even before it announced on 11 November that it would not contest Conservative seats. In this context the Conservative vote is consistent with an overall majority, which on three post-war occasions has been won with less than 40% of the UK vote (Labour in October 1974 and 2005 and Conservatives in 2015).
At the same time, confidence has been falling away in uniform national swing as a predictor of Conservative/Labour marginal seats. In 2015 the Conservatives gained 8 seats from Labour and lost 10 to them; in 2017 they gained 6 (all in the Midlands and North) while losing 28 (in Scotland, no seat has changed hands between the two parties since 1997). Of these 52 seats, only three appear twice, switching in both elections (Derby North, Gower and Vale of Clwyd).
Even the broadcasters’ exit poll, dead right in 2005 and 2010, has not coped with this, leading to the underestimation of Conservative seats by 15 in 2015 (the difference between a hung parliament and overall majority) and four in 2017. Uniform swing has also been poor in predicting the location of Liberal Democrat gains and losses. Just like old times, election night is becoming a matter of counting off seats one by one.
Polling relies on getting an accurate sample of the population you are studying. In this case, voters who will actually cast a ballot - about two-thirds of the eligible electorate. Getting the sample – usually via online or telephone means - is a problem, knowing whether they will vote is another one. ‘Margin of error’ (now more common in US than UK polling presentation) is a statistical expression of the risk of a totally wrong result arising from a given sample size and is less important than the quality of the sample in the first place. It would be a pity if the media were put off by these issues to the point of shying away from polling interpretation and allow party presentations of the ‘mood on the doorstep’ to dominate.
With a month to come and the holiday season beckoning, the risk is of a tedious campaign in which little new is likely to emerge about Johnson and Corbyn. The two of them have shut out Jo Swinson, Nicola Sturgeon and Nigel Farage from the first TV debate (ITV on 19 November) which always gets the most attention, and was the basis of ‘Cleggmania’ in 2010.
In the meantime, a lead’s a lead. Labour’s consolation is that at the corresponding stage in the 2017 campaign polls were showing Conservatives in the mid-40s and Labour around 30%; the climb in Labour’s ratings did not became clear until about ten days later and was always underestimated. In 2019 the dynamics are different. The LibDems are stronger but are likely to need more than one election to rebuild their constituency base. The Brexit Party part-withdrawal and implicit alliance with the Conservatives may be a game changer and here the polls in the next few days will come into their own in telling the story. The Conservative vote will have to soften if it is to be a close result, and a key sign will be poll numbers starting with a 3 rather than a 4.