Jeremy Corbyn

Every time the opinion polls seem to get it wrong, we journalists say to ourselves that we mustn’t allow our reporting to be so dominated by polls next time. And then it happens again.

In the 1992 election, I worked on the BBC election results programme, so I witnessed first hand the panic as the prediction from the exit poll switched (about 10 minutes before we went on air at 10pm) from Labour being the largest party in a hung parliament to the Conservatives. The huge screen behind David Dimbleby was changed from a picture of a happy Neil Kinnock to one of a beaming John Major.

The coverage of that election campaign was dominated by the assumption, generated by an average Labour lead in the opinion polls of only two points, that Kinnock was likely to be prime minister, probably supported by minor parties.

Labour policies were scrutinised closely not just for themselves but for their acceptability to the Liberal Democrats; there were endless discussions of the mechanics and possible horse-trading of a hung parliament, and in the final week, a huge fuss about electoral reform. Similar things happened in the 2015 election, the 2016 referendum and the 2017 election. In no case were the opinion polls very wrong, but in each case the assumptions built on them coloured the reporting of the campaign, and the result came as a surprise.

So far, we are not making those mistakes this time. The coverage of this election generally assumes that Jeremy Corbyn and Boris Johnson have a roughly equal chance of being prime minister – with the remote possibility sometimes noted that it might be Jo Swinson, or an alternative Labour or Tory leader acceptable to her.

I think this is the right approach. Labour policy was hardly taken seriously in 2017 because few journalists expected Corbyn to win. This time Labour’s plans are already under greater scrutiny, although the initial surprise is that John McDonnell’s tax and spending plans are regarded by many economists as more sustainable than Sajid Javid’s.

But we ought at least to understand that this even-handedness defies what the opinion polls are saying. The current polls suggest a Conservative majority. Even the least favourable to Johnson, Survation, suggests he would have 18 seats more than all the other parties put together. The most favourable, Opinium, gives him a majority of 172, almost as big as Tony Blair’s 179 in 1997.

Naturally, many people refuse to believe any of these numbers at all. The polls got it wrong last time, they say, so we can’t trust them. But we should recognise that the pollsters have corrected the over-corrections they made last time, and that their results are now in line with the new polling method that was a success in 2017.

One memorable poll at the last election was the YouGov seat-by-seat model that suggested a hung parliament. Many people thought The Times made a mistake by running it on its front page, including me, but also including YouGov itself, which based its final prediction on its conventional poll. But the conventional poll was wrong and the Multilevel Regression and Poststratification (MRP) poll was much closer to the result.

So far in the run-up to this election, there has been only one poll using that method. It was carried out for Best For Britain for its tactical voting campaign, called GetVoting.org, which advises people on how best to cast a vote against Brexit in each seat.

This polling is a little out of date, having been carried out in September and October. Since then, the Conservatives have gone up four points, on average, and Labour has gone up two. That suggests that if the Best for Britain MRP model were updated, it would point towards a Tory majority of 70, which is the same as is suggested by the current average of conventional polls. MRP is not necessarily better than conventional polling. There were two other MRP polls in the last election, and they got it about as wrong as the normal polls. So we cannot be sure about any of them this time, but, as ever, they are a better guide to what people are thinking now than any other way of assessing public opinion.