How times change. In bygone days, polls were rare treats. Mid-term figures for voting intention would appear two or three times a month. Pundits and strategists would savour each one, as if sipping a rare malt whisky. Today they flow like running water. At least one hundred have been published since that start of this year. This has led to some highly selective reporting.
Last week, for example, the Daily Telegraph trumpeted a Deltapoll survey reporting that Labour’s lead had more than halved in a week to 10 per cent. The paper paid rather less attention to a PeoplePolling survey saying Labour’s lead had jumped six points to 25 per cent.
What, then, is really going on? We all know – at least, I’m sure regular readers of these blogs know – that even the most perfectly-conducted poll is subject to a margin of error. Probability theory tells us that 19 times out of twenty, a poll should be correct to within three points for each main party – and one time out of twenty, purely random factors will give us a “rogue” poll that is outside that range.
Of this year’s hundred polls, around five are likely to have been such accidental rogues. Precisely because they are outliers, they are likely to produce dramatic figures. News editors love them; the rest of us should beware.
To combat this trend, various websites combine the results to produce a rolling average – what we used to call a “poll of polls”. This smooths out the wrinkles caused by random variations. Currently this process puts Labour’s lead at around 20 per cent. Is that where the electorate is today? Perhaps; but perhaps not.
Here’s the problem. Half a century ago, when “poll of polls” became fashionable, the main pollsters used broadly similar methods. Differences between different companies conducting polls at the same time were likely to flow from random factors. Averaging made sense.
Today, however, polling methods vary from company to company. Instead of three or four working in much the same way, we have at least fourteen, and their results vary systematically. They can be divided into two groups, those that tend to report above average Labour leads, and those that tend to report below-average leads. Let’s call them red-group and amber-group polls. They are:
Red-group polls: FindOutNow, Focaldata, Ipsos, Omnisis, PeoplePolling. Redfield & Wilton, YouGov
Amber-group polls: BMG, Deltapoll, Kantar, Opinium, Savanta, Survation, Techne
Here are the separate averages of the two groups from the start of the year to last weekend:
As that table shows, there is a seven-point gulf in Labour’s lead between the two groups of polls. The amber group polls has provided 29 of the 30 poll showing Labour to be less than 18 points ahead, while the red group has provided all of the 25 polls showing Labour more than 23 points ahead. The biggest difference concerns the Conservative vote share (24 versus 29 per cent). Both groups of polls show the Tory share well down on the 45 per cent they won in 2019, but they disagree on the scale of the haemorrhage.
Maybe the truth does lie in between the two groups, and that Labour’s lead in the past three months has been around 20 per cent. But what if one group is right and the other wrong? If the red group is right, then Labour is currently in landslide territory, and the Tories have little chance of preventing Keir Starmer winning an outright majority.
However, if the amber group is right, then the next election could be much closer. Depending on what happens in Scotland, and the extent of tactical voting in England and Wales, Labour will need a lead of at least 8 per cent in the popular vote, and possibly as much as 13 per cent, to win an overall majority in the new House of Commons. On the amber group figures, a modest Tory recovery could well produce a hung parliament.
The good news as that we shall eventually pin down the truth. The bad news as that we must wait until after the next election for a definitive verdict. Meanwhile, we shall soon have an opportunity to shine at least some light on the matter. The local elections on May 4 will tell us a fair amount about the public mood. They will leave some important gaps: there are no elections this time in Scotland, Wales or London. Some big cities, such as Birmingham, will also sit this one out, as will a range of more rural areas. But enough of England will hold elections to give us plenty of data to digest.
How should we judge these? First, let’s cast aside the figures that generate most attention as the results flow in: the number of seats that each party gains and loses. To see why, come to Devon. In the east of the county is the new town of Cranbrook. It will elect three councillors to East Devon District Council. Last time round, in 2019, 818 electors cast their votes – or 273 per councillor. An hour’s drive away is Peverell, close to the centre of Plymouth. Four years ago, its 4,605 voters elected a single councillor. When variations are so great, simple numbers of gains and losses can be misleading – especially this year, when many councils have rejigged their ward boundaries.
Instead, the figures to watch are the “National Equivalent Vote” (NEV) shares for each party. These explore the pattern of gained and lost votes for each party across the country, and estimate how the whole of Britain would have voted had elections been held everywhere. The BBC is likely to provide its NEV estimate on Friday May 3. Colin Rallings and Michael Thrasher of Nuffield College, Oxford, will produce their figures over the weekend. In the past, there have been slight differences between the two sets of estimates, but they generally agree on the broad picture.
As a rule, Labour does worse, and the Liberal Democrats better, in local elections than in opinion polls. Bearing this in mind, what will success look like for each party?
For Labour, an NEV vote share of around 40 per cent, and a lead of 15-plus points over the Conservatives would be an excellent result, and suggest that the red group polls are in the right ballpark.
The Conservatives should be relieved if they can keep Labour’s lead to around ten points. This would also be more consistent with the amber-group polls. For technical reasons flowing from slightly different methodologies, the BBC has recently reported lower Conservative shares than Rallings and Thrasher; so Tory relief would come with 25-plus per cent (BBC) and 28-plus per cent (R&T). This would still leave the party a long way from victory at the general election, but it would reduce the risk of catastrophe.
As for the Liberal Democrats, they have recovered from their dismal performances of a decade ago. Last year their NEV share was 19 per cent (BBC) or 17 per cent (R&T) – roughly double their opinion poll rating. Success would be to reach 20 per cent in at least one estimate, and preferably both, for the first time since 2010.
This analysis was first published by Prospect