33%? 17%? What is Labour’s real lead?

Almost 14 million people voted Conservative at the last election. How many of them would vote Conservative today? Just five million, says YouGov; up to eight million say its rivals.  Both are dreadful figures for the Tories, but the gulf between them is extraordinary. It’s this gulf that lies behind the huge differences in the latest polls. Are the Tories 33 points behind Labour, as YouGov says, and heading for catastrophe, or less than 20 points behind, as others say, with some hope of recovering from their mid-term blues? Given the mayhem that has followed recent polls, YouGov’s in particular, it’s worth getting to the bottom of this.

Our starting point is that the contrast between YouGov and other established polling companies is far too big to be explained by sampling fluctuations. Is there something about its methods that could be the cause?

Actually, there is; and if YouGov has got it wrong, blame me. Back in 2005, when I was its Chairman, I arranged for every member of our panel (then around 100,000) to be asked in election week either how they would vote or had just voted. Not everyone responded, but we acquired a database of tens of thousands of people whose subsequent voting intentions could be compared with their  2005 vote – as they had reported it at the time.

YouGov has maintained this practice, gathering vast amounts of baseline data at each election for use in polls through the subsequent parliament. Other companies also look at past vote; but most of them collect this information afresh, at the time of each new poll. So last week’s YouGov poll anchored its analysis in the 2019 votes of respondents as reported in 2019 – while most other companies anchored it in how respondents remembered last week how they voted in 2019.

Why does this matter? Because all polls try to ensure that their published figures represent the electorate as a whole, and most polls (Ipsos is an exception) include past vote in the way they adjust their raw data. We know that the Conservatives defeated Labour by 45-33% among those who voted; it stands to reason that the past vote in any new poll should match those figures.

Here’s the problem. I’m sure that pretty well everyone reading this blog remembers clearly how they voted in 2019; but many voters misremember. What is more, research down the years has found that when a party falls out of favour, the number of people who say they supported it last time also falls.

If people are asked about their current and past vote in the same poll, many will want to give the same answer. They say, they might even think, that they saw through the Tories as clearly three years ago as they do today, whereas in fact many of them contributed to Boris Johnson’s victory. In the 1990s there was much talk of pollsters missing out on “shy Tories”. Maybe some of them are now missing out on shy ex-Tories.

This false memory syndrome may have played a part in last week’s figures. These are the “consistency” figures for each party as reported by different pollsters – the proportion of 2019 party supporters who say they would vote the same way today.

Survation: Conservative 55%, Labour 87%,

Deltapoll: Con 59%, Lab 88%,

Opinium: Con 44%, Lab 82%

YouGov: Con 37%, Lab 75%

YouGov’s figures for consistency are significantly lower than those of the other companies. This is precisely what we would expect at a time when the Government is so unpopular, and the number of shy ex-Tories is on the rise. Interestingly, Opinium’s sample contains a mixture of respondents who reported their 2019 vote at the time, and those whose vote has been remembered subsequently. We would expect its consistency figures to be higher than YouGov’s and lower than Survation’s and Deltapoll’s – and indeed they are.

How does this affect the latest figures, showing the Tories in all kinds of trouble? Here’s how. At the last election 45% of us voted Conservative. Today, it’s likely that fewer than 40% will now say they voted Tory. A perfect sample would report this lower figure. Any poll that increased its figures for the Tory vote in 2019 to match the actual result would then be making its results less accurate.

If this hypothesis is correct, it helps to explain not only why YouGov reports higher leads than other companies, but why they have suddenly become more out of line, as voters turn away from Liz Truss and her party. It could be that false memory syndrome got worse last week, and dented the accuracy of the polls that rely on the answers people give today about how they voted last time.

Therefore, I would defend the system that YouGov introduced 17 years ago. However, before Labour becomes too jubilant or the Tories too suicidal, another factor should be taken into account, and its impact is huge. People who voted Conservative at the last election are more likely than supporters of other parties to say they don’t know how they would vote if an election were held now, or say they would not vote at all. They are omitted from the voting figures, even though all past experience tells us that a high proportion of them will actually turn out in an election – and it’s fair bet that most of them, having not crossed to another party when the Government is in so much trouble, will end up returning to the Conservatives.

Although the pollsters agree that there are more ex-Tory than ex-Labour and ex-Lib Dem  “don’t knows” and “won’t votes”, they disagree on the numbers. Let’s return to the latest round of polls. These are the proportions of 2019 party supporters who now say “don’t know” or “won’t vote”:

Survation: Conservative 15%, Labour 4%,  

Deltapoll: Con 14%, Lab 2%,  

Opinium: Con 29%, Lab 5%

YouGov: Con 35%, Lab 11%,  

According to YouGov, almost five million people who voted Tory last time are currently “don’t knows” or “won’t votes”. Survation and Deltapoll put the figure at two million. These differences are the flipside of the loyalty figures set out the start of this blog. Once again, they flow from the different times at which data on the vote in 2019 was collected. (As with the consistency figures, and for the same reason – the way it collects data on how people voted three years ago – Opinium comes in between.)  And even if YouGov’s figures are technically right, the chances that anything like five million Tories from 2019 would sit out the next election are zero. Omit them if your passion is statistical purity, but include at least some of them if you prefer electoral realism.

Opinium is the one company that adjusts its data, and assumes that a high proportion of each party’s “don’t knows” would in practice return home in an actual election. It does this partly by asking a second, “squeeze” questions, asking “don’t knows” who they would vote for “if you were forced to choose”, and partly by drawing on turnout patterns in past elections, which show that slightly more Labour than Conservative supporters tend to end up staying at home.

Both adjustments lift Tory support and narrow Labour’s lead. In its latest poll, Opinium reports a Labour lead of 19 per cent. But without its “don’t know” and turnout adjustments, Labour’s lead would be 25 per cent – a difference of six points. As YouGov reports significantly more ex-Tory don’t knows, equivalent tweaks might well reduce their latest Labour by around ten points. Think of it as a practical adjustment to plausible reality. It would leave YouGov and its rivals not as far apart as last week’s headlines suggested.

In the light of all this, it’s tempting to say that the polls are all over the place and tell us nothing we can rely on. That’s to dismiss them too easily. They disagree on the current size of Labour’s lead, but agree that it has risen sharply since Liz Truss became Prime Minister. Kwasi Kwarteng’s give-away mini budget has backfired spectacularly. The Tories have lost their reputation for economic competence, and will find it had to regain it; I doubt whether this morning’s u-turn on the top rate of tax will make much difference. If nobody switched parties between now and election day, Labour will would with a lead in the popular vote of around 20%.

Normally we would expect the governing party to gain ground as the election approaches. However, “normally” is not a word that sits comfortably with the way things are just now. In any event it’s hard to see the Conservatives recovering enough to avoid defeat. Keir Starmer is on course to become Prime Minister. The uncertainty, at least in my mind, is whether he will end up with an overall majority (which would need a lead in the popular vote of around 12%)  or leading a minority government.

This blog was originally written for Prospect