How many “Don't knows” could a pollster know if a pollster knew unknowns?
How many “Don't knows” could a pollster know if a pollster knew unknowns?
Despite a laser focus on polling uncertainty in CCHQ, such signals are in fact only mixed for the Conservatives
There is nothing especially unique about the overall level of uncertainty in vote intention polls right now, but there is something unique about the partisan composition of that uncertainty. This could give hope to, but should also worry, the Conservatives.
Hello everyone, and welcome to my first Substack newsletter in a… little while.
Fair to say it’s been a busy first few weeks of the year over at YouGov towers. I won’t say much more on those MRPs released by the Telegraph this January, other than that it’s worth keeping in mind the difference between what YouGov choose to field and choose to release (and how we release it), and what clients choose to do with the work YouGov does for them (and how others report it).
And no, I don’t know who the donors are!
In this issue, I’m going to pick over the data on those people currently telling us they “don’t know” who they would vote for if a UK general election were being held tomorrow, and place it in a longer-term perspective. This is an extension of the 1000-worder I wrote for Con Home in January, for whom I’ll be doing a monthly piece this year up until whenever Rishi Sunak decides he’s had enough of keeping us all in suspense and sends us to the polling stations.
For this article, I’ve made use of having no word limits to expand on the Con Home article points, plus add in figures from 2010 and 2015 for extra comparisons and context.
So, let’s get into it.
Right now, according to the latest YouGov poll, Labour are on 44% of current vote intention, while the Conservatives are down on 23%. This 21-point gap is down from the eye-watering 27 points we saw earlier in January, but would nonetheless spell nothing short of Conservative Party wipe out up and down the country if it fell out this way the next general election.
While the ‘wipe out’ possibility is front and centre of many minds within the Conservative party, the media, and the commentariat, some of those closer to Sunak’s inner-circle are less down and out. They point to one eye-catching set of people within current vote intention data which gives them what they say is cause for extreme hope – those currently telling pollsters they “don’t know” how they would vote if a general election were happening tomorrow.
In recent months, anywhere between around 15-20% of people polled in YouGov vote intention surveys are telling us that they simply do not know who they would vote for if there were marking a ballot paper in a polling booth tomorrow.
Is that a lot? Well put it this way: it suggests that the jury is out on all parties – including the Conservatives and Labour, and Sunak and Starmer – for somewhere between 1-in-6 and 1-in-5 of the electorate. That’s around between eight and ten million voters.
But is it a lot in historical perspective? In short, not really, no.
The table below shows the average (across three polls) percentage of people telling YouGov they don’t know who they would vote for if a general election were being held tomorrow, at this same stage at each of the last five election cycles1. Ahead of 2024 election, that figure currently stands at 17%.
Looking at the table, 2024 levels of uncertainty do not look at all particularly large in recent context, being the same as average levels of voter uncertainty recorded in 2017 and in fact lower than uncertainty ten months out from 2019.
Uncertainty now does however look a little larger than 2015, and notably larger than in 2010. What is significant about 2010 of course is that it was the last time an opposition party successfully supplanted a government and took the keys to Number 10.
Voters were, it seems, more sure about what was in front of them before they last replaced a government than they are now.
Even still, 17% is not worlds away from 12%, and it is not even the highest amount of voter uncertainty we have had in recent electoral history.
But there is something quite particular and unusual about this current crop of uncertain voters - if not their relative composition within the electorate: their previous voting behaviour.
A number of pieces have been written lately about how undecided voters look more likely to break for the Conservatives than any other party. One of the key reasons why is that just under half of all those who currently don’t know who they would vote for if a general election were being held tomorrow voted Conservative in 2019.
Yes, you read that right - of all people who are uncertain about how to vote at the next election, almost half are 2019 Tory voters.
While over one-fifth (22%) of 2019 Conservative voters are uncertain about what to do at the next general election, that figure is only 9% for 2019 Labour voters.
In the graph below is the percentage of each party’s voters from the previous election who were (or, in the case of 2024, are) telling YouGov that they did (or do) not know who they would vote for in an imminent hypothetical general election at (presumably) a similar distance out from the actual vote itself.
And as we can see, that 22%-9% Con-Lab gap is completely unprecedented in recent electoral cycles.
This is uncharted (aside from the one below… I’ll get me coat) territory.
The next highest figure of uncertainty in recent elections comes from 2017 Labour voters ahead of the 2019 contest (23%). That election of course turned out to be Labour’s worst performance at a general election since 1935.
With that in mind, we might actually take the high uncertainty figure among 2019 Tory voters as (yet another) sign of impending doom for the Conservatives, rather than a potential saving grace.
But there are good reasons to think that if indeed many of these uncertain voters did come out to vote at the next election (which there is no guarantee that they will)2, they would be more likely to vote for the party they backed previously than to switch to another.
What consequence might this have for the election to come? If that assumption bears out, then the Conservatives may well have a much larger pool of the electorate currently likely to actually vote for them when the chips are down, and the “who governs?” question is asked, than most headline vote intention figures actually suggest.
But how realistic is that? And how much could it eat into Labour’s lead?
Some pollsters already attempt to answer this question in their headline vote intention figures. Take Opinium for example, who after collecting their vote intention samples subsequently remove all who say “don’t know” from this sample and re-weight the figures back to what they estimate to be the ‘likely voter’ population. This, in theory, gives a handle on what, according to Opinium’s model, is likely to happen with the “don’t knows”.
But we can appeal to a much fuller, probabilistic model of this, which also gives us a much stronger sense of how “don’t know” behaviour will actually effect things seat-by-seat.
Isaac Levido was of course addressing Tory MPs, telling them not to ‘divide and fall’, shortly after a brand new YouGov MRP model was dropped into the mixing bowl by the Telegraph. That, based on over 14,000 interviews, showed that, if an election were being held now (or rather with current patterns of vote intention), the Labour party would win a majority of 120 seats.
In short, Starmer would take his party from that historic 1935 electoral defeat to not only overturning the hefty Conservative majority won by Boris Johnson in 2019, but building a majority of his own the size of which not seen from any party since Blair’s years in government.
What is particularly important about the MRP poll in reference to this discussion is that it accounts for what might happen with the “don’t knows”, baking it into estimates you see published at the constituency level.
This is in fact why, unlike other projections using headline vote intentions, our MRP model did not show Labour north of 400 seats, or the Conservatives below 100.
For reference. it is also a significant part of the reason why the implied national vote shares from the MRP are so different to headline YouGov vote intention polls.
The MRP technique works out what it believes, given all the data is has, to be is the most likely pattern of voting for each type of voter in country by building models of relationships between voters who have given a vote intention in the sample data, and using them to project how all different types and tribes of voters across the country would vote in an election being held at that time.
This can be seen as much more statistically advanced and sophisticated variation (and, one which has been around much longer) of the sort of “don’t know reallocation/re-weighting” efforts noted above by pollsters such as Opinium. It takes out “don’t knows” and uses information from expressed vote intention to account for the likely behaviour of everyone in the electorate.
Altogether, that means that our YouGov MRP projection is already taking into account all of the above reasons and rationale for the Conservatives to maybe have a bit more optimism looking at the current polling than they otherwise might.
This is, in effect, what things would look like in a best-case scenario regarding the “don’t knows” right now. What YouGov are saying is that, even “accounting for the don’t knows”, as the polling currently stands, the Conservatives are on to a hounding.
If the Tories are to stop Labour sweeping to victory, they not only need to bring back in even more “don’t knows” than models suggest they currently will, but also win back a large number of voters who have abandoned the party for their rivals to the right and left.
In other words, “don’t knows” are just one (significant, and large) piece of the puzzle the Conservatives must solve if they are going to deny Labour victory this year.
Here, I am assuming a mid-November election date.
In fact, in my analysis of turnout during the 2019 General Election for the Ford, Bale, Jennings, and Surridge book The British General Election of 2019, I suggested that there was quite some evidence in the data to suggest that a large number of 2017 Labour Leavers in particular stayed at home.
Good graphs. And of course those high Lib Dem don't knows in 2015 didn't save the party...