The final countdown

A roundup of the final polling data of the 2024 campaign

Well, here we are. The election is imminent.

In my last post I argued that the Conservatives would make a concerted push to get out the core vote, rolling back some of the gains Reform has made at their expense.

As I’ve said before, not everything can be scripted. In addition to these efforts we’ve seen Reform forced to drop candidates who had made racist remarks, while others decided to jump ship.

What has this wrought? Let’s see:

That ‘lump’ in the Conservative vote share looks like a belated tightening of the polls to me. This time I haven’t normalised out the pollster-specific differences: at this point I think it’s better to see the raw polls (each dot is an individual result) and the spread between them¹.

That choice slightly obscures the trend. Pollsters differ in their assessment of party support. Some of the last to publish their results do not generally favour the Conservatives’ chances; this makes it look a bit like the Conservatives rose, but then fell slightly, to Reform’s benefit.

To control for this, I double checked the pollsters publishing the last few results (the ones gathered on or after the weekend) and compared each pollster to their previous result from the week before. Normally, if nothing has changed, we’d see random variations of ±2 percentage points; they should average out to zero. Instead, all pollsters showed a rising or flat vote share, with an average rise of nearly two percentage points.²

In short, there seems to be a clear pattern: a meaningful shift to the Conservatives in the last couple of days.

As I’ve said previously, there’s not enough time left for the Conservatives to close the gap between them and Labour. The Conservatives are still polling worse than they were at the start of the campaign! However, the size of the Labour majority could be very dependent on small changes in the Conservative vote share.

Is it prediction time?

Yes, it is.

I downloaded and compared the final MRP—sorry, Megapoll!—-based predictions from no fewer than eight different pollsters. These companies are projecting really different outcomes! However, a result of around 100 Conservative seats and about 450 Labour seats wouldn’t be too embarrassing for any of them.

Here’s a sample:

• Survation seat prediction: 34—99 Conservative, 447—517 Labour, 49—73 Lib Dem

• YouGov: 78—129 Conservative, 391—466 Labour, 57—87 Lib Dem

• More In Common: 126³ Conservative, 430 Labour, 52 Lib Dem

This more-or-less represents the full spectrum of results. We can see that the upper and lower estimates don’t overlap—but all are predicting a very large Labour majority (Supermajority!).

On the other hand: Survation are invoking a scenario where the Liberal Democrats could, on a good day, overtake the Conservatives and seize the constitutional role of His Majesty’s Loyal Opposition. Meanwhile, More in Common are projecting a humiliation, but not an annihilation.

Polls are never perfect. They come with statistical uncertainty, but also with assumptions. These assumptions lead to a large scatter in the results between pollsters. This scatter is larger than the statistical uncertainty. And if you want to do a clever statistical model—all the cool kids are building MRP models these days—this point holds even more strongly.

So here’s what I did. I looked at all the constituencies where every single model predicts that the same party will be in the lead:

I make that 377 constituencies where these pollsters’ models, with all their varying assumptions and widely-varying outputs, all agree that Labour will come first⁴. You need 326 for a majority⁵.

In other words: if Labour don’t win a majority of more than 100 seats tomorrow, the polling industry might as well collectively pack up overnight and flee the country.⁶

As an aside, the fact that these predictions are so sturdy in the face of different assumptions is because the majorities in those constituencies are large. The late swing I talked about can’t affect these seats.

Other constituencies

Of course, that’s only the unanimous predictions. What about consensus predictions—i.e. where five or more of the eight companies agree on the most likely result?

Here we have to be a bit more cautious. The data for these models were collected over the course of several weeks. That late swing to the Conservatives—which could continue to grow—makes a bigger difference at the margins. Indeed, as I said a few days ago, that’s now the whole ball game for the Conservatives.

The wisdom of crowds: seat predictions where a majority of pollsters agree. With (1) no late swing, (2) a small swing, and (3) a big swing.

Here is the ‘consensus map’. Or rather, three versions of it.

In the first version you’re getting the actual consensus.

In the second version I’ve applied a small swing to the Conservatives of the kind I think we’ve seen over these last few days. You may be able to see at a glance that a bunch of places switch hands. I think pricing this in is the right call.

In the third version I’ve imagined that this late swing continues to grow over the last couple of days of the campaign, doubling in size. This is a big ‘if’ but worth exploring—even if merely to show you visually what a big difference it makes!

In terms of seats the Conservatives go from 83 to 99 to 117 seats. The number of seats where pollsters are evenly split in their predictions goes up, rising from 20 to 29⁷. We might expect at least some of those to go the Conservatives’ way.

Cards on the table

All in all, if I were to bet on it, based on this ‘wisdom of crowds’ approach, I would guess something like:

• Labour: around 450 seats

• Conservatives: around 110 seats

• Liberal Democrats: around 60 seats

For the smaller parties:

• Greens: either one or two. Brighton Pavilion is likely to remain Green. Pollsters are evenly divided on whether Bristol Central will stick with Labour or not.⁸

• Reform: Nigel Farage will most likely win in Clacton. I need at least one hostage to fortune, so: my gut says that, while they have a few possible prospects, Reform are being squeezed too hard elsewhere to break through in many places. Low single digits.

There are some other results that matter. I expect Plaid Cymru to win perhaps three Welsh seats. Most pollsters believe the SNP in Scotland will be rolled back to under 20 seats (from the current 48), throwing a fire blanket on the separatist cause. And Northern Ireland never gets enough attention, including from pollsters—it certainly doesn’t make it into any of their Megapolls.

It’s worth emphasising that this is the first election campaign in which lots of different companies have published these constituency models. My consensus-based approach may not be the right call; it’s quite possible that Survation have captured something important in their model that eludes their competitors. We won’t have to wait long to find out.

For tomorrow

It’s a real privilege to be able to vote in regular, free, and fair elections. Politics is often viewed cynically, but it would be a shame if we were to forget that. So, whatever your politics: go and vote. Enjoy it. And whatever the shape of our country’s future may be, we’ll face it together.

1. Worth noting that this spread is quite a bit larger than the shaded region, which indicates expected variation from the average based on statistical uncertainty. That’s because pollster assumptions affect their results by more than the variation due to random chance.

2. Oh, you want the individual numbers? Fine:

• WeThink: 23%, up from 20% on the 28th of June (+3)

• Opinium: 21%, up from 20% on the 28th of June (+1)

• Deltapoll: 22%, up from 21% on the 29th of June (+1)

• PeoplePolling: 16% (spot the outlier!), up from 15% on the 26th of June (+1)

• Whitestone: 21%, up from 18% on the 27th of June (+3)

• BMG: 22%, up from 20% on the 26th of June (+2)

• Techne: 21%, up from 19% on the 27th of June (+2)

• Redfield & Winton: 22%, up from 19% on the 27th of June (+3)

• Verian: 21%, no change from the 24th of June (+0)

• JL Partners: 24%, up from 22% on the 25th of June (+2)

• Savanta: two polls, on 20% and 24%. Average of 22%, up from 21% on the 24th of June (+1)

• More in Common: 24%, up from 23% on the 26th of June (+1)

3. More In Common aren’t providing an official range, just a central estimate, but we can still compare their best estimate with the ranges given by other pollsters.

4. The numbers for the other parties are 26 for the Lib Dems, 13 (!) for the Conservatives, 7 for nationalist parties, and 214 seats where the models disagree. Many of these seats without consensus between models will most likely go Conservative.

5. As a reminder, a government’s ‘majority’ is the number of surplus MPs they have relative to the number they need to equal all the other MPs combined. There will be 650 MPs elected in July. If you have 326, you have a majority of two (two more MPs than all the rest put together). If you have 377 MPs, that leaves 273 MPs you don’t control: a majority of over 100.

6. And I will need to find a new hobby.

7. Interestingly, quite a few of these are Conservative/Lib Dem marginals.

8. Fun fact—these two constituencies are adjacent in the alphabetised list of constituencies. Maybe the Greens should target ‘Brigg and Immingham’ next.