Moving the needle: what kind of swing to expect at the next election
Labour look set to win big at the next election. Attention has started to turn to the question of just how big that win might be. We have been provided with a succession of MRP polls showing Labour getting eye-popping majorities of more than 200. More recently, we have seen the Economist produce a nowcast with a central estimate of Labour getting a majority of just (just!) 94. While even that would be a fine mandate for a new Prime Minister, there’s a big difference between the two. Since both are using broadly the same figures, how can the difference be explained?
There are questions about tactical voting and the extent to which the polls are accurately picking up Conservative support but at the centre of the question is how the Conservatives’ loss of votes will be distributed. Historically, swings between the two main parties at general elections have tended to track a uniform national swing: that is, if we apply the national swing between the parties uniformly in each constituency (UNS), we’ll expect to get more or less the right overall seat count.
When you stop and think about it, there’s no logical reason for this. In seats where one party has done poorly in the past, you’d think they’d have fewer votes to lose than in seats where that party has done well. Still, over the years uniform national swing has worked fairly well in practice. It’s not a law like gravity. UNS has sometimes been quite far adrift of the mark – for example in 1997 it overestimated the Conservative seat count by 42. Most of the time it’s been near enough.
The objection to MRP polls is that intrinsically they rely on swing working in a quite different way. Their methodology produces an effect of proportional swing: that is, votes are gained or lost in constituencies proportionaly to their pre-existing shares. These accentuate swings between parties, resulting in greater seat turnover.
Chris Hanretty has looked at this discrepancy. I strongly urge you to read his article, here. It’s excellent. To summarise, you’d expect MRP polls to produce proportional swing before the tug of war of the election campaign has begun. Once the election campaign starts and the parties pick up their ends of the rope, MRP polls should produce less proportional and more uniform results as voters absorb the messages the parties are giving them in their local constituencies.
I find Chris Hanretty’s argument compelling. Yet I still think the swing next time is likely to look more proportional than usual.
That’s a bold claim. Bold claims need strong evidence. Fortunately, I have it.
In fact, I have two strong bits of evidence. One is mathematical and one is situational from the last election. More weakly, I have some polling evidence that accentuates the points arising from the maths and the situation.
The maths
The mathematical point is touched on in Chris Hanretty’s article. Where a party has a low vote share in a constituency, a national swing of greater than that vote share would, if you naively apply uniform national swing, result in that party being deemed to get a negative vote share in that constituency.
Normally so far as Labour and the Conservatives are concerned that’s a theoretical point only. It’s exceptional for one of those party’s vote share to change by more than a few percentage points.
In 2024, however, we face exceptional circumstances. In 2019 the Conservatives tallied 44.9% of the vote share in Great Britain. Currently their average vote share in the polls is somewhere around 23.5%. That means they look as though they’re losing over 20% of vote share.
There are 69 constituencies in Great Britain where the Conservatives tallied less than 20% of the vote share (I’m leaving the Speaker’s seat out of account). These are mostly in Scotland and inner city England, especially London. If the Conservatives end up with less than 25% of the vote share at the next election, uniform national swing breaks down so far as these 69 constituencies are concerned.
You might think this is a technical point to be brushed over. But the average vote share in these 69 constituencies is 13.9%. Even if the Conservatives don’t get a single vote in any of these seats, the remaining 562 seats need to absorb an additional loss if the maths is going to work. That additional loss is about 0.75% off the Conservative vote share in each of those constituencies.
Of course, the Conservatives will not get 0% in 69 constituencies. At the last election, Labour lost their deposit in just 11 constituencies. All 11 were long-standing two way marginals with a non-Conservative option where Labour’s vote had been systematically squeezed over successive elections. No such squeezing mechanism will take place on the Conservative vote in any constituency.
You can make your own guess what the expected baseline will be for Conservative vote share in such constituencies if they tally say 25% nationally. I will make an assumption that they will get 5% on average in these 69 constituencies. I regard that as being very much at the low end for the Conservatives for the reason I’ve given. That means that something like 1.3% on average would need to be lopped off the Conservative vote share in each of the remaining 562 constituencies.
Oh, but that’s not the end of it. If you agree with me that it’s implausible that the Conservatives will drop below a certain level in any given constituency, you also need to think about the constituencies in which the Conservatives tallied between 20 and 25% of the vote last time – there’s another 40 of those. You’d expect the Conservative average vote share in these seats to be a touch higher than in their very worst seats. Let’s say that the average for these bottom 109 seats is bumped up to 6% (I still regard this as low, by the way). That means a hit of c1.9% from the Conservative vote share in the remaining 522 constituencies.
Now, obviously these figures are imprecise and you really shouldn’t be using decimal places when thinking about this. In the context of marginals, however, the effect is potentially meaningful. Modifying UNS to take account only of this mathematical point is potentially the difference of a dozen seats.
We happen to have quite a bit of polling for both London and Scotland, where many of these seats are clustered. That polling consistently suggests that the Conservatives have seen smaller falls in these areas than across Great Britain as a whole.
The sum total of all this is that anyone applying uniform national swing naively is going to underestimate the number of Conservative seat losses unless the Conservatives start picking up vote share quite markedly.
The situation
There’s been a lot of attention lavished on Reform UK’s rise in the polls. There has, however, been surprisingly little attention given to what that might mean in practice in the ballot box.
I have repeatedly expressed my scepticism about some of the polling for Reform UK. But let’s put that to one side for now and assume that it is broadly correct. Where is their vote coming from?
When that question is asked, it is usually meant metaphorically and the answer metaphorically is given: disaffected Conservatives. 72% of Reform UK supporters in the most recent Opinium poll voted Conservative in 2019. In the most recent Redfield & Wilton poll, two thirds of Reform UK supporters voted Conservative in 2019. The general picture is clear.
But let’s think about that question geographically. In 2019, the Brexit party (as it then was) chose not to stand in any constituency with a sitting Conservative MP. In fact, it only contested 276 seats. Its national vote share of 2% at that election is thus seriously misleading: it equates in reality to a national vote share of over 4.5%, if we make the assumption that its vote would have been evenly distributed. As I explain below, this is a dangerous assumption.
The impact of Reform UK’s strategic decision not to stand in any constituency with a sitting Conservative MP was, as you would expect, felt largely in seats that the Conservatives do not currently hold. Indeed, of the 38 seats where the Brexit party tallied more than 10% of the vote, just four are now held by the Conservatives (including Hartlepool, which the Conservatives later picked up at a by-election).
It is possible to imagine a world in which the Brexit party largely took votes from Labour, and where by standing in Conservative seats in future Reform UK (as it now is) might assist the Conservatives to hold those seats. That, however, does not seem to be the world we live in. Indeed, as I note above, polling consistently shows that Reform UK’s reservoir of support is drawn in the main from former Conservative voters.
Reform UK have pledged to stand in every constituency at the next election. That in practice means there are over 300 Conservative seats which didn’t have a Brexit party candidate in 2019 that will have a Reform UK candidate this time. Even if they stood still from last time, we could expect Reform UK to take at least 4.5% of the vote in those constituencies.
In practice, we should expect to see them do better than that on average. The Brexit party competed in 50 of the 69 seats where the Conservatives got less than 20% of the vote share. The Brexit party kept its deposit in just six of those seats (they kept their deposit in 92 of the seats they stood in) and got more than 10% of the vote share in just three of them. Brexit party performance was positively correlated with Conservative party performance in 2019. Its decision not to stand in Conservative-held seats probably artificially suppressed its vote share, even when pro-rating to allow for the seats in which it did not stand.
All the polling suggests that Reform UK are set to make an advance on the Brexit party performance last time round. As noted above, the great bulk of its support now comes from 2019 Conservative voters.
So the impact on the Conservatives will be two-fold. First, an artefact of the 2019 election shape of voting, by which Reform UK vote share was suppressed in Conservative-held seats by the decision not to stand in such seats, will be removed. This will inevitably give a hit to the Conservatives in exactly the constituencies (the ones they hold) where they least want to get it. This will not be uniform because it cannot be uniform.
I suggest that this hit is treated as 5% off the Conservative vote share on average in each constituency it held in 2019.
Secondly, we have polling evidence that Reform UK’s new support is drawn strongly from previous Conservative support. If that new support for Reform UK endures, it is not obvious that it will be distributed uniformly across constituencies. There seems no obvious reason why it should follow a pattern established for swings between Labour and the Conservatives. And indeed, there is scant polling evidence of a Conservative to Reform UK swing in Scotland or London.
Caveats
The observant will have noticed two major caveats. First, I have assumed that the Conservative vote share will not substantially rise. Secondly, I have assumed that Reform UK’s vote share will hold up. If either of these assumptions fails (and they are correlated contingencies) then my conclusions will also be undermined. The extent to which they are undermined depends on the extent to which the assumptions fail.
That said, MRP polls and the Economist’s forecast are based off information as we currently have it. This post should be considered in the same light.
Conclusion
The distinction between the two camps, UNS and proportional swing, can be overstated. Most of those advocating a UNS approach accept that it will not be a straight line adjustment. For my part, I accept that the swing will – after adjustments – probably owe more to UNS than to proportionality. But for the reasons set out above, I think that if the end result is anything like current polling, we should be making substantial adjustments for the extreme nature of the swing and the confounding impact of Reform UK before applying UNS.
It’s important not to double count. The loss of support to Reform UK in constituencies held by the Conservatives in 2019 is logically prior to the mathematical problem with UNS on large swings. Even after making adjustment for that (and without making any allowance for the Conservatives’ relatively better polling in Scotland and London or for Reform UK support being correlated with 2019 Conservative support), the effect of this is going to be at least an additional 3% swing over and above UNS, and probably rather more than that. So if the end result is something like current polling I expect the Conservative seat losses to be dozens more than a strict application of UNS would predict.