If you’ve followed me for a while, you probably know I post a lot of my own constituency-level projections. Most electoral systems in Europe involve distributing seats at a sub-national level, and the UK is one of the most ‘extreme’ examples of that with six hundred and fifty FPTP seats. We have one of the least proportional electoral systems for this reason — which is why popular vote alone won’t tell you everything you need to know about the outcome of an election.
Between general elections, unless you’re willing to commission a MRP (Multilevel Regression & Poststratification) poll, the only way to estimate constituency level results is by building a sort of ‘one size fits all’ model that can crudely convert popular vote into a seat count. Britain Elects (@BritainElects), Election Maps UK (@ElectionMapsUK), Stats for Lefties (@LeftieStats), and Electoral Calculus all have some variation of this — all with their own health warnings and caveats. Here’s how mine works:
The load bearing part of my model is the idea that every party in Britain has strong seats and weak seats. In every constituency, you can calculate a crude ‘strength’ rating by taking a party’s vote share in the seat and dividing it by their vote share across the county. For example, the strength ratings in Hazel Grove (a Lib Dem-held seat in Greater Manchester) are:
LAB: 69.9%, CON: 82.7%, RFM: 105.7%, LDM: 308.8%, GRN: 57.2%
Everything beyond this is where the ‘assumptions’ come in. For example, we have to assume a party’s distribution of strength (AKA votes) doesn’t dramatically shift between elections. From there you can basically just reverse the initial equation, multiplying these figures by the latest voting intention polls. The Lib Dems' 308.8% multiplied by ~14% becomes 43.2%, meaning they increase their vote share in Hazel Grove by roughly 4.5pts.
This is what the main page of my spreadsheet looks like. Input on the left, output on the right.
After step one, you’re left with a result that is mathematically viable and well within the range of possibility. However there are still some realistic tweaks that can, and should, be made.
One of these is party ‘vote efficiency’ — basically whether their votes are concentrated on safe seats they already hold or more spread out across marginal seats they could win. In my model, that’s expressed as a % between -100 and 100 where negative figures spread out a party’s vote and positive figures focus it on safer seats.
The Lib Dem vote is more concentrated than most other parties
The second main “additional” calculation is for incumbent advantages (or disadvantages). This mostly affects Green Party and Reform-held seats that were gained as a result of targeted campaigns. In a situation where the Green Party doubles their vote share nation-wide, it would be unrealistic to see them double their vote share in Bristol Central — which would leave them on… 113.2%. For this reason, Greens, Reform and even Plaid Cymru all face incumbent penalties in my model.
Another feature of the UK electoral system (and FPTP systems in general) is tactical voting. My model has an option to apply tactical voting to the broad ‘left’ and ‘right’ blocs by taking a set % of votes away from lesser parties and transferring them to the best placed winner on a constituency basis. Currently this feature is applied at a very marginal level — 5% maximum. This also includes the option to set up pacts between parties, mainly Reform and the Conservatives, but at the moment that’s just hypothetical.
In the event of a pact, you’d expect the ‘tactical voting’ element to also shoot up to ~50% or greater. In that case it just draws from the lesser party’s *potential* votes rather than their actual votes
The final set of options is something that isn’t applied strongly, but still can be set for hypothetical scenarios. Most of these relate to a party’s performance among particular demographics, but it’s also where a party’s position in the rural/urban divide and north/south divide can be adjusted.
Every constituency is either stronger than average or weaker than average in regard to these categories. For example, you can decide the Lib Dems are actually winning more urban voters this time around and set a slight ‘boost’ for the party in those seats. For me to apply these modifications to the standard monthly projections, there would have to be a concrete justification for it — otherwise it’s just for hypotheticals. Still fun though.
The final component of this seat calculator, only added a few weeks ago actually, is the inclusion of YouGov MRP data. Currently the seat ‘strength’ ratings I mentioned earlier are influenced by recent MRPs to the extent of ~20% their implied strength vs ~80% of mine (based on the results of last year’s general election). For the most part there hasn’t been much difference between my model and MRPs, but I would expect them to diverge the further we get from GE:2024. Maybe then I’ll be relying on their data a bit more.
As I mentioned at the start of this, 95% of the projection is built on the basic foundations of a ‘proportional swing’. The rest is just minor tweaks hear and there. It’s far from perfect, and it’ll never be able to objectively take into account your own local knowledge, but I believe it to be a solid model.
This is the part of the spreadsheet where everything comes out looking neat and fully rounded.
Ultimately it’s up to you how much faith you put into these projections, but I’ll try to make everything as transparent as possible while providing my best possible ‘guess’ of what’s happening at the constituency level. Ballot Box Scotland sums this up much better than I can on their website:
"Even with those more nuanced forms of projection, local issues, personalities, and campaigns have a substantial impact, as do changing demographics. So it simply isn’t possible to take a national poll and use it to predict with perfect accuracy the result in each constituency
the Ballot Box Scotland ethos is to try and improve understanding of elections whilst also assuming people know enough to start with that they don’t need to be beaten over the head with caveats and health warnings every time I post something"
(WiP)
For Senedd elections in Wales, the formula is very similar. Each constituency is calculated the same as I would under FPTP, however, the seat 'strength' ratings had to be calculated based on the results of the 2024 General Election. Normally I wouldn't cross between two different types of election, but the new Senedd boundaries are comprised of Wales' 32 Westminster constituencies paired together. Until there's notional results for these new boundaries, GE 2024 is the only data that fits - and is also recent enough to include Reform UK as a competetive party.
(WiP)
How would you forecast a party that doesn’t exist yet?
This is a good question, and is obviously something that needs to be done if Jeremy Corbyn and Zarah Sultana launch their new party. The benefit of a ‘party strength’ model is that you don’t necessarily need to rely on previous election results. In the case of Corbyn and Sultana, it’s again based on assumptions, but I believe the only fair assumptions you can make right now.
From what little polling data we have, it looks like the new JC/ZS party is breaking though with the same demographic the Greens are, and Labour were in 2017/2019. This is also backed up by the fact most of their potential new voters will be drawn from those who voted Green and Labour in 2024. Additionally there’s the ‘Independent Alliance’ group of MPs who weren't running under a Corbyn Party banner at the last election, but their campaigns might give some insight into how a Corbyn-led party performs with Palestine as one if it’s top priorities.
In short, the seat strength ratings for the new Corbyn/Sultana party are based on a combination of Labour, Green and Independent performance — leading to quite a strong urban-rural divide above all else.
I would love nothing more than fresh data on this though! Currently waiting on more polling.
There’s no way X constituency is going Y in a million years
If you’re local and you disagree with the results then fair enough, I can’t challenge that. However I would not be surprised by anything these days.
X Party did well in local elections here, why isn’t that reflected in the model?
This is a fair point, and something I actually want to work on in the near future. Of course the underlying data in this model has to be from previous general elections, but I will look into introducing a local government element to this.
You’re a GPEW member. How can I trust this model is impartial?
I wouldn’t blindly trust any electoral model by anyone, in any country, for any election, but trust me — rigging it in favour of the Green Party would be a lot of effort for very little payoff. Credibility is very important to me, so I make sure every party is treated the same.
Is there a larger company / organisation behind this?
Nope. I just started doing this because I genuinely find it fun and interesting. If you want to support the work I do here though you can buy me a coffee via Ko-fi: https://ko-fi.com/josh_housden
