Thread by @benwansell

How should we be thinking about the confusing array of UK General Election polls right now? As @leonardocarella pointed out the other day, it all depends on how votes split within Remain and Leave groupings. I've tried to model this a little more. Thread follows 1/n

@leonardocarella So @leonardocarella begins with dividing constituencies up into the Leave/Remain spectrum. I'll add to that by taking into account where Labour performed well vs Tories in 2017. We have @chrishanretty's estimates for Brexit vote and the GE results. That looks like this. 2/n

@leonardocarella @chrishanretty We can see already an in-built Leave supporting party advantage. As @chrishanretty finds - there are more constituencies that had a majority for Leave than Remain. Oh BTW, everything that follows is England and Wales only. Sorry. 3/n

@leonardocarella @chrishanretty What I'm going to do is take the poll of polls from @britainelects and see if I can use this as a baseline for simulating how each constituency might vote. And then I'm going to alter that by giving different parties an edge due to Leave / Remain vote. 4/n

@leonardocarella @chrishanretty @britainelects So let's begin at the beginning. I take current Cons 32, Lab 24, Lib 21, and BXP 13 , and then I use the GE17 proportion of vote in each constituency but imagine it scaled up to today's polls. I then see for each constituency who wins. And I get this: C 305, L 193, LD 70 BX 0 5/n

@leonardocarella @chrishanretty @britainelects Look where Labour are strong. The whole right-hand side, which they won in GE 2017. The LDs take some seats since their average polling is higher but there's no direct Brexit effect yet. Tories + DUP (not shown!) lack a majority. 6/n

@leonardocarella @chrishanretty @britainelects But now let's add a Brexit effect. I'll make it the same across parties and the way I do it is give the Tories and BXP more votes from constituencies above the Brexit mean and LDs and Lab votes from those below. And I get this. Con 304, Lab 104, Lib 157, Bxp 3. 7/n

@leonardocarella @chrishanretty @britainelects Labour got seriously damaged. Why? Lib Dems are already polling similarly and start taking a lot of pro-Remain but anti-Cons constituencies. Labour also lose a lot of Brexity marginal constituencies. BXP don't really benefit a great deal 8/n

@leonardocarella @chrishanretty @britainelects So let's help Nigel out. I'll give the BXP double the effect the Tories have (and keep Lab / LD the same in terms of their ability to pick up Remainers). Now we have Cons 273, Lab 95, Lib 160, BXP 40. Tories are hurt but Labour even more so. 9/n

@leonardocarella @chrishanretty @britainelects What if we really are in a 4 party world and BXP and LDs are both best at picking up people on Leave / Remain dimension. Then we get this. Cons 256, Lab 76, Libs 194, BXP 42. So now the LDs really start eating into the Tory vote, though not in a way that really helps Remain 10/n

@leonardocarella @chrishanretty @britainelects Can Labour resolve this by becoming the party of Remain? I'll make their attractiveness to Remain strongest and put everyone else back at moderate effects on Leave/Remain dimension. Doesn't help. Con 320, Lab 105, Lib 141, BXP 2. Cons just take the Brexit seats Lab now lose. 11/n

@leonardocarella @chrishanretty @britainelects Finally, what if Boris shoots the Brexit Party fox and Cons become most attractive to Brexit voters (everyone else moderate). Well not a lot happens: Cons 314, Lab 102, Lib 152. What he takes from BXP he loses to Libs. 12/n

@leonardocarella @chrishanretty @britainelects Alright. Enough graphs. Time for caveats. This is England and Wales only (and I've ignored Plaid - sorry!). The modelling I am doing is very simple - just creating distributions from GE 2017 and Brexit vote that match current polling on average. 13/n

@leonardocarella @chrishanretty @britainelects But, if you're interested I have posted the code and csvs with data online at github.com/benwansell/Gen…. Hope it works! But you should be able to see the assumptions I have made and be able to alter them by adding new polls or different parameters 14/n

@leonardocarella @chrishanretty @britainelects What I've learned from this (other than that simulating is a nightmare), is that its very easy for parties - well Labour - to get caught in the middle. The seats they need to hold are Brexity marginals but the seats they need to gain are Remainy marginals, where LDs are. 15/n

@leonardocarella @chrishanretty @britainelects And maybe the other lesson is that I don't see how Boris wins big just by winning over Leave voters. He needs to crack into very very Labour seats to get a majority (move right on my axis) or Remainy Tory seats (move down). Killing BXP not enough. n/n

@leonardocarella @chrishanretty @britainelects Oh and one more caveat. I think my model is too friendly to the LDs but can't quite figure out why. It may be that it over-favours them in anti-Labour pro-Remain areas, ie Raab, IDS, Boris even might be losing their seats in my model. So I'd hold off on the Swinsonmania yet n+1/n