A projection of the 2017 election
I did a lot of work before the 2015 election trying to anticipate what might happen seat by seat in that election, looking at local factors, candidates, demographics, recent local election results, Lord Ashcroft‘s constituency opinion polling and all of that. Like most prediction efforts in 2015, it was not very successful. It was swamped by three things –
(1) the broad national tide in England and Wales that submerged quite a lot of the local differences I anticipated. Part of the Conservative success was built on ‘nationalising’ the election choice, to the detriment of the Lib Dems in particular who were hoping for personal votes to salvage them as happened in elections when their tide has been receding, such as 1979.
(2) the completeness of the SNP sweep in Scotland.
(3) the extraordinary success of the Conservatives with their protective targeting of the marginals they gained in 2010 and a handful of Labour seats they picked off.
I’ve tried out a different approach this time, with a mechanistic formula applied to all the English and Welsh Con/ Lab seats and nowhere else. Martin Baxter at Electoral Calculus has done a vastly more sophisticated and complete version of the same sort of exercise, but this is a crude formula incorporating rather few variables. It’s based on the swing to the Conservatives being partly proportional to the UKIP vote, an element of uniform national swing, a bump in one direction or another for seats at the outliers of referendum voting behaviour (Remain or 60 per cent plus for Leave). There’s therefore a double weighting for strongly Leave seats, which tended for obvious reasons to have high UKIP votes in 2015. The source for constituency referendum voting is Chris Hanretty’s compilation of published local results and his work modelling the others. An allowance has been made where the Greens have stood down or intervened.
The initial model had a national baseline of the Conservatives winning by a 15-point margin, which is where a fair few polls were at the time. That would produce a Con gain of 44 from Labour. Allowing for a few gains from SNP in Scotland, and from UKIP in Clacton and possibly a couple from the Lib Dems, this would be a Con gain of around 52 seats for an overall majority of around 120.
The way the model works skews the Conservative gains to the North and white working class England. Labour would hold apparently perilous marginals in London such as Ealing Central & Acton and Brentford & Isleworth, but lose some former heartlands. Strikingly, the Tories would win all three seats in Stoke-on-Trent (Labour since 1935) plus Newcastle-under-Lyme (Labour or radical Liberal since 1906) for a clean sweep in Staffordshire.
Given that 15-point leads have not featured in the most recent polling, I did a simple modification to see what might happen if the Conservative lead were 10 points – i.e. a relatively small national swing of 1.7 per cent to the Conservatives. The modification is a ‘dumb’ 5 point drop in the Conservative lead in each seat, except Copeland where I have assumed a lingering effect from the by-election in February. The number of Conservative gains from Labour would fall to a mere 15 and there would be one offsetting Labour gain (Brighton Kemptown); with additional gains from other parties the Conservative majority would be 50-60.
If the model is right about the contours of party support in this election, the difference between Con +10 and Con +15 is very important for Theresa May’s future authority over her party – a three figure majority or a margin that hardly seems worth it won at the cost of considerable political capital.
I think the model would need a bit of tinkering under the bonnet to cope with a scenario like Con +5, which I still don’t think is at all likely, but is within the current polling range. I might run that model if it seems worth it after the weekend. But it’s interesting that the YouGov seat model shows unlikely places like Battersea as ‘lean Labour’ with its baseline of Con +4 while my Con +10 model would still give the Conservatives Battersea by 5 points. It suggests that the YouGov model is weighting the demographic and referendum factors even more strongly than I do.