TL;DR summary: if you adjust the uniform swing so that voting patterns reflect echoes of past Labour or LibDem strength, the predicted Tory majority vanishes. If you add onto this a measure of tactical voting, their seat share might fall by dozens of seats more. But detecting whether this is realistic is very, very hard.
Before launching into this, a recap.
I have been on quite a journey, hopefully towards a decent model for the impending* General Election. It began with a straight arithmetical exercise, intended to turn headline voting numbers into seats, done in a very naive way: take a certain chunk of Conservative and Labour votes, and reassign them uniformly so that you get the national vote share. Like this:
The result of this kind of exercise was set out in this post which reflected on the sheer volatility and sometimes arbitrariness of the results. For example, the numbers above produce for me a 33%-26%-18% win for the Conservatives over Labour, but 338-202-35 in terms of seats. Brutal. The method was destined to deliver a very poor outcome to a split opposition with the Tories in a clear lead.
However, it also looked naive, in at least two ways.
First, LibDem votes just head heedlessly to every seat in an even manner. This struck me as unlikely: recent European election results showed a much lumpier, more motivated voting surge. For example, there are Labour seats where the model spits out a significantly better result for the LibDems than they achieved in May: places like Blackpool South were showing a LibDem vote share of 13%, despite their only scoring 9% just four months ago. And that meant that in other places the LibDem surge was being undermeasured in the model – places like Harrogate that went 28% in EU2019, and 43% as recently as 2010.
So I designed a factor to reflect this “LibDemmyness“, and found that a modest application of that factor might raise the LibDem gains off the Conservatives by around 20-30-40 seats.
This also applied to Labour. The steep fall in their vote share (from 40% in 2017 to 26% now) meant a quite vertiginous fall everywhere, even places that are historically very pro-Labour. Is this realistic? Lord Mandelson in a recent event cited Hartlepool, his old seat, as an example – in each of the past three elections, Labour had beaten the Conservatives by a minimum of 14% – yet my model had that shrinking to 5%. Now, maybe that is possible: the Brexit Party took 52% of the vote in May, so who knows. Mandelson may be out of touch. But the Conservative party took just 5% in EU2019 and so a model suggesting they are competitive looks a bit odd.
So I added a “Labourishness” factor, and found that a modest application of this might raise the Labour seat total by 10 – mostly, taken from the Conservatives’ total. Here are two examples: a seat that stops turning blue, and a seat that goes LibDem, thanks to these factors.
To emphasize, this is not a prediction. It merely says that if these older voting propensities come good, then you get results 30-40 seats worse for the Conservatives. Put another way, the “unfair” luck they enjoy from the voting system is partially eroded.
Now, getting to the point. What about tactical voting? You could argue that these factors already take it into account – they basically instruct voters to emphasize their past Labour and LibDem patterns, which quite inevitably pushes in a tactical direction. But given the stakes, it is not unreasonable to wonder if voters will think hard about whether their vote will have the effect they want and change accordingly. Matthew Goodwin, the expert academic, has written about this and modelled a situation where dozens of LibDem and Labour candidates just stand down (and, presumably, just hand their votes to the other one). The result – Conservatives collapse from 366 seats to around 100 less.
For me, that is too extreme. Candidates don’t stand down, and their voters do not obey like sheep. Instead, I have set up a milder version like this:
- Choose four categories of seat where LD-LAB tactical voting may take place. They are
- Labour held, less than 50% of the vote (39 seats)
- LibDem held in 2017 (12 – I appreciate I must now update this!)
- Conservative held, and even if the LibDems were polling 25% nationally they would still be third (72)
- Conservative held, and even if Labour were polling 35% nationally, they would still be third. (53)
- Then apply a % to the votes that the ‘conceding’ party would pass to the other party.
The result? For every 10% of tactical voting, there is a loss to the Conservatives of around 8-9 seats. Here is a chart:
Incidentally, Labour gain 5 seats for every 1 that the LibDems gain – what you would expect, but still a reason to stop and think about its political saleability as a bargain.
Which brings me to 1997 and 2001. These are the elections that give us the best sense of what degree of Tactical Voter-iness is possible.
I wanted to work out how much TV went on there, according to my model, and so rebuilt the machine using 1992’s data, and went to work trying to reverse engineer a 1997-style Labour majority. (This is a very ugly way of operating, with all sorts of assumptions – the 1990s were very different from today.) I found that without any tactical voting aspect, the Tories would have won 190 seats. So to deliver them their 165 seat nightmare, you would have needed a 45% tactical voting switch across 200 seats that they won in 1992.
Apply that much tactical voting this time round and you obviously produce a very poor result for the Conservatives, even if they gain some of the higher national vote share totals they have recently scored (around 33-4%). This may explain why the Conservatives internal polling was weaker.
Apply it to some of the weaker outcomes recently polled – e.g. 31% Con, 28% Lab – and the result is a total rout against the Conservatives – seat numbers in the low 200s.
Bottom line: it is impossible to predict, but if this highly confrontational behaviour by the Conservatives inspires tactical voting against them anywhere near what we saw in the 1990s, their chance of a majority vanishes. I cannot tell if that is a realistic assumption; I hope to illustrate many more specific seat model-predictions in order that the hive mind can tear it to pieces (or perhaps validate).
*Though the odds of a 2019 vote have slipped to around 65%, at time of writing, down from 90%