Predicting the UK 2019 Election

Election Dec 11th 2019

See Update at bottom

Following on from our first look on the 7th of December (see here), there has been a small rise in Tory fortunes and a small drop in LibDem and the minor parties, Labour roughly static. The Tory rise has been due to a sharp uptick over the last 3 days, for the first time they have hit over 50% of the Zeitgeist.

This makes prediction “Interesting” as the smoothed curves (of all ilks – the main one we use, the cumulative curve, is shown above) still show the Tories below 50% of the vote but the momentum of the last few days may well propel them above on the day – or, of course, it may revert to the mean.

Therefore, to make our prediction we will embrace the first Weaselly Caveat, of probability mathematics.

In a nutshell, the most probable outcome is still a Tory Victory, but with less than half the seats in the house. In other words a Hung Parliament is the most likely outcome. This we calculate has a probability of about 60%

However the odds of a Tory victory have gone up to about 35% (from c 15%) and while that probably means a small majority of a few seats, Britain’s first-past-the-post system could deliver a larger majority depending on where the votes fall (see further Weaselly Caveats below).

In addition, the shift of the Tory support upwards at the expense of the LibDems has reduced the chance of a Labour + Libdem hung Parliament to c 5%.

Other Weaselly Caveats that apply are:

  • We still do not know what will happen to the Liberal Democrat vote on the day – at c 9% of the vote in a close race it can still distort the picture, especially in first-past-the-post vote distribution. Will it stay firm or will it collapse into (probably mainly) Labour support?
  • Tactical voting – what % of the population will indulge? So far the Remainer parties have steadfastly embraced the Judean Popular Front strategy of hating each other more than the Romans (Brexiteers) but the voters may not
  • “Shy Tories” – mainly upper class demographics who talk liberal and vote Tory – we don’t think there are any  anymore, but there may well be some shy Champagne Socialists who will vote to keep their taxes low.


Well we got the right party winning, but we had only c  1/3rd probability given for an outright win. Our analysis of why we didn’t get a higher probability showed that:

  • The social data analytics system got the vote share just about spot on, once the numbers sre nomalised for the c 4% of minor parties we didn’t track, so the data analytics prediction engine works perferctly well.
  • The error was mainy in the seat allocation from the vote difference, and the analyst (me) is to blame there, got my equation translating vote share to number of seats wrong.  Essentialy I under-weighted the impact of vote share differential on seat allocation at high (c 9%) differentials (I knew it was a geometric equation, but not that much!)
  • The system saw the late spike of momentum, but the analyst under weighted it’s impact.

From the Weaselly Excuses Dept comes – (i) predicting seats in “first past the post” is more difficult than proportional voting or the USA’s “two horse race” system, and (ii) there hasn’t been a vote share to seat allocation ratio like that for a generation!

That last point about “not seen that ratio in a generation” is informative – as with Trump’s win (and Thatcher’s), a large demographic that once voted for one side shifted to the other. Typically in any system dynamic, that shift implies some major factor has reversed. In this case, a large number of traditional labour supporters have reversed to support the Conservatives. Making another prediction (caveat emptor ….) it takes a lot of effort to reverse a reversal.