The 650 | 23rd May 2017

How our constituency forecasts model works

By

Constituency Forecasts is our model for projecting the 2017 UK general election result. Unlike forecasts based on uniform, or adjusted uniform swings, we use a mathematical tool called Regional Variants, or RegVar (after the brass band). RegVar attempts to aggregate together data from previous elections to predict how a particular seat will behave relative to the UK as a whole, and to nearby seats. In theory, this means that if Southwark sneezes, then Kate Hoey catches a cold.

The aim is to avoid the kinds of systematic polling errors that can be caused by small errors in the national polls. As Ed Miliband will tell you, more votes doesn’t always mean more seats under first past the post. Similarly, we saw Hillary Clinton rack up a two million popular vote lead while losing the Electoral College – she would have needed her voters to be less than 2 per cent more efficiently distributed to win the day, and the presidency.

RegVar works first by taking the national vote share and comparing it with the regional vote share, simulating the unequal distribution of votes. In the South East, you would expect the Tories to pick up twelve votes for every ten they received nationally, while they might get only seven in the North West. Consequently, for every 100 Tory votes nationally, we expect perhaps 120 Tory votes in Guildford.

These disparities are the basis for RegVar, which handicaps each party in each seat against historical, national, and regionally adjusted performance.

In practice, this allows us to suggest what kind of task each party is up against in each constituency. For instance, in Gower, a seat the Tories won by just 27 votes in 2010, their generous polling lead is making the country’s slimmest majority look unusually comfortable.

We have also incorporated the results of the recent local elections, and the EU referendum into our model, as a Brexit Bonus. The Brexit Bonus further adjusts the overall handicap of each seat depending on the Leave margin, and the competing parties’s stances.

We are working to improve RegVar over the coming weeks, incorporating tactical voting patterns, turnout variables, and long-term polling trends from the archives – going right back to 1970 – as well as the exit polls on the night itself. RegVar does not pretend to tell you who will be the next Prime Minister – we all know who she is – but we do hope to give you a sense of just how big a majority she might win.