Forecasting the 2015 UK General Election

CON 252, LAB 231, SNP 52, LD 41, UKIP 38,  DUP 9, GRN 9, PC 8, SF 6, SDLP 2, Independent 1, Speaker 1

Required for majority: 322 (CON 70 short, LAB 91 short)

The upcoming UK General Election has been called “the most unpredictable for almost a century”. There are several reasons for this; the rise of UKIP, the Greens, and the SNP, the collapse of the Liberal Democrats, who nevertheless seem to maintain highly localized popularity in areas where they have sitting MPs, and the fact that Labour and the Conservatives are going in to the election essentially on level pegging.

The focus of the media is on the UK-wide vote shares published by the main opinion pollsters, but as Mike Smithson from politicalbetting.com correctly says, the winner of the election will be settled by 650 first past the post races, and not the winner of the national popular vote. Given the fact that some seats have become three, four or even five way marginal seats, forecasting the winner has become an incredibly difficult task.

The latter-day god of psephology, Nate Silver at fivethirtyeight.com, correctly forecast the outcome of all fifty states in the 2012 United States Presidential election. However, his model was based on aggregating results from state opinion polls. This approach would not work in the UK, as there is not (and could never be) the same level of constituency opinion polls as there are for states in US Presidential elections. An alternative approach is needed.

I have built a forecast model based on the results of council elections across England, Wales, and Northern Ireland since 2011. This has meant combing through thousands of individual council races, and matching the ward level results to constituencies. For Scotland, I have taken Holyrood Election results and matched them to Westminster constituencies in a similar fashion.

I have had to “fill in the blanks” in a few seats, wherever a party did not stand in every ward in a constituency, which was especially prevalent for the Greens and UKIP in 2011 and 2012.

The final part of the model is to mark parties up or down depending on the prevailing national opinion polls. For England, I have taken the simple average of the four most recent UK-wide opinion polls, and the simple average of the four polls after polling day for each election day in 2011, 2013 and 2014, and used these to create a scalar factor to apply to each party in each constituency. For example, Labour’s polling average after the 2014 local elections was 33.8%, whilst at the time of writing it is 32.8%. Therefore all Labour results for 2014 have been multiplied by (32.8 / 33.8 = 0.97), i.e. a three percent penalty to reflect their decline in the polls. I applied similar adjustment to polls in Scotland and Wales, although polls are a lot fewer and far between, so I simply used the single most recent poll and the poll closest to election day to calculate the adjustment scalars. I didn’t apply any adjustment for Northern Ireland at all; polling is unreliable and public opinion tends to be relatively consistent.

Results and Forecast 2

The map above shows the results of the 2010 General Election, and the forecast result for this year’s election using the local election results to forecast Westminster results. A spreadsheet showing the forecast winner, forecast second and third places, and their forecast votes shares can be found here.

There are a few things that are immediately obvious. Firstly, both the forecast totals for UKIP and the Greens are a lot higher than consensus estimates. I think that it is possible that the lack of precedent of the “big two” of Labour and the Conservatives polling such a low combined share (less than 60% in some recent  polls) means that commentators are being (small “c”) conservative in their estimates of how well the Greens and, particular, UKIP may do. I have spot checked some of my forecast results with the constituency level polls published by Lord Ashcroft, and they are in the same ballpark. Or I may be completely wrong. We will see.

So, what would happen if these forecast results became reality? In all likelihood, there would probably be a second election in short order, because it would be very difficult for either Labour to form a coalition, and all but impossible for the Conservatives, even though they would have “won” in terms of seats (but not the popular vote).

Given Sinn Féin’s abstentionist policy, the magic number for a majority in the House of Commons is 322. A Labour-led coalition in such a scenario would require both the Liberal Democrats and the SNP to get 324 seats, giving the coalition a majority of two. A Conservative-led coalition would require all of the Liberal Democrats, UKIP and the DUP to get 340 seats, a majority of 18. However, it is almost impossible to imagine Nigel Dodds, Nigel Farage, and Nick Clegg (or whomever gets the gig of Lib Dem leader) all agreeing to be in a coalition with each other. It would be weird.

Both David Cameron and Ed Miliband will be praying that the minor party surge is only a flash-in-the-pan, and that in the election itself people will return to their “natural” tribe and vote for one of the big two. Otherwise the next parliament could end up being an ungovernable mess. It is very possible to envisage a scenario where Nigel Dodds or (probable SNP Westminster party leader) Alex Salmond wield real power, as their Westminster blocs may be needed to prop up a government. One can imagine that both men are busy drafting their wish lists for what additional resources they would like to see being dispatched to Northern Ireland and Scotland in the event that they find themselves at the centre of national attention in May. They could call it the coalition of the shilling.

A qualified accountant and data analyst, interested in politics, economics and data. Twitter: @peterdonaghy