Forecasting Local Election Seat Gains/Losses 2018

by Stephen Fisher

My forecasting model for seat gains/losses at local elections has previously been a simple model based on change in party support in the polls. While the historical data show that changes in the percentage of the council seats that a party wins is reasonably strongly correlated with changes in that party’s poll share, that basis for forecasting this year would not work.

The Conservatives are up by 8 points and Labour up by 6 points in the polls since just before the 2014 local elections, when most the seats up for election this week were last fought. As in last year’s general election, support for both main parties has increased largely because of a big drop in UKIP support. Forecasting the parties separately based on changes in their respective poll shares would produce misleading predictions of big gains for both main parties. In first-past-the-post elections it is typically relative not absolute performance that matters for seat outcomes. Last year, despite winning more votes the Conservatives lost their parliamentary majority because the Labour vote share went up by more than did that for the Conservatives.

For this reason, my forecasting models this year are based on changes in the gaps between polls shares.  For the Conservatives, who have traditionally faced many contests with the Liberal Democrats, their leads over both Labour and the Liberal Democrats matter. For Labour, the model is primarily based on the Labour lead over the Conservatives. Meanwhile, for the Liberal Democrats, their changing opinion poll performance relative to the Conservatives, but not Labour, has historically been correlated with headline local election seat changes.

As usual, the models exclude local elections on the same day as a general election and also 1992 when the local elections were held just a month after. The models also reflect the fact that the Conservatives have typically done less well and Labour slightly better in local elections than can be accounted for by opinion polls alone since the Tories came to power in 2010.

The following table shows my model based forecast, a 95% probability prediction range, and the Colin Rallings and Michael Thrasher estimates from their analysis of local by-elections.

 2018 Forecast Range Rallings & Thrasher
Con +8 -448 to +465 -75
Lab +131 -126 to +388 +200
LD -83 -335 to +169 +12

Despite the stronger rationale for looking at relative rather than absolute change in party support, the historical data show weaker correlations with the former than the latter. As a result, the range of possibilities with 95% probability of happening has widened somewhat by comparison with my previous forecasting models. This is especially so with the Conservative forecast. There the range is so wide the model is essentially saying that the Tories could win many more or many fewer seats than they did in 2014.

While the uncertainty range might look ridiculously large it does reasonably reflect the very weakness of the relationship between swing in the polls and local election seats changes. Just as this year, in both 1994 and 1999 there was a small swing to the Conservatives in the polls compared with 4 years previously. In 1994 the Conservatives’ net losses amounted to 10 per cent of the seats up for election that year. In 1999, the Tory seat tally as a percentage of the total increased by 10 points. With 4410 seats being contested this year, the range forecast this year similarly reflects the possibility that the Conservative tally might be up or down by 10 percentage points, that is 440 or so seats.

In this context, the central forecast of +8 for the Conservatives may as well be 0. Despite the polls showing a small swing to the Conservatives from Labour and a modest 4.5 Liberal Democrat to Conservative swing since 2014, the model predicts no real change for the Conservatives because they are in government. Governments typically suffer council seat losses in addition to those that can be accounted for by the lower poll ratings they also usually experience.

Conversely, because they are in opposition Labour are more likely to make gains than losses despite the swing to the Tories in the polls.

The Liberal Democrats are likely to suffer net losses in council seats because they are a point down in the polls while the Conservatives have gone up since 2014. By contrast, Rallings and Thrasher predict that the Liberal Democrats will make gains because they have been doing well in local by-elections and so expected to be up 7 points in their National Equivalent Vote (NEV). There was a similar discrepancy in the direction of our forecasts last year. On that occasion, my model rightly predicted Liberal Democrat losses. In truth, it was wrongly based on the party being down in the polls. In fact, the Liberal Democrats’ Projected National Share (PNS) and NEV for 2017 were up from the comparable figures for 2013. It was not so much their own failings to win new votes but the greater rise in the Conservative vote that cost the Liberal Democrats council seats last year.

Colin Rallings and Michael Thrasher also expect UKIP to lose nearly, if not actually, all 125 of the seats they won in 2014 that are up for election this year. I do not have a statistical forecasting model for UKIP seat changes, but drastic losses seem very likely given they are down from 15 points in 2014 to just 3.5 points in the average of recent polls. Moreover, they are fielding many fewer candidates and not even defending some of the seats they won last time.

UKIP’s gains in 2014 came mainly at the Tories’ expense, but that does not mean that any UKIP losses this week will necessarily translate into Conservative gains. Even if the Conservatives pick up most of the seats UKIP lose, a potentially bigger question is where the former UKIP vote goes in places where they did well last time but did not win. My analysis of the 2013 and 2014 local elections argued that the rise of the UKIP vote came almost equally at the expense of both main parties. Similarly, there was little sign of differential damage to the two main parties from the constituency variation in the UKIP surge in 2015. While most of the former UKIP voters from the 2015 general election went to the Conservatives in 2017, a substantial minority went to Labour. As a result the Conservatives benefitted, but not entirely, from differential swing in areas of former UKIP strength. It could well be that this year the Conservatives again benefit disproportionately from the collapse of the UKIP vote, but maybe not by much.

The movement of voters from UKIP to the Conservatives last year was the most striking part of a broader trend for former Leave voters to switch to the Tories. Conversely, Remain voters have become more likely to vote Labour. The net effects of this trend are reflected in the vote intention polls, but geographical implications for this year’s local elections are not part of my model.

Some 150 of the 326 councils in England have elections this week. On average there is a small difference in the 2016 Leave vote between the places that do (51%) and do not (53%) have elections. But among the places that have local elections this week, there is a big difference in the Leave vote between London (40%) and elsewhere (55%). As John Curtice has wisely pointed out, if, as some polls suggest, the swing to Labour is greater than the national average in the Remain voting capital, then it must be lower than average in Leave voting provincial England. Gains for Labour in London might be somewhat offset by gains for the Conservatives elsewhere.

However, there are more seats being contested this week in places where the Remain vote was higher in 2016 than in the council areas where it was relatively low. Most strikingly London accounts for 1830, 41%, of the 4407 seats up. The over-representation of London is partly due to the city having all-out elections while most of the smaller district councils in Leave leaning areas have just one-third of their council seats up for election. Relative to the number of seats being contested, the average Remain vote across councils with elections this week is 52%. On this basis the Conservatives might well suffer net losses and Labour might comfortably exceed the forecasts. Just as last year’s English shire council elections proved especially fertile ground for the Tories, so the disproportionately urban geography this year should advantage Labour.

Performance of the forecasts for England last year

The table below shows the forecasts from last year and the actual outcome. In 2013 and 2014 the Rallings and Thrasher estimates were better than my first two attempts. And so they should be, given their model applies predicted share changes at the seat level and not the national level. Last year my predictions were better. This may have been, as I suggested at the time, because I could take into account the most recent polls that showed a big boost for Conservatives well after the last of the local by-elections that Rallings and Thrasher had to work with. Also, as noted above, my Liberal Democrat forecast was right for the wrong reasons.

2017 Forecast Range Rallings & Thrasher Actual
Con +430 +190 to +670 +115 +561
Lab -315 -555 to -75 -75 -393
LD -30 -265 to +210 +85 -49

 Acknowledgements: Thanks to David Cowling, Colin Rallings, and Michael Thrasher for help with the data.

Advertisements

3 thoughts on “Forecasting Local Election Seat Gains/Losses 2018”

  1. Am I correct in believing that all poll forecasts have been adjusted to take into account that those who say they will vote Labour are now more likely to vote than in previous polls before last year’s GE. This change in intention being the reason why so many got Labour’s predicted share wrong last year.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

w

Connecting to %s