Forecasting Local Election Net Seat Gains/Losses 2019

By Stephen Fisher.

Updated 23:00 on 1st May to include a poll-based forecast I did not previously have access to, and to clarify figures for the Rallings and Thrasher local by-election model based forecast.

My predictions for this week’s bumper crop of English council elections are given in Table 1 below, along with those of others. The changes are relative to (sometimes notional) results for 2015 when the seats were last fought.

Table 1. Forecasts for English local election net seat changes 2019

Forecast Range R&T (local by-election model) R&T (local vote intention poll) Hayward
Con -700 -90 to -1,300 -405 -1,100 -800
Lab +590 +200 to +990 +150 +840 +300
LD +350 -10 to +720 +400 +170 +500

R&T refers to Colin Rallings and Michael Thrasher. Their figures are, I’m told, in the hard copy version of their Sunday Times article. (They differ slightly from those on Sophy Ridge on Sunday and others being attributed to them on Twitter.) Their local by-election model takes the results of recent local by-elections to predict the National Equivalent Vote (NEV) and projects the implied NEV changes since 2015 at the ward level (perhaps using notional results where required.) The local-election vote-intention poll Rallings and Thrasher used was conducted by Opinium which had the following shares of the vote: Con 28, Lab 36, LD 10, UKIP 9. That poll-based forecast is also mentioned in this excellent blog on the background to the local elections, which includes great data visualisation.

Lord Robert Hayward is a Conservative peer who works for the party on various psephological issues. I take it that his predictions reflect his expectations given his information rather than an exercise in Tory expectations management. I don’t know of any other predictions for headline gains/losses other than those described as personal guesses.

My predictions align with the others in anticipating substantial Tory losses and gains for both Labour and the Liberal Democrats. There are differences in scale but the main qualitative differences between the three forecasts is on who benefits most. Both Rallings and Thrasher local-by election model, and Hayward expect the Liberal Democrats to make more gains than Labour, whereas my model and the local-election vote-intention model predict bigger Labour gains.

My predictions are derived from regression analyses of historical patterns in the net changes in seats at local elections (compared with four years previously) with corresponding changes in Westminster vote intention opinion polls, government status and prior seat changes as key predictors. So my predictions can be thought of as expectations for party performance given what we know from the recent opinion polls and the historical record of local elections.

The Liberal Democrats have made no advance in the polls from the 9 points there were on in 2015. Nor have Labour, who are again on 33 per cent in the polls. However, the Tories are down from 34 to 29 points in the average of the polls in April (taking only the most recent poll from each pollster and considering only polls from British Polling Council members.) So the polls suggest the swing from the Conservatives to Labour is the same 2.5 points as that from the Conservatives to the Liberal Democrats since 2015. However, based on the historical experience of local elections, the Labour seat tally is more sensitive to the Con-Lab swing than is the Liberal Democrat tally to Con-LD swing.

Whether or not the swings have been measured appropriately (or are sensible indicators of change in popularity of the parties) is a big source of potential error this year. The 2015 local elections were on the same day as the general election, and the polls were famously wrong on that occasion. The polls made different mistakes in 2017 and so perhaps it is not appropriate to simply take the change in the polls from four years ago as a predictor. Maybe I should have used the actual GB share of the vote from the general election. Had I done so the implied swing from the Conservatives to Labour would be a substantial 5 points, resulting in a prediction of even bigger Tory losses and Labour gains.

However, a principal reason why the Conservatives are doing so badly in recent polls is the very recent advent of the Brexit Party, which, despite doing well in the polls, is not competing in the local elections. Meanwhile UKIP have taken some of the Conservative poll share over the last year, but they are fielding many fewer candidates this year than they did in 2015. If recent polls are too pessimistic for the Tories, at as they pertain to local election performance, that will be compensated for in the model somewhat by the fact that the 2015 polls underestimated the Conservatives. That, along with a preference for consistency, is why I’ve continued to use the simple method of taking the change in the polls on 4 years previously for the forecasting model.

Another reason the models are relatively more optimistic than other forecasts for Labour than the Liberal Democrats is the historical tendency for Labour to do well in local elections under periods of Conservative government. The same is somewhat true for the Lib Dems but to a lesser extent. The Liberal Democrats lost many seats while in coalition and have failed to make a substantial recovery since then. That may change this year, but my model now includes a factor that reflects the relatively weak Lib Dem results since 2016 and so expects them to continue a more lacklustre performance, especially by contrast with the 1990s. Labour’s failure to make the kinds of local election seat gains since 2010 that they made in the 1990s seems to be in line with their failure to build a substantial lead over the Conservatives in the polls.

The point estimates for my forecast do not sum to zero. Each party has a separate regression model and so this is an unsurprising possibility on a technical level. Somewhat by chance the prediction of 250 net gains for the three main parties, while probably an over-estimate, is likely to be helped by the decline of UKIP since 2015. That party is fielding many fewer candidates this year than they did then and they are polling at half the level they were in 2015. The party is expected to lose the vast majority of the notional 176 seats they are defending.

Yet again the prediction intervals (ranges) for my forecasts are extremely large. This is because there is inherently a lot of noise in a model like mine. For instance, as I mentioned last 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.”

Usually the huge prediction intervals encompass Rallings and Thrasher’s predictions, but not so for Labour this year. That suggests my model has over-estimated Labour’s gains. Since my model operates only at the macro level, Rallings and Thrasher’s ought to work much better. However, mine does have the advantage of using more recent polls compared with local by-elections, and there have been big changes in public opinion over the last month.

One reason to believe my model is not too bullish for Labour is the history of this year’s election cycle, as shown in Table 2 below. Labour did not just make losses last time, but, apart from in 2011, at every election for this cycle since 1995, when John Major was prime minister and Tony Blair was riding high in the polls. The cumulative loss of seats for Labour for these councils since 1995 provides some reason to think that there is considerable room for recovery.

Table 2. Forecast and historical net seat changes for the 2019 cycle of English local elections

2015-19 Forecast 2011-15 2007-11 2003-7 1999-2003 1995-9
Con -700 +541 +86 +911 +562 +1293
Lab +590 -203 +857 -505 -812 -967
LD +350 -411 -748 +246 +173 -119

You really do have to go back to 1995 to find a set of local elections in this cycle at which the Conservatives suffered net losses. This is extraordinary given they were in government with austerity policies in 2011 and were fighting for re-election in 2015. The 2015 result is not surprising in light of the general election result, but it reminds us that the 2015 general election was remarkable not just for the failure of the opinion polls but more importantly because the incumbent government managed to increase both their vote share and lead after a full term in office.

It would be extremely impressive if the Conservatives managed to continue their winning streak. Given the travails of the government and their failure to deliver Brexit on time, or at all so far, it seems all but inevitable that the Tories will suffer losses. Given their cumulative gains over the last 23 years, they have a lot to lose. Net losses of over 1000 do not seem far-fetched given they are defending 4900 seats.

Past performance of the predictions

 Table 3 shows the forecasts for last year together with the actual net seat changes. In 2017 my forecast was, unusually, more accurate for all three parties than Rallings and Thrasher’s predictions. In 2018 mine was better for the Labour, just slightly worse for the Conservatives and much worse for the Liberal Democrats. While my forecast for the Tories was so close to zero it can be forgiven for having the wrong sign, predicting Lib Dem losses when they made gains is an important problem. My model failed to anticipate that the Lib Dems would make gains in (Remain leaning) Con-LD contests despite a modest overall national swing in the other direction (from LD to Con), visible in the polls, NEV and BBC Projected National Share of the vote last year. The same pattern might occur this year, in which case my model is likely to under-estimate Tory losses and Lib Dem gains.

Table 3. Forecasts and actual net seat changes in 2018

  Actual Forecast Range Rallings & Thrasher
Con -35 +8 -448 to +465 -75
Lab +79 +131 -126 to +388 +200
LD +75 -83 -335 to +169 +12

There is more discussion of this pattern of results in 2018, and the more general tendency for the Conservatives to do better and Labour to do worse in Leave areas, in my blog posts for Prospect and Deltapoll. That pattern, with similar consequences for seats, is likely to transpire again this year.

Discussions of past local election forecasts and other aspects of local election analysis can be found by following the “Local Elections” archive link in the category list on the right hand side of this site.

Acknowledgements: Thanks to David Cowling, Colin Rallings, Michael Thrasher, the BBC and House of Commons Library for help with the data.

Leave a Reply

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

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

Facebook photo

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

Connecting to %s