by Stephen Fisher

Local elections are supposed to be about local matters, but the electoral fortunes of councillors and would-be councillors depend very heavily on the popularity of their parties nationally. So the outcome of this year’s local elections depend on how the current standing of the parties compares with four years ago when the seats up for election this year were last contested.

In the run up to the 2012 local elections Labour were polling above 40% with a solid 9 point lead over the Conservatives, in part because of the so-called omnishambles budget. Meanwhile the Liberal Democrats had slumped in the polls after the formation of the coalition government with little over 11%.

According to the average of the most recent polls from six companies, Labour have dropped since 2012 to 35% now, only narrowly ahead of the Conservatives who are on 34%. The Liberal Democrats have slid even further to just 7%. On this basis alone we should expect both Labour and the Lib Dems to lose substantial numbers of council seats while the Conservatives should make gains from their 3-point recovery.

Of course one reason to doubt this is that polls might have now corrected a pro-Labour anti-Tory bias that may have existed in 2012 but only became visible at the general election. If that is the case then the Tories are effectively level-pegging. But Labour and the Liberal Democrats would still be down on their 2012 levels of popularity. Under these circumstances we should still expect to see the Conservatives making gains just off the back of decline in support for the other two main parties of English local government.

Another tendency visible in the varying numbers of council seats won by each party over time is a pattern of reversion to the mean. If a party does particularly well, given their levels of support in the polls at the time, then that success is likely to unwind a bit when those seats are contested again four year’s later. Similarly unusually large losses are somewhat recovered. 2012 was a great success for Labour. They went into the local elections with 27% of the seats up for election but ended with 44%. This 17-point rise was the largest for the party since 1981. What it means for this year is that Labour are likely to make even greater loses than might be expected just from the fall in their poll ratings alone. Conversely heavy loses for the coalition parties in 2012 should be somewhat recovered if the patterns of the past are maintained.

Above and beyond these factors, governments tend to be punished in local elections to the benefit of opposition parties. This pattern was clearly visible in 2012. Now the Conservatives are still in government and on this basis we should expect them to suffer further seat losses. The Liberal Democrats suffered particularly heavy losses in local elections while they were part of the coalition government; much more than could have been expected just from their poor poll ratings, even though they already reflected a lot of the dissatisfaction with Lib Dem coalition membership. A key question for these local elections is whether, as a result of no-longer being in government, the Liberal Democrats will recover some of their substantially lost ground even though their poll ratings are their lowest going into local elections since 1990.

The rise of UKIP is much more recent and so cannot be analysed in quite the same way. Despite commanding double-digit poll shares and local election vote shares since 2013, UKIP still has less than 3% of council seats across Britain. Given they are up from 8% in the polls in 2012 to 14% now they should make further gains. A similar rise for UKIP in the polls, from 2009 to 2013, was accompanied by a net increase of 139 councillors: just 6% of those up for election at the time. So a similar rate of success in translating poll shifts into council seats this time would mean gains of 165 seats. But there are questions about whether UKIP will put so much effort into local elections now as they did in 2013. Many of the party’s supporters are more focused on the EU membership referendum. The party may have less momentum and apparently lower media fascination after their failure to win more than one seat at the general election.

The implications of the UKIP rise for the three main parties of English local government are difficult to fathom. Doubtless UKIP have taken more votes from former Conservatives than from other parties. But all three have suffered so the comparative disadvantage for the Tories has been relatively muted. If this issue were a problem for my forecasting models we should see the Conservatives systematically underperforming the estimates in recent years. That is not the case.

Forecasts

The table below shows my forecasts from statistical models based on the factors discussed above, together with predictions from Colin Rallings and Michael Thrasher (available here). They also predict UKIP +40. I do not have a prediction for UKIP. Their predictions ought to be more accurate than mine since theirs are based on recent local by-election performance and use ward level data to estimate the consequences for seat outcomes from their projected National Equivalent Vote (NEV). My forecasting models are aggregate (GB) level only and based on just 27 years of local election and polling data. Patterns of party competition for local government have changed a lot over that time, jeopardising the exercise. Nonetheless I think it is interesting to see what can be gleaned from such models and how they compare with the Rallings and Thrasher predictions. This year I am surprised how similar the estimates are, especially for the Liberal Democrats given that they are projecting an increase in the NEV despite them being down in the polls since 2012.

2016	Forecast	Range	Rallings & Thrasher
Con	+19	-73 to +112	+50
Lab	-151	-276 to -26	-150
LD	+93	+54 to +132	+40

What numbers of seats should each party get?

The Rallings and Thrasher predictions are often misinterpreted as targets for the parties, or benchmarks, as if beating them constitutes good performance and falling short bad. Given their methodology their forecasts can only really be understood as telling us how the parties are likely to do if they replicate their levels of performance in recent local by-elections. But of course recent local by-elections reflect the popularity of the parties and track Westminster vote intention polls quite closely.

There is one sense in which the forecasts provide a guide as to how well the parties should do. They tell us how the parties should do if what we know about them, and how local elections work, already continue to hold. So if the Conservatives suffer substantial losses instead of making gains it does not mean that they are doing terribly badly for a party that has been in power for six years, just that they did worse than their recent polls and local by-election results suggest they should have done. But even this kind of interpretation depends on the quality of the models. Past predictions from both Rallings and Thrasher and especially myself have not been that accurate so they should not be taken very strictly.

The more profound questions about how well a party that has been governing for six years should do, or how well the principle opposition party should do a year after losing a second general election, are much more tricky. Still less is there much of a useful historical basis for thinking about the how the Liberal Democrats or UKIP should do in their current positions.

It has been argued, by Danny Finkelstein, that local elections tell us nothing we do not know already from the polls about the state of the parties nationally. But we are now all more painfully aware that polls can be wrong. Local election results proved one of the best methods for forecasting the 2015 (see here and here). Chris Prosser even managed to predict a Conservative win in 2015 from what were considered excellent results for Labour in 2012 (here and here).

It is admittedly hard to say precisely what local election outcomes indicate about the state of the parties, never mind the accuracy of polls or a far away general election. Nonetheless local elections are exercises in which millions of people cast their collective verdict on politicians mostly from national parties. Even though they matter much more for local government, we know from past experience that they local elections provide a guide as to popularity of the parties nationally. More than they should do.

Background and technical details

Details of my first efforts at local election forecasting in 2013 are here with the results and 2014 application discussed here. The figures for the 2014 elections with actual results are in the table below.

2014	Actual	Forecast	Range	Rallings & Thrasher
Con	-236	-130	-235 to -30	-220
Lab	+324	+130	+30 to +220	+490
LD	-310	-400	-570 to -230	-350

The Rallings and Thrasher forecasts come from their PSA briefing.

I was too busy with the general election last year to do local election results forecasting. Doubtless any forecast for locals then would have suffered from the same polling miss that troubled general election forecasts.

This year I have developed the local election results models further. They all still contain the core element of change in opinion polls since the last comparable round of local elections as the main predictor. I have now discovered that the models fit much better if general election years are discarded (but the data from them are still used for predictors of results in off-election years.) In addition I have added indicators for who was in government to the Conservative and Labour models, when previously they were in the Liberal Democrat model only. Also new to the Conservative and Labour models is the share of seats they won or lost in the previous round (i.e. a 4-year lagged dependent variable). There is some evidence for this variable for the Liberal Democrats, but since it was not statistically significant it was left out of the model. Finally the Labour model also includes a variable accounting for the fact that they do less well in the English shire counties election, controlling for the other factors.

The models now have R-squareds of 0.90, 0.93 and 0.83 for Con, Lab and LD respectively. The Root Mean Squared Errors (RMSE) are 3.5, 2.9 and 2.0 respectively. These indicators of model fit are much better than those for the original models. So they should perform better if I have not over-fitted the data. We shall see!

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