|Party||2017 Votes||2017 Seats||Pred Votes||Pred Seats|
Prediction based on opinion polls from 07 Aug 2017 to 28 Aug 2017, sampling 6,404 people.
|Lab choice of Lib/Nat|
|Con choice of Lib/Nat|
|Nat choice of Con/Lab|
|No overall control|
The future is never certain. But using our advanced modelling techniques, we can estimate the probability of the various possible outcomes at the next general election. ('Nat' means SNP+PlaidC)
The last election was another bad one for pollsters. The overall poll error in the Conservative lead was over 4pc, which incorrectly suggested that the Conservatives would win a sizeable majority. That was the second time in a row that the polls got the election result wrong.
But there was one notable exception. Late in the campaign, YouGov pioneered a new method of analysing polls which gave a very different result. The new method gave a final campaign prediction of 304 seats for the Conservatives. This was pretty close to the actual result of 318 seats, and correctly predicted that the Conservatives would not have a majority. YouGov called their method "multi-level regression and post-stratification", which baffled many people.
But what actually is this method? And does it really work, or were the YouGov results just accidentally right by lucky chance? The answers are that this method is similar to those successfully employed in other contexts and that it is very suitable to analysing polling data as well. It works pretty well in practice, not just for the most recent election but also for earlier British elections. Here are the details.
Posted 12 September 2017
Press reports in early September 2017 (see Times or Guardian) suggest that the government is unlikely to push forward with the reduction in seats to 600. It is suggested that the Boundary Commissions will be asked to repeat their reviews but keeping the total number of seats at 650.
That will require primary legislation, since the 600 seats are set down in the Parliamentary Voting System and Constituencies Act 2011. If new legislation is brought forward it could usefully also change the electorate tolerances from 5pc to 10pc, as this would make seats more homogeneous and reduce the need to cross traditional boundaries.
A repeat of the reviews could not be done very quickly, since all the proposed boundaries would have to be redrawn and the consultation processes re-run on the new boundaries.
Taken together, this greatly increases the chances that the next general election will be run under the existing boundaries.
See the full analysis of New Boundaries 2018.
Posted 11 September 2017
|Election Result 1955|
The new and improved Electoral Timeline map gives an interactive journey through all British elections from 1955 onwards.
The map shows how each area of the country voted along with political commentary about the election including its context, the campaigns and key statistics such as seats won and swing.
Repeated themes include new Prime Ministers trying to secure mandates, oppositions re-inventing themselves as fresh modernisers, sterling crises, miners' strikes, economic problems, and questions about Europe.
Posted 16 August 2017
|London Wards: Party winner in 2015 and 2017|
The new "Data Map" is a fully browsable map which can display a dozen indicators from the Electoral Calculus geographical database. The key indicators include:
Uniquely these indicators are available at four geographic levels:
The map allows you to explore the political and demographic make-up of the entire country, regions or individual localities, and opens up a whole new way to see political data.
The Data Map is available here now.
Posted 25 May 2017, Revised 26 June 2017
|Swing from Lab to Con against 'Leave' share, by seat|
In a new analysis, we look at the swing voters who decided the last election. Using regression techniques, the swing voters who chose the Conservatives are revealed to be 'Leave'-supporting, nationalist, older and working class, whilst voters who swung to Labour were 'Remain'-supporting, internationalist, younger and middle class.
Siginficantly, economic issues did not seem to be a key priority for many of these voters. This was definitely the "Brexit" election, where other issues were pushed to the side.
See the full analysis here.
Posted 26 June 2017
The traditional Electoral Calculus review of the 2017 prediction is now available. The prediction was overall of poor quality, over-predicting the Conservatives by forty seats.
Post-election analysis shows that the largest cause of this was polling error in the pre-election campaign polls, but other factors were also at play including strong tactical voting and idiosyncratic local factors.
See the full analysis here.
Posted 22 June 2017
Last week's surprise election result was bad news for the Conservatives as they lost their parliamentary majority and now have to rely on the support of the Democratic Unionists. But many Conservatives thought that their position would be improved by the new consituency boundaries. Originally proposed in 2010, the new seats are scheduled to come into effect in 2018. And using the 2015 voting patterns, the changes would have increased the Conservative majority by about 26 seats.
But the situation is very different now. Using the new 2017 general election results, Electoral Calculus has re-run the new boundary calculations to calculate the implied 2017 result using the initial new boundary proposals from the Boundary Commissions.
These calculations are based on the actual 2017 general election result, plus recent local election results up to 2015, census demographic data, and the initial proposals from the Boundary Commissions. They are not dependent on opinion poll data, and are not subject to opinion poll error. They are provisional estimates, and subject to change as the Boundary Commissions publish revised recommendations, and to using more recent local election data.
But these results will bring little cheer to the Conservatives.
With the new implied results the Conservatives are still short of a majority. They change from being eight seats short to being three seats short, winning 298 out of the new 600 seats. Their DUP partners themselves drop two seats down to seven, so the combined Conservative/DUP grouping increases its Commons majority very slightly from six to ten seats.
election result 2017
|Implied result at 2017|
under new boundaries
Although many Labour seats in Wales and the North disappear, the Conservatives lose several now-marginal seats in London and eastern England. Their recent gains in Scotland also leave them vulnerable to the decreasing number of Scottish seats.
The Boundary Review is not going to play the role of the seventh cavalry in rescuing the Conservatives from their electoral problems.
Posted 12 June 2017
This headline in the Economist seems to sum up what happened on Thursday. For yet another time, the pre-election campaign polls contained considerable error. The final average of the campaign polls showed an average Conservative lead over Labour of 6.8pc, whereas the actual difference in vote share was only 2.5pc.
That gives a poll error of over 4pc, which is only partially better than the poll error in 2015 of around 6pc. Though this error over-stated the Conservatives rather than under-stating them.
And, as most people know, if you put the wrong inputs into the prediction model, then the prediction will not be accurate. So the Electoral Calculus model predicted the Conservatives would win 358 seats, based on the incorrect poll data. In the end that was forty seats too high, and the Conservatives ended up with 318 seats and will be a minority government.
The polling data also seems to have confused the attempts to quantify the effect of the EU referendum. The data suggested that Remain voters would stick with their 2015 party, but the pattern of seat results suggest that seats in Remain areas saw significant defections away from the Conservatives.
Allowing for both of these problems, the basic model, given the correct inputs would have predicted the result to within about ten seats. (See the prediction.) This is similar model error to that seen in previous years.
In the final post on election day, the issue of polling error was flagged up and confidence bounds were given on the Electoral Calculus prediction. The actual result was contained within these bounds, with the exception of the Liberal Democrats who managed to gain four seats even while losing national vote share.
In the updated Electoral Battleground, it was shown that if the "low-lead" pollsters were correct (and they were closest to the right answer) then a Conservative minority government was a definite possibility with around a one-third chance of happening.
The main advance warning of the shock result came from YouGov's new methodology which uses models and demographics to predict voter behaviour. Electoral Calculus wrote approvingly of this methodology at the time (see below, 31 May 2017) and said that it could be the path of the future for market research. Congratulations to YouGov for getting it so right.
For the conventional polls, the winner of the most accurate pollster goes to Survation, who had three separate polls all showing a Conservative lead of just 1pc.
And the Exit Poll conducted for the TV broadcasters was, as usual, the most accurate. The Exit Poll is not only conducted on a fairly large sample, but it avoids the two main problems of the campaign polls which are uniform sampling and turnout. If someone has just come out of a polling station, then you know they have turned out.
It may be that this year's polling error was the mirror image of 2015. Last time, the pollsters asked respondents whether or not they intended to vote, and many young people over-confidently predicted that they would vote. To compensate, pollsters started disregarding what people said and used their own models to work out whether someone would vote. But the younger voters seem to have been so enthused by Jeremy Corbyn's Labour party that they voted in larger numbers than predicted. Truly "voters astonish pollsters". And us.
Posted 10 June 2017
it's worth remembering that all the predictions on Electoral Calculus are subject to error. The actual results will be different from the estimated predictions.
The two main sources of error are polling error and model error. The first of these happens when the opinion polls conducted by the polling organisations are inaccurate. This happened particularly in 2015 when the Conservative lead over Labour was mis-estimated by more than 6pc. The polls are a key input to the Electoral Calculus model, so if the polls are wrong then the output predictions will also be wrong.
The other error is model error. The model has been refined to be as accurate as possible, but it is still an approximation to behaviour of millions of individual voters across 632 seats in Britain. Even if the polls are completely correct, there will still be some mis-predictions due to model error. To give an idea of scale, the model error in 2015 was less than 10 seats.
This election campaign has been notable for the strong divergence in polling numbers from some of the major pollsters. At one stage, there was a large gap of more than 7pc between YouGov/Ipsos-MORI and ICM/ComRes in their estimates of the Conservative lead over Labour. This has narrowed with the final campaign polls, as both YouGov and Ipsos-MORI are reporting leads closer to the average.
The Electoral Calculus estimate of polling error is 3.5pc standard deviation for the major parties. This corresponds to a 90% significance level for the 2015 error amount. Model error is estimated at a further 0.5pc. Feeding these uncertainties into the Electoral Calculus model gives these ranges for the possible number of seats won at the election:
|Party||Low Seats||Pred Seats||High Seats|
Overall there is a 88pc chance of a Conservative victory, but the majority can range from wafer-thin up to 186 seats. The Labour party might lose seats, but also has a chance of gaining some. The Liberal Democrats are unlikely to gain any seats and look set to lose some of their current eight seats (or just possibly all). In Scotland there are a lot of closely-fought seats, so the SNP's range of seats running from 34 to 55 is relatively wide. In any event they should still be the largest party in Scotland.
Though the actual results will fall where they will.
Posted 8 June 2017
In an eye-catching Times story on 31 May, YouGov revealed a new model which predicts the Conservatives will only get 310 seats, leaving them short of a majority.
Not only is this prediction notably different from the predictions from Electoral Calculus and others, but the methodology used is very different from normal polls. Traditional polls simply aim to calculate the fraction of the overall population which supports each of the major parties. And they do this with the standard market research techniques of asking questions and adding up how many people gave each answer.
YouGov have tried something much more ambitious and modern. Using what they describe as multilevel regression and post-stratification analysis (which is reminiscent of the machine learning techniques of the big tech companies), they are trying to model how each individual voter in the country thinks. They need a big poll to do this (with a sample size of 50,000) plus regression against census demographics and British Election Study data.
This approach looks similar to that used by Electoral Calculus to calculate EU Referendum voting at the locality and ward level, as well as the other political measures as described in our Thirty Extreme Places in Britain article.
So Electoral Calculus salutes YouGov for their modern approach of combining polls and models to get richer and more insightful predictions. We will know next week whether their approach has got it exactly right this time. If it's right, then they will be justly celebrated. But even if it isn't, it is still the right thing to do and the method can be refined in future years to be more accurate. One day, maybe all polling will be like this.
Nonetheless, and for the record, the Electoral Calculus prediction is still that the Conservatives will get a sizeable majority.
Posted 31 May 2017