General Election Prediction

May 2015 Election Results: Conservative Majority of 12

Party2010 Votes2010 Seats2015 Votes2015 Seats
CON 37.0%307 37.8%331
LAB 29.7%258 31.2%232
LIB 23.6%57 8.1%8
UKIP 3.2%0 12.9%1
Green 1.0%1 3.8%1
SNP 1.7%6 4.9%56
PlaidC 0.6%3 0.6%3
Minor 3.4%0 0.8%0
N.Ire 18 18

Voter Migration Flows: post-election update

Many thanks to everyone who took the time to say how much they liked the voter migration flow graphic. Several people also suggested it would be even better with the real support figures, rather than the incorrect picture given by the pre-election polling.

Here is the revised graphic. Compared with the previous version, it has the correct support figures, which means that both the Conservatives and Labour are each one voter better off than in 2010. The Liberal Democrats are reduced to eight voters. There is a new flow of one voter from Labour to Conservative, and two fewer voters leaving the Conservatives for UKIP.

The UKIP group of 13 voters includes three voters who previously voted for other minor parties such as the BNP in 2010, but these voters are not shown on the diagram.

Although Liberal Democrat voters split two-to-one in favour of Labour over the Conservatives, the Conservatives gained twice as many seats (27) from the Lib Dems as Labour did (12).

15 May 2015

New Boundary Estimates: Conservative Majority of 50

There has been recent interest in the likely effect of new boundaries which may be brought in under this parliament. Electoral Calculus prepared a full set of notional implied results under the 600-seat "Sixth Periodic review" of boundaries which was conducted around 2013.

Although these boundaries were not used in 2015, they can still give a good approximation of the likely effect of the boundary changes. If we use the actual election result (adjusted slightly to compensate for model deficiencies) and feed it into the user-defined predictor, then we can see the effect of the boundaries.

Using these figures and the old boundaries gives CON 331, LAB 232, LIB 9, UKIP 1, Green 1, SNP 55, and Plaid 3, which is almost exactly correct. Then when we switch to the proposed 2013 boundaries we get

CONLABLIBSNPPlaidN.Ire
325202549316

This gives the Conservatives a majority of 50 seats, well ahead of their current majority of 12. This is equivalent of nearly another twenty seats for the Conservatives.

Without any change to legislation, the Sixth Review should restart this year for completion in 2018. It looks unlikely that the Conservative government would want to slow this process down.

13 May 2015

Prediction Error 2015

The final pre-poll Electoral Calculus prediction for the election was not accurate. Although the prediction correctly said that the Conservatives would be the largest party, it significantly underestimated the number of Conservative seats by about fifty seats.

Several users have naturally expressed their disappointment(*) in this error, and asked what its cause is.

The primary cause of this prediction error is error in the national opinion polls. The actual election showed that support figures were: Con 37.8%, Lab 31.2%, Lib 8.1%, UKIP 12.9% and Green 3.8%, which implies a Conservative lead over Labour of 6.6%. The average of the final polls from all the pollsters only gave a Conservative lead of 0.2% This poll error of about six per cent was the main driver of the prediction error.

We can check this by feeding the actual support levels into the Electoral Calculus model, using also the actual Scottish support levels of: Con 14.9%, Lab 24.3%, Lib 7.5%, UKIP 1.6%, Green 1.3% and SNP 50.0%. Then we get the following seat prediction (actual seats in brackets): Con 322 (331), Lab 240 (232), SNP 55 (56), Lib 10 (8), UKIP 1 (1), Green 1 (1). This gives the big two parties correct to within ten seats and all the smaller parties correct to within one or two seats.

So the prediction would have been relatively accurate if the polling inputs themselves had been more accurate.

This polling error is similar in size to the polling error in 1992 when the pollsters had an error of 9.7% in the lead of the Conservatives over Labour.

The pollsters will likely conduct their own investigation into this matter, and Electoral Calculus will follow those investigations closely to try to reduce these problems in the future.

UPDATE: The British Polling Council has announced an independent enquiry under Prof Patrick Sturgis into the causes of the polls' apparent bias.


* An example comment received on Friday 8 May was: "given the events of yesterday I would now have difficulty believing you if your poll indicated tomorrow was Saturday".