Original URL: https://www.theregister.com/2010/04/14/elections_stats_matters/
Election algorithm gives Tories sunnier outlook
Still short of a majority though
Posted in Software, 14th April 2010 11:02 GMT
As the election campaign hots up, could a model built to demonstrate the benefits of the Python programming language hold the key to a successful election forecast?
The Tories almost certainly hope so, as one feature of VIPA - Voting Intentions Predictive Analysis – is that it consistently predicts a better outcome for that party than other electoral predictors available online. Unfortunately for them, even the more optimistic forecasts set out by this model suggest they will end up just 5,000 votes short of forming an administration on May 7.
VIPA is built using the Resolver One spreadsheet-style desktop application.
Analysts can use this software as a simple means to generate Python code, a powerful and open-source alternative to Visual Basic, which is instantly executed to display results – or they can add to this generated code by inserting their own Python instructions at different points in the program.
VIPA is the brainchild of Resolver Systems director Robert Smithson and Giles Thomas. Smithson’s choice of politics as the focus for Resolver’s demo system is that much less surprising when one realises he is also the son of Mike Smithson, creator of the PoliticalBetting site for those interested in insightful political analysis and gambling on possible political outcomes.
The main difference between the VIPA model and other election predictor schemes is that its forecast rejects the concept of a Unified National Swing (UNS), which is used by most predictors in one form or another, and assumes the same swing in votes from party to party in every constituency.
Instead, VIPA factors in not just the swing to or from a particular party, it also uses data drawn from recent Guardian / ICM polls (pdf) to determine the precise shift likely between each party. Since the composition of votes at the last election varied by constituency, this leads to some very different predictions when contrasted with the UNS approach.
Most significantly, VIPA has tended to produce slightly better forecasts for the Tories. Thus, as recent polls have shown the Tory lead narrowing, two of the main online forecasters have shown the Tories remaining stuck in second place when it comes to seats.
A recent ICM poll predicts the Conservatives will receive 36.5 per cent of the national vote, Labour to get 31.9 per cent and Lib Dems to receive just 20.6 per cent.
According to UK Polling Report, this is disastrous for the Tories, putting them on 273 seats, leaving Labour on 291, and giving the Lib Dems 54 seats.
The result from Electoral Calculus is little better putting the Tories at 273, Labour 287, and Lib Dems on 57.
With the same figures, VIPA gives the Tories 310 seats. Labour 270 seats and Liberal Democrats 41.
The end result leaves the Tories just 16 constituencies short of an overall majority – or scrabbling around for a further 5,357 votes in the most marginal constituencies, including Morecombe, Warrington South and Leeds North-West.
Is this bad news for David Cameron? Possibly not. We spoke to Robert Smithson, who expressed rather more confidence in his predictive algorithm than in the accuracy of current polls. He believes that Labour support is overstated, and believes the true gap between Labour and Tories is of the order of 10 points.
Ever a betting man, Smithson added: "If someone would give me decent odds, I’d be prepared to venture a wager on Lib Dems out-polling Labour." ®