Election forecasting, that is the using data such as party standings, leader ratings and economic indicators to predict the result of the next general election, is still in its infancy. The combination of general elections only every four years or so with the relatively recent innovation of regular detailed polling figures means that there has often been a shortage of data and election results with which to create and test models.
However, as each general election passes the volume of data accumulates and predictive models get more sophisticated. Of course, this begs the question about how well you can predict an election in advance based on statistical factors. What about events and the human touch?
In order to shed a bit of light on this question (and to have a bit of political speculating fun), Liberal Democrat Voice has arranged for a team of academics to create an exclusive series of election predictions.
The team – Richard Nadeau, Michael Lewis-Beck and Eric Belanger – published a detailed paper in the August 2009 edition of Journal of Elections, Public Opinion and Parties. For those interested in the analytical approach, their introduction explained:
First, a prediction equation is offered, based on a powerful empirical proxy variable for the election outcome itself. Second, an explanatory equation is offered, accounting theoretically for that proxy variable. This recursive system of equations is estimated, evaluated and found, by various diagnostics, to be extremely robust. Then, forecasts are rendered for multiple measures of UK election outcomes, in order to bring together the various measures that have appeared in the literature: government vote (and seats) share, opposition votes (and seats) share, government vote (and seats) lead.
Their predictive model works on a three-month lagged structure; i.e. their model uses current information and says “if a General Election were held in three months time, here’s what the result would be.” That is because according to their work, looking at previous general elections, the situation in terms of figures such as leader ratings and government satisfaction three months out from polling day has provided a reliable guide to what then happens on polling day.
Here is the current prediction:
Conservative lead of 10% with 322 seats (4 short of an overall majority)
The forecasters explain: “These forecasts are based on the Ipsos MORI polling data gathered mid-October (16-18). The key variables from which our votes and seats forecasts are derived are the average approval rate of the government and PM (27.5) and the approval rate of the leader of the opposition (49.0) measured three months before an hypothetical election.”
But with the most likely election date being May next year, there is yet time for those ratings to change before we get to three months out from the election and the point at which the prediction gets ‘frozen’.
A three point shift up in Cameron’s approval and down in the Government and Brown’s approval would produce in their model instead a Conservative lead of 13% and 342 Conservative seats (a majority of 34 seats). On the other hand, a shift of three points in favour of Brown and the Government and off Cameron’s approval would produce only a 7% Conservative lead and leave them 24 short of an overall majority.
How reliable are these predictions? The accuracy of the model tested out on past elections in the article quoted above is impressive, though of course creating a model which gets the past right is easier than getting the future right.
We hope to update the predictions as the general election nears. Assuming a May 2010 election, February’s prediction will be the one on which the model stakes its credibility.