The really curious thing about this year's US presidential election is that it looks set to defy all political forecasts. While the most respected political science models have predicted victory for Mitt Romney, polls have consistently suggested otherwise.
Political science and predictive models seldom receive much attention in the UK but they enjoy a strong tradition in the US. And they normally get it right. Most spectacularly, the University of Colorado’s forecasting model in 2000 successfully predicted that Al Gore would win the popular vote but lose the electoral college. Since its inception in 1980, the University of Colorado’s forecasting model has successfully predicted the results of the last eight US presidential elections.
Colorado’s model is based on forensic analysis of state-by-state factors, especially unemployment rates and real per capita income. Contrary to conventional wisdom, the model finds that there is no statistical significance in 'the location of a party’s national convention; the home state of the vice president; or the partisanship of state governors'. It really is ‘the economy, stupid’. Somewhat hearteningly, Kenneth Bickers, who has co-designed the model, says that 'gaffes, political commercials or day-to-day campaign tactics' are all subsumed by economic concerns.
Based on Colorado’s prediction, Obama will win only 218 electoral college votes, with swing states includes Pennsylvania, Ohio and Florida (all leaning Obama-wards according to polls) expected to give their votes to Romney. But the economic data upon which the findings rest will be five months out of date by 6 November – the model will soon by updated to take into account more recent economic data.
The experiences of Colorado’s forecasting model mean only Luddites would decry its worth. But, right now, polling numbers suggest the model will not extend its run of success to nine consecutive elections; indeed, defying all traditions, the odds on a Romney victory actually lengthened after his nomination-accepting speech at the Republican Convention - although we have yet to hear from Obama at his party's convention.
Prediction models like Colorado’s clearly have their place: for forecasting several months out, they should be considered more reliable than polling data. This election, however, has several particular circumstances, and the modellers accept these can always have an impact. Romney’s choice of Paul Ryan as running mate means the election has ceased to be a referendum on Obama’s first term and, increasingly, developed into a referendum on the Congress in which Ryan has been so prominent – and which has consistently polled worse approval ratings than Obama. It's also worth considering the increasing alienation of women from the Republicans; the superiority of Obama as a campaigner compared to Romney; and that crucial economic indicators have improved in the past two years.
It all points to an election that might just defy the models. But if Obama does win on 6 November, that will be no reason to add the forecasters to America's unemployment figures just yet.