Pre-Election Polling in Britain, 1950-1997
In: Electoral Studies, Band 22, Heft 1, S. 1-20
Eve-of-election polls in GB have a reasonably good track record in predicting the outcome of the ensuing general election. Over the 1945-1997 period, the average error in estimating parties' vote share was roughly 2 percentage points. Where UK pollsters -- unsurprisingly -- have encountered the most difficulty has been in estimating the size of the vote share gap between the first- & second-placed parties. The average error in estimating this gap between 1945 & 1997 was 3.2 points. This is clearly enough, in a close contest, significantly to impair the pollsters' ability to identify the winning party -- though the paper shows that the closeness of the contest does not affect the accuracy of the pollsters' estimates. The paper considers the extent to which British voters, in response to published opinion polls, have engaged in either bandwagoning or anti-bandwagoning behavior. It finds that anti-bandwagoning is far more common in GB than bandwagoning. The paper also explores the extent to which statistical models of government support might improve politicians' abilities to forecast party support in the run-ups to elections. It finds that, given the information available to Prime Ministers at the time they make decisions to call elections, a long-term autoregressive model provides more accurate forecasts of government support than simple forward projections of support in the months preceding elections. 7 Tables, 2 Figures, 13 References. Adapted from the source document.