Monday, March 22, 2010
Opinion Polls and the Election- November 2009 Best Predictor?
Counter intuitive? It certainly is, but the data is compelling. This method produces spot-on accuracy for 1966 and 2001, pretty damn close for 2005 and an average of 3% margin of error for all the elections after 1945 when polling was in its infancy; that's as good as most polls gathered by top pollsters using the most up to date methods.
The statistical problem, apart from nobody being able to predict the future of course, is, according to the author, that the number seats to be won by the minor parties is so 'unpredictable'. Assuming they will win the same as in 2005- 31- he proceeds to make his own prediction based on the polls of November 2009.
So what does it produce? Labour-291 seats; Conservative 282; Lib Dems 42. This would be a hung parliament of course but with Cameron with a majority of 9 if he can cobble together an alliance with the Lib Dems. My bet is, as I've said before, he would opt to govern as a minority government for some months before asking us again if we think he is the man to lead us.
Or... he might not get more seats than Dave and Whiteley's prediction proves wrong- 3% margin of error remember. Finally, the above does not take account either of how the minor parties might vote; e.g. Ulster Unioniusts are in a kind of 'alliance' with Dave and could provide the margin he'd need if Whiteley's predictions prove accurate.
In any case, we have FPTP, not PR, so you can't just take the percentages and throw it into a Uniform National Swing predictor. If you read the politicalbetting.com articles on it, such an act, even with the actual voting percentages, will give you an error of up to 50 seats.
I'd bet he could be right on the percentages, but on which polls? We range from Angus Reid to YouGov. And I'd bet against him on the seat predictions, as are the betting markets.
Links to this post: