Another day, another raft of opinion polls — but how to interpret them? We know that applying a uniform national swing is unsatisfactory, as it takes no account of concentrated canvassing in marginals, regional issues, or the incumbency factor, likely to help popular sitting MPs regardless of the relative unpopularity of their party.
Yet, only the very occasional survey polls opinion on a constituency-by-constituency basis, and even then only in key marginals.
In an attempt to make more sense of national numbers as they apply across the country’s 650 constituencies, Resolver Systems has developed a forecasting model that looks to make more of “where votes come from”.
For example, Liberal Democrat voters in 2005 who say they are switching to another party are much more likely to vote for the Conservatives than they are not to vote at all. However, people who voted for parties that were not one of the “big three” are much more likely to stay home than they are to vote for the Conservatives. (The modellng is more complex than that, but that’s the gist.)
16 seats that could deprive Cameron of his majority
Applying this method to the most recent Guardian/ICM poll (crucially ICM is one of the few polling firms to ask a question about past voting behaviour), the predicted outcome of the election would leave the Tories up by 100 seats but still 16 short of an overall majority.
The constituencies that would have deprived the Tories of power are listed above. The tiny majorities in all 16 amount to 5,357 — in other words, the number of votes between a hung parliament and an overall majority.
It’s worth noting that Populus has begun asking the “previous voting” question and Daniel Finkelstein and his Fink Tank team at the Times are attempting to do something similar. Yesterday, based on the latest Times/Populus poll, they predicted the Tories would be seven seats short of power.
We’ll be returning to the Resolver numbers between now and election day.
Follow the New Statesman team on Facebook.
Content from our partners
Related
