Based on pretty much the same opinion polls, how can UKPolling report predict the Conservatives will be 52 seats short of an overall majority, while FiveThirtyEight predict the shortfall will be only 18?

My colleague Alex Barker on the Westminster Blog sums up the arguments very neatly here, and if you want to read the full threads of the political nerds at war, all the links are here on the FiveThirtyEight site. I will just point to the evidence for both sides and then add some observations about the 2010 exit poll, which should give a pretty good prediction of the result at 10pm tomorrow.

Although there are quite a few differences between the models the big difference is that traditional UK predictions on seats are based on a uniform national swing, while the new kids on the block suggest a proportionate loss model is best. Uniform swing assumes that the national percentage point change in votes for a particular party will be replicated in every constituency. Nate Silver at FiveThrityEight argues that it is better to assume Labour loses the same proportion of its vote in each constituency, with the result that the percentage point losses are greater where Labour scored well in 2005.

What evidence exists? Read more