Earlier quoted context omitted.
FYI, I use the data from electoral-vote.com to turn it into a probability of winning the election at http://prespredict.com
I like that your site shows the 2008 history right below the 2012 (although it is distracting that the scale in time is different between them). I couldn't even find that comparison when I looked for it on Nate Silver's site. I'm curious about the methodology as well. Why didn't you choose more standard gaussian statistics for the differences, instead of the combination linear/step model? How much difference would it…
I prefer to have the flat region of 50% probability when the state poll difference is small, which the gaussian model doesn't have, to take into account (a) the various errors in the polling data and (b) the likelihood that people's minds can change by the time the election comes.
As to how much difference it would make, with the current model, Obama's probability is now at 81%, whereas with the Gaussian model it would be around 95%.