Andrea Moro of the University of Minnesota and Sam Wang of Princeton appear to have independently implemented the same good idea. They make a simple illustrative models in which the only difference between polls and the actual election is sampling error. They simulate elections with the models and they calculate who wins. Here for Moro and Here for Wang.
Oddly they get different results. As of 2:06 Rome time October 15, Moro calculates the probability that Kerry will win at least 270 electoral votes as 41.828 % Wang calculates it as 27%. Why the difference ?
There is a methodological difference. Moro uses the most recent poll, Wang averages recent polls. This means that Wang has a smaller sampling error after averaging so his simulation is closer to assuming that the candidate who is ahead wins for sure.
I don't think this explains the difference. I think the explanation is that Moro uses data from 2.004k.com which does not report online polls by Zogby WSJ or the partisan polls of Strategic Vision (Republican). The Strategic vision polls are key, because their latest poll has Bush ahead by 8 % in Ohio.
Wang calculates a very high probability that the winner in Ohio will win the election.
Another meta meta point, the last tick in Moro's hourly simulations is bad for Kerry (who was briefly over 50%). This is the result of the most recent results from 2.004k.com which include a national poll with Kerry ahead by 5% and a poll with Bush ahead by 2% in Ohio. I take the modelling seriously enough that, just looking at those numbers, I thought "this is bad" meaning "this is bad for Kerry". For the actual outcome, Ohio matters, national popular vote does not matter.
Finally, I am not convinced that the probability that Bush will win is over 50% (an electoral college tie means the election goes to the house of representatives voting one state delegation one vote so Bush would win). The reason may be wishful thinking and Kos probaganda, but I don't think so. There are three factors missed by the polls. First a very large number of new Democrat registered voters who appear as unlikely voters to pollsters but who are likely to vote. Second the same organisation and energy shown by registration will probably mean Democrat get out the vote efforts are strong. Finally experts say undecided voters tend to decide in favor of the challenger.
It is very hard to explain this last point to a computer because experts are numerophobic. The undecided voter claim is made two ways. What you see is what you get, that is, the share of votes for incumbants is equal to the share of people who say they will vote for them in polls, that is, all undecided voters vote for the challenger. The other is the 60:40 rule which says that 60% of undecided voters vote for the challenger. That is a big difference. I assume the 60:40 rule is meant seriously. This is about 1% (20% of 5% undecided) more for Kerry than the polls show. Such a swing would make a huge difference. Wang calculates that a 1 point swing to Kerry implies his probability of winning is 60% not 27%.
On the other hand I have no idea about the Smith Rove Bin Laden effects. That is the effect of Sinclair broadcasting, new dirty tricks by Rove and/or a terrorist attack by Bin Laden. I hasten to add that I am not suggesting that Smith and Rove are morally equivalent to Bin Laden. Not at all. Nothing of the kind. Never said that.