Each week I've been doing Monte Carlo simulations of the general election based on data from the Votemaster's www.electoral-vote.com. In addition to a simulation using a 4% margin of error for the polls, for the past few weeks I've been using a wider margin of error,  this week about 12.5%, taken from a regression of 2004 polling data errors compared to the final results. For each margin of error, I've run 10,000 simulated elections. This week's results:
4% Margin of Error
Obama wins 99.5%, averages 308.4 EV
McCain wins 0.5%, averages 229.6 EV
No electoral ties

12.5% Margin of Error
Obama wins 93.4%, averages 307.3 EV
McCain wins 6.2%, averages 230.7 EV
Electoral tie 0.4%

McCain is doing a little better than in the last two weeks because he leads by 7 in the most recent Florida poll, a Rasmussen poll from June 26th. It's not clear that this poll represents a genuine shift - the previous Rasmussen poll actually gave McCain a larger lead (by 8), whereas Obama led in the latest Florida polls from ARG and Quinnipiac. The Votemaster's algorithm (which I use) averages the latest poll with any other polls from other pollsters within 7 days of that most recent poll. Since the ARG and Quinnipiac polls were a little older than that, they're ignored in my analysis, which makes Florida swing from a toss-up to quite likely McCain. If McCain is taking a large lead in Florida, it's certainly good news for him.

Obama's good news for the week is taking a 5 point lead in Montana. Sure, it's only 3 electoral votes, but it's yet another formerly solid red state where Obama may well be competitive. This is the advantage of the 50-state strategy: the more states Obama might be able to pick up, the less dependent he is on winning previously key swing states like Florida, Ohio, or Pennsylvania.

Pollsters may well differ on Georgia: Insider Advantage gives McCain just a 2 point lead this week, and their prior poll in mid-June had McCain up 1. By contrast Rasmussen's last poll gives McCain a 10 point lead, the same lead he had in their poll in early June. I suspect that they're differing in how they weight the samples, and that the race is probably more like a 4-7 point difference that has remained rather stable, rather than swinging between a 10-point McCain lead and a virtual dead-heat. Georgia (and Montana) should be solid McCain country if he's going to win; that he's perhaps in trouble in both is certainly good news for Obama.

Louisiana may be in the same situation, but with different pollsters: here Rasmussen's two last polls were showing McCain up 9 and 11, wheras Opinion Research shows a 16 point McCain lead in their last two polls. Unlike Georgia, no pollster shows this a close race, so these fluctuations don't affect the simulations at all.

Most of the other changes this week favored Obama, although none had a
significant effect on the simulations. Obama increased already large
leads in New York, New Jersey, Connecticut, Massachusetts and Rhode Island. And while McCain still leads in Alabama, the last poll shows a 15 point lead, down from 20-30 point leads in earlier polls.

I'll add the usual caveats: we're still four months away from the elections, and much can and will change. McCain's chances of winning are probably more like 30% than 5% or less, but the current polling certainly still favors Obama. Despite currently losing Florida, he still projects to win by about 80 EVs.

McCain is shaking up his campaign staff, giving more responsibility to Karl Rove protege Steve Schmidt. Current polling data suggest some sort of change is needed.