We have another factor, over and above the competitive polling between Ras and Mason-Dixon and the others, namely the lag from national polls to state level polls (which always seems to be about a week).
At any rate, that's why I don't accept any one poll, but always look at realclearpolitics' average. There, McCain is one state short of winning (CO), and is just a couple points down in PA and MI in the cumulative polling.
Finally, we've debated the "Wilder/Bradley" effect, which I have dismissed for most states (but NOT Ohio). However, I'm thinking it might be real---because I was reminded that here in OH in 2006, Ken Blackwell was only down in the worst of the polls 12% and in the best, 5%, yet he lost by 25%! Now, that's some serious Bradley effect going on.
Rasmussen and Gallup sauce their polls depending on what they THINK the mood of the public is. Not exactly science. They assumed Barack should get a bump from the economic crisis so they change their dem/rep weighting to reflect that. The problem is that while they did that in their state polls, they now know that their assumptions were false which is why Gallup has McCain/Obama tied among registered voters. The only way for this to happen is for McCain to get a significant advantage the last few days with independents. And I’ll bet that’s exactly what happened.