I've been reading all the write-ups I can find on the possible reason for the differences between the state polls which seem to show a clear advantage for Obama and the National Polls, which seem to show a lot tighter race.
Jonathan Bernstein, Nate Silver, Sean Trende, and Sam Wang are all good reads.
Somewhat of a consensus seems to be that historically state polls have been more accurate than national polls (except for Trende), and that ultimately the state polls are a better measure of where things are at and it's likely that the national polls are collectively off.
Anyway, my possible explanation...
If you look nationally at registered voters vs likely voters, the registered voters are stronger for Obama than likely voters. I do understand why it's the case that likely voter screens tend to shift points toward Republicans.
Perhaps, though, the state polls do a lot better job of identifying likely voters, especially the battleground states. If the Presidential race is a product, like Pepsi, and you advertise the hell out of it in Ohio, Florida, and Virginia, aren't you going to have better polling numbers on whether people are likely to try Pepsi in those states vs nationally where it might be harder to determine because so many more people are less engaged.
I've got to imagine in Ohio a lot more people understand a) there's an election going on b) their preference and c) most importantly, whether or not they're going to vote.
Common sense seems to say that a likely voter screen in Ohio, where spending on TV, Mail, Phone, and actual personal contact is in the tens of millions, is going to be a lot better than a voter screen of Illinois where only people already engaged have any real exposure to this race.
If someone says they're likely to vote, or not, in Ohio, I think I believe them more than I would someone in Illinois.
So, I think it's harder to determine likely voters nationally because non battleground states have people that are generally less engaged than the average when compared to battleground states.
One thing that shows this quite plainly is to compare polls from battleground states to non-battleground states from previous elections (as opposed to comparing national to state polls like others have done). When you do that, you see that non-battleground state polls are a lot less accurate than battleground state polls.
Just a thought.