By now we have so much state-level and national-level polling data that the divergence between them can no longer be dismissed as probable sampling error. I don't think Sean Trende has it nailed, but he is wrestling with the right problem:
Sean Trende: What's Behind the State-National Poll Divergence?: [T]e national surveys point to a Romney win, while the state polls collectively point to an Obama win. Both can’t be correct. The RCP Average currently has Mitt Romney up by 0.8 points… a Republican enjoying a one-point lead nationally should expect a three-to-four-point lead in Florida, a two-to-three-point lead in Ohio, and a tie in Iowa. Instead we see Romney ahead by roughly one point in Florida, and down by two in Ohio and Iowa….
To account for what we see, some have hypothesized that Romney is simply over-performing in the blue states and blowing the roof off in the Southern red states, both of which are untouched by Obama’s ground game. But this theory immediately runs smack into the face of some inconvenient facts…. Florida, North Carolina, and Virginia collectively represent about one-third of the population of the South, so if the “state poll view” of the races is correct, then it becomes difficult for Romney to run up huge margins in the South. He would have to really be doing well in places like Texas, Mississippi and Georgia….
[W]e can reverse-engineer a national poll from the state polls…. I took the states with RCP Averages… poll data post-Oct. 4, our data set is confined to a fairly stable period in the race. For the 19 remaining states [without RCP Averages], I assumed Romney would do as well as George W. Bush…. I downgraded Romney a few points from Bush's showing in Texas, and upgraded Obama a few points from Kerry's Hawaii performance, owing to the “home-state advantages” the former now lacks and the latter now holds.
I took these percentages and multiplied them by the actual number of votes cast in each state in 2008. This produced an expected vote total for Obama and Romney in each state under current polling (or, if not polls were available, under Bush’s 2004 vote shares). If the state polls are right… Obama should lead by 1.18 points in the national vote….
[T]he state and national polls really are saying different things, at least for now. In other words, if you are calling for the state polls to be right, you are pretty much necessarily calling for the national polls to be wrong, and vice versa….
[T]here are several good arguments for favoring the state polling: (1) you have more polls -- a much larger collective “n”; (2) you compartmentalize sampling issues -- pollsters focused exclusively on Colorado, for example, seem less likely to overlook downscale Latinos than pollsters with a national focus; and (3) the state pollsters were better in 1996 and 2000…. But… national pollsters… [are] a battle-tested group with a long track record…. Of the 14 pollsters surveying Ohio in October, only four did so in 2004 (five if you count CNN/USAToday/Gallup and CNN/Opinion Research as the same poll). Pollsters such as ABC/Washington Post, Gallup, Pew, Battleground, and NBC/WSJ are well-funded, well-staffed organizations. It’s not immediately obvious why the Gravises, Purple Strategies and Marists of the world should be trusted as much as them, let alone more…. Finally, remember that in 2008, the national polls were pretty much spot-on; the state polls were off by a couple of points. Although they actually tended to favor McCain that year, it’s just further proof that the larger “n” isn’t a guarantee of greater accuracy.
And Nate Silver on the same question:
Oct. 30: What State Polls Suggest About the National Popular Vote: [P]retty much every method for evaluating the election based on state polls seems to hint at a very slight popular vote lead for Mr. Obama, along with an Electoral College one…. [S]even different Web sites that use state polls, sometimes along with a modicum of other information like a state’s past voting history, to produce predictions of the popular vote in each state…. [T]he various projections strongly agree with another…. The only state where different sites show different candidates ahead right now is Florida, where Talking Points Memo gives Mr. Obama a nominal 0.2-percentage point lead while the others (including FiveThirtyEight) have Mr. Romney slightly up instead….
I’m more interested in looking at this data in a macroscopic way. Suppose, for example, that you take the consensus forecast in each state… this implies that Mr. Obama leads nationally by 1.9 percentage points — by no means a safe advantage, but still a better result for him than what the national polls suggest.
What if turnout doesn’t look like it did in 2008? Instead, what if the share of the votes that each state contributed was the same as in 2004, a better Republican year That doesn’t help to break the discord between state and national polls, unfortunately. Mr. Obama would lead by two percentage points in the consensus forecast weighing the states by their 2004 turnout. Or we can weigh the states by their turnout in 2010, a very good Republican year. But that doesn’t help, either: instead, Mr. Obama leads by 2.1 percentage points based on this method….
Whether the state polls or the national polls characterize the election correctly could well determine its outcome…. [O]ur decision to cast our lot mostly with the state polls is not arbitrary. In recent years, they’ve been a slightly more unbiased indicator…. In recent elections — since state polling data became more robust — it’s the state polls that have done a bit better. This was especially so in 1996, when national polls implied a double-digit victory for Bill Clinton over Bob Dole (and Ross Perot) but state polls were more in line with the single-digit victory that he actually achieved….
[I]f Mr. Romney wins the popular vote by more than about two percentage points… he’ll be very likely to cobble together a winning electoral map, somehow and some way…. But the historical evidence weighs in slightly more heavily on behalf of the state polls, in my view, when they seem to contradict the national ones. If the state polls are right, than Mr. Obama is not just the favorite in the Electoral College but probably also in the popular vote….
Mr. Obama made gains in the FiveThirtyEight forecast on Tuesday… Ohio, where three polls released on Tuesday gave Mr. Obama leads by margins ranging from three to five percentage points. Two of the polls, from Grove Research and the Mellman Group, generally show strong results for Democrats, which give them less impact in the forecast after applying our adjustment for pollster “house effects”. Still, the three polls taken collectively were enough to widen Mr. Obama’s projected lead in Ohio to 2.4 percentage points from 2.1 on Monday. Given how central Ohio is to each candidate’s electoral strategy — and how little time remains in the race — this was enough to improve Mr. Obama’s Electoral College chances. (The forecast does not yet account for the poll by Quinnipiac University for The New York Times and CBS News, which had Mr. Obama five points ahead in Ohio but which was released after we had run the model for the night.)…