Sean Trende has been pushing the idea that the state and national polls in the presidential race are not compatible, and he makes the point in detail in a worthwhile post today. The basic argument is that if you take the best estimates from polling averages of where every state should be, and multiply each by likely turnout numbers for each state, then you should get a number equal to the national polling results — but in fact, the numbers don't match. This suggests that the situation isn't an Obama advantage in the Electoral College, as it would if the vote was just distributed well for the president. Instead, it suggests that either the national polling averages or the state polling averages are just wrong.
Trende doesn't really try to resolve the split, although he does mention that many state pollsters are new to the field, while most of the national surveys are done by experienced groups. Nate Silver basically agrees that the gap is there, but argues that the state polls typically perform better.
I have three general comments about this, beyond pointing out that what Trende is doing here is perfectly legitimate — this isn't “unskewing the polls” at all.
First, we're not talking about a massive difference. Nate Silver estimates that the correct national Obama lead that goes with the state polling leads he sees is about 2 percentage points; Trende calculates it at 1.1 points. HuffPollster currently has a Romney lead in the national vote over 0.6%. So that's either just barely within or just outside of the normal 1-2% error we might expect to see in the national vote (or, perhaps, in the Ohio vote in the other direction).
Second, while Trende is of course exactly right that the state votes have to add up to the national vote, it wouldn't really be unthinkable for one party to have an advantage of this size in the Electoral College. Currently HuffPollster is showing a bias of about 3 points — it would take a 2.3% uniform national swing to give Romney Ohio and, with it, the Electoral College, which would give him a 2.9% national lead. That is unusually large, but remember that just last cycle the state that put Obama over the top, Colorado, was about 1.7 points more Democratic than the national average.
And most importantly, it’s quite possible that both the swing state polls and the national surveys are correct, and that it’s the safe state polls that are wrong. Remember, you need all of it to add up, and an Obama electoral college advantage can occur despite either Obama doing extremely badly in Republican safe states, worse-than-expected in Democratic safe states, or both. Neither Trende or Silver find that happening, but safe states are (unsurprisingly) dramatically underpolled. Huffpollster is working from 560 national polls, and is tracking 97 polls in Ohio, but only has 29 total polls in California – just seven of them, with Obama leads ranging from 12 to 23 points, since the first debate. And there have apparently been only two – two! – polls of Texas since the first debate, and ten overall throughout this cycle.
In other words, we really have no good estimate of what’s happening in most of the safe states, including very large ones. So the question really is which is more likely: that there’s a systematic difference between heavily surveyed swing states and the very heavily surveyed national vote, or that the polling averages have a few large safe state estimates wrong.
Of course, Trende certainly could be correct. And Silver’s look at it includes his own estimates and others which incorporate previous history, so there’s more to go on than just very sparse polling. Moreover, there’s certainly a possibility that it’s a little of each, which would put the true results somewhere in the middle. We don’t know! But there’s not enough information, in my view, to reject the possibility that both the national and swings state polls are correct.