American polls

With the Canadian election so interesting, I haven't been looking much at American stuff for the last few days. Josh Marshall writes about the recent polls, showing Bush at 4 out of 10 approval.
I guess this is OK, though I would be much happier if it were 3 out of ten.
As Canadian polls showed, sometimes polls don't get you very far -- they may help people decide who to vote for, or against, but I'm not sure how useful they are in predicting results. The last Canadian polls were done about 5 days before the election, and the thinking is that there was a bit of a turnaround over the last weekend, enough to give Martin an extra 20 seats -- I guess his last-minute coast-to-coast sprint was a vote-getter -- he came across as pumped up, positive, energetic. So even a day can make a difference, I guess.
So when it comes to the American polls, I don't know how much faith to put in them. Also, by their nature, national polls are not reflective of the riding results or, I presume, the Electoral College results -- after all, it doesn't matter whether Kerry gets 95 per cent of the New York vote, or 55 per cent, he still gets 31 electoral college votes. And likewise, whether Bush gets 52 per cent in Texas, or 98 per cent, it still gets him 34 electoral college votes.
So when you read those national polls that have asked a thousand voters nationwide how they will vote, the result has to be read with a bale of salt -- a 30/70 split is large enough to be significant in the electoral college, but a 40/60 split may not be, depending on where those votes are. And a state-by-state breakdown of a thousand-person poll is meaningless.