There are many reasons why I've never understood coverage of presidential races in the US. There seems to be a concerted effort to focus on everything that doesn't matter--specifically broad national polls. As we all know, it's the electoral college that determines the president. We all know how problematic this system is--it effectively disenfranchises Republicans in deep blue states such as CA and NY, and Democrats in deep red states such as TX and GA, but that's a complaint for another day. Right now the EC bears forth the crown.
I really thought McCain would pick a running mate based on the EC maths that meant he had to wrest back one or two midwestern Kerry states. When he picked Sarah Palin, I pretty much wrote off his chances. I've had a lot of other Obama supporters think I'm crazy. "But did you see how much people ate up Palin's convention speech? etc. etc." I responded: how is Palin going to win McCain MI or PA? The shallow assessment is that she would deliver the Hillary Clinton contingent of working-class women and blue-collar men, which completely ignores the depth of the policy differences between Clinton and Palin. Sure, Palin will help turn Texas an even deeper red this year. big whoop. It was always going red, anyway. This is why undifferentiated national polls are not much use.
What is useful are the sites that detail polling along EC lines. My first favorite is Electoral-vote.com, run by the renowned comp sci professor Andrew Tannenbaum a.k.a. the Votemaster (who unveiled his identity to great interest soon after the 2004 election). Its commentary has an avowedly Democratic bent, but it uses irreproachable methods of poll analysis for its statistics. Here is today's map:
Earlier this year I learned about another such site, FiveThirtyEight (the number referring to the total number of EC votes available). Electoral-vote.com is more streamlined, using the main map as a gateway to much more detail on subsidiary pages. FiveThirtyEight takes the opposite approach of information explosion on the front page. Sometimes that suits me, and sometimes it doesn't, so I've been in the habit of using both sites for ongoing reference. Here is today's FiveThirtyEight map: 
The two sites also use different criteria for poll vetting and statistical analysis, which provides alternate perspectives. Both sites today have Obama in the 330 vote range, which would be a big win, but nowhere near as notable as I'd at first thought. Nixon and Reagan have had truly eye-popping EC margins of victory the likes of which we may never see again. But I do think Obama's margin will be higher than most current estimates. Both the above sites point out the systematic under-counting of the youth vote. True, voting reliability comes with age, but many of those 18 year olds who were never polled in 2004 because they had no land lines still don't have land lines at 22. All indications are that they are much more likely to vote this year than in 2004. Balancing that might be some manifestation of the Bradley effect, where polls fail to reflect a reluctance to vote for a minority candidate. In the end the only EC maths that really matters is what the boards show November 5th. Assuming, of course that it doesn't once again take Supreme Court action to settle matters (oh perish the very vestige of thought!).