The Horserace Blog makes a pretty good case for Bush winning Florida by crunching registration and turnout numbers.
Bush looks pretty good in Minnesota too.
Pennsylvania no telling.
Here is a tip on how you'll know how Dubya is doing in PA. If you see them tabulating the votes and early on Kerry has only a small-to-modest lead (maybe 4-5% max), that is a very good sign for the president. That means that Dubya is holding his own in the suburbs, counter-balancing Kerry's larger numbers in the rural areas. The rural votes, which break overwhelmingly Republican, trickle in later. This is what happened in 1994 when Santorum beat Harris Wofford. All night long Wofford held a slim lead over Santorum, and it broke Santorum's way around 1 AM or so.
Iowa favors Bush.
And for kicks look at how well Missouri mirrors the rest of the nation.
About the Polls:
So...what does this mean? Essentially, it means that the reliable polls are, roughly speaking, Time, Battleground, Gallup. All of the rest skew toward Democrats, and should be viewed with caution. A surprising number of the rest are either using questionable methods or are using unadvertised methods. This is absolutely taboo among social scientists. Methodology is the only element that the researcher can control, and thus it is critically important -- indeed it is an ethical responsibility -- for the researcher to make his/her methods avaiable.
And finally, how Kerry's slip in the black vote can't be offset anywhere else. And why Clinton's job is to secure this base.