I generally believe in averaging all polls. That is how RCP does it, and it is rather interesting that the averaging of ALL polls (good or bad, Republican-leaning and Democratic-leaning) usually comes up with a better result than taking one or the other poll by itself. Polls are not foolproof, they regularly suffer from bad one-day samples. If you average out all polls you basically remove the margin of error.
If you look at 2004 the generic RCP average was closer to the end result than any one one polling firm was. The same is true for almost every single Senate race in 2006 and many House races.
I don’t believe “the polls are right.” I just don’t believe ALL polls are as wrong as is claimed here. Even the one poll that shows a close race still has McCain losing this thing. Let’s see first whether IBD/TIPP stays tight through today instead of moving away again (as happened until yesterday) before making that one poll the be-all. If that one moves to, say, 4% for Obama, there would not be a single poll even close to a McCain victory, unless GWU/Battleground gives us something good today.
Ahhh... but a poll at 4% can be incorrect as well. McCain needs to only be within 2% of the popular vote to win the electoral college. The recent state poll releases from Mason Dixon and SuveyUSA suggest an Obama national lead of 2 to 3%, not the blowout number suggested by Gallup.