Too many Democrats (38%), too few Republicans (28%)in this sample. This is the favorite Jedi mind trick of media polls. In every Presidential election since 1988, neither party had more than a four point advantage over the other party. You can safely disregard any poll with a spread of more than 4 points between Democrats and Republicans. It isn’t scientifically valid because the sample used is not representative of the actual voting population (it contains too many Rats and not enough GOPers).
To put this in perspective, there is evidence in Rasmussen’s tracking poll that Obama polled better on 9/19 and 9/20, while McCain polled slightly better the next two nights. He had the race even this morning.
In Gallup, Obama polled slightly better on 9/19-9/21, but McCain polled even last night. This is among registered voters. Gallup says that its likely voter pool is about 4-5 points more Republican than its registered voter pool (which it uses for its daily tracking). This morning all this translated to a 3 point lead for Obama among registered voters, suggesting that McCain has in fact pulled even or even taken a small lead among likely voters.
"Show me just what Mohammed brought that was new, and there you will find things only evil and inhuman, such as his command to spread by the sword the faith he preached." - Manuel II Palelologus