Another important thing to keep in mind is that partisanship sometimes kicks in late, or manifests itself in the behavior of undecided voters and so doesn't become apparent until election day.
So e.g. if you have a poll showing the Republican up 3 in a very blue district, the poll may be a perfectly accurate snapshot of currently-decided voters, but it may nevertheless be true that the Democrat is favored because undecideds will break heavily D.
Be more wary of this in the case of polls with high numbers of undecided voters and/or candidates with low name recognition. It's more likely to be an issue in WV-3 than in the Texas senate race, for example.
