The metric I like to look at when they include it is the "who did you vote for last time" question.
Colorado:
Poll: O+2
Actual: O+9
Virginia:
Poll: O+9
Actual: O+7
Wisconsin:
Poll:O+13
Actual: O+14
Granted, people can move between states, forget who they voted for, or simply lie to the pollster, But those numbers are another indication that the Colorado result is skewed towards the Republicans since the difference is beyond the MoE for the poll.
If one were to adjust the toplines to have the poll samples conform to actual 2008 results, you'd end up with: CO: O+2, VA: O+2, WI: O+7.
You can't do that. Poll samples on the "who did you vote for in the last election" question often don't match the previous results for various reasons - people who voted in the previous election fail the likely voter screen because they aren't as enthusiastic about voting this time or moved in or out of the state or died or have buyer's remorse about who they voted for the last time and lie to pollsters or whatever. The 2008 electorate isn't the same as the 2012 electorate.
That said, the CO poll seems a bit off. But I wouldn't doubt a small Romney lead. Obama is campaigning there today for a reason. The state is at least close, not some Democratic blowout.
