From 3 Quarks Daily:
I’d like to see a list of those “national polls” and what each is saying, because the individual poll results vary widely. I’ve read and heard that in 2008 the Rasmussen poll was the most accurate. I mean, if poll A says Dems by 10 points and poll B says Repubs by 10 points, the “average” would say it’s a dead heat, but somebody in there is dead wrong. Another example: If 5 polls say X by 2, X by 4, X by 6, X by 8 and X by 10, the average (2+4+6+8+10) is 6 (that is 30 divided by 5) but if X ends up winning by 6, only one of those polls was accurate, regardless of what the “average” shows. In 2008, that poll was Rasmussen, so that’s who I’m paying attention to at the moment.
Correction: I think it was 2010, not 2008, when Rasmussen was the most accurate poll.
It also appears that the “overestimated Republican” polls have a much smaller standard deviation than the “overestimated Democrat” polls, though it is hard to say with such a small sample.
Interestingly, there was only one election where the actual votes switched party from the polling average. In other words, the polls seem to usually predict the right winner.
Deets @4, And even in that case (2000), the polls sort-of predicted the winner — the D won the popular vote but the electoral college winner after all the hanging chads and butterfly ballots and Supreme Courts were done was the R.