The exception was the Rasmussen Reports tracking poll, which gave Mitt Romney a two-point lead among likely voters. (This was in the version of the poll that included voters who leaned toward a candidate, which is the one that FiveThirtyEight uses for all surveys.)
What to think of the Rasmussen poll? Their surveys usually have a Republican lean, but it seems to have gotten stronger in the last few weeks. It has also been stronger in some years than others. Rasmussen got reasonably good results in years like 2006 and 2008 when their polls were close to the consensus. However, their polls were
the least accurate of the major polling firms in 2010, when they had an especially strong Republican house-effect.
The same was true in 2000, when they had a three- or four-point statistical bias toward Republican candidates.
This feature is not unique to Rasmussen Reports: a poll that substantially differs from the consensus, whether in a Democratic or Republican direction, is usually not one that you’ll want to bet on. And there is even less reason to do so when
a poll is taking a number of methodological shortcuts, while others are being more thorough. But there have been years when the whole polling average has been off in one direction or another, and the “outlier” polls turn out to look good.
It’s also the case that a broken clock is right twice a day.