AAPOR has its report out today on what went wrong with the polling in some of the early primary states last year. Most of the attention is being given to the findings in New Hampshire, where we didn't poll, but there's also some discussion about Wisconsin and South Carolina where we did.
We were actually the most accurate pollster for the Democratic primary in both of those states but we still underestimated Barack Obama's support. For us I'm pretty sure the culprit was poor weighting. We were using a turnout model predicated too much on past elections, not sufficiently taking into account the effect Obama's unique candidacy was having on the demography of the electorate. That led to us substantially undersampling the Obama friendly group of voters under 30 in both Wisconsin and South Carolina, and in South Carolina it also led to us underestimating the African American share of the electorate.
We made adjustments on those counts for the rest of the primary season that worked pretty well in all but one of the final six contests we polled.
Last year was a unique election cycle and I think pollsters who were able to adapt to changing realities over the course of the contest and not just fall into the trap of doing what they were accustomed to in the past ended up being pretty accurate by the end.