"The three issues that I think we're dealing with [in the polls]: first of all is the partisanship of the electorate. In the last 25 years, the electorate has ranged from plus-4 Democrat to plus-2 Republican in '02. In the most recent poll's partisanship, USA/Gallup is plus-9 Democratic electorate, ABC News is plus-11 Democratic electorate, CBS/New York Times plus-5, Newsweek plus-8, Time plus-8, AP/Ipsos plus-8. So, every one of these polls has an electorate that looks more Democratic than any electorate has looked in 25 years."
And:
"The third issue of course is the relevance of the national polls in predicting House races and the challenge that Democrats always have is that our voters are more efficiently distributed. You saw that in the recent battleground that came out between [pollsters] Celinda Lake and Ed Goeas, which showed an 8-point Democratic advantage on the generic ballot. But in the Republican districts that the Democrats have to win to win back Congress, it was even. In the Democratic districts, it was a 21-point Democratic advantage."
Read the whole thing in US News & World Report.
Just real quick and dirty.
If a generic poll has Dems +9, and the actual electorate is Dems +2, that could bring a poll (assuming 50-50 split of independents) from 54%-45% to 51%-49%. (and I doubt the public is Dems +2)
So, that bit of oversampling, can make a poll that's truly within the MOE look like a 9-10 point deficit.