So Newsweek put out a new Generic Ballot. The magazine's polling had drawn notice before in my House projection reports (this week's edition coming later today), but this new one just seems completely out of line: Democrats +5 among Registered Voters. That filtering is expected to lean to the left after the 2008 anomaly, but this is ridiculous.
I have two major points of contention with this poll that call into question how useful it is. First off, out of a pool of 1,025 adults, it finds that 902 are registered to vote (MoE 4.1%). According to the US Census Bureau the figure is actually 63.6% in 2008 (MoE 0.3%), which matches the 2004 figure, and is up about 5% from 1996 and 2000. In other words, this poll's basic filtering is outside the margin of error.
On top of that, it finds that 768 respondents (MoE 4.4) will definitely vote, or 85% of the registered voters. Again, turning back to the Census Bureau, that figure for a Presidential election, which is grossly optimistic for an off-year election, is 90%. Only a 5% dropoff in voting from the 2008 Presidential election to the 2010 midterm would be newsworthy in itself. I'm not expecting that.
So even without digging into partisan makeups or any of the fancy stuff, we can already question this poll's worth just by seeing improbable results right in the voter screening questions. This poll is worth about a dollar if someone wanted to buy it.