Depends on the state. A reasonable sample would be Dem +4 nation wide. That assumes that the election will fall some place between 2004 and 2008 as far as turn out goes.
However here is the problem for the pollsters. Since 2010 Reps have been posting significantly higher registration numbers in most of the battle ground states. So it is possible that even a +4 Dem sample is still too high based on current party registration numbers
So Dems point to 2008 and claim they should get a bigger sample Repbs point to 2010 and say the Dems should get a smaller sample
Problem is the pollsters are using 2008 as a base line and assuming Obama will do better in 2012.
See any evidence on the ground that Obama is doing better with the voters in 2012 then he did in 2008?
OK, I think I get it now.
In a word, no. He is preforming worse in every group. Even african americans who came out in big number in 2008 will fall some. Democrats always get 90% of the votes, it will just be a bit smaller total votes this time.
OK, so we say that the bias, or sampling should reflect the most recent turnout numbers to get most accurate results. But then, turnout varies from election to election, as you point out, and is not going to be in 2012 like 2008 or 2010, which in turn tells me what I’ve always thought that is that these polls are in the end more or less worthless.
I think D+2 is safe. That is a max number for the Dems IMHO. As long as Romney can hold onto his 10% + margin with the Independents he should win a tight race. If we repeat the 2010 turnout he will coast home.