I’ve wondered the same thing: if you call up 1,000 voters on a supposedly random basis, and you get 70 percent Dems and 30 percent Republicans, and they all tell you they are going to vote straight party line, does that mean you have established accurately that the Democrat is going to receive something close to 70 percent of the vote?
But I know that pundits are always talking about “sampling” and “weighting” the polls. So I assume the “weighting” comes in when they get the rough numbers, and then apply turnout ratios (and probably other factors) to the 70-30 split and come up with something a bit different.
What I see conservatives hanging their hats on this time is that many pollsters are using 2008 turnout numbers to weight the results; and in 2008 the proportion of Republicans voting was far less than in, for instance, 2010.
Then again I could be completely wrong.
Fair polls reflect the actual ratio of D/R/I, whatever that may be: ratio of registered voters or likely voters (which obviously takes into account how likely they are to vote.).
Oversampling means that the polls ask more of one party, than their actual ratio.
It is NOT just more of one side vs the other.
As an example, say that some place they have 60% D and 40% R registered. If the poll is weighted to have 60% D and 40% R in that sample, than the poll is properly weighted.
If they poll 70% D and 30% R, when there are 60% D and 40% R registered, then they are oversampling Democrats.
Oversampling means that the polls don’t reflect the actual ratio.