Here is Jay Cost -- once an election analyst with RCP, now with The Weekly Standard -- pointing out what should have been obvious weeks ago: that the two different kinds of poll results for Romney v. Obama, whether in individual states for incorporation into an electoral vote model, or nationally for estimating the popular vote, are the result of sampling different populations.
How do you do that, you ask? Like this:
- Ring. Ring.
- Potential Pollee: "Hello?"
- Live Interviewer.: "May I speak to the oldest female voter in the house?"
- P. P.: "That's me!"
- L. I.: "Are you between the ages of 18 and 55?"
- P. P.: "Uh. No."
- L. I.: "Thank you very much for your time. Good evening."
So you get the interviewees you want. The right racial, the right age, and the right socioeconomic profile. Everything. But the answers you get are dependent on the model of the electorate with which you start. And you don't waste L. I.'s valuable time on interviews you won't use.
So here comes National Journal's Editor-in-Chief Reid Wilson with a complete explanation of the dichotomous distribution of sample means that Jay Cost showed us.
The mainstream media and their pollsters are extrapolating the change from the 2004 electorate to the 2008 to get a 2012 electorate that's even more pro-Obama than 2008 on the basis of the notion that the electorate has been becoming more diverse, so why couldn't 2008 be nothing but the result of the trend.
Meanwhile Rasmussen and other autodialer pollsters are just sampling, and adapting to the electorate they find, or weighting their results to get a different electorate, but in either case getting one more like 2010's.
Who's right? Well, we could wait till next Tuesday... but that would be no fun!
I tend to think Gallup's results support Rasmussen much more than they support CNN, CBS/NYT, NBC/WSJ, etc. Gallup has a carefully tuned likely-voter model. Not just "How likely are you to vote," but seven questions: Do you know where you vote? Where you vote early? Did you vote in the last general election? And so forth. [Sherry got that call, too, but Gallup has told us before what the model is.]
And remember that Gallup foretold the 2010 mid-term when the Generic Congressional Ballot results for their likely voter population were 13 points more Republican than the registered voter population.
What will this election be like if the electorate is more 2010 than 2008?
Well... what is the Rasmussen no-tossup result tonight?
Umm, well, Romney gets 206 (fairly) sure plus Virginia's 13 (+3%), Colorado's 9 (+3%), Florida's 29 (+2%), New Hampshire's 4 (+2%), Ohio's 18 (+2%), and Iowa's 6 (+1%), for a total of... drum roll, please... 285.
Close. But it definitely wins the cigar. And there's room to lose Virginia, or Iowa and Colorado. Bur not Ohio.
