Polls including cell-phones are highlighted in orange.
Wondering about the variation across survey organizations in estimating Obama's national lead? As I have been discussing in my research methods course here at AU, much of the variation is likely accounted for by differences in likely voter models but also in sampling differences that include either cell phones or landline phones only.
As Wired reports, over at the blog 538.com, Nate Silver shows that surveys that include cell phones average a 10pt lead for Obama, reflecting stronger support among younger voters and minorities. Among surveys that include only landline phones, Obama's lead averages only 5pts.
To read more about the impact of the profusion of polls on coverage of the election, see this short chapter on the topic that I recently published at the Encyclopedia of Survey Research Methods.
I'm frankly a bit surprised that anybody would even bother running a poll without cellphones. The logistics are trickier, sure; but "without cellphones" is basically a nice way of saying "We definitely didn't bother to ask any college students, and we missed a nontrivial percentage of everybody under 30."
I can see that a lot of this would, historically, be less relevant in politics, since younger people generally vote less, and landlineless people are a pretty new phenomenon; but surely pollsters would have updated their methods, if only for market research's sake.
Nice to see that Pew and Zogby learned their lessons from 2004. They were very close to the final popular vote margin of 6.3%. Pew's methodology has definitely earned my trust by their inclusion of cell phone wielding participants.
It's also worth remarking that the gross mean of these polls is almost exactly 7%, a number quite close to the final margin of 6.3%. I guess Pollster.com had the best idea all along, especially their use of regressive trend analysis instead of "the latest numbers" as a raw input.