"According to the release describing the Iowa State University poll we linked to earlier today, that poll "presents a much different picture
the race than other recent polls." More specifically, four recent surveys from the Des Moines Register, ARG, Rasmussen Reports and Strategic Vision all show Barack Obama and Hillary Clinton running within the margin of sampling error of each other, with John Edwards just a few points behind. The ISU survey shows Clinton (at 30%) with a comfortable lead over Edwards (24%) and Obama (20%). Why the big difference?
As we noted earlier today, the poll was conducted from November 6-18, which makes its results older than the four most recent surveys on our Iowa chart. However, according to ISU political science professor Jim McCormick, who directed the poll, "the biggest explanation for that is the volatility that still exists among those people who are likely to caucus." A better explanation is the poll itself, which is very different than other recent Iowa caucus surveys.
One difference involves the sample sizes on which the results were based. These numbers were omitted from the ISU release. According to Sarah Nusser, director of ISU's survey center, the results were based on interviews with 287 likely Democratic and 241 likely Republican caucus goers. Those are much smaller samples than used on other recent polls. Smaller sample sizes make for greater random sampling error, a form of "volatility" that is about the poll's design, not the voters.
The biggest difference involves the sample. It was drawn from the Iowa Secretary of State's list of registered voters, but unlike every other Iowa poll that I'm aware of, ISU sampled only registered Democrats and Republicans, excluding the 36% of Iowa voters with no party registration.
Here's why that omission is important:...."
http://www.pollster.com/blogs/more_on_the_iowa_state_poll.php