What’s wrong with using a list of Republican voters for a Republican caucus poll? The answer is that it’s extremely easy for independent and Democratic voters to register or re-register as Republicans at the caucus site. Historically, a fair number of independent voters do this.
And as we've seen, Paul draws a wide advantage among independents and Democrats, so the CNN poll is "quite simply missing these voters and therefore will probably underestimate Mr. Paul’s support, perhaps by several percentage points." But by that same standard, things are looking worse for Paul in New Hampshire:
There is some unambiguously good news for Mr. Romney in New Hampshire, however, where the CNN poll did include independents and shows him with a very healthy-looking 27-point lead over Mr. Paul. Although the New Hampshire numbers still have the potential to change significantly both before and after the Iowa, that is a pretty healthy cushion, and Mr. Romney’s rivals appear to be losing ground at the very moment when they need to gain it. Our forecasting model now gives Mr. Romney a 90 percent chance of winning New Hampshire, up from about 75 percent previously.
One important potential shortcoming for both surveys: Neither samples mobile phones. As the Washington Post reports, the most recent estimates [pdf] find that 29 percent of Iowa adults had a cell phone but no landline in their household. The Washington Post/ABC survey conducted in early December found that Paul did better than Romney on interviews conducted over cell phones, while Romney did better than Paul on landline interviews.