A recent CNN poll appears to run counter to recent polling showing Ron Paul leading the Iowa caucus. The CNN poll shows Mitt Romney in the lead, and neoconservative champion Rick Santorum rising. But the CNN poll excluded from its sample Democrats and independents: precisely the group of people whom previous polling showed giving Ron Paul the lead. Thus, in suggesting a Romney victory, CNN is repeating the same polling mistake that many believe helped cause the Chicago Tribune to falsely claim that Thomas Dewey had defeated Harry Truman: they are polling a skewed sample of the electorate.
Here are the numbers for the four top candidates from the CNN poll:
Romney 25%
Paul 22%
Santorum 16%
Gingrich 14%
CNN reports these results with the headlines, "Poll: Romney on top in Iowa," and "Stunning turnaround for Santorum in Iowa."
But, as Nate Silver notes at FiveThirtyEight, CNN only polled "using a list of registered Republican voters and registered Republicans only."
To present a poll of people limited to those currently registered as Republicans as predictive implies a belief that on caucus day, there will not be a significant group of people there who are not now registered as Republicans. (According to the rules, voters can register as Republicans at the caucus.)
But no one outside of CNN believes this. As Silver noted, Public Policy Polling put the Democratic and independent share of the caucus electorate at 24%. The Washington Post put it at 18%. Public Policy Polling showed Paul beating Romney 39-12 among Democrats and independents. Of course, the key reason that these Democrats and independents are backing Paul is Paul's opposition to the wars. So, by excluding them from its sample, CNN is silencing antiwar voices -- nothing new there.
Suppose that Public Policy Polling was right about the composition of the electorate, and about the preferences of Democrats and independents. Let's see what happens to the CNN poll if you add the Democrats and independents back in.
Here are the numbers for the "adjusted" CNN poll:
Paul 26%
Romney 22%
Santorum 14%
Gingrich 14%
So, both of CNN's headlines in reporting its poll were substantially driven by CNN's decision to exclude Democrats and independents from their polling.
It's certainly true that in any event, the gap between Paul and Romney is not large relative to sampling error, regardless of the composition of the electorate; and it's certainly true that Santorum has risen and Gingrich has fallen relative to previous CNN polling.
Nonetheless, it remains true that the stories CNN told by their headlines were substantially driven by their decision to exclude Democrats and independents from their polling, and therefore, it remains true that the CNN headlines were substantially misleading.