As Nate Silver observes today, there’s been a significant divergence between the national and state polls. In the national polls, Romney is ahead. But if you poll each state individually and then weight the results by state turnout, Obama is ahead. What explains the discrepancy?
Here’s one idea: Let’s say you’re doing a poll of New York, so you call a bunch of random 212 numbers. Some of those people have moved their numbers to California — either because they’ve taken their 212 cell phone to California or they’ve used the new number-porting feature of the phone system to redirect a 212 number to a California phone. When the pollster calls them and asks them if they’re registered to vote in New York, they say “No, I’m registered to vote in California.” The pollster hangs up. Meanwhile, the California pollster never calls them, since they don’t have a California area code. The only time they appear in a poll is if a national pollster calls them and asks if they’re registered to vote in the United States (they say yes).
In other words, there’s a group of people who show up only in national polls and not state polls. You’d expect this group to be much bigger than it was four years ago, since cell phones and local-number portability have gotten dramatically more popular. If these people were primarily for Romney, this would explain the discrepancy. (On the other hand, I’d intuitively expect mobile and mobile-using people to skew Obama.)
You could test this by looking at whether there’s a systematic difference between random-digit dialing (RDD) polls and alternative methods of sampling that would be immune from this problem.