Without a doubt, a big source of the discrepancy between the pre-election surveys and the election outcome in New Hampshire is the order of candidates' names on the ballot and in the surveys.
Our analysis of all recent primaries in New Hampshire showed that there was always a big primacy effect — big-name, big-vote-getting candidates got 3 percent or more votes more when listed first on the ballot than when listed last.
Until this year, New Hampshire rotated candidate name order from precinct to precinct, which allowed us to do that analysis.
This year, the secretary of state changed the procedure so the names were alphabetical starting with a randomly selected letter, in all precincts.
The randomly selected letter this year was Z.
As a result, Joe Biden was first on every ballot, Hillary Clinton was near the top of the list (and the first serious contender listed) and Barack Obama was close to last of the 21 candidates listed.
Thus, I'll bet that Clinton got at least 3 percent more votes than Obama simply because she was listed close to the top.
Most, if not all, of the pre-election telephone polls rotated name order from respondent to respondent, which meant name order did not distort their overall results. Failing to incorporate the name order effect that probably happened in the voting booth is therefore probably partly responsible for the polls' inaccuracy.
More importantly, if New Hampshire had rotated name order in the voting booth as it has always done in the past, the race would probably have been too close to call without a recount and might even have been an Obama victory.
Jon A. Krosnick is the Frederic O. Glover professor in humanities and social sciences at Stanford University. He got his Ph.D. from the University of Michigan.