Pollsters who failed to predict the size of John McCain's victory in Tuesday's New Hampshire primary--and the narrowness of Bill Bradley's defeat--are not hanging their heads, but they are scratching them.
"We're all saying, 'What the hell happened?' " said Andy Smith, director of the University of New Hampshire's Survey Center.
Though only one of six day-by-day tracking polls failed to predict the winners--American Research Group had Texas Gov. George W. Bush beating McCain--none foresaw the margins of victory. Going into the Republican primary, McCain led Bush in New Hampshire by no more than 12 percentage points in those polls. McCain ended up with an 18-point win.
On the Democratic side, Vice President Al Gore had a 16-point lead over Bradley in one poll and averaged just under a 10-point edge among six polls. Gore ended up beating the former New Jersey senator by only 4 percentage points.
The surprising margins, pollsters say, resulted from higher-than-expected turnout among New Hampshire's independent voters and voters who were undecided until primary day strongly favoring insurgents McCain and Bradley. Why those factors failed to turn up in the tracking polls has pollsters like Smith reconsidering their formulas.
"For some reason we all were systematically missing a large group of voters from our samples," Smith said, "and it's up to us to figure out why that happened."
CNN polling director Keating Holland said he's "kicking" himself for not assuming that undecided voters, who typically favor challengers, would end up supporting McCain over the Texas governor and Bradley over the sitting vice president.
"If you've had eight years to decide whether or not you're going to vote for Al Gore and you're still undecided the day of the primary, you're not going to vote for him," Holland said.
Tracking polls, which try to gauge daily changes in public opinion, can be helpful in predicting trends, but they are "quick and dirty," said Henry Brady, director of UC Berkeley's Survey Research Center. They can very easily miss segments of a population.