Correction to This Article
- A Jan. 10 A-section article incorrectly identified Neil Newhouse as half of the polling team behind the NBC-Wall Street Journal poll. He served in that capacity for about six months last year.

Were N.H. Polls a Repeat of 1948 Failure . . . or Not?

By Jon Cohen
Washington Post Staff Writer
Thursday, January 10, 2008

What if the polls were right, and the dynamics changed dramatically on Election Day?

Democratic pollster Peter Hart has a contrarian view on the latest polling kerfuffle. Hart, one-half of the polling team behind the NBC-Wall Street Journal poll, contended yesterday in an online comment on a Washington Post article that we have just witnessed a redux of 1948, the year of the polling industry's most famous failure.

That year, public polls stopped collecting data on the presidential campaign in early- to mid-October, thinking that New York Gov. Thomas E. Dewey had the race with President Harry S. Truman sewn up. After Election Day, the Detroit Free Press ran the box score "Truman 304, Pollsters 0."

In the vastly accelerated political world of 2008, perhaps polling only two or three or 3 1/2 days of what was a five-day campaign in New Hampshire was the equivalent of stopping polling weeks before the 1948 election.

Unfortunately, data to fully test this hypothesis are not public; the American Association of Public Opinion Research and others have called on pre-election pollsters to release their data for analysis.

The network exit poll's "time of decision" question is too crude a gauge to offer a clean read on late-deciders. Polls are good at measuring opinions, but they are less accurate in assessing past behavior. And of the question's five-answer options, three are within the week, potentially leading to an overstatement of the percentage of those making up their minds just before the elections.

One intriguing tidbit in the available data is that comparing exit-poll numbers with those from the last CNN-WMUR-University of New Hampshire poll shows a much bigger movement to Sen. Hillary Rodham Clinton (D-N.Y.) among women than among men. In the pre-election poll released Sunday, 34 percent of women said they supported Clinton, while the exit poll showed that 46 percent of women voted for the senator. The change among men was plus four percentage points.

Is this partial evidence of the "tear" effect that so many have latched onto? A return to the "gender gap" that polls showed before Iowa and the media storm that followed? As was clear last night, much more data analysis -- and more elections -- await.

"In a five day campaign, it was a mistake to think the final decision would be made 'so early,' " Hart wrote. "It is after all, New Hampshire. The motto of the state should be: 'where big mo comes to die.' Add Senator Obama's name to a long list [of] other 'sure NH winners' -- George H.W. Bush in 1980, [Walter] Mondale in 1984, and George W. Bush in 2000."

Neil Newhouse, the other half of the NBC-Journal polling brain-trust (which did not poll in New Hampshire during the primary contest), said: "We just didn't see it coming!"

Newhouse went on to say in an e-mail: "With Hillary Clinton's victory last night, any shred of reputation that pollsters have for being accurate barometers of public opinion goes out the window.

"The problem I have with most explanations for why the polling was so far off is that the very same pollsters that blew the call on the Democratic Presidential Primary last night nailed the GOP primary results. So, how could they get one so wrong and the other so right? Wouldn't a late shift toward Hillary have impacted the GOP primary, too?

"Actually, we've seen a similar pattern of polling not matching actual vote in a couple of instances involving well-known African American candidates where, in a competitive election, pre-election polling shows African American candidates receiving significantly more white votes than the candidate actually gets on election day. The most immediate problem with that theory is that the Iowa polling and caucus results sure didn't seem to reflect that behavior.

"Regardless of the specific reason why the polls were wrong last night, one thing's for sure -- this is an election campaign for the books, providing an abrupt back-to-earth lesson for all of us who study public opinion for a living."

© 2008 The Washington Post Company