boston.com your connection to The Boston Globe
THE NUMBERS GAME

Late polls are seen as largely accurate

WASHINGTON -- After an election season marked by unprecedented criticism of mainstream opinion polls, pollsters were able to breathe easier over the final election results, which largely vindicated their predictions.

Most final polls published in the last 48 hours of the campaign accurately forecast President Bush winning with a narrow lead over John Kerry. Among the big national polls, Gallup gave Bush a 2-percentage-point edge among likely voters, 49 to 47. The Washington Post/ABC, Zogby, and NBC/Wall Street Journal polls all gave Bush a 1-percentage point advantage.

With almost all votes counted, Bush had a 3-percentage-point lead over Kerry in the popular vote, 51 to 48. Independent Ralph Nader finished with less than one-half of 1 percent of the popular vote. The rest of the popular vote went to an assortment of third-party candidates.

"Virtually all of the polls came within a few percentage points of the actual margin," said Michael Dimock, the research director for the Pew Research Center for the People & the Press in Washington. "That's almost more than what you can expect." In its final poll, Pew predicted exactly the 51 to 48 breakdown in the popular vote.

The results compare favorably to 2000, when many polls before Election Day showed a small Bush lead over Al Gore, who ended up narrowly winning the popular vote.

"This was an excellent election for the polls," said Humphrey Taylor, the head of Harris Interactive, whose last telephone poll showed Bush up 49 to 48.

Not every survey picked Bush. The last Fox News poll gave Kerry a 2-percentage-point advantage, 48 to 46, and Marist measured a 1-percentage-point edge for Kerry in its final poll, 50 to 49. Two Internet-based surveys, touted by some political scientists as the future of polling, were also off, both calling the race for Kerry by identical margins, 50 percent to 47 percent.

Still, the overall agreement of the final polls, most of which are based on much larger samples than surveys earlier in the campaign, stood in contrast to the variation in September and October polls, which had spurred many partisans to question their accuracy. Right after the Republican convention, the poll numbers for Bush and Kerry varied widely.

Some critics worried that the prevalence of cellphones, which pollsters do not call, and the declining response rates to polls might have undermined their accuracy. A small but growing percentage of Americans do not have a land-line phone, and even among those who do, pollsters acknowledge that, increasingly, people they call are unwilling to be interviewed.

But at least in the national polls, none of those problems seemed to be a factor.

"The cellphone issue is a very serious issue for pollsters, but this year it wasn't a sizable enough problem to affect the accuracy of the polls," Dimock said. "The cellphone-only population is small enough not to have an impact overall on the polls. Four years from now, if the proportion is significantly higher, our industry is going to have to find a way to account for that."

Response rates to surveys continued their decline this year.

The election results also offered little evidence to support accusations during the campaign from Democrats who said that the methodology used by many opinion polls underrepresented their strength. Critics had charged that randomly drawn poll samples often end up giving disproportionate weight to Republicans and should be adjusted to reflect the partisan breakdown of the country as a whole based on exit polls from the 2000 election. In particular, Gallup came under fire for publishing surveys that included significantly more Republicans than Democrats; the liberal group MoveOn.org took out full-page newspaper ads attacking the "longstanding problem" in the firm's methodology.

Gallup maintained that it could not assign a statistical weight for partisan identification, because party isn't a fixed demographic characteristic like sex or race.

"To impose the party breakdown from four years ago was just the wrong thing to do in this situation," said Dimock, who also does not assign a weight by party identification. "We had a president that has reshaped to some extent the image of the Republican Party and appealed to groups that otherwise might not have considered themselves Republican."

Tuesday's results seem to back them up. Exit polls show that 37 percent of voters Tuesday were Democrats, and 37 percent Republicans. In 2000, the breakdown was 39 percent Democrat and 35 percent Republican.

Ruy Tiexiera, a fellow at the Center for American Progress, whose weblog regularly criticized Gallup during the campaign for failing to assign weight by party, said the election results do not necessarily close the argument.

"This doesn't settle the question of whether during the campaign some of these samples were drawing in too many Republicans. They had big advantages, more than 37 percent," he said.

Taylor dismissed criticism of survey methodology, especially complaints about volatility. "If you have six, eight, 10 polling firms doing polling, you would expect there to be a number of outliers," he said. "If there were no outliers, the only thing that could mean is that we were all in collusion. I think all of the critics have been totally unfair."

SEARCH THE ARCHIVES
 
Today (free)
Yesterday (free)
Past 30 days
Last 12 months
 Advanced search / Historic Archives