With a week before the election, Democratic presidential candidate Barack Obama has not trailed Republican John McCain in a national public poll in more than a month, but his lead fluctuates greatly in some surveys as pollsters mix art and science as they try to predict who will actually vote.
Pollsters are trying to reflect the potential effects of race, demographic, and registration trends, and the shifting political climate as they assemble the samples for their surveys. Some also make special efforts to reflect the proliferation of voters who only use cellphones.
The result is a mixed bag. The RealClearPolitics website yesterday listed Obama's lead as 7.4 percentage points in the average of 11 national polls completed in the week before, but depending on the survey, his edge ranged from 3 percentage points to 13.
The Globe interviewed five major pollsters yesterday and found that each has a wrinkle in methodology, a "special sauce," one pollster called it. There are different methods to identify likely voters, and most pollsters tweak the results to fit perceived models of demographic groups or party identification within the electorate.
Gallup, which surveys 1,000 likely voters per night, and reports each day on the combined results from the previous three nights, now issues its findings three different ways to present different possible scenarios. Over three nights ending Sunday, Obama led by 5 percentage points among traditional likely voters - those who said they will vote and have voted with some regularity in the past - and by 10 points among both registered voters and a universe of likely voters expanded to include those who say they will vote but do not have a history of voting. That includes many younger and black voters Obama is counting on.
"We're more traditional, in that vanilla mode," said Eric Nielsen, Gallup's senior director of media strategies. "That's why we're showing different models, allowing other people to predict and argue about what's going to end up being the best number."
John Zogby is a firm believer in weighting his samples to reflect what he sees as trends. Zogby International conducts a nightly tracking poll for Reuters and C-SPAN.
Unlike some other pollsters, he tweaks his sample to reflect a 38 percent-to-36 percent edge for Democrats in party identification, and said surveys that place Republicans below 30 percent skew results in favor of Obama.
"There's a fundamental stability to party identification, and that's my difficulty with my colleagues who don't weight for party ID, resulting in some wild fluctuations in some polls," Zogby said. His three-day survey of a total of 1,203 voters through Sunday night showed Obama leading McCain by 5 points.
Scott W. Rasmussen of Rasmussen Reports said "the best indicator of how someone is going to vote is partisan identification," which "eliminates some of the statistical noise." Through Sunday night, Obama was up by 5 points in his survey.
Chris Anderson, vice president of Opinion Dynamics of Waltham, which polls for Fox News, said his surveys include likely voters based on whether they say they plan to vote and does not screen out those without a voting history. He said he does not weight by party.
"This is arguably where art meets science, and it varies from pollster to pollster," Anderson said. "As a result, more black, Hispanic, and young people are making it past the screens than in the past, and it comes out naturally in our models." In his firm's survey of 936 likely voters on Oct. 20-21, Obama led McCain by 9 percentage points.
By contrast, the George Washington University/Battleground Poll has consistently reflected a closer national race, with McCain trailing by only 3 points in a five-night survey of 1,000 voters completed Sunday night.
Ed Goeas said the survey does not sample on Friday or Saturday nights because he has found results those nights skew in favor of Democrats. The survey is weighted to reflect an increase in young voters, and takes into account the intensity of the voter, which he said has been increasing for Republicans recently.![]()



