BOSTON — Two things are clear from the wildly divergent polls in this year’s presidential election — President Bush is ahead, but nobody knows by how much.
The swirl of surveys over the last week has shown the race anywhere from a 13 percentage point lead for Mr. Bush to a dead heat between him and Democratic presidential nominee Sen. John Kerry of Massachusetts. That gap, which pollsters said is unusually wide, has prompted debate inside and outside of the campaigns over where the race really stands.
“If I had to make a guess today, it would be Bush,” said Harry W. O’Neill, who polled for Roper and before that for Barry Goldwater and Richard Nixon. “Even if it’s only a point, he’s the one who’s ahead, and the internals and all those other questions are in favor of Bush, except the economy.”
The most discussed result may be last week’s Gallup poll, which showed a 13-point lead for Mr. Bush among likely voters, an improvement even over the seven-point lead the same poll showed the president held after the Republican National Convention. A CBS-New York Times poll released late Friday showed Mr. Bush with a nine percentage point lead among likely voters surveyed.
That compares with the poll from the Pew Research Center for the People and the Press, which showed Mr. Bush with only a one point lead among likely voters. And a Harris poll released last week even showed Mr. Kerry holding a one point lead among likely voters.
Pollsters, politicians and the campaigns are flummoxed.
“I can’t explain the huge variance,” said Adam Clymer, political director for the Annenberg Election Survey. “I don’t understand why Pew and Harris are at one place and Gallup and the New York Times-CBS are at another, and I’m not sure that if I had complete access to all their data and their methods I would, either.”
Humphrey Taylor, chairman of the Harris Poll, said the variance is particularly curious because the polls in question are all done by reputable organizations. “The first thing I look for is who’s done it. These are well-known people who have a history,” he said.
Pollsters said when looking at polls, in addition to the source, to keep in mind whether the survey included enough respondents; whether it covers too many days, during which opinion could have shifted, or too few, which could leave a skewed count of who was home; and how the questions were posed.
But in the case of recent major surveys, those factors were all similar. One potential culprit, Mr. O’Neill said, is the way the surveys identify likely voters.
There is almost always a difference of a few percentage points between the opinions of the general population, or registered voters, and the opinions of those who actually show up to vote. Polls try to weed out nonvoters through a “likely-voter screen.”
Gallup, for example, uses a series of questions, such as whether a voter knows his polling place, is knowledgeable about the issues, and has a history of voting, to decide whether someone is a likely voter this election. But Mr. O’Neill said it may be too early in the season for that to be effective.
“While that is probably a very good way to screen likely voters in your absolutely last poll before the election, if you use that earlier on, it kind of is narrow, and tends to eliminate a lot of people,” said Mr. O’Neill, a member of the National Council on Public Polls’ polling-review board. “Right now, I’d focus more on the registered voters than the likely voters for comparing polls.”
The Gallup poll’s 13-point lead for Mr. Bush shrinks to eight points when all registered voters are considered. The Pew lead for Mr. Bush, though, jumps from one point to six points — another oddity, since Republicans usually perform better among likely voters than registered voters.
Mr. Clymer said another curiosity is that the horse-race numbers are more volatile than some of the other poll numbers, such as whether the country is on the right track, or how the candidates fare on handling the issues.
This year’s results may be skewed because of the higher level of interest in the campaign.
“If in fact the level of interest is higher, if the degree of polarization is stronger, it may be that different ways of defining likely voters produce different results,” Mr. Taylor said.
The Kerry campaign sought to shoot down the Gallup results in a conference call with reporters on Friday.
“We do have some evidence from the 2000 race that the Gallup poll often was the outlier among other polls, and there was more volatility in those numbers, perhaps based on the somewhat complicated likely-voter model that that poll uses, perhaps for other reasons,” said Tom Kiley, a pollster for the Kerry campaign.
Still, he conceded that the preponderance of polls indicates Mr. Bush is slightly ahead.
“If we do not include the Gallup survey, the average of all these polls points to a two-point race, with the president ahead by two; obviously, a very tight race,” he said.
Mr. O’Neill said averaging polls together is probably a decent method of figuring out where a race is, as long as the polls ask similar questions and are taken during the same time period.
“Basically, they’re within a few points, and I’d say that’s where the elections stands,” Mr. O’Neill said. “Maybe somewhere between four to eight [points] is probably the range.”
In addition to sampling error, which every poll reports, Mr. Taylor said there’s an amount of uncertainty built into polls.
One factor is that, with the proliferation of polls, people are increasingly reluctant to participate. Mr. Taylor said that to have 1,000 respondents, it will take about 4,000 or 5,000 calls over a two- to three-day period.
The National Council on Public Polls conducted a survey of national polls in 2000 and evaluated their accuracy compared with Election Day results.
They found the Harris Poll to be the closest to the actual result. Harris’ final poll showed Mr. Bush and Al Gore even, at 47 percent each. On Election Day, Mr. Gore and Mr. Bush each took 48 percent.
The most inaccurate survey was the Rasmussen poll, conducted by computer with a recorded voice. It had Mr. Gore winning by nine points, 49 percent to 40 percent.