Could the polls be wrong?
They have misled before. The most famous time came in 1948 when polls showed Republican Thomas Dewey on his way to winning the White House but missed the late surge that carried Democrat Harry Truman to victory.
More recently, polls showed Obama ahead of Democratic rival Hillary Clinton in the January New Hampshire primary by an average of 8 percentage points. Clinton won.
McCain’s campaign thinks it could happen again on November 4.
"All signs say we are headed to an election that may easily be too close to call by next Tuesday," McCain pollster Bill McInturff wrote in a memo released on Tuesday.
Pollsters are careful to say their work does not predict a race’s outcome but only captures a snapshot of the electorate at a certain point in time.
And there is always the possibility of error in a discipline that combines science with a certain amount of guesswork.
"We are engaged to some degree in some artwork and assumption," said pollster John Zogby, whose Reuters/C-SPAN/Zogby tracking poll shows Obama leading by 7 percentage points.
Pollsters can’t simply tally up the results of their telephone surveys but must make educated guesses about who will actually show up to vote.
These "likely voter" models vary from poll to poll, leading to results that can vary as well.
Rasmussen Reports’ daily tracking poll on Wednesday showed Obama leading McCain by 3 percentage points, while a survey released on Tuesday by Pew Research Center for the People and the Press showed Obama leading McCain by 15 percentage points.
HIGHER TURNOUT AMONG BLACKS?
Gallup publishes two separate polls based on varying likely-voter models. Its traditional model showed Obama ahead by 2 percentage points on Wednesday, while its "expanded" model, which assumes higher turnout rates among minorities and young people, showed Obama leading by 7 percentage points.
Zogby’s poll and the Pew poll also assume that black voters, inspired by the chance to elect Obama as the first black president, will make up a higher percentage of the electorate this year.
But the McCain campaign argues that turnout will be high among all demographic groups, diluting any impact from black voters.
Another question is the "Bradley Effect" — the notion that white voters fearful of being labeled racist overstate their support of black candidates in polls.
That theory, named for a black 1982 California gubernatorial candidate who narrowly lost to a white opponent after leading in opinion polls, has been widely discounted.
But polls might be underweighting any racial backlash because intolerant voters tend to hang up on pollsters, said Pew Research Center president Andrew Kohut said.
"That could be a factor in a close election, but it’s probably not a factor in an election that seems as wide open as this," Kohut said.
The increasing number of people who decline to participate in opinion polls, and the difficulty of reaching cellphone-only households pose challenges as well.
"When you get a 15 percent response rate, the people you get aren’t necessarily representative of the population as a whole," said Nate Silver, a statistician who tracks opinion polls at the Web site FiveThirtyEight.com.
Despite these concerns, pollsters say they’re confident in their work. After all, it’s hard to overlook the fact that major polls have lined up closely with the actual vote in every presidential election since 1980.
"We bring a whole lot more science to who’s ahead and who’s behind than a handful of old white men sitting at the Mayflower Hotel in Washington having breakfast," Zogby said.