This year's polling at greater risk of getting it wrong

A few days before the election, national opinion polls confound:

— The popular vote in the presidential race will be either razor close or a double-digit blowout.

— The 14 percent of Americans who don't use landline phones at home, and thus aren't called by many pollsters, could upend all pre-election prognoses, or they won't.

— If first-time voters and African-Americans turn out in crushing numbers, stamp "never mind" over some of the surveys you're seeing today.

The caveats and variables of modern political polling are enough to cram a clipboard. The most glaring asterisk of all — that two out of three of us refuse to be polled — is just a fact of life in public pulse-taking.

This year, pollsters agree, the dangers of getting it wrong may be greater than ever — even if in recent elections they've had it mostly right.

"Our own polling in 2004 was spot-on" calling a 3-point edge to President Bush over Democrat John Kerry on Election Day, said Scott Keeter of the nonpartisan Pew Research Center for the People and the Press.

But back then, just 5 percent of the public relied only on cell phones. And recent Pew studies show today's cell-only users are disproportionately young, African-American and supportive of Democrat Barack Obama — 55 percent to Republican John McCain's 36 percent, as of a month ago.

Certainly, the reasons for one poll differing from others are numerous. Some polling groups, including Pew and the Gallup Organization, are reaching out to cell-phone users while others, such as Zogby International, are not.

Zogby spokesman Fritz Wenzel said requiring respondents to pay for the 40 minutes of calling time needed to complete some surveys "may corrupt the sample," so only those who can afford it participate.

Pew gets around that by reimbursing cell-users for their minutes.

The variations in survey methods don't end there. Some pollsters "weight" their samples to reflect the nation's party affiliations — Zogby is among them — but others don't.

Some surveys base the projected turnout of voters, and how that electorate will look, on past presidential elections. Others anticipate next week's turnout to be dramatically different in terms of age and race.

"Do I set my numbers to reflect an electorate that's 9 percent African-American — or will it be 13 percent this time?" asked Neil Newhouse of the GOP pollster Public Opinion Strategies. "Just a 1-percent difference in that question will have an impact on a poll's results."

Gallup is navigating the uncertainty by releasing two versions of its daily tracking poll — one "traditional" and the other "expanded."

The traditional poll assumes that turnout will mirror past contests, and the latest numbers give Obama a 3-percentage-point lead. In the expanded method his lead jumps to 7 percentage points — 51 percent to 44 percent — based on presumptions of higher black participation and first-time voters actually voting.

"Typically, we just take the responses of the people who say they're 'definitely' voting," said Ann Selzer, an Iowa pollster whose caucus forecasts have impressed the industry. "There are others who use the historical model that says the best predictor of future behavior is past behavior."

Under that historical model, a pollster may write you off if you haven't voted in the past — even if you say you'll definitely vote this time.

Selzer's polling before the January caucuses accurately foretold an advantage to Obama over Democrats John Edwards and Hillary Clinton. The Clinton camp charged that Selzer was overestimating the turnout of young voters.

As it happened, caucus sites statewide were jammed and nearly 60 percent who attended did so for the first time.

Selzer discounts the need to reach cell-phone users.

Knowing that they're younger than the electorate as a whole, she said the most reliable way of including their preferences is to "weight" the answers of young voters in the sample. Selzer called "modest or insignificant" the evidence that young cell-only users are politically different as a group than young voters who use landlines.

The biggest group of all — and the most mysterious — is the bloc that can't be reached or won't take part in polls.

For every 100 households called, 75 to 80 refuse to be polled or don't answer their phones.

The 66 percent who refuse is too large not to cover wide political ground. But when compared with the electorate as a whole, they're known to be more "racially conservative" and might hurt Obama at the polls, said Pew's Keeter.

"That's a difference that makes sense: It's comes down to how much you trust people you don't know."