Deborah Leavy: Don't always believe the polls

HE'S UP by three points! Down by five! They're dead even!

That's what the media tell us, breathlessly, every day.

But don't get too excited - these polls don't matter that much. Not just because the only poll that matters is on Election Day - and even with exit polls, the only one that matters is still the one in the voting booth.

"A national horse-race poll is like People magazine - it shows us what's hot at the moment," said William Rosenberg, a professor at Drexel who's an expert on presidential polling.

And this race is harder to track than usual. Polling results could be easily distorted by the Bradley effect, the cell-phone effect and a host of other factors.

The Bradley effect is named after Tom Bradley, the mayor of Los Angeles who in 1982 ran for California governor against George Deukmeijian, the state attorney general. Polls showed Bradley ahead right up to Election Day - but Deukmeijian won.

Social scientists theorized that, because Bradley was black, some white folks felt uncomfortable telling pollsters they weren't voting for him. Instead, they said they were for Bradley. In the privacy of the voting booth, however, they pulled the lever for Deukmejian.

Will the same scenario prove true for Barack Obama, who's been mostly up in the polls? Or might it be the reverse, with polls overestimating the black vote for Obama, missing African-Americans who don't want to admit they support McCain? We just don't know.

And then there's the difficulty pollsters have in calling cell-phone numbers, which aren't in the phone book. And the biggest users of cell phones, who may not even have a landline, are young people, who are entering the electoral process in droves to enthusiastically support Obama. By not counting their opinions, are pollsters underestimating the Obama vote?

Could be. When phones were first used for polling, results were skewed because only the affluent, who tended to be Republican, could afford them. In the same vein, one early poll relied on sample ballots inserted in a magazine. Subscribers, better-educated and better-off financially than voters in general, tilted the results in favor of the loser.

The trick to polling accurately is to ask a sample that accurately represents all the voters, but there are many variables. Pollsters usually work from lists of registered voters, which could mean missing the newly registered. In Pennsylvania, Democrats have added 320,000 voters so far this year. Republicans have lost 60,000.

Some polls sample only likely voters, but which ones will actually vote if it rains on Election Day? Pollsters rely in part on past voting patterns, but again will miss new voters. Different pollsters use different statistical models to predict who will actually vote. They pose the questions differently, interpret answers differently and end up with different results - and no one knows if any of them are right. They say they try to correct for these flaws, but it's an inexact science.

And there's something even more fundamentally wrong with national polling: - It reports only national results. Gathered from samples across the country, a national poll at its most accurate represents only the popular vote.

But, of course, that's not how presidents are elected. We elect them through the Electoral College, with results tallied state by state. In the 2000 election, Al Gore won the popular vote, but Bush ended up with enough electoral votes to win.

This distorts the results of a national poll. For example, if Obama wins Pennsylvania by one vote, he gets all of the state's 21 electors. If John McCain wins Alaska by 100,000 votes, he gets Alaska's three electors. A poll of those two states would show McCain ahead, but Obama would be ahead in electoral votes and closer to winning the presidency.

AND DON'T forget about the other candidates: independent Ralph Nader, Libertarian Bob Barr, and the Green Party's Cynthia McKinney. I haven't seen their names reported in a poll, yet even the small number of votes they get can be significant. The votes Ralph Nader took that otherwise might have been cast for Gore cost Gore the election in 2000.

It's likely that candidates will go up and down between now and Election Day. Just remember, as Yogi Berra once said, "It ain't over till it's over." *

Deborah Leavy is a regular contributor to the op-ed page and an associate member of the Daily News editorial board. E-mail her at deborah.opinion@gmail.com.