Volatile mood of voters, turnout among reasons polls differ so much

Voters who follow political news closely may feel whipsawed by new poll results in the Florida governor's race, showing Democrat Alex Sink with a narrow but significant lead over Republican Rick Scott.

That's because within the last week, two other polls by credible pollsters have shown Scott with narrow but significant leads.

That raises a question being heard a lot in political circles this year: What's up with the polls?

Polls in Florida's top races, governor and U.S. Senate, have been all over the map, varying widely even among respected pollsters.

For example:

•A Sept. 23 poll by Mason-Dixon Polling & Research, sometimes called the state's most reliable published political pollster, showed Sink with a 7-point lead over Scott, 47-40 percent. But over the next two days, Rasmussen Research and Quinnipiac University polls showed Scott ahead by 6 points, and CNN/Time showed Scott up 2 points.

•The same CNN/Time and Quinnipiac polls showed Republican Marco Rubio with comfortable, 11- and 13-point leads over no-party candidate Gov. Charlie in the Senate race.

But the following week, Zogby and the Republican-oriented Public Opinion Strategies, polling for the Florida Chamber of Commerce, showed Crist in striking distance, 6 or 7 points down.

•The Aug. 24 primaries, Scott v. Bill McCollum for the GOP governor nomination, and Jeff Greene v. Kendrick Meek in the Democratic Senate race, produced complete polling befuddlement.

A week before the voting, Mason-Dixon showed Meek 14 points ahead of Greene and McCollum narrowly leading Scott -- while an Ipsos poll showed Greene 4 points ahead of Meek, and Scott 10 points ahead of McCollum.

Three days before the primary, Mason-Dixon showed McCollum with a comfortable, 9-point lead and Quinnipiac gave him a 4-point margin -- but Public Policy Polling, a North Carolina-based, Democratic-oriented firm, showed Scott 7 points ahead.

When Scott won -- by less than 3 points -- a dejected Brad Coker of Mason-Dixon said it was the first time in 26 years of Florida political polling that his final poll called the outcome of a race wrong.

Political polls are never exact, but this year's variations are worse than usual, pollsters say.

"If polls are within 3 or 4 points, that's OK, but when you're seeing 10- and 12-point variances, somebody's wrong," said Jim Kitchens of the Maitland-based, Democratic-oriented firm The Kitchens Group, who has been doing political polls in Florida for 26 years.

The reason, he and other pollsters said, is an unusually volatile, passionate voter mood this year, making it harder for pollsters to produce an accurate turnout "model" - a forecast of how many people, of what ages, races and parties, will turn out to vote.

Pollsters use those models to adjust the samples of respondents in their polls.

"Usually polls are different because their samples are different," said Kitchens.

At the same time, there's been a sharp increase in the number of political polls published in Florida, from various sources, said Marian Johnson, vice president for political operations at the Florida Chamber of Commerce.

"There's a lot from universities and a lot from people I've never heard of before," she said. "Until I see their methodology, I just read them for the interest, because I'm a political junkie."

In some cases, the polls come from civic or other interest groups that
want to get in on the political action.

The union of Miami-Dade College faculty members, for example, has just begun publishing polls as part of an attempt to draw attention to higher education issues, said spokesman Keith Donner.

But some polls, Kitchens said, come from political operatives or others seeking attention and publicity, and using polls of questionable quality to do it.

"I see people trying to cheap it out," he said.

Automated dial polling, or "robopolling," in which computer make phone calls and respondents punch in their answers on the phone keypad, is rapidly gaining popularity among new pollsters.

The technique is far less expensive than human callers, and those who use it say it's just as accurate -- but traditional pollsters disagree.

"You may have a 12-year-old on the line, and you won't know it," Kitchens said.

Without an accurate turnout model, a pollster would spend time and money sampling the opinions of people who don't vote. But a turnout model depends on making political judgments - which party's voters are more enthusiastic, and which candidates will draw minority or elderly voters.

Primaries, with lower turnout, are often harder to predict than general elections, which contributed to the varying results in the primary, Coker said.

A change in state law last year may have affected this year's primaries, Kitchens said.

The state made absentee ballot applications good for two years rather than one, so voters who ordered absentee ballots during one of the most exciting presidential races in years are getting them this year as well - during an off-year election when turnout normally drops as much as 50 percent.

The result was thousands of unexpected voters in the Aug. 23 primary.

Coker's poll conflicted with Ipsos in the primaries mainly because Ipsos sampled all registered voters, while Mason-Dixon screened its respondents for likely voters.

Even Ipsos pollster Julia Clark said at the time that likely voters were a better sample for a primary, with its expected low turnout. She said her client, the St. Petersburg Times, wanted a registered voter sample for other questions on the poll.

In the Democratic race, that turned out to be true - Meek won. But in the GOP race, with an usually high turnout, Ipsos was right -- Scott won.