A day before two of the most-watched primary races in Florida's gripping political history, the question at work, lunch or the gym may be, "Who's going to win?"
Kendrick Meek will defeat Jeff Greene in the Democratic U.S. Senate race, you say. And Bill McCollum will beat Rick Scott in the GOP gubernatorial primary.
But wait, this is the year of the outsider. Greene and Scott will prevail.
Picking the outcome of a close race is always a tossup, but this year is particularly vexing. A blizzard of polls has produced different winners and losers, often days apart.
"I feel like I'm left to analyze races by Braille," said Jennifer Duffy, senior editor of the Cook Political Report, who is tracking both Florida races.
Polls. They are the ultimate insider's tool, part science, part art, treasured and disdained, and, increasingly, part of Florida's political landscape.
In newspapers, on TV or radio or websites, come numbers from Quinnipiac, Mason-Dixon, Ipsos, Rasmussen, McLaughlin & Associates, Tarrance Group, Public Policy Polling, Voter Survey Service and others. Not to mention polls for political parties and candidates.
"We're one of the most interesting states in the nation, so there is more polling," said Dave Beattie, whose Fernandina Beach company, Hamilton Campaigns, has long conducted polls in Florida.
But, as Beattie points out, not all polls are created equal. The disparate results of recent weeks have touched off a debate about the value of the surveys and the different methods used to produce them.
A little over a week ago, the St. Petersburg Times, Miami Herald and Bay News 9 published Ipsos Public Affairs polls showing Greene edging Meek and Scott with a commanding 10-point lead over McCollum.
That news came amid a new Mason-Dixon survey showing Meek trouncing Greene by 14 points and McCollum leading Scott by a few.
Then came a Voter Survey Service poll showing Scott defeating McCollum. But a Quinnipiac University poll showed McCollum up by 9 percentage points.
Over the weekend, came a poll from Mason-Dixon with McCollum and Meek ahead 45-36 and 42-30, respectively. On Monday, came two new polls: One showing Scott out front; the other narrowing McCollum's lead.
Is Florida that fickle? In short, no.
The main reason the Ipsos poll had such different numbers was it looked at registered voters, while the Mason-Dixon poll narrowed down to "likely voters."
Julia Clark, who conducted the Ipsos poll, said the objective was to take a broad look at issues such as oil drilling and voter sentiment in Florida, and not just focus on the horse races between Meek and Greene, and McCollum and Scott.
But coming so close to the primary, the horse race is what generated attention — giving way to complaints that the poll painted an inaccurate picture or at least was an outlier.
Take the Meek and Greene contest. Greene has outspent Meek by millions on TV, going from unknown to serious challenger in weeks. The Ipsos poll confirmed that.
Turnout in midterm primaries is typically low, however, drawing only the most energized voters. In other words, likely voters. That would seem to favor Meek, who may not be a household name but is known among party activists and has a stronger ground organization.
"Meek's base is African-American and white upscale liberals," said Mark Blumenthal, editor of Pollster.com. "I think there's real strong evidence those two populations are more in the hard-core primary group."
That's why Mason-Dixon and other polls show him on top and why Meek stands a good chance at winning on Tuesday. The same factors hold true for McCollum, a better-known quantity among hard-core Republican voters.
Clark countered that many voters were still undecided, so trying to narrow down likely voters could have been "premature." The Ipsos poll, conducted Aug. 6-10, showed nearly one-third of Democrats had not made up their minds between Greene and Meek.
If turnout is higher than expected Tuesday, that could favor outsiders Greene and Scott.
Polling is expensive, and even more so to capture likely voters. To get there, pollsters use a variety of "screens."
A questioner asks a person a series of questions: Are you a registered voter? Are you a registered Democrat or Republican? How likely are you to vote? Very, not very, somewhat? Some ask voters to rate their interest in the election on a scale of 1 to 10.
Even then, some people say they will vote — to not seem apathetic — but never do.
"Selecting likely voters involves a lot more art than science," Blumenthal said. "There's a lot of judgment calls."
And there's a lot of ways to slice responses and balance them for geographic and demographic areas of the state. Women vote more, so a poll that does not take that into account may not be as accurate. There is a high concentration of Hispanics in South Florida, but not all pollsters use bilingual questioners.
In fast-paced campaigning, where a gaffe or hard-hitting TV commercial can hit any day, timing of polls also plays a role. Making calls on Monday instead of Wednesday could yield different results. Same with calling on a weekend vs. a weekday.
All of which explains why polls — even those who reach for the same group of likely voters — can produce different results.
"Most of it's not worth the paper it's printed on," said Rich Heffley, a Florida Republican insider who has helped with polling, most recently the Tarrance Group survey conducted for the Florida Medical Association. It showed McCollum up on Scott.
"There needs to be some level of quality control," Heffley said, suggesting the media establish set standards before agreeing to publish results, either in print or online. "There are so many ways to screw up polling and very few ways to do it right."
One of the more controversial trends in the field is Interactive Voice Response polls. Rasmussen Reports and Public Policy Polling are two companies that have jumped into Florida in recent years. Voters who pick up the phone are greeted by a recorded voice that guides them through questions that are answered by pressing phone digits.
The knock on "robo" polls, which are less expensive to produce, is that a person with casual interest in politics is more apt to hang up on a recording, which can make for a less random sample.
But robo polls have vastly improved over the past decade, and comparisons show they match the accuracy of those with live questioners, at least in the horse race. "I've come to appreciate over the years that they're not so bad," Blumenthal said.
Blumenthal's company thinks it has found the right way to assess races: by plotting various polls to form a trend line. As of Sunday, Pollster.com showed Meek beating Greene and McCollum beating Scott.
Tuesday, we'll know for sure.
Alex Leary can be reached at email@example.com. Follow him on Twitter @learyspt.