Why the networks' call kept changing

By ADAM C. SMITH

© St. Petersburg Times, published November 9, 2000

The alarming and cryptic e-mails started hitting newsrooms at 9:38 p.m. election night:

"WE'RE CANCELING THE VOTE IN COUNTY 16," said Voter News Service's first message, refering to Florida's Duval County. "THE VOTE IS STRANGE."

Then the bombshell at 10:13: "WE'RE RETRACTING OUR CALL IN FL BECAUSE WE DON'T HAVE OUR PREVIOUS CONFIDENCE."

By then, of course, the TV networks had already begun leading America on an election night roller coaster ride. Just before 8 p.m., anchors declared Al Gore the winner in Florida, only to retract the projection two hours later because of dubious exit polling data, including from Tampa. Then, at 2:18 a.m., the networks called Florida for Bush, only to back off again less than two hours later.

It was a stunning blunder by the networks and a reminder of the vagaries of polling. Skewed numbers are possible with any poll before Election Day, but the stakes are much higher with exit polls used by news outlets to call elections.

"It's a high wire act," said Frank Newport, a pollster with Gallup, which does no exit polling. "Exit polling, unlike anything we do, is an incredibly intense situation where you take data and operate with the clock ticking right in front of you."

Networks used to hire their own exit polling outfits, but for the past decade they have all relied largely on the same organization to gather and crunch the numbers for calling winners before all the votes are counted. Voter News Service is the consortium run by NBC, ABC, CBS, CNN, Fox News and the Associated Press, and has dozens of subscribers, including the St. Petersburg Times.

The staff of the New York-based corporation grows to nearly 45,000 for national elections. On Tuesday, its small army of pollsters included people surveying 1,818 randomly selected voters in 45 Florida precincts, as well as people phoning in vote tallies from select county and voting precincts across the state. They take those numbers, compare and massage them, and normally pick the winner.

But a combination of problems helped VNS miss the mark on one of the most crucial states in the presidential election. At 4 p.m. VNS showed Gore leading Bush 49 percent to 47 percent, and by 7:30 p.m., the exit poll showed Gore leading 51 percent to 46 percent. Shortly before 10 p.m., they had Gore leading 49 percent to 48 percent.

Warren Mitofsky, a veteran exit pollster who used to head VNS and worked with CBS and CNN analyzing the polling data Tuesday, said that a sampling of six precincts in Tampa included too many Democrats and that VNS received inaccurate vote count reports from the Jacksonville area. That inflated the number of Gore votes in the normally Republican-leaning area.

VNS issued a brief statement Wednesday evening saying that initially VNS and its members considered the race too close to call. But their confidence grew after reports of actual vote counts in sample counties and precincts supported their exit survey numbers in those areas.

"These models, based on sampling precincts, have served us well through many elections. However, we will investigate why they did not work properly in this specific situation," VNS said.

VNS is well-regarded in the industry. Unaffiliated pollsters Wednesday suggested that if anyone blew the Florida call, it was the networks who declared the winner too fast.

This wasn't the only time VNS fell short on its exit polls, however.

In a Democratic U.S. Senate primary in Texas in March, it wrongly reported that the third place finisher had finished in second place. In 1996, it wrongly projected the winner a U.S. Senate race in New Hampshire and overestimated Pat Buchanan's showing in both the New Hampshire and Arizona presidential primaries.

© 2006 • All Rights Reserved • Tampa Bay Times
490 First Avenue South • St. Petersburg, FL 33701 • 727-893-8111

Contact the Times | Privacy Policy
Standard of Accuracy | Terms, Conditions & Copyright