Why the Fla. Exit Polls Were Wrong

In today's column:
• Behind Those Bad Exit Poll Predictions in Florida
• Grading the Polls
• How Swing Groups Swung

Samples of voters that contained too many Democrats, a mysterious miscount of the votes in the Jacksonville area and some bad assumptions led Voter News Service, the television network exit poll consortium, to make and then retract two dramatic election-night predictions on the winner of the presidential race in Florida.

"When something goes wrong, it's never just one thing that goes wrong," said Kathy Frankovic, director of surveys for CBS and a member of the VNS board of directors.

The first error led CBS and the other major television networks to declare Vice President Gore the winner in Florida shortly before 8 p.m. – and then take it back two hours later.

Frankovic said the problem centered in a handful of precincts in the Tampa area and a miscount of the vote from precincts in Duval County in the northern part of the state.

"In the Tampa area, the exit poll results from the [sample] precincts turned out to be more Democratic than the vote turned out to be," Frankovic said.

VNS yesterday interviewed 1,818 randomly selected voters in 45 carefully chosen precincts around the state. The results from these sample precincts as well as actual vote totals after the polls close, are the basis on which the networks make their election-night "calls" of the winner. After the polls close, VNS compares the results of their precinct samples with the actual vote to make sure their sample accurately reflects the vote.

Almost always, it does. But not last night in Florida. Before the networks made the call for Gore, VNS showed Gore with anywhere from a 3 to 6-point lead over Bush while the actual vote counts were showing a much closer contest.

There was another problem that led to the Gore call: "The vote tabulation also favored Gore and it should not have. It was either a miscount or a mis-entry of the data. I don't know who exactly made the mistake, VNS or the county."

But as actual vote counts streamed in from around the state, the discrepancy became increasingly more apparent, and the networks took back their call.

Then, shortly after 2 a.m., the networks made their second bad prediction of the night. "There was a lot of the tabulated vote already in, and we were paying close attention to the counties that had not yet been fully counted," Frankovic said. "Based on that, and the expected vote and the historical vote in these counties, we felt it was a very good call for Bush."

In fact, internal VNS estimates at the time suggested Bush would carry Florida by "tens of thousands" of votes – and win the election, Frankovic said.

But as the votes in these counties, primarily in heavily Democratic south Florida, started coming in, it became clear that "we had made assumptions about the vote that turned out not to be true. It was a lot closer than we thought it would be and we saw the lead get narrower and narrower" instead of widening.

Shortly after 3 a.m., CBS and the other networks reversed their Bush call, even as Gore prepared to make his concession speech in Nashville.

"We are embarrassed, obviously," Frankovic said. "And distressed. But at the end, we did the right thing. We took it back and admitted it."

Poll Watchers Grade the Polls

OK, so nobody really cares about the polls anymore. Well, almost nobody.

So how did those wacky polls do this year? To refresh your memory, here are the final calls by the major public pollsters:

Voter.com/Battleground: Bush 46, Gore 41, Nader 4, Buchanan * = BUSH +5
(After undecided voters are apportioned to the candidates, the estimates are Bush 50, Gore 45)
The Washington Post: Bush 48, Gore 45, Nader 3, Buchanan 1 = BUSH +3
ABC: Bush 48, Gore 45, Nader 3, Buchanan 1 = BUSH +3
NBC/Wall Street Journal: Bush 47, Gore 44, Nader 3, Buchanan 2 = BUSH +3
Pew Research Center: Bush 45, Gore 43, Nader 4, Buchanan * = BUSH +2
(with undecided voters: Bush 49, Gore 47)
CNN/USA Today/Gallup: Bush 48, Gore 46, Nader 4, Buchanan 1 = BUSH +2
CBS: Gore 45, Bush 44, Nader 4, Buchanan 1 = GORE +1
MSNBC/Reuters: Gore 48, Bush 46, Nader 5, Buchanan * = GORE +2

Our take: the pre-election polls don't look half bad from the vantage point of the indeterminate day after. (Come on, the election experts can't even tell you what the popular vote is and they've got actual ballots in their hands.)

All but one – the vaunted voter.com/Battleground poll (whose decision to weight their data to an even split in voters' party ID might need to be rethought after this election, see below) – had the race within 3 points.

CBS News looks to be the winner and new champion of pre-election polling: Their final poll estimated the race at 1 percentage point.

Averaging the numbers above gives you Bush 46.5 percent to Gore's 44.6 percent. So the "poll of polls" gives you a 2-point race that is, in words destined to become the mantra of this election, too close to call.

Worth Noting

Remember those groups we said yesterday we were watching? Well, here's what we saw:

• The Gender Gap. It was huge among men – and it worked to Bush’s great advantage. White men voted 60 percent to 36 percent for Bush, while white women split 49 percent for Bush and 48 percent for Gore. President Clinton won among white women by 5 points four years ago.

• White Catholics. Bush won 'em 52 percent to 45 percent for Gore, a reversal from the past two presidential elections when this group narrowly went for the Democratic candidate.

• Clinton Fatigue. The majority of voters – about seven in 10 – said their vote had nothing to do with the First Bubba. But among those who were trying to send a message to 1600 Pennsylvania Ave., the edge went to those who didn't have anything nice to say. In all, about two in 10 said their vote was meant to express opposition to Clinton, and about one in 10 said their vote was meant to express support.

We were watching those voters who like Clinton's work but not his persona. As predicted, one in three of these voters defected to Bush. Moreover, "honest" ranked as the single most important trait voters this year were seeking in the next president – and eight in 10 of these voters supported Bush.

• Turnout: The Small Pictures. Gore folks were hoping for a strong turnout among African Americans and they are likely reasonably satisfied. Black turnout held steady at 10 percent of overall turnout, identical to 1996 and two points higher than in 1992.

As predicted, blacks went strong for the vice president: 90 percent, compared to 8 percent for Bush. This is an even bigger Democratic vote than in 1996, when Clinton got 84 percent of the black vote.

It appears Hispanics made up a slightly smaller proportion of the electorate in 1996 (4 percent compared to 6 percent), even though in absolute numbers more may have voted than last cycle if this turns out to be a high turnout election. Gore won here 62 percent to 35 percent – decisive, but a narrower margin than Clinton in 1996.

• Young people: Generation Whatever-they-are-called-now split the vote rather than rocked it this year. Voters aged 18-29 made up 17 percent of the electorate, same as four years ago and down from 1992. And they divided their support between Gore (48 percent) and Bush (46 percent). This is worth more than an asterisk, since Clinton won in this group by 19 percentage points last go-around. Nader grabbed 5 percent of the vote here – fairly underwhelming.

• Nader Voters: Speaking of . . . Hard to believe some Nader supporters aren't ruing the day they told themselves their vote for the Green party candidate wouldn't affect the Gore-Bush contest.

According to exit polls, 47 percent of Nader voters would have gone for Gore if it had been a two-man race, and only 21 percent for Bush. (Three in 10 say they would not have voted.)

'Nuff said.

• Which Party Shows Up at the Party? Steady state here. The partisan breakdown of the electorate yesterday mapped exactly onto that in 1996: 39 percent Democrat, 35 percent Republican. And 1992 figures were also almost identical.

What this means: The parties were at parity in terms of energizing their voters.

Another thing this means: Pollsters need to have a serious, open (and not hostile) discussion about whether or not their data should take these numbers into account in some fashion.

• The SuperModels. Finally, those political scientists near and dear to our geeky hearts who slaved over mathematical models to predict this election a big win for Gore.

Well, they still could be right about Gore, and if so they'll get credit for going against the grain. But they called it big, and it ain't gonna be big, so it may be back to the drawing board in the ivory towers.