The following appeared at brillscontent.com December 18, 2000:
On Election Night 2000, something went horribly wrong. Here, the story of how bad numbers and network laziness caused a media meltdown.
by Seth Mnookin
n a November 14 memo to the board of directors of Voter News Service, Murray Edelman, the polling consortium’s editorial director, discusses an internal investigation “that the members and the lawyers have asked us to conduct.” The investigation is meant to uncover how and why VNS failed at the one thing it is designed to do: accurately call presidential and state elections. This year, counting only statewide races in which the top two candidates finished within 5 percentage points of each other, VNS miscalled about 10 percent of the races it was hired to project, including the presidential race and a key Senate race.
VNS, and the television networks it works for, failed so spectacularly because it didn’t factor in the massive shifts in how Americans vote. Brill’s Content gained access to VNS documents, including screen grabs of the VNS numbers that resulted, first, in the calling of Florida for Vice-President Al Gore and, later, the calling of the state, and the presidency, for Texas governor George W. Bush. Network and VNS officials have been predictably parsimonious with their comments, but the VNS documents, combined with interviews and transcripts of the networks’ election-night coverage, explain how a multimillion-dollar project designed to serve the public ended up doing exactly the opposite.
The Marriott at the Capitol in Austin, Texas, was calm the morning of November 7. The hotel was booked solid — hundreds of newspaper reporters, dotcom scribes, and television producers had reserved rooms months in advance — but there was little of the frantic energy that fuels campaign life. No predawn baggage call. No urgent press releases or pool reports. No earnest spinning or misplaced luggage.
The weather was miserable — rainy and cold. The last several blocks of Congress Avenue, leading to the steps of the capitol building, were cordoned off. An enormous riser faced the capitol, and TV crews were setting up their klieg lights and cameras. Behind the riser and past a security checkpoint, a spacious media tent housed two rows of tables; each table, assigned to four or five reporters, was equipped with phones and electrical outlets.
Just after 2 p.m. central time (CT), I called my editors in New York. The election looked tight, at least for the time being. Bush wasn’t doing as well across the South as had been projected, and Gore didn’t look good in Pennsylvania. Most important, the battleground states that would likely determine the election — states like Michigan, Iowa, and Florida — were all close.
These numbers and projections were being generated out of New York by VNS. The Associated Press and five television networks — ABC, CBS, CNN, Fox News, and NBC — fund VNS, and many other print and broadcast outlets pay to receive its data. Brill’s Content does not subscribe to VNS, but editors here, like editors and reporters at other newspapers, magazines, and websites across the country, were leaked exit-poll information on election day.
The networks don’t rely solely on VNS to make projections; VNS supplies raw data and analysis, from which network employees make projections. Each of the networks has its own “decision desk,” where a team of analysts reviews the VNS reports as they come in, hoping to call states minutes or even seconds before others. It’s unclear what the networks hope to gain. As Martin Plissner, a former political director at CBS News, writes in The Control Room: How Television Calls the Shots In Presidential Elections, “The principal beneficiaries of this entire exercise...would be the egos of the news executives; hardly anyone in the greater world, even in the world of media and politics, knew, let alone cared, about this rat race among the networks.”
At 1 p.m. eastern time (ET) the first round VNS data was released. It came from 28 states, including Alabama, Florida, Illinois, Kentucky, Michigan, Mississippi, New Hampshire, Ohio, Pennsylvania, and West Virginia. This data contained about a third of the total exit-poll questionnaires from those states; another third would come in around 4 p.m., and the rest would trickle in until after the polls closed. This first round of numbers didn’t look good, at least not from the perspective of the decision desks. State after state showed four or five “bads,” or precincts in which the data seemed too aberrant to be trusted. A maximum of two bads is considered acceptable out of 40 or 50 polling points, but these numbers meant that early on, up to 10 percent of the networks’ information was unreliable.
This was troubling but not disastrous. For one thing, there was still time to correct bads. Furthermore, exit polls are just one part of a complicated equation used to make projections. Historical factors, such as a state’s voting record and changes in its population, are included. As polls close, sample precinct data of real results — data culled from poll workers by VNS — are added in. Finally, the actual county-by-county vote is used. This information is used by VNS to alert subscribers to a state’s status: win, call, lead, even, and rev (reverse). Before the 2000 presidential election, VNS had had only one rev: In 1996 it declared Democrat Dick Swett the winner of a New Hampshire seat in the U.S. Senate; in fact, the seat was won by Republican Bob Smith.
Jonathan Alter, a Newsweek columnist and commentator at NBC and its cable affiliate, MSNBC, remembers when he saw the first round of numbers. “At 1 o’clock, I went to the decision desk at 30 Rock [NBC’s headquarters in New York City’s Rockefeller Plaza],” Alter says. “They were getting the early exit polls...and Jeff Zucker [executive producer in charge of election night] said, ‘If Pennsylvania doesn’t change, it’s over, Gore lost’....It was a sign of how wrong the early exit polls were. They already had a bunch of states wrong.” Alter noted that the people at NBC all knew early exit polls often changed dramatically before election day was over. (Gore won Pennsylvania by more than 200,000 votes, 51 percent to 47 percent.) Zucker says that “all throughout that afternoon...the VNS model was showing that Gore had lost Pennsylvania....Everyone assumed...there was no way Gore was going to win because he was going to lose Pennsylvania.”
Still, there was time to correct the Pennsylvania error and other errors in VNS’s models. The country’s earliest poll closings were hours away. At 2:47 p.m. ET, VNS sent out an alert to its subscribers that read: “The problems with the state survey weighting are cleared up. We have cleaned out the bad precinct problem.”
By the time the second round of data started coming in, at 4 p.m. ET, there were at most one or two bads, out of 40 to 50 precincts for the large states. But the news that VNS was correcting its models on election day had some network employees worried. At CNN, Judy Woodruff remembers when she was told about the 2:47 p.m. ET alert. “There was a report that VNS was relooking at its models....It was a signal to me that something was different about this election,” she says. “The fact that VNS, that we and others were paying a lot of money to subscribe to this, and the fact that on Election Day they were looking at whether their model was correct, it told me that something different was going on this year. It wasn’t coming out the way they expected.”
This apprehension — the sense that VNS’s models might be faulty or that the results weren’t going to be as easily analyzed as the relatively lopsided races of 1992 and 1996 — was not new to Election Day. Some former network executives believe that VNS’s model, which had never been tested on a close election, was doomed to blow up in a race tighter than Clinton-Bush in 1992.
But none of these concerns was shared with viewers. In the early evening, Dan Rather assured the CBS audience that “if we say somebody’s carried a state you can pretty much take it to the bank, book it, that that’s true.” He said, “Let’s get one thing straight from the get-go. We would rather be last in reporting returns than to be wrong. And again, our record demonstrates that [to be] true. If you hear someplace else that somebody’s carried a state and you’re off, as you shouldn’t be, watching them, then come back here.”
By 5 p.m. CT, the temperature in Austin had dropped further, and the press tent had filled up. The three large-screen TVs were tuned to CNN. “In the 7 p.m. eastern hour,” political analyst Bill Schneider was saying, “polls close in nine more states — Florida, Georgia, New Hampshire, South Carolina, Vermont, Virginia, North Carolina, Ohio, and West Virginia.” These states were as good a sampling as journalists could hope for: Florida, New Hampshire, and Ohio were all hotly contested; West Virginia had been traditionally Democratic but looked like a possible pickup for Bush; and the four other southern states would indicate whether Bush was “making his numbers” in parts of the country in which he was favored. At the time it didn’t look good for the Texas governor: He wasn’t doing well in the South, and all the swing states remained in play.
While VNS was supplying data and analysis, the networks were supplying on-air commentary. At 7 p.m. ET, every network was talking about the poll closings in nine states. And every network was wrong: The polls were closing in only eight states. Not a single network seemed to realize that the Florida panhandle is in the central time zone and the rest of the state is in the eastern time zone. The polls in that heavily Republican part of the state wouldn’t close for another hour — 8 p.m. ET. That wasn’t the only factual error the networks made on election night, mistakes that in other years would have been quickly forgotten but that heighten a sense that the networks weren’t as prepared or informed as they should have been.
In the Austin press tent, no one was thinking about the Florida panhandle. That the state wasn’t called immediately was a signal that the race there was tight. As Schneider ran down the poll closings across the country, he had warned CNN viewers that Florida might not be decided until after midnight. Rather was saying the same thing on CBS: “We’re waiting on a possible decision in Florida, but you’ve got time to put on another cup of coffee and pour it because, in Florida, it’s generally considered to be so close that it may be a long while before anybody is able to call it.”
Back at the decision desks, analysts knew it wouldn’t be that long. By 7:30 p.m. ET, the state was looking like it could tip for Gore. By 7:40, it was almost ready to call. VNS had one bad precinct, and VNS’s model estimated that Gore would carry the state 51.1 percent to 46.5 percent. In general, VNS methodology dictates that to call a state, a candidate’s “crits,” or individualized projections based on VNS data and analysis, should be at least 2.7; at that point, there is less than a 1 in 200 risk of error. At 7:40, Gore’s minimum crit was 2.7; his most important crit indicator was at 3.3. The indicators were based on a combination of exit polls from 45 precincts, a small amount of the state’s tabulated raw vote, county models, past voting patterns, current projections, and a smattering of unofficial countywide results supplied to VNS by poll workers at 120 precincts. (There are 5,885 precincts in Florida.) VNS gave Florida a call status.
Warren Mitofsky, the founding head of VNS and a decision desk analyst for CBS and CNN, recalls the scene: “The exit poll was showing a very small lead for Gore. When you combine it with real returns, it’s showing a real lead for Gore. When you look at the exit-poll calculation of error and the first handful of county votes, they’re all telling you that Gore is ahead.... So now I’m looking at all of that, and I couldn’t imagine a safer call if I had to make it up myself. A projection is not made until the chances of making a mistake are at 1 in 200 or less. And when we called Florida, just before 8, it was significantly less than that.”
At 7:55 p.m., VNS changed Florida from a CALL to a WIN GORE; at this point, NBC, CBS, and CNN had already declared Gore the winner there. Gore still had to win both Pennsylvania and Michigan to have a decent shot at the presidency — what the networks were referring to as the “trifecta” — but that was looking increasingly possible.
Within minutes, Mary Matalin, the conservative CNN commentator, began raising doubts about the Florida call. “Well, you’re — I’m going to go out on a limb here,” she said on-air. “We have early data. The spread is 2 percent. The raw total is 4,000 votes at this point. If it continues at this pace, there are half a million absentee ballots out there. I’m just telling you, this reminds me of [Governor George] Deukmejian in California [in 1982]: lost on Tuesday, won on Thursday.” But Matalin was dismissed as a partisan stalwart. “Well, when we do call the state, we’ve taken the absentee ballot count into account,” Schneider said. “When we call the state, we’re pretty sure that state is going to go for the winner.”
Matalin wasn’t the only person questioning Florida. Over the next couple of hours, members of Bush’s team in Austin spoke up as well. Karl Rove, Bush’s chief political strategist, went on NBC at about 9:30 p.m. ET and admonished the network. “I would also suggest that Florida has been prematurely called,” he said. Rove pointed out that Florida has two time zones. “First of all, I thought it was a little bit irresponsible of the networks to call it before the polls closed in the western part of Florida. Florida is still split among two time zones, eastern and central. You all called it before the polls had closed in the central part of the country.”
Despite the fact that they were sharing data, the networks had acted as if they were alone in projecting winners. But after Rove’s chastising — and a defiant appearance by Bush — anchors strained to point out that if there was a mistake, everyone had made it. Right before 10 p.m. ET, Peter Jennings said on ABC, “You know that we have projected Florida for Mr. Gore. I think everybody’s now projected Mr. Gore winning in Florida. Mr. Bush says he’s not yet ready to concede Florida.”
In fact, as Jennings was talking, the networks had known for almost an hour that Florida was in trouble. “I can’t tell you what a bad feeling that was,” Mitofsky says. “But when you start to notice that the margin you felt was there starts to decline, you don’t immediately yell ‘Pull it off the air.’ So we watched it for a short time to make sure we were going to be wrong. We wanted to make sure some aberrant piece of data didn’t come in.” Around 9:30 p.m. ET, VNS began warning subscribers about bad data in Florida. At 9:38, VNS sent a message that read, “We are canceling the vote in Cnty 16 — Duval Cnty, FL — vote is strange.”
“It did look like it wasn’t going for Gore,” Mitofsky says. “I don’t remember how small the margin was, but...it didn’t look like he was going to eke out a small victory either. We sure had no confidence that Gore was going to hold on to a small lead, and this was too important a decision to let it sit there.” Mitofsky, in a conference call with CNN and CBS, told them to pull back Florida.
On CNN, Jeff Greenfield was talking when Bernard Shaw broke in.
Shaw: “Stand by, stand by — CNN right now is moving our earlier declaration of Florida back to the too-close-to-call....”
Greenfield: “Oh, waiter...”
Shaw: “Into the too-close-to-call column.”
Greenfield: “One order of crow.”
Bill Schneider: “One order of crow, yes.”
Woodruff, who was also on CNN at the time, remembers when Tom Hannon, the executive producer on election night, spoke into her earpiece. “It was very direct,” Woodruff says. “Tom is always very succinct. He just said, ‘We’ve got to pull back Florida.’ And that was it. I just thought, Oh, my God, how horrible. I was not aware so much of what other nets were doing. We have to crane our necks to look all the way across the newsroom, and so I could sort of make it out, but I couldn’t tell exactly who they were calling. I was concerned about us. I didn’t want us to be making mistakes.”
At CBS, Rather was aware of what was going on at the other networks, and he was careful to share the blame: “Based on what we believed, and most other people believed at the time — I know of nobody who didn’t believe it.... It turns out some of the data is suspect.”
Indeed, there were some suspect data, most notably in Duval County, but that didn’t seem to be the main problem; in fact, VNS’s correction of the Duval County vote probably increased the time it took to pull Florida back from Gore. The main problem, as Matalin had indicated, seemed to be that VNS underestimated the number of absentee ballots, both in Florida and the rest of the country. That was the reason VNS gave to its subscribers for pulling back Florida. In a computer message that went out at 10:13 p.m. ET, VNS said, “We are retracting our call in FL because we don’t have the confidence we did and we are still examining the absentee vote.” This was followed up by the November 14 memo to the VNS board from Edelman, who wrote, “I still believe the biggest problem in the model is that we did not correctly anticipate the impact of the absentee vote.”
Although Edelman’s postmortem memo addresses only Florida — and most of the country’s attention was focused on Florida — VNS muffed a number of other calls around the country. At about the same time that the networks pulled Florida back from Gore, they awarded New Mexico to the vice-president, only to retract it later. VNS also declared Maria Cantwell the winner of a Senate seat from Washington state, saying she had beaten incumbent Slade Gorton, only to call back that race later. Here’s the VNS message, which went out to subscribers at 4:58 a.m. ET on November 8, that explained the pullback: “We have to retract Washington Senate — a good part of the absentee vote remains to be counted over the next 10 days. The current margin for Cantwell is not enough to maintain the call over that period.” On CNN at about 5 a.m. ET, Shaw explained, “Officials in Washington state say the absentee ballots are being counted and they’re not finished counting, so we back off.” (Although it looks as if, ultimately, both the New Mexico and Washington calls were correct, they weren’t accurate calls at the time. New Mexico went back and forth for weeks, and Cantwell wasn’t declared the winner in Washington until December 1.)
These mistakes combine to paint a picture of VNS’s model as seriously flawed. On the basis of about 30 close races nationwide — senatorial, gubernatorial, and statewide presidential races that were decided by 5 percentage points or less — VNS fumbled three calls, or 10 percent of those it was hired to make.
On election night, the early Florida gaffe was pushed to the side in what became an increasingly tense drama. There hasn’t been a close presidential election in more than 20 years, and the television anchors, as well as the journalists assembled in Austin and Nashville, where Gore was holed up, were feeding off the excitement. In the press tent, reporters, who had “electoral calculators” on their screens, were figuring out which combination of states would translate into 270 electoral votes for one candidate or the other. The Washington Post’s website was printing raw vote totals as they came in, and I toggled among screens: the Missouri Senate race, in which the incumbent, John Ashcroft, was locked in a tight battle with a dead man, Mel Carnahan; the Washington, Oregon, Iowa, and Florida presidential vote totals; the national vote totals. It was freezing in the press tent — Reuters’s Patsy Wilson had wrapped one of the paper tablecloths around her for warmth. Rain was pouring down on the increasingly intoxicated crowd that had gathered in front of the governor’s mansion. Just before 11 p.m. CT, Rather said, “Welcome back to CBS News election headquarters. Let’s pause and take a deep breath, appreciate it for what it is. This is the dance of democracy. This is as close as we come to a kind of sacred time in this country.”
The crowd in Austin had expected Bush to win ever since Florida had been pulled back from Gore. But people were wet and tired and cold, and enthusiasm waned as the night wore on. At 11:30 p.m. CT, Jeb Bush’s eldest son, George P. Bush, tried to pump up the crowd, exhorting “guys” and “females” to chant “We want Florida.” Barely anyone responded. But people did respond to the vote totals, and in Florida, Bush was flirting with a 100,000-vote lead.
At 2:05 a.m. ET — 1:05 in Austin — with 96 percent of the vote counted, VNS showed Bush with a 29,386-vote lead in Florida, with 185,157 votes not yet counted. Gore would need 57 percent of those votes to win, which was possible but unlikely. Networks were discussing a Bush presidency. By 2:15, VNS was showing that Bush’s lead had jumped to 47,861 votes, with 102,204 to be counted. Now Gore needed 72.4 percent of the remaining votes to win. VNS divided Florida into five regions; Gore was doing best in the “Miami/Gold Coast,” in southern Florida, but was drawing just 60.1 percent of the vote there. VNS said it was 99.9 percent certain that Gore would lose Florida — the state on which the presidency hinged. A minute later, Fox made the call. Brit Hume was on-air at the time: “Got to interrupt you. We are now calling — Fox News now projects George W. Bush the winner in Florida, and thus it appears the winner of the presidency of the United States. Fox News projects George W. Bush the winner of the presidency of the United States based on the call we now make in the state of Florida. And, so there it is, Bush the apparent winner. I must tell you, everybody, after all this, all night long, we put Bush at 271, Gore at 243. I feel a little bit apprehensive about the whole thing. I have no reason to doubt our decision desk, but there it is.”
The other networks followed suit. Doris Kearns Goodwin, the presidential historian, was speaking on NBC when Tom Brokaw interrupted her.
Brokaw: “Stop. Stop. Doris, Doris, Doris, Doris, Doris, Doris.”
Goodwin: “Uh-oh; something’s happened.”
Brokaw: “George Bush is the president-elect of the United States. He has won the state of Florida.”
On ABC, Jennings asked Cokie Roberts if “this is it?” She replied, “Yes. It’s been a very cautious night, Peter, after the initial first bad call, and so I think that, unlike other times, instead of rushing to make calls, here it is, what, 2:30 in the morning. So this is likely to be it. Yes.”
CNN made its call after cutting to correspondent Candy Crowley in Austin; viewers could see the crowd celebrating the announcement on Fox, though CNN had not made a call of its own. By 2:20 a.m. ET, every network had declared Bush the winner. The media filed out of the press tent and into the holding pen in front of the governor’s mansion. Red, white, and blue lights bathed the white columns. A Bush video set to Stevie Wonder’s “Signed, Sealed, Delivered” showed on the JumboTron, and a young woman sang “America the Beautiful.” As the JumboTron flitted from image to image, the beery crowd — Bud and Bud Light had been selling for $3 a cup — cheered at dissonant images: A defiant Ralph Nader got some lusty applause, as did a shot of the depressed, equally wet crowd outside the War Memorial in Nashville.
Then we waited in the cold. Mark McKinnon, a former Democrat who had helped craft Bush’s media strategy, made a tipsy appearance on CNN, then went back, he said, to “have some more tequila.” The networks began their postmortems, and the dailies scrambled to get in one last dispatch. And unbeknownst to the reporters huddled outside in Austin, Bush’s lead was dropping.
By 3:15 a.m. ET — 2:15 in Austin — I began to wonder what was going on. The crowd had been told 45 minutes earlier that Bush would be making an appearance in about 20 minutes, and hadn’t been told anything since. I walked past the security guards and back to the press tent. I saw that Bush’s lead had dropped to about 11,000 votes. On NBC, Brokaw was talking about a recount: “You know, it’s — it’s technically possible that there could be a recount that could flip that [11,000-vote Bush lead]....And we haven’t seen Vice-President Al Gore. Perhaps they’re waiting for all the votes to be counted, all the — all the i’s to be dotted, all the t’s to be crossed. That would be something if the networks managed to blow it twice in one night.”
On CNN, Shaw and Woodruff were getting anxious while waiting for Gore’s concession speech.
Woodruff: “I think we’ve waited now a sufficient number of minutes since we know his motorcade arrived at this location; it is appropriate for us to ask questions about what is the delay all about....”
Shaw: “He could be talking...”
Woodruff: “...because presumably there were remarks that were prepared either before he left the hotel or on the way over.”
Shaw: “Well, John [King, CNN’s correspondent in Nashville] reported that he wanted to add — that he actually wanted to put some finishing touches to his remarks. He could be on the phone with Bill Clinton.”
Bush’s lead continued to drop, and reporters in Austin, in Nashville, and on the networks were talking about figures from the Florida secretary of state’s office; those numbers showed an even smaller lead for Bush. On-air at ABC, political director Mark Halperin began to question the call. “I — I feel like I’m the sole person here standing in the way of the Bush transition,” he said. “But, you know, we see here, the Florida State Department Division of Elections website shows a very small margin — this is brought to our attention by, amongst other people, the Democrats down in Nashville, who say they look in their computer at this website and they see 48.9 to 48.9, a very small margin of a thousand votes between the two men....” But Jennings told Halperin to ignore the website: “And I think I’m hearing in my ear that we believe that the Voter News Service is ahead of the Florida Department of State’s numbers at this point.” But at VNS headquarters in Manhattan, folks weren’t so sure. A 3:27 a.m. ET message from VNS to subscribers read, “Florida — The Sec of State website has a much narrower margin for Bush. We are comparing county by county trying to determine discrepancies.”
Back on CNN, Shaw, Woodruff, and Schneider were talking about what was now a 600-vote margin, a figure that, as Schneider pointed out, “is like one floor of a condominium in South Florida. It’s a very, very small number of people.”
Then word came down that Gore’s campaign manager, William Daley, would make a statement. As Daley walked onstage, Woodruff said, “Before he speaks, John, CNN will confirm there will be a recount in the state of Florida. And CNN moves it to the undecided column and back from George W. Bush.”
Daley was succinct. “Just an hour or so ago, the TV networks called this race for Governor Bush. It now appears that their call was premature,” he said. Reporters in Austin were speechless, clustered around the television sets in the press tent. “Let me be very clear about this....Without being certain of the results in Florida, we simply cannot be certain of the results of this national election....And until the results — the recount — is concluded and the results in Florida become official, our campaign continues.”
After Daley left the stage in Nashville, Jennings announced that ABC was moving Florida back to the too-close-to-call column. “So now we have our second major switch of the night,” he said.
On CBS, Rather pleaded with his audience not to blame him for this second huge gaffe: “I’m always reminded of those west Texas saloons where they had a sign that says, ‘Please don’t shoot the piano player; he’s doing the best he can,’“ he said. “That’s been pretty much the case here tonight over this election.”
While we waited in the cold in Austin, the networks continued to scramble. On ABC, Jennings and Halperin were trying to figure out what would happen next; they finally decided that the Florida secretary of state would oversee the recount. Jennings asked Halperin for the name of the secretary of state. “Her name is Katherine Harris,” Halperin responded. “She’s a Republican, and we’re going to learn all about her. Fourth-generation Floridian.” Around 3:30 a.m. Austin time — 4:30 in New York and D.C. — we realized that this was not going to be resolved soon. There were 224 votes separating Bush and Gore in Florida: 2,902,733 to 2,902,509 — a difference of about four one-hundredths of 1 percent.
Over the next hour, the press tent cleared out, and the TV anchors who had been on-air through the night finally were replaced. In Atlanta, Judy Woodruff left the CNN studio at 6 a.m. ET; she had been “sitting in those seats” since 4:45 p.m. the day before. “I only got up once to go to the bathroom in those 13 hours,” Woodruff says. “You couldn’t take your eye off this story. You knew it was a once-in-a-lifetime thing.”
In the days and weeks following November 7, plenty of blame went around for the bad calls made on election night. John Ellis, a cousin of George W. Bush’s and the head of Fox’s decision desk, was accused of pressing his fingers to the scale: In this scenario, Fox’s call, the first among the networks’, prompted everyone else to follow suit. But Ellis had no control over the VNS numbers that showed it was 99.9 percent definite that Gore would lose. As Mitofsky says, “this business about Fox pressuring other people to call, I never made a projection in my life because of some other network. When I heard they put it out there, I was disappointed, because I wanted to do it, but I was in the process of reviewing the counties, one at a time....I wanted to make sure there were no bad numbers. We were about to make the projection.”
Mitofsky also says calling Florida for Bush would have happened — and happened unanimously — even if there wasn’t one polling consortium. “That second call would have happened no matter what,” he says. Looking back at the VNS data from election night, I’m not sure I agree. At 2:05 a.m. ET, Bush was leading by 30,000 with 185,000 votes left to be counted. At 2:10 a.m., VNS reported that Bush’s lead had increased to 51,000 votes, with only 180,000 left to be counted — meaning Bush had picked up 21,000 votes, even though only 5,000 additional votes had been counted, according to VNS’s calculations. That discrepancy seems to have been overlooked in the rush; it was the 2:15 a.m. VNS data that prompted the final call. But if every network had been number-crunching independently instead of just interpreting the same batch of data, it’s conceivable someone would have caught the apparent 16,000 phantom votes.
As it turns out, it was those 16,000 votes that led to the faulty call. In his memo to the VNS board, Murray Edelman, the polling group’s director, talks about those votes: “There was a ‘correction’ in Volusia County at 2:08 that showed a major drop in votes for Gore. The screen at 2:05, just prior to the change in Volusia County, showed a lead of 29,386 for Gore [sic — the lead was actually for Bush] which increased to 51,433 five minutes later.” At 2:15, VNS was showing that there were 1,405 votes remaining to be counted in Volusia County; Gore would pick up about 21,000 of those votes, or almost 20,000 more votes than VNS thought were outstanding. As Edelman explains in his memo, “This ‘correction’ was corrected at 2:48. At the 2:50 time point, the [Bush] lead dropped.” And more votes remained to be counted: “At 2:10,” Edelman writes, “the end of the night model estimated 179,713 votes outstanding. This turned out to be an underestimate since there were over 359,000 votes yet to come in.” Volusia County was not the only county that VNS had miscalculated. The VNS model projected 41,000 votes outstanding in Palm Beach County at 2:10. “[B]ut 129,000 votes actually came in. This difference could have been due to errors in the county or because many absentee ballots were included at the end. It could also have been that the very large precincts in the county reported at the end.”
After reading this explanation, Mitofsky’s claim that Florida would have been called for Bush even if each network had its own polling operation is even less persuasive. It seems clear that a major problem with VNS’s models is that they failed to take into account that how Americans vote has changed dramatically since 1996. Many more people vote by absentee ballot, which was as easy to anticipate as it was neglected. Absentee ballots skew exit polls: People using those ballots don’t show up at the local school on election day and can’t be asked how they voted. Jeff Gralnick, a former executive at ABC and CNN, says the VNS system was “waiting to explode.” He compares the networks’ faith in the reliability of the VNS exit-poll models with the pre-Challenger mind-set at NASA. “They were launching space shuttles,” he says. “Nothing can go wrong. You become so secure in your belief in your own technology that you just keep doing it until it blows up.”
With additional research by Justina Kessler.