On Election Day, in a room somewhere in New York City cut off from the outside world, a small group of media representatives will spend hours poring over polling data from around the country. No cell phones or Internet connections will be allowed, and the group will not emerge until 5 p.m. E.T. to share what they have learned with their bosses. These people are part of the National Election Pool (NEP) and they owe their monastic retreat to a long-running debate on how early election reports can affect the outcome of a race.
Exit polling surveying people leaving voting locations about the ballots they cast debuted in the 1960s, as news organizations (and on a small scale, candidates) sought to gather demographic data about voters that could be used to predict election results. Legendary polling pioneer Warren Mitofsky conducted the first major exit poll for a network during the 1967 Kentucky governor's race and by the 1970s, exit polling had become an industry practice. But in 1980, NBC reported Ronald Reagan's 1980 victory over Jimmy Carter nearly three hours before polls closed on the West Coast, leading to a large-scale examination of exit polling and Congressional hearings on whether it depressed voter turnout. As a result, networks vowed not to project a state's winners until polls there are closed. States have tried and failed to restrict exit polling, which is protected by the First Amendment. (Ironically, the U.S. government is a big supporter of exit polling abroad: the practice is widely used by pollsters hired by NGOs and monitors to verify that elections are being conducted legitimately. The U.S. government has even financed exit polls in former Soviet republics and satellites to ensure votes are counted accurately.)
In the 1990s, the major news networks and the Associated Press formed a polling consortium called Voter News Service (VNS) to cut costs, eliminating the redundancy of reports from multiple sources. But redundancy isn't always a bad thing, as proved, disastrously, in 2000 when VNS (and the networks soon afterward) declared the race for Al Gore around 8 p.m., only to switch to George W. Bush at 2 a.m. and declared the race locked at "too close to call" two hours later.
An embarrassing computer glitch in 2002 sealed the consortium's fate; it was shuttered soon after and replaced by a different set of pollsters that serve the National Election News Pool. But this organization suffered its own scandal in 2004 when exit poll data was leaked online around midday on Election Day, prompting bloggers to declare John Kerry the presumptive winner. In 2006, the pollsters began quarantining representatives of the NEP to prevent such leaks from occurring.
The current exit poll system uses a careful methodology that includes sampling voters during various periods of the day (certain demographic groups tend to vote at different times) and conducting telephone surveys of those who voted early or by absentee ballots. The pollsters for this year's election also say they are making an effort to ensure the organization's more than 1,000 surveyors are diverse in previous elections, surveyors tended to be young, and presumably attracted younger participants (who are more likely to be Democratic). The questionnaires are filled out anonymously and deposited into boxes, which pollsters say helps decrease the Bradley effect, in which voters don't to tell pre-election pollsters that they're planning to vote for white candidates over black candidates. About half of all those asked to fill out exit poll questionnaires decline.
Actual projections are made by "decision desks" at the networks small groups of journalists and polling experts who use exit data and actual returns to decide when to announce winners. So even though all the networks have access to the same exit poll data, they often don't broadcast projections at the same time. During the primary season, the networks used exit poll data to slice the electorate into various demographic groups giving viewers proof, for example, of Barack Obama's strength over Hillary Clinton among black voters and Clinton's popularity among older voters. These tidbits help fill airtime when networks have exit poll data, but can't release figures on who's winning until polls are closed in a given state. Expect to see this again on Election Day, in between the time the NEP representatives are released from quarantine and a winner emerges.
But quarantine or no quarantine, news outlets still remember Florida in 2000; if swing-state races appear tight when the last polls close, odds are the media will be cagier about releasing early results no matter how good the data look.