Having covered political campaigns closely over the past decade and a half, I'm used to hearing the same questions over and over from friends and family members less immersed in the minutia of elections. The hardest question by far to answer is this one: How can the networks possibly know who won a race before the votes are counted? The answer isn't simple but let's give it another shot.
First, let's start with the exit polls. Six organizations – the Associated Press, CBS News, ABC News, NBC News, CNN and Fox News – pool their resources to support the Election Day survey of voters who show up at polling locations all over the country. This is the National Election Pool Exit Poll, and is the one used by the networks and most commonly cited in post-election examinations (there are other exit polls and some of these organizations supplement them with independent efforts but let's stick with the NEP). The NEP is run by the firms of Edison Media Research and Mitofsky International.
Edison's Joe Lenski told CBS correspondent Russ Mitchell that this year's poll will be conducted at 1,000 polling places and will question over 100,000 voters. Selected voters are asked to complete a questionnaire after having already voted about their attitudes toward both national and local/state issues of importance. Mystery Pollster Mark Blumenthal has an excellent explainer providing more details about how the surveys are done, which anyone interested should check out. Here's how Edison Media describes the exit poll on its website:
Exit polls are interviews with voters after they have cast their votes at their polling places. The polling places are a scientifically selected sample of polling places that collectively represent a state, or for the national exit poll they represent the nation. An interviewer gives every nth voter exiting the polling place a questionnaire to complete. There are questions about demographic such as gender, age, race, and issues related to the person' s vote and questions about the person' s vote choice in the different contests. Participation is voluntary and anonymous. Interviewing starts when the polls open in the morning, continues throughout the day until about an hour before they close at night.This data is compiled throughout the day and, in the past, preliminary numbers have circulated throughout the afternoon and have shown up on the Internet. The big change for 2006 is the institution of the "quarantine room." Because the exit poll numbers that end up on the Web are raw and incomplete, they are often an inaccurate snapshot of voting behavior. Blumenthal explains:
Once the polls close, the interviewer will attempt to obtain actual turnout counts, and if possible, actual vote returns for their precinct. One of the unique aspects of the exit poll design is the way it gradually incorporates real turnout and vote data as it becomes available once the polls close. The exit poll designers have developed weighting schemes and algorithms to allow all sorts of comparisons to historical data that supports the networks as they decide whether to "call" a state for a particular candidate. When all of the votes have been counted, the exit poll is weighted by the vote to match the actual result.Because representatives of all six participating media organizations are present at Edison during the day, looking at the numbers as they flow in, the early results have been passed around among correspondents, producers and others involved in election coverage. Ostensibly, this helps prepare journalists and gives them a bit of a headstart on figuring out where the political winds are blowing. This year, those representatives will be locked away and unable to communicate with their colleagues until 5:00pm. The result should be no leaked numbers.
Contrary to what some may believe, the exit polls are not the sole source of calling races, as Blumenthal notes above. Each organization using exit polls has a team of experts pouring over them and pulling in other indicators. They look at the vote tabulations compiled by the AP, examine the historical voting patterns take into account absentee voters (who are surveyed by phone before Election Day) and just about any other factor you can imagine. For example, if an exit poll in a precinct which has traditionally favored one party by wide margins suddenly has the other side winning or if the exit polls indicate larger or smaller numbers of women voting than usual, the red flags should be waving. Once satisfied that the data is solid, the "call" is made.
Despite huge problems with the system in Florida during the 2000 campaign and a couple hiccups in other campaigns, election night calls have an outstanding track record for accuracy. Because of 2000 (where networks called Florida for Gore, then Bush before pulling back altogether) networks have an incentive to be more cautious than they may have been in the past. Competitive pressures are also involved (everyone wants to be the first to call a state or election), but don't play as big a role as in the past. Since the networks began pooling their resources, they all have access to the same data at the same time as everyone else. That's why they all "call" races at just about the same time in the vast majority of races.
At CBS News, they do not "call" races, preferring to use the word "estimate" – as in, CBS News estimates that so-and-so will win the race. That's a pretty accurate description for the process. Because some of the prediction is based on what amounts to a random survey of voters, there can be errors and unpredictable outcomes so it's not an iron-clad guarantee. But it's not a wild guess either -- it's a very well-educated projection of a final outcome.