After exit polls got a deservedly bad reputation for initially predicting a John Kerry win in 2004, pundits and politicians alike were hesitant to put their faith in the early-voting numbers. So when the exit polls tonight came out showing very positive data for Barack Obama, Democrats weren't, understandably, ready to declare victory yet.
The Kerry fiasco was just the worst example of a problem that has long been endemic to exit polls. Over the past 20 years, they have slightly favored the Democratic candidate for President in every election. Many observers believe that's because Democrats are more willing to talk to exit pollsters. Exit polls also suffer from an unusually high margin of error due to the difficulty of choosing sample precincts that mirror all of the diverse precincts in the area.
But if this year's exit polls turn out to be more accurate than in past years, as they appear to be, it's not by accident. The National Election Pool (a consortium of ABC News, Associated Press, CBS News, CNN, Fox and NBC News), working with Edison Media Research and Mitofsky International, made changes this year in an effort to do better.
For one thing, there has been a strict quarantine in place since 2006 to avoid early leaks of totally unreliable data which is what happened in 2004. This year, no exit poll data was released until 5 p.m. E.T., and only a handful of National Election Pool officials were allowed to see the early data.
The exit pollsters are also getting better training to help them avoid oversampling Democrats. Here is how an exit poll is supposed to work: pollsters stand outside more than 1,000 precincts around the country, all of which have been scientifically chosen to represent a particular area. As voters leave their polling places, the interviewer tries to randomly select about 100 of them to fill out a questionnaire.
For example, the interviewer might approach every ninth voter. But this is a lot easier said than done when you are standing more than 40 feet outside of a busy polling place at rush hour. About half of the people approached decline to participate (a percentage that has been steadily rising over the past several decades).
In recent years, exit pollsters have received better training to help them stick to the random selection. This is not easy to do with a large labor force that works for you only one day a year. But it may also help that exit pollsters now include some older people which may entice more older voters to participate.
In some places, pollsters are allowed to stand closer to polling places than they were before, which may also help improve the response rate. And the questionnaire is a little shorter, so more voters may be willing to participate.
Still, the challenge of choosing precincts that accurately reflect the broader region remains immense, as does the quest to pull a truly random sample of voters. The 2006 congressional election included a bias for the Democrats once again, and several of the Democratic exit polls during the primary contest between Obama and Hillary Clinton ended up being wrong, even though some of these reforms were already in place. Still, for this election at least, the exit polls do not appear to have gone too far astray.