Exit polls have been a source of controversy in recent elections. While there will still be challenges, changes have been made since the last presidential election.

Senior Fellow Karlyn Bowman

Exit polls have been a source of controversy in the past four elections. Will there be problems again on Tuesday?

A little background: Exit polls got their start 40 years ago in the 1967 Kentucky gubernatorial contest. Then, the late Warren Mitofsky of CBS News conducted the poll. Edison Media Research and Mitofsky International will conduct this year's poll for the National Election Pool, a consortium of the five major networks (ABC, CBS, NBC, FOX and CNN) and the Associated Press. About 3,000 people will be involved in the 50-state effort, and more than 100,000 voters will be surveyed.

Interviewers are sent to pre-selected precincts around the country, and they hand a questionnaire to every nth voter (n, or the skip interval, is computer-generated) leaving the polling place. The poll is usually printed on both sides of a single piece of paper, and voters are asked to check boxes that provide basic demographic information about themselves and answers to election-related questions. Interviewers phone their results into exit poll headquarters several times during the day.

While the interviewing is going on, the Associated Press deploys a small army to collect actual vote returns throughout the United States. Because many Americans will vote absentee or early--possibly as many as a third this year--the exit pollsters also survey people before Election Day to ask whether they have voted and, if so, how. Those results are combined with the results collected on Election Day. Edison then provides its estimates and the rest of the exit poll data to the consortium members; the desk analysts at the networks review it and decide when to call an election.

In 2004, Edison/Mitofsky provided exit poll data and election estimates for the presidential race and for 120 other races across the 50 states. They surveyed voters at 1,480 locations and conducted 500 phone interviews in 13 states with a high proportion of absentee/early voters.

But exit polls aren't always reliable, and the controversy surrounding the 2000 election is the most infamous example. The exit pollsters and news organizations awarded Florida to Al Gore before the polls closed and then retracted it. Later, the networks (but not the exit pollsters) awarded Florida to George W. Bush--and then retracted that call, too.

There was a lot of blame to go around. Absentee votes were undercounted, and the election model was flawed. A simple mistake in data entry inflated Gore's vote. Fierce network competition to be the first to call the contest triggered the premature calls. After the election, the networks vowed not to release results in any state until all the polls in that state had closed.

The machinery was overhauled for the 2002 elections, but that year, massive technical failures botched exit poll reporting. In 2004, and again in 2006, the exit poll overstated Democrats' performance. In the 1,460 exit poll precincts where Edison/Mitofsky collected both exit poll tallies and actual final vote returns in 2004, the exit poll results overstated the actual difference between John Kerry and Bush by 6.5 points in Kerry's favor.

This problem was due, according to a post-election analysis by Mitofsky, "to Kerry voters participating in the exit polls at a higher rate than Bush voters." This was the largest problem of its kind in the past five presidential elections. Although it was a serious problem for the pollsters, he said, this error did not "lead to a single incorrect NEP [National Election Pool] winner projection on election night."

Can these past problems be overcome on Tuesday night? To compensate for the potential oversampling of Democrats, Edison/Mitofsky have improved interviewer training and are using fewer young interviewers. They have compromised with states that had kept interviewers far from the polling locations, which made data collection difficult. And as in 2006, the networks won't get the data until 5 p.m., which will help to prevent the kind of leaks that suggested--too early--that Kerry would win in 2004. Additional safeguards will check the poll against actual votes. These changes should improve the reliability of the only tool we have to look at the views and values of actual voters and how they have changed over time.

Karlyn Bowman is a senior fellow at AEI.