NEW YORK – Two firms that conducted Election Day exit polls for major news organizations reported Wednesday that they found a number of problems with the way the polls were carried out last year, resulting in estimates that overstated John Kerry's share of the vote.
Edison Media Research (search) and Mitofsky International (search) found that the Democratic challenger's supporters were more likely than President Bush's supporters to participate in exit polls interviews. They also found that more errors occurred in exit polls conducted by younger interviewers, and about half of the interviewers were 34 or under.
The news organizations — ABC, CBS, NBC, CNN, FOX News and The Associated Press — formed the consortium to get exit polling data for the 2004 election after a previous group known as the Voter News Service was disbanded.
In November 2000, flawed information from VNS twice led television networks to incorrectly declare a winner in the presidential race in Florida, the state that proved to be key to the outcome. And in the 2002 election, VNS was unable to provide its members and other clients with results from exit poll surveys.
Exit poll material is used to help make projections of winners and to supplement the vote count with an analysis of why people voted as they did. The data is not meant to be made public, but several Web sites posted leaked exit poll material on Election Day 2004 suggesting a Kerry lead.
In an effort to stem any leaks, the news organizations already have agreed to withhold the distribution of exit poll information within their organizations in future elections until late in the day, instead of releasing the data in earlier batches.
Edison and Mitofsky said problems contributed to exit poll data that overstated the vote for Kerry nationally and in 26 states, while data for four states overstated the vote for Bush.
They noted that in a number of precincts, interviewers were kept 50 feet or more away from polling places, potentially skewing results toward people motivated to go out of their way to participate in exit polls. They also found suggestions that interviewers may not have carefully followed rules for selecting voters at random, which may have skewed results.
The polling firms said they believed the exit poll errors were not the result of the selection of precincts where the interviews took place or the analysis of the data once it was received. They also said they found no evidence to suggest fraud by rigging of polling equipment.
Despite the problems, the firms noted that they still made correct calls for all races on election night.
In their report, the firms suggested several steps to mitigate errors in future exit polling efforts, such as using fewer young interviewers in races where that might make a difference; taking steps to assure that interviewers can conduct exit polls closer to polling stations; stepping up training procedures for interviewers; and revising the length and format of exit polls to see if more voters can be enticed to participate.
The report noted that discrepancies between exit polls and actual voting results also have occurred in previous elections, but not to such a great degree. Joe Lenski, the head of Edison Media Research, said the error tended to show up in elections with a high level of passion among the electorate, such as the 1992 vote in which Bill Clinton defeated the first President Bush and Ross Perot.
Younger interviewers often get lower response rates from exit polls, Lenski said, but what was different this time around was that that factor resulted in data overstating the results for one candidate.
"You look at the factors out there, and young voters in this election were the strongest supporters of Kerry by age group," he said. "Older voters seeing a younger interviewer may have been less likely to participate because they might believe that interviewer might not agree with them politically."
Lenski said raising the level of voter participation in future exit polls would reduce error. Only 53 percent of those asked agreed to complete the questionnaire in 2004, in line with the past.
Lenski said the firms were delivering the full polling data Wednesday to research centers at the University of Connecticut and the University of Michigan, following procedures from previous exit polls.
News executives said they were encouraged by the polling firms' suggestions for reducing error in future polls, but noted additional work remained to be done.
"We're pleased that the report is finally out and that people will have an opportunity to know exactly what we know happened on election night," said Kathleen Carroll, executive editor of the AP. "But the report clearly identifies some problems that need further scrutiny, and we support Edison and Mitofsky continuing to devote some serious energy to understanding those problems."
Bill Wheatley, a vice president at NBC News, said his network was glad the firms were working to eliminate future problems but that NBC would be keeping a close watch on the issue.
"In any public opinion polling, there's going to be a margin of error," Wheatley said. "We just need to make sure that margin of error is reasonable."