YOU ARE HERE: LAT HomeCollections


Valuable but Volatile Exit Poll Shows Its Weakness

Statistics: It took a close election like Tuesday's to expose the limitations of sampling opinion after votes are cast. People can lie, demographics change.


The wild swings in predicting who won Florida in Tuesday's presidential vote are highlighting the danger of relying too heavily on an inexact but crucial tool of modern election coverage: the exit poll.

"We've gotten used to the success of these polls because we haven't had a close election in 20 years. We haven't had a test case. Last night we did," said Margo Anderson, an expert on the history of polling at the University of Wisconsin at Milwaukee.

Of her own state, where Gore squeaked by with a 6,124-vote margin, she said: "Frankly, it was more efficient to count the ballots than do a prediction."

Voter News Service conducts the exit polls that Associated Press and the five major networks use to make their projections. VNS pollsters declined to be interviewed about the problems in Florida, on which this year's presidential election is pinned.

But in a statement about the foibles of their Florida predictions, they said their models "have served us well through many elections. However, we will investigate why they did not work properly in this specific situation."

Nevertheless, predictions have become standard fare--sometimes coming out just minutes after polls close. In 1980, Jimmy Carter was (correctly) declared the loser, and conceded before West Coast polls closed. How do the experts make such predictions long before the ballots are counted?

Largely because such polls are so expensive, only two organizations conduct national election night exit polls. They are Voter News Service and the Los Angeles Times. The two organizations send battalions of interviewers to voting precincts around the country to gather information on votes that have been cast. The information has to be collected, sent in and analyzed--within about 12 hours.

"It's a very, very hard day and very hard to do," said Susan Pinkus, director of the Los Angeles Times Poll.

The real work starts long before the election, Pinkus said, when pollsters select precincts from around the country that will be representative of the entire nation. They pull these together based on the historical voting trends there. Then they create models that accurately reflect current voting patterns.

"We put maps over maps over maps," she said. "It takes four to six months."

Culling this sample of representative precincts can be a challenge for the most demographically volatile states--like Florida.

"Essentially, every four years in Florida a lot of people die and a lot of new people move in and a lot of new people become citizens. So doing comparisons with the past is difficult because the place changes more than other parts of the country," said David Lublin, a political scientist at American University.

Donald Saari, an expert on the mathematics of voting at UC Irvine, agreed. "The techniques we've been using are based on past experiences. They're full of errors simply because society changes."

Collecting data also poses problems. Interviewers take data from voting places in at least three waves--morning, afternoon and evening--because each time slot may have its unique population of voters.

Some trends identified by pollsters: White-collar workers vote early in the day, while blue-collar workers vote after their shifts end. Retirees vote midday. Men are more likely to vote in the morning and women in the afternoon.

Taking a swath of voters from only one time of day runs the risk that the sample will be skewed and will not represent the public at large. "The basic problem is you're trying to guess how 10 million people will vote based on a small sample size," said Mary Gray, a professor of math and statistics at American University.

An unexpectedly large election turnout, as was seen in some areas Tuesday, can also skew exit polls, which are based on historical voting levels. Exit polls may have underestimated the votes coming out of inner cities, for example.

Polling has come a long way since 1936, when a poll conducted by the Literary Digest picked Alf Landon over Franklin D. Roosevelt. The problem was a classic sampling error: They mailed out the poll to telephone owners. In 1932, most people who had telephones were quite rich and not likely to vote for the New Deal Democrat.

While sampling techniques have gotten far more sophisticated, pollsters are desperately trying to stay abreast of the rapidly changing demographics, economics and psychology of the voting populace. "You have to run fast just to keep in place," Gray said.

Other problems center on the voters themselves.

"People lie," said Gray, especially when there's a sensitive issue involved--say, a candidate who's a racist. Many pollsters say a rising number of voters are refusing to give out any information.

Los Angeles Times Articles