News Analysis : Inaccurate Poll Results Laid to Bad Polling

Fingers were pointing in all directions Wednesday over why the news media polling in this week’s elections was so inaccurate, particularly in historic races in New York and Virginia, where polls put black candidates much further ahead than they actually were.

The commonly heard explanation is simply that people lie, telling pollsters they will vote for a black candidate when in fact they will not. That is the same explanation some offered in 1982 when Los Angeles Mayor Tom Bradley lost his bid to become the nation’s first elected black governor.

But, in fact, the evidence this week and interviews with experienced pollsters suggest that the idea that people lie to pollsters out of racism is mostly a myth.

The problem is not bad people but bad polling, often conducted for the media. Pollsters and election analysts said the faulty polls may fail to predict the ethnic makeup of voter turnout in such races, fail to account for those who refuse to answer questions after voting and fail to interpret other factors. Unexpectedly high turnout by white voters in both New York and Virginia, for instance, did skew polls conducted on and before Election Day, analysts said.

The controversy over Tuesday’s election is most closely focused on Virginia, where exit polls conducted for media organizations showed Democrat L. Douglas Wilder, the black candidate for governor, beating Republican J. Marshall Coleman by 10 points. The media regularly uses such polls, which interview voters as they leave polling places, to project winners and losers.

Led by a Fraction

When the Virginia vote was actually counted, however, Wilder led Coleman by fractions of a percentage point, fewer than 8,000 votes, a margin so close that a recount is expected.

To a lesser extent, results in the New York City mayor’s race have raised questions. The CBS-New York Times exit poll showed black Democrat David N. Dinkins beating Rudolph W. Giuliani by 6.5 points. The actual margin was between two and three points.

“I never expected an exit poll to be any more precise than that,” said Warren Mitofsky, director of polling for CBS.

But in Virginia, “The bottom line is, people lied,” contended Tim Gallagher, director of political and news media affairs of Mason-Dixon Opinion Research Inc., which conducted an exit poll paid for by five television stations. “I don’t know how you protect against it,” he said.

If Gallagher is right, it calls into question the accuracy of any exit polling conducted in races involving ethnic and racial candidates, and casts a cloud over a major tool used by the media in covering politics.

But others disagree with Gallagher.

To improve accuracy, “there are absolutely things they can do,” said Paul Maslin of the Democratic polling firm of Hickman-Maslin, which conducted polling for both of W. Wilson Goode’s successful mayoral campaigns in Philadelphia and the late Mayor Harold Washington’s in Chicago.

“There are lots of reasons for having bad results,” said Mitofsky of CBS. But voters’ lying on exit polls is not one of them. “If they are lying, this is a new phenomenon.”

While the news media rely heavily now on exit polls, such polls are among the most expensive to do and the most vulnerable. And too often the media, trying to save money, hire polling firms that lack the expertise to do them well.

“Exit polls look so simple and everybody does them and they are a mine field,” said California pollster Mervin Field.

One problem is that a growing number of people are refusing to answer exit polls. This is particularly true in California, where there is antipathy toward the media’s projecting the outcome of national races before the state’s polls have closed.

Whatever their reasons, most of those refusing to answer exit polls tend to be older white voters. And in campaigns where race is an issue, these white older “refusals” tend overwhelmingly to vote for the white candidate, Maslin said.

So unless pollsters account somehow for “refusals,” their polls will be wrong.

Field’s California Poll, for instance, now “weights refusals” by inserting a blank or dummy ballot for anyone leaving a voting place who refuses to answer a question. The pollster notes the likely race and approximate age of that voter on the dummy ballot. And when the results are tabulated, Field in effect adds such voters to the total, predicting their vote based on their demographic makeup.

‘Cluster Sampling’

A second problem with exit polls is that because of their cost they require “cluster sampling,” or conducting interviews with people in only a handful of sample precincts. The Virginia exit poll, for instance, interviewed voters at 60 out of 1,900 precincts.

A traditional telephone poll, in contrast, contains samples of every voter group from every region.

In conducting exit polls, “in effect you have to violate some fundamental sampling laws,” Field said.

Unless the pollster has picked the sample precincts carefully and knows how to weight the data to accurately reflect the eventual turnout, the chances of error are great.

And a third problem is accounting for absentee voting in exit polls.

In California, for instance, absentee voting now represents close to 15% of all voters, and tends to be 80% Republican.

In 1988, the failure to account for that in exit polls led ABC and NBC to initially characterize the Senate race incorrectly as going to Democrat Leo T. McCarthy and the presidential race in the state as going to Democrat Michael S. Dukakis.

The problem of polling in campaigns with a racial component is not limited to exit polls either.

A week before the election, for instance, the Washington Post poll had Wilder 11 points ahead in Virginia, while other polls showed the spread closer to four points.

‘People Don’t Talk’

A key reason is the failure of the pollster to accurately predict who will vote, especially by ethnic or religious group.

“People don’t talk about this very much, but all polls, including exit polls, do corrections for turnout based on informed judgments,” said William Schneider, political consultant for The Times.

Usually, pollsters base estimates on past elections--how many Catholics, Jews and blacks, for example, have gone to the polls. “But if there are short term factors, such as a black candidate bringing out an unusual number of white or black voters,” Schneider said, it can become very difficult.

What happened in New York and Virginia, pollsters said, is that the presence of a black candidate brought out an extraordinarily high percentage of white voters. Hence, the historical models the pollsters used to calculate their results were faulty.

Finally, another polling problem in campaigns with a racial component is that an overwhelming number of “undecided voters” are whites who are struggling with the idea of voting for a black candidate. And, history indicates, almost all of them will choose the white candidate on Election Day.

“My rule of thumb is all the white undecideds will go to the white candidate’s column,” Maslin said.

With polls in campaigns involving race, it is not so much lying voters as undecided voters who can skew poll results.

But pollsters can, if careful, attempt to account for that.

And then there are the media. One problem is that reporters now use partial exit polling data gathered early in the day to do reporting for early editions and newscasts, polling data that is necessarily sketchy and thus erroneous, pollsters said.

“I have given up,” said CBS’s Mitofsky. “They shouldn’t pay attention to (such early figures). I don’t pay attention to them.”

If the idea that voters lie to pollsters is unfounded or at least exaggerated, what happened in Tom Bradley’s gubernatorial race in California in 1982?

Several polls, including The Los Angeles Times Poll, showed Bradley winning handily. With a week to go, Field’s California Poll showed Bradley ahead by seven points.

One problem, which cropped up again Tuesday, was that pollsters overestimated minority turnout.

Field and others expected it to be about average, 20%. But since Bradley chose to run as a mainstream candidate, not mobilizing his base, minority turnout was only 15%.

Another factor was the failure to measure undecideds.

And when exit polls failed to predict George Deukmejian’s victory over Bradley, that was partly because pollsters had not anticipated that relaxed restrictions on absentee voting would cause a rise in that segment of the vote, and that it would go 80% Republican.

Field now asks people in pre-election votes if they are voting absentee, and weights those figures into his exit polls.

“I have always felt that black-white races are the most challenging test of what people in the polling business do,” said Democratic pollster Maslin. “We should all learn from this. It was probably a good thing that happened yesterday.”