Polls: they're useful tool, but take with a grain of salt. Polls: both their accuracy and impact are overstated

Since 1960 polls have figured more prominently in the reporting and discussion of each succeeding presidential election. With this increased use has come increasing concern - both about how competently the survey findings are being handled, and whether in some deeper sense the electoral process is diminished by our attending so much to its ``horse race'' dimension. The biggest problems I see in how the poll data have been interpreted in the current campaign stem from two common errors: (1) attributing to polls a greater degree of precision than they can possibly deliver, and (2), relatedly, assuming that all responses people give to survey questions equally reflect firmly held judgments.

Those of us who practice survey research have not done nearly enough to impress upon our ``consumers'' that a great variety of different factors conspire to make the findings very soft.

Instead, we have often encouraged the mistaken notion that the findings of a well-designed survey are akin in their accuracy to the readings of a finely calibrated thermometer.

For example, a survey taken by the New York Times and CBS News Sept. 21-23 found George Bush leading Michael Dukakis by six percentage points, while another survey by the same organizations done Oct. 1-3 put Mr. Bush's lead at two points. Commenting on these results, the Times said that their polls showed the race tightening.

It would be difficult to overstate how wrong-headed that conclusion was. That the race was getting closer was one possible hypothesis - and not a very likely one at that. The most likely explanation of the shift was that it resided somewhere in normal imprecision of poll-taking. A great many other polls were being conducted, and the overall pattern they showed was not that of a tightening race.

Good polls taken by competent professionals have been shown repeatedly to yield imprecise results. In October 1984, for example, major national polls varied enormously in their pictures of the Reagan-Mondale contest. One put Mr. Reagan up by nine points, while another, done at the same time, had him ahead by 25 points. The range in recent national polls - where Bush's margin has fluctuated between 7 and 17 points - simply duplicates that experience.

The further away from election day a poll is conducted, of course, the softer its findings are likely to be - because many of those polled are likely not to have reached any real decision. If a voter isn't bound by party loyalty, and if as well he doesn't follow politics closely, he is unlikely to decide on his presidential vote in May or even in August.

The decision comes later, when real world events - above all, the approach of election day - demand it. Before that, this voter will give you an answer if you ask him his expected vote, but that answer is likely to reflect only superficial features of the campaign's rhythm or flow.

Even if we were ready to acknowledge the frailty of poll findings, the problem of polls' intrusiveness would still remain. Literally hundreds of election surveys have been conducted at the state and national level this month alone. If the results of any one of them are prone to error, the composite picture - which shows George Bush well ahead - is almost certainly accurate. In the age of polling, one of the candidates is likely to have to conduct the last weeks of his campaign in an environment where almost everyone believes on good grounds that he will lose.

This is polling's ultimate intrusiveness. Something within many of us rebels at being told a week or two before election day - and told scientifically, no less - what the outcome is likely to be.

It is one thing to resent polls' intrusiveness, however, and quite another to claim we know what the consequences are of their telling us who is going to win.

New York Times columnist A.M. Rosenthal asserts that ``there may be journalists who believe that the heavy emphasis given to the predictions of a Bush victory has had no bandwagon effect [but] I just don't know any.'' To which I can only reply: There may be political scientists or public opinion analysts who believe that the predictions have a bandwagon effect, but I don't know any.

From all our experience with election polls, we can extract no persuasive evidence that poll findings drive voters' decisions.

Much evidence points the other way. In the current campaign, for example, polls taken in July and August showed Mr. Dukakis well ahead, some by as much as 15 to 20 points. A week after the Republican convention, however, virtually every poll showed Bush ahead. How did the convention override the bandwagon effect created by the earlier polls? It's hard to resist the conclusion that there was in fact almost nothing of the kind to overcome.

As members of a society that has practiced democratic government for more than two centuries, Americans seem to have little difficulty remembering that leading in a poll does little to qualify a candidate for the nation's highest office.

Are some people discouraged from voting at all by the fact that polls show one contender well ahead? We simply don't know. If they are, we have no basis for claiming that despondency among backers of the candidate shown trailing is more powerful than complacency among those who favor the nominee who appears well ahead.

Are discussions of present-day elections increasingly poll-driven? Absolutely. Are the election results themselves skewed by poll reports? Probably not.