The Polls: 1948 Looks Better and Better

THE Great Polling Debacle,'' of 1948 has assumed almost mythic status. The last Gallup survey that year, done Oct. 15-25, had shown Democrat Harry Truman trailing Republican Thomas Dewey by 5 percentage points. But Nov. 2, in the ``only poll that counts,'' Truman won the presidency - by 4.5 points. That America's most famous pollster showed in his final survey the president getting five points less of the vote than he was in fact to receive threw the survey business into a tizzy. (Elmo Roper and Archibald Crossley, Gallup's principal competitors, fared just as badly.) Otherwise sensible men and women wondered aloud whether the opinion research would - and should - survive.

Just eight days after this ``disaster,'' the prestigious Social Science Research Council (SSRC) convened a panel to explore what had gone awry. Working furiously, the group finished in just five weeks, issuing its report on Dec. 27. It moved with such dispatch because the SSRC feared that ``extended controversy regarding the pre-election polls ... might have extensive and unjustified repercussions upon all types of opinion and attitude studies and perhaps upon social science research generally.''

I relate this charming tale of lost innocence because, viewed against the polls' performance in recent elections, including last week, the 1948 record seems more triumph than disaster. Both pre-election surveys and those done election day among voters (known as exit polls) have frequently missed the mark by margins well in excess of those that earned polling such ridicule four decades ago. This is more remarkable when it is remembered that in 1948 all interviewing was ended so far in advance of election day that we can't be sure the polls were wrong.

But there is no doubt the polls were spectacularly wrong this year. For example, in the much-studied New York City mayor's race, late pre-election surveys and exit polls alike showed Democrat David Dinkins besting Republican Rudolph Giuliani by a wide margin. A late reading from the New York Times/WCBS-TV exit poll - one drawn, that is, from data gathered far enough into the voting so that the pattern should have been clear - showed Dinkins winning by a landslide: 56 to 42 percent. In fact, he won by just two points.

Similarly, in the hotly contested Virginia governor's race, late polls found Democrat Douglas Wilder comfortably ahead of Republican Marshall Coleman. Their composite picture put Wilder's margin at about 10 percentage points. Based on its exit poll, the ABC affiliate in Washington, D.C., said the Democrat was cruising to victory in a contest ``not even close.'' In fact, it was a dead heat.

Because Virginia and New York had contests pitting black and white candidates, some of the early scrambling for explanation (and exoneration) suggested that race might somehow have produced the error, perhaps by creating a climate that encouraged some respondents to misrepresent their vote. In fact, for a variety of reasons we know that race had little to do with it.

Perhaps the most conclusive datum here is that the polls' difficulty in approximating the actual vote did not suddenly appear in contests between white and black candidates. Instead it has been escalating over the last decade and was dramatically on display in the 1988 presidential election. Many of the 1988 exit polls misread the actual Bush-Dukakis totals in states across the country by margins even surpassing those seen in '89.

After its 1948 ``debacle,'' the survey community engaged itself in anguished self-scrutiny leading, as noted, to a thorough study of ``what went wrong.'' Even more today, we need systematic evaluation by opinion researchers, journalist users, and other participants.

Some will want simply to throw up their hands and say of the polls, ``It just proves you can't trust them.'' Others will be satisfied with offering numerous hypotheses on the sources of escalating error in election polling. My own favorite hypothesis finds the single greatest source of the error in the increasing refusal rates all surveys encounter. More of the public are just saying no to pollsters.

But the fact is we don't know just why it is that polls are having a harder time estimating the vote. Too, if election polling is having trouble, it's likely other forms of opinion research are as well, though in less evident ways. We need to know what has happened.

I don't think the polls should go away. High among their merits, they require the political community - press included - to reach outside its own closed circle and confront vox populi.

It's now the case, however, even more than in 1948, that opinion research must demonstrate that it is in fact able to amplify - not distort - the public's voice.