Since 1948, public opinion polls have not been much in the public eye. The Great Miscalculation of 1948 drove them from their favored places in the daily newspapers, and the experts in opinion and attitude research, commercial and academic, retired to the haven of the professional journals and specialized publications, where they could meditate at leisure on what had happened and why.

Despite its recent fall from grace, there is no doubt that public opinion polling in its contemporary form is a far sturdier and solider plant than the straw poll, which was completely finished off by another great error, the Literary Digest’s prediction in 1936 of a Landon “sweep.” The straw poll, based on the wholesale and indiscriminate distribution of ballots, had no way of finding out what exactly had gone wrong when its prophecies fell flat. In contrast, the public opinion poll, even though it covered a few thousand individuals as against millions, was systematic enough to make it possible to figure out just where it had fallen into error. It could retrace the steps of its operations: had the question been asked incorrectly?—it could then compare the effect of asking two different types of question; were interviewers’ biases at fault?—it could compare the results obtained by different interviewers; had the sample been selected incorrectly?—it could calculate what the effect of differently constructed samples would have been.

All this, and more, was done shortly after the election in a 396-page study published by the Social Science Research Council: The Pre-Election Polls of 1948, by Frederick Mosteller, Herbert Hyman, Philip J. McCarthy, Eli S. Marks, and David B. Truman.

The discussions in this volume are impressive in their mathematical clarity and precision. At every point of the polling process, errors were possible, and at every point some checks exist by which it is possible to find out whether such errors actually occurred. How to evaluate the role of the different errors in the total picture is a different and more complicated matter, and indeed it is not attempted. But one over-all conclusion is perfectly clear: nothing really exceptional happened in the 1948 election. Working with the customary samples and procedures, errors of this degree must be expected. As Frederick Mosteller sums it up: “In the present state of polling, where systematic errors of 3 or 4 percentage points are common, it is not to be wondered that polling operations incorrectly predict elections. The wonder is that polls do as well as they do.”

This is, from the point of view of the social history of polling, a rather remarkable conclusion. After all, the reader is bound to exclaim, no one told us this before 1948!

The mathematics leading to Mr. Mosteller’s conclusion are interesting. Because of the peculiarities of the electoral college system, to forecast elections correctly one must correctly predict the vote in each of a large number of states. Many of these states split almost exactly 50-50 between the two parties, so that a small percentage shift can throw any of them into one or the other party’s lap. Now statistical theory tells us how large a sample we need to achieve a certain level of precision in describing some characteristic of a population from a sample of this population. Thus, to get the same precision for the state of Iowa that a sample of 3,000 would give us for the entire United States, we would need a sample of no less than 2,994! And if we wanted to get the same precision for a town of approximately 32,000 people, we would still need a sample of 2,743! It is financially quite out of the question to use samples for every state as large as the one we ordinarily use for the nation. Consequently, the estimates for the various states are bound to be precarious.

There is a further theoretical complication. Any sample drawn from a larger population will not be an exact miniature of the larger population. For, leaving out all considerations of bias that may enter into the interviewing or question-construction or interpretation, which will produce what is called “systematic error,” we must expect a certain “random error,” the variation that we will get from sample to sample, even if we get perfect random samples with no bias. Let us suppose a not uncommon random error of 3 per cent (which means that any sample will be off from the true figure on the average 3 per cent) and consider a state which splits 51-49—a not unlikely occurrence. Under these circumstances, the puller would have six chances out of ten of hitting the winner right he would not do much worse if he flipped a coin. A 48-52 split would give him seven chances out of ten, and only with a 55-45 split would he have nine chances out of ten.

This is on the basis of the statistical theory alone—which assumes a pure random sample, and no bias. When we add the distortions inevitably produced by the variation in the sample actually interviewed from a true random sample, as well as other biases, it is clear that polls can predict elections with no great certainty.

One further item of analysis by Mr. Mosteller is of interest. In order to test how good any system of prediction is, we have to compare it with some other system of prediction. This, it seems, is done in meteorology, where one compares the excellence of prediction with the results one would have got if one had simply assumed that any day’s weather was going to be just what the preceding day’s weather had been. Now suppose, says Mr. Mosteller, we try the same with election forecasts. Suppose we predicted at each presidential election that each state would go the way it went last time. Embarrassingly enough, this rather mechanical system of predicting does about as well as polling for all the elections in which polling has been used, which means all those since 1936. It is true that the Democrats won every one of these elections—in other words, that things have not changed too much from one presidential election to the next since 1936, and we would have to expect this “persistence forecasting” to do quite well under the circumstances. If we go backwards, we find that “persistence forecasting” would have done a horribly botched job in such elections as 1920 and 1932, where there were changes of administration; polling might have done better in these elections.

_____________

But, after all, the predicting of elections is a rather special task, and an exceptionally difficult one, and it would be unfair to judge polling by it alone. To quote Mr. Mosteller again: “It is a rare study that has a serious interest in the change from 49 to 51 per cent.” It is in these fields other than elections, where the change from 49 to 51 per cent is of little interest, that polling has undoubtedly made a quite considerable contribution. The polls have performed a task similar to that of a geographical survey: they outline the social landscape, telling us how people in different areas and countries, of different age groups and classes, differ on a vast variety of questions. Of course they do only a rough topographical job, and we should only depend on them for broad outlines: they tell us there is a mountain here and a lake there, without giving exact dimensions.

It is the poll as social survey that provides the chief interest-indeed, the fascination-of a huge volume, Public Opinion, 1935-1946, published by the Princeton University Press, under the editorial direction of Hadley Cantril, and prepared by Mildred Strunk. It contains 1,191 large double-columned pages and costs $25.00. It reports, with breakdowns, results of nationwide polls conducted by twenty-three polling organizations in 16 countries that concern themselves with regular polling on public issues. Even at that, it does not include all the poll questions asked by these organizations; and it does not attempt to deal with the questions asked by market research agencies or public polling organizations for private clients.

There is in this book a great deal that is astonishing, a great deal that is amusing, a great deal that reminds one, though hardly nostalgically, of those distant days when the depression, Martin Dies, and Father Coughlin (in 1938 nine per cent of the population listened to him regularly) were all live issues. Many of the results are cause for speculation. Why do 39 per cent of Americans report they are very happy, and only 8 per cent of Frenchmen? Is it really possible that 22 per cent of the American people had read Gone With the Wind in 1939, and that at one point, in 1937, five per cent of the American people were engaged in reading it? One is impressed by the cultural level of the Scandinavian peoples: in Sweden, 42 per cent of the people own oil paintings (including 31 per cent of the workers); 21 per cent of the population in Sweden and Denmark own more than a hundred books; 25 per cent of the Swedes read one or more foreign languages (not counting Danish or Norwegian), and 12 per cent two or more; in 16 per cent of Swedish homes someone plays the piano, in 8 per cent the violin, in 6 per cent the organ. One is charmed to discover that more Danes like modern furniture than traditional furniture. And awed to discover the number of Bible readers in the United States: 5 per cent of the population read the Bible every day (that would be some five or six million); a further 35 per cent have read some of it in the past month; 26 per cent have read it through. And one sees the lay of the land in these United States when one discovers that 91 per cent of those in the South believe in immortality as against only 63 per cent of the residents of the Pacific Coast.

_____________

To this reader, at least, it was more interesting to read about these matters, generally of only peripheral interest to pollsters, than about attitudes towards reciprocal trade agreements; naturally it is polls of this latter sort that make up the bulk of the volume. But, inevitably, one must turn from this type of intellectual grazing to consider what this huge volume, and the many-branched enterprise it represents, adds up to. How does it modify our intellectual horizons? What does it make possible? Perhaps the most interesting approach to these problems is to ask, as Hadley Cantril does in his brief introduction, what would have been the significance of similar types of polls conducted during the French or American Revolution, or during the Civil War, or in Germany before Hitler. Mr. Cantril does no more than raise this question; another leading figure in the field, Paul F. Lazarsfeld, goes into the question in some detail in an address to the American pollsters’ professional association, “The Obligation of the 1950 Pollster to the 1984 Historian (Public Opinion Quarterly, Winter, 1950-51).

Professor Lazarsfeld’s article is filled with several examples of the possible role of the polls in the illumination of history. For example: Lord Macaulay makes a distinction, in his essay on Machiavelli, between the dominant values of Renaissance Italy and those of contemporaneous Northern Europe. Italy, a land of commercial towns, had developed a culture in which ingenuity was a dominant value; the still backward North continued to prize courage. Carrying through his distinction, Macaulay suggests that to a Florentine audience seeing Othello, Iago would have been the hero.

Here, suggests Professor Lazarsfeld, Macaulay proposed a hypothesis that falls squarely into the field of opinion research, and one which could have been directly tested by a poll. After rather facetiously considering the preliminary technical operations, Professor Lazarsfeld says: “The crucial question would have been: how many Florentines and Londoners, respectively, approve of Iago, and how many of Othello, and how many say ‘don’t know’? Nothing less, but hardly much more, would have been needed to provide empirical evidence for Macaulay’s brilliant conjecture.”

This is a rather breath-taking possibility. One thinks of all the historical hypotheses that have been debated these many years and whether poll data could have confirmed or refuted them. Certainly the extension of the polling process into the past—even speculatively—raises some interesting questions: who, for example, among the Londoners and Florentines would we have polled? Would we have questioned the theatergoers alone? Would we have compared only the merchants of the two towns? What about the workers and lumpen? And what about the clergy and nobility? Whose opinion mattered on this question? Our present-day election pollers have at least this problem settled for them: they poll those who are eligible to vote.

What Professor Lazarsfeld is after, of course, is to get pollsters to look at history, both past and current, so they can be helpful to the historian of the future when he comes to consider the mood, the attitudes, and the opinions of the men of the 1950’s. Professor Lazarsfeld sees the polling profession in a position where it can afford to let up on its almost single-minded concentration on technical problems and consider how it could be of value to other social scientists. A great deal of what the pollsters have already done is unquestionably of great value in this respect. In the volume Public Opinion, 1936-1945, for example, one can find series of questions, asked of the British at intervals of a month or two, to find out what they thought of Neville Chamberlain: we can thus trace the rise and fall of his popularity more closely than by a study of the British press. Similarly, the series of questions testing American attitudes toward entry into the war, asked every few months in 1939, 1940, and 1941, forms valuable data for a historian of this period.

_____________

But it is in the field of the rise and fall of values—the history of a “Protestant ethic,” or of a “success ideology,” or of the conflict of “courage” and “ingenuity”—that Professor Lazarsfeld is most interested in seeing the polls make their contribution: after all, the polls study attitudes and it is to the history of attitudes that they have the most to offer. Perhaps this highly speculative branch of history, based generally on isolated bits of literary evidence, might be transformed by polls into something as substantial as diplomatic or economic history. For example, Professor Lazarsfeld suggests that the polls may be able to test David Riesman’s hypothesis (in The Lonely Crowd) that there has been a shift in America from dominance by an “inner-directed” type of person—self-willed, certain of his goals, relatively impervious to the views of those around him—to an “other-directed” type, to whom goals are less significant than approval by others. Now as a matter of fact a great deal of existing poll data can be taken as supporting such a thesis. Riesman has written that the greatest fear of the “other-directed” type is of his own temper: both because the explosion of temper may lose him approval, and because the need to rein himself in builds up a pressure to explode, which he must restrain. And one poll does report that 14 per cent of the American people say that “temper” is their worst fault—twice as many as choose the next most popular fault. Can we consider this as offering validation for Riesman’s thesis? Certainly no opponent of Riesman’s would be convinced: he will simply reject the interpretation that “temper” is a good index of “other-direction” or that those questioned meant by “temper” what Riesman means.

This, essentially, is the problem: we know, after a poll, that about three thousand persons in all spheres of life have been approached and had a single rather simple question put to them. But can one take their “yes” or “no” as sufficient proof for the existence of a subtle and complex attitude that it may take a book to expound? Consider another example from Professor Lazarsfeld’s article. He refers to a finding in The American Soldier that those soldiers who do not pray seem to be more mature and get along better with their fellow soldiers than those who do. What might explain this? Professor Lazarsfeld suggests that in an industrial society there is a bifurcation in attitudes to religion: one group clings to it because society is so incomprehensible, uncontrollable, and frightening;-another group, more mature and capable of unblinkingly facing the disenchanted world, rejects it. Now no poll could have even suggested such an extensive theory: nor could any poll or even many of them demonstrate it. The poll may give plausibility to the theories developed by a historian; it adds a new type of evidence; but as long as men are interested in comprehensive interpretations of history, we cannot expect polls of themselves to refute or prove any single theory.

Professor Lazarsfeld is quite aware of this: he does not think the introduction of evidence from polls does away with the historian’s need to interpret. Polls are new evidence, and in the social sciences evidence has always played a rather ambiguous role. Theories arise, are fiercely held, and pass away—and after it is all over, it is often hard to see that the weight of evidence was so overwhelming as to have led to their acceptance in the first place, or their abandonment in the second. Persuasiveness of argument, apt examples from history and experience, inner logic, and perhaps our simple need to have a part of our experience given satisfactory meaning—these have played a far greater role in the history of theories in the social sciences than strict canons of evidence and proof. It seems that we human beings want to believe we know more about our human state than a strict adherence to scientific method would allow. The thought that we will now be able to buttress our ideas with the solid foundations of the opinion poll is an exciting one, and in many individual cases we will be able to do just this. But when it comes to those large constructions, the few glories of the social sciences, that pull together many things in one flash of illumination, polls will give only some additional data, no worse than we had before, and not much better, but in any case very far from justifying the assurance with which each of us holds his theories, or seeks to persuade others to relinquish theirs.

_____________