Quantitative Methods in Politics[1]

Stuart A. Rice
Chairman, United States Central Statistical Board

THE TITLE of this paper was the title of a book published in 1928, and I am speaking as its author. In that book I sought to replace a number of general descriptive statements concerning American politics and political occurrences with precise numerical statements. Simple statistical methods were applied to the analysis of commonplace data, especially to data that related to political attitudes and public opinion.

The accurate measurement of public opinion is a central problem for a science of politics. In authoritarian states such measurement would be difficult, whether officially or unofficially sponsored. In democratic states, on the other hand, the idea and the habit of measuring opinion are firmly established. Representative government assumes that the will (or opinion) of the people should be ascertained and given effect through elections, General and primary elections, referenda and straw votes are a part of the same culture complex, and all are statistical measurings of public opinion. I believe that Mr. Gallup's important statistical data have interested newspaper readers because the American public is habituated to elections and polls.

The economic importance of opinions and attitudes is generally recognized. To illustrate, the National City Bank asserts in its December, 1937, review of economic conditions:

The business of the country is an aggregate of the actions and policies of millions of individuals and corporations, all following the best judgments they can bring to bear on their own problems; and it might be supposed that their difference of opinion would cancel out. But that is not always the case. They are all subject to the same economic and psychological influences. By and large they tend to go forward together, planning and spending confidently; and conversely, they curtail and economize together. The changes from one policy to the other mean a vast difference in the volume of trade and the level of prices.

While such statements as this stress the importance of opinion as an economic factor, there has been comparatively little effort to evaluate this factor quantitatively, except, perhaps, in the case of market analyses. Yet I believe that many aspects of the opinion factor in business are measurable. In the citation from the National City Bank just quoted there is suggested a counterpart in business to what I have termed the "landslide tendency" in politics. Landslides of opinion, both as to politics and as to prudent economic behavior, probably

( 127) result from the rapid growth of mass communication through newspapers, motion pictures, and the radio. As agencies of mass communication extend their influence, the landslides become greater. Here is an hypothesis of seeming importance, susceptible, I think, of quantitative testing. Unfortunately for the purpose, the votes of stockholders at meetings of corporations are less adequate indexes of opinion than their analogues, the popular votes in political elections.

It need not be argued further that opinions and attitudes have important consequences, justifying attempts to express them in numerical terms. We turn then to methods of measurement. The traditional or rule-of-thumb method is an election.

Elections presuppose the complete summation of opinion upon cur-rent political issues within a universe (the electorate); just as straw votes assume representative sampling. Actually, in democratic countries, elections never obtain complete "coverage" of the electorate. The non-voter is a perennial problem of American politics. Nor do elections provide opportunity for the full expression of opinion by the voter upon even a single subject. There is seldom a chance for him to say more than "yes" or "no" upon what may be a complex question. Most frequently his opinion is not expressed upon a question at all, but upon a list of candidates whose individual opinions he is unlikely to know. If he did know them, he would probably find that the opinions of any candidate on some questions coincided with, and on other questions differed from his own.

As a statistical method, an election may be described as a voluntary self-recording of opinion by some members of the electorate upon a greatly simplified statistical schedule. Totals depend upon voluntary appearances by voters at polling places on a single date. There are no follow-ups, no recanvasses, a minimum of instructions and explanations, and no checks upon the understanding by the respondent of the questions asked. The tabulating process is a simple hand-count, with-out cross tabulation with other known data in ways customary in statistical machine tabulation.

The statistical method of election may be contrasted with the method of census. Censuses have usually been employed for the collection of objective data, but their use to collect subjective information is by no means rare, and they offer an alternative to the method of election as a means of summating opinions and attitudes. By the census method an approximately complete coverage of the universe is obtainable. Each individual, without initiative on his part, is approached by an enumerator and invited, or required under penalities for non-compliance, to answer the questions asked. These may be

( 128) elaborated where the issues in an election must be oversimplified. Trained enumerators may elicit more accurate formulations of opinion from some individuals than these would be able to construct for them-selves independently.

An election will not provide an accurate summation of public opinion unless, by chance, the opinions of those who vote and the issues upon which they vote are representative samples of the opinions of all persons having opinions upon all pending issues. For example, it has been contended that the President holds the support of the electorate but that the latter is out of sympathy with certain New Deal policies for which he stands. This contention, if true, illustrates the unadaptability of the election method to the measurement of public opinion. It does not discredit the representative form of government or the use of elections to select public officials. Moreover, the method of election is in accord with our national traditions of individualism. The method of census is less "democratic"for it does not have the voluntary aspect of an election.

It is currently proposed that a national referendum election should precede our entrance into foreign war. According to this plan, if a majority of voters acquiesced all of the people might then be called upon to risk their lives and property, but not otherwise. It is assumed that the voters have competence, without special knowledge or educational preparation, to pass upon this question; also that the lives and property of non-voters and of an opposing minority may justly be put in jeopardy by the votes of the majority. Waiving these debatable assumptions, let us assume that in a matter of such consequence an accurate summation of opinion is essential.

For this purpose, the method of census is superior to the method of election. The census could reach some millions of people who would be directly affected by war but who would be ineligible to vote in an election. There could be specially trained enumerators, with a carefully prepared schedule or ballot, covering alternative statements of pro-posed public policy. These statements could be arranged upon a scale from the most belligerent to the most pacifistic. Items of information concerning the respondent himself could be obtained, such as sex, age, marital status, and whether or not a legal voter. Opinions could then be correlated with these items. For example, it could be ascertained whether the young men of 18 to 31 were more or less in favor of war than their elders, their wives and sweethearts, or other groups within the electorate.

All answers would be compulsory but confidential. Enumerators would be instructed to avoid suggestions that would influence the

( 129) replies, though they might make factual statements, officially authorized, which would serve as premises for the questions asked. Thus, the enumerators might state that a hostile air fleet was on its way toward our shores across the Atlantic. The official statement on following days could be revised in accordance with any change in the factual premises. The possibility of daily changes in the official statement introduces an uncontrolled and unfortunate variable into the census procedure. However, the absence of such an official statement would introduce an even greater possibility of bias, because of the variety of propaganda and other influences to which the respondents would be subjected.

A census of opinion on a proposed declaration of war would require considerable time, during which the prospective enemy might obtain military advantages; but this factor of time is also involved in the election method, to which alone I am offering the census method as an alternative.

The recent National Unemployment Census, despite its name, is a good illustration of the election method of measuring opinion, and of the contrast between that method and the method of census. It was believed by many students, including the present speaker, that only a true census—a door-to-door canvass by trained enumerators—would provide a relatively accurate account of unemployment in the United States. In part this belief was based upon the seeming impossibility of obtaining either a complete coverage of the unemployed, or a representative sample within the universe, by any other method. The belief was also based upon the need for information regarding the employment status of the entire gainfully occupied population, in order to interpret data pertaining to the unemployed part of the population.

The method of census, however, was rejected in favor of the method of self-registration, or election. In effect, the National Unemployment Census was a poll, in which all who wished to do so were given an opportunity to vote upon the question whether they were unemployed or partly unemployed. Their votes were recorded in answer to Questions 2(a) and 2(b) which were key questions on the schedule. Other, more objective inquiries, of which 2(c)-relating to emergency workers -is an illustration, were also on the schedule. To "get out the vote," many of the devices familiar in a political campaign were utilized. It is interesting to note, however, that the unemployment election approached as nearly as possible to the method of census, without actually employing that method. Ballots (Unemployment Report Cards) were distributed to every home in the country. The voting did not necessitate, as some had proposed, that the voters (the unemployed)

( 130) should apply at an election booth or similar place of registration.

Perhaps the chief scientific value of the National Unemployment Census will be the experience provided with the election method of obtaining data hitherto regarded as obtainable only through the method of census.

Straw votes may be obtained by either the method of election or the method of census, or by a combination of the two. The Literary Digest polls appear to have been of the election type, since, like the Unemployment Census, they rested heavily upon voluntary returns from mailed questionnaires. The Digest poll of 1936 appears to have illustrated the possibility of bias contained in the election method. Mr. Gallup, on the other hand, employs the method of a representative sample census. To the extent that he is able to master the problems of sampling which he encounters, his statistical findings may summarize American attitudes and opinions more accurately than do elections or any other method yet devised.


  1. A paper presented at the Ninety-ninth Annual Meeting of the American Statistical Association, Atlantic City, New Jersey, December 28, 1937.

Valid HTML 4.01 Strict Valid CSS2