Opinion polling is a method of analysis for drawing inferences about the attitudes or behaviors of a population by studying a random sample of persons from that population. Nonrandom surveys are sometimes used by social scientists for theory building, but only random samples can produce valid estimates of population traits. Since social scientists are typically interested in population traits, and since few other methods are as useful for studying population traits at the level of individuals, opinion polling with random samples has become one of the most common methods of data collection in the social sciences. Space does not allow a full treatment of this topic internationally and across the social sciences, but detailing the early adoption and continued use of opinion polls for political science research in the United States provides a useful introduction to the reasons why opinion polling remains such an important tool for the social sciences.
Political scientists have long studied elections and public opinion, but before the 1940s they tended to emphasize how elections and public opinion should work in theory more than how they actually did work in practice. In part this emphasis reflected the scholarly paradigm of this period, which saw the core questions of democratic politics residing in the institutions, processes, and outcomes of political systems more than in the ways ordinary citizens made sense of them. But this emphasis was also encouraged by limitations in early methods of social analysis, which often forced scholars to speculate about factors influencing election outcomes because they lacked systematic evidence for studying many questions of interest. During this period, the best forms of voting data available to political scientists were aggregate election results that could be compared across voting districts or geographic areas, but such data had little to say about the reasons why individuals decided to participate in elections or why they supported particular candidates.
The development of random sampling surveys in the 1930s had a lasting impact on the ways that political scientists studied elections and public opinion. Nonrandom straw polls that included anywhere from dozens to millions of citizens had been used to predict elections and describe public opinion since the 1820s, but these proved to be invalid methods of social inquiry. Their results could not be generalized to larger populations because the people who answered such polls were not a random cross-section of society. The demise of the straw poll came in the presidential election of 1936, when the Literary Digest magazine incorrectly predicted a Republican win over Democratic incumbent Franklin D. Roosevelt (1882–1945). George Gallup’s (1901–1984) new random sample survey correctly predicted a Roosevelt win, and he demonstrated that the Digest Poll’s biased sample was the source of its inaccuracy.
The success of random sampling at predicting elections quickly encouraged its adoption by political scientists, and the 1944 publication of The People’s Choice —the first systematic analysis of American voting behavior to use modern opinion polls—marked a major turning point in political science research. The transformation of this research tradition is widely considered to have been completed with the 1960 publication of The American Voter, which became a cornerstone of political behavior research in America. By this time the opinion poll—particularly the American National Election Study—had become the standard tool for political science research on mass opinion and voting behavior, as it remains a half century later.
The rapid adoption of polling came during a transitional period in political science known as the behavioral revolution, and stems in part from two parallel developments that were occurring at the same time. The first is the rise of methodological individualism as a critique of the dominant style of inquiry common to social science scholarship before the 1940s, which tended to explain any particular social phenomenon as a product of other social phenomena. For example, an election outcome might be explained as resulting from the state of the national economy. Methodological individualism held to the contrary that any social phenomenon was a collective product of individual-level behavior. For this school of thought, explaining election outcomes as a product of the economy requires understanding how economic factors influence the choices made by individual voters. Because opinion polls could measure not only political behavior but also the underlying attitudes that precede and shape such behavior, polls fit squarely with the tenets of methodological individualism and with the new paradigm of scholarship championed by behavioral researchers. The second development was the rapid growth of the science of statistics during the first half of the twentieth century. Polls not only provided information about the attitudes and behaviors of individuals, but when sampled using the new random probability methods they produced findings that could be generalized to entire populations. It was this unique combination of developments, arriving as the behavioral revolution was getting underway, that quickly transformed the opinion survey into the standard method for opinion research in the United States.
The ability to generalize survey findings from a sample of 1,500 respondents to a population of millions comes from strict adherence to the statistical principles of random probability sampling. Researchers often want to study the attitudes or behaviors of a population, such as adults in the United States. Although they might prefer to include every member of the population in a survey, conducting a census for any but the smallest of populations is prohibitively expensive and fraught with difficulties. The solution offered by random probability sampling is a compromise. If a small subset of the population is selected at random to take the survey, then the central tendencies of that sample—such as the percentages of people holding various opinions—will tend to be quite similar to those of the population from which they were drawn. The compromise comes in how closely the sample estimates are likely to match the true characteristics of the population. Random sampling allows for a small amount of error between the mix of answers given by the sample and those that would be revealed by a census of the population, and also allows for the possibility that once in a while the sample may have quite different characteristics from the population.
Because random sampling allows researchers to accurately determine the probability that both kinds of error will occur in a given sample, they can assign a level of confidence to the likelihood that the sample results approximate the population’s actual characteristics. The probability estimate for the first type of error is called the margin of error, defined as the range of values around a sample estimate in which the true population value is likely to be found. The size of the margin of error is determined by the number of persons included in the sample, with larger samples allowing more precise estimates. The margin of error for a sample of one thousand persons is plus or minus three percentage points, meaning that the proportion of respondents in the sample holding a particular opinion should be within three percentage points either way of the true population value. The probability estimate for the second type of error is called the confidence level, which is the likelihood that a population’s true value falls within the range given by the sample’s margin of error. Typically, random samples are drawn with a 95 percent confidence level, meaning that the population’s true value should fall within the sample’s margin of error ninety-five times out of one hundred.
These error estimates presume that each individual in a population has an equal and random chance of being selected, but other types of error can also influence poll results. Any violation of equal and random selection may produce sampling errors that can bias survey estimates away from the population’s true value. Nonresponse errors can occur when persons selected to be in the sample are never contacted, decline to be interviewed, or refuse to answer particular questions. Measurement errors can be introduced by the wording of questions and the order in which they are asked. For example, answers obtained by asking respondents to select from a list of pre-determined responses—using what are called “forced-choice” questions—can produce different estimates of public opinion than answers obtained by recording the verbatim responses provided by respondents using “open-ended” questions. Errors of conceptual validity can occur when a question fails to adequately measure the concept of interest to survey researchers. These other types of error are not taken into account by the sample’s margin of error or confidence level.
Opinion polls are used by political scientists for four main purposes. First, polls are used to measure and predict political behavior. Pre- and postelection surveys help political scientists understand why citizens support particular candidates or parties. Election-day exit polls provide insights into the demographic and social characteristics of voters. Second, polls are used to chart trends in behavior and attitudes over time. Tracking polls, which consist of small samples taken every few days during the course of an election campaign, clarify how citizens respond in the short term to campaign activities. Trend analyses pose the same questions to different samples every few months or years to study long-term changes in the attitudes and behaviors of a population. Third, polls are widely used for correlational analysis, which examines how attitudes and behaviors are related to one another at the individual level. Fourth, polls are used to conduct general population experiments, where samples are divided into treatment and control groups to produce experimental findings. Unlike traditional laboratory experiments, the findings from survey experiments can be generalized to populations, which makes this combination of methods increasingly appealing to social scientists.
Polling has helped political scientists understand public opinion processes, but it also has changed the ways that political scientists think about public opinion. Up until the middle of the twentieth century, a sociological paradigm emphasizing the activity of organized groups informed the mainstream of opinion research in the social sciences. Before the behavioral revolution, the phenomenon of public opinion tended to be associated with action or barriers to action rather than merely with a potential to act. In contrast, newer psychological interpretations of public opinion associated with the behavioral revolution and informed by survey research have tended to view attitudes rather than actions as the primary phenomenon of interest. Likewise, while the earlier sociological conceptions of public opinion were concerned with action conducted by interested groups rather than the population as a whole, the method of random sampling has cultivated a perspective that views public opinion as an attribute of unorganized masses or entire societies. As a byproduct of this paradigm shift, the rise of polling has encouraged political scientists to neglect sociological and philosophical dimensions of public opinion research that once had been vibrant areas of social inquiry.
SEE ALSO Exit Poll; Hypothesis and Hypothesis Testing; Methods, Quantitative; Methods, Research (in Sociology); Methods, Survey; Polling; Public Opinion; Quantification; Survey; Surveys, Sample
American Association for Public Opinion Research.http://www.aapor.org.
American National Election Studies. http://www.umich.edu/~nes.
Asher, Herbert. 2004. Polling and the Public: What Every Citizen Should Know. 6th ed. Washington, DC: CQ Press.
Campbell, Angus, Philip Converse, Warren Miller, and Donald Stokes. 1960. The American Voter. New York: Wiley.
Converse, Jean M. 1987. Survey Research in the United States: Roots and Emergence, 1890-1960. Berkeley: University of California Press.
General Social Surveys. http://www.norc.org/projects/gensoc.asp.
Gunn, J. A. W. 1995. “Public Opinion” in Modern Political Science. In Political Science in History: Research Programs and Political Traditions, eds. James Farr, John S. Dryzek, and Stephen T. Leonard, 99-122. New York: Cambridge University Press.
Lazarsfeld, Paul F., Bernard Berelson, and Hazel Gaudet. 1944. The People’s Choice: How the Voter Makes Up His Mind in a Presidential Campaign. New York: Columbia University Press.
Natchez, Peter B. 1985. Images of Voting/Visions of Democracy. New York: Basic Books.
Price, Vincent. 1992. Public Opinion. Newbury Park, CA: Sage.
Scott L. Althaus