The results of these private polls, either in politics or in marketing, are usually confidential and not available to the public. Many national and statewide polls, however, regularly sample public opinion on many different topics and publish the results in newspapers and magazines. In the United States the best known of these are the Harris Poll and the Gallup Poll. A large body of research information on consumer behavior and political attitudes and behavior has also been developed and published by academic organizations such as the Survey Research Center at the University of Michigan and the National Opinion Research Center at the University of Chicago.
OPINION POLLS--HISTORICAL BACKGROUND
Opinion polls were used as early as 1824 by two newspapers, the Harrisburg Pennsylvanian and the Raleigh (N.C.) Star, to test the strength of political candidates. These were "straw polls," in which a haphazardly selected group of citizens were asked their opinions to see which way the "political wind" was blowing. In the 1920s and 1930s the magazine Literary Digest became famous for its huge political polls. It sent as many as 18 million postcards to potential voters in the United States, asking their preference among the presidential candidates. As many as 2 million replies were received, but despite their immense size these samples had two sources of error. The lists of people to whom postcards were sent were biased (that is, they were not representative of all citizens) because they excluded many people of lower socioeconomic status. Also, those who replied were a self-selected sample, and individuals who took the trouble to return the postcard were often more extreme in their opinions than the average citizen.
Despite these sources of error the Literary Digest poll correctly predicted the presidential election winners up through 1932. In 1936, however, it predicted that Franklin D. Roosevelt would lose to Alf Landon, whereas Roosevelt actually won a landslide victory. Partly as a result of this error, the magazine soon went out of business, and the principle was clearly established that a biased sample, no matter how large, cannot be trusted.
OPINION POLLS--SAMPLING METHODS
In the 1936 election a more scientific sampling method was introduced into politics--the method had been used in business since the 1920s--by three different polling pioneers, George Gallup, Elmo Roper, and Archibald Crossley. All three correctly predicted Roosevelt's victory and thus launched scientific public opinion polling toward its subsequent great popularity. They used the quota method of sampling, in which individual members of the sample are chosen in accordance with a quota so as to roughly match the national population on factors such as geographic area of the country, urban versus rural residence, sex, age, race, and socioeconomic status.
The major problem with the quota method of sampling is that the interviewers are allowed discretion in choosing the individual respondents within the quota categories. This discretion introduces a possible source of bias, because the resulting sample can largely omit some types of people, such as those who are difficult to contact.
A much better approach is the probability method of sampling, in which specific respondents are chosen by random selection methods. The result of this method is that no type of individual is systematically omitted from the sample, and the likely amount of error in the resulting data can be calculated.
Statistical laws have established that no matter how large the population being studied (from a small city to a whole country), the size of the sample is the main factor that determines the expected range of error in a probability sample. Most current polls use samples ranging in size from 1,000 to 2,000 individuals. A sample of 1,500 has an expected (that is, a 95%-certain) margin of error of plus or minus 3%, and larger samples yield only slightly smaller errors. Many polling organizations have adopted probability methods in selecting their samples, but the less-reputable polls still use quota methods or even nonscientific haphazard methods of sample selection, and the quality of their findings suffers accordingly.
OPINION POLLS--VALIDITY OF POLL RESULTS
Several other factors, in addition to sampling methods, can cause errors in poll results. First, the pollster must determine whether respondents have any information about the topic on which to base their opinions. Second, the questions must be carefully worded and pretested in pilot studies to ensure their clarity and impartiality. Questions must avoid biases in wording that suggest a socially desirable answer or lead respondents to agree with one side of an issue. Finally, interviewers must be carefully trained to avoid influencing respondents' answers.
In political polling, several factors can cause errors in predicting election results. Would-be respondents who are not at home produce uncertainty in the data, and many respondents are often undecided which way to vote. Last-minute changes in voting intentions sometimes occur between the time of the poll and the election. Differential turnout on election day is also a problem, and pollsters have developed techniques for estimating respondents' likelihood of voting.
The primary goal of opinion polling is to describe the distribution of public opinion at a given point in time, not to predict election results. Each national election, however, provides an excellent chance to validate the polls by comparing their results with the actual election outcome. By this standard the major national polls in the United States and Great Britain have an outstanding record. Since the start of scientific polling in 1936, they have only predicted two national elections incorrectly--the 1948 U.S. presidential election in which Harry S. Truman scored a last-minute victory over Thomas E. Dewey, and the 1970 British election in which the Labour party was unexpectedly defeated. Factors responsible for these wrong predictions included ceasing polling too long before the election to catch last-minute trends, the use of quota samples, and difficulties in estimating differential voter turnout. Improvements in the polls' methods have been made since 1948, and in congressional and presidential elections since then the leading polling organizations have had only a tiny average error in predicting the national vote. Although the polls have been accused of affecting the political process by creating a "bandwagon effect" benefiting the leading candidate, research studies have shown no evidence of poll results influencing voters' political choices.
Stuart Oskamp
Bibliography: Asher, Herb, Polling and the Public (1988); Bogart, Leo, Polls and the Awareness of Public Opinion (1985); Gallup, George H., ed., The Gallup Poll: Public Opinion 1984 (1985); Mann, Thomas E., and Orren, Gary R., eds., Media Polls in American Politics (1992); Oskamp, Stuart, Attitudes and Opinions (1977); Wheeler, M., Lies, Damn Lies, and Statistics (1976); Worcester, Robert M., ed., Political Opinion Polling: An International Review (1983).