As elections draw closer, we see more reports on the results of opinion polls. In this article, I would like to introduce the significance of opinion polls in democratic societies and the methods by which they are conducted.
Public Opinion and Opinion Polls
Without doubt, opinion polls are a very useful tool for ascertaining people’s current opinions. While there is some truth in the contention that an over-reliance on opinion polls undermines democracy, we must also ask ourselves what would happen if opinion polls could not be conducted at all. Politicians, the mass media, and lobby groups would simply make up their own versions of “public opinion.” In an even more frightening scenario, only certain institutions (like the government) would be able to conduct opinion polls. For a democracy to function in a healthy way, it is important that everyone be able to conduct opinion polls (based on an appropriate procedure and sense of ethics) and that the results of these polls be published.
The results of an opinion poll are not the same as “public opinion,” however. Strictly speaking, there is no set definition of “public opinion,” even among researchers. They differ, for example, in their views on whether public opinion means the opinion of the majority, or opinions that stands out in the media, or socially influential opinions.
Whatever your view, however, I think we can all agree that public opinion is something intrinsically tied to public issues. Being aware of people’s opinions and situations is vital to thinking about the public good. And opinion polls are an important tool for this.
The First Opinion Polls
Newspapers began publishing rudimentary opinion polls for predicting elections in the middle of the 19th century. At first, these were nothing more than straw polls conducted of participants at political meetings and journalists, but they gradually evolved into large-scale polls.
The Literary Digest was a weekly magazine that increased its circulation in the 1920s by correctly predicting the outcome of presidential elections. In its poll for the 1936 U.S. presidential election, the magazine surveyed 10 million voters. Based on the over 2 million responses received, the magazine predicted the Republican candidate, Alfred Landon, would beat the Democratic candidate, Franklin Delano Roosevelt. Meanwhile, Gallup, a new research firm, predicted that Roosevelt would win after polling about 50,000 people.
In the end, the victory went to Roosevelt and Gallup. The Literary Digest’s failure is attributed to its selection of voters from lists of readers, car owners, and telephone owners. As a result, the polled group was skewed toward a comparatively wealthy demographic (which tended to vote Republican), leading to an incorrect prediction. On the other hand, Gallup relied on a more scientific theory of sampling to select a polling group that was as representative of the entire voting population as possible. Not only the size of the sample but also the method for selecting it is important when using opinion polls to correctly guess people’s opinions.
Methods for Conducting Opinion Polls
It would be easy to find out the opinions of Japanese voters if we could just “poll all Japanese voters at once.” This is not practical, however, so usually a sample of the population is extracted and polled (this is called sampling). Individuals are selected so that the polled sample is as representative of the statistical population (in this case, Japanese voters) as possible. The most important sampling method is called random sampling.
Random sampling leaves the selection of individuals from the population to chance. This means that all members of the population are selected with the same probability. Suppose, for example, there was a city with 200,000 voters, and 2,000 of the voters were to be selected and polled. To put it very simply, a list of voters’ names could be assigned consecutive numbers, 2,000 numbers could be generated at random, and the corresponding voters polled. (This is called simple random sampling.) Telephone surveys often use a method (Random Digit Dialing, or RDD) in which phone numbers are generated at random and then called.
When you have a large population, however, assigning numbers to everyone is difficult. Thus, in actual opinion polls and social surveys, districts are first selected at random in proportion to their respective populations (the more heavily populated an area, the higher the probability of selection), and then the same number of individuals is selected from each district (this is called multistage sampling). In such polls, districts are grouped by factors like region and population before the polled locations are chosen to avoid selection of only specific regions and large cities. This is referred to as “stratification.” Devices like these can enhance the accuracy of polls.
A margin of error is inevitable, however, even when random sampling is used. One should not read too much into a small difference in percentage in a standard poll of 2,000 people. With random sampling, though, the margin of error can be estimated based on sample size. To reduce the margin of error by half, the sample must be quadrupled in size.
Difficulties in Conducting Opinion Polls
No matter how much attention is given to the selection of polled individuals, the reliability of a poll drops when the response rate is low. In that sense, opinion polls face a challenging environment today. For example, the response rate for the “Public Opinion Poll on Social Awareness” conducted by the Cabinet Office has fallen from over 80 percent in the 1970s to about 60 percent in recent years. The shift is attributed to changes in lifestyle, an increase in self-locking housing complexes and similar technology, and concern over protection of personal information.
As someone involved with opinion polls, I am particularly troubled by the distrust felt by those selected for such polls. When people are polled, they sometimes express their uneasiness with comments like “Why was I chosen?” Such concerns are natural in our dangerous world. At the same time, polls must be as representative as possible to have any meaning. I hope to figure out a way to restore people’s trust in opinion polls.