Opinion Poll
Overview
An opinion poll (여론조사; 輿論調査) is a survey method that asks structured questions to a sample designed to represent the entire population and statistically estimates a group's opinions, attitudes, and preferences on a specific issue. During election periods, it serves as a key indicator of candidate approval ratings and party support, and it is also widely used for policy evaluation, consumer awareness, and diagnosing social conflict. The reliability of survey results depends on various factors, including sample design, sampling method, question wording, survey timing, and response rate.
Main Content
Historical Background
Modern opinion polling began in the United States in the 1930s. George Gallup gained public trust by correctly predicting Franklin Roosevelt's re-election in the 1936 U.S. presidential election, using quota sampling matched to population composition instead of increasing sample size. After the failure to predict Truman-Dewey in 1948, probability sampling theory was established, and combined with the spread of telephones and computerized statistical processing, it grew into a large-scale survey industry.
Types of Survey Methods
- Interview survey: Face-to-face (F2F) interviews have high response rates and allow complex questions, but cost and time are high.
- Telephone survey: Wired and wireless RDD (random digit dialing) methods were traditionally used, but securing valid samples has become difficult due to the sharp decline in landline telephone ownership and the spread of spam-blocking apps.
- Online survey: Panel-based web surveys have become mainstream, and demographic weights such as sex, age, region, and education are applied to correct bias in non-probability samples.
- ARS (automated response) survey: A method in which respondents press buttons following voice prompts; it is inexpensive, but there is criticism that respondent tendencies may be biased.
Sample Design and Error
Survey results always contain sampling error and non-sampling error. Sampling error is statistical uncertainty arising from random sampling; at a 95% confidence level, with a sample of 1,000 people, it is roughly ±3.1%p. Non-sampling error arises from leading question wording, refusal to respond, nonresponse bias, event effects at the time of the survey, and the like, and can distort results more than sampling error.
Representative Indicators
- Approval rating: Favorability toward or voting intention for a candidate or party
- Evaluation of state affairs: Percentage of positive/negative evaluations
- Party support: Proportional representation voting intention, etc.
- Support/opposition to policy: Whether one agrees with a specific bill or policy
- Average opinion poll approval rating: An indicator that reduces volatility by averaging multiple surveys
Points to Note When Interpreting
When reading survey results, one must check ① the commissioning organization and conducting organization, ② the survey date and method, ③ the sample size and sampling error, ④ the response rate, ⑤ whether weights were applied, and ⑥ the full wording of questions. In particular, results extracted only from the "active voting bloc" or from responses saying they "will definitely vote" can differ considerably from results for all voters.
Regulation and Publication Standards
In Korea, under Article 108 of the Public Official Election Act, to publish or report an election opinion poll, the survey commissioner, survey organization, survey date, sample size, sampling error, response rate, question wording, etc. must be disclosed together. The Central Election Opinion Poll Deliberation Committee (중앙선거여론조사심의위원회) discloses and deliberates registered survey results, and publication is restricted for a certain period before election day (6 days for presidential elections, 6 days for National Assembly elections, etc.).
Recent Trends
In 2024–2025, the opinion polling industry is undergoing the following changes. First, as the share of mobile phone virtual number-based ARS and telephone surveys grows, response rates have fallen to single digits, deepening the bias problem in which only respondents with certain tendencies remain. Second, online panel surveys are growing rapidly, and sophisticated correction techniques that reflect not only demographic weights but also political orientation and past voting history have been introduced. Third, aggregation of multiple surveys has spread, and the tendency to value averages and trend lines more than individual surveys has become pronounced. Fourth, response quality verification, detection of anomalous responses, and nonresponse imputation models using AI and machine learning are being applied in practice. Fifth, as the risk grows that generative AI will circulate large amounts of fake public opinion data or manipulated graphs, the importance of fact-checking and source verification is being highlighted. In addition, due to frequent errors in survey results, "distrust of opinion polls" is expanding, and calls are growing for changes in the media's survey reporting practices and for survey organizations to disclose their transparency.
Related Topics
- [[Sample Survey]]
- [[Sampling Error]]
- [[Public Official Election Act]]
- [[Central Election Opinion Poll Deliberation Committee]]
- [[Approval Rating]]
- [[Statistical Weighting]]
- [[Election]]
- [[Media Literacy]]