An opinion poll is a survey designed to estimate the views, attitudes, preferences or intentions of a defined population by asking questions of a sample of people. Political opinion polls commonly measure attitudes towards parties, candidates, governments, policies, political issues or elections. Because interviewing every member of a population is usually impractical, pollsters select a sample and use statistical methods to estimate the distribution of responses in the wider population. A poll is therefore an estimation instrument rather than a direct census of public opinion. Its meaning depends on who was surveyed, how participants were selected, what questions were asked, when interviews occurred and how the resulting data were analysed.
A typical opinion poll begins with a target population and a sampling design. The target population might be all adults, registered voters, likely voters, residents of a particular area or members of another defined group. The pollster then recruits or selects respondents using a probability based or non probability method. Participants answer a questionnaire, their responses are recorded and the resulting dataset is analysed. In many surveys, weights are applied so that the sample better corresponds to known characteristics of the population. The final published estimate may report percentages, averages, levels of approval, issue preferences or other statistical summaries.
Opinion polls can measure many different forms of political opinion. Election intention polls ask respondents which candidate or party they would support under specified circumstances. Approval polls ask whether respondents approve or disapprove of an office holder or institution. Issue polls ask whether respondents support or oppose a particular policy. Importance questions ask which issues respondents consider most important, while favourability questions measure positive or negative views of political actors or institutions. These are not interchangeable measures. A person can favour a party but oppose one of its policies, approve of a government while disliking a particular decision, or regard an issue as important without supporting a particular proposed solution.
Poll results are highly dependent on question wording. Asking whether people support a policy in isolation can produce a different distribution of responses from asking them to choose between competing arguments or options. Researchers have documented wording effects in which seemingly small changes alter measured levels of agreement. Question order can also affect responses because earlier questions can make particular considerations more salient. For this reason, a credible poll should disclose its questionnaire or exact question wording, fieldwork dates and relevant methodological information. These details allow readers to determine what the reported percentages actually measure.
Polls can be conducted through telephone interviews, online questionnaires, postal surveys, face to face interviews or mixed modes. Modern polling increasingly uses online probability samples and non probability opt in panels as well as telephone and address based approaches. Each method has different strengths and potential weaknesses involving coverage, recruitment, cost, response rates and mode effects. AAPOR recommends transparency about how participants were contacted, selected, recruited and screened. Consequently, two polls reporting the same question can differ because their sampling frames, recruitment methods, weighting procedures or interview modes are different.
Polls have uncertainty because a sample is not the entire population and because respondents may differ from nonrespondents. In probability sampling, sampling error can be estimated under the relevant design and is often expressed through a margin of error or confidence interval. However, sampling error is not the only possible problem. Question wording, nonresponse, coverage limitations, inaccurate answers, weighting decisions, mode effects and timing can all affect results. AAPOR specifically cautions that error margins should not be reported for non probability samples in the same manner as for probability samples because the statistical assumptions underlying conventional sampling error do not apply in the same way.
Election polls require additional methodological choices because voting behaviour involves turnout as well as preference. A pollster may distinguish all adults from registered voters and likely voters, and may use models or questions intended to estimate who will actually vote. These choices can materially affect reported results. Timing is also important because campaigns, debates, scandals, economic developments, major events and changes in candidate visibility can alter attitudes between fieldwork periods. A poll therefore describes responses during the period in which it was conducted rather than guaranteeing what people will think or do later. Polls should consequently be interpreted as measurements with specified populations and dates rather than timeless statements about public opinion.
An opinion poll is most useful when its methodology is transparent. Important information includes the organisation conducting the poll, its sponsor, the target population, sample size, sampling method, recruitment method, fieldwork dates, survey mode, question wording, response options, weighting procedure and treatment of undecided or missing responses. Readers should also distinguish national estimates from subgroup estimates because subgroup samples are often much smaller and therefore less precise. AAPOR and major polling organisations emphasise methodological disclosure precisely because percentages cannot be properly interpreted without knowing how they were produced. A poll is therefore not simply a number in a headline but the measurable output of a particular research design.
