Opinion Polling is the systematic collection and analysis of information about what a defined population thinks, believes, prefers, expects or intends, usually by asking questions of a sample rather than every member of the population. It is used to measure public opinion on political candidates, parties, policies, institutions, social issues, consumer preferences and current events. The central statistical problem is inference: researchers observe a sample and use the resulting responses to estimate characteristics of a larger target population.
Opinion polls therefore produce estimates rather than direct measurements of every person's views. The quality of an estimate depends on the target population, sampling design, recruitment process, response rate, questionnaire design, fieldwork period, weighting, mode of interviewing and analysis. A poll is consequently not simply a collection of percentages. It is the final product of a chain of methodological decisions, each of which can affect the estimate. AAPOR specifically recommends examining sampling methods, question wording, response options, weighting and the dates of data collection when interpreting polling results.
The history of Opinion Polling developed from earlier forms of systematic social measurement, political canvassing and statistical sampling. Modern polling became particularly important during the twentieth century as researchers developed methods for selecting smaller samples capable of representing much larger populations. Early polling relied heavily on face to face interviews and telephone interviewing, while later practice incorporated random digit dialling, postal questionnaires, probability based online panels, opt in internet samples and mixed mode designs.
Changes in communications technology have altered both how people can be contacted and who is willing to respond. Declining response rates have therefore become an important methodological issue because people who participate can differ systematically from people who do not. Contemporary pollsters use weighting, stratification, sample controls and other statistical procedures to compensate for known imbalances, but these adjustments cannot guarantee the elimination of every source of bias. AAPOR notes that nonresponse can bias results even when the initial sample is probability based.
A fundamental distinction in Opinion Polling is between probability and non probability sampling. Probability sampling gives members of the target population known or calculable probabilities of selection and provides a statistical basis for estimating sampling error. Non probability approaches select respondents through mechanisms such as opt in panels or other recruitment procedures without known selection probabilities, so conventional margins of sampling error cannot simply be applied.
AAPOR cautions that reporting a conventional margin of error for a non probability sample can be misleading unless a suitable statistical model and assumptions are provided. This distinction matters because a large sample does not automatically compensate for a systematically unrepresentative recruitment process. Ten thousand respondents selected through a biased mechanism may provide less reliable population inference than a substantially smaller carefully designed sample. Sample size reduces some forms of random sampling variation, but it does not automatically remove nonresponse bias, coverage error, measurement error or model error.
Questionnaire design is a major component of Opinion Polling because the wording and ordering of questions can influence responses. Researchers must decide whether questions should be open ended or closed ended, which response categories should be offered, whether respondents should be permitted to say that they do not know, and whether potentially unfamiliar concepts require explanation. Leading or emotionally charged wording can produce different answers from neutral wording.
Question order can also create context effects because earlier questions can make particular considerations more salient when later questions are answered. AAPOR identifies question wording and question sequence as factors capable of affecting poll results. This means that two professionally conducted polls can produce different estimates even when they study the same population and general subject. Comparability therefore requires attention not only to headline percentages but also to the exact questionnaire, response options, fieldwork dates and sampling procedures.
Opinion Polling can measure several different forms of opinion that should not be treated as interchangeable. Attitudinal questions measure evaluations or beliefs, behavioural questions ask about past actions, intention questions ask what respondents expect or plan to do, knowledge questions assess factual awareness, and demographic questions describe characteristics of respondents. Political polling may ask about voting intention, but a voting intention percentage is not identical to a prediction of an election result.
AAPOR explicitly describes election polls as snapshots in time rather than forecasts and notes that actual turnout is generally lower than self reported voting intention. Opinion polls can also measure intensity, certainty and issue salience, which can be important when respondents give superficially identical answers for very different reasons. Analysts therefore need to distinguish what the question actually measures from the broader interpretation subsequently attached to the result.
Weighting is another essential component of Opinion Polling. A sample may contain too many or too few respondents from particular demographic or political groups relative to the target population. Weighting adjusts the contribution of respondents so that the final dataset more closely matches specified population benchmarks. Common variables can include age, sex, education, geographical location and other characteristics relevant to the research design. Political polls can also consider political affiliation or past vote when constructing an estimate of the electorate.
Pew Research Center's September 2026 methodological work describes weighting as an adjustment designed to bring a sample into closer alignment with the population and discusses the increasing use of political weighting in United States polling. The process can improve estimates but also introduces assumptions, especially when the benchmark itself is uncertain or when weighting variables are correlated with the outcomes being measured.
Interpreting Opinion Polling requires understanding uncertainty. For probability samples, the margin of sampling error describes uncertainty attributable to sampling under specified statistical assumptions. It does not account for every possible source of error. AAPOR stresses that the reported margin of error does not capture problems such as nonresponse bias or incorrect turnout models. Other potential sources include coverage error, inaccurate population benchmarks, questionnaire effects, interviewer effects, mode effects, weighting assumptions and changes in opinion during fieldwork.
A poll showing one option at 51 percent and another at 49 percent therefore should not automatically be interpreted as demonstrating a meaningful difference. Polling averages can sometimes provide a broader picture because individual poll specific random variation may partly cancel out, but averages can still inherit systematic biases from the polls included.
Opinion Polling has significance beyond elections because it provides governments, researchers, journalists, organisations and the public with systematic evidence about attitudes and preferences. Longitudinal polling can reveal changes over time, while repeated cross sectional surveys can track population level trends without interviewing the same individuals repeatedly. Panel studies can instead follow individuals and examine how particular attitudes change. Subgroup analysis can identify differences by age, education, region, socioeconomic position or political affiliation, although small subgroup samples have greater uncertainty.
Polling can inform research into political participation, trust, polarisation and policy preferences, but it does not itself determine what policy should be adopted. The most reliable interpretation treats polling as empirical evidence about measured responses under specified conditions, rather than as a direct representation of an immutable public will. Transparent reporting of methods allows readers to judge what conclusions the evidence can reasonably support.
