Public opinion is measured primarily through systematic surveys in which researchers ask a sample of people questions designed to estimate the views of a wider population. The process begins by defining the target population, such as all adults in a country, residents of a particular region or eligible voters.
Researchers then select respondents using probability or non probability methods, collect answers through modes such as telephone, online, postal or face to face interviewing, process the responses and apply statistical adjustments where appropriate. The purpose is to infer population level attitudes from observed responses in a sample. The quality of the inference depends on sampling, questionnaire design, response patterns, weighting and other methodological factors.
Probability sampling is a traditional foundation of representative survey research because it gives members of the target population known or estimable probabilities of selection. Examples include random digit dialling, address based sampling and selection from suitable population frames. Random selection reduces the risk that researchers choose respondents according to convenience or self selection. It does not eliminate all error because people may fail to respond, sampling frames may have coverage limitations and respondents may give inaccurate answers. Nevertheless, probability sampling provides a statistical basis for estimating sampling error and generalising from the sample to the target population.
Non probability sampling is also used, particularly in online research, where participants may be recruited through opt in panels or other volunteer based systems. Such samples can be less expensive and faster to field, but the probability that any particular population member enters the sample is not necessarily known. A large sample size does not automatically solve representativeness problems because thousands of respondents recruited through a systematically unrepresentative process can still differ from the population. AAPOR therefore distinguishes probability and non probability designs and emphasises transparency about recruitment, sampling frames, eligibility and coverage. Results from non probability samples require particular care in interpretation.
Weighting is a statistical procedure used to adjust survey responses so that the final estimates better reflect known characteristics or selection probabilities of the target population. A respondent's weight can be thought of as representing a certain number of people in the population. Weighting can correct for unequal probabilities of selection and differences between the achieved sample and population benchmarks. The Office for National Statistics uses weighting when drawing inferences from survey samples to the wider eligible population, while major research organisations use multistep weighting procedures that can account for selection probabilities and population characteristics. Weighting improves representativeness but cannot magically correct every form of bias.
Questionnaire design is another major component of opinion measurement. Researchers must decide exactly what is being asked, how concepts are defined, which response options are offered and whether respondents can indicate uncertainty or lack of knowledge. Small changes in wording can alter results because respondents may interpret apparently similar questions differently. The order of questions can also affect answers if earlier questions make particular considerations more salient. AAPOR specifically recommends examining exact question wording and question order when evaluating poll results. Consequently, two surveys about the same political issue can produce different findings even when conducted around the same time if their instruments frame the issue differently.
Survey mode can also affect measurement. Telephone interviews, online questionnaires, postal surveys and face to face interviews place respondents in different social and technological environments. People may answer sensitive questions differently when speaking to an interviewer than when completing a private online questionnaire. Coverage can also differ because some people are easier to reach through one mode than another. Modern surveys may therefore use mixed mode designs or carefully designed sampling systems intended to provide broader population coverage. Mode effects are a recognised source of potential variation, meaning that methodological comparisons should consider not only the wording of questions but also how respondents were contacted and interviewed.
Sampling error is another component of survey measurement. If a properly selected sample is used to estimate a population proportion, different samples would produce somewhat different results simply because different individuals happened to be selected. Researchers can estimate this uncertainty under appropriate sampling designs and often report margins of sampling error or confidence intervals. Sampling error is only one source of uncertainty, however. Nonresponse, coverage problems, inaccurate reporting, question wording, interviewer effects, weighting and other practical difficulties can produce additional error or bias. A poll with a small numerical margin of error can therefore still be misleading if its underlying design has serious non sampling problems.
Public opinion can also be measured through repeated surveys, panel studies, focus groups, qualitative interviews, experiments, administrative data and behavioural indicators, although these methods answer different questions. A repeated cross sectional survey measures different samples of a population at multiple points in time, while a panel follows the same individuals and can examine individual level change. Focus groups can reveal reasoning and language that fixed response options may miss, but they are not normally designed to produce population estimates. Experiments can test whether information or framing changes responses. The method must therefore match the research question, and researchers should avoid treating every measure of political attitudes as equivalent to a representative national poll.
