A poll sample is the group of people whose responses are actually collected in a survey and used to estimate the opinions or characteristics of a larger target population. If a poll seeks to measure the political views of all eligible voters in a country, the poll sample is the subset of eligible voters who are selected or recruited to answer the questionnaire. The sample is therefore distinct from the population.
The population is the complete group about which the researcher wants to make an inference, while the sample is the observed group from which information is obtained. Sampling exists because surveying every member of a large population is often impractical, expensive or unnecessary. The statistical validity of the resulting estimate depends heavily on how the sample is constructed and how closely it permits the observed respondents to inform the target population.
A poll sample has several important dimensions, including its size, source, eligibility rules, selection procedure and composition. Sample size refers to the number of respondents whose answers are included in the analysis, although the exact effective sample size can be lower when weighting or complex designs reduce statistical efficiency. The sampling frame is the source or mechanism from which potential participants can be selected.
In probability sampling, each eligible unit has a known non zero probability of selection from a defined frame. In non probability sampling, respondents may be recruited through opt in panels, advertisements, social networks or other mechanisms without known selection probabilities. These distinctions affect what statistical inferences can legitimately be made from the sample.
A sample is not automatically representative merely because it contains a large number of respondents. A sample of one hundred thousand people selected through a strongly self selecting process can systematically differ from the population, while a much smaller carefully designed probability sample can produce a useful population estimate. Sampling quality therefore depends on selection mechanisms as well as numerical size. Coverage is another issue.
If some members of the target population have little or no chance of entering the sample, the survey can suffer from coverage error. For example, a sampling system based on a particular contact method may fail to reach groups who are difficult to contact through that method. AAPOR identifies coverage, recruitment, nonresponse and self selection as important considerations when assessing the quality of survey samples.
Probability samples and non probability samples represent two major methodological approaches to constructing poll samples. In probability sampling, selection is based on a random mechanism and the probability of selection is known or can be calculated. Traditional approaches have included random digit dialling and samples drawn from defined lists or address based frames.
Non probability samples do not provide each member of the target population with a known probability of selection and may use opt in online panels or other recruitment methods. AAPOR notes that non probability samples do not have a simple classical margin of sampling error because the assumptions underlying probability based sampling are not present in the same way. Such surveys can still be analysed, but their uncertainty must be considered using appropriate models and disclosed assumptions rather than automatically treating them like simple random samples.
Poll samples can also be deliberately structured to provide sufficient observations from particular groups. Oversampling means selecting more members of a subgroup than their proportion in the population would ordinarily produce so that the subgroup can be analysed with greater precision. The resulting data can then be weighted so that the final estimates reflect the subgroup's actual population proportion. This illustrates an important distinction between the composition of the raw sample and the composition of the weighted analytical sample.
A raw sample does not necessarily need to mirror the population exactly in every characteristic if the research design includes appropriate weighting and if the assumptions underlying that weighting are reasonable. However, weighting cannot correct every form of bias, particularly when important differences exist between respondents and nonrespondents that are not captured by the variables used in adjustment.
Nonresponse is one of the central problems affecting poll samples. Some selected people cannot be contacted, some are ineligible, some refuse to participate and some begin a survey without completing it. If people who participate differ systematically from people who do not participate on variables relevant to the political questions being measured, the resulting sample can be biased even when the original selection process was probability based.
AAPOR explicitly notes that nonresponse can bias poll results when respondents and nonrespondents differ systematically, including when their political preferences differ. This means that a poll sample should not be evaluated solely by asking whether participants were randomly selected initially. The behaviour of the selected population during recruitment and interviewing also matters.
Sample size affects statistical precision but does not determine representativeness by itself. Under a conventional probability sampling framework, larger samples generally reduce sampling variability, which is why increasing the number of respondents can narrow the margin of sampling error. However, the improvement follows diminishing returns, and methodological defects can matter more than additional respondents.
AAPOR states that larger samples generally produce smaller sampling error but also emphasises that a larger sample is not necessarily better when other methodological factors are weak. Complex weighting, clustering, nonresponse, poor coverage and badly worded questions can all reduce the practical quality of an estimate. The correct interpretation is therefore that sample size is one component of survey quality rather than a complete measure of it.
A poll sample ultimately functions as the evidential bridge between observed responses and claims about a wider population. Every percentage reported from a survey is based on the people included in the analytical sample, potentially after weighting and other adjustments. Understanding that relationship is essential because a poll result is not simply a count of opinions among respondents. It is an estimate intended to describe a specified population under specified assumptions.
Readers should therefore ask who was sampled, who was excluded, how respondents were recruited, how many participated, whether the sample was weighted, and whether the methodology supports the population claim being made. A sample is scientifically useful when its design and analysis provide a defensible basis for generalisation, not merely because it contains many respondents.
