A good sample can still land unevenly
Even a carefully drawn sample may differ from the population after fieldwork. Availability, willingness to answer and the mode of contact all affect who completes an interview. Some groups can be harder to reach than others, and response can vary between counties. The achieved mix therefore needs checking rather than assuming it matches the plan.
Compare with a known benchmark
Weighting compares the achieved sample with known population benchmarks, such as census distributions of age, gender and county. It adjusts each respondent’s contribution up or down so that the results better reflect those measured characteristics. The benchmarks must describe the population the poll claims to cover, and the report should identify them.
One interview, a different contribution
Arithmetic illustration, not a research finding: a respondent from an under-represented group might receive a weight of 1.4, while someone from an over-represented group might receive a weight of 0.7. These are contributions to a calculation, not extra people interviewed. The achieved sample size does not change. The balance of contributions changes.
Understand the limits
Weighting can address known, measured imbalances. It cannot recover the views of people who never appear in the frame when those missing views differ in ways the benchmarks do not measure. It cannot repair dishonest answers or a badly worded question. Very uneven weights can also reduce the precision of an estimate, which is why their effect belongs in the assessment of uncertainty.
Look for the workings
An honest report says whether results are weighted and identifies the benchmarks used. ResearchCannon’s reporting standards require that transparency. Keep the unweighted base alongside a result: it tells you how many people actually contributed interviews, even when their statistical contributions differ.
Weighting is not cooking the books. It is showing your workings for how a sample becomes a population.