Why surveys use weights
A survey usually collects answers from a sample rather than from everyone in the population it aims to describe. Even a carefully selected sample may not match that population exactly. Some groups may be sampled at different rates by design, while others may be more or less likely to respond. Weighting assigns a value to each completed response so that the combined answers better reflect a defined target population.
Imagine a survey intended to describe all adults in a community. If younger adults make up a smaller share of respondents than they do of the community, researchers might give their responses more weight. The weighted result then counts each of those answers somewhat more heavily. This is an adjustment to how responses are summarized; it is not a change to what anyone said.
What a weight can adjust
Weights can account for known differences in selection chances. For example, a survey that deliberately samples more people from a small geographic area may need to reduce that area's influence when reporting a population-wide estimate. Researchers may also adjust for differences between respondents and the target population on characteristics for which they have suitable population information, such as age groups or regions.
These adjustments depend on choices about the population and the characteristics used. A survey of eligible voters, for instance, has a different target population from a survey of all adults. Researchers must also decide how to handle respondents who belong to several categories at once. Weighting methods can balance multiple characteristics, but a match on each separate characteristic does not guarantee a match on every combination of them.
The limits of correction
Weighting works best when people who responded are reasonably similar, on the question being studied, to people who did not respond but share the characteristics used in the adjustment. If willingness to participate is related to an opinion that the weighting information does not capture, bias may remain. Giving more weight to the available respondents cannot reveal what absent people would have answered.
Weights also cannot fully repair a survey that fails to reach part of its target population. If some people had no practical chance of being contacted, researchers have no direct answers from that group to reweight. Nor can weights fix a confusing question, a misleading set of response options, or answers influenced by how an interviewer asks. Those are problems with coverage or measurement, not simply with the proportions of respondents.
Why weighted estimates still have uncertainty
When some responses receive much larger weights than others, the estimate can become more sensitive to the answers of a relatively small number of people. Weighting may improve the balance of a sample while also increasing its variability. That is one reason a weighted survey should not be judged only by its raw number of respondents. The sampling method, the spread of weights and other design choices matter too.
A reported margin of error, when one is provided, addresses only the kinds of uncertainty covered by its calculation. It does not automatically include every possible error from nonresponse, coverage or question wording. Estimates for a subgroup can be especially uncertain if few people in that subgroup responded, even when the full survey has many participants. Small differences between estimates therefore deserve cautious interpretation.
How to read a weighted survey
Start with the survey's stated target population: whose views is it meant to describe? Then look for an explanation of how respondents were recruited, which characteristics were used in weighting, and where the population information for those adjustments came from. If the survey reports both unweighted and weighted sample details, they can help show how much adjustment was needed. A large adjustment is not proof that a result is wrong, but it is worth understanding.
Finally, connect the method to the claim being made. A weighted estimate of an opinion is more persuasive when the survey reached the relevant population, asked a clear question and disclosed enough information to assess its limitations. Weighting is a useful tool for making responses better reflect a population. It is not a guarantee that every source of survey error has disappeared.
