How survey weighting works—and what it cannot fix

Start with the population, not the answers

Before weighting begins, a study must define whom it intends to represent: adults, registered voters, likely voters, residents of a particular place, Filipino citizens abroad, or another specific population. A benchmark that is appropriate for one population may be wrong for another.

The benchmark record should identify the authority, table, universe, geography, release date, and version. Election studies may require an electorate benchmark rather than a general-population benchmark.

Respondents receive different analytical influence

If a demographic group appears less often in the sample than in the approved benchmark, respondents in that group may receive more analytical weight. Groups appearing too often may receive less. Modern workflows can calibrate several dimensions while limiting unstable or extreme weights.

Responsible reporting shows the variables used, the weighting method, the range and trimming of weights, the effective sample size, and any design effect. Weighted percentages should never be presented without explaining the underlying sample and method.

Weighting has real limits

Weighting cannot create information from people who had no meaningful chance to participate. It cannot correct a misleading question, an unmeasured source of nonresponse, coordinated manipulation, or an eligibility error.

Demosight therefore combines weighting with recruitment-source monitoring, eligibility rules, response-quality review, benchmark versioning, sensitivity checks, and a permanent methodology passport for every releasable study.

• Predeclare the target population and benchmark source.
• Monitor recruitment and quota cells during fieldwork.
• Inspect weight concentration and effective sample size.
• Publish both the adjustment and its limitations.