What is Weighting?
Weighting adjusts survey results so they better reflect a target population when some groups are over- or under-represented in your responses. It gives each respondent a weight that increases or decreases their influence on reported totals and averages.
Weighting is a statistical technique used after data collection to correct imbalances between a survey sample and the population you care about. If certain groups (for example, parents who speak a particular language, age bands, or neighbourhoods) answered at different rates, weights scale responses so the survey estimates match known population totals or proportions. Common approaches include simple post‑stratification (matching on a few categories), raking (iteratively matching several margins), and probability weights (based on sampling design or known selection probabilities). Weighting does not create new information — it redistributes influence among existing answers to produce estimates that are more representative.
Usage example
A school surveys 600 families but finds only 40% of respondents are from households where Spanish is spoken, while school records show 60% of families speak Spanish. The administrator applies weights so responses from Spanish-speaking households count more, making school-level estimates (like satisfaction rates) reflect the actual parent population.
Practical application
Why it matters: many community surveys underrepresent specific language groups or demographic segments. Weighting helps correct those imbalances so decisions are based on a more accurate picture of the whole community. Practical points: you need reliable population benchmarks (e.g., school registration data, census counts or administrative records) for the characteristics you weight on; weighting can increase the variance of estimates (and so widen confidence intervals); and it cannot fix bias caused by differences on unmeasured characteristics. In practice, keep weighting simple (a few key variables), document your method and benchmarks, and report both weighted and unweighted results so stakeholders understand the effect of the adjustment.
FAQ
What information do I need to create weights?
You need two things: (1) variables in your survey that identify groups (for example language spoken at home, age group, or area), and (2) trustworthy population benchmarks showing the true distribution of those groups (from school records, local authority data or the census). Without reliable benchmarks you can’t produce defensible weights.
Can weighting fix all survey bias from low response in some communities?
No. Weighting helps when the difference between respondents and the population is captured by the variables you weight on. It won’t correct bias from unmeasured differences (for example, if people who don’t respond systematically have different opinions even within the same demographic group). Combine good outreach and translation (to reduce nonresponse) with weighting for best results.
Will weighting change individual answers or just the reported results?
Weights don’t alter individual responses; they change how much each response counts when calculating totals, averages or percentages. You can and should preserve and inspect both the raw (unweighted) data and the weighted estimates.
Do I need a statistician to weight my survey?
Basic weighting (one or two variables) can be done with guidance and common tools (spreadsheet formulas or survey software). For complex designs, multiple weighting variables, or when you need to estimate uncertainty correctly, getting statistical help is recommended to avoid mistakes and to report results transparently.