What is Survey weighting?
Survey weighting adjusts responses so a survey sample better matches the known characteristics of the population you care about. It rebalance results when some groups are over- or under-represented in your collected responses.
Survey weighting is a statistical correction applied to survey answers to make the analysis reflect a target population (for example a town, school, or customer base) when the people who responded are not a perfect cross‑section of that population. Weighting assigns each response a numerical weight — typically population proportion divided by sample proportion for a subgroup — so groups that are under-represented count for more and over-represented groups count for less. Common approaches include simple post‑stratification (one variable), raking (iterative adjustment across multiple variables), and use of design or probability weights when sampling was structured. Weighting requires reliable population benchmarks (census, enrolment lists, customer records) and enough responses in each subgroup to produce stable estimates.
Usage example
A council runs an online consultation and knows the town is 30% people aged 18–34 and 70% older residents, but their respondents are 10% aged 18–34 and 90% older. To correct this imbalance they compute weights: weight for 18–34 = 0.30/0.10 = 3.0, weight for 35+ = 0.70/0.90 ≈ 0.78, then apply those weights when calculating overall percentages so the final results better reflect the actual population.
Practical application
Why it matters: Without weighting, survey results can mislead decision‑makers — for example making an issue look less important because under‑represented groups were less likely to respond. For Hearo users, weighting is particularly useful when certain language groups or demographics respond more or less often (e.g., English speakers dominating an online survey). Weighting can help estimate what the whole community thinks, support defensible public consultations, and prioritise outreach. Important caveats: weighting does not create missing responses or fix biased question wording; it increases estimate uncertainty (especially for small subgroups); and it depends on accurate population benchmarks. Best practice is to gather good demographic/language benchmarks, ensure adequate sample sizes in key groups, report both weighted and unweighted results, and be transparent about the methods used.
FAQ
Can weighting create opinions we didn't actually hear from under-represented groups?
No — weighting scales the responses you did receive so they count proportionally to the population, but it does not invent answers or replace missing voices. If a group is absent or has very few responses, weighted estimates will be unstable and outreach should focus on collecting more real responses.
Do I need to weight every survey?
Not always. Weighting is helpful when your sample clearly differs from the population in important ways (age, language, gender, geography) and you have reliable benchmarks. For small, targeted, or convenience surveys where representativeness isn’t the goal, unweighted summaries may be appropriate.
How does weighting work with multilingual surveys?
Multilingual surveys can produce language-related nonresponse (some language groups respond less). You can use weighting by language — or by a combination of language and demographics using raking — to adjust for that imbalance, provided you have trustworthy population figures for each language group and enough responses per group to be reliable.