What is Post-stratification?
Post-stratification is a weighting technique applied after data collection to make survey results better reflect a known population. It adjusts responses so groups that were under- or over-represented in your sample count proportionally in the final estimates.
Post-stratification means grouping survey respondents into categories (strata) such as age, gender, region or language, then giving each group a weight so the sample matches known population proportions. You choose the variables you can benchmark against reliable population data, calculate each strata’s expected share in the population and compare that to its share in your sample, then multiply responses by the ratio (population % ÷ sample %) to produce weighted results. The method corrects for differential response rates or convenience samples, but it depends on accurate population benchmarks and enough responses in each strata — it cannot fix biases tied to characteristics you didn’t measure.
Usage example
A council runs an online consultation and finds that 70% of responses come from English speakers while community data shows only 40% of residents prefer English. The analyst post-stratifies by language: weight for English = 0.40 ÷ 0.70 ≈ 0.57, weight for non-English = 0.60 ÷ 0.30 = 2.0. Applying these weights adjusts the consultation results so they better reflect the full community’s language mix.
Practical application
Post-stratification matters because real-world surveys often over- or under-represent parts of the population — for example, some language groups or age bands may be less likely to respond. By aligning the sample to known population distributions, organisations get more accurate estimates of overall attitudes and needs, which helps make decisions fairer and reporting more defensible. In multilingual engagement, post-stratification helps ensure smaller-language communities aren’t drowned out by a larger responding group. Remember: it improves estimates when you have good benchmarks and enough respondents per strata, but it’s not a substitute for inclusive outreach or for measuring important variables you didn’t collect.
FAQ
When should I use post-stratification?
Use it when your sample differs from known population distributions on variables you care about and you have reliable benchmarks (census, registration lists, administrative data). It’s especially useful after convenience or online surveys where some groups respond less often.
Can I post-stratify by language?
Yes. If you know the share of the population speaking each language, you can weight responses so language groups are represented proportionally. Be cautious if some language groups have very few respondents — large weights can increase uncertainty and distort results.
Does post-stratification remove all survey bias?
No. It corrects bias only for the variables you stratify on (and only if your benchmarks are accurate). It cannot fix bias from unmeasured factors, nor fully compensate for systematic differences in how groups interpret questions.
How is post-stratification implemented in practice?
Weights are calculated as population proportion divided by sample proportion for each strata and applied during analysis. You can do this in spreadsheets or statistical software. When many benchmarking variables are needed, techniques like raking (iterative proportional fitting) are an alternative to keep cells from becoming too small.