What is Stratified sampling?
Stratified sampling is a survey method that divides a population into distinct subgroups (strata) and draws random samples from each so all key groups are represented. It improves precision and ensures minority or important subgroups aren't missed.
In stratified sampling you split the population into non-overlapping strata defined by characteristics relevant to your study (for example age groups, neighbourhoods, or language spoken). After creating those strata you draw a sample from each — usually at random — rather than sampling the whole population uniformly. You can allocate sample sizes proportionally to each stratum’s population share or deliberately oversample smaller strata to ensure enough responses for analysis. When you oversample, you later apply weights so overall results reflect the true population.
Usage example
A council wants feedback on local services but knows residents’ views differ by ward and home language. They divide the population into strata by ward and primary language, then randomly select households within each stratum. They oversample small-language groups so those communities’ responses are large enough to analyse, and weight responses back to the population when reporting overall results.
Practical application
Stratified sampling matters because it reduces sampling error and guarantees representation of groups that might be under-represented in a simple random sample. For organisations running multilingual surveys (for example with Hearo), stratifying by language ensures responses from smaller-language communities are captured and understood, avoiding biased conclusions based only on majority-language replies. It also makes subgroup comparisons more reliable and can be a cost-effective way to get precise estimates for key groups without surveying the entire population.
FAQ
How is stratified sampling different from quota sampling?
Stratified sampling uses random selection within each stratum, which supports statistical inference and unbiased estimates. Quota sampling fills fixed numbers of respondents in each subgroup but typically relies on non-random selection, which can introduce bias and limits how confidently you can generalise results.
How do I choose which variables to stratify by?
Pick characteristics that are known before sampling and that are likely to affect the survey outcome or participation (for example region, age, socioeconomic status, or language). Keep the number of strata manageable and use combinations only when necessary, because many small strata increase complexity and cost.
Do I need to weight my results after stratified sampling?
If you sampled each stratum in proportion to its share of the population, weighting isn’t necessary for overall estimates. But if you oversampled or used unequal allocation to ensure subgroup sample sizes, you should apply weights so the final results reflect the true population distribution.