What is Stratified Sampling?
Stratified sampling is a method that splits a population into distinct subgroups (strata) and draws samples from each so that key groups are represented. It improves representativeness and precision compared with simple random sampling when groups differ on the things you care about.
In stratified sampling you first divide the population into meaningful subgroups — for example by language, age band, neighbourhood or income level — then select samples separately from each subgroup. The selection within each stratum is usually random. You can choose to sample proportionally (each stratum represented in the sample in the same share as in the population) or to oversample smaller but important strata so you can analyse them reliably. The method ensures that the sample includes enough people from each group to make valid comparisons or estimates.
Usage example
A council wants feedback on a new service and knows residents speak English, Spanish and Somali. Rather than randomly contacting everyone (which might miss smaller-language communities), the council divides its contact list into three strata by primary language and randomly samples 500 English speakers, 250 Spanish speakers and 250 Somali speakers so each community is included in the results.
Practical application
Stratified sampling matters because many public consultations and surveys need reliable results for specific subgroups — for example, different language communities, age groups or neighbourhoods. It reduces the chance that smaller or harder-to-reach groups are underrepresented, improves the precision of subgroup estimates, and supports fairer, more inclusive decision-making. For multilingual work, stratified sampling helps ensure minority-language communities are heard without having to run separate surveys for each group.
FAQ
How is stratified sampling different from quota sampling?
Both aim to get representation from subgroups. Stratified sampling selects respondents randomly within each stratum (supporting statistical inference), while quota sampling fills predefined targets non-randomly (often by convenience), which can introduce bias. Stratified sampling is preferable when you need reliable, generalisable estimates.
When should I oversample a stratum?
Oversample when a subgroup is small in the population but important to analyse on its own (for example a minority language community). You can oversample and later apply weights in analysis so overall estimates remain representative of the population.
How many strata should I use?
Use as many strata as needed to capture meaningful differences (language, age, geography) but avoid creating so many small strata that there aren’t enough people in each. A handful of well-chosen strata usually works best.
What do I need to run a stratified sample?
You need a way to classify your population into the chosen strata (a contact list or frame with the relevant attributes) and a sampling method to select respondents inside each stratum. Accurate population counts or estimates are important if you plan proportional sampling or weighting.