Definition · Plain-language
Stratified sampling
Stratified sampling is a probability sampling method that divides a population into mutually exclusive subgroups, or strata, and then draws a random sample from within each one.
The step most authors miss
Doing CRediT right? Don’t stop at the statement.
A CRediT statement credits you inside one paper. The recognition CRediT was built for happens when those roles are tied to you, persistently. Sign in with your ORCID — free — and claim your CRediT contributions on casrai.org, the home of the standard. They become a verified, portable part of your identity, not a line that disappears into one PDF.
Free: claim your contributions, then export a journal-ready CRediT statement, schema.org structured data, JATS XML, CSV or BibTeX — and preview your public profile. A membership publishes that profile publicly and verifies the journals you serve.
Divide into strata, then sample within each
Stratified sampling works in two steps. First, the researcher partitions the whole population into strata — mutually exclusive and collectively exhaustive subgroups defined by a characteristic relevant to the study, such as age band, region, or department. Then a random sample is drawn independently from within each stratum. Because every subgroup is sampled, no stratum can be missed by chance, which is the key difference from simple random sampling. The characteristic used to form strata should be related to the outcome being measured, or stratifying gains little.
Proportionate and disproportionate allocation
In proportionate stratified sampling, the number drawn from each stratum mirrors that stratum’s share of the population, so a group that is 20% of the population supplies 20% of the sample. In disproportionate allocation, researchers deliberately over-sample small but important strata — for instance a rare subgroup — to obtain enough cases for separate analysis, then weight the results to restore overall representativeness. Proportionate allocation is the common default; disproportionate allocation trades simplicity for statistical power within small subgroups.
Why it improves precision
When strata differ markedly on the variable of interest but members within each stratum are relatively similar, stratified sampling yields more precise estimates than simple random sampling of the same size, because it removes between-stratum variability from the sampling error. It also guarantees representation of every subgroup, which matters for fairness and for subgroup analysis. The cost is that it requires a sampling frame with known stratum membership in advance. This contrasts with cluster sampling, which divides the population into naturally occurring clusters and samples whole clusters, chosen mainly for convenience and cost.
Key facts
At a glance
- Definition: probability sampling that splits the population into strata, then samples within each
- Type: a probability (random) sampling method
- Strata: mutually exclusive subgroups sharing a relevant characteristic
- Allocation: proportionate (mirrors population) or disproportionate (over-samples small strata)
- Benefit: better representativeness and precision when strata differ
- Contrast: cluster sampling samples whole naturally occurring groups instead
Common misconceptions
What people often get wrong
Often heard: Stratified sampling and cluster sampling are basically the same method.
Actually: They differ fundamentally. Stratified sampling samples within every subgroup to ensure representation; cluster sampling randomly selects whole clusters and is chosen mainly for cost and convenience.
Often heard: Stratified sampling is a non-probability technique.
Actually: It is a probability method: within each stratum, units are selected at random, so every member of the population has a known, non-zero chance of selection.
Often heard: You can stratify on any variable and still gain precision.
Actually: Gains come only when the stratifying variable is related to the outcome and strata differ on it. Stratifying on an irrelevant characteristic adds complexity without improving estimates.
