On this page
Inspect group participation
| Feature | Stratified within-group sampling | Cluster sampling |
|---|---|---|
| Groups participating | Defined sampled strata represented by allocation | Selected subset of clusters |
| Primary choice | Units within each stratum | Which clusters enter |
| Within chosen groups | Specified random sample | All units or a second-stage sample |
Use the same fictional branches
Selecting five records from each of eight branches uses branches as strata in this simple description. Selecting two branches and twenty records in each uses a two-stage cluster description. Both collect forty records, but the allowed selections and dependence differ.
Recognize designs that use both ideas
A design can stratify branches by region and then select clusters within each region. Calling it only “random sampling” loses important structure. Record strata, first-stage units and later-stage selections explicitly.
Keep cost and precision as separate evaluations
Sampling every group may improve subgroup coverage, while selecting fewer groups may reduce collection travel. Actual precision depends on group differences, within-group similarity, allocation and estimator. This comparison is a classification guide, not a universal error ranking.