What is stratified sampling?
Stratified sampling divides a population into non-overlapping groups called strata (for example girls and boys, or rural and urban schools) and then draws a separate random sample inside every stratum. It guarantees that each group is represented in the sample, instead of leaving representation to chance.
The point of stratification is protection. In a plain random draw, a small but important group, say, children in the remotest block, can end up barely present in your sample purely by chance. Stratifying by that characteristic makes their presence a design guarantee, not a lucky outcome. It is the difference between hoping every group is heard and ensuring it.
The allocation formula
A worked example
Suppose an education program covers 12,000 children: 7,200 in rural schools (60 percent), 3,600 in semi-urban schools (30 percent), and 1,200 in urban schools (10 percent). You need a total sample of about 385 for a 5 percent margin of error at 95 percent confidence.
| Stratum | Population | Share | Sample |
|---|---|---|---|
| Rural | 7,200 | 60% | 231 |
| Semi-urban | 3,600 | 30% | 116 |
| Urban | 1,200 | 10% | 38 |
| Total | 12,000 | 100% | 385 |
One caution the table makes visible: 38 urban children give a wide margin of error for the urban estimate on its own. If you must report urban results separately, you need more urban children than their proportionate share, which is where equal or optimum allocation comes in, and the overall estimate then needs weighting.
When to stratify
- When a subgroup matters for the findings and might be underrepresented by chance: girls, children with disabilities, the lowest-performing schools.
- When the outcome differs across groups. Sampling within homogeneous strata removes between-group variation from the error, so stratification often buys slightly better precision for free.
- When you must report subgroup estimates separately. Then allocate more than proportionately to small strata.
Frequently asked questions
What is the difference between stratified and cluster sampling?
Stratified sampling divides the population into groups and samples within every group; it is done to guarantee representation and usually improves precision. Cluster sampling selects only some groups (schools, villages) and surveys inside the selected ones; it is done to save fieldwork cost and usually reduces precision.
Does stratified sampling need a larger sample size?
No, not for an overall estimate. With proportionate allocation it needs the same or a slightly smaller total than simple random sampling at the same precision. The sample only grows if you want each stratum estimated separately with its own tight margin.
Proportionate or equal allocation, which should I use?
Proportionate for one overall estimate: split the sample as the population splits. Equal (or optimum) allocation when every stratum must be reported on its own with similar precision; small strata then get boosted, and the overall estimate must be computed with weights.
See stratification working, live
Sampling Shala's Learn mode draws stratified samples before your eyes and shows what changes. Free, in your browser.
Open Learn mode