In random sampling, sometimes the sample is drawn separately from different disjoint subsets of the population. Each such subset is called a stratum. (The plural of stratum is strata.) Samples drawn in such a way are called stratified samples. Estimators based on stratified random samples can have smaller sampling errors than estimators computed from simple random samples of the same size, if the average variability of the variable of interest within strata is smaller than it is across the entire population; that is, if stratum membership is associated with the variable.

For example, to determine average home prices in the U.S., it would be advantageous to stratify on geography, because average home prices vary enormously with location. We might divide the country into states, then divide each state into urban, suburban, and rural areas; then draw random samples separately from each such division.