MerQuR+Statistics Glossary

Statistics Glossary → Stratified sampling (design stratum)

Stratified sampling (design stratum)concept

A sampling design in which the population is first divided into mutually exclusive subgroups (strata) and a sample is drawn independently from each stratum.

Strata are formed from characteristics known in advance, such as region, rural/urban status or school type. Because a separate sample is drawn from each stratum, differences between strata do not contribute to sampling error; the variance is computed only from variability within strata. When strata are homogeneous inside and differ from each other, estimates are clearly more precise than under simple random sampling. Making sure that small but important subgroups are represented by enough cases is another aim of stratification.

In complex survey designs, primary sampling units (PSUs) are selected within strata. The variance is then computed from the deviations of PSU totals within strata, and the design degrees of freedom are the number of PSUs minus the number of strata. If strata were sampled disproportionately, each observation must carry a sampling weight. Standard errors computed with ordinary methods that ignore the design are wrong: stratification on its own usually reduces error, while clustering increases it. Stratification should not be confused with post-stratification, an adjustment made after sampling.

An everyday example

To estimate household income in a country, households are selected separately from each of 7 regions. Because incomes differ greatly between regions, stratification prevents the sample from over- or under-representing a region by chance, and the national average is estimated more precisely. Enough households can also be selected from a small region to give a reliable estimate for that region.

In MerQur

In the weighted mean, weighted frequency, weighted total and weighted linear and logistic regression, the 'Stratum' field makes the variance be computed from PSU totals within strata (design df = PSUs − strata). The regressions also offer 'Also add the stratum to the model as a fixed effect (dummies)', and Survey-PHREG has the same role in its 'Design stratum' field.

Analyses where this concept appears

Weighted Mean (Survey Means)Weighted Frequency (Survey Freq)Weighted Total (Survey Total)Weighted Linear Regression (Survey Regression)Weighted Logistic Regression (Survey Logistic Regression)Survey-PHREG (Weighted Cox)

Related concepts

Primary sampling unit (PSU) and cluster samplingSampling weight (design weight)