Stratified sampling
Sample separately within meaningful population strata
Stratified sampling partitions the population into non-overlapping strata and draws probability samples within each. Proportionate allocation mirrors population sizes; disproportionate allocation can improve precision or ensure enough observations for small subgroups. Population estimates must reflect unequal selection probabilities when allocation is disproportionate. Strata are sampled throughout, unlike cluster designs that first select only some groups.
Use when subgroup estimates matter or reliable auxiliary information identifies strata related to the outcome, and each frame unit can be assigned to a stratum.
Strengths
- Ensures planned coverage of important subgroups
- Can improve precision when strata explain outcome variation
Limitations
- Requires accurate stratum information before selection
- Poor weighting can distort overall estimates
Know the boundary
Matching population proportions alone is not stratified random sampling if within-stratum recruitment is non-random.