Stratified sampling
Stratified sampling divides the population into non-overlapping subgroups (strata) by a relevant characteristic, then draws a random sample within each. Precision improves when the stratifying variable is related to the outcome of interest.
FrameworkSampling design
See it move
A population of 1,000 customers splits into three spend tiers: 600 low-spend, 300 medium-spend and 100 high-spend. Stratified sampling draws separately within each tier, in proportion to its size, so the small high-spend group is never swamped by chance the way a single random draw from the whole 1,000 might swamp it.
The formula
Variables
- Sample size allocated to stratum h
- Total planned sample size
- Number of units in stratum h
- Total population size
Proportional allocation. Each stratum contributes observations in proportion to its share of the population, ensuring representation without oversampling any subgroup.
Check yourself
A retailer has 10,000 customers categorised as low, medium, or high spenders. It randomly selects 100 customers from each spending category. Which sampling method is this, and what is its main advantage?