Introduction
Stratified random sampling is a probability-based sampling method that divides a population into homogeneous subgroups (strata) and then draws random samples from each stratum. By ensuring that each important subgroup is represented, stratified sampling can increase precision, reduce sampling error, and produce estimates that better reflect the population’s structure than a simple random sample of the same size.
Key takeaways
– Population is split into mutually exclusive, meaningful strata (e.g., age, gender, income).
– Random samples are drawn from each stratum — either proportionally or disproportionately.
– Stratified sampling improves precision, especially when strata differ from one another.
– It requires a complete list of the population and reliable classification into strata.
– Common uses include surveys, market research, public-health studies, and index replication (e.g., bond portfolio sampling).
How stratified random sampling works (conceptual)
1. Identify a population and one or more variables that meaningfully partition it into subgroups (strata).
2. Ensure strata are mutually exclusive and collectively exhaustive — each unit belongs to exactly one stratum.
3. Decide how many observations to take from each stratum:
• Proportionate stratification: sample size in each stratum is proportional to its size in the population.
• Disproportionate stratification: sample sizes differ from proportional allocation (used to oversample small but important strata).
4. Within each stratum, select a simple random sample (or another random selection method).
5. Combine the stratum samples into the overall sample; apply weights in analysis if sampling was disproportionate.
Simple random sample vs. stratified random sample
– Simple random sampling: every member of the population has equal probability of selection; easier and cheaper when population is homogeneous or sample is small.
– Stratified sampling: ensures representation across key subgroups and is more precise when population heterogeneity is aligned with chosen strata.
Types of stratified random sampling
– Proportionate stratified sampling: n_h = (N_h / N) * n
• n_h = sample size for stratum h, N_h = population size of stratum h, N = total population, n = total sample size.
• Yields a sample that mirrors population proportions; often simpler to analyze.
– Disproportionate stratified sampling: n_h is chosen deliberately not in proportion to N_h (e.g., oversample a small but important stratum to reduce variance or to ensure minimum sample sizes for subgroup analysis).
• Requires weighting in analysis to produce unbiased population estimates.
Practical step-by-step guide (how to implement)
1. Define the population and research objectives
• What population do you want to represent? What estimates do you need (overall and by subgroup)?
2. Choose stratification variable(s)
• Pick variables strongly related to the outcome (e.g., age, gender, geography, income). Prefer variables that explain between-strata variance.
3. Obtain a sampling frame and classify units
• You need a list of all population units and a reliable way to assign each unit to exactly one stratum.
4. Decide sample size (total n)
• Based on desired precision, budget, and statistical power.
5. Allocate sample to strata
• Proportionate allocation: n_h = (N_h/N) * n
• Neyman (optimal) allocation (if you have stratum variances): n_h = n * (N_h * S_h) / Σ(N_j * S_j)
• S_h is the standard deviation in stratum h for the variable of interest.
• Disproportionate allocation: pick n_h to ensure enough observations in small but important strata (prepare to weight results).
6. Draw random samples within each stratum
• Use simple random sampling, systematic sampling, or random-digit selection within each stratum.
7. Conduct data collection
• Track response rates by stratum; low response in a stratum may require follow-up or reweighting.
8. Analyze and weight (if needed)
• If disproportionate sampling was used (or differential nonresponse occurred), apply weights w_h = N_h / n_h to restore representativeness for population-level estimates.
9. Report methods and limitations
• Document how strata were defined, allocation method, response rates, and weighting approach.
Worked example (proportionate allocation)
Suppose:
– Total population N = 180,000
– Strata by age: 24–28 (N1 = 90,000), 29–33 (N2 = 54,000), 34–37 (N3 = 36,000)
– Desired total sample n = 50,000
Proportionate allocation:
– n1 = (90,000 / 180,000) * 50,000 = 25,000
– n2 = (54,000 / 180,000) * 50,000 = 15,000 (Investopedia example showed 16,667 implying different N2; adjust per actual N_h)
– n3 = (36,000 / 180,000) * 50,000 = 10,000
(Adjust numbers to match actual stratum counts; the Investopedia example yields 25,000 / 16,667 / 8,333 if underlying N_h differ.)
If disproportionate sampling is used (e.g., sample half of the smallest stratum), you must weight results back to population proportions:
– Weight for stratum h = N_h / n_h (use normalized weights when combining estimates).
Advantages of stratified random sampling
– More precise estimates than simple random sampling when strata are internally homogeneous and differ from each other.
– Ensures representation of small but important subgroups.
– Can reduce sample size (and cost) for a given level of precision.
– Facilitates subgroup (stratum-level) analysis.
Disadvantages and limitations
– Requires a complete sampling frame and accurate classification of units into strata.
– Strata must be mutually exclusive; overlapping categories invalidate probability sampling assumptions.
– More complex design, implementation, and analysis (especially with disproportionate allocation and weighting).
– If strata are poorly chosen (not related to the outcome), gains in precision can be small or nonexistent.
– Nonresponse or classification errors within strata can bias results.
When to use stratified random sampling
– When the population has heterogeneity that is well-explained by known characteristics (e.g., region, age, income).
– When you need precise subgroup estimates or want to ensure coverage of small subgroups.
– When a sampling frame with strata labels is available.
– Less useful when you cannot reliably classify the population or when the variables for stratification are unrelated to your outcome of interest.
How to choose strata (practical tips)
– Choose variables strongly correlated with the primary outcome(s) or that are of analytic interest.
– Keep strata few and meaningful — too many strata complicates sampling and may create tiny strata.
– Make strata mutually exclusive and collectively exhaustive.
– Consider operational constraints: availability of data, cost, and ease of classification.
– If uncertain, pilot studies or historical data can help estimate within-stratum variance to inform allocation (e.g., Neyman allocation).
Common pitfalls and how to avoid them
– Overlapping strata: ensure each unit belongs to exactly one stratum.
– Missing or incomplete sampling frame: obtain or construct a reliable frame before designing the stratification.
– Small sample sizes in strata: use disproportionate sampling or combine similar strata.
– Failure to weight after disproportionate allocation: leads to biased population estimates — always apply appropriate weights.
– Ignoring nonresponse: track response by stratum and adjust weights or follow up aggressively where response is low.
Fast fact
Stratified random sampling is sometimes described as “proportional random sampling.” Note: “quota sampling” is a different concept (non-probability); don’t conflate probability-based stratified sampling with non-probability quota sampling.
Which sampling method is best?
– No single “best” method fits every study. Use stratified random sampling when subgroup representation and precision are priorities and when you can reliably stratify the population. Use simple random sampling when the population is homogeneous or when you lack a reliable frame for stratification.
The bottom line
Stratified random sampling is a powerful and widely used technique to improve precision and ensure representation of important subgroups. It requires careful choice of strata, a complete sampling frame, and attention to allocation and weighting. When implemented properly, it yields more informative and reliable estimates than an unstratified simple random sample of the same size — especially when strata account for meaningful heterogeneity in the population.
References and further reading
– Investopedia: “Stratified Random Sampling”
– Cochran, W. G. (1977). Sampling Techniques (3rd ed.). Wiley. (for Neyman allocation and sampling theory)
– Lohr, S. L. (1999). Sampling: Design and Analysis. Duxbury Press.
Editor’s note: The following topics are reserved for upcoming updates and will be expanded with detailed examples and datasets.