Disproportional Cluster Sampling — Unequal-Probability Cluster Sampling
Disproportionate Cluster Sampling · Also known as: disproportionate cluster sampling, unequal-probability cluster sampling, variable-rate cluster sampling, non-proportional cluster sampling
Disproportional cluster sampling is a probability-based survey design in which naturally occurring groups (clusters) are selected as primary sampling units, but the number of clusters or elements drawn from each group is not proportional to that group's share of the population. By deliberately over- or under-sampling certain clusters, researchers gain analytic flexibility and precision where it matters most, at the cost of requiring post-hoc weighting for population-level inference.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use disproportional cluster sampling when: (1) sub-group analyses of small or rare clusters are required — proportional allocation would leave those sub-groups with too few cases for reliable estimates; (2) clusters vary dramatically in size and strict proportionality would cause impractical fieldwork burdens in very large clusters; or (3) precision targets differ across sub-domains and variable sampling fractions are the most efficient way to meet each target simultaneously. Do NOT use it when all clusters are similar in size and composition, the analysis does not require sub-group disaggregation, and the added complexity of weighting and design-based variance estimation cannot be justified — in those cases, proportional cluster sampling or simple random sampling is preferable and easier to defend.
Strengths & limitations
- Guarantees adequate sub-group sample sizes for domains that would be under-represented under proportional allocation.
- Allows analytic precision to be tuned independently for each cluster group, improving overall study efficiency.
- Retains the cost and logistical advantages of cluster sampling — elements are collected in geographic or administrative batches.
- Flexible: works with any secondary selection mechanism (census within cluster, SRS sub-sampling, PPS).
- Transparent probability framework enables unbiased population-level inference through weighting.
- Requires accurate auxiliary information about cluster sizes to compute valid design weights; errors in frame data propagate into biased estimates.
- Analysis is more complex than proportional designs — weighted estimators and design-based standard errors are mandatory.
- Design effect from intra-cluster correlation (ICC > 0) inflates effective sample size requirements compared to simple random sampling.
- Weight variability (high coefficient of variation of weights) reduces effective sample size and inflates variance, partially offsetting the precision gained from targeted over-sampling.
Frequently asked
What is the key difference between disproportional and proportional cluster sampling?
In proportional cluster sampling the number of elements drawn from each cluster mirrors that cluster's share of the total population, so all elements have equal inclusion probabilities and no weighting is needed. In disproportional cluster sampling the fractions differ deliberately across clusters, giving some elements higher selection chances than others. Disproportional designs require inverse-probability weights to produce unbiased population estimates but offer greater analytic flexibility for sub-group comparisons.
Do I always need to weight the data?
Yes, if your goal is population inference. Without weights, your estimates describe only the collected sample — which, because of the unequal selection fractions, does not mirror the population. Weights restore representativeness. The one exception is purely descriptive within-cluster analysis where the goal is to characterise each cluster on its own, not to generalise to the population.
How do I calculate the design effect (DEFF)?
The design effect is the ratio of the actual variance of an estimator under the complex design to the variance that would arise under simple random sampling of the same total size: DEFF = Var_complex / Var_SRS. A DEFF greater than 1 means the cluster design is less efficient than SRS; a DEFF less than 1 (possible when disproportional over-sampling targets high-variance sub-groups) means it is more efficient for that estimator. Software packages such as Stata (estat effects) or R (survey::svymean) compute DEFF automatically.
When should I prefer disproportional stratified sampling instead?
If you can enumerate individual population members into strata before sampling — and field access is not a constraint — stratified sampling is generally more efficient and simpler to weight. Cluster sampling is preferred when a complete enumeration of individuals is impossible or prohibitively expensive and sampling of geographic or administrative groups is the only practical approach. In many large-scale surveys both strategies are combined: stratification at the first stage, cluster selection at the second, with disproportional allocation across strata.
What software handles disproportional cluster sampling correctly?
Any software with a complex survey module: Stata (svyset with cluster() and strata() options), R (survey package, svydesign()), SAS (PROC SURVEYMEANS / PROC SURVEYLOGISTIC), SPSS Complex Samples, and Mplus (TYPE=COMPLEX). These tools apply design weights and compute design-based standard errors via Taylor linearisation or replication, which is mandatory for valid inference from cluster designs.
Sources
- Kish, L. (1965). Survey Sampling. John Wiley & Sons. ISBN: 978-0471489009
- Cochran, W. G. (1977). Sampling Techniques (3rd ed.). John Wiley & Sons. ISBN: 978-0471162407
How to cite this page
ScholarGate. (2026, June 3). Disproportionate Cluster Sampling. ScholarGate. https://scholargate.app/en/survey-methodology/disproportional-cluster-sampling
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Cluster SamplingSurvey Methodology↔ compare
- Disproportional Stratified SamplingSurvey Methodology↔ compare
- Multistage SamplingSurvey Methodology↔ compare
- Proportional Cluster SamplingSurvey Methodology↔ compare
- Systematic SamplingSurvey Methodology↔ compare
- Weighted SamplingSurvey Methodology↔ compare