Proportional Cluster Sampling — Probability Proportional to Size
Proportional Cluster Sampling (Probability Proportional to Size) · Also known as: PPS cluster sampling, proportional-to-size cluster sampling, size-proportional cluster sampling, probability proportional to size sampling
Proportional cluster sampling selects naturally occurring groups (clusters) from a population with probability proportional to each cluster's size, so that larger clusters have a higher chance of selection while every individual element retains an equal overall inclusion probability. This design efficiently handles large, geographically dispersed populations and is the backbone of national health, education, and social surveys worldwide.
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 proportional cluster sampling when (1) the population is naturally grouped into identifiable clusters whose sizes are known or estimable from a frame, (2) individual-level listing is costly or infeasible before sampling, and (3) clusters vary markedly in size — PPS corrects the bias that would arise from ignoring size differences. It is the standard design for national or regional surveys of households, schools, health facilities, and similar hierarchical populations. Do not use it when cluster sizes are unknown or highly unreliable, as inaccurate MOS data invalidate the PPS mechanism. Avoid it also when the analytical goal requires very precise sub-group estimates, because clustering inflates variance relative to simple random sampling of the same total n; in that case, stratified sampling with independent strata is preferable.
Strengths & limitations
- Produces a self-weighting sample when a fixed number of elements is drawn per cluster, simplifying estimation and reducing the need for complex weighting.
- Operationally efficient for large, geographically dispersed populations — interviewers work within clusters, reducing travel costs dramatically.
- Gives every population element an equal probability of inclusion regardless of cluster size, satisfying the core requirement of probability sampling.
- Scales naturally to multistage designs; additional stages (e.g., classrooms within schools, households within villages) can be added with consistent PPS logic.
- Well-established theoretical foundations and variance estimation formulas available in standard survey sampling texts.
- Requires accurate, up-to-date measures of cluster size for the entire sampling frame before selection; outdated registers introduce selection bias.
- Estimates of totals and means are unbiased only if the MOS accurately reflects the actual cluster sizes at the time of data collection — population drift between frame construction and fieldwork increases error.
- Cluster-based designs yield higher variance than equivalent simple random samples due to within-cluster homogeneity (the design effect, DEFF, is typically 1.5–3.0 in social surveys).
- Variance estimation is more complex than for SRS, requiring specialised software (e.g., survey packages in R, Stata, or SAS) that accounts for the clustered structure.
Frequently asked
What is the difference between proportional cluster sampling and proportional stratified sampling?
In proportional stratified sampling the population is divided into strata and individual elements are drawn directly from each stratum in proportion to stratum size — no clustering is involved. In proportional cluster sampling entire groups (clusters) are drawn as primary sampling units with probability proportional to their size, and individuals are sub-sampled only from the selected clusters. Stratified sampling is generally more precise for a given n; cluster sampling is more operationally feasible when a complete element-level frame does not exist.
When does a PPS cluster design become self-weighting?
A design is self-weighting when every element in the population has the same overall probability of inclusion. In a two-stage PPS design this occurs when a fixed number of elements (b) is sampled from every selected cluster, because the two-stage inclusion probability n*b / N is constant regardless of cluster size. When the within-cluster sampling fraction varies across clusters, the design is no longer self-weighting and explicit design weights must be computed and applied.
How do I account for clustering when analyzing the data?
Use survey-aware analysis software that accepts a complex survey design specification — typically the primary sampling unit identifier, stratum identifier, and design weight. In R, the 'survey' package (Lumley) handles this via svydesign(); in Stata, the svy prefix commands are standard; SAS PROC SURVEYMEANS/SURVEYREG also support clustered designs. Ignoring the clustering and analysing with standard OLS or t-tests produces standard errors that are too small, inflating Type I error rates.
What sample size do I need for a PPS cluster design?
Start from the required sample size for a simple random sample (based on target precision, alpha level, and expected proportion or mean), then multiply by the anticipated design effect (DEFF). For typical social surveys DEFF values between 1.5 and 2.5 are common, meaning you may need 50–150 % more observations than an SRS of the same target precision. Pilot data or published DEFF values from similar surveys in the same domain are the best guide for the design effect assumption.
Can PPS cluster sampling be extended to more than two stages?
Yes. Multistage PPS designs are standard in large national surveys. In a three-stage design you might select districts PPS by population, then villages PPS by household count within selected districts, then households by simple random sampling within selected villages. Each stage applies PPS selection so the overall inclusion probability remains constant, preserving the self-weighting property. Each additional stage adds a further component to the design effect, so precision decreases; this is offset by the large gains in operational feasibility.
Sources
- Cochran, W. G. (1977). Sampling Techniques (3rd ed.). John Wiley & Sons. ISBN: 978-0471162407
- Kish, L. (1965). Survey Sampling. John Wiley & Sons. ISBN: 978-0471489009
How to cite this page
ScholarGate. (2026, June 3). Proportional Cluster Sampling (Probability Proportional to Size). ScholarGate. https://scholargate.app/en/survey-methodology/proportional-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
- Multistage SamplingSurvey Methodology↔ compare
- Proportional Stratified SamplingSurvey Methodology↔ compare
- Simple random samplingSurvey Methodology↔ compare
- Stratified SamplingSurvey Methodology↔ compare
- Systematic SamplingSurvey Methodology↔ compare