Weighted Sampling — Weighted Probability Sampling
Weighted Probability Sampling · Also known as: probability proportional to size sampling, PPS sampling, unequal probability sampling, importance sampling
Weighted sampling is a probability-based design in which units are selected with unequal probabilities proportional to a known auxiliary measure of size or importance. Sampling weights — the inverse of inclusion probabilities — are applied during analysis so that each sampled unit correctly represents the population units it stands for. The approach underpins large-scale government, health, and social surveys where simple random sampling would be inefficient.
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.
+8 more
When to use it
Use weighted sampling when the population is heterogeneous and an auxiliary size variable is available that correlates with the outcome, making equal-probability sampling inefficient; when rare but important sub-groups must be represented without over-sampling the whole population; or when combining frames with different coverage probabilities. It is standard for national household surveys, establishment surveys, health surveillance, and multi-stage assessments. Do not use it when no reliable auxiliary variable exists — a weakly correlated size measure inflates rather than reduces variance. Avoid it also when the analytic team cannot apply design-based variance estimation: ignoring weights yields biased point estimates, and using ordinary standard errors on weighted data produces anti-conservative inference.
Strengths & limitations
- Increases statistical efficiency when the size measure is strongly correlated with the outcome, reducing variance compared to simple random sampling at the same total n.
- Allows representative coverage of rare sub-groups by boosting their selection probabilities without proportionally inflating total sample size.
- Produces design-unbiased estimates for the whole population and planned sub-domains when weights are correctly applied via the Horvitz-Thompson framework.
- Base weights can be post-stratified or calibrated to external benchmarks to further reduce bias from frame imperfections or differential non-response.
- Widely supported by established statistical theory and mature software implementations across all major survey-analysis platforms.
- Requires a reliable auxiliary size variable; a poorly chosen or outdated measure can increase variance rather than reduce it.
- Highly variable weights inflate the design effect and can make weighted estimates less precise than a simple random sample of the same size.
- Design-based analysis requires specialist software and survey-aware estimation procedures; routine regression output on weighted data is incorrect.
- Non-response and frame undercoverage can bias estimates even after weighting if the auxiliary variable does not predict response propensity.
- Weight construction and documentation are more complex and costly than equal-probability designs.
Frequently asked
What is the difference between a sampling weight and an analytic weight?
A sampling (design) weight is the inverse of a unit's probability of selection; it tells the analyst how many population units each sampled unit represents. An analytic weight is a user-defined precision weight used in some model-fitting routines to down-weight noisy observations. In survey analysis you need design weights — or calibrated final weights that incorporate design, non-response, and post-stratification — not arbitrary analytic weights.
Do I still need to use weights if my sample is very large?
Yes. Large sample size does not remove bias from ignoring weights; it simply means you estimate a biased quantity more precisely. If selection probabilities correlate with the outcome — which is the whole point of the design — unweighted estimates are systematically wrong regardless of n.
What software correctly applies survey weights?
R's survey package (Thomas Lumley), Stata's svy suite, SAS PROC SURVEYMEANS and SURVEYREG, SPSS Complex Samples, and Mplus (for structural equation models) all implement design-consistent point and variance estimation. Always declare the design — weights, strata, and primary sampling unit identifiers — before running any estimation command.
When should I trim extreme weights?
When a handful of units carry very large weights (e.g., exceeding five times the mean weight), they dominate estimates and inflate variance. A common practice is to cap weights at the 95th or 99th percentile and renormalize so total weights still sum to the population size. Always report the trimming threshold and compare key estimates before and after trimming.
How does weighted sampling differ from post-stratification?
Weighted sampling refers to the design phase: units are selected with unequal probabilities. Post-stratification is an analytic adjustment applied after data collection to force sample margins to match known population totals. Both produce survey weights; post-stratification is commonly applied on top of design weights to reduce residual bias from frame imperfections or differential non-response.
Sources
- Cochran, W. G. (1977). Sampling Techniques (3rd ed.). John Wiley & Sons. ISBN: 978-0471162407
- Horvitz, D. G., & Thompson, D. J. (1952). A generalization of sampling without replacement from a finite universe. Journal of the American Statistical Association, 47(260), 663-685. DOI: 10.2307/2280784 ↗
How to cite this page
ScholarGate. (2026, June 3). Weighted Probability Sampling. ScholarGate. https://scholargate.app/en/survey-methodology/weighted-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