Direct answer
Stratification samples within defined subgroups, clustering samples groups of related units, and weighting changes each observation’s contribution; all can alter precision and require design-aware estimation.
Visual explanation
Sampling design changes represented information
What this calculation tells you
Complex sampling designs determine who can be observed and how sample records represent a target population. Their structure is part of the estimator, not a detail added after analysis.
Record strata, clusters, selection probabilities, stages, finite-population information, and weight meaning. Use software and methods that carry those design variables into variance estimation.
Where it is used
Public surveys
Explain why released estimates include design-aware standard errors.
Experiments
Account for treatment delivered by site, class, or household clusters.
Market research
Document weighting and stratified allocation choices.
Education
Visualize representation versus raw record count.
Common situations
- Planning cluster sampling.
- Allocating a sample across strata.
- Interpreting survey weights.
- Explaining effective rather than literal sample size.
Start with the statistical question
Record strata, clusters, selection probabilities, stages, finite-population information, and weight meaning. Use software and methods that carry those design variables into variance estimation.
One population is sampled three ways: scattered simple units, colored strata with within-stratum samples, and whole clusters. Weight sizes then change represented population mass.
Worked example
If respondents come in similar clusters of average size m with intraclass correlation ρ, the equal-cluster approximation DEFF=1+(m−1)ρ shows why 100 clustered observations may carry less information than 100 independent ones.
Assumptions that carry the result
The simple design-effect formula assumes equal cluster size and one correlation approximation. Kish effective sample size addresses weight variation only and is not a complete survey variance estimator.
Interpret the result without overreaching
These arithmetic summaries do not replace replicate weights, Taylor linearization, multistage design variables, nonresponse adjustment, calibration, or specialist survey analysis.
- Analyzing clustered records as independent.
- Treating frequency weights as survey weights.
- Reporting effective sample size as the literal respondent count.
Practical questions
Frequently asked questions
Are weighted results automatically representative?
No. Weight construction, coverage, nonresponse, calibration, and the target population all matter.
Does clustering always reduce precision?
Positive within-cluster similarity commonly does; the effect depends on correlation, cluster sizes, allocation, and estimator.
Can I use n_eff in every ordinary formula?
Not safely. Effective-size summaries are approximations and do not reproduce every feature of a complex variance estimator.
Further reading
Authoritative sources
Use these primary and professional resources to check definitions, conventions, or requirements that may extend beyond this guide.