Survey weighting diagnostic
How concentrated are your survey weights?
Check effective sample size, weighting efficiency, and the influence of the largest weights. See the full sample alongside the groups you plan to report.
Start with your final respondent weights.
Use one row per person. Data stay in this page's memory until you clear them or reload. Optional analytics and marketing tags are disabled on this page. Contact links do not carry your data.
Weight-only diagnostic
See what your weights imply
Add a weight file or try the demonstration. You will see the positive-weight base, Kish effective sample size (ESS), distribution, and concentration here.
Weight variation answers one part of the question.
Weight concentration
Kish effective sample size (ESS) summarizes unequal weights. Equal positive weights give an ESS equal to the number of respondents. Greater dispersion reduces it. Weighting efficiency is the ratio of Kish ESS to the positive-weight base.
Adjustment to the population
Unequal weights may correct differences in who participated. High efficiency alone cannot establish whether those corrections worked. This release does not compare your weights with population targets or measure bias.
Uncertainty for a result
Actual uncertainty depends on the outcome and sampling design, including strata, clusters, and calibration. A weight-only summary cannot supply every estimate's design effect, confidence interval, or margin of error.
Review the weighting decisions behind the numbers.
Bring the sample source, weighting variables, population controls, and planned reporting groups to the discussion. A large weight can be justified; its purpose matters.
Discuss a weighting reviewDoes a low effective sample size mean the weighting is wrong?
No. It identifies dispersion in the weights. The adjustments may reduce bias in important outcomes while increasing weight variation. Assess the weighting variables, target alignment, and sensitivity of the findings before changing the weights. There is no universal efficiency score that certifies quality.
Why do my subgroups have different efficiencies?
Each group can contain a different range of weights. This calculator recomputes Kish ESS within each selected group, keeping its unweighted positive base visible. Group ESS values cannot be added to recover the overall ESS. These are weight diagnostics; formal survey domain estimates need the original design information.
Should we trim the largest weights?
The largest weights deserve inspection, not an automatic cutoff. Trimming can reduce dispersion while weakening the adjustment to population controls or changing estimates. This release does not trim, recalibrate, or recommend replacement weights. Benchmark alignment and trimming sensitivity are planned extensions.
Can I use this result in a sample-size calculator?
Do not automatically substitute this unequal-weighting effect for a validated design effect. A planning adjustment must fit the outcome, sampling design, and inference assumptions.
The detectable-difference calculator supports independent-group planning under stated assumptions. It does not turn a weight-only diagnostic into full survey uncertainty.
What happens to my file and respondent IDs?
The browser reads the file locally. This tool does not upload records, place them in browser storage, or attach them to a contact request. Use Clear all data to remove the loaded input and results. The optional text report contains aggregate diagnostics and selected group labels; respondent IDs and raw rows are omitted. Review the report before sharing it.
Calculation policy
A reproducible weight-only calculation.
Kish ESS = (Σw)² / Σ(w²). The unequal-weighting effect is n / ESS, and weighting efficiency is ESS / n. Here n counts only positive-weight records after any explicitly chosen exclusions. The coefficient of variation uses the population standard deviation, with divisor n, so the unequal-weighting effect equals 1 + CV².
The calculation is descriptive for any supported set of positive weights. Interpreting it as a variance approximation requires assumptions about outcome variances and their relationship to weights. It does not incorporate strata, clusters, finite-population corrections, or estimator-specific calibration effects. It produces no confidence intervals.
Percentiles use R's type-7 linear interpolation. Concentration counts round up to whole respondents, with ties leaving that count unchanged. Distributions are displayed at mean weight 1. No rows are trimmed or deduplicated; negative weights are unsupported. Select a single final weight, not a frequency count or a set of replicate weights.
Sources reviewed September 29, 2026:
- PracTools: Kish unequal-weighting effect
- AAPOR: Disclosure standards
- NCHS: Reliability of NHANES estimates
- R: Sample quantiles (type 7)
For a separate planning question, the survey sample-size and precision planner explains its own sampling assumptions. This diagnostic does not automatically transfer a weight-only adjustment into that planner.