Survey Weighting and Calibration
Also known as: Survey Calibration, Post-Stratification Weighting, Raking Adjustment, Ağırlıklandırma (Anket)
Survey weighting is a statistical procedure that assigns a numeric weight to each sampled unit so that the weighted sample reproduces known population totals. Rooted in classical sampling theory and systematically synthesized by Sharon Lohr (2010), the approach corrects for unequal selection probabilities, unit nonresponse, and coverage gaps, producing estimates that are more representative of the target population than raw sample means or totals would be.
Key highlights
- Corrects simultaneously for complex sampling designs, nonresponse bias, and coverage undercoverage in a single coherent weight.
- Calibration to census or administrative totals typically reduces estimator variance relative to Horvitz-Thompson estimation.
- Weighted estimates are consistent with published population benchmarks, enhancing credibility for public-sector and policy reports.
- The framework is model-assisted rather than model-dependent, preserving design-based inference guarantees even when calibration models are misspecified.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Apply survey weighting whenever a probability sample departs from simple random sampling, when unit nonresponse is nonnegligible (typically >10%), or when reliable auxiliary population totals exist that the sample should reproduce. Key assumptions include known or estimable inclusion probabilities, ignorable nonresponse within adjustment cells, and accurate auxiliary benchmarks. The method is less suitable when auxiliary data are outdated, when the outcome variable is unrelated to the calibration variables (no variance gain), or when extreme weights inflate variance more than calibration reduces it.
Strengths & limitations
- Corrects simultaneously for complex sampling designs, nonresponse bias, and coverage undercoverage in a single coherent weight.
- Calibration to census or administrative totals typically reduces estimator variance relative to Horvitz-Thompson estimation.
- Weighted estimates are consistent with published population benchmarks, enhancing credibility for public-sector and policy reports.
- The framework is model-assisted rather than model-dependent, preserving design-based inference guarantees even when calibration models are misspecified.
- Extreme calibration weights can inflate variances and standard errors, sometimes exceeding the variance of the uncalibrated estimator.
- Quality depends entirely on the accuracy and currency of the auxiliary population totals used as benchmarks.
- Nonresponse adjustment assumes that respondents and nonrespondents within each adjustment cell are exchangeable — an untestable assumption.
- Weighting addresses mean-level representativeness but does not recover unobserved subgroup heterogeneity or correct measurement error.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
What is the difference between post-stratification and calibration?
Post-stratification is a special case of calibration where the auxiliary constraints are cell population counts forming a partition of the sample. Calibration is more general: it can incorporate marginal totals from multiple variables simultaneously (raking) or continuous auxiliary totals, and it allows any distance function to control weight departure from the base weights.
How do I handle very large or very small weights?
Large weights signal that a small number of respondents represent many nonrespondents or hard-to-reach units. Common remedies include weight trimming (capping at a percentile threshold), bounded calibration algorithms that constrain the weight ratio, or redefining adjustment cells to reduce within-cell heterogeneity. Each remedy trades some bias reduction for variance control and should be documented transparently.
Can survey weighting eliminate all nonresponse bias?
No. Weighting removes nonresponse bias only to the extent that the calibration or adjustment variables are correlated with both response propensity and the survey outcome. If nonrespondents differ from respondents on unmeasured characteristics even after conditioning on available auxiliaries, residual bias remains. Sensitivity analyses and follow-up studies of nonrespondents are recommended for high-stakes surveys.
Sources
- 1.Lohr, S. L. (2010). Sampling: Design and Analysis (2nd ed.). Brooks/Cole.ISBN 978-0-495-10527-5
You have read it. What now?
Cite this page
ScholarGate. (2026, June 2). Survey Weighting. ScholarGate. https://scholargate.app/survey-methodology/survey-weighting