Audit What a Nonresponse Weight Adjustment Can Address

Identify the missing participation stage

Nonresponse is not a single event. Someone may decline the entire interview, complete an interview but miss an examination, or attend the examination while leaving one measurement unavailable. Write a participation map that distinguishes these stages and identifies which response process the supplied weight addresses.

The NHANES weighting tutorial separates selection weights, nonresponse adjustments and population calibration. It also notes that interview or examination nonresponse adjustments do not automatically account for missing individual items. Check the relevant component documentation rather than assuming that one final weight resolves every missing-data problem.

Audit an adjustment cell

For a simple cell-based adjustment, identify the population of sampled eligible units and the respondents within each prespecified cell. Confirm whether the adjustment uses weighted totals or unweighted counts. Investigate cells with no respondents, very low response or implausibly large adjustment factors before carrying them into outcome estimation.

Choose cell variables or response-model predictors using information available for respondents and nonrespondents. Record the substantive argument for why adjustment within those groups could reduce response-related bias. A response model cannot recover information about an entirely unmeasured selection mechanism merely because it predicts participation well.

Work through a hypothetical response pattern

Suppose 100 eligible sampled people each have base weight 10, but only 80 respond. Within this deliberately simplified equal-weight cell, the adjustment factor is 100/80 = 1.25, and each respondent receives weight 12.5. The respondents’ adjusted weights sum to 1,000, matching the original base-weight total.

Now suppose 10 of those 80 respondents lack a laboratory outcome. The original unit-response adjustment does not explain how to analyze those missing laboratory values. Dropping them without further thought can change the represented population and introduce another selection problem, even though the remaining records still carry valid interview weights.

Preserve the limits in the report

State which stage was adjusted, which predictors or cells were used and what assumptions connect respondents to nonrespondents. Review weighted auxiliary distributions against credible population information, recognizing that agreement on those variables does not establish agreement on every outcome.

Keep a record of collapsed cells, extreme adjustments and sensitivity analyses. Use the provider’s component-specific weights when required, and avoid applying a second adjustment that duplicates an existing one. Report item-missingness handling separately so readers can see exactly which source of missing information each procedure addresses.

Source verification: 6 November 2026. Examples are hypothetical.