Audit Survey Strata and Primary Sampling Unit Identifiers

Recover identifiers from the release documentation

A primary sampling unit identifier describes sampling structure, not necessarily an individual participant or the institution appearing in a clinical record. Start with the survey release’s design notes. Record which variables represent strata, clusters, weights and any later sampling stages; distinguish public masked identifiers from confidential original identifiers.

The survey design documentation explains the role of cluster identifiers and the option for identifiers reused across strata. A nesting option can represent a documented coding scheme, but it should not be used automatically to silence a data-quality warning.

Build an identifier audit table

Before fitting a model, count records, unique clusters and missing design fields within each stratum. Check whether an identifier unexpectedly crosses strata. Trace repeated values to the codebook and merge history: they may reflect intended local numbering, a failed join or concatenation of incompatible survey cycles.

Create a small table showing each stratum’s cluster count and the number of records with zero, negative or missing weights. Keep the table in the analysis log rather than exporting participant-level identifiers into a public report. Confirm that design variables remain unchanged after recoding outcomes and constructing analytic covariates.

Inspect a hypothetical repeated code

Suppose stratum A contains clusters numbered 1 and 2, and stratum B also contains clusters numbered 1 and 2. There are four distinct stratum–cluster combinations, not two independent clusters across the whole survey. If the provider documents local numbering, a nested identifier or an appropriate nesting option represents that structure.

Now suppose a merge removes every record from cluster 2 in stratum B. The remaining analytic file has a singleton stratum. That is a different problem from reused labels. Identify whether it arose from domain restriction, missing covariates or genuine sampling design before choosing a variance treatment.

Document decisions that affect uncertainty

A certainty unit and an accidentally isolated cluster do not have the same interpretation. Follow documented survey guidance for singleton handling and record the selected software option explicitly. Where several defensible treatments exist, examine their effect on the main interval and explain the sensitivity.

Finish by checking the design summary against expected counts from the release. Report the variance method and relevant design limitations, without implying that a successful software command validates the sampling model. A reviewer should be able to distinguish provider-specified structure from analyst-created assumptions and to identify the point in the pipeline where any observations were excluded.

Source verification: 6 November 2026. Examples are hypothetical.