Keep every sampling stage behind the final interview

Draw the selection path before flattening the data

Write the stages in order, naming the population and sampling unit at each stage. A study might select business locations, then eligible customer accounts associated with selected locations. State whether the second stage includes every eligible account or selects another sample. Those procedures create different evidence, even when the final file has identical-looking account rows.

CDC describes NHANES as selecting participants through four sampling stages, retaining distinct primary and later-stage units. That named survey illustrates a selection hierarchy; its population and operational rules are not a template for this business study.

Create one diagram and one ledger that express the same path. For every final unit, the ledger should identify its parent selection and the frame from which it could be drawn. Preserve the actual hierarchy rather than reconstructing it from today’s organizational structure. A location merger after collection must not silently rewrite which primary units the original study selected.

Record the frame and selection rule at every stage

For each stage, retain the eligible-unit definition, frame version, selected references and the analyst-approved selection procedure. Ask the supplier to provide stage-specific inclusion information in the form the analyst needs, not an unexplained final weight alone. Distinguish an absent frame from an empty eligible population and a declined organization from one never selected. Those states explain different gaps in the design.

Keep permission and data minimization attached to the operational handoff. An organizational gatekeeper may provide an approved eligibility count or facilitate a sampling process without giving the research team unrestricted access to employee records. Do not ask the gatekeeper to nominate the most enthusiastic participants in place of the specified selection. If the approved route cannot implement the design, document the limitation and obtain a revised research decision.

Test identities and substitutions through a synthetic ledger

Build a small fictional example with repeated local account numbers in different parent organizations. Verify that the combined identities remain unambiguous and that the receiving analyst can distinguish the parents. A local identifier such as account seven may be unique within one location but ambiguous across the study. Avoid resolving that ambiguity by treating records as duplicates and deleting one without evidence.

Include a selected parent that declines, an eligible final unit that cannot be contacted and a parent with an unexpected eligible count. Have the design owner specify how each is recorded and whether any further selection is permitted. Do not let fieldwork staff replace a difficult parent with a nearby cooperative one or increase interviews elsewhere simply to recover the headline count. Preserve deviations even when the final number of interviews matches the original target.

Accept the analysis handoff as a design package

Reconcile final records to their actual parents and selection stages. Check that no interview appears without a supported path and that parent-level dispositions survive the export. Give the analyst the frame definitions, stage identifiers, selection information and exception record together. Ask for a small acceptance readout demonstrating that the intended design is represented, without treating a software object’s successful creation as proof that the supplied design facts are correct.

Report the count of selected primary units separately from the number of final interviews. A large interview total concentrated in a few organizations is not the same evidence as the same total spread across many independently selected organizations. Keep this limitation connected to the study’s reporting scope. The completed ledger makes the actual collection path inspectable; it does not guarantee population coverage, truthful answers or causal interpretation.

Make the reasoning concrete.

A fictional customer-support study selects service centers and then samples customer accounts within each participating center. Its supplier delivers a spreadsheet containing only completed account interviews.

The team chooses the design-package handoff. It does not invent missing parent probabilities from the number of interviews delivered.

Take the questions into your next brief.

Use these prompts in your own notes. Nothing is submitted from this page.

1. Stages

   Which units are selected at each stage?

2. Frames

   Which versioned eligible list supports each selection?

3. Identity

   How does each final unit retain an unambiguous parent?

4. Exceptions

   How are refusals and unauthorized substitutions distinguished?

5. Acceptance

   What evidence shows the analyst received the actual design?

Watch for the wrong conclusion.

• Flattening the dataset before preserving its sampling hierarchy.
• Letting a gatekeeper choose convenient replacements.
• Inventing earlier-stage selection probabilities from achieved interview counts.

Sources & scope

Workflows, worksheets and fictional examples are original editorial analysis. Sources support only specifically attributed principles. Platform guidance can change; legal requirements depend on jurisdiction and context. No measured results or named human reviewer are claimed.

• CDC NHANES Tutorials: Sample Design →Primary methodology documentation · Supports: NHANES-specific example of multistage selection, not a universal business-survey prescription.