Preserve the first-phase sample when selecting detailed follow-up measurements

Separate the measurement plan from the follow-up invitation

Identify the broad first-phase measurements and the additional information that is costly or burdensome to collect. State why the second phase is needed for the research decision. Specify which permitted first-phase variables can inform selection, and when their values are frozen. A later correction to one of those variables may affect the recorded selection rationale, so retain the version actually used.

The survey package’s two-phase documentation describes sampling a population and then taking a subsample from that sample, with the first-phase sample forming the second phase’s population.

Keep the intended research design separate from the participant-contact plan. Being observed in phase one does not by itself authorize a new interview, recording or data request. Confirm the approved invitation and consent route with the research owner before any real follow-up. Synthetic planning records are enough to test the selection and handoff mechanics.

Specify who can enter phase two and with what chance

Have the analyst document eligibility, phase-two selection variables, selection probabilities and any planned differences across groups. Distinguish purposeful qualitative follow-up from probability subsampling when the intended inference requires the latter. Selecting unusually articulate or enthusiastic respondents can be useful for a different research purpose, but it must not be relabeled as the approved two-phase draw.

Retain a membership ledger for every first-phase unit indicating whether it was eligible, selected and actually measured in phase two. Keep selection separate from later refusal or noncontact. Do not replace a declining selected unit with an available unselected one without a justified design decision. Record the reason for each absence without collecting unnecessary personal explanations or treating missing measurements as zero values.

Preserve the relationship between phases in the data package

Use a stable authorized unit reference to connect first-phase information with later measurements. Keep the full phase-one scope available to the analytical workflow in the approved form, rather than exporting only successful follow-ups. Include phase-specific design information and the actual selection-variable version. The final file must make it possible to distinguish a measurement not collected by design from one missing after selection.

Rehearse a synthetic handoff containing a selected-and-measured unit, an eligible-but-unselected unit and a selected unit without the additional measurement. Have the analyst confirm how the intended method uses each status. Check that ordinary joins or complete-case filters do not silently collapse the package into only fully observed records. The test concerns selection and measurement availability, not an instruction to fabricate the unobserved second-phase answers.

Accept the analysis scope and preserve the selection history

Ask the analyst to state which population quantity the two-phase analysis estimates and how both phases’ selection information enters the method. Obtain a reference output or design summary that demonstrates the receiving workflow represents the actual plan. A successful model fit on the follow-up subset alone is not sufficient acceptance. Keep unresolved selection departures and nonresponse limitations alongside the result.

If the project changes its second-phase criteria, version the change and decide whether it creates a new selection group or requires a different analysis. Do not rewrite earlier criteria to match the people who happened to participate. Report the phase-one, selected phase-two and measured phase-two counts separately. The practical outcome is an inspectable nested selection process, not a claim that additional detail from fewer people automatically represents everyone surveyed initially.

Make the reasoning concrete.

A fictional service-design study collects brief account-process descriptions, then plans a detailed, approved workflow measurement on a probability subsample. The team is tempted to choose only the most interesting initial narratives.

The team retains the planned selection process or explicitly changes the research purpose; it does not disguise a purposeful follow-up as the original design.

Take the questions into your next brief.

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

1. Measurements

   What additional information justifies the second phase?

2. Selection

   Which frozen first-phase variables and probabilities govern entry?

3. Membership

   Are eligibility, selection and measurement distinct statuses?

4. Linkage

   Does the package retain the full authorized phase relationship?

5. Scope

   What inference survives departures and phase-two nonresponse?

Watch for the wrong conclusion.

• Calling interesting-case follow-up a probability subsample.
• Dropping all first-phase units without later measurements.
• Treating initial participation as automatic recontact permission.

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.

• Thomas Lumley, survey documentation: Two-phase designs →Primary statistical software documentation · Supports: First-phase sample as the population for deliberate second-phase subsampling.