# Define who a feedback sample can actually describe

Original constructed teaching cases with complete context. No real financial records, sponsorships, contracts, people, interviews, tests, preferences or payments.

Name the target population, eligible unit, accessible sampling frame and selection process before making a population claim. Inspect duplicate/ineligible records, omitted groups and nonresponse separately. A random subset of a restricted list does not restore the people missing from that list, and a balanced quota or many opt-in replies does not by itself create probability sampling.

## Define the population and eligible unit

A question about recent users who attempted an export needs a different population from a question about everyone who follows a public Page. Specify relevant experience, reference period, language and access needs without collecting unnecessary identifiers. A post impression, account, person, session and response record are not interchangeable units.

Write the intended statement before choosing an available list. If you can only describe voluntary respondents, say that directly. Do not begin with a convenient contact export and rename it all LinkedIn creators. This guide creates no list, sends no invitation and does not authorize private-data collection.

## Inspect the frame before drawing a sample

AAPOR’s sample-design guidance distinguishes the population from the frame used to reach potential respondents, and probability from nonprobability designs. Document the actual source, date, access basis, coverage and unit mapping. The same person appearing twice is not automatically two eligible sampling units; similar names are not proof of duplication either.

List target groups not covered by the frame and groups that cannot use the chosen language, device or participation route. Adding a random selection step inside the available frame does not give omitted groups a nonzero selection chance. A frame-quality review is a design decision before questionnaire wording, not a license to acquire missing identities.

## Trace the complete constructed selection chain

The teaching scenario assumes a target population of 100 eligible units and an available frame covering 80 distinct units. Twenty units are selected uniformly without replacement from that frame. Each framed unit has a 20/80 selection probability under that declared design; the 20 omitted target units have zero chance through this frame. These are invented design inputs, not an actual sampled audience.

Suppose only eight selected units later supply usable outcomes in the constructed example. Keep target 100, frame 80, selected 20 and usable eight as separate counts. Eight divided by 20 is a declared usable-return fraction of 40%, not automatically an AAPOR response rate, a representative sample or the probability that a target person participated.

| Constructed stage | Units | Scope |
| --- | --- | --- |
| Target | 100 | Declared eligible population |
| Frame | 80 | 20 target units omitted |
| Selected | 20 | Uniform draw from the frame only |
| Usable outcomes | 8 | Not a formal response-rate claim |

## Resolve three frame and selection decisions

F1 — Constructed duplicated list: a frame has 82 rows but two are confirmed repeat records of existing units. Decision: retain the original rows and documented unit resolution; use 80 distinct framed units if that identity basis is valid. Outcome: no assumption that all 82 rows represent different people or that similar-looking names can be silently merged.

F2 — Constructed opt-in post: an open post obtains 300 voluntary answers and no known selection probabilities. Decision: describe the invitation context and respondents, and hold a probability-sample margin of error. Outcome: the large answer count does not repair self-selection or prove preferences of silent followers.

F3 — Constructed quota: a plan fills equal counts in two experience groups through convenient volunteers. Decision: record the quota and recruitment mechanism separately. Outcome: a balanced selected set, not random selection or automatic coverage of all eligible people; uncertainty needs an appropriate design/model beyond that balance.

## Keep selection, contact and response dispositions

For permitted real work, retain eligibility, selection, noncontact, refusal, partial and completed outcomes using the appropriate actual definitions. Do not convert an unknown eligibility state into a known refusal or a missing answer into disagreement. Report mode, dates, language, exclusions and any weighting with the source and design.

AAPOR’s disclosure standards ask for coverage, selection and precision/model details. A randomized ordering of two reading tasks addresses an order issue inside that task design; it does not by itself make the participants a probability sample of all readers. Keep comparative task completion and explicit preference distinct from sample representativeness.

## Save the design before interpreting outcomes

The worksheet keeps target/unit/period, actual frame/access basis, exclusions and overlap decisions, selection probabilities if known, mode/language, dispositions and allowed population claim. It contains no real contacts or recruitment script. Use only genuinely authorized, voluntary participation for a real study.

AAPOR’s relevant sample-design and disclosure sections were read on 1 November 2026. All counts and record scenarios are constructed, with no draw from real people, message or response collection. For actual research, preserve the registered instrument and study-specific protocol instead of changing them after seeing answers.

## Primary sources

- [AAPOR: Designing a sample](https://aapor.org/standards-and-ethics/best-practices/)
- [AAPOR: Disclosure standards and precision scope](https://aapor.org/standards-and-ethics/disclosure-standards/)