Sampling Frames and Recruitment Sources

Overview

A sampling frame is the operational representation from which sampling units are selected. A recruitment source is the place, system or pathway through which potentially eligible participants are identified or approached. The two may be the same, but they often are not.

For example, a hospital delivery register can be a sampling frame for selecting recent deliveries, while a postpartum clinic can be one recruitment source. Conversely, a social-media advertisement is a recruitment source but rarely provides a complete frame because the number and characteristics of all eligible people exposed to the advertisement are unknown.

This lecture explains how to construct and appraise frames and recruitment sources, including coverage, duplication, outdated records, multistage sampling, privacy and documentation.

Why this matters

A probability-sampling procedure is only as defensible as the frame to which it is applied. Drawing random numbers from an incomplete list does not make the final sample representative of the target population. The procedure may be random within the list while systematically excluding people absent from it.

Frame problems can cause:

• undercoverage, when eligible units are missing;
• overcoverage, when ineligible or out-of-scope units are included;
• duplication, when one unit appears more than once;
• incorrect linkage, when records do not correspond to the intended person or unit;
• outdated information, leading to failed contact or misclassification;
• unequal opportunity for selection; and
• privacy or governance breaches during pre-screening.

Key concepts

Sampling frame

A sampling frame should identify every in-scope sampling unit once, as far as practicable, and include sufficient information to apply the sampling design and locate selected units. Depending on the study, a frame may be:

• a list of people;
• a register of households;
• a map of enumeration areas;
• a list of facilities or clusters;
• a time-location schedule;
• a set of telephone numbers;
• linked administrative records; or
• a sequence of frames used at different sampling stages.

A frame is not necessarily a spreadsheet. It is the complete operational system that connects the source population to selectable units.

Sampling unit and element

A sampling unit is selected at a particular stage. The element is the ultimate unit about which information is sought. In a multistage household survey, districts, enumeration areas and households may be sampling units, while adults are elements.

Recruitment source

A recruitment source identifies or reaches potential participants. Examples include:

• clinic lists;
• clinician referrals;
• community organisations;
• advertisements;
• research registries;
• electronic health-record searches;
• birth or disease registries; and
• respondent referrals.

A recruitment source may support consecutive or purposive recruitment without constituting a probability frame.

Defining the frame population

The team should first specify:

1. the target population;
2. the source population;
3. the sampling unit at each stage;
4. the reference period;
5. the minimum variables needed for eligibility and selection;
6. the method for identifying duplicates;
7. the update schedule; and
8. the difference between frame inclusion and final eligibility.

The frame population comprises the units represented on the final frame. Researchers should compare this population with the source population. Discrepancies should be prevented where possible and described where unavoidable.

DECISION GUIDE

Appraise a proposed frame

Ask whether the frame:

• covers all important subgroups and settings;
• contains out-of-scope or ineligible units;
• lists any unit more than once;
• uses a stable and unique identifier;
• reflects the correct calendar period;
• permits calculation of selection probabilities;
• supports contact without unnecessary disclosure;
• can be updated and audited; and
• can be linked to strata, clusters or eligibility information without changing the intended population.

Common frame architectures

Complete list frame

A complete list frame enumerates all known units. It is suitable for simple random, systematic or stratified sampling when the list is sufficiently complete and current.

Area or cluster frame

When no person-level list exists, researchers may select geographical or organisational clusters first and construct household or person lists within selected clusters. This approach reduces field cost but creates clustering and requires retention of selection probabilities at every stage.

Multiple or composite frames

A composite frame combines sources. The team must define:

• how records are matched;
• which source takes precedence;
• how duplicates are removed;
• whether the sources cover mutually exclusive domains; and
• how inclusion probabilities are calculated when a unit can enter through more than one source.

An overlapping dual-frame design requires special estimation. It should not be approximated by merging lists and ignoring duplicate opportunities.

Time-location frame

For populations congregating at identifiable venues and times, researchers may sample venue–time units and then eligible people encountered within them. The frame is a calendar of place–time combinations, not a complete person list. Coverage excludes people who never attend included venues.

Event or episode frame

Some clinical databases list encounters rather than people. If the person is the analysis unit, repeated episodes must be linked. Selecting encounters without accounting for multiple opportunities can overrepresent frequent users.

Frame-construction log

illustrative

Table 1. Construction of a composite maternity frame

Step	Records remaining	Rule
Maternity-register extraction	1,540	Deliveries during 2025
Link gestational-diabetes flag	1,286	Diagnostic record or verified antenatal result
Remove duplicate episode records	1,248	One delivery episode per woman
Exclude out-of-period deliveries	1,236	Delivery date within study period
Verify hospital and residency fields	1,210	Source-population definition met
Final frame	1,210	Eligible for sampling, pending detailed screening

Note: Counts are illustrative. A frame-construction log should preserve the rules and date of each operation.

Coverage error

Coverage error is not measured only by the percentage of missing records. Its importance depends on how omitted or duplicated units differ in variables relevant to the study outcome.

For the postpartum study:

• women diagnosed outside specialist clinics may have different follow-up;
• women without linked identifiers may be more mobile or socioeconomically disadvantaged;
• private-hospital patients may have different access to testing;
• frequent service users may appear in several lists; and
• women who already attended postpartum care must not be the only source for an attendance outcome.

Researchers can assess coverage by comparing counts and characteristics with independent sources, auditing samples of unmatched records, consulting service staff, and piloting contact. These checks do not prove completeness but reveal plausible mechanisms.

Frame maintenance and field use

Frames change. Births, deaths, migration, diagnoses, staff changes and facility closures can make a list obsolete. The protocol should specify a freeze date or update schedule. In rolling recruitment, monthly or weekly frame batches may be needed.

Selected units should retain:

• frame identifier;
• selection stratum and cluster;
• order or random number;
• initial eligibility status;
• approach attempts;
• final recruitment outcome; and
• reason for non-enrolment.

Researchers should not replace a selected non-responder with the next convenient person unless the sampling design explicitly prespecifies a replacement rule. Such replacement can distort selection probabilities and hide non-response.

PROTOCOL CHECKLIST

Frame and recruitment-source information to document

• source population and reference period;
• all data sources used to construct the frame;
• sampling unit at each stage;
• linkage and deduplication rules;
• exclusions applied during frame construction;
• date of extraction and update procedures;
• known coverage limitations;
• selection variables, strata and clusters;
• contact information available and governance basis;
• handling of selected but ineligible or unreachable units; and
• separation between frame construction, eligibility screening and consent.

Ethical and governance considerations

Interface with statistical analysis

The frame and design determine whether inclusion probabilities are known. These probabilities may be required to construct survey weights. Strata, clusters and finite-population corrections may also affect variance estimation.

Researchers must retain design variables from selection through analysis. A dataset containing only participant responses and no stratum, cluster or selection-probability information may make correct analysis impossible.

For non-probability recruitment sources, model-based adjustment or calibration may improve alignment with benchmarks, but it does not convert the sample into a probability sample without additional assumptions.

See the Biostatistics lectures on complex survey analysis and clustered data.

Frame multiplicity and unequal selection probabilitiesMMed extension

A person can have multiple opportunities for selection when the frame lists episodes, telephone numbers, facilities or overlapping registries. The selection probability then depends on multiplicity. One solution is to define a unique person-level frame before sampling. Another is to retain multiplicity information and incorporate it into weights. Ignoring multiplicity can overrepresent frequent users or people registered in several systems.

Worked example

WORKED EXAMPLE

Building a defensible postpartum sampling frame

Question: What proportion of women with gestational diabetes complete postpartum glucose testing within 12 weeks after delivery?

Inadequate proposal: recruit women from the postpartum clinic.

Revised frame:

1. Extract all 2025 delivery episodes from the three maternity registers.
2. Link antenatal diabetes diagnoses and qualifying glucose results.
3. Deduplicate women while preserving separate pregnancies where applicable.
4. verify the hospital, delivery date and residency definition.
5. assign a unique frame identifier.
6. stratify by hospital and quarter of delivery.
7. select women within strata.
8. screen detailed eligibility and document outcomes.
9. ascertain postpartum testing through linked laboratories and participant contact.

Benefit: Women can be included regardless of whether they attended postpartum care, so the outcome pathway does not define entry into the sample.

Remaining limitation: Deliveries outside the three hospitals remain uncovered and must be acknowledged.

Suggested protocol or reporting wording

Evidence basis and critical appraisal

General research-methodology guidance places the frame between the source population and selected sample [1]. The WHO STEPwise approach shows how survey frames, sampling stages and field procedures must be planned together rather than treated as separate tasks [2]. Large household-survey manuals likewise emphasise listing quality, unique identifiers and retention of stage-specific probabilities [3].

Survey-methodology texts distinguish coverage error from random sampling error: a perfectly executed random selection can still be biased when the frame omits systematically different people [4]. Reporting standards require investigators to explain participant selection and the setting from which the sample arose [5].

A critical appraisal of a frame should examine not only completeness but differential completeness. Missing 5% of a frame may be serious if the omitted group carries most of the outcome risk. Reviewers should request the frame date, construction sources, duplicate rules, linkage success and independent coverage checks. They should also determine whether the frame lists persons, episodes, facilities or time–location units, because the analytical implications differ.

When several sources are combined, the report should show whether they are mutually exclusive, how overlaps were resolved and whether a unit had more than one path to selection. Without this information, inclusion probabilities and the direction of coverage bias may be impossible to judge.

Common mistakes

Knowledge check

Is a laboratory database automatically a person-level sampling frame?

Answer and reasoning

No. It may list tests rather than unique people, contain repeated observations, omit people tested elsewhere and lack information needed to define the source population. It can contribute to a frame after linkage and appraisal.

What is undercoverage?

Answer and reasoning

Undercoverage occurs when members of the source population are absent from the frame and therefore have no chance of selection through that frame.

Why must cluster and stratum identifiers be retained?

Answer and reasoning

They describe the probability design. They are needed to estimate correct standard errors and, together with selection probabilities, may be required for weighted population estimates.

Can a social-media advertisement be a recruitment source?

Answer and reasoning

Yes. It can identify volunteers, but the exposed eligible population and individual selection probabilities are usually unknown. It should therefore not be reported as a complete probability frame.

Key take-home messages

SUMMARY

Key take-home messages

• A sampling frame operationally represents the source population; a recruitment source is a pathway for identifying or approaching people.
• Assess coverage, duplication, outdated records and multiplicity before selecting units.
• Do not use a downstream outcome pathway as the only frame for estimating that outcome.
• Preserve all design variables from frame construction through analysis.
• Document linkage, deduplication, updates and non-enrolment.
• Protect privacy during pre-consent identification and contact.

References

1. World Health Organization Regional Office for the Western Pacific. Health research methodology: a guide for training in research methods. 2nd ed. Manila: WHO Regional Office for the Western Pacific; 2001. https://iris.who.int/items/0765fa3d-af1d-43b3-bb36-6576e88a7871
2. World Health Organization. WHO STEPS surveillance manual: the WHO STEPwise approach to noncommunicable disease risk factor surveillance. Geneva: World Health Organization. https://www.who.int/teams/noncommunicable-diseases/surveillance/systems-tools/steps/manuals
3. ICF. Demographic and Health Survey sampling and household listing manual. Rockville, MD: ICF; 2012. https://dhsprogram.com/publications/publication-dhsm4-dhs-questionnaires-and-manuals.cfm
4. Groves RM, Fowler FJ Jr, Couper MP, Lepkowski JM, Singer E, Tourangeau R. Survey methodology. 2nd ed. Hoboken, NJ: Wiley; 2009.
5. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP; STROBE Initiative. The Strengthening the Reporting of Observational Studies in Epidemiology statement: guidelines for reporting observational studies. Lancet. 2007;370(9596):1453–1457. https://doi.org/10.1016/S0140-6736(07)61602-X