Survey sampling frame checklist: coverage, duplicates and eligibility

A survey sampling frame is the operational list or procedure from which sample units are selected. Before drawing a sample, compare the frame population with the target population, quantify known undercoverage and overcoverage, resolve duplicates without collapsing distinct units, freeze eligibility and update rules, and preserve a versioned audit trail that can reproduce every unit's chance of selection.

What is a survey sampling frame?

A sampling frame identifies the units available for selection and supplies information needed to locate, stratify or cluster them. It may be a register, customer database, address list, school roster, geographic area procedure or a combination of sources. The frame unit must map to the analysis unit: a household frame used for person estimates needs an explicit within-household selection step.

Statistics Canada's current sixth-edition Quality Guidelines treat frame and sample design as a distinct survey-design stage: define the population, develop the frame, assess whether it covers the target population, keep it current and implement procedures for detecting and correcting coverage errors. The frame audit below turns that guidance into a pre-selection release gate.

Coverage error categories

Fifteen-point frame audit

1. Write the target population in one testable sentence.
2. Name the sampling unit, observation unit and analysis unit.
3. Record every source system, owner, extraction date and version.
4. Define inclusion, exclusion and unknown-eligibility rules.
5. Compare frame counts with compatible external totals.
6. Profile missing identifiers, contact fields and stratifiers.
7. Check impossible dates, stale statuses and invalid locations.
8. Search exact duplicates using stable identifiers.
9. Search probable duplicates using documented linkage fields.
10. Review one-to-many relationships such as people with multiple accounts.
11. Measure records added, updated and removed between frame versions.
12. Test whether key subgroups are absent or unusually sparse.
13. Freeze the frame before random selection and hash the extract.
14. Store the randomization method, seed and inclusion probabilities.
15. Reconcile selected records, fieldwork dispositions and the final analysis file.

Worked duplicate and eligibility example

A customer frame contains 10,200 rows but only 10,000 unique verified customer IDs. A total of 180 customers appear twice and 10 appear three times, which accounts for the 200 extra rows. If rows are sampled equally without correction, duplicated customers have two or three selection records and therefore a higher chance of selection.

The audit does not delete every match blindly. It reviews conflicting records, shared contact details and business rules, then creates one eligible unit per verified customer or records multiplicity in the inclusion probability. Suppose 140 unique IDs are confirmed former customers and 60 have unresolved eligibility. The frozen eligible frame contains 9,800 units, with the 60 unresolved cases held outside selection under a predeclared rule. If a compatible external total is 10,250, the 450-unit difference is a coverage warning—not proof that all 450 are missing eligible customers, because definitions and dates may differ.

Frame specification template

Store target_population, reference_period, sampling_unit, analysis_unit, source_systems, extract_timestamp, frame_version, inclusion_rule, exclusion_rule, unknown_eligibility_rule, stable_identifier, deduplication_rule, coverage_benchmarks, stratification_fields, selection_method, random_seed, inclusion_probability, frame_owner, quality_owner and known_limitations.

Coverage reconciliation table

Boundary and reproducibility tests

1. Test records exactly on every date, age, tenure or geographic boundary.
2. Test one unit present in two source systems.
3. Test two distinct people sharing a contact field.
4. Test a record with missing eligibility.
5. Test a unit updated after the frame freeze.
6. Test a restored historical record.
7. Rerun linkage on the same frozen inputs and configuration.
8. Rerun selection with the same seed and confirm unit IDs and inclusion probabilities match.

How does the frame connect to later survey steps?

Define and audit the frame before using the sample-size worksheet. Carry inclusion probabilities into the weight specification, record sampled-case outcomes in the response-rate worksheet, and apply post-collection identity checks in the data-cleaning checklist. These stages address different errors; none substitutes for a documented frame.

Transparent method

The current Statistics Canada guidance says to define the target population, assess whether the selected frame covers it, keep the frame current, use supplementary sources when needed, detect and correct potential coverage errors, and save sampling and collection information. Its archived 2009 coverage-and-frames guidance names missing in-scope units, included out-of-scope units, misclassification and duplicates as frame problems and recommends monitoring against alternative sources and survey returns. The fifteen checks, specification, reconciliation example and reproducibility tests are editorial synthesis, not an official certification standard.

Limitations

External totals may be old or define the population differently. Probabilistic linkage can create false matches and missed matches. A web intercept or open-link survey may not have a conventional person-level frame. Multi-frame, area and indirect designs require specialized estimators. Passing this checklist does not prove complete coverage, unbiased response or valid inference beyond the documented frame and design.