03 · What You Need to Know
How Sampling-Frame Errors Affect Probability Sampling
What Is a Sampling Frame, and Why Must It Be Accurate?
A sampling frame is the operational list or other selection mechanism used to identify units eligible for sampling. Depending on the study, those units may be students, teachers, households, schools, organizations, or records.
For simple random sampling, a common frame is a list containing one entry for each eligible population unit.
For stratified sampling, the frame must also support accurate assignment of units to their respective strata. In cluster and multistage designs, separate frames may be needed at different selection stages.
The quality of the frame affects whether the intended population is adequately covered and whether the selection probabilities can be calculated correctly.
A technically flawless random-number generator cannot compensate for a frame that systematically omits eligible population members.
What Types of Sampling-Frame Errors Should You Look For?
Sampling-frame problems are not limited to duplicate names. Several distinct errors can affect selection and inference.
| Frame problem |
Example |
Potential consequence |
| Duplicate records |
A teacher appears twice because of two institutional assignments. |
The teacher may have more opportunities for selection than intended. |
| Outdated eligibility |
A former employee remains on the active staff roster. |
An ineligible unit may be selected. |
| Undercoverage |
Recently enrolled students are missing from the registry. |
Some eligible units have no selection opportunity. |
| Incorrect classification |
A school is assigned to the wrong sector or region. |
Stratum assignment or allocation may be incorrect. |
| Outdated contact information |
An eligible participant has an inactive email address. |
Contact failure may increase nonresponse. |
| Incorrect unit definition |
Multiple course enrollments are treated as separate students. |
The frame may sample enrollments rather than individuals. |
These errors require different responses. A stale email address does not necessarily make a person ineligible, while a duplicate record may require correction even when every associated field is current.
How Do Duplicate Records Affect Selection Probabilities?
Duplicate records can alter the relationship between the probability of selecting a frame entry and the probability of selecting a distinct population unit.
Suppose a frame contains 1,000 records, but one eligible person appears twice. If the researcher draws 100 records through simple random sampling without replacement, each record has an inclusion probability of 100 / 1,000 = 10%.
However, the duplicated person has two opportunities for inclusion. The probability that at least one of those records is selected is greater than 10%.
This calculation illustrates why duplicate records should not be treated as harmless clerical errors.
More generally, when a person appears in several frame entries, their inclusion probability depends on the number of entries and the selection design. In some specialized multiple-frame or multiplicity sampling designs, such probabilities can be calculated deliberately. Accidental duplication, however, should not be mistaken for a properly designed unequal-probability sample.
How Can You Identify Duplicate Records Reliably?
Begin with stable, unique identifiers when available. Institutional student numbers, employee identifiers, facility codes, or other verified identifiers are generally more reliable than names alone.
Exact matching can identify records sharing the same unique identifier. However, different identifiers may sometimes refer to the same underlying unit, particularly when data come from several administrative systems.
Additional checks may use combinations of relevant fields, such as institutional assignment, date of enrollment, or other authorized identifying information.
Probabilistic record linkage or fuzzy matching may help identify potential duplicates when exact identifiers are unavailable, but these procedures can produce false matches and missed matches.
Potential duplicates should therefore be reviewed according to a documented decision rule, especially when incorrectly merging two distinct people would remove an eligible unit from the frame.
Watch Out
Do not delete records solely because two people share a name. Names are not necessarily unique identifiers, and incorrect deduplication can create undercoverage by removing legitimate population members.
Should Duplicate Records Always Be Merged?
Not until the sampling unit has been established.
Suppose a study aims to estimate the experiences of individual teachers. A teacher appearing twice because of employment at two schools may need to be represented once, with appropriate handling of multiple institutional memberships.
However, if the study's sampling units are teacher-school assignments, the two records may represent distinct eligible units rather than accidental duplicates.
Similarly, a student enrolled in three courses should appear once in a student-level sampling frame but may legitimately appear three times in a course-enrollment frame.
The appropriate correction depends on what the population unit represents. Deduplication should follow the research definition, not simply the appearance of repeated identifiers.
How Should Outdated Records Be Handled?
Outdated records should be evaluated against a clearly defined eligibility reference date or period.
For example, if the target population consists of teachers employed on a specified census date, the frame should identify which teachers met that criterion on that date.
A teacher who transferred after the reference date may still belong to the defined population, depending on the study's eligibility rules. Conversely, a former teacher who left before the reference date may be ineligible.
Researchers should distinguish outdated eligibility information from outdated contact information.
An eligible teacher with an inactive institutional email address remains eligible. The appropriate response is to seek an authorized, current contact channel rather than remove the teacher simply because the original email no longer works.
Where eligibility cannot be verified, the record should be flagged for review rather than automatically classified as ineligible.
How Do You Conduct a Sampling-Frame Audit?
A systematic frame audit should occur before the random draw whenever feasible.
Step 1: Define the sampling unit and reference population.
Specify exactly what constitutes one eligible unit and the relevant eligibility date or period.
Step 2: Preserve the original frame.
Retain an unmodified, access-controlled version so that every correction can be traced.
Step 3: Check identifiers and duplication.
Identify exact duplicate identifiers and investigate possible duplicate units represented by different records.
Step 4: Verify eligibility and classification.
Check active status, population membership, institutional assignment, and stratum variables against authoritative records.
Step 5: Assess population coverage.
Compare frame counts with reliable external totals and investigate missing or unexpectedly overrepresented groups.
Step 6: Resolve and document corrections.
Record which entries were merged, removed, updated, or retained, together with the reasons and evidence.
Step 7: Freeze the selection frame.
Assign stable identifiers, record the final frame size and version, and conduct the planned probability selection.
For large or complex frames, automated checks can support the audit, but decisions that materially affect eligibility or unit identity may still require manual verification.
Frame cleaning should be reproducible. Another researcher should be able to understand how the original administrative records became the final selection frame.
What If Frame Errors Are Discovered After the Sample Has Been Selected?
Postselection errors require particular caution because changing the selected records can alter the implemented sampling design.
If a selected record is confirmed ineligible under the original population definition, document the ineligibility and assess its implications for the sampling frame, eligible population size, and estimation.
If two selected records refer to the same eligible person, do not simply discard one and assume the remaining observations constitute an ordinary simple random sample. The person's original inclusion probability may differ from that of other units.
If an eligible selected participant has outdated contact information, attempt to update the contact information through authorized channels. Contact difficulty does not itself justify deleting the unit.
When errors are extensive, it may be preferable to correct the complete frame and draw a new sample before recruitment, provided this is feasible and the decision is documented.
Once recruitment or data collection has begun, corrections may require specialized estimation adjustments or a carefully designed supplemental probability sample.
Unplanned substitution can compromise the design. The methodological concerns associated with replacing selected nonrespondents also apply when researchers attempt to repair frame problems through informal replacement.
Can Sampling Weights Correct Duplicate or Outdated Records?
Sometimes, but only when the selection mechanism and relevant inclusion probabilities can be reconstructed.
If duplicate records create known unequal inclusion probabilities, appropriate design weights may account for those probabilities in certain estimators.
However, weighting cannot automatically recover eligible units that had no opportunity for selection because they were omitted from the frame.
Nor can ordinary inverse-probability weights correct inaccurate measurements or unknown eligibility errors without additional information and assumptions.
Researchers should distinguish errors that can be addressed through known-probability estimation from those that require improved coverage, additional data, or explicit limitations.
Where unequal inclusion probabilities are present, design weights must reflect the actual selection mechanism rather than the mechanism originally intended.
How Do Frame Errors Affect Different Probability Sampling Methods?
| Sampling design |
Consequential frame error |
Possible implication |
| Simple random sampling |
Duplicate or omitted individual records |
Incorrect or zero inclusion probabilities for affected population units |
| Systematic sampling |
Duplicate records or changes in list ordering |
Altered selection positions and possible interactions with periodic patterns |
| Stratified sampling |
Incorrect stratum membership |
Incorrect allocation, classification, or weighting |
| Cluster sampling |
Missing or duplicated clusters |
Distorted cluster selection probabilities and population coverage |
| Multistage sampling |
Errors in higher- or lower-stage frames |
Incorrect overall inclusion probabilities or missing ultimate units |
The underlying principle is consistent across designs: the frame must correspond to the intended population and the probability selection mechanism.
How Should Frame Corrections Be Documented?
Researchers should retain an audit trail describing the original frame, validation procedures, corrections, and final selection version.
Useful documentation includes the frame source, extraction date, eligibility reference period, original record count, duplicate-resolution rules, number of confirmed ineligible records, and final number of eligible units.
Where external population totals were used to assess coverage, identify those sources and explain any unresolved discrepancies.
For privacy-sensitive records, documentation should preserve the necessary methodological information without disclosing personal identifiers unnecessarily.
A reproducible audit trail supports research integrity and allows readers to assess whether the sample-selection procedure was implemented as claimed.
04 · A Practical Example
Cleaning a University Student Sampling Frame Before Selection
Hypothetical Example
Preparing an Enrollment Registry for a Student Survey
A researcher plans to survey undergraduate students about their experiences using an institutional learning management system.
The registrar provides a file containing 5,200 records. The intended population is all students actively enrolled in undergraduate programs on a specified reference date.
The researcher needs a probability sample of 500 distinct eligible students.
An audit of the original frame identifies several issues:
| Audit finding |
Records affected |
Action |
| Repeated student identifiers |
120 additional records |
Verify identity and retain one eligible record per student |
| Students not enrolled on the reference date |
80 records |
Remove after eligibility verification |
| Eligible students with outdated email addresses |
60 records |
Update contact information without removing students |
Assume the 120 duplicate entries and 80 ineligible records are distinct, nonoverlapping problems. The 60 outdated email addresses belong to otherwise eligible students and therefore do not reduce the frame count.
After correction, the final frame contains 5,200 − 120 − 80 = 5,000 distinct eligible students.
The researcher verifies the final count against the official enrollment total and then draws 500 distinct students through simple random sampling without replacement.
The researcher retains the original registry securely, documents all corrections, and preserves the final frame version and random selection record.
If the 120 duplicate entries had been left in the frame, affected students could have received higher selection probabilities. If the 60 students with outdated email addresses had been removed, eligible students would have been excluded merely because they were difficult to contact.
The example illustrates why frame validation must distinguish unit identity, eligibility, and contactability rather than treating every problematic record as a reason for deletion.
