03 · What You Need to Know
How to Plan Source Data Verification for Reliable Research Records
What Is Source Data Verification?
Source data verification (SDV) is the process of comparing information recorded in a research database or data collection record with the relevant original source information.
Its purpose is to determine whether selected data have been transcribed, transferred, or represented accurately.
For example, a researcher may compare an achievement score entered into a central database with the original scored assessment or an authorized electronic source record.
Source verification is widely discussed in clinical research quality management, but the underlying procedure can also be useful in educational, behavioral, social, and other research involving transferred or transcribed data.
The applicable regulatory requirements differ across contexts. Clinical research guidance should not automatically be treated as a binding requirement for every nonclinical study.
What Counts as a Source Record?
A source record is the original or otherwise authoritative record from which research information is obtained or against which it can be verified.
Its form depends on the research design and data collection system.
| Research information |
Possible source record |
Database information to verify |
| Achievement assessment |
Original scored assessment or authorized assessment system record. |
Participant identifier, score, and assessment date. |
| Survey response |
Original questionnaire or electronic response record. |
Selected responses and coding. |
| Attendance |
Authorized attendance record. |
Attendance counts or session participation. |
| Participant eligibility |
Approved screening or eligibility documentation. |
Eligibility classification. |
| Intervention assignment |
Authorized allocation record. |
Assigned study condition. |
| Laboratory measurement |
Instrument output or original laboratory record. |
Measurement value, units, and specimen identifier. |
| Interview metadata |
Authorized interview log or recording metadata. |
Participant code, date, and recording status. |
The research team should identify the authoritative source before collection. Otherwise, different personnel may treat conflicting documents as equally authoritative.
How Is Source Data Verification Different From Data Validation?
Data validation checks whether information satisfies specified rules. Source verification examines whether recorded information corresponds to its source.
For example, a range check may confirm that an achievement score falls between 0 and 40. Source verification may reveal that the recorded score of 36 differs from the original score of 26.
Both procedures are useful, but they detect different problems.
Data validation
Evaluates whether values satisfy defined range, format, consistency, or logic requirements.
Source data verification
Compares recorded information with the relevant authoritative source.
Measurement validation
Evaluates evidence supporting whether a measurement procedure is suitable for its intended scientific interpretation.
A database entry can pass validation and match its source while still representing an invalid measurement.
The distinction is important when planning range and logic checks during data collection.
Why Should Verification Be Planned Before Collection?
Planning establishes which records will serve as sources, how they will be retained, and how discrepancies will be investigated.
Without advance planning, the team may discover that source documents were not preserved, identifiers cannot be linked reliably, or personnel lack authorization to access the records needed for verification.
For example, if original assessment sheets are discarded after scores are entered, subsequent verification may be impossible.
Planning also allows the research team to allocate resources according to the importance and vulnerability of different data fields.
Verification is most useful when it is integrated into data management procedures rather than introduced as an improvised activity after errors have already accumulated.
Which Variables Should Receive the Strongest Source Verification?
Priority should generally be given to variables whose errors could materially affect participant protection, the primary research question, or the validity of the analysis.
These may include primary outcomes, intervention assignment, eligibility indicators, participant identifiers, critical assessment dates, and important exposure measurements.
For example, an incorrect participant identifier may link several otherwise accurate measurements to the wrong individual.
Similarly, an incorrect primary outcome score can directly affect the estimated intervention effect.
However, the appropriate control depends on the source of error. If a score is generated automatically by a validated system, verifying every exported score manually may offer limited benefit compared with validating the scoring algorithm and transfer process.
The principles for prioritizing critical research variables should inform the verification plan.
Does Every Record Need 100% Source Verification?
No universal requirement applies to every research project.
Complete verification of selected critical fields may be justified in some studies, particularly when the data are manually entered and errors could have serious consequences.
In other circumstances, targeted or sampled verification may provide a more proportionate approach.
Research on risk-based clinical monitoring has questioned the efficiency of routine exhaustive verification when resources could instead be directed toward more consequential risks.
However, evidence from clinical trials should be applied cautiously to other research settings, and applicable regulatory or contractual requirements must still be followed.
The appropriate verification coverage depends on error likelihood, scientific consequences, existing controls, available resources, and the intended use of the data.
What Are the Main Approaches to Source Verification?
| Approach |
Description |
Potential use |
| Complete verification |
Every eligible value in a defined set is compared with its source. |
Small datasets or selected high-consequence fields. |
| Random sample verification |
Records are selected using a documented probability-based procedure. |
Estimating discrepancy patterns across a defined population of records. |
| Risk-based verification |
Records or fields are selected according to predefined risks. |
Concentrating effort on critical variables or vulnerable procedures. |
| Triggered verification |
Additional checks follow identified anomalies or quality concerns. |
Investigating unusual discrepancies or site-level problems. |
| System-level reconciliation |
Data transfers or exports are checked against authoritative system outputs. |
Assessing integrity of electronic data movement. |
These approaches can be combined. For example, a team may verify all manually entered primary outcome scores while checking a random sample of lower-risk administrative fields.
Triggered verification can provide additional assurance when monitoring reveals a problem, but it should not be treated as an unbiased estimate of the overall error rate.
How Should a Verification Sample Be Selected?
The selection procedure should reflect the purpose of verification.
If the objective is to estimate the frequency of transcription errors across a dataset, a probability-based sample may be appropriate.
If the objective is to detect consequential errors in high-risk records, targeted selection may be more useful.
Researchers should distinguish these purposes because a sample deliberately enriched with suspicious records cannot directly estimate the overall discrepancy rate without accounting for its selection design.
The plan should specify the sampling unit, selection procedure, timing, and criteria for expanding verification.
Sample size should be justified according to the intended precision or quality decision rather than selected solely because a fixed percentage is administratively convenient.
How Many Records Should Be Verified?
There is no universal percentage suitable for every study.
A fixed 5%, 10%, or 20% sample may be convenient, but its adequacy depends on the number of records, expected discrepancy frequency, desired precision, and consequences of undetected errors.
For example, a small sample may provide little information about rare but consequential errors.
Where verification is intended to estimate an error rate, statistical sampling principles can help determine the required number of records.
Where the objective is to protect a small set of critical fields, complete verification of those fields may be more appropriate than sampling.
The research team should document the rationale and avoid treating arbitrary percentages as established methodological standards.
Should Verification Be Performed During or After Data Collection?
Verification during collection can identify problems while source records and responsible personnel remain available.
For example, checking a sample of manually entered assessment scores each week may reveal a recurring transcription problem before hundreds of additional records are entered.
Final verification may still be necessary after data transfer, merging, or transformation.
A staged approach can therefore combine early checks with later reconciliation.
The schedule should reflect collection volume, error risk, and the time required to resolve discrepancies.
Where a critical problem is detected, additional verification may be necessary before collection continues under the same procedure.
Who Should Perform Source Data Verification?
Verification should be assigned to personnel who understand the source records, variable definitions, and authorized correction procedures.
Independent review may be useful when the original data entry process is vulnerable to transcription errors or when consequential corrections require additional oversight.
However, independence is not an absolute requirement for every verification activity.
In small studies, the same research team may perform data entry and verification, provided responsibilities and audit procedures are appropriate to the risks.
For high-consequence corrections, separating the person proposing a change from the person authorizing it may strengthen accountability.
The plan should identify who can inspect source information, resolve discrepancies, and approve modifications.
What Happens When a Database Value Does Not Match Its Source?
A discrepancy should trigger investigation rather than immediate replacement without documentation.
First, confirm that the correct participant, observation, and source record are being compared.
Then determine whether the difference reflects transcription, coding, unit conversion, data transformation, or another process.
For example, a recorded score of 36 may differ from a source score of 26 because of a typing mistake. Alternatively, the database value may represent a transformed score rather than the raw result.
The correction should follow the approved data management procedure and preserve enough information to reconstruct the decision.
When the source itself is ambiguous or potentially incorrect, the issue may require additional investigation rather than automatic acceptance of the source value.
Does the Original Source Always Contain the Correct Value?
Not necessarily. Source verification establishes correspondence with the designated source, but the source may itself contain errors.
For example, a manually scored assessment may have been scored incorrectly before its total was entered into the database.
Similarly, an instrument may produce a systematically biased reading that is accurately transferred into the research dataset.
Verification therefore addresses a particular aspect of data integrity, not the entire validity of the measurement process.
Additional controls may include instrument testing, scoring validation, calibration, duplicate measurements, and procedural monitoring.
These controls should be coordinated within the broader data quality assurance plan.
How Does Source Verification Work With Electronic Data Capture?
Electronic systems can reduce manual transcription, but they do not eliminate every risk of data inconsistency.
When participants enter responses directly into an electronic system, the original electronic record may itself be the source.
In such cases, repeatedly comparing the database with an identical copy may provide little additional assurance.
More useful controls may include validating data capture, testing scoring algorithms, checking exports, preserving audit trails, and reconciling records across connected systems.
For example, a learning management system may calculate achievement scores automatically and export them into a research database. Verification may focus on whether the export preserves identifiers, values, and assessment dates accurately.
The appropriate process depends on the data architecture and where authoritative information originates.
What About Source Data Verification in Qualitative Research?
Source verification can be useful for some qualitative research records, but its purpose should be defined carefully.
For example, checking a transcript against an authorized audio recording may identify transcription omissions or inaccuracies.
Similarly, verifying interview dates, participant codes, and file identifiers can protect the integrity of the research record.
However, source verification should not be confused with establishing that a qualitative interpretation is objectively correct.
Interpretive analysis requires methodological coherence, reflexivity, and appropriate engagement with the evidence.
Researchers should also consider whether access to recordings, transcripts, or field notes creates confidentiality risks.
How Should Confidentiality Be Protected During Verification?
Source records may contain direct identifiers, sensitive information, or contextual details that could reveal participants' identities.
Verification personnel should access only the information necessary for their authorized responsibilities.
Secure systems, controlled permissions, pseudonymized identifiers, and documented access arrangements may reduce privacy risks.
Researchers should not create unnecessary copies of sensitive records merely to facilitate verification.
Where records are held by another institution, access should follow the applicable consent arrangements, data-sharing agreements, ethics approvals, and legal requirements.
Verification quality must be balanced with legitimate obligations to protect participants and confidential information.
What Should Be Recorded in a Verification Log?
A verification log should provide sufficient information to reconstruct the activity and its findings.
Useful fields may include the record identifier, variable checked, verification date, source used, reviewer, discrepancy status, corrective action, and resolution date.
Where a value is corrected, the original and revised values should be traceable through an appropriate audit mechanism.
The log should distinguish unresolved discrepancies from confirmed errors and authorized corrections.
Documentation should be proportionate to the study's risks and avoid unnecessary duplication of sensitive source information.
What Should Happen When Verification Reveals Repeated Errors?
Repeated discrepancies may indicate a systematic problem rather than isolated mistakes.
For example, a high frequency of incorrect assessment scores at one site may reflect inadequate training, ambiguous scoring rules, or a defective export process.
The research team should investigate the underlying cause and determine whether additional records may be affected.
Corrective action may involve repairing a data transfer process, retraining personnel, revising instructions, or expanding verification.
The team should also evaluate whether the problem could affect the primary analysis or participant protection.
Simply correcting the sampled records without investigating the broader failure may leave substantial errors undetected.
How Do You Determine Whether Verification Is Worth the Cost?
Verification should address risks that matter to the study.
For manually entered critical outcomes, source comparison may provide substantial value. For validated direct electronic capture, additional manual checking may contribute relatively little.
The appropriate balance depends on the likelihood of error, consequences of an undetected discrepancy, effectiveness of existing controls, and cost of verification.
Researchers should not assume that more source verification always means better scientific evidence.
The broader principles for balancing data quality control with research feasibility can help establish proportionate procedures.
What Should the Verification Plan Contain?
A practical source data verification plan should specify the authoritative records, critical variables, coverage, sampling method, timing, responsible personnel, and discrepancy resolution procedures.
It should also establish how corrections are authorized, how original information is preserved, and when repeated errors require escalation.
Where the study involves multiple sites, the plan should address consistency of verification practices and the management of local source records.
The verification plan may be incorporated into the research protocol, data management plan, or a dedicated quality assurance procedure.
Its level of detail should reflect the study's complexity and the consequences of data errors.
Watch Out
Source data verification is not equivalent to verifying the scientific truth of a measurement. A database can match its source perfectly while the original observation remains inaccurate, biased, or methodologically unsuitable. Verification must be combined with controls addressing how the source data were generated.
