Data Verification – Definition, Methods, Process, and Examples

Data verification is the systematic process of confirming that recorded, entered, transferred, transformed, or reported data agrees with its original source, an authorised reference, or documented requirements. It helps researchers detect transcription errors, missing records, unauthorised changes, processing mistakes, and inconsistencies before the data are analysed or used to support conclusions.

Introduction

Research findings are only as dependable as the data on which they are based. A sophisticated statistical model cannot correct a participant response that was entered incorrectly, a laboratory value that was assigned to the wrong sample, or a quotation that was inaccurately transcribed.

Data verification provides a structured way to check whether research data represent what was originally observed, collected, recorded, or transmitted. It may involve comparing a spreadsheet with paper questionnaires, reconciling database totals, reviewing an interview transcript against an audio recording, or checking whether a migrated dataset matches its source.

This article explains the meaning of data verification, its main methods, its relationship with validation and data cleaning, and the steps researchers can use to create an auditable verification process.

Key Takeaways

• Data verification checks data against a source, reference, specification, or audit trail.
• Verification and validation are related but are not always defined identically across disciplines.
• A verified value is not necessarily true if the original source itself is inaccurate or biased.
• Verification may be manual, automated, complete, sample based, or risk based.
• Researchers should preserve raw data, record discrepancies, document corrections, and recheck corrected values.
• Artificial intelligence can flag anomalies and match records, but it cannot independently establish ground truth.

What Is Data Verification?

Data verification is a quality-control process used to determine whether data accurately and completely represent their source or comply with predefined collection and processing requirements.

In a simple data-entry study, verification may mean comparing the values entered into a spreadsheet with the original questionnaires. In a data-migration project, it may mean confirming that every source record appears correctly in the target database. In laboratory research, it may include checking sample identifiers, instrument files, calibration records, calculations, and reporting steps.

The U.S. Environmental Protection Agency defines data verification in its environmental-quality framework as evaluating the completeness, correctness, and compliance of a dataset against methodological, procedural, or contractual requirements. Its central goal is to document that reported results reflect what was actually done (U.S. Environmental Protection Agency [EPA], 2002).

In clinical research, source data verification, commonly abbreviated as SDV, usually refers more specifically to comparing study data with their original source records. In database management, verification often means source-to-target comparison following entry, transfer, transformation, integration, or migration.

A practical working definition

For general academic use, data verification can be defined as:

The documented comparison of data with an identified source, reference, requirement, or audit trail to determine whether the data are accurate, complete, consistent, traceable, and unchanged where change was not authorised.

Because terminology differs among fields, researchers should define what they mean by “verification” in their protocol, data-management plan, standard operating procedure, or methodology section.

Why Is Data Verification Important?

Data verification is important because undetected errors can pass into statistical analysis, interpretation, publication, and decision-making. Verification identifies discrepancies early and creates evidence showing how the accuracy and integrity of a dataset were assessed.

It detects transcription and transfer errors

Errors can occur when handwritten responses are entered into a spreadsheet, recordings are transcribed, codes are assigned, files are merged, units are converted, or data are moved between systems.

Examples include:

• Entering 51 instead of 15.
• Assigning a result to participant 104 instead of participant 140.
• Omitting one page of a questionnaire.
• Changing a decimal value from 0.27 to 27.
• Losing leading zeroes from an identifier.
• Importing dates in the wrong regional format.

It protects the credibility of analysis

An analysis may be computationally correct but substantively misleading if the input data do not match the original records. Verification reduces the risk of producing precise answers from incorrectly recorded information.

It supports transparency and reproducibility

Verification logs show:

• What was checked.
• Who performed the check.
• Which source was used.
• What discrepancies were found.
• How corrections were authorised.
• Whether corrected values were rechecked.

Such documentation enables supervisors, collaborators, reviewers, auditors, and future researchers to understand how the dataset was produced.

It identifies process weaknesses

Repeated discrepancies may indicate a larger problem rather than isolated mistakes. For example, frequent date errors may reveal unclear form instructions, an import configuration problem, or incompatible date formats across sites.

It supports ethical and regulatory responsibilities

Some studies involve data that influence participant safety, treatment decisions, environmental action, financial reporting, or public policy. In these contexts, verification may form part of a formal quality-management system.

Current clinical-trial guidance emphasises data integrity, traceability, security, proportionate quality management, and risk-based monitoring of data that are critical to participant safety and reliable results (International Council for Harmonisation [ICH], 2025).

What Does Data Verification Examine?

A verification process may assess several characteristics. Not every project needs every characteristic, and each should be defined operationally.

Characteristic	Verification question	Example
Accuracy	Does the recorded value match the source?	Does the spreadsheet show the same age as the questionnaire?
Completeness	Are all required records and fields present?	Were all 250 questionnaires entered?
Consistency	Do corresponding records agree across files or systems?	Is the participant’s study ID identical in the survey and laboratory file?
Integrity	Has the data been altered, corrupted, or lost?	Does the file checksum match the value recorded at transfer?
Traceability	Can the value be traced to its source and processing history?	Can a corrected measurement be linked to the original value and correction note?
Conformance	Were documented procedures and specifications followed?	Was the required measurement protocol used?
Uniqueness	Are records duplicated when they should be unique?	Does one participant appear twice under slightly different identifiers?
Timeliness	Is the reference current enough for the intended use?	Was an administrative dataset updated before the analysis period?

These characteristics overlap with broader data-quality dimensions. Verification is one mechanism for assessing them; it is not the whole of data-quality management.

Data Verification vs. Data Validation

Data verification generally asks whether data match their source or documented requirements. Data validation generally asks whether data satisfy rules, constraints, or the intended use. However, the terminology varies across research, computing, engineering, environmental science, and regulated clinical work.

Dimension	Data verification	Data validation
Main question	Does the data agree with its source, reference, or documented procedure?	Does the data meet defined rules or appear suitable for its intended use?
Typical errors found	Transcription, transfer, omission, duplication, mapping, or unauthorised-change errors	Type, format, range, logic, relationship, or fitness-for-purpose problems
Example	Checking that 52 in a spreadsheet matches 52 on the questionnaire	Checking whether an age of 252 is outside the permitted range
Reference	Source document, source system, instrument file, protocol, audit trail	Validation rule, logical constraint, data model, or intended analytical requirement
Typical timing	After entry, capture, transfer, transformation, or reporting	Before, during, or after collection and processing
Possible outcome	Matched, discrepant, missing, altered, or unresolved	Valid, invalid, accepted with warning, or rejected

Why the distinction can be confusing

Different fields reverse or modify these meanings. For example, an environmental-quality framework may treat verification as checking procedural compliance and validation as a deeper evaluation of analytical quality. A business database guide may define validation as format checking and verification as confirmation against real-world information.

Therefore, writers should not assume that readers share one universal definition. State the definition being used and describe the actual procedures rather than relying only on the labels.

A value can be verified but invalid

Suppose a participant writes their age as 250, and the researcher enters 250 correctly.

• The value is verified because the electronic record matches the source form.
• The value fails a reasonable validation rule because the age is implausible.

A value can be valid-looking but not verified

Suppose the participant writes 52, but the researcher enters 25.

• The value 25 may pass a rule permitting ages from 18 to 100.
• It is not verified because it does not match the source.

Data Verification vs. Cleaning, Quality Control, and Research Validity

Data verification versus data cleaning

Verification detects or confirms discrepancies; cleaning changes, standardises, removes, or corrects data according to documented rules.

Verification should normally occur before or alongside cleaning. Researchers should not silently overwrite raw values during cleaning, because doing so removes the evidence needed to understand what changed.

Data verification versus quality assurance

Quality assurance is the planned system designed to prevent quality problems. It may include staff training, standard operating procedures, form design, database permissions, instrument maintenance, and verification plans.

Verification is one activity within that broader system.

Data verification versus quality control

Quality control includes operational checks used to detect problems during or after research activities. Data verification is usually considered a quality-control activity, although terminology may vary.

Data verification versus reliability

Reliability concerns the consistency or repeatability of measurement. A researcher can accurately transfer an unreliable measurement into a database. The transfer is verified, but the measurement process may still be unreliable.

Data verification versus research validity

Research validity concerns whether a study supports the interpretation or conclusion being made. Verification cannot by itself establish construct validity, internal validity, external validity, or statistical conclusion validity.

A dataset can match its source exactly while the sampling method remains biased or the instrument measures the wrong construct.

Types and Methods of Data Verification

1. Source-document comparison

Source-document comparison checks recorded data against the original document or record from which it was obtained.

Examples include comparing:

• Spreadsheet entries with questionnaires.
• Case-report forms with clinical records.
• Coded observations with field sheets.
• Published figures with official statistical tables.
• Interview transcripts with recordings.

This method is direct but may be slow and requires controlled access to the source.

2. Proofreading or visual checking

A reviewer visually compares the entered value and the source. It is simple and inexpensive for small datasets but can be affected by fatigue, expectation, and the tendency to overlook repeated errors.

Read-aloud checking, in which one person reads from the source while another reviews the entered data, can provide greater separation between entry and review.

3. Double data entry

Double data entry involves entering the same source data independently twice and comparing the resulting datasets.

A discrepancy report identifies fields that differ. The reviewer then returns to the original source to determine the correct value.

Research has shown that double entry can reduce transcription errors compared with single entry, although it requires additional staff time and does not correct errors already present in the source document (Atkinson, 2012; Paulsen et al., 2012).

Independent entry is preferable to having one operator re-enter data while viewing the first version, because the latter may reproduce the same mistake.

4. Record matching

Record matching compares corresponding records using a unique identifier or a combination of fields.

It can identify:

• Missing records.
• Duplicate records.
• Incorrect identifiers.
• Values that differ across files.
• Records that were linked to the wrong participant.

Probabilistic or fuzzy matching may be used where names, addresses, or identifiers contain variations, but uncertain matches should be reviewed rather than automatically accepted.

5. Source-to-target reconciliation

This method is common after database transfer, transformation, integration, or migration.

Researchers may compare:

• Row counts.
• Unique identifiers.
• Field-level values.
• Category frequencies.
• Minimum and maximum values.
• Missing-value counts.
• Numeric totals.
• Summary statistics.
• File sizes or checksums.

Matching totals are useful but not sufficient. Two datasets can have the same number of records while containing different records or offsetting errors.

6. Cross-source verification and triangulation

When one authoritative source is unavailable, researchers may compare information from multiple independent sources.

For example, a population estimate might be checked against:

• A national statistics office.
• A ministry report.
• An international database.
• A peer-reviewed demographic study.

Agreement increases confidence, but it does not prove correctness if the sources rely on the same underlying dataset or repeat the same error. Researchers should examine provenance and methodological independence.

7. Aggregate reconciliation

Aggregate verification checks whether totals and summaries can be reproduced from the underlying records.

Examples include confirming that:

• Participant subgroup counts add to the total sample.
• Monthly values sum to the annual figure.
• A reported mean can be reproduced from the analysis dataset.
• Table values match the statistical output.
• Percentages use the correct denominator.

This method is especially useful before publication.

8. Integrity checks using hashes or checksums

A cryptographic hash or checksum can show whether a digital file changed between two points in time.

For example, the researcher can generate a SHA-256 hash before transferring a file and compare it with the hash generated after transfer.

A matching hash supports file integrity, but it does not establish that the original file was accurate, unbiased, or correctly collected. It proves sameness, not truth.

9. Audit-trail review

An audit trail records who created, viewed, changed, approved, or exported data and when the action occurred.

Reviewing the trail can reveal:

• Unauthorised changes.
• Corrections made without explanations.
• Changes made after database lock.
• Repeated edits to critical fields.
• Missing attribution.
• Unusual access patterns.

Electronic data-capture systems such as REDCap can support attribution, validation rules, access controls, and audit capabilities (Harris et al., 2009).

10. Automated comparison scripts

Researchers can use R, Python, SQL, spreadsheet formulas, or specialist data-quality tools to compare datasets systematically.

Automated scripts are useful because they:

• Apply the same rule consistently.
• Scale to large datasets.
• Produce repeatable reports.
• Reduce fatigue-related oversight.
• Can be rerun after corrections.

However, the script itself must be tested. A coding error can produce a reassuring but incorrect verification report.

11. Sample-based verification

Rather than checking every item, researchers may verify a planned sample.

Approaches include:

• Simple random sampling.
• Stratified sampling.
• Risk-based sampling.
• Targeted checking of critical fields.
• Increased checking for sites or operators with high discrepancy rates.

Sampling reduces workload but creates a risk that errors outside the sample will remain undetected. The sampling method, unit, rationale, and escalation rules should be documented.

12. Risk-based verification

Risk-based verification allocates more checking to data whose errors would have the greatest consequences.

High-priority items may include:

• Primary outcomes.
• Eligibility criteria.
• Participant consent.
• Serious adverse events.
• Key exposure variables.
• Sample identifiers.
• Values used in regulatory or safety decisions.

In clinical research, current guidance favours proportionate monitoring focused on critical data and processes rather than assuming that every field always requires identical checking (FDA, 2023; ICH, 2025).

How to Verify Research Data Step by Step

Step 1: Define the purpose of verification

State what the process is intended to establish.

Examples:

• Confirm accurate questionnaire entry.
• Confirm that imported records match the source database.
• Confirm that reported tables match statistical output.
• Confirm that interview transcripts represent the recordings.
• Confirm that critical clinical fields are supported by source records.

Step 2: Identify the verification unit

Specify what will be counted and checked:

• Individual field.
• Record.
• Questionnaire.
• Transcript.
• File.
• Sample.
• Participant.
• Site.
• Table or figure.

A “2% error rate” is ambiguous unless the unit is stated.

Step 3: Identify the authoritative source

Possible sources include:

• Original paper form.
• Digital collection system.
• Instrument output.
• Audio or video recording.
• Laboratory notebook.
• Official database.
• Signed report.
• Protocol or standard operating procedure.
• Certified copy.

The source should be named and its limitations documented. When several sources disagree, a predetermined hierarchy or adjudication procedure is needed.

Step 4: Define criteria and tolerances

Specify what counts as:

• A match.
• A discrepancy.
• A missing item.
• A critical error.
• An acceptable numerical tolerance.
• A correctable formatting difference.
• An unresolved discrepancy.

Tolerance may be appropriate for rounded or continuously measured values, but categorical identifiers normally require exact matching.

Step 5: Select complete, sample-based, or risk-based coverage

Use complete verification when:

• The dataset is small.
• Errors would have severe consequences.
• The source is easy to compare automatically.
• Regulations or protocols require it.
• Critical identifiers or outcomes are involved.

Use sampling when:

• The dataset is very large.
• The expected risk is low.
• Resources are limited.
• Automated checks already cover major failure modes.
• Escalation rules are available if the sample performs poorly.

Step 6: Preserve the original data

Store raw data separately and restrict editing. Corrections should be made in a working copy, through a controlled system, or with a documented change log.

Never replace the only copy of the original record.

Step 7: Perform the comparison

Use the chosen method consistently. Where possible, the verifier should be independent of the person who entered or transformed the data.

Independence reduces the chance that the same assumptions or habits will be repeated.

Step 8: Record every discrepancy

A discrepancy log should include:

• Record or field identifier.
• Source value.
• Recorded value.
• Type of discrepancy.
• Date detected.
• Reviewer.
• Resolution.
• Person authorising the correction.
• Date corrected.
• Reverification status.

Step 9: Investigate the cause

Classify the discrepancy where possible:

• Transcription error.
• Source ambiguity.
• Missing source.
• Duplicate entry.
• Import or mapping error.
• Unit-conversion error.
• Coding error.
• System defect.
• Unauthorised alteration.
• Outdated reference.
• Unresolved source conflict.

Root-cause classification helps prevent recurrence.

Step 10: Correct through a controlled process

Corrections should be traceable. The original value should remain recoverable where required.

Do not “correct” a value merely because it appears unusual. An unusual but accurately recorded observation may be genuine.

Step 11: Reverify corrected data

A correction should not be assumed correct merely because it was made. Recheck it against the source and update the verification log.

Step 12: Summarise and report the results

Report:

• Data checked.
• Method.
• Coverage or sample.
• Number and type of discrepancies.
• Critical discrepancies.
• Corrections.
• Unresolved items.
• Limitations.
• Preventive actions.

Data-Verification Metrics and Formulas

Verification match rate

[
\text{Verification match rate} =
\frac{\text{Number of checked items matching the source}}
{\text{Total number of items checked}}
\times 100
]

If 1,970 of 2,000 checked fields match their source:

[
\frac{1,970}{2,000}\times100=98.5%
]

Discrepancy rate

[
\text{Discrepancy rate} =
\frac{\text{Number of discrepant items}}
{\text{Total number of items checked}}
\times100
]

Using the same example:

[
\frac{30}{2,000}\times100=1.5%
]

Completeness rate

[
\text{Completeness rate} =
\frac{\text{Required items present}}
{\text{Total required items}}
\times100
]

If 980 of 1,000 required observations are present:

[
\frac{980}{1,000}\times100=98%
]

Relative reconciliation difference

[
\text{Relative difference} =
\frac{|\text{Target total}-\text{Source total}|}
{|\text{Source total}|}
\times100
]

Suppose the source total is 50,000 and the target total is 49,500:

[
\frac{|49,500-50,000|}{50,000}\times100=1%
]

Important interpretation warning

There is no universal acceptable verification rate. The appropriate threshold depends on:

• The verification unit.
• Consequences of error.
• Data type.
• Measurement precision.
• Study design.
• Regulatory requirements.
• Whether the source itself is reliable.
• Whether discrepancies are random or systematic.

A 99.9% match rate may still be unacceptable if the few errors affect participant identity or the primary outcome. A lower rate may be manageable if discrepancies involve noncritical formatting differences that are fully resolved.

Examples of Data Verification in Research

Survey research

A researcher receives 500 paper questionnaires. Two research assistants independently enter the responses. Software compares both datasets and produces a discrepancy file. Each discrepancy is resolved by returning to the original questionnaire.

The process verifies transcription. It does not establish that participants answered honestly or that the questionnaire validly measured the intended construct.

Qualitative research

A researcher checks a sample of interview transcripts against the audio recordings. Names, numerical information, technical terms, pauses relevant to interpretation, and inaudible sections receive special attention.

The verification log records corrections and uncertain passages. Participants may also be invited to clarify factual information, although participant checking is conceptually different from transcript verification.

Laboratory research

A laboratory team verifies that:

• Sample identifiers match the collection log.
• Instrument files correspond to the correct samples.
• Calibration and quality-control records are present.
• Units were transferred correctly.
• Calculations can be reproduced.
• Reported results match the final approved output.

Verification cannot compensate for a poorly calibrated instrument or an inappropriate experimental design, but it can reveal whether required procedures were documented and followed.

Secondary-data research

A researcher downloads an official dataset and records:

• Publisher.
• Dataset title.
• Release date.
• Version.
• Download date.
• File name.
• Checksum.
• Documentation.
• Transformations.

After cleaning, the researcher uses a script to confirm that the original file remains unchanged and that every exclusion, recode, and derived variable can be traced through code.

Clinical research

Source data verification compares information in a study database with original clinical or study records. Contemporary practice increasingly combines targeted source checking with central monitoring, data analytics, audit-trail review, and risk-based oversight.

A recent scoping review concluded that SDV methods and error-rate reporting remain heterogeneous, reinforcing the need to define the verification unit, process, and outcome clearly (Hamidi et al., 2024).

Data migration

A university transfers student records from one system to another. The team compares:

• Total record counts.
• Unique student IDs.
• Programme codes.
• Enrolment status.
• Credit totals.
• Missing fields.
• Duplicate identifiers.
• Financial balances.

Field-level mismatches are investigated before the old system is retired.

Data Verification in Modern Research

Verification should begin during planning

Researchers should decide before collection:

• Which data are critical.
• What the source of truth will be.
• Which checks will be automated.
• Whether independent review is required.
• How discrepancies will be recorded.
• What level of coverage is proportionate.
• Who can authorise corrections.

Research-data guidance recommends planning quality procedures before data gathering and documenting collection, processing, and checking activities throughout the project (UK Data Service, n.d.).

Verification belongs in the data-management plan

A data-management plan can specify:

• File naming.
• Version control.
• Access permissions.
• Raw-data preservation.
• Entry procedures.
• Verification frequency.
• Backup and integrity checks.
• Audit-trail retention.
• Correction procedures.
• Responsibilities.
• Retention and sharing.

Reproducible scripts strengthen verification

Manual checking may be necessary, but repetitive comparisons should be scripted when possible.

A reproducible script can:

• Compare source and target record counts.
• Check identifiers.
• Recalculate totals.
• Generate discrepancy tables.
• Confirm expected variable types.
• Produce a dated report.
• Be rerun after corrections.

The script, software environment, and input versions should be preserved.

Verification should be proportionate

Not every field has equal importance. A practical framework classifies data according to risk:

Risk level	Examples	Suggested approach
Critical	Consent, eligibility, primary outcome, participant ID, safety event	Complete or highly intensive independent verification
Important	Major covariates, treatment dates, sample IDs	Automated checks plus targeted or stratified review
Routine	Administrative or descriptive fields	Automated checks and sample-based review
Low consequence	Optional notes not used in analysis	Limited review unless anomalies appear

This is a planning model, not a universal regulatory rule.

Digital Tools for Data Verification

Spreadsheets

Excel, Google Sheets, and similar tools can support:

• Exact-match formulas.
• Conditional formatting.
• Duplicate detection.
• Lookup comparisons.
• Record counts.
• Data-validation rules.
• Protected raw-data sheets.

Spreadsheets are accessible but become difficult to audit when formulas are overwritten, changes are not logged, or multiple versions circulate by email.

Electronic data-capture systems

Systems such as REDCap may provide:

• Form-level validation.
• User permissions.
• Field histories.
• Audit trails.
• Double-entry workflows.
• Query management.
• Export controls.

Tool availability and configuration vary by institution.

Statistical programming languages

R and Python are suitable for:

• Dataset comparison.
• Duplicate detection.
• Recalculation.
• Rule-based checks.
• Missingness summaries.
• Reproducible discrepancy reports.

SQL and database tools

SQL can compare source and target tables using joins, counts, checksums, grouped totals, and exception queries.

Data-cleaning tools

OpenRefine and comparable applications can help identify inconsistent categories, clusters, whitespace differences, and formatting variations. These functions support verification but do not independently establish the correct value.

Version control

Git and research-oriented versioning systems can preserve code history and document changes. Large data files may require specialised storage or data-versioning tools.

Checksums

Checksum utilities can confirm that files transferred without change. Record the algorithm, value, date, and file version.

Laboratory and research-information systems

Laboratory information management systems, electronic laboratory notebooks, clinical data-management platforms, and institutional repositories may provide audit trails, permissions, metadata, and workflow controls.

Artificial Intelligence and Data Verification

Artificial intelligence can assist data verification by identifying anomalies, matching records, extracting information, comparing documents, and prioritising cases for review. It should not be treated as an independent source of truth.

Tasks AI can support

AI-assisted systems may help with:

• Optical character recognition.
• Entity and record matching.
• Transcript comparison.
• Duplicate detection.
• Outlier prioritisation.
• Image or document comparison.
• Pattern-based fraud detection.
• Classification of discrepancy types.
• Identification of inconsistent text.

Risks of AI-assisted verification

AI may:

• Produce false matches.
• Miss unusual but genuine records.
• Repeat bias in its training data.
• Infer information not present in the source.
• Generate confident but unsupported explanations.
• Expose confidential information.
• Make decisions that are difficult to reproduce.
• Change behaviour following a model update.

Good practice for AI use

Researchers should:

1. Validate the AI-assisted procedure against a human-verified benchmark sample.
2. Record the model, version, settings, and date.
3. Set thresholds before reviewing results.
4. Preserve the original data.
5. Require human review of consequential discrepancies.
6. Monitor false-positive and false-negative rates.
7. Avoid uploading confidential data to an unapproved service.
8. Reassess performance when the data domain changes.
9. Document the role of AI in the methodology.

AI is most useful as a prioritisation and comparison assistant. Accountability for the final decision remains with the researcher or authorised reviewer.

Advantages of Data Verification

Data verification can:

• Detect errors before analysis.
• Improve confidence in data transfer and reporting.
• Preserve traceability.
• Support reproducibility.
• Reveal weaknesses in collection or entry procedures.
• Reduce repeated correction work.
• Strengthen audit readiness.
• Improve collaboration across research teams.
• Provide evidence for the limitations reported in a paper.

Limitations of Data Verification

The source may be wrong

Perfect agreement with a biased, fabricated, outdated, or incorrectly measured source does not make the data true.

Complete checking can be expensive

Manual 100% verification may require substantial time and can divert resources from more important quality activities.

Human reviewers make errors

Fatigue, expectation, familiarity, and ambiguous handwriting can affect manual review.

Automated checks find only specified patterns

A script cannot flag a problem that its rules do not represent. Incorrect code can also create false assurance.

Independent sources may not be independent

Two websites may agree because both copied the same underlying report.

Verification can threaten privacy

Reviewers may need access to identifiable or sensitive source records. Access should be limited to what is necessary and authorised.

A high match rate can hide critical errors

Error severity matters as much as frequency. One incorrectly linked participant record can be more serious than many harmless formatting discrepancies.

Verification does not prove validity

It does not automatically establish that the research design, sample, instrument, model, or interpretation is appropriate.

Common Data-Verification Mistakes

Treating validation rules as proof of truth

A correctly formatted email address, date, or identifier may still be factually wrong.

Failing to identify the source of truth

Verification becomes impossible when reviewers do not know which record is authoritative.

Editing raw data directly

Overwriting raw values removes traceability and may make errors impossible to reconstruct.

Using the same person for entry and checking without safeguards

The original operator may repeat or overlook the same mistake.

Checking totals only

Matching totals do not prove that individual records match.

Ignoring systematic discrepancies

A small overall error rate may conceal a high rate in one site, one variable, or one operator.

Applying one threshold to every field

Critical and noncritical variables require different levels of attention.

Correcting unusual values without evidence

Outliers should be investigated, not automatically changed or deleted.

Failing to reverify corrections

A corrected value can introduce a new mistake.

Reporting “the data were verified” without details

Readers need the method, source, coverage, verifier, discrepancy definition, and outcome.

Data-Verification Checklist

Before verification begins, confirm that:

• The purpose is documented.
• The verification unit is defined.
• The authoritative source is identified.
• Critical variables are classified.
• Match and discrepancy criteria are specified.
• Numerical tolerances are justified.
• Raw data are preserved.
• Access is authorised.
• The verifier’s independence is considered.
• Complete, sample-based, or risk-based coverage is justified.
• Automated scripts have been tested.
• Discrepancies are logged.
• Corrections retain an audit trail.
• Corrected values are reverified.
• Unresolved discrepancies are reported.
• Results and limitations are summarised.

Data-Verification Log Template

Field	Example entry
Project	Student Well-being Survey
Verification date	27 June 2026
Record ID	P-0148
Variable	Weekly study hours
Source value	15
Dataset value	51
Discrepancy type	Transposed digits
Criticality	Important
Reviewer	Reviewer B
Resolution	Corrected to 15 after source review
Authorised by	Data manager
Reverified	Yes
Root cause	Manual transcription
Preventive action	Double-entry check for numeric fields
Notes	Original questionnaire retained

How to Report Data Verification in a Thesis or Paper

A methodology section should state:

• The source data.
• The software or system used.
• Who entered the data.
• Who performed verification.
• Whether checking was independent.
• The verification unit.
• The proportion or sample checked.
• Selection method.
• Rules and tolerances.
• Discrepancy-resolution procedure.
• Results.
• Treatment of unresolved discrepancies.

Example methodology statement

Questionnaire responses were entered into an electronic database by one trained research assistant. A second assistant independently verified all participant identifiers and primary outcome fields against the original questionnaires. A simple random sample of 20% of the remaining fields was also checked. Discrepancies were recorded in a dated log, resolved by reference to the original questionnaire, and independently rechecked. The raw database was retained unchanged, and all corrections were applied to a version-controlled analysis copy.

The percentage should reflect the actual study procedure rather than being copied from a generic template.

Conclusion

Data verification is a documented process for determining whether research data match their source, reference, procedural requirements, or recorded history. Effective verification identifies critical data, preserves the original record, uses proportionate manual and automated checks, records discrepancies, and makes every correction traceable.

Verification improves confidence in data handling, but it cannot prove that the original source, measurement, sample, or research design is valid. It is therefore best treated as one component of a broader research-quality and data-governance system.

References

• The following references are formatted for an APA 7 reference list. Website-access dates may be added where required by your editorial policy.
• Andersen, J. R., Byrjalsen, I., Bihlet, A., Kalakou, F., Hoeck, H. C., Hansen, G., Hansen, H. B., Karsdal, M. A., & Riis, B. J. (2015). Impact of source data verification on data quality in clinical trials: An empirical post hoc analysis of three phase 3 randomized clinical trials. British Journal of Clinical Pharmacology, 79(4), 660–668. https://doi.org/10.1111/bcp.12531
• Atkinson, I. (2012). Accuracy of data transfer: Double data entry and estimating levels of error. Journal of Clinical Nursing, 21(19–20), 2730–2735. https://doi.org/10.1111/j.1365-2702.2012.04353.x
• Food and Drug Administration. (2023). A risk-based approach to monitoring of clinical investigations: Questions and answers—Guidance for industry. U.S. Department of Health and Human Services.
• Hamidi, M., Eisenstein, E. L., Garza, M. Y., Morales, K. J. T., Edwards, E. M., Rocca, M., Cramer, A., Singh, G., Stephenson-Miles, K. A., Syed, M., Wang, Z., Lanham, H., Facile, R., Pierson, J. M., Collins, C., Wei, H., & Zozus, M. (2024). Source Data Verification (SDV) quality in clinical research: A scoping review. Journal of Clinical and Translational Science, 8(1), e101. https://doi.org/10.1017/cts.2024.551
• Harris, P. A., Taylor, R., Thielke, R., Payne, J., Gonzalez, N., & Conde, J. G. (2009). Research electronic data capture (REDCap)—A metadata-driven methodology and workflow process for providing translational research informatics support. Journal of Biomedical Informatics, 42(2), 377–381. https://doi.org/10.1016/j.jbi.2008.08.010
• International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use. (2025). ICH harmonised guideline: Good clinical practice E6(R3).
• Monash University. (n.d.). Data quality. Monash Research Infrastructure.
• Paulsen, A., Overgaard, S., & Lauritsen, J. M. (2012). Quality of data entry using single entry, double entry and automated forms processing—An example based on a study of patient-reported outcomes. PLOS ONE, 7(4), e35087. https://doi.org/10.1371/journal.pone.0035087
• UK Data Service. (n.d.). Quality: Ensuring quality of data at all stages.
• U.S. Environmental Protection Agency. (2002). Guidance on environmental data verification and data validation (EPA QA/G-8; EPA/240/R-02/004).
• World Health Organization. (2022). Data quality assurance: Module 3—Site assessment of data quality: Data verification and system assessment.