03 · What You Need to Know
How to Protect Online Research From Duplicate Submissions and Fraudulent Participation
What Counts as Duplicate Participation?
Duplicate participation occurs when the same individual contributes more than one participation event in circumstances where the research protocol permits only one.
For example, an online cross-sectional questionnaire may require one response per eligible participant. Two submissions from the same person could therefore constitute duplicate participation.
However, repeated submissions are not always inappropriate.
Longitudinal studies, repeated-measures experiments, and diary studies may intentionally collect multiple responses from each participant.
Researchers must define the permitted participation structure before identifying duplicates.
Distinguish Duplicate, Fraudulent, and Low-Quality Responses
Duplicate participation
Repeated participation by the same individual when the study permits only one enrollment or submission.
Fraudulent participation
Deliberate misrepresentation or fabrication intended to obtain entry, compensation, or another benefit under false pretenses.
Low-quality response
A submission containing insufficient, inattentive, inconsistent, or otherwise unreliable information, without necessarily involving deliberate deception.
These categories overlap but should not be treated as equivalent.
A genuine participant may accidentally submit a questionnaire twice. Another may respond carelessly without intending to deceive. An automated submission may contain plausible-looking answers.
Different problems require different prevention and review procedures.
Why Are Online Studies Vulnerable to Invalid Participation?
Open recruitment links can be distributed beyond the intended audience, and participants may enroll without direct interaction with the research team.
Financial incentives may increase motivation to submit multiple responses or misrepresent eligibility.
Automated systems may also generate submissions, while online participant marketplaces and recruitment platforms can introduce their own quality-control challenges.
However, the prevalence of these problems varies across studies and recruitment arrangements.
Researchers should not assume that every open online survey experiences extensive fraud or that a particular recruitment platform guarantees authentic participation.
What Types of Suspicious Participation Might Occur?
| Potential problem |
Illustrative pattern |
Important qualification |
| Repeated submission |
One person submits several responses |
May be accidental or permitted by design |
| Eligibility misrepresentation |
Applicant falsely claims a required qualification |
Inconsistent answers may reflect misunderstanding |
| Automated submission |
Software generates form responses |
Technical indicators can produce false positives |
| Fabricated identity |
Participant invents credentials or personal characteristics |
Verification requirements must remain proportionate |
| Inattentive responding |
Responses are entered without adequate consideration |
Not necessarily intentional fraud |
| Compensation-driven repetition |
Multiple enrollments seek repeated payment |
Incentives alone do not establish misconduct |
| Coordinated invalid responses |
Several submissions contain suspicious shared patterns |
Shared characteristics may have legitimate explanations |
Researchers should investigate patterns using evidence appropriate to the study rather than assign fraudulent intent from one unusual characteristic.
How Can Study Design Reduce Duplicate Participation?
Prevention begins with defining whether participants may enroll once or repeatedly.
Where one-time participation is required, suitable controls may include unique invitation links, authorized participant identifiers, or platform mechanisms that limit repeated enrollment.
For example, a university study using an approved student roster may distribute one unique invitation to each selected participant.
However, such controls depend on the quality of the underlying identifiers and the platform's implementation.
They should also be designed to preserve confidentiality and accommodate participants who need to resume interrupted submissions.
Are Unique Survey Links Better Than Open Links?
Unique links can provide stronger control over invitations and participation when a defined recruitment frame exists.
They may help distinguish completed responses from repeated attempts and support follow-up with selected participants.
However, unique links can be forwarded, misused, or associated with identifying information.
They may also be unsuitable for anonymous open recruitment where no participant list exists.
Researchers should choose invitation controls according to the sampling design and privacy requirements rather than assume that unique links are always necessary.
Can IP Addresses Identify Duplicate Participants?
IP addresses may provide limited information about network connections, but they are not reliable individual identifiers.
Multiple legitimate participants may share an institutional, household, or public network address.
Conversely, one individual may use different connections or addresses across submissions.
IP information can therefore contribute to a broader investigation but should not independently determine whether participation is fraudulent.
Researchers should also consider whether collecting or retaining IP addresses is necessary and authorized under applicable privacy requirements.
What About Cookies and Device Fingerprinting?
Cookies and related mechanisms may help identify repeated browser sessions under particular conditions.
However, participants can use different browsers or devices, clear stored information, or share devices with others.
Device fingerprinting can also create significant privacy concerns because it may permit tracking across interactions.
Researchers should not assume that technically available tracking methods are ethically justified.
Any use should be proportionate, disclosed where required, and consistent with the approved research protocol.
Can Completion Time Reveal Fraud?
Unusually short completion times may indicate that a participant did not engage adequately with the questionnaire.
However, completion speed varies with reading ability, familiarity with the topic, device characteristics, accessibility tools, and questionnaire design.
For example, a short questionnaire may legitimately be completed quickly by an experienced respondent.
Researchers should establish empirically or conceptually justified thresholds and evaluate completion time alongside other indicators.
Extremely fast completion is a reason for further review, not automatic proof of fraud.
Are Identical Answers Evidence of Duplicate Participation?
Not necessarily. Participants may genuinely provide identical responses, especially when questionnaires contain few items or limited response categories.
For example, several teachers may select the same options concerning institutional assessment practices.
Identical answer patterns become more informative when considered alongside independent evidence of repeated participation.
Researchers should avoid deleting responses solely because they resemble other submissions.
The probability of coincidental similarity depends on the instrument and population.
Can Attention Checks Improve Data Quality?
Attention checks may identify certain forms of inattentive responding, but they do not directly establish participant identity or eligibility.
They can also disadvantage respondents who misunderstand instructions, experience language barriers, or use accessibility technologies.
Researchers should evaluate whether attention checks are appropriate for the participant population and questionnaire.
Exclusion rules should be specified in advance where feasible and applied consistently.
Failing an attention check should not automatically be described as fraudulent behavior.
Can Open-Ended Questions Help Identify Invalid Responses?
Open-ended questions may provide additional evidence about participants' understanding and experiences.
For example, a study recruiting university teachers may ask participants to describe a relevant teaching practice.
However, genuine participants may provide brief, formulaic, or linguistically imperfect answers.
Automated text generation also complicates the interpretation of unusually polished or generic responses.
Researchers should not treat writing style, grammatical accuracy, or apparent sophistication as definitive indicators of authenticity.
Qualitative response assessment should use clear criteria and consider the intended research purpose.
Can Automated Bot-Detection Systems Be Trusted?
Automated detection systems may help identify patterns associated with scripted or automated submissions.
However, their performance depends on the platform, detection rules, and population.
False positives can occur, particularly when legitimate participants use shared networks, privacy technologies, or accessibility tools.
Researchers should understand the available validation evidence and limitations of the detection system.
Automated scores should ordinarily be treated as indicators requiring appropriate interpretation rather than unquestionable determinations of fraud.
How Should Researchers Combine Multiple Indicators?
A layered review procedure can consider several independent sources of evidence.
For example, a submission may warrant review when it combines contradictory eligibility information, an apparent repeated invitation identifier, and an unusual response pattern.
However, adding more indicators does not automatically improve accuracy when the indicators share the same underlying limitations.
Researchers should establish clear decision rules and distinguish evidence supporting confirmed duplication from weaker indicators of possible invalidity.
Uncertain cases may require further review rather than immediate deletion.
Should Fraud Detection Be Conducted Before Data Collection?
Prevention procedures should be planned before recruitment begins whenever possible.
This allows researchers to define participation limits, eligibility requirements, response-quality criteria, and review procedures without knowing which exclusions would favor particular findings.
However, unexpected problems may emerge during collection and require protocol amendments or additional safeguards.
Such changes should be documented, justified, and reviewed through appropriate institutional procedures when necessary.
Researchers should avoid modifying exclusion rules merely because particular responses are inconvenient for the analysis.
How Can Researchers Avoid Excluding Legitimate Participants?
Fraud prevention must account for the possibility of false-positive classifications.
For example, two students using the same campus network may share an IP address while representing distinct eligible participants.
Likewise, participants using assistive technologies may exhibit interaction patterns different from those anticipated by automated screening tools.
Researchers should evaluate whether detection rules disproportionately affect particular groups.
Where feasible, uncertain cases should be reviewed using additional evidence rather than excluded automatically.
What Is the Difference Between Fraud Prevention and Data Cleaning?
Fraud prevention attempts to reduce invalid participation before or during collection.
Data cleaning identifies and addresses problems in the records after they have been generated.
Both may be necessary, but they serve different purposes.
For example, a unique invitation system may reduce repeated enrollment, while post-collection checks may identify duplicated submissions caused by technical errors.
Researchers should distinguish invalid participation from duplicate database records created by synchronization or export processes.
The latter may require data reconciliation rather than exclusion of a participant.
How Should Suspicious Responses Be Classified?
A useful procedure distinguishes confirmed invalid cases, cases meeting prespecified exclusion criteria, and unresolved suspicious cases.
Confirmed duplication may be supported by a reliable unique participation identifier and source-system evidence.
Other cases may show unusual patterns without sufficient evidence to establish deliberate misconduct.
Researchers should record the reasons for classification and the procedures used to reach decisions.
Labels such as fraudulent should be reserved for cases where the evidence supports that interpretation.
Should Suspicious Responses Be Deleted Permanently?
Generally, researchers should preserve the authorized original dataset and document exclusions in the analytical workflow.
Confirmed invalid submissions may be excluded from analysis according to prespecified or appropriately justified rules.
However, permanent deletion may interfere with auditability or reproducibility when retention is otherwise authorized.
Conversely, identifiable or unnecessary information should not be retained beyond applicable requirements merely for convenience.
Data handling should balance research integrity with privacy and retention obligations.
How Can Fraud-Related Exclusions Affect Statistical Results?
Invalid responses may distort descriptive estimates, associations, and uncertainty calculations.
However, incorrectly excluding genuine responses can also introduce bias.
For example, removing all very fast responses may disproportionately exclude participants with high familiarity with the questionnaire topic.
Researchers should evaluate the number and characteristics of excluded submissions and whether conclusions depend on uncertain exclusion decisions.
Where appropriate, sensitivity analyses may compare results under defensible alternative handling rules.
Does Participant Compensation Increase Fraud Risk?
Compensation may create incentives for repeated enrollment or eligibility misrepresentation, particularly when participation is open and verification is limited.
However, compensation is also an important means of recognizing participants' time and reducing financial barriers.
Researchers should not assume that compensated participants are inherently less trustworthy.
Instead, compensation arrangements should be proportionate and supported by suitable enrollment controls.
Ethical considerations include fairness, undue influence, accessibility, and protection of participant information.
Can Research Remain Anonymous While Preventing Duplicates?
Some duplicate-prevention procedures can operate without collecting direct names, but strong assurance that every response comes from a unique person may be difficult in fully anonymous open recruitment.
For example, anonymous invitation tokens may limit use of each token while not necessarily proving who used it.
Researchers should distinguish anonymity from confidentiality and explain the actual technical arrangements.
If identity verification is essential, the study may require a confidential rather than fully anonymous enrollment process.
The appropriate balance depends on the scientific and ethical requirements.
How Should Researchers Report Fraud Prevention?
Reports should describe relevant eligibility screening, duplicate controls, response-quality criteria, and procedures for reviewing suspicious submissions.
Researchers should report the number of records excluded for different reasons when these quantities are available and relevant.
For internet surveys, the CHERRIES checklist provides useful reporting guidance concerning survey administration and response handling.
Where detection rules are uncertain, the limitations should be acknowledged rather than implying perfect classification.
Transparent reporting allows readers to evaluate both the protection of data integrity and the possibility of erroneous exclusions.
When Can Fraudulent Participation Make an Online Dataset Unusable?
A dataset may be unsuitable when invalid participation is extensive, the identity or eligibility of essential respondents cannot be established, or contamination cannot be separated from genuine observations with sufficient confidence.
For example, an open survey intended to represent licensed professionals may be compromised if a substantial proportion of submissions cannot be distinguished from ineligible respondents.
However, unusual response patterns alone do not establish that the entire dataset is fraudulent.
Researchers should evaluate the evidence, consequences for the intended inference, and feasibility of defensible remediation.
If reliable conclusions cannot be supported, additional recruitment or a revised design may be necessary.
Watch Out
Do not equate unusual responses with fraud. Shared IP addresses, fast completion, identical answers, and automated detection scores can all have legitimate explanations. Fraud prevention should protect data integrity without converting imperfect technical indicators into unsupported accusations or automatic exclusions.
