A research question can be scientifically interesting and still be unethical to pursue in the proposed way. The problem may be excessive risk, hidden coercion, unfair selection, poor consent, unnecessary collection of sensitive information or a design too weak to justify the burden placed on participants.

That is why research ethics cannot be added at the end as a form to sign. It belongs inside the design from the first question onward.

Human-participant research ethics is the discipline of protecting the rights, welfare, dignity and legitimate interests of people whose lives, bodies, behaviour, information or experiences become part of research. It asks whether the research should be done, whether it is designed well enough to justify the burden, whether participation is genuinely voluntary, whether risks are proportionate, whether selection is fair, whether privacy is protected and whether independent oversight is appropriate.

This is a general research-methods owner. It does not replace the separate canonical owners for Medicine, Biology or Veterinary science. Clinical trials, human biomedical research, animal research and clinical care have specialist legal, scientific and ethical frameworks that must be followed in their own domains.

The ethical research loop

QUESTION
→ SCIENTIFIC / SOCIAL VALUE
→ NECESSITY OF HUMAN PARTICIPATION
→ FAIR RECRUITMENT
→ RISK / BURDEN ASSESSMENT
→ CONSENT OR OTHER LAWFUL ETHICAL BASIS
→ PRIVACY + DATA MINIMISATION
→ INDEPENDENT REVIEW WHERE REQUIRED
→ CONDUCT
→ CONTINUING OVERSIGHT
→ HANDLE DEVIATIONS / HARMS / COMPLAINTS
→ REPORT HONESTLY
→ PRESERVE / SHARE / DESTROY DATA AS AUTHORISED
→ RETURN RESULTS OR KNOWLEDGE RESPONSIBLY

The loop is ethical because it remains accountable after approval. A protocol can begin ethically and later drift. A participant can withdraw. New information can change the risk balance. A data breach can alter the privacy state. Ethics is therefore a lifecycle, not a one-time gate.

1. Research ethics begins before consent

It is tempting to reduce ethics to informed consent: if people agree, the research is ethical. That is too narrow.

People cannot consent to make a scientifically worthless design valuable. Consent does not automatically justify unnecessary risk. A highly paid offer can create undue influence. A participant may agree to a privacy intrusion without understanding what future linkage makes possible.

Consent is one protection inside a broader ethical architecture.

2. The Belmont principles remain a useful ethical grammar

The U.S. Belmont Report, published in 1979, identifies three principles: respect for persons, beneficence and justice. They remain influential far beyond one jurisdiction.

Respect for persons recognises autonomy and the need for additional protection when autonomy is diminished. Beneficence requires attention to benefits and harms. Justice concerns the fair distribution of research burdens and benefits.

These principles do not mechanically solve every disagreement. They provide questions that a protocol must answer.

3. Singapore has its own ethical and regulatory landscape

Research rules depend on jurisdiction and domain. In Singapore, the Bioethics Advisory Committee examines ethical, legal and social issues arising from research on human biology and behaviour and develops policy recommendations. Its current public materials distinguish its advisory role from the role of institutional review boards and regulators.

The BAC’s 2021 revised Ethics Guidelines for Human Biomedical Research provide a Singapore-specific resource for human biomedical research. The BAC also states that it does not adjudicate individual proposals; researchers should use their relevant institutional review process.

This article does not turn those biomedical guidelines into rules for every social, educational or historical study. It uses them as an example of why jurisdiction, domain and institutional governance must be checked rather than inferred from a generic ethics article.

4. Ethical principles and legal rules are related but not identical

A study can satisfy a minimum legal requirement and still raise ethical concerns. It can also be ethically defensible in principle while requiring legal authorisation that has not yet been obtained.

Researchers therefore need both layers: what should be done ethically, and what must be done under the applicable law, regulation, contract and institutional policy.

The U.S. Common Rule and 45 CFR 46 are examples of jurisdiction-specific regulatory structures. They are not global law.

5. Define whether the activity is research

Not every activity that collects information is research. Quality improvement, teaching evaluation, journalism, service operations and research can overlap in methods while differing in purpose and governance.

Changing the label does not remove ethical duties. If an institution calls something “service evaluation” while publishing generalisable conclusions from identifiable participant data, governance questions remain.

Where classification affects review requirements, use the responsible institutional or regulatory authority rather than self-exempting because the project feels low risk.

6. Scientific validity is an ethical issue

If a study cannot answer its question, participants may bear inconvenience or risk without a reasonable prospect of producing useful knowledge.

This does not mean every exploratory study needs a huge sample or definitive design. It means the claims and burden should match the information the design can realistically produce.

Sample-size planning, measurement validity and analysis quality therefore have ethical dimensions. See Statistical Power and Sample Size Planning and Construct Validity and Measurement Models.

7. Social value does not mean guaranteed benefit to participants

Research can be valuable because it improves knowledge for future learners, communities or institutions even when individual participants receive no direct benefit.

That distinction should be clear in recruitment and consent. A participant should not be led to believe that joining an educational study guarantees personal academic improvement or that joining a health study guarantees treatment benefit.

8. Risk includes more than physical harm

Human-participant research can create psychological, social, economic, reputational, legal and privacy harms. A survey question about family conflict may be physically harmless while emotionally distressing. A leaked dataset can expose sensitive behaviour. A classroom study can stigmatise a group if results are reported carelessly.

Risk assessment should follow the actual data and context rather than assume “non-medical” means “no risk”.

9. Probability and severity are separate

A very unlikely severe harm can matter. A common minor inconvenience can also matter when imposed on thousands of people.

Risk assessment should describe both likelihood and magnitude, plus uncertainty. “Minimal risk” is a defined regulatory category in some systems and should not be used casually when a local rule supplies a specific definition.

10. Minimise risk before asking whether it is acceptable

Do not begin by deciding that a risk is worth taking. First ask whether the same research question can be answered with less intrusion, fewer sensitive variables, a safer recruitment process, lower burden or a non-human data source.

Risk minimisation is design work.

11. Consent is a process, not a signature

A signed form does not prove understanding, voluntariness or capacity. Informed consent requires relevant information, comprehension and a genuinely voluntary choice under the applicable ethical and legal framework.

The information should be understandable to the intended participant. Dense legal language can satisfy documentation needs while failing the human communication job.

OHRP provides human research protection training, including participant-centred informed-consent resources, as one current public reference.

12. Consent should explain what participation actually involves

Participants need to know what they will be asked to do, for how long, what data will be collected, the foreseeable risks or discomforts, relevant benefits, privacy protections, voluntary nature, withdrawal arrangements and whom to contact with questions or concerns.

Additional information may be required by law or domain. The ethical principle is transparency sufficient for an informed choice.

13. Consent language should distinguish research from ordinary service

If a learner receives normal teaching plus optional research measurements, make that distinction clear. Refusing research participation should not be represented as refusing the ordinary educational service when that is not true.

Likewise, teachers or employees should not feel that participation is a hidden employment requirement merely because recruitment comes from someone with authority.

14. Power relationships can make apparent voluntariness fragile

Students, employees, prisoners, patients and dependent populations can face explicit or subtle pressure. The person asking for consent may also control grades, work opportunities, care or privileges.

Possible protections include neutral recruitment, delayed disclosure to instructors about participation, alternative activities, independent consent staff or other domain-specific safeguards.

The exact solution depends on the context. The ethical task is to identify the dependency rather than pretend a signed form makes it disappear.

15. Payment can reimburse, compensate or influence

Payment can recognise time, inconvenience and expenses. It can also create concerns about undue influence, especially where participants are economically vulnerable or risks are substantial.

The amount, structure and timing should be reviewed in context. Withholding all earned compensation from a person who withdraws early can itself create pressure to remain.

16. Withdrawal is not always the same as erasing all prior data

Participants may have a right to stop future participation. Whether already collected data can or must be removed depends on consent terms, legal requirements, de-identification, study integrity and jurisdiction.

Do not promise automatic deletion if the project cannot lawfully or technically honour that promise. Explain the actual withdrawal boundary before participation.

17. Children require age-appropriate respect as well as adult permission

Research involving children often requires permission from a parent or guardian and, where appropriate, the child’s assent. The exact legal rules vary.

Ethically, a child is not merely an object transferred into research by an adult signature. Explain the activity in developmentally appropriate language and take resistance seriously within the applicable framework.

18. Capacity is decision-specific

A person’s ability to understand and decide can vary with the complexity of the decision, timing, illness, communication support and other conditions. Capacity should not be inferred solely from diagnosis, age or disability label.

When a person lacks the required decision-making capacity, authorised surrogate processes and additional protections may apply. Domain-specific legal advice and institutional review are essential.

19. Accessibility is part of informed participation

A consent process that is unreadable to a person because of language, vision, hearing, literacy or cognitive accessibility barriers is not fully inclusive merely because the information exists somewhere.

Provide accessible formats, interpretation, communication support and sufficient time where needed. Accessibility can also improve research quality by reducing systematic exclusion.

20. Justice asks who bears the burden and who receives the benefit

A study may recruit from one group because it is easy to reach while the eventual benefits are intended for another. That can be ethically acceptable in some cases and exploitative in others.

Ask why this population is being recruited. Is it scientifically relevant? Is it merely convenient? Are groups unfairly excluded from potentially beneficial research? Are risks concentrated among people with less power?

21. Exclusion can be ethically consequential too

Protecting a group by excluding it from research can also create an evidence gap. If older adults, disabled people or language minorities are routinely excluded, later decisions may rely on evidence that does not represent them.

The ethical task is not “include everyone”. It is to justify inclusion and exclusion in relation to risk, scientific need and fair access to the benefits of knowledge.

22. Privacy begins with deciding what not to collect

The safest sensitive variable is often the one the study never needed. Data minimisation reduces breach impact, consent complexity and future misuse risk.

Ask whether precise address, full date of birth, names, device identifiers or free-text responses are necessary for the research question. If a coarser field is sufficient, collect the coarser field.

23. Anonymous and de-identified are not synonyms

Data may be directly identified, coded, pseudonymised, de-identified under a defined standard or truly anonymous in a way that prevents reasonable re-identification. The terms have jurisdiction-specific meanings.

Replacing names with IDs can reduce exposure but does not automatically make a dataset anonymous if a linkage key or distinctive combination of attributes remains available.

Route to Data Security and Privacy for the broader information-protection framework.

24. Linkage can create new ethical risk from old data

Two individually low-risk datasets can become sensitive when linked. A school attendance file and a counselling dataset may reveal patterns neither source disclosed alone.

Consent, legal basis, data-owner permissions, governance and re-identification risk may need to be reconsidered for the linked asset. See Administrative Registers and Record Systems and the separate record-linkage owner in this research collection.

25. Secondary use can be valuable and still require governance

Reusing existing data can reduce participant burden and increase scientific value. It can also move beyond what people reasonably understood when information was first collected.

Whether re-consent, waiver, notification or another legal basis is appropriate depends on jurisdiction, identifiability, original permissions, risk and institutional policy. Researchers should not infer permission from technical access alone.

26. Publicly available information can still raise ethical questions

A social-media post may be publicly visible but authored for a limited audience. Quoting it in a searchable publication can increase exposure. Combining posts can identify people who were obscure in the original context.

Public availability is relevant but not always ethically decisive. Consider expectations, sensitivity, identifiability, vulnerability and potential harm.

27. Deception requires more than “the study works better if we do not tell them”

Some behavioural research uses incomplete disclosure or deception because full disclosure would invalidate the measurement. That creates a conflict with informed consent and trust.

A defensible protocol needs a strong scientific reason, minimised risk, independent review where required, and an appropriate debriefing plan unless debriefing itself would create additional justified concerns.

Deception should not conceal material risks or replace a more ethical feasible design.

28. Debriefing repairs knowledge, not every possible harm

A debrief can explain the true purpose, correct false beliefs and allow questions. It does not automatically undo distress, embarrassment or reputational harm caused by the procedure.

Ethical review therefore evaluates the full experience, not merely whether the study ends with an explanatory sheet.

29. Covert observation raises special questions

Observation in genuinely public settings can differ ethically from recording people in spaces where they reasonably expect privacy. Identifiability, sensitive behaviour, recording technology and publication format all matter.

A note that “people were in public” is not a complete ethical analysis.

30. Qualitative research has distinctive relational ethics

An interview can produce unexpected disclosure about trauma, illegal activity, family conflict or identifiable third parties. The researcher cannot always predict the content in advance.

Plan confidentiality boundaries, distress procedures, quotation practices and the treatment of third-party information. See How Qualitative Research Works for the methodological owner.

31. Direct quotations can re-identify participants

Removing a name from an interview quotation may not make it anonymous. Distinctive events, occupations, schools or phrasing can reveal identity.

Researchers may need to paraphrase, mask contextual details or obtain explicit quotation permissions. Each approach trades fidelity, readability and privacy differently.

32. Focus groups cannot guarantee participant-to-participant confidentiality

The research team can promise how it will handle data. It cannot guarantee that another participant will never repeat what was heard.

Explain this limitation before participation and avoid inviting highly sensitive disclosure when the format cannot protect it adequately.

33. School research must separate teaching authority from research authority

Teachers have legitimate authority over learning activities. Research participation is a different relationship. If a study is optional, students should not lose ordinary educational opportunity for declining.

When the educational activity itself is part of a cluster-randomised or institution-wide study, consent and waiver questions become more complex and require the appropriate institutional process. This article cannot determine those questions for a specific school.

34. Research on employees requires attention to retaliation and confidentiality

Employees may worry that managers can see their responses. Even a truly anonymous survey can fail if staff do not believe the protection.

Independent collection, aggregation thresholds, clear governance and separation from performance management can be important protections.

35. Communities can be harmed even when individuals are de-identified

A publication can stigmatise a neighbourhood, school, ethnic group or occupational community without naming any participant.

Group-level harm belongs in ethical review and dissemination planning. A technically anonymous dataset is not automatically socially harmless.

36. Community engagement is not a substitute for individual consent

Consulting community leaders can improve relevance, trust and cultural understanding. It does not normally authorise participation on behalf of autonomous individuals unless a lawful and ethically justified framework provides otherwise.

Community and individual protections can complement each other.

37. Independent review exists because researchers have interests too

Researchers may care deeply about their hypothesis, funding, publication, career or programme. Good intentions do not remove conflict.

An institutional review board or research ethics committee provides independent scrutiny of participant protection under its governing framework. The U.S. Common Rule and Singapore biomedical governance offer different examples of formal review structures.

38. IRB approval is not a scientific warranty

An ethics committee can determine that a study meets its ethical and regulatory criteria without guaranteeing that every scientific conclusion will be correct.

Researchers remain responsible for statistical integrity, data quality, transparent reporting and domain competence.

39. Exempt, expedited and full review are governance categories

Some systems distinguish activities that are exempt from specified regulatory requirements, eligible for expedited procedures, or require full committee review. These categories have technical definitions.

Do not self-classify from a generic internet summary where local policy requires institutional determination. Low perceived risk is not automatically the same as regulatory exemption.

40. Amendments matter because approval applies to a defined protocol

Changing recruitment, adding a sensitive variable, expanding the population or altering consent can change the ethical state. Many governance systems require approval before material protocol changes except where immediate action is needed to eliminate an urgent hazard.

A protocol version should therefore be identifiable and controlled.

41. Adverse events and unanticipated problems need routes

A participant experiences unexpected distress. A device stores identifiers contrary to the protocol. A researcher accidentally emails a participant list. These events need predefined escalation, documentation and corrective action.

The relevant reporting duties depend on the governing system. Ethical operations require knowing who receives the signal before the failure occurs.

42. Complaints are evidence about the research system

Participants need a route to raise concerns independent of the researcher where appropriate. A complaint can reveal consent failure, coercion, unexpected burden or misunderstanding that routine monitoring missed.

OHRP’s current complaint guidance illustrates how oversight systems can receive and evaluate concerns within their jurisdiction.

43. Confidentiality promises need technical implementation

A consent form can promise confidentiality while the spreadsheet remains on an open shared drive. Ethical language must correspond to actual access control, encryption, retention, backup and disclosure practice.

Promises should be tested against the real data path.

44. Data retention should have a purpose and endpoint

Keeping every identifiable file forever “just in case” increases privacy risk. Destroying everything immediately can undermine audit, reproducibility and legitimate future use.

Retention periods should follow law, institutional policy, consent, research need and preservation obligations. Separate identifiers from analytic data where appropriate.

45. Open science and participant protection can conflict

Sharing data and code improves transparency and reuse. Human-participant data can contain disclosure risk that makes unrestricted sharing unethical or unlawful.

Possible responses include controlled access, synthetic data, restricted variables, data-use agreements or sharing only code and metadata. “Open” should mean as open as responsibly possible, not automatically public.

Route to Research Data Management and FAIR Principles and Open Data and Responsible Data Sharing.

46. Re-identification risk changes as external data grows

A dataset considered difficult to re-identify today may become easier to link tomorrow as public records, commercial databases and machine-learning tools expand.

Privacy review should therefore consider the data environment, not only direct identifiers in one file.

47. Algorithmic research can involve human participants without direct interaction

A model trained on human behaviour, messages, images or educational records can affect or reveal people even if no researcher meets them.

Questions include lawful access, consent or other basis, group harms, bias, re-identification, secondary use and downstream deployment. Data ethics and human-participant ethics overlap but are not identical.

48. AI-generated participant simulations are not human evidence

A language model can generate plausible interview responses or synthetic records. Those outputs may help test a pipeline or explore question wording. They do not become observations of real human experience because the prose sounds realistic.

Synthetic data can reduce some privacy risks while introducing validity and disclosure questions. Route to Synthetic Data and Privacy-Preserving Data.

49. Returning individual results can help or harm

Participants may want personal results. Some measurements are not validated for individual interpretation, and returning them can create anxiety or misunderstanding.

The protocol should decide which results can responsibly be returned, by whom, with what explanation and under which professional framework.

50. Returning aggregate results supports respect and trust

Participants often contribute time without later hearing what the study learned. A clear aggregate summary can honour that contribution where feasible and permitted.

The summary should not overstate findings or accidentally identify small subgroups.

51. Publication ethics is adjacent but separately owned

Authorship disputes, plagiarism, fabricated data, undisclosed conflicts and correction practices concern research integrity and publication ethics. They can harm participants indirectly but are a distinct governance layer.

Use How Research Integrity and Publication Ethics Work for that owner.

52. Clinical research belongs to Medicine

Research involving patients, medicines, procedures, diagnosis or treatment carries specialist medical ethics, safety and regulatory requirements. This general article should not be used to design or approve a clinical trial.

eduKate’s separate Medicine route includes The Clinical Trials & Research Participation Web. That specialist owner remains separate.

53. Veterinary research belongs to Veterinary

Research on animals raises welfare, veterinary, scientific and regulatory questions that are not solved by human consent principles.

eduKate’s Veterinary owner remains separate, including Veterinary Ethics.

54. Oral history has its own relationship to consent and archives

Oral history can be research, archival practice, historical documentation or some combination depending on institution and jurisdiction. Consent may include preservation and future access rather than only short-term analysis.

Use How Oral History Archives Work for the archival owner.

55. A participant-information audit should work backwards

PUBLIC CLAIM
→ ANALYTIC DATA
→ CODED RECORD
→ SOURCE RECORD / INTERVIEW / OBSERVATION
→ CONSENT OR OTHER AUTHORISED BASIS
→ RECRUITMENT
→ REVIEWED PROTOCOL
→ RESEARCH QUESTION
→ JUSTIFIED HUMAN BURDEN

If a researcher cannot reconstruct that path, ethical accountability has been weakened.

56. A consent-form audit should ask what could surprise the participant

• Would the participant be surprised that this variable is collected?
• Would they be surprised by who can access it?
• Would they be surprised by future linkage?
• Would they be surprised that withdrawal does not erase prior de-identified data?
• Would they be surprised that quotations may appear publicly?
• Would they be surprised by commercial, international or future research use?

Surprise is not a complete legal test. It is a practical way to expose gaps between technical consent language and reasonable human understanding.

57. A risk register should include research operations

Risk source	Possible harm	Possible control
Recruitment by authority figure	Coercion or perceived obligation	Neutral recruiter, alternative route
Free-text sensitive data	Re-identification	Minimise, redact, restricted access
Group reporting	Stigma	Context, aggregation, community review where appropriate
Linked administrative data	Expanded privacy exposure	Governed linkage, separation, minimisation
Deception	Distress, loss of trust	Necessity test, review, debrief
Small sample	Burden without useful knowledge	Design repair or narrower claim

58. Ethical proportionality changes with consequence

A short anonymous questionnaire about study habits is not ethically equivalent to collecting genomic data, mental-health histories or precise movement traces.

Oversight, consent detail, security and justification should scale with risk, sensitivity, vulnerability and irreversibility.

59. “No names collected” does not end privacy analysis

A combination of school, age, class, rare condition and timestamp can identify a person even without a name. Free text can contain names accidentally.

Assess identifiability at the dataset level and within the data environment.

60. Ethical research should be able to stop

Predefine conditions that would pause or terminate data collection: unexpected serious harm, privacy breach, evidence that risk is greater than anticipated, loss of scientific validity or inability to maintain required protections.

Stopping is not failure when continuation would be unethical.

61. Ethical research should also be able to correct itself

If consent wording proves confusing, fix it through the authorised amendment process. If data access is broader than promised, restrict it and report as required. If a published description accidentally identifies a participant, correct the publication and address the underlying process.

Trust is strengthened by visible correction more than by pretending mistakes never occur.

62. A practical pre-study checklist

1. Define the research question and social or scientific value.
2. Explain why human participation or human-derived data is needed.
3. Identify the applicable jurisdiction, institution and domain.
4. Assess scientific validity and sample adequacy.
5. Identify physical, psychological, social, economic, legal and privacy risks.
6. Minimise risk before balancing risks and benefits.
7. Justify inclusion and exclusion.
8. Map authority relationships that could impair voluntariness.
9. Design accessible consent or the appropriate alternative authorised process.
10. Minimise identifiable data.
11. Define security, retention, linkage and sharing rules.
12. Obtain the required independent review before starting.
13. Version the protocol and participant materials.
14. Define adverse-event, breach and complaint routes.
15. Plan participant and public communication of results.

63. A practical conduct checklist

1. Recruit only through approved methods.
2. Use the current approved information and consent materials.
3. Record consent or other authorisation correctly.
4. Collect only approved data.
5. Respect withdrawal and contact preferences.
6. Protect confidentiality in daily operations, not just policy documents.
7. Report required deviations and unexpected problems.
8. Do not change the protocol silently.
9. Monitor whether burden or risk differs from what was predicted.
10. Preserve an auditable record.

64. A practical publication checklist

1. Report participant flow and exclusions honestly.
2. Protect identities in tables, quotations and examples.
3. Do not overstate direct participant benefit.
4. Disclose material conflicts.
5. Report protocol deviations relevant to interpretation.
6. Share data only within the permitted governance framework.
7. Correct privacy or factual failures promptly.
8. Return an appropriate aggregate summary where feasible.

65. The Library ownership boundary

This page owns the general ethics of research involving human participants and human-derived information. It deliberately leaves specialist domains separate.

• Research Integrity and Publication Ethics owns authorship, misconduct, conflicts, corrections and publication conduct.
• Data Ethics and Responsible Use owns broader data power, fairness and human consequence.
• Qualitative Research owns interviews, observation, coding and reflexivity.
• Surveys and Sampling owns questionnaire design and population inference.
• Medicine Clinical Trials & Research Participation remains the specialist medical route.
• Veterinary Ethics remains the specialist veterinary route.

66. What a learner should remember

Ethical research is not simply research with permission. It is research whose question, design, recruitment, consent, risk, privacy, oversight, conduct and reporting respect the people who make the knowledge possible.

The strongest ethical question is often ordinary: if I were the participant, what would I reasonably need to know, what could harm me, who has power over me, and who is responsible if something goes wrong?

Sources and further reading

Sources were checked for this edition in September 2026. Rules vary by jurisdiction and research domain; always use the current responsible authority for a real protocol.

• U.S. HHS Office for Human Research Protections, The Belmont Report.
• HHS OHRP, Federal Policy for the Protection of Human Subjects (Common Rule).
• HHS OHRP, 45 CFR 46.
• HHS OHRP, Ethical Codes & Research Standards, page reviewed June 2026.
• Singapore Bioethics Advisory Committee, Ethics Guidelines for Human Biomedical Research, 2021 revised.
• Singapore Bioethics Advisory Committee, FAQ on role, review and human biomedical research.

Continue through eduKate: Research Methods and Source Evaluation → Statistical Power and Sample Size Planning → Research Integrity and Publication Ethics → Data Ethics and Responsible Use → Research Collections Directory.