Toolbox · Field Note 01 · Research Ethics

Before the field: what ethics review actually asks you.

审查先于田野

Notes from a Yale Human Research Protection Program training session, September 2026 — rewritten for people whose research involves archives, records, and communities that have already been studied enough.

播种光明，收获永恒

Why this note exists

The rules were written for clinical trials. Most of us are not running one.

The US system of human-subjects protection was built by physicians and lawyers, out of medical scandals, for medical research. Archival work, ethnography, and evaluation were fitted into it afterwards. That mismatch is not a reason to treat review as an obstacle — but it does mean you have to translate your project into a language it was not designed to hold.

The 1979 report that grounds the whole system says as much about itself. Its authors declined to make any determination about social experimentation, noting that the problems it raises may differ substantially from those of biomedical and behavioural research, and left the question to a successor body. That question was never fully answered. What follows is a practical map of the system as it stands, and where research on records rather than bodies tends to snag.

The threshold question

Two definitions decide everything else.

Review is triggered by a conjunction. Fail either half and you are outside the system; meet both and no framing gets you out.

ResearchA systematic investigation designed to develop or contribute to generalizable knowledge. If your findings are meant to travel beyond one setting, you are here.

Human subjectA living individual about whom you obtain information through intervention or interaction — or whose identifiable private information you obtain, use, study, analyse, or generate.

Human subjects researchSubmission required. The only open questions are which level of review, and how long it takes.

The word that catches archival researchers is generates. Added in the 2018 revision of the Common Rule, it means you fall inside the definition even when your source material arrives without names, if your analysis is what makes people identifiable again. Piecing a place, a date, and a sequence of events back into a person is exactly that.

Outside the definition: data about deceased persons; data about facilities, businesses, or organisations; genuinely de-identified data. Note that quality-improvement work escapes on a different ground — it is local by design and not intended to be generalizable — which is precisely what most research cannot claim.

Levels of review

You do not choose your level. The office assigns it.

You submit; a regulatory analyst pre-reviews for completeness and compliance; the file is then triaged to a chair or designee for exempt and expedited review, or to a convened committee for full board. Your influence runs entirely through how precisely you describe risk and mitigation.

Exemptions	8 categories	Minimal risk, and fits a listed category — public-behaviour observation, educational methods, benign behavioural interventions, and secondary use of data that is either publicly available or recorded without identifiers. Exempt from the regulatory requirements, but still determined by the office, not by you.
Expedited	9 categories	Minimal risk, fits a listed category, and meets the approval criteria. Reviewed by a chair or designee rather than a meeting.
Full board	Convened	Greater than minimal risk, or fitting no expedited category, or touching contested subject matter. Reviewed by a committee in session. At a large institution this is less slow than it sounds — Yale runs twenty-nine committees meeting weekly — but it adds a revision cycle.

Minimal risk is a term of art: the probability and magnitude of harm no greater than those ordinarily encountered in daily life or in routine examination. Judged against the participant's daily life, not yours.

Principles

Three principles, three applications.

Respect for persons → consent

Two obligations, not one: acknowledge autonomy, and protect those whose autonomy is constrained. The founding text works this through with prisoners, and concedes it is a genuine dilemma — excluding them disrespects them; including them may coerce them.

Beneficence → risk assessment

Minimise harm, maximise benefit. Harms are listed as psychological, physical, legal, social, and economic — and the text warns against attending only to the first two. For records research, the live categories are usually the last three.

Justice → selection

Are these participants chosen because the question requires them, or because they are available, compromised, or easy to reach? Communities that have been researched repeatedly are exactly where this question bites.

Consent is a process, not a form. The three required elements are all verbs: disclose what is needed for a decision, facilitate understanding of what you disclosed, and promote voluntariness. Comprehension must be adapted to the participant's language and literacy, and the investigator is responsible for confirming it — a signature is not evidence of understanding. Where a signed form is itself the most dangerous document in the study, a consent script with waiver of documentation is a standard instrument, not a special favour.

On compensation. The founding text is explicit that an inducement which would ordinarily be acceptable becomes undue influence when the participant is especially vulnerable, and that pressure from figures of authority — particularly where sanctions are possible — is unjustifiable. If access to participants runs through local officials, your design needs a route by which someone can decline without the official learning of it.

Working internationally

The slow line is the one you do not control.

Protocols are expected to align with local law and local standards of practice; the review board considers local context, including legal, cultural, and logistical factors. In practice this means two review systems and one sequence.

Local ethics approval	First	The international checklist asks for the local IRB or ethics committee's approval alongside it. Approval, not an application in progress. National research permits often sit on top of that. Budget months, and start before you think you need to.
Letters of support	Parallel	From recruitment or research sites. These are relationships, not paperwork, and they cannot be produced on demand.
Faculty PI or advisor	Parallel	Students generally cannot serve as principal investigator without special permission; a faculty advisor agreement is required. Identify the person a year out.
Home institution	Last	Protocol, submission form, consent document or script, instruments, recruitment materials. The fastest part of the process, and the only part fully in your hands.

Also worth knowing which framework your counterpart committee actually uses. Belmont is a US document; ethics committees elsewhere more often work from the Declaration of Helsinki — which since 1964 has covered research on identifiable human material and data, and whose 2024 revision moved further toward participation and toward returning results to the communities studied — and from the CIOMS guidelines.

The 1940s Guatemala experiments were funded, conducted, and then filed. They surfaced in 2010 because a historian read a researcher's papers in a university archive and understood what she was looking at. A record can sit for sixty years and still change a government's conduct when it finds a different reader.

Why records research is not a lesser ethical category

A worked case

Where a post-conflict archive snags.

Rwanda's gacaca courts tried hundreds of thousands of cases in community hearings after 1994 and left a written record of roughly sixty million pages of testimony. The courts closed in 2012; the record did not. Reading six concerns off that case shows how the general rules land on a specific project.

De-identification does not work on narrativeStrip the names and the testimony is still identifiable by content — the hill, the date, the order of events. Anyone from that community knows who is described. The regulatory standard for de-identification was built for structured datasets and does not transfer.

The secondary-use exemption may not reachThe exemption for existing data turns on the material being publicly available or handled without identifiers. A record held in custody by a foreign state is neither.

The archive is mixedSome of the people named are dead and some are living. Research confined to the deceased falls outside the definition; a single living identifiable person brings the whole analysis back in. Separating the two can, however, make a faster preliminary component possible.

The risks are social and legalNot physical, not psychological: a witness re-identified inside their own community, in a legal environment where speech about the genocide is itself regulated. This is the section of the application that determines the level of review.

Access runs through intermediariesIntroductions by local officials are often the only practical route to participants, and are also the classic structure of undue influence. Design the exit before you design the interview.

Some participants may be incarceratedResearch involving prisoners has its own subpart and its own procedural consequences. Decide early whether that population is necessary to the question.

Practice

Four things worth doing a year early.

Book a consultation before you submit anything. Most programmes offer one. An hour with the office costs nothing, commits you to nothing, and returns a realistic timeline — which is the single most useful number you can put in a funding application.

Read the protocol template before you write the proposal. Its section headings are a complete list of what reviewers want to know. It is the best available skeleton for a research-design section.

Start the foreign approvals first. Everything you cannot write yourself is slower than everything you can.

Answer revision requests the same week. What actually delays approval is rarely the committee. It is the applicant's reply time — the one variable entirely under your control.

Tool 01

Is it human subjects research?

Five questions, in the order a regulatory analyst would ask them. This gives you a working expectation, not a determination — only the review office can make one of those.

QUESTION 1 OF 5

Is your project a systematic investigation designed to develop or contribute to generalizable knowledge?

Findings meant to travel beyond the single setting you studied. Internal evaluation intended only to improve one programme in one place is not this.

Yes No

QUESTION 2 OF 5

Does your material concern living individuals?

Research confined to people who have died falls outside the definition. Archives are usually mixed: if even one identifiable living person is in scope, answer yes.

Yes No — deceased only

QUESTION 3 OF 5

Will you obtain, use, study, analyse, or generate identifiable private information about them — or interact with them directly?

Remember generate: if your analysis is what makes an anonymised source identifiable again, you are inside the definition. Test it by asking whether someone from that community could name the person from your description.

Yes No — genuinely de-identified

QUESTION 4 OF 5

Could participation plausibly expose someone to harm beyond that of ordinary daily life?

Count legal, social, and economic harm, not only physical and psychological. Judge against the participant’s daily life, in their setting, not yours.

Yes, or I am not sure No — minimal risk

QUESTION 5 OF 5

Is your source material either publicly available, or handled without recording identifiers?

Records held in the custody of a government, an institution, or a family are not publicly available, however freely they were once spoken.

Yes No

Probably not researchWithout the aim of generalizable knowledge, this falls outside the definition and no submission is required. Two cautions: the exemption is lost the moment you decide to publish, and work that relies on it must not be described as research in its own materials. If publication is even possible, plan as though you were inside.

Start over

Probably not human subjects researchInformation about people who have died does not make them human subjects. Confirm the file is genuinely clean of living identifiable people — testimony about the dead routinely names the living. Ask the office for a written not-human-subjects determination rather than assuming one; it is quick, and it is the document a funder will want to see.

Start over

Probably not human subjects researchGenuinely de-identified material sits outside the definition. Apply the harder test first: could someone from that community identify the person from what remains — the place, the date, the sequence? For narrative sources the answer is usually yes, in which case you are inside the definition after all.

Start over

Likely exempt, but determined by the officeMinimal risk plus publicly available or non-identified source material is the standard exemption pattern. You still submit an exemption request; you do not declare it yourself. Expect a short turnaround and light documentation.

Start over

Likely expeditedMinimal risk with identifiable material generally falls to a chair or designee rather than a convened committee. Write the risk section as precisely as you can — the distinction between expedited and full board is made on how well the harms and mitigations are specified, not on the topic.

Start over

Plan for full boardGreater-than-minimal risk, contested subject matter, or uncertainty about either sends a file to a convened committee. At a large institution this is weeks rather than months, but it adds a revision cycle. Three things follow:

Submit early enough to absorb one round of requested changes.
Write harms as legal, social, and economic — not only physical and psychological.
Answer revision requests the same week. Applicant reply time is what actually delays approval.

Start over

Tool 02

Work backwards from the field.

Enter the month you intend to start fieldwork. The dates below are planning heuristics for a project with a foreign site, drawn from how long each step takes when nothing goes wrong. Treat them as the latest acceptable dates, not targets.

12 months out	Book a consultation with the review office Costs nothing, commits nothing, returns a real timeline.
11 months out	Confirm a faculty PI or advisor Students generally cannot be principal investigator alone.
10 months out	Open the foreign approval file Identify the committee, obtain its forms, learn its meeting cycle.
9 months out	Request letters of support from sites These are relationships, not paperwork; they cannot be rushed.
7 months out	Submit to the local ethics committee National research permits often sit on top of this.
4 months out	Local approval in hand The international checklist asks for approval, not an application in progress.
3 months out	Submit the home protocol Placed to absorb one full revision cycle.
1 month out	Approval and final logistics Visas, affiliations, translation of consent materials.
Month zero	Fieldwork begins

Two rules this schedule encodes. Everything you cannot write yourself is slower than everything you can — so foreign approvals, support letters, and a faculty signature start first. And the home submission is placed to absorb one full revision cycle, because one is normal.

Tool 03

What to ask in a consultation.

An hour with the review office is the cheapest thing available to you and almost nobody books it. Bring your project in five minutes and spend the rest on questions whose answers are facts, not opinions. Below is a set that works for records-based and international projects; take the ones that apply.

TIMELINE AND PATH

1. For a social-behavioural protocol of this shape, what is the realistic submission-to-approval window here?
2. What would push this to full board rather than expedited, and is there anything in the design that would change that?
3. Does the international checklist require the local approval in hand at submission, or can it be in progress?

THE MATERIAL

4. How does this office treat records held by a foreign government — not publicly available, not covered by our own health-privacy rules?
5. De-identification does not work on narrative testimony, which is identifiable by content. Has the office handled that before, and how?
6. Can the portion concerning deceased persons be carved out for a separate not-human-subjects determination?

PEOPLE

7. Is a consent script with waiver of documentation standard in a setting where a signed form is itself the risk?
8. If access to participants runs through local officials, how does the board assess voluntariness, and what designs have satisfied it?
9. If some participants are incarcerated, what does the prisoner subpart change procedurally?

PROCESS

10. What does the faculty advisor agreement require of the signer, and who can serve?
11. What can I do in the next three months so that a funding application can say something credible about the review pathway?

Bring it on one sheet of paper with a one-paragraph project description at the top, and hand a copy across the table. Whoever you are meeting will write on it, and that sheet becomes the record of the meeting.

播种光明，收获永恒