Education Research Ethics and Evidence Synthesis

Schools are full of questions that deserve serious investigation. Does a reading intervention help children interpret unfamiliar passages? Does a timetable change reduce lost instructional opportunity? What happens to learning when a digital platform gives incorrect explanations? Which students remain excluded from an otherwise successful programme?

A Ministry of Education needs education research ethics, evidence synthesis and knowledge-to-policy assurance: a public system for asking answerable questions, protecting students and staff who participate, evaluating the complete relevant evidence, and translating findings into authorised, correctable decisions.

The governing question is: Can the ministry produce and use research that is credible, ethically obtained, relevant to its actual schools and capable of changing a decision—without treating learners as raw material or selecting only studies that support a preferred policy?

This is an independent eduKateSG Ministry of Education V2.0 reference, not a statutory research ethics code or official government evaluation policy. Research approval, privacy, human-participant protections and legal duties depend on the competent jurisdiction and institution.

What this reference does differently

Reference 19 explains how to evaluate a defined policy intervention and scale it responsibly. Reference 08 explains how evidence enters a national decision ledger.

This guide addresses the institutional machinery before and between those stages: what questions deserve research, who may study students, how conflicting findings are synthesised, what the evidence really supports, how studies are corrected or retracted, and how policymakers avoid turning a promising abstract conclusion into an ungrounded classroom mandate.

The OECD’s Who Really Cares about Using Education Research in Policy and Practice? (2023) examined more than 30 education systems and identified important gaps in the coordination, accessibility and synthesis of education evidence.

A national education research service should therefore satisfy seven requirements:

A clearly defined public educational question.
Appropriate ethical and institutional permissions.
Sound, transparent and proportionate research methods.
Traceable evidence with accurate limitations.
Competent synthesis across relevant studies.
A decision route with named authority and alternatives.
Long-term correction, follow-up and research integrity.

1. Begin with a problem teachers and learners actually face

Research agendas can become dominated by topics that are fashionable, easy to measure or attractive to funders.

A practical public research question should identify:

Learner or service: Who is affected?

Current problem: What has been observed, using evidence rather than assertion?

Relevant mechanism: Why might a change matter?

Decision: What real policy or teaching choice could the result inform?

Alternatives: Which other explanations or interventions should be compared?

Ethical burden: What would research participants need to experience?

Feasibility: What resources and follow-up would make a trustworthy investigation possible?

For example, “Does AI improve education?” is too broad. “When teachers use a bounded AI explanation tool in a defined fraction unit, do students perform better on fresh independent tasks than under a credible existing approach?” is more useful—provided the study design and permissions are appropriate.

The ministry should invite relevant teacher and learner perspectives without expecting those groups to design or authorise the entire research process.

2. Distinguish service monitoring, programme evaluation and human research

Not every analysis of school information is identical.

Ordinary service monitoring checks whether authorised obligations are being delivered, such as teacher coverage or library opening hours.

Programme evaluation asks whether a defined intervention was implemented and what outcomes followed.

Research involving people may systematically investigate new questions, collect additional data or assign treatments in ways that require formal research ethics oversight under applicable rules.

Secondary data research uses existing records but can still create privacy, re-identification or consent concerns.

The boundaries and approval duties differ across jurisdictions. A ministry must adopt the actual legal and institutional definitions rather than claim that calling an activity “quality improvement” automatically removes all research responsibilities.

A single activity can raise both service and research questions and may require appropriate review.

3. Make children’s interests and agency central to ethical design

UNICEF’s Policy on Ethics in Evidence Activities Involving People, linked to its May 2026 ethics work, emphasises ethics review, safeguarding, risk management, meaningful consent or assent where applicable, confidentiality and privacy.

That UNICEF policy governs UNICEF-supported evidence activities. It is an important comparative reference, not a national education statute.

A public education research protocol should address:

whether the question justifies involving children;
foreseeable risks and disproportionate burden;
consent, assent or another lawful basis as applicable;
privacy and data minimisation;
child-friendly explanation of participation;
protections against pressure from teachers or institutional authority;
safeguarding and referral arrangements;
equitable selection of participants;
safe withdrawal or change in participation where permitted;
oversight and response if unexpected harm occurs.

Children should not feel that their grades, access to teaching or relationship with school staff depend on joining a study.

4. Protect ordinary educational entitlement during experimentation

A school may wish to compare two reading methods. It cannot assume that researchers may remove legally required provision from a comparison group merely because doing so would simplify an analysis.

Before approval, identify the educational standard each participant must continue to receive.

Where allocation between interventions is appropriate, use competent ethical and methodological review. Ensure students can access necessary safeguarding, support and suitable instruction under applicable rules.

Researchers must disclose relevant differences in time, staffing, teaching materials and other implementation conditions.

The purpose is to learn while protecting learners, not to design an elegant experiment that ignores public obligations.

5. Give researchers access only to information they legitimately need

Educational records can include attendance, examination marks, assessments, disability-related arrangements, family communications and safeguarding concerns.

A research team does not automatically need or have permission to see all these records.

An authorised data access plan should specify:

Research question and approved purpose.
Dataset and source authority.
Fields actually required.
Lawful access basis and relevant permissions.
Data minimisation and pseudonymisation or other protection where appropriate.
Who may access, link or export records.
Security, retention and deletion requirements.
Disclosure and publication safeguards.
Correction and incident response.
Limits on use for later unrelated studies or AI processing.

Aggregated or de-identified data can sometimes be appropriate but are not universally anonymous, particularly for small school cohorts. Qualified data governance review remains necessary.

6. Demand a defensible study design

A research question, method and result must fit together.

For every project, document:

Population: Which learners, schools or educational processes are being studied?

Comparison: What alternative explains what might have happened otherwise?

Outcomes: Which capabilities or services will be measured, and how?

Timing: How long is needed to assess implementation, retention or progression?

Bias: What selection, missing-data or measurement problems are foreseeable?

Analysis: Which methods are appropriate to the design?

Limitations: Which conclusions will remain unsupported?

Decision use: What evidence would be strong enough to justify an actual change?

A descriptive school study can reveal a problem but may not establish its cause. A carefully designed experiment can strengthen causal inference but still have limited relevance to other classrooms.

The British Educational Research Association’s Ethical Guidelines for Educational Research, fourth edition, published in 2018, emphasise responsibility to participants, ethical judgement across different research approaches and the integrity of research communication.

7. Register important questions and outcomes before observing results

Research becomes less trustworthy when analysts search many outcomes and publish only the ones that moved in a favourable direction.

A suitable protocol can predefine the main question, primary measure, comparison and analysis approach where appropriate.

Not every qualitative or exploratory study follows the same preregistration model. The principle is to make consequential methodological choices visible and explain any changes rather than present a revised measure as though it was always the original plan.

An ethics or research register may record:

study identifier and approval status;
principal question and justification;
participants and recruitment rules;
intended data and outcome measures;
methodological plan and relevant uncertainties;
changes with dates and reasons;
risks, protection arrangements and oversight;
reporting and correction duties.

A published protocol cannot replace professional judgement as new evidence arises, but it can help readers understand which conclusions were anticipated and which were exploratory.

8. Separate a strong study from a complete evidence base

One well-conducted study can still occur under unusually favourable conditions.

Perhaps a reading intervention was tested with highly trained teachers and strong family support. Perhaps another report measured only short-term recall. A third study from another language system may use a different definition of the skill.

Evidence synthesis asks what can be learned from the relevant body of research, not from the most appealing result.

A systematic review or other structured synthesis should:

State the question and population.
Define inclusion and exclusion criteria.
Search appropriate sources transparently.
Check study methods and risks of bias.
Compare outcomes and definitions carefully.
Examine inconsistent results and contextual differences.
State the confidence and remaining uncertainty.
Explain limits of transfer to the target jurisdiction.
Record the date and later correction requirements.

The OECD’s 2023 study identifies evidence synthesis as an important but often underdeveloped component of education research engagement.

9. Create an evidence ladder that prevents false certainty

The eduKateSG V2.0 knowledge system can distinguish levels of claim maturity without presenting them as a universal scientific standard.

A practical evidence ladder might be:

E0 — Question: An important problem or hypothesis exists.

E1 — Observation: A relevant pattern has been documented, with limitations.

E2 — Mechanism: A plausible explanation links the problem and proposed response.

E3 — Implementation: The intervention can be delivered as intended in bounded conditions.

E4 — Evaluation: Appropriate evidence indicates what changed and how confidently causes can be inferred.

E5 — Transfer: Results hold or have been appropriately adapted across meaningful settings.

E6 — Institutional assurance: The programme is governed, monitored and correctable at scale.

A finding at one stage does not automatically prove the next. A study can offer strong causal evidence in a specific context without demonstrating national capacity or long-term sustainability.

The ladder is a proposed decision framework and should not be confused with a formal national research-grading rule.

10. A worked case: the reading programme with impressive results

Imagine a hypothetical early-reading programme that produces a large improvement on its own weekly word-recognition quizzes.

A minister is encouraged to announce a national expansion.

A proper synthesis asks whether the quiz was independently valid, whether students improved on different reading tasks, whether comprehension changed, and whether the result persisted beyond the rehearsed material.

It also examines which teachers delivered the programme, which students were excluded, and how much instructional time the method consumed compared with the existing alternative.

Suppose the evidence supports improvements in a specific word-level skill but says little about general comprehension.

The appropriate policy could be a targeted pilot with independent outcome measures and qualified teacher preparation. It would be misleading to advertise the programme as a proven cure for every literacy gap.

Good research narrows a claim to what the evidence can actually carry.

11. A worked case: an AI tutor pilot with hidden costs

A second hypothetical study reports that students using an AI homework system completed more questions than students in a comparison class.

The research team should ask whether independent understanding improved, how often answers were inaccurate, what teacher checking was required and whether students had comparable access to devices.

If the system improved task completion but created extra correction work or greater dependence on worked examples, the policy judgement is not straightforward.

Review data privacy, licensing, safeguarding and permitted assessment use as independent approval gates.

The result may justify a redesign, a narrower intervention or continued evaluation rather than national deployment.

An educational tool should be judged by capability and legitimate service outcomes, not engagement counts alone.

12. Examine conflicts of interest and research funding

A commercial curriculum provider may commission evidence about its own product. A ministry may evaluate a politically visible reform. A university researcher may have legitimate career incentives connected to publication.

Conflicts do not automatically invalidate research. They need declaration and appropriate management.

A review should identify:

funder and commissioning institution;
relevant author or evaluator interests;
control over design, data, analysis and publication;
access to independent review;
whether negative results may be reported;
contractual rights and intellectual property;
permissible use of participant data;
the authority responsible for policy adoption.

Where the public consequences are material, a suitable independent evaluation or replication may be justified.

The research should be able to report an inconvenient result without being silently suppressed.

13. Make local context part of the evidence judgment

An intervention developed in one jurisdiction may depend on language, teacher preparation, class size, facilities, cultural context or official curriculum structures that differ elsewhere.

The policy team should ask:

What is the mechanism?

Which prerequisites make it work?

Which school environments were studied?

What would change in our schools?

Can we preserve its essential components?

What evidence would show a mismatch?

A cross-country study can provide useful direction, but copying a programme without its staffing or learning prerequisites is not evidence-informed policy.

The OECD and UNESCO both emphasise stronger links between research, local institutional capacity and decision-making.

14. Give teachers and learners a legitimate route into research

Teachers often see gaps before a national statistical report does. Students can describe experiences that administrative records miss.

An education research service should have a proportionate route for research questions from schools and communities.

Educators can provide anonymised descriptions, collaborate on appropriate studies and interpret practical constraints. Learner participation should follow age-appropriate ethical and safeguarding requirements.

Do not transfer unpaid research responsibilities to teachers without time or institutional support.

A useful participatory process produces better questions and implementation evidence while retaining proper professional and ethical authority.

15. Track corrections, retractions and changing conclusions

A published study may later be corrected, challenged or retracted. A policy based on that study may then require review.

A research source registry should record:

Author, title, publisher and date.
DOI or other stable identifier where available.
Research design and relevant population.
Outcome measures and source data access conditions.
Review or publication status.
Known corrections, expressions of concern or retractions.
Important limitations and replication findings.
Applicable curriculum or education-policy relevance.
Licensing and permitted use.
Last review and policy owner.

A citation manager or automated retrieval system must not treat a DOI alone as proof that an article remains valid.

When a supporting claim changes materially, the institution should trace which recommendations, teaching resources or decisions depend on it.

16. Translate research into an actionable policy brief

Policy staff need more than a long bibliography.

A useful brief contains:

Decision: What problem must the competent authority address?

Evidence: Which findings are most relevant and why?

Confidence: What is known, what is uncertain and what depends on context?

Mechanism: How is the proposed action expected to work?

Alternatives: Which other feasible responses exist?

Ethics and equity: What protections or distributional effects matter?

Capacity: What teachers, materials, time and finance would be required?

Decision rights: Who may approve, defer or refuse the proposal?

Return: Which observations would confirm benefit or require correction?

The brief must distinguish research conclusions from value judgments or legal duties. A technically effective intervention is not automatically the fairest or most affordable national policy.

17. Build a research-to-decision register

An institutional ledger can include:

research question and public purpose;
ethics and legal approval status;
responsible institutions and declared interests;
design, participant protections and data rights;
primary findings and uncertainty;
relevant evidence synthesis;
local applicability and limitations;
policy alternatives and resource implications;
competent decision authority;
approved, rejected or deferred disposition;
delivery and monitoring owner;
source corrections and scheduled review.

The system should preserve unfavourable, null and mixed findings appropriately. Deleting them can cause future schools to repeat avoidable failures.

18. A 90-day research-readiness pilot

Days 1–30 — Define one real policy question. Identify the problem, relevant learners, existing evidence, permissible data and competent ethics review requirements. Check whether new human-participant research is necessary at all.

Days 31–60 — Build a structured evidence packet. Search and appraise relevant studies, compare contexts and document conflicts or unresolved questions. Where authorised, design a bounded new study or implementation test.

Days 61–90 — Make a provisional evidence decision. Produce a transparent synthesis and policy brief. Decide whether to commission further work, test a practical change, maintain the status quo or abandon an unsupported idea.

This pilot does not replace a formal ethics review or establish long-term educational effects. It tests whether the ministry can handle evidence responsibly before policy scale.

19. Common research-system failures

Research questions chosen for publicity rather than educational need: Begin with a real decision.

A service audit treated as permission for unrestricted experimentation: Verify legal and ethics requirements.

Children pressured to participate because the school is involved: Protect legitimate consent or assent and participation rights.

Sensitive records transferred without a clear purpose: Apply lawful minimisation and access control.

One positive study treated as a national mandate: Synthesise the relevant evidence and context.

Outcome measures chosen after results are known: Document original plans and changes.

Commercial funding hidden: Declare and manage conflicts.

A statistically significant result described as universally important: Examine practical size, uncertainty and transfer.

Corrected or retracted findings remain in teaching guidance: Use a source registry and dependency review.

Research conclusions never return to schools: Connect findings to a legitimate decision and implementation check.

The institutional standard

Evidence-informed education policy is not a contest to find research supporting a decision already made. It is a disciplined way to learn from uncertainty while respecting the people whose lives and education produce the evidence.

A Ministry of Education V2.0 succeeds when research is ethically gathered, critically synthesised, honestly interpreted and connected to decisions that can be revisited when the evidence or real world changes.

Sources and connected reading

Editorial scope: Independent policy and research governance reference. Every real investigation must follow applicable law, qualified ethics oversight, data-protection requirements and authorised institutional procedures.

Ministry of Education V2.0 Reference 56 | How to Govern Education Research Ethics and Evidence Synthesis