The gold standard of research services is not a report that proves what a client already believes. It is a methodical service that asks a meaningful question, collects suitable evidence, handles people and data responsibly, and reports what the results can—and cannot—establish.
This guide follows a fictional education programme trying to evaluate a new science tutorial. The case moves from an initially biased commission to a credible question, appropriate research design, transparent limitations and a decision the organisation can actually use.
The sector perspective complements The Gold Standard of Research, The Gold Standard of Data Analytics and Clinical Research. These domains overlap in integrity but differ in their methodologies, risks and legal requirements.
Find the part you need
Choose a chapter group or follow the complete research journey. Each section returns to the index.
Chapter index
Ask the right question · Chapters 1–4
Design and recruit · Chapters 5–8
Protect data and inference · Chapters 9–12
Report and act · Chapters 13–16
1. Research begins with a decision someone cannot yet make
A fictional education organisation wants to know whether a new science tutorial improves understanding. Its director asks a research company to demonstrate that the new format is more effective. That request contains a hidden conclusion before any evidence has been collected. A responsible researcher reframes it: which learning outcomes matter, for which learners, relative to which alternative, and how could those outcomes be measured?
The gold standard of a research service is not a glossy report that supports the client’s preferred story. It is a defensible method that produces evidence appropriate to a real question, respects the people and information involved, and describes uncertainty honestly. The organisation and tutorial are invented. This worked example is designed to expose the difference between finding an answer and commissioning applause.
Back to chapters · Next chapter
2. Identify the type of research and its applicable rules
Research services include market studies, public-opinion surveys, social research, scientific experiments, archival investigation, programme evaluation and commissioned technical analysis. The goals and ethical requirements differ. A market-research interview is not a clinical trial, and routine service analytics do not automatically become a scientific experiment because someone labels a dashboard ‘research’.
The recently published ISO 20252:2026 specifies vocabulary and service requirements for market, opinion and social research, including insights and data analytics. It replaces the withdrawn 2019 edition within that scope; it does not govern every laboratory or medical study. The OECD Frascati Manual 2015 defines research and experimental development for statistical purposes. Specify the actual research category before borrowing standards from another field.
Back to chapters · Next chapter
3. Write a question that can be answered
A useful research question identifies a population, subject, outcome, comparator and timeframe where those elements are relevant. ‘Does the tutorial help?’ is vague. ‘How does performance on unfamiliar science explanations change after four weeks of the new tutorial, compared with a credible alternative?’ is closer to a question that can guide design.
Researchers should distinguish exploratory questions from pre-specified tests. Exploratory work can generate ideas without pretending that every discovered association was predicted in advance. The protocol should state what is being investigated, which conclusions the design could plausibly support, and what the client will do if the evidence is inconclusive. A sound question also includes the possibility of a result nobody hoped to see.
Back to chapters · Next chapter
4. Separate methodology from sales commitments
A research firm can agree to a timetable, scope, quality process and deliverables; it cannot responsibly promise what the data will reveal. If the contract pays only when a particular conclusion is reached, the incentive conflicts with the purpose of independent inquiry. The service agreement should disclose the sponsor’s role and the research team’s authority over method and truthful reporting.
Clarify ownership of instruments, anonymised outputs where applicable, raw data access, intellectual-property rights, confidentiality and publication arrangements. Identify who can request a change to the method and how such changes will be recorded. The research organisation should not silently revise sample selection because the preliminary results disappoint the buyer.
Back to chapters · Next chapter
5. Choose a design that matches the strength of the claim
Different methods answer different questions. Interviews can reveal how learners experience a lesson; structured observation can reveal what teachers and students actually do; surveys can estimate views with suitable sampling; well-designed comparisons may support cautious claims about effectiveness. No method is automatically superior without a question and context.
If the education organisation only compares this term’s top-performing volunteers with last year’s full cohort, a higher score cannot be attributed safely to the new tutorial. The groups may differ before teaching begins. A research plan should consider appropriate comparison methods, baseline differences and plausible competing explanations. When causal identification is weak, use language that reflects association or experience rather than claiming proof of impact.
Back to chapters · Next chapter
6. Sample people who represent the question
A sample should be designed around the intended population, not merely around who is easiest to reach. Parents who reply quickly to an online survey may have different circumstances from families who need telephone help; advanced students may volunteer more readily for optional enrichment. These selection effects can distort conclusions.
Describe recruitment criteria, coverage, exclusions, sample size reasoning, nonresponse and the limits of generalisation. A small qualitative sample can be valuable for discovering problems, while it may be inappropriate for estimating a population percentage precisely. More responses do not automatically repair systematic bias. Show who was absent and why their absence might matter.
Back to chapters · Next chapter
7. Treat consent and safeguarding as substantive work
People involved in research should understand the purpose, participation choices, foreseeable burdens and privacy arrangements in language appropriate to them. Additional protections may be necessary for children, people in dependent relationships and other contexts involving heightened vulnerability. Research teams must use applicable ethics review, institutional requirements and data-protection rules.
The World Medical Association Declaration of Helsinki, revised 2024, addresses medical research involving human participants and relevant identifiable material or data. It is not a general approval for school surveys or commercial studies; those require their own appropriate oversight. In the fictional tutorial study, families should not feel that refusing a research questionnaire jeopardises their child’s education.
Back to chapters · Next chapter
8. Build instruments that measure what matters
A test, interview guide or questionnaire can look professional while failing to capture the intended construct. If the purpose is transferable scientific explanation, a quiz of memorised vocabulary may produce an attractive score without demonstrating the desired capability. A good instrument links each question to the learning objective and checks clarity and difficulty.
Pilot prompts with people similar to the target audience, then document revisions. Beware of leading wording such as ‘How much more confident are you after our excellent programme?’ Neutral questions, suitable response choices and alternatives for those who cannot answer can reduce avoidable distortion. In observations, define what counts as the behaviour rather than leave coding to each researcher’s intuition.
Back to chapters · Next chapter
9. Plan data quality before collecting records
Research data can fail through duplicate entries, missing values, inconsistent units, wrong dates, incomplete responses and faulty links between records. Design validation rules, access controls and reconciliation processes while the dataset is small. Maintain enough information to understand each variable, source and transformation.
Suppose one tutorial site records assessment marks out of 20 while another records percentages. Combining them as if the units were identical could create a false improvement. A well-managed service catches the mismatch, documents how records are harmonised and leaves original values auditable. Cleaning data should not mean deleting inconvenient observations without a defensible rule.
Back to chapters · Next chapter
10. Protect personal and commercially sensitive information
Research projects may include opinions, student records, interviews, business data or personal identifiers. Collect only what is necessary for the approved purpose, control access, restrict exports and plan retention, transfer and secure disposal. Explain what de-identification can and cannot achieve, especially in small groups where descriptions might reveal an individual even without a name.
The research report should not display a rare learner’s detailed quote next to enough background facts to make identification obvious. The team can paraphrase or aggregate findings where appropriate without fabricating the substance. Public release, sponsor access and re-use of the dataset may require different permissions and protections. Privacy is part of trustworthy evidence, not a barrier that can be ignored for a compelling story.
Back to chapters · Next chapter
11. Analyse comparisons without hiding uncertainty
Researchers should select analyses that fit the design and measurements. Report denominators, baseline differences, missingness, appropriate variability and uncertainty. Averages can conceal important variation; changes in which learners completed follow-up can mislead. Testing many relationships and presenting only a favourable one creates an unreliable impression.
For the fictional tutorial study, an average score improvement may reflect practice with a familiar test rather than deeper understanding. Compare new tasks when appropriate, inspect who withdrew, and explore competing explanations. A result can be useful even without a dramatic statistical finding: it may show which part of the teaching process needs improvement or why a larger study would be required.
Back to chapters · Next chapter
12. Make results reproducible within ethical boundaries
Another competent researcher should be able to understand how the work was performed. Preserve versions of instruments, sampling decisions, analysis code where used, transformation rules and a clear record of amendments. Restricted data may not be publicly sharable, but controlled reproducibility and independent checks remain possible through appropriate governance.
The eduKate Scientific Reproducibility guide explains this idea in education. A research service should identify where verification was performed and where it was not. Copying the same chart into three reports is not replication; applying the documented method to evidence and checking the result is closer to an independent test.
Back to chapters · Next chapter
13. Use AI tools without inventing evidence
Automation and generative AI can help draft questions, organise transcripts, search literature or produce analytic code, but these tools can also invent citations, collapse distinctions or misclassify responses. Researchers remain accountable for verifying sources and interpretations. Confidential data should not be placed in unapproved services merely because a convenient tool can summarise it.
Maintain a review process for AI-assisted steps that could materially change results. Where appropriate, document which tools and versions were used, what data was supplied and how outputs were checked. An AI-generated thematic summary should be compared with actual transcript extracts, including contradictory cases. Speed is valuable only when it does not outrun the organisation’s ability to establish what the evidence supports.
Back to chapters · Next chapter
14. Report negative, inconclusive and mixed findings
Research reports should distinguish findings, limitations, interpretations and recommendations. If the tutorial helps one type of explanation but not another, report both. If the sample is too small to distinguish a useful effect from noise, do not turn uncertainty into either a triumphant claim or a declaration that the intervention cannot work.
Disclose funding, relevant conflicts and methodological changes. Use charts with meaningful baselines and accessible explanations. The 2024 Declaration of Helsinki places particular emphasis on accurate and complete reporting, including negative and inconclusive results, in medical human-participant research. Outside that domain, the same broad discipline of nonselective reporting helps clients avoid decisions driven by convenient omissions.
Back to chapters · Next chapter
15. Turn evidence into decisions without selling certainty
The client’s decision may be whether to expand a programme, revise it, conduct a better trial or stop an ineffective approach. Research recommendations should connect the strength of evidence with the cost and risk of the proposed action. A low-risk change in teaching materials may be worth testing even when certainty is limited; a costly system-wide claim may require stronger support.
Offer a decision table that states what was observed, what is unknown, what alternative explanations remain and what next action would reduce uncertainty. Do not blur the boundary between a research conclusion and a commercial recommendation. A client should leave able to explain what the investigation adds to existing knowledge, not simply what the researcher wants to sell next.
Back to chapters · Next chapter
16. Judge the research service by transparent usefulness
Return to the fictional tutorial study. The responsible firm may find some encouraging improvements, weaker evidence in other areas, and important sampling limitations. The final report includes the instrument, analysis method, limitations, data governance and a feasible follow-up test. The client has learned how to make the programme better without pretending every learner has already benefited.
The gold standard is research that people can inspect, challenge and use responsibly. Sound questions, ethical recruitment, careful measurement, truthful analysis and transparent limitations are not administrative extras. They are what makes a research service trustworthy even when its findings are inconvenient.
Worked example: a confident claim versus credible evidence
These observations are fictional and illustrative. The table shows why research should separate an attractive result from the conclusion it can validly support.
| What the client notices | What remains uncertain | Better research action |
|---|---|---|
| Volunteers score higher | They may have been more prepared before the programme | Measure baselines and compare appropriate groups |
| Average rises by six points | Test familiarity, missing learners and variation | Use appropriate new tasks and disclose denominators |
| Parents report satisfaction | Satisfaction is not identical to learning transfer | Separate experience measures from capability measures |
| One subgroup shows a strong benefit | Small numbers and multiple comparisons | Report uncertainty and plan independent verification |
| Client wants to advertise success | Findings may not support a universal claim | Approve bounded wording or conduct stronger evaluation |
A responsible research-service checklist
- The question permits an honest negative or inconclusive answer.
- Research type, methodology, sampling and limitations are declared clearly.
- Participants are recruited ethically under applicable oversight.
- Measures represent the intended construct and are piloted.
- Data sources, units, cleaning, versions and access are controlled.
- Analyses preserve meaningful uncertainty and alternative explanations.
- All material findings, including unfavourable results, are reported.
- Client decisions are matched to the strength and boundaries of the evidence.
Current source standards and eduKate references
- ISO 20252:2026 — Market, Opinion and Social Research Service Requirements
- OECD — Frascati Manual 2015
- World Medical Association — Declaration of Helsinki (2024)
- eduKate — Research
- eduKate — Data Analytics
- eduKate — Scientific Reproducibility
- The Gold Standard | EduKate Singapore
Standards listed have different scopes. ISO 20252:2026 concerns specified market, opinion and social research services; the Declaration of Helsinki concerns medical research with human participants. Research with children, regulated medical data or other sensitive circumstances requires its own proper ethics, legal and professional safeguards.