eduKate Singapore

Research Data Management and FAIR Principles | From Research Question to Reusable, Traceable Evidence

Research Data Management and FAIR Principles

Research data management is the deliberate stewardship of data across the research lifecycle so that evidence remains understandable, secure, traceable, reproducible where appropriate and reusable under legitimate conditions. The FAIR Guiding Principles strengthen that stewardship by making digital resources more Findable, Accessible, Interoperable and Reusable for both people and machines.

Research data becomes durable evidence when a future receiver can identify it, understand how it was produced, determine whether it is appropriate to use and trace it back to the research process that created it.

Research data is not simply the final spreadsheet attached to a paper. It can include observations, measurements, images, recordings, survey responses, interview transcripts, code, simulation outputs, laboratory instrument files, field notes, annotations, derived datasets and documentation. Managing that evidence begins before collection and continues after publication.

ARTICLE ID: DATA.MANAGEMENT.020
Canonical function: research stewardship, reproducibility and responsible reuse
Series route: The Data Lifecycle → Research Data Management and FAIR Principles.

Research Data Management Starts Before Data Exists

Weak data management often begins when collection starts before the research team has decided how files will be named, how participants or samples will be identified, which metadata will be recorded, where sensitive data will be stored or how versions will be distinguished.

A stronger route begins with a plan:

Question → Plan → Collect → Describe → Validate → Store → Analyse → Preserve → Share → Reuse → Cite → Correct

The purpose of planning is not bureaucracy. It is to prevent future ambiguity while decisions are still cheap to change.

What Counts as Research Data?

The answer depends on the discipline and research question. Research data can include:

The management question is not only “what files do we have?” but “which digital objects are required to understand, verify or appropriately reuse the research?”

The Data Management Plan

A data management plan can define:

A good plan evolves with the project. It should be updated when methods, instruments, collaborators or sharing conditions change.

FAIR: Findable, Accessible, Interoperable and Reusable

The FAIR Guiding Principles provide a framework for improving the ability of humans and machines to discover and reuse digital resources. They were introduced in the 2016 paper “The FAIR Guiding Principles for scientific data management and stewardship” and remain a central reference for data stewardship.

FAIR is not a single software product, repository or checklist. It is a set of principles that guide implementation choices.

F — Findable

Data cannot be reused if nobody can discover that it exists.

Findability is strengthened by:

Findability is a receiver problem. Metadata should use enough generic and domain-specific description that a future researcher—or a machine—can locate the resource without already knowing its exact filename.

Persistent Identifiers

A persistent identifier gives a research object a stable identity even when storage locations or display names change. Digital Object Identifiers are widely used for publications and datasets, while other persistent identifier systems serve researchers, organisations and research objects.

The identifier is an anchor. Metadata, citations, versions and provenance can attach to that anchor through time.

Rich Metadata

Metadata helps future receivers decide whether a dataset is relevant and trustworthy.

Research metadata may include:

Metadata richness should follow what a legitimate future receiver needs to assess suitability for reuse.

A — Accessible

Accessibility means that once a resource is found, the route for retrieving it—or determining the conditions under which it may be retrieved—is clear and standardised.

FAIR does not mean that all data must be open. The FAIR principles explicitly allow authentication and authorisation where necessary. Sensitive, confidential, commercial or ethically restricted data can still be FAIR when metadata and access conditions are well described and legitimate access is handled predictably.

This distinction is fundamental: findable and appropriately accessible is not the same as publicly downloadable.

Metadata Can Outlive Data Access

FAIR encourages metadata to remain accessible even when the data itself is no longer available. This preserves the scholarly record that a resource existed, what it represented and why access changed.

That principle aligns strongly with archival thinking: absence should not erase provenance.

I — Interoperable

Interoperability means data and metadata use representations, vocabularies and relationships that allow them to work with other data, tools and workflows.

Interoperability can be strengthened through:

Interoperability does not require every discipline to become identical. It requires enough explicit structure that differences can be translated without semantic collapse.

Community Standards

Reuse improves when researchers follow relevant disciplinary standards for metadata, minimum information and data representation. Community standards capture knowledge that a generic repository cannot invent automatically.

The correct standard depends on the research community and data type. FAIR is therefore implemented through domain-aware choices rather than one universal schema.

R — Reusable

Reusability means future receivers have enough context to judge whether data can support another legitimate purpose.

Reusable resources benefit from:

Reusability does not mean every dataset is suitable for every question. It means the receiver has enough evidence to make that judgement.

Provenance

Research provenance records how a resource was generated, by whom, from which source material, under which methods and conditions, and which transformations were applied.

A useful route is:

Question → Method → Instrument or Source → Raw Data → Processing → Derived Data → Analysis → Figure or Result → Publication

When that route survives, future researchers can inspect how a result came to exist.

Raw Data

“Raw” does not always mean untouched reality. Instrument files may already contain internal processing. Survey exports may encode platform decisions. Digitised material may depend on scanning and transcription.

The stronger principle is to preserve the earliest practical research representation together with enough metadata to explain how it was produced.

Derived Data

Derived data is produced from earlier data through cleaning, filtering, normalisation, coding, aggregation or modelling.

Derived outputs should remain linked to source versions and transformation logic. Otherwise a clean final table can become disconnected from the evidence it claims to summarise.

Version Control

Research data changes. Errors are corrected, annotations improve, code evolves and classifications are revised.

Versioning should preserve:

Correcting data should improve current knowledge without making published history impossible to reproduce.

File Naming and Organisation

File naming is a small decision with large cumulative effects. Names should distinguish object, date or version without relying on personal memory.

Folders and filenames are not substitutes for metadata, but clear conventions reduce local confusion during active research.

Open Formats and Preservation

Long-term preservation benefits from well-documented, widely supported formats where practical. Proprietary formats may still be necessary during active research, but preservation planning should consider whether future receivers will be able to open and interpret the files.

Format choice is therefore part of the data lifecycle.

Repositories

A research data repository provides infrastructure for deposit, metadata, identifiers, preservation and access. Suitable repositories may be institutional, disciplinary or general-purpose depending on the research context.

A repository is not merely storage. A trustworthy repository should help preserve identity, metadata, access conditions and long-term stewardship.

Repository Selection

Useful selection questions include:

Data Citation

Data citation gives research datasets visible scholarly identity. A citation can credit creators, identify the exact resource and version used, and allow future readers to follow the evidence route.

Good citation practice treats data as a research output with provenance, not as anonymous background material.

Licensing and Reuse Conditions

A future receiver should know whether and how a dataset may be reused. Clear licences or access conditions reduce ambiguity.

Licensing does not override privacy, consent, confidentiality or other obligations. A licence is one layer of the governance stack.

Sensitive Research Data

Research may involve personal, confidential, culturally sensitive, commercially sensitive or otherwise restricted data.

Responsible management can include:

FAIR can coexist with these controls because accessibility can be conditional and authenticated.

FAIR Is Not the Same as Open

This deserves repetition. Open data focuses on unrestricted access and reuse under stated terms. FAIR focuses on discoverability, access routes, interoperability and reuse context. Data can be both FAIR and open, FAIR but restricted, or open but poorly FAIR if it lacks identifiers, metadata and interoperable description.

FAIR Is Not Automatically Ethical

A dataset can satisfy technical FAIR principles and still raise ethical questions about consent, power, community rights, harm or inappropriate reuse.

FAIR improves the stewardship and reuse pathway. Ethical governance still determines what responsible reuse means.

See Data Ethics and Responsible Use.

FAIR and Machine Actionability

FAIR places strong emphasis on machine-readable metadata because modern discovery, integration and reuse increasingly involve automated agents and computational workflows.

Machine actionability requires more than putting a PDF description next to a dataset. Identifiers, vocabularies, relationships and access routes should be represented predictably enough for software to interpret them.

Research Data and Software

Software can be essential research evidence. A dataset may be impossible to interpret or reproduce without the code that generated, cleaned or analysed it.

Research stewardship should therefore consider source code, dependencies, environments, parameters and workflow definitions where they materially affect results.

Computational Environments

Reproducibility may depend on software versions, libraries, operating systems and hardware assumptions. Environment files, containers or documented dependencies can reduce future ambiguity.

The goal is proportional preservation: capture enough of the computational environment to reproduce the job that matters.

Reproducibility

Reproducibility asks whether another authorised researcher can regenerate a result from the relevant data, code, parameters and methods.

Not every research process can be reproduced exactly. Field conditions change, unique historical events do not repeat and some source material is restricted. Good stewardship makes the limits explicit rather than pretending complete repeatability.

Replicability

Replicability often concerns whether an independent study using new data or observations can test a similar claim. Data management supports replication by making original methods, variables and analytical choices understandable enough to compare.

Quality and Research Data

Research data quality is inseparable from method. A technically valid table can still be poor evidence if sampling, calibration, measurement or annotation was weak.

Useful quality metadata can record:

See Data Quality.

Research Data Lineage

Lineage connects raw observations to derived data, analyses, figures and publications. It allows a suspicious result to be traced backward and a corrected source to be propagated forward.

See Metadata and Data Lineage.

Research Data Security

Security controls should reflect sensitivity and research need. Strong research security protects confidentiality and integrity without preventing legitimate analysis.

See Data Security and Privacy.

Research Data Backup and Recovery

Irreplaceable observations deserve particular protection. A processed table may be reconstructable from raw data and code; an original field observation or unique interview may not be.

Backup priority should therefore reflect reconstructability, consequence and research value.

Active Data vs Archival Data

Active research data changes frequently and needs collaboration, version control and operational backup. Archival data is stabilised for long-term preservation and future access.

The transition from active to archival state should preserve documentation, identifiers, final versions and provenance.

Retention and Disposal

Not every research file should be kept forever. Retention decisions should consider reproducibility, consent, policy, funder or institutional requirements, privacy, preservation value and cost.

Disposal should be documented so later receivers understand why data is unavailable.

Research Collaboration

Collaborative projects should clarify responsibilities early: who owns source data, who can modify shared datasets, which repository is canonical, how versions are merged, and what happens when a collaborator leaves.

Many research data failures are governance failures disguised as file-management failures.

Data Stewardship

Data stewards help connect research practice with metadata, repositories, standards, quality, preservation and governance. Their role is not simply administrative. Good stewardship protects the future interpretability of evidence.

Researcher Identity and Contribution

Persistent researcher identifiers and contribution metadata can improve attribution. Research data often has contributors whose work does not fit traditional authorship neatly: collectors, curators, software developers, data stewards and annotators.

Good metadata can make those contributions visible.

Negative and Null Results

Research stewardship should not preserve only successful outputs. Negative, null or inconclusive results can prevent repeated work and help future researchers understand the evidence landscape.

Whether such data can be shared depends on context, but absence of a desired result is still part of research memory.

Data Reuse

Reuse creates new research possibilities, but reuse should respect the original data-generating context.

A future researcher should ask:

FAIR Evaluation Without Score Theatre

FAIR assessment can be useful, but one composite score can hide important weaknesses. A resource may be highly findable but poorly reusable because provenance is weak. Another may be well described but inaccessible under an unclear process.

Assessment should guide improvement rather than produce a decorative badge.

Worked Example: Experimental Research

A laboratory experiment produces instrument files, calibration records, sample identifiers, processing code and derived figures.

A strong research data route preserves:

The figure in the paper now has a return path to the observation.

Worked Example: Survey Research

A survey project may hold direct identifiers, contact records, response data, codebooks and analysis files. Responsible management can separate identity from analysis, document sampling and missingness, preserve questionnaire versions and deposit appropriately de-identified or controlled-access data with clear metadata.

FAIRness is achieved through discoverability and a legitimate access route—not by publishing sensitive raw responses openly.

Worked Example: Computational Research

A simulation study may generate terabytes of output that are expensive to preserve indefinitely. Researchers may preserve source code, configuration, random seeds where relevant, representative or final outputs, metadata and the recipe needed to regenerate intermediate states.

Good stewardship is not always “keep every byte”. It is “preserve enough evidence and capability to reproduce the scientific job that matters”.

Worked Example: AI Research

AI research may depend on training datasets, preprocessing, labels, model checkpoints, evaluation sets and prompts. Reproducibility requires versioning across the stack.

Datasets should preserve provenance and usage conditions. Generated or synthetic data should be distinguishable from observed data. Evaluation sets should be protected from accidental training leakage.

Common Failure Modes

A Research Data Management Checklist

  1. What data and research objects will be created?
  2. Which are irreplaceable?
  3. What identifiers will be used?
  4. Which metadata and community standards apply?
  5. How will raw and derived data remain linked?
  6. How will versions be distinguished?
  7. How will code and computational environments be preserved where needed?
  8. What quality and uncertainty metadata must travel with the data?
  9. Where will active data be stored and backed up?
  10. Which data is sensitive or restricted?
  11. What access route is legitimate?
  12. Which repository is appropriate?
  13. Will the resource receive a persistent identifier?
  14. What licence or reuse conditions apply?
  15. What should be preserved, retained or disposed of?
  16. Can a future receiver reproduce the evidence route?

A FAIR Readiness Checklist

  1. Findable: Does the resource have a persistent identifier and rich searchable metadata?
  2. Accessible: Is the retrieval or access-request route clearly described using appropriate protocols?
  3. Interoperable: Are formats, vocabularies and relationships understandable across relevant systems and communities?
  4. Reusable: Are provenance, licence or access conditions, quality, methods and community standards sufficient for a future receiver to judge reuse?

A Maturity Ladder

  1. Stored: research files are retained somewhere.
  2. Organised: naming, versions and responsibilities are defined.
  3. Described: metadata explains content and method.
  4. Traceable: raw data, transformations and outputs are linked.
  5. Protected: backup, access and sensitive-data controls are deliberate.
  6. Preserved: stable versions are deposited with durable identifiers.
  7. FAIR: resources are meaningfully findable, appropriately accessible, interoperable and reusable.
  8. Adaptive: reuse, correction and new standards improve future stewardship.

Further Reading

The Deeper Principle: Preserve the Route from Claim Back to Evidence

Research is cumulative. Every new claim enters a world of earlier observations, methods, datasets and interpretations. Data management protects that chain by keeping evidence identifiable, contextualised and recoverable.

FAIR strengthens the handoff to future receivers. It does not demand that every dataset be open or every byte be preserved. It demands that digital resources be managed well enough that people and machines can find them, understand the access route, interpret them across systems and judge whether they can be responsibly reused.

Data Management Series

Final idea: research data management protects the continuity of evidence. FAIR makes that evidence easier for future people and machines to find, access appropriately, connect and reuse without severing it from the provenance, responsibility and context that make research trustworthy.

eduKate Asia

Subscribe to
our newsletter

Discover more from eduKate Singapore

Subscribe now to keep reading and get access to the full archive.

Continue reading