Guides for Researchers

How Do I Prepare My Data for Sharing?

Selecting research data and documentation for long-term preservation

BACK TO GUIDES

You Cannot Keep Everything

As a project progresses, large numbers of files are often created: raw data, processed datasets, analysis scripts, draft figures, temporary files, meeting notes, duplicate copies and intermediate results. By the end of a project, deciding what should be preserved and shared can be challenging.

Preservation does not mean keeping every file forever. The aim is to identify the research outputs that are needed to understand, verify, reproduce or build upon the research in the future.

Preparing data for preservation and sharing involves more than simply uploading files to a repository. You do not know who might benefit from your research data in the future. To be useful, shared data should therefore be accompanied by sufficient context to explain what the data represent, how they were collected and processed, and any conditions governing their reuse.

A useful question is not "What can I keep?" but rather "What would someone need to understand or reuse this research in five or ten years?"

A Simple Way to Think About Preservation Decisions

In most projects, decisions about preservation could be grouped into three categories:

Keep	Consider	Discard
Files required to understand, verify or reuse the research.	Files that may have disciplinary, historical or future value.	Temporary, duplicate or superseded files with no continuing value.

What Should Usually Be Kept

Type of material	Examples	Why it matters
Final research data	Final datasets used to support publications or conclusions.	Allows verification and reuse.
Documentation	README files, codebooks, protocols, data dictionaries.	Makes the data understandable.
Analysis code and workflows	Scripts, notebooks, pipelines.	Supports reproducibility.
Consent and ethics documentation	Where retention is permitted and required.	Documents how data were collected and managed.
Metadata	Repository descriptions and structured information.	Improves findability and reuse.

A useful rule of thumb is that preservation decisions should focus on future value. If removing a file would make it difficult for someone to understand, verify or build upon the research, it is usually worth keeping. If the file no longer serves a clear purpose and has no foreseeable value, it may be a candidate for disposal.

Materials That May Require Further Consideration

Example	Why it may be worth keeping
Intermediate datasets used to generate published results	May help explain how final results were produced.
Laboratory notebooks, field notes or research records	May provide important context for interpreting the research.
Materials subject to retention requirements	May need to be preserved to comply with institutional, legal or disciplinary obligations.
Teaching, training or secondary-analysis materials	May support future educational or research activities.

What Might Not Need To Be Kept?

Example
Temporary working files
Duplicate copies
Intermediate exports that can be recreated
Superseded versions
Software installation files
Cached or automatically generated files

The precise boundary between these categories will vary between disciplines, institutions and projects. When in doubt, it is often safer to retain materials temporarily and seek guidance from a data steward, archivist or institutional support service before disposal.

Tip: The fact that a file can be recreated does not always mean it should be discarded. If recreating a file would require substantial time, expertise or resources, preserving it may still be worthwhile.

It Is Not Just About The Data

One of the most common reasons research data become unusable is not the loss of the data themselves, but the loss of context.

A spreadsheet containing hundreds of variables may remain perfectly readable years later, yet be impossible to interpret if no documentation explains what the variables represent, how the data were collected, or how values were coded.

For this reason, preserving documentation is often as important as preserving the research data themselves.

Many repositories also ask researchers to provide descriptive information when depositing data. This information helps others discover the dataset, understand its contents and determine whether it is relevant to their needs.

Before sharing data, it can be useful to consider whether someone unfamiliar with the project would be able to answer the following questions:

Question	Examples of supporting information
What do these data represent?	Dataset descriptions, abstracts, README files.
How were the data collected?	Protocols, methods sections, laboratory or field procedures.
What do the variables, codes or abbreviations mean?	Data dictionaries, codebooks and variable descriptions.
How were the data processed or analysed?	Analysis scripts, workflows, software documentation.
Under what conditions can the data be reused?	Licences, access conditions and repository information.
How is this dataset related to other research outputs?	Publications, software, protocols and project information.

A useful rule of thumb is that if a future researcher would need to contact you to understand the dataset, additional documentation may be needed. The goal is not to document everything, but to provide enough context for others to understand what the data represent and how they can be used.

Follow Community Standards

Research communities often develop shared ways of collecting, organising, describing and exchanging data. These practices, sometimes referred to as community standards, help researchers within a discipline understand each other's data more easily and make it easier to combine information from different studies.

One common misconception is that every research project needs to develop its own way of organising and describing data. In reality, many disciplines have already established conventions that researchers can adopt and adapt. Using existing standards not only saves time, but also increases the likelihood that data can be understood and reused by others.

Community standards can take many forms. In some fields, they may specify how data should be organised or which information should accompany a dataset. In others, they may define standard vocabularies, metadata schemas, reporting guidelines or preferred repositories.

Using recognised standards is not usually about compliance for its own sake. Rather, it helps ensure that others can interpret and reuse data without having to guess what variables mean, how measurements were recorded or which conventions were followed.

Type of standard	Example
Metadata standards	Agreed information used to describe datasets within a discipline.
Controlled vocabularies	Standardised terms used to describe concepts, species, locations or research topics.
Reporting guidelines	Agreed information that should accompany published research.
File organisation conventions	Common ways of structuring files and folders within a research community.
Data formats and schemas	Standard structures that allow data to be exchanged between systems and researchers.

Following community standards can make data more interoperable and reusable. It can also reduce the amount of explanation needed because future users may already be familiar with the conventions used within the discipline.

Example: Researchers sharing biodiversity data often use standards such as Darwin Core, which provides a common way of describing species observations. Because many organisations use the same standard, datasets from different sources can be more easily combined and compared.

Researchers do not need to become experts in every standard used by their community. However, before sharing data, it is often worth checking whether funders, journals, repositories or professional societies recommend particular standards or practices.

Community standards also evolve over time. What matters most is not necessarily choosing the most sophisticated standard available, but adopting approaches that are widely recognised and supported within the relevant research community.

Finding relevant standards can sometimes be challenging, particularly in interdisciplinary fields. A useful starting point is often the repository where the data will eventually be shared, as many repositories recommend or require specific standards. Funders, journals, professional societies and research infrastructures may also provide guidance on disciplinary practices and expectations.

Researchers looking for disciplinary standards may find the following resources helpful:

Resource	What it provides
FAIRsharing	A curated catalogue of data standards, databases and policies across research disciplines.
Colleagues and data stewards	Practical advice on which standards are commonly used within a field.
Repositories	Recommended metadata schemas, file formats and disciplinary conventions.
Journal data policies	Expectations for data sharing, reporting and documentation.
Research infrastructures and disciplinary organisations	Community-developed standards, vocabularies and best practices.

Choose Appropriate File Formats

The format used to share research data can have a significant impact on whether others are able to access and reuse them in the future. While many software applications use proprietary formats that require specific programs or licences, other formats are openly documented and supported by a wide range of tools.

When possible, it is generally preferable to share data in formats that are widely used, well documented and unlikely to become obsolete in the near future. This does not necessarily mean abandoning the formats used during the research process. In many cases, researchers choose to preserve both the original working files and a more accessible version intended for sharing and long-term preservation.

Instead of only sharing	Consider also sharing	Why it may help
Proprietary spreadsheet files (e.g. Excel `.xlsx`)	CSV (`.csv`) or TSV (`.tsv`) files	Plain text formats that can be opened by many software applications and programming languages.
Proprietary text documents (e.g. Word `.docx`)	PDF/A, plain text (`.txt`) or Markdown (`.md`)	PDF/A is designed for long-term preservation, while plain text can be read by almost any system.
Lossy image formats (e.g. JPEG `.jpg`)	TIFF (`.tif`) or PNG (`.png`)	Lossless formats preserve image information without introducing compression artefacts.
Proprietary statistical software files (e.g. SPSS, Stata, SAS)	Open tabular formats such as CSV, together with documentation	Allows access to the underlying data without requiring specialised software.
Software-specific outputs	Exported formats that can be opened independently of the original software	Reduces dependence on a particular software package or version.

There is rarely a single "best" file format. Researchers often preserve both the original working files and a more open or widely accessible version intended for sharing and long-term preservation. The most appropriate choice will depend on the type of data, disciplinary practices and repository requirements.

It is important to note that the goal is not necessarily to replace the formats commonly used within a research community. In some disciplines, specialised or proprietary formats remain the standard way of exchanging data and may contain information that would be lost during conversion. Sharing data in the format that researchers in the field expect can therefore be entirely appropriate.

However, when possible, it may be useful to provide an additional version in a more accessible or widely supported format. This can reduce barriers for researchers outside the discipline, facilitate long-term access, and allow the data to be reused without requiring specialised software. Rather than choosing between one format or another, researchers can often benefit from providing both.

TSV versus CSV:

TSV (Tab-Separated Values) and CSV (Comma-Separated Values) are both open, non-proprietary formats for storing tabular data. TSV may be preferable when data fields contain many commas, such as free-text responses, metadata, or bibliographic information, as it reduces the need for quotation marks and escaping characters.

TSV is commonly used in scientific and technical fields, including bioinformatics and large-scale data processing. CSV, however, remains the more widely recognised and supported format, particularly in spreadsheet software and web applications.

Define Access Conditions

Preparing data for sharing involves deciding not only what will be shared, but also how it will be accessed and reused. While some datasets can be made openly available to anyone, others may require restrictions because of privacy, confidentiality, commercial interests, security concerns or legal obligations.

As discussed in the guide Should My Research Data Be Open?, making data available does not always mean making them openly accessible. Different access conditions can help balance openness with ethical, legal and practical considerations.

Before sharing data, it can be useful to consider the following questions:

Question	Why it matters
Can the data be shared openly?	Open access maximises visibility, reuse and transparency.
Do the data contain personal or sensitive information?	Additional safeguards, anonymisation or access controls may be required.
Are there legal, contractual or commercial restrictions?	Agreements with partners, funders or third parties may affect sharing options, and the licence applied.
Will users need to apply for access?	Controlled-access mechanisms can allow reuse while protecting sensitive information.
Have reuse conditions been made clear?	Users should understand what they are permitted to do with the data.

In practice, access conditions often fall into one of three broad categories:

Access type	Description
Open access	Data can be accessed and downloaded by anyone without restrictions.
Restricted access	Access is limited to specific users, groups or conditions, although metadata remain publicly visible.
Controlled access	Researchers must submit a request or application before access is granted.

Choosing appropriate access conditions is often more important than deciding whether data should be simply "open" or "closed". Many repositories support a range of access options that allow researchers to balance openness with ethical, legal and practical considerations.

Tip: For datasets containing personal or sensitive information, preparing data for sharing may involve anonymisation, pseudonymisation or the use of controlled-access repositories. These approaches are discussed in more detail in 'What Changes When My Research Involves Sensitive Data?'

Choose an Appropriate Licence

A licence helps future users understand how research data may be accessed, reused, shared or adapted. Without a licence, potential users may be uncertain about what they are legally permitted to do, even when the data are publicly available.

Open licences are legally binding instruments, not simply statements of intent. They define the permissions granted to future users and the conditions under which the data may be reused.

For research data, many repositories, funders and Open Science organisations recommend applying the most permissive licence that is appropriate for the dataset (e.g. CC0). Research data are frequently combined and reused across projects, institutions and disciplines, and additional restrictions can create barriers to reuse. In particular, legal attribution requirements may become difficult to manage when many datasets are combined, a challenge sometimes referred to as attribution stacking.

A clear licence can:

Clarify reuse permissions.
Reduce uncertainty for potential users.
Encourage legitimate reuse of the data.
Help users understand the conditions under which the data may be reused.

Licensing and citation are not the same thing

Researchers should still continue citing datasets they use, regardless of the licence applied. A licence governs legal reuse conditions, while citation supports transparency, acknowledgement and scholarly credit.

When selecting a licence, it is also important to verify that the dataset does not contain third-party materials, contractual restrictions, personal data or other obligations that may affect what permissions can be granted to future users.

A Final Check Before Sharing

Most barriers to reuse are not caused by a lack of effort. Rather, they arise because researchers are deeply familiar with their own data and may underestimate how much context future users will need. Taking the time to review a dataset from the perspective of someone outside the project can help identify and address many of these issues before sharing.

Here are some questions you can check out before sharing your data on a repository.

Question	Why it matters
Have I selected the data that should be shared?	Not every file created during a project needs to be preserved or shared. Focus on the data and supporting materials needed to understand, verify or build upon the research.
Are the data understandable without relying on memory?	Future users may not have access to the people who created the data. Important information should therefore be documented rather than assumed.
Is sufficient documentation available?	README files, protocols, codebooks, data dictionaries and methodological notes help others interpret and reuse the data correctly.
Are variable names, labels and abbreviations explained?	Information that seems obvious today may be difficult to interpret in the future.
Are the data stored in appropriate file formats?	Appropriate formats can improve accessibility, interoperability and long-term usability.
Have relevant community standards been considered?	Established disciplinary practices can improve interoperability and facilitate reuse.
Have access conditions been defined?	Data may be shared openly, under controlled access, or with specific restrictions depending on legal, ethical or commercial considerations.
Has an appropriate licence been selected?	A licence clarifies the conditions under which others may reuse the data.
Have personal or sensitive data been addressed appropriately?	Personal or sensitive information may require anonymisation, restricted access or additional safeguards before sharing.
Can someone understand how the data were generated and processed?	Information about data collection, cleaning, processing and analysis helps future users assess the quality and suitability of the data.

Key Takeaways

Preparing data for sharing involves more than selecting files to preserve.
Documentation, metadata and context are often as important as the data themselves.
Community standards and appropriate file formats can improve interoperability and reuse.
Access conditions and licences should be chosen before data are shared.
Removing barriers to reuse increases the long-term value and impact of research data.