# How to Encourage Data Reuse in Your Research Community By the end of this article you will be able to build a concrete plan for research data sharing in your lab, department, or consortium. You will know how to prepare a dataset so that strangers can actually work with it, how to run a data repository comparison before depositing, how to write metadata that machines can read, and how to set up incentives so that people reuse data instead of quietly ignoring it. You need a dataset you control, a plain text editor, access to a command line terminal, and about an afternoon for the first walkthrough. The core argument is simple. Mandating deposition increases the amount of data available, but mandates alone do not create reuse. The WorldWide Antimalarial Resistance Network (WWARN), after ten years of operating as a data platform, concluded that genuine secondary analysis requires efficient curation processes and real rewards for the people who share and reuse data [1]. Everything below follows from that observation. ## Why Data Reuse Matters in the Life Sciences Data reuse means that a dataset collected for one study is analyzed again, by other people, for new questions. Reuse takes many forms. Pooled analyses combine individual-level data from several studies to answer questions no single study can address. Replication checks confirm that published findings hold. Modelers fit new computational models to existing data, as happens in pharmacometrics, where shared data, models, and workflows support open innovation across organizations [2]. Computational biomechanics researchers share models so that others can verify reproducibility, repurpose them for new simulations, and blend expert-built models across scales and physical domains [3]. The payoff is measurable. OpenNeuro, a BRAIN Initiative archive for brain imaging data, shares more than 600 datasets covering more than 20,000 participants, and those shared datasets have already supported more than 150 publications that reused the data [4]. That is research output that would not exist if every dataset had stayed on a lab hard drive. Reuse also multiplies the value of public investment. Sharing data supports reproducibility, replication, consortium-level pooled analyses, and the ability to answer multiple research questions from a single expensive study, which is why epidemiologists have made FAIR practices a central recommendation for their field [5]. ## The Foundation: FAIR Data Before You Promote Reuse You cannot encourage data reuse in your research community if the shared data are unusable. The FAIR principles, findability, accessibility, interoperability, and reusability, define what usable means [4]. Brain-CODE, the Ontario Brain Institute neuroscience platform operating since 2012, turned each principle into a concrete service. Findable meant interactive tools for requesters to build data cuts matched to their research question. Accessible meant distinct mechanisms for metadata access, secure-environment access, and data export. Interoperable meant standardization both at data capture and at data release. Reusable meant documentation and context that let a newcomer start analysis. Two preparation steps carry most of the weight. **First, make metadata machine-readable.** Metadata describe and give context to data, and machine-readable metadata let both computers and humans discover, access, integrate, and reuse data across platforms. A framework developed for population health data uses a three-stage approach: document datasets with the Data Documentation Initiative (DDI) Codebook standard, harmonize variables into the Observational Medical Outcomes Partnership (OMOP) Common Data Model, and expose everything through standardized formats and vocabularies [7]. **Second, avoid the minimum-effort metadata trap.** A study of metadata use in electronic lab notebooks found that many users adopt a "minimum required" approach to metadata, which endangers their own ability to recover data later, let alone anyone else's [8]. If your community members fill in metadata only to satisfy a form, the resulting records will not support discovery or reuse. ## Step-by-Step Strategies to Encourage Data Reuse in Your Research Community The following walkthrough takes a dataset from a lab folder to a deposited, citable, reusable resource. Commands are shown for a Unix-style terminal. Every template is generic, so adapt the field names to your field. ### Step 1. Inventory what you already have List every data file, its format, its size, and its access restrictions. From your project directory: ```bash find . -type f \( -name "*.csv" -o -name "*.xlsx" -o -name "*.dta" -o -name "*.nii.gz" \) -exec du -h {} \; ``` Expected output looks like this: ``` 12M ./raw/cohort2019_measurements.csv 48M ./raw/imaging_subject14.nii.gz 3M ./processed/cohort2019_clean.csv ``` Record the results in a file called `inventory.csv` with columns for path, format, human subjects yes or no, and current location. This inventory becomes your deposit checklist and your risk screen. Research programs that involve data stewards early in the data lifecycle, at the onset of the research program rather than at the end, preserve data far more successfully [9]. ### Step 2. Organize files in a documented structure Create a folder layout that separates raw from processed data and add a `README.md` at the top level. A plain file listing works better than an elaborate structure nobody maintains: ``` README.md data_dictionary.csv raw/ processed/ code/ docs/protocol_v2.pdf ``` The README should state, in plain sentences, what each study visit measured, what the cohort was, and what each folder contains. Common error: researchers write the README last, after deposit, and never get around to it. Write it before deposit, while the decisions are still fresh. ### Step 3. Build a data dictionary A data dictionary is a machine-readable table describing every variable. Keep it as a CSV so any tool can read it: ```csv variable,label,type,units,allowed_values,description subject_id,Participant identifier,string,,,"Anonymous code assigned at enrollment" age_years,Age at enrollment,integer,years,18-85,"Self-reported at first visit" treatment_arm,Randomized group,string,,control|low_dose|high_dose,"Assignment from central randomization" outcome_score,Primary endpoint score,numeric,points,0-100,"Measured at week 24 by blinded assessor" ``` This is the single highest-value artifact for reuse. A person downloading your data can map your columns to their analysis in minutes instead of emailing you and waiting. ### Step 4. Convert to open, nonproprietary formats Epidemiologic FAIR guidance recommends moving from proprietary files to machine-readable, nonproprietary formats and open-source analysis code as datasets progress toward wider reuse [5]. In practice this means CSV instead of a proprietary spreadsheet format for [tabular data](/blog/research-skills/tabular-data-what-it-is-and-how-to-analyze-it), plain text for code, and community standards for domain-specific data. In neuroimaging, the Brain Imaging Data Structure (BIDS) standard is the reason OpenNeuro can curate, share, and support reuse across hundreds of heterogeneous datasets [4]. Check whether your field has an equivalent standard before inventing your own layout. A frequent conversion error looks like this in Python: ``` UnicodeDecodeError: 'utf-8' codec can't decode byte 0x96 in position 1203 ``` This means the CSV was exported with a Windows legacy encoding. Fix it by re-exporting as UTF-8 from the source application, or read it explicitly with the matching encoding in pandas, for example `pd.read_csv("raw/file.csv", encoding="cp1252")`, then write it back with `encoding="utf-8"`. ### Step 5. Write metadata in a standard, machine-readable format ![Path from a metadata record to a parse check, with fixes for a scanner error, a CSV encoding error and a checksum check.](/article-images/illustrations/encourage-data-reuse-research-community-strategies-metadata-errors.webp) Structured documentation beats a prose paragraph. A minimal machine-readable record in YAML looks like this: ```yaml title: "Longitudinal cohort measurements, cohort 2019" creators: - family: "Rivera" given: "A." description: "Primary endpoint measurements and covariates for 240 participants followed 24 weeks." keywords: ["[cohort study](/blog/guides/cohort-studies-a-practical-guide-to-design-conduct-and-interpretation)", "primary endpoint", "longitudinal"] license: "CC-BY-4.0" access_level: "open" citation: "Please cite the accompanying dataset DOI in derivative work." ``` If your data support clinical or population health reuse, mapping variables to the OMOP Common Data Model, and documenting the study with DDI Codebook, makes integration with other datasets straightforward rather than a months-long negotiation [7]. To verify your metadata file parses correctly: ```bash python3 -c "import yaml; print(yaml.safe_load(open('metadata.yml'))['title'])" ``` Expected output: ``` Longitudinal cohort measurements, cohort 2019 ``` If instead you see `yaml.scanner.ScannerError: mapping values are not allowed here`, a colon is unquoted inside a value. Wrap that value in quotation marks and rerun. ### Step 6. Map variables to standard vocabularies where they exist Interoperability improves dramatically when your variable names point at shared definitions. The Vaccine Ontology project shows the pattern: the CanVaxKB database cataloged more than 670 manually annotated cancer vaccines, but a database alone does not provide structured relationships and standardized definitions, so the team expanded the ontology to define and interrelate those vaccines, then used SPARQL query scripts to retrieve vaccines by their features [10]. For your own data, check whether an established ontology covers your domain, and record the ontology term identifier alongside each variable in your data dictionary. ### Step 7. Deposit in the right repository and get a persistent identifier ![Decision path asking whether data are sensitive, field-standard, or pooled, leading to controlled-access, domain archive, or preservation archive.](/article-images/illustrations/encourage-data-reuse-research-community-strategies-repository-choice.webp) Run a data repository comparison before you deposit, not after. The table below summarizes the main options using real platforms from the life sciences literature. | Repository type | Best for | Access model | Documented examples | |--|--|--|--| | Domain-specific open archive | Field-standard data types with curation expertise | Open, following FAIR principles | OpenNeuro for brain imaging [4] | | Disease-specific consortia | Pooled, harmonized individual-level data | Often managed or application-based | ADNI, WWARN, PPMI and similar pharmacometric and disease databases [2] | | Controlled-access platform | Sensitive human data with governance requirements | Tiered access for metadata, secure-environment analysis, or export | Brain-CODE [6] | | Institutional or preservation archive | Long-term stewardship after a project ends | Open or mediated | OAIS-based institutional repositories [9] | Match the repository to your data's sensitivity, not to convenience. Human subject data with confidentiality constraints may require a controlled-access model, and epidemiologic FAIR practice explicitly pairs accessible data with access limited to authorized users where appropriate [5]. Resolve legal, ethical, and proprietary questions before the deposit conversation starts, because these questions behave the same across fields that share individual-level longitudinal data [2]. Common error during deposit: the repository rejects a file with a message like `400 Bad Request: file exceeds maximum size`. Split the dataset into logical chunks with `split -b 900M bigfile.csv chunk_` (note: this command's flag is two dashes, written as split -b with a size limit, or archive the files first with `zip`), or ask the repository about large-file transfer. A second common error is a DOI that returns not found in the hours right after deposit. [Persistent identifiers](/knowledge/bioinformatics/persistent-identifiers-in-bioinformatics-selection-criteria-and-implementation-workflow) often take a day to activate, so verify the DOI the next business day before advertising it. Verify your upload integrity with checksums before and after transfer: ```bash sha256sum processed/cohort2019_clean.csv ``` Expected output: ``` 9f86d081884c7d659a2feaa0c55ad015a3bf4f1b2b0b822cd15d6c15b0f00a08 processed/cohort2019_clean.csv ``` Run the same command on the downloaded copy. If the two hashes differ, the transfer corrupted the file. If you see `sha256sum: processed/cohort2019_clean.csv: No such file or directory`, your working directory is wrong, so run `pwd` and `ls` to reorient. ### Step 8. Make the data citable and linked Attach a license, publish the DOI, and cite the dataset in every related paper using the standard pattern: author name, year, dataset title, repository name, and DOI. Add a data availability statement to the manuscript that names the repository and the access route. This closes the loop between the paper, which readers see, and the data, which they need. ### Step 9. Run at least one active reuse activity Passive availability produces quiet repositories. WWARN's study group approach brings researchers together around a shared question, collates and curates the relevant data, and credits everyone involved, with deliberate attention to researchers in resource-limited settings who otherwise get excluded from secondary analyses [1]. A study group can be as small as three labs, one curated dataset, and one jointly authored analysis plan. The activity creates both the reuse and the social proof that reuse is normal in your community. ### Step 10. Track reuse and report it back Measure what you can: dataset downloads, views, citations, and published secondary analyses. OpenNeuro tracked reuse through published papers, which grew past 150 publications, and used that evidence to justify continued development [4]. Report these numbers to your community. People repeat behaviors that get counted and celebrated. ## Build the Culture: Incentives That Actually Change Behavior Academic advancement seldom recognizes or rewards data sharing, and this gap is the deepest obstacle to reuse [1]. Mandates produce deposits. Incentives produce good deposits and genuine reuse. Four mechanisms have support in the literature. **Credit and authorship.** WWARN's model rewards contributors through the study group process, where curators and data providers earn visible roles in downstream analyses [1]. Epidemiologic FAIR recommendations likewise call for recognition and incentives for sharing both data and code, with appropriate credit to resource creators [5]. **Organizational support and training.** Achieving a high degree of FAIRness requires funding, training, organizational support, and incentives, and these costs are outweighed by the benefits of more reproducible research [5]. Concretely, your department can fund a data steward role. Data stewards who engage with repositories and curators at the start of a research program develop data sustainability plans that keep data alive after the grant ends [9]. **Standards as shared infrastructure.** Communities that organize around standards lower the cost of every subsequent reuse. Pharmacometric organizations have worked on data formats, exchange standards, and database properties precisely so that sharing becomes routine rather than heroic [2]. Computational biomechanics researchers treat model sharing as a necessity rather than a utilitarian interest, driven by the desire to combine expert-built models across systems and scales [3]. Openness recommendations for data-driven life science studies similarly emphasize making code and data public and transparent to maximize impact, reproducibility, and reuse [11]. **Leadership signals.** In workplace settings, leadership-driven knowledge-sharing opportunities and explicit encouragement shape whether professionals exchange information at all, a pattern documented among hospital health professionals where such opportunities varied by profession and remained limited overall [12]. The research analog is clear: if department heads never mention data sharing in hiring, promotion, or seminar culture, junior researchers read the silence accurately. ## Sustaining Reuse Over Decades, Not Months A dataset deposited and forgotten decays. Long-term preservation of biomedical data requires an explicit strategy, and the Open Archival Information System (OAIS) model provides the standard functional structure with six entities: Ingest, Access, [Data Management](/blog/guides/data-management-basics-principles-processes-and-best-practices), Archival Storage, Administration, and Preservation Planning [9]. You do not need to run the archive yourself, but you should ask candidate repositories which of these functions they perform and who funds them. Placing equal importance on initial activities such as collection and on downstream activities such as preservation keeps data a continued resource after the research project reaches the end of its lifecycle [9]. Ask three questions of any repository before committing. Does it assign persistent identifiers. Does it commit to format migration as file formats age. Can it support the access model your data require, whether open, tiered, or secure-environment. Brain-CODE's tiered design, separating metadata access from secure-environment analysis from export, shows how one platform can serve both openness and governance at once [6]. ## Common Mistakes and Limitations The most frequent failure modes in research data sharing are predictable, and each has a direct fix. - **Treating the mandate as the strategy.** Mandating sharing increases the volume of data available without enabling the secondary analyses that justify the policy [1]. Pair every mandate with curation support and credit. - **Minimum-effort metadata.** Users who fill in only the required metadata fields compromise their own future ability to find and recover their data [8]. Audit a random sample of your community's deposits quarterly and show examples of good records. - **Proprietary formats and closed code.** Nonproprietary files and open-source code are explicit FAIR recommendations, and skipping them blocks integration before anyone even tries [5]. - **Ignoring the human subject questions until deposit day.** Legal, ethical, and proprietary issues for individual-level data need resolution early, with data stewards and repository curators involved from the start [2]. - **Depositing without a data dictionary.** A dataset without variable definitions forces every reuser to reconstruct context from the methods section, and most will not bother. - **Expecting reuse without community activity.** Availability is not the same as reuse. Structured activities such as study groups create the first wave of secondary analyses and the norms that sustain them [1]. - **No feedback loop.** If contributors never see download counts, citations, or co-authorship from reuse, participation fades. Track and report reuse metrics. One honest limitation. The evidence base for what drives reuse is strongest in fields with mature infrastructure, such as neuroimaging, pharmacometrics, and epidemiology [2][4]. If your field lacks a BIDS-equivalent standard, you may need to adopt a neighboring field's conventions and advocate for domain standards in parallel. ## Frequently Asked Questions ### What is the fastest first step to encourage data reuse in my research group? Write a data dictionary for your most-used dataset this week. It costs hours, requires no new infrastructure, and removes the single largest barrier every downstream reuser faces. ### Do I have to share everything openly? No. FAIR accessibility means accessible to authorized users, and controlled-access or secure-environment models are accepted mechanisms for sensitive data, as Brain-CODE and epidemiologic practice both demonstrate [6][9]. ### How do I choose between repositories? Match the repository to data type, sensitivity, and required curation. Domain archives such as OpenNeuro provide field-standard curation, controlled-access platforms such as Brain-CODE handle governed human data, and disease-specific consortia support pooled harmonized analyses [2][4][6]. ### What does machine-readable metadata actually mean? It means metadata stored in standardized formats and vocabularies that software can parse, such as DDI Codebook documentation or OMOP Common Data Model mappings, rather than free-text descriptions that only humans can interpret [7]. ### How do I get academic credit for sharing data? Deposit with a DOI, cite the dataset in your papers, and join or organize reuse activities such as study groups where contributors earn authorship. Recognition and incentives for sharing remain underdeveloped, so building them into your community is part of the work [9]. ### Is sharing human subject data risky? The risks are managed, not eliminated, through governance. Resolve legal and ethical questions early, engage data stewards at project onset, and choose an access model that matches the sensitivity of the data [2]. ### How do I know if reuse is actually happening? Track downloads, citations, and published secondary analyses of your datasets, and report the numbers back to contributors. Published reuse papers are the outcome measure OpenNeuro used to demonstrate impact [4]. ### What is the difference between data sharing and data reuse? Sharing is the act of making data available to others. Reuse is what happens afterward, when other researchers analyze that data for new questions. Sharing without reuse usually signals missing metadata, missing incentives, or missing community structures [1]. ## Sources 1. [Strength in Numbers: The WWARN Case Study of Purpose-Driven Data Sharing.](https://pubmed.ncbi.nlm.nih.gov/30350772/) 2. [Open innovation: Towards sharing of data, models and workflows.](https://pubmed.ncbi.nlm.nih.gov/28684136/) 3. [Perspectives on Sharing Models and Related Resources in Computational Biomechanics Research.](https://pubmed.ncbi.nlm.nih.gov/29247253/) 4. [The OpenNeuro resource for sharing of neuroscience data.](https://pubmed.ncbi.nlm.nih.gov/34658334/) 5. [Moving Toward Findable, Accessible, Interoperable, Reusable Practices in Epidemiologic Research.](https://pubmed.ncbi.nlm.nih.gov/36804665/) 6. [FAIR in action: Brain-CODE - A neuroscience data sharing platform to accelerate brain research.](https://pubmed.ncbi.nlm.nih.gov/37274750/) 7. [Making Metadata Machine-Readable as the First Step to Providing Findable, Accessible, Interoperable, and Reusable Population Health Data: Framework Development and Implementation Study.](https://pubmed.ncbi.nlm.nih.gov/39088253/) 8. [Creating context for the experiment record. User-defined metadata: investigations into metadata usage in the LabTrove ELN.](https://pubmed.ncbi.nlm.nih.gov/25405258/) 9. [Long-term preservation of biomedical research data.](https://pubmed.ncbi.nlm.nih.gov/30356367/) 10. [Empowering standardization of cancer vaccines through ontology: enhanced modeling and data analysis.](https://pubmed.ncbi.nlm.nih.gov/38890666/) 11. [Recommendations for open data science.](https://pubmed.ncbi.nlm.nih.gov/27195107/) 12. [Knowledge-Sharing Practices Among Dentists, Pharmacists, and Allied Health Professionals: A Cross-Sectional Study in Eastern Cape Public Hospitals, South Africa.](https://pubmed.ncbi.nlm.nih.gov/41595860/)