Data governance frameworks from IASSIST 2017

The archived IASSIST 2017 conference, held at the University of Kansas in Lawrence from 23 to 26 May 2017, brought together researchers working across information science, agriculture, health and international development. Its theme, “Data in the Middle: The Common Language of Research,” captured the practical problem behind the event’s governance discussions: how shared data can remain understandable, usable and accountable across disciplinary and organisational boundaries.

For an Australian audience, the conference is especially relevant because research data crosses familiar divides between government, universities, industry and community organisations. Canberra concentrates national agencies, while universities and research institutes operate across states with different privacy regimes and infrastructure. The Australian research market works through competitive grant schemes, block-funded university portfolios and collaborative projects, so stewardship affects both compliance and the value partners can obtain. ARC and NHMRC policies have made data planning and sharing part of reporting for many funded projects, and repositories such as Research Data Australia help connect institutional holdings.

Governance here means more than a privacy notice or repository catalogue. It covers the full lifecycle of data: collection, consent, documentation, storage, analysis, publication, reuse and disposal. It also assigns responsibility for quality, security and legitimate access. That wider definition is essential for deep learning, where a trained model may circulate long after its source datasets have moved, and for digital agriculture, where observations about land, water and production can affect commercial interests and communities.

The archived program links plenary discussion with contributed work on big data, machine intelligence, digital agriculture and global food security. Read together, the sessions present governance as shared infrastructure rather than an administrative afterthought. The strongest thread is interoperability: common metadata, durable identifiers, clear licences and explicit provenance allow people with different technical vocabularies to assess and combine evidence without surrendering local control.

From conference themes to governance principles

Looking across the program, the governance frameworks discussed at IASSIST 2017 can be grouped into three connected concerns: stewardship of research assets, responsible use of sensitive information, and technical interoperability. The plenaries supplied a common vocabulary, while specialist presentations tested that vocabulary against large-scale computation and food-system research. This combination matters because policies that work for a curated social-science dataset may fail when streaming sensor records or proprietary farm data are involved.

At the centre is the familiar FAIR pattern: data should be findable, accessible, interoperable and reusable. The conference material is best read as an early expression of those principles in a research setting, not as a single rigid standard. A framework also needs named custodians, documented decisions, risk-based controls and mechanisms for review. Without those organisational elements, good metadata can describe a dataset while leaving nobody accountable for correcting it or explaining a restriction.

Australian workplace discussion often favours plain questions and a “fair go” for all participants, sometimes framed as “let’s be clear”. Translated into governance, that means researchers should be able to see why data are collected, who may use them and how benefits are shared. It also means data governance cannot be reserved for senior academics or information specialists. Students, field technicians, partner organisations and regional participants need workable routes to ask questions and influence decisions.

Governing data through its research lifecycle

A lifecycle model turns broad principles into responsibilities. Before collection, a data management plan should state purpose, formats, volume, storage, naming conventions, retention, sharing options and the ethical or legal basis for use. During active research, a steward records changes and ensures that files and metadata remain linked. At publication, versioned deposits, persistent identifiers and licences separate the citable record from working copies. Closure should trigger review, archiving or defensible destruction rather than indefinite, unmanaged storage.

Deep learning raises the cost of weak documentation. Labels, training-set selection, preprocessing code and model versions determine what a result means, yet a paper may mention only the final architecture. Good provenance follows the chain from raw observation to trained parameter, preserves decisions about exclusions and notes uncertainty where it cannot be removed. This supports reproducibility, while also allowing reviewers and public agencies to inspect bias, leakage and conflicts that a headline accuracy figure can conceal.

Access should reflect both usefulness and harm. Open formats and clear licences enable reuse, but open release is not appropriate where data identify people, reveal confidential farm operations or expose locations that could create risk. Australian projects must also consider the Privacy Act 1988, applicable state and territory rules, ethics conditions, contractual limits and Indigenous authority. A tiered model can separate public metadata, approved controlled access and restricted records, with review dates and auditable reasons for each restriction.

Stewardship across agriculture and food systems

Digital agriculture makes these choices tangible. In the grain and livestock districts around Toowoomba and the Darling Downs, sensors, satellite imagery, weather records and farm-management software can be combined to improve yields, water use and disease response. The same combination can expose land values, production methods or business performance. A domain framework therefore needs agronomic standards and data stewards who understand seasonal workflows, alongside ordinary security controls; generic repository rules alone will not reveal whether a date, soil reading or yield estimate is fit for a particular decision.

Global food-security research adds an equity dimension. A platform may contain abundant data about well-connected farms while overlooking pastoralists, smallholders or households facing the greatest constraints. Coverage gaps can then be mistaken for evidence that a problem is small. Fieldwork across inland Australia also depends on connectivity that is intermittent at blackspots, so collection, consent and quality checks must survive offline work and delayed transfer. Where Aboriginal and Torres Strait Islander knowledge is involved, governance must recognise community authority, negotiated access and benefit sharing rather than treating cultural information as an ordinary open resource.

Comparing governance approaches

No single framework covers every stage or type of data. The comparison below places the main lenses discussed in the conference program side by side, showing where each adds control and where another layer is required.

Framework or lens	Primary purpose	Main mechanisms	Best suited to	Common weakness
FAIR data principles	Make research assets discoverable and reusable	Rich metadata, persistent identifiers, standard licences and interoperable formats	Cross-platform discovery, replication and integration	Says little about ethical access or organisational accountability
Institutional data management policy	Assign responsibility across a project	Data management plans, named stewards, retention schedules and repository deposit	Universities and grant-funded research teams	Can become a compliance exercise if workflows are not supported
Ethics, privacy and security review	Protect people, communities and lawful interests	Consent, minimisation, de-identification, agreements and tiered access	Health, personal, commercially sensitive or location-sensitive data	Can slow delivery when review happens after design is fixed
Domain stewardship	Maintain shared meaning and quality over time	Community governance, data standards, curation, provenance and feedback	Agricultural, environmental and food-system datasets	Requires stable funding and trusted relationships with contributors
Algorithm and model governance	Make derived outputs reliable and explainable	Versioned training data, model documentation, validation, bias tests and monitoring	Deep learning and other high-volume predictive systems	Cannot repair poor source coverage or undocumented assumptions

These lenses work best as a stack. FAIR improves interoperability; institutional policy creates ownership; ethics and law determine permissible use; domain stewardship protects meaning; and model governance tests the claims produced from the data. Applying only one layer creates blind spots. A dataset can be meticulously FAIR yet ethically unusable, while a heavily controlled collection can remain inaccessible even to the people responsible for its quality.

The appropriate balance also changes with risk and reuse potential. A public environmental time series may justify open release with quality flags. A linked health dataset may need secure infrastructure and detailed agreements. Cooperative farm data may require a trustee, purpose limits and revenue-sharing terms before individuals see benefits. Governance should therefore be proportional, documented and revisited when the research question, dataset combination or intended audience changes.

Recommendations for accountable data stewardship

Implementation starts with operations, not aspirational wording. Australian institutions can use role-based committees that include domain experts, research software engineers, privacy and ethics specialists, Indigenous governance expertise and representatives from partner communities. Each dataset should have a responsible custodian, escalation route, access-review cycle and service-level expectation for corrections. The archived conference’s practical guidance for reaching its venues in Lawrence remains available through Lawrence transport information. That practical emphasis carries over for Australian teams: reliable travel, clear arrival instructions and accessible participation are part of whether governance enables knowledge to circulate.

Oversight also needs a rhythm. A short launch review can confirm purposes, risks and responsibilities; periodic audits can test whether metadata still match the data, restrictions remain justified and access decisions were applied consistently; and a closing review can record what was learned for future projects. These checks should produce evidence that the committee can inspect, not merely minutes that no one revisits. Responsibility must be distributed enough to handle real work while remaining clear enough to prevent every issue disappearing into a collective inbox.

• Assign a named data steward and an accountable business owner for every governed dataset, with clear authority to approve access and resolve quality problems.
• Require a proportionate data management plan before collection, covering lawful and ethical purpose, formats, storage, retention, sharing and destruction.
• Capture metadata, provenance, version identifiers and licences as part of the research workflow, using automated checks where feasible and manual review where meaning cannot be reduced to a field.
• Use tiered access, data-use agreements and review dates for sensitive records, and explain restrictions in plain language to people whose data are affected.
• Assess representativeness, label quality and likely bias before training or publication, then document limitations and monitor important changes after deployment.
• Fund repositories, curation, stewardship and preservation as continuing research infrastructure, and measure success through reuse, data quality, compliance and demonstrable public value.

Good governance at IASSIST 2017 was therefore a disciplined arrangement for stewardship, context and accountability rather than a declaration of open data. Its enduring value for Australian research is a practical standard: each important dataset should have an owner, documented conditions, traceable decisions and a credible route for responsible reuse.