provide comprehensive information about the necessary data systems and environments required to support data.

A Research Data Census will provide some specifics critical to cost-effective planning for stewardship of federally funded research data, however, and it will allow us to infer some key requirements for data cyberinfrastructure. In particular, a Research Data Census could help inform:

˲ • Useful estimates of the storage capacity required for data stewardship, and a lower bound on data that must be preserved for future timeframes. Data required by regulation or policy to be preserved is a lower bound on valued preservation-worthy research data—additional data sets will need to be preserved for research progress (for example, National Virtual Observatory data sets).

˲ • The types of data services most important for research efforts. Knowing the most common types of useful services and tools can help drive academic and commercial efforts.

˲ • Estimates of the size, training, and
skill sets that will be needed for today’s
and tomorrow’s data work force.

Getting it Done

A Data Census sounds like a big job and it is, however there is potential to use existing mechanisms to help gather the needed information efficiently. We already provide annual and final reports to federal R&D agencies to describe the results of sponsored research. One could imagine a straightforward addition to annual reporting vehicles and/or sites such as grants. gov to collect this information ( preferably electronically). Although U.S. Census information is gathered every 10 years, the Research Data Census would require frequent updating in order to provide useful information for planning purposes about our dynamically changing data landscape. The right periodicity for reporting is a topic for discussion, but an annual update probably provides the best resolution for the purpose of tracking trends.

Note also that there is real complexity in doing an effective Data Census: much of our data is generated from collaborative research, which can cross institutional, agency, and na-

an effective Research
Data Census should
provide a quantitative
snapshot of the
research data
landscape at
a given point in time.

tional boundaries. The Data Census reporting mechanisms must take this into account to produce relatively accurate counts. Data sets are often replicated for preservation purposes—do we count the data in all copies (all of which require storage), or do we count only the non-replicated data? (It is interesting to note that the U.S. Census has a related problem and covers it as question 10: “Does person 1 sometimes live or stay somewhere else?” If yes, check all that apply….). As with any survey, careful design is critical in order to ensure the results are accurate and useful as the basis for making predictions and tracking trends.

 

using the Research Data Census to Create Effective Data Stewardship

An important outcome of the Research Data Census would be evidence-based information on the amount of data in the research community that must be preserved over time. This would help in understanding and meeting our needs for archival services and community repositories.

Such information can help cut data management and preservation problems down to size. Knowing that data valued by a particular community is typically of a certain type, a certain size, and/or needed over a certain timeframe, can help the community plan for the effective stewardship of that data. For example, accurate estimates of the digital data emanating from the Large Hadron Collider at CERN have been instrumental in the development of a data analysis and management plan for the High Energy Physics community.

It is likely that some of the capacity needed for stewardship of research data will come from university libraries reinventing themselves to address 21st century information needs; some of the capacity may come from the commercial sector, which has responded to emerging needs for digital storage and preservation through the development of commercial services. In some cases, the federal government will take on the stewardship responsibilities for research data (for example, the NIST Science Reference Data). It is clear that the size, privacy, longevity, preservation, access, and other requirements for research data preclude a “ one-size-fits-all” approach to creation of supporting data cyberinfrastructure. It is also true that no one sector will be able to take on the responsibility for stewardship of all research data. A national strategic partnership spanning distinct sectors and stakeholder communities is needed to effectively address the capacity, infrastructure, preservation, and privacy issues associated with the growing deluge of research data. The development of a Research Data Census can provide critical information for more effectively developing this partnership.

 

no Time Like the Present

The 2010 requirement for a data management plan at the National Science Foundation (http://www.nsf. gov/news/news_summ.jsp?cntn_id= 116928&org=NSF) joins existing requirements for data sharing and management at NIH and elsewhere. Such requirements expand community awareness about responsible digital data stewardship and will exacerbate the emerging need for reliable, cost-effective data storage and preservation at the national scale.

A Research Data Census will provide a foundation for estimating the data cyberinfrastructure required for strategic stewardship. It can lay the groundwork today for access to our most valuable digital research assets tomorrow, and the new discoveries and innovation they drive.

 

Francine Berman ( [e-mail]) is Vice president for research at rensselaer polytechnic institute, the former director of the San diego Supercomputer Center, and the co-chair of the blue ribbon task Force for Sustainable digital preservation and access ( http://brtf.sdsc.edu).