Organizational Structures Panel Report: 2018 Data Science Leadership Summit

Data science is fast emerging as a new discipline. Data science initiatives across universities and colleges in the US and beyond are emerging at a rapid rate. That data science is as much a practice as it is a discipline, raises challenging organizational questions - for example, whether and how data science should be its own major, department, or division at a university. If data science is siloed into one academic unit, how does one effectively integrate data science skills and best practices into curricula in other departments, particularly those necessary for students in other disciplines to reach the expectations of 21st century employers? This panel was an outgrowth of discussions held during the inaugural Data Science Leadership Summit held in March 2018 at Columbia University, with the goal of furthering understanding of the approaches to data science developing at US institutions. For this Summit, the Program Committee, led by Srinivas Aluru, chose “Data science organizational structures” as one of the current challenge areas for academic departments and cross-departmental initiatives. The session was organized to meet the objectives of the Summit and learn from those leading exemplar programs about success, challenges, and lessons learned. The goal of the panel was to move beyond the structured information already gathered on entities and their institutions (from the Summit report) to questions and decisions encountered along the way. How do we make data science entities work for our institutions? How do we make them sustainable? What approaches seem
to be working well? What are the challenges?

Key topics covered include:

• Models of campus engagement
• On-going challenges for the DS entity at the Institutional level
• Key tactical decision points - perspectives from “looking back”
• What’s important as the DS community moves forward?

Significant take-aways:

• A number of institutions have achieved considerable success in establishing a “virtuous cycle” in which the “producers” of data science methodology are closely partnered with the “consumers” of that methodology, with advances in methodology driving advances in discovery, which in turn drive further advances in methodology. While the institutions belonging to the Moore-Sloan Data Science Environments project (Berkeley, NYU, and the University of Washington) had a head start in this regard, many other institutions, such as Columbia University (started in 2012 with NYC funds) have met with success.
• Many institutions are experiencing tremendous demand from all corners of the campus for data science education of all forms, ranging from consulting and tools-oriented short courses to degree programs at the bachelors, masters, and doctoral level. Education can be a good way to create connections, collaborations, and goodwill.
• Campus and local community engagement - i.e. active stimulation events or interactions - is a significant aspect of most of the institutional models discussed on this panel. While timing and processes vary, there are widespread efforts to bring together faculty, students, and projects to leverage expertise and resources. Differences in the ways that research and scholarship are funded across domains require different approaches to bringing together faculty “under one roof.”
• Variation in institutional and domain-based incentives, as well as compensation (e.g. grant-funded or not), will impact the level of interest and engagement across departments and faculty.
• Only a small number of institutions are organized to generate revenue.

Editorial Note: Some sections have been modified by panelists to update perspective and provide additional context. We are grateful for these additions. (March 12, 2020)