The New Reality of Reproducibility: The Role of Data Work in Scientific Research

Although reproducibility--the idea that a valid scientific experiment can be repeated with similar results--is integral to our understanding of good scientific practice, it has remained a difficult concept to define precisely. Across scientific disciplines, the increasing prevalence of large datasets, and the computational techniques necessary to manage and analyze those datasets, has prompted new ways of thinking about reproducibility. We present findings from a qualitative study of a NSF--funded two-week workshop developed to introduce an interdisciplinary group of domain scientists to data-management techniques for data-intensive computing, with a focus on reproducible science. Our findings suggest that the introduction of data-related activities promotes a new understanding of reproducibility as a mechanism for local knowledge transfer and collaboration, particularly as regards efficient software reuse.