Making research ‘really reproducible’

Listening to Victoria Stodden, Assistant Professor of Statistics at Columbia University, give the keynote speech at the recent Open Repositories conference in lovely Prince Edward Island Canada, I realised we have some way to go on the path towards her idealistic vision of how to “perfect the scholarly record.”

Victoria Park, Charlottetown

Charlottetown, Prince Edward Island

As an institutional data repository manager (for Edinburgh DataShare) I often listen and talk to users about the reasons for sharing and not sharing research data. One reason, well-known to users of the UK Data Archive (now known as the UK Data Service), is if the dataset is very rich and can be used for multiple purposes beyond those for which it was created, for example, the British Social Attitudes Survey.

Another reason for sharing data, increasingly being driven by funder and publisher policies, is to allow replication of published results, or in the case of negative results which are never published, to avoid duplication of effort and wasting public monies.

It is the second reason on which Stodden focused, and not just for research data but also for the code that is run on the data to produce the scientific results. It is for this reason she and colleagues have set up the web service, Run My Code. These single-purpose datasets do not normally get added to collections within data archives and data centres, as their re-use value is very limited. Stodden’s message to the audience of institutional repository managers and developers was that the duty of preserving these artefacts of the scientific record should fall to us.

Why should underlying code and data be published and preserved along with articles as part of the scholarly record? Stodden argues, because computation is becoming central to scientific research. We’ve all heard arguments behind the “data deluge”. But Stodden persuasively focuses on the evolution of the scientific record itself, arguing that Reproducible Research is not new. It has its roots in Skepticism – developed by Robert Boyle and the Royal Society of the 1660s. Fundamentally, it’s about “the ubiquity of error: The central motivation of the scientific method is to root out error.”

In her keynote she developed this theme by expanding on the three branches of science.

  • Branch 1: Deductive. This was about maths and formal logic, and the proof as the main product of scientific endeavor.
  • Branch 2: Empirical. Statistical analysis of controlled experiments – hypothesis testing, structured communication of methods and protocols. Peer reviewed articles became the norm.
  • Branch 3: Computational. This is at an immature stage, in part because we have not developed the means to rigorously test assertions from this branch.

Stodden is scathing in her criticism of the way computational science is currently practiced, consisting of “breezy demos” at conferences that can’t be challenged or “poked at.” She argues passionately for the need to facilitate reproducibility – the ability to regenerate published results.

What is needed to achieve openness in science? Stodden argued for the need for deposit and curation of versioned data and code, with a link to the published article, and  permanence of the link.This is indeed within the territory of the repository community.

Moreover, to have sharable products at the end of a research project, one needs to plan to share from the outset. It’s very difficult to reproduce the steps to create the results as an afterthought.

I couldn’t agree more with this last assertion. Since we set up Edinburgh DataShare we have spoken to a number of researchers about their ‘legacy’ datasets which – although they would like to make them publicly available, they cannot, either because of the nature of the consent forms, the format of the material, or the lack of adequate documentation. The easiest way is to plan to share. Our Research Data Management pages have information on how to do this, including use of the Digital Curation Centre’s tool, DMPOnline.

– Robin Rice, Data Librarian

Leave a Reply

Your email address will not be published. Required fields are marked *