[Experiment] Exploring code review for science

Over the years, science has evolved many quality control mechanisms, the best known of which is probably peer review. This first appeared nearly 300 years ago, and was initially put into place to allow trusted peers to scrutinize new findings, ideas, and societal implications as part of the publication process.

Here at the Mozilla Science Lab, we’re interested in finding ways of pushing the limits to what we think of as “science on the web” and instilling better practices for open, reproducible research. We aim to do this through community building, skills training (like Software Carpentry), and building prototypes and running pilots with other groups and organizations.

Our latest experiment is exploring a model for code review in science. As research becomes increasingly dependent on computation, we want to see whether the code review practices that are now commonplace in open source software development will also work in research.

Why code review?

The code used to produce the results in a paper is not usually reviewed when a paper is published, beyond a basic “sense-check” (or, as a colleague put it to me, “plausibility versus defensibility, not reproducibility”).  As code follows the trajectory of data in being integrated and recognized as a first-class research object, we want to work with the scientific community to figure out how scientists can and should check each other’s computational methods without increasing the time and cost of reviews unsustainably.

How this will work

The staff at PLOS Computational Biology helped us identify a set of already-published papers that include snippets of code in MATLAB, Python, R, and C++ that may be suitable for code review. We’ve put the code in front of a group of Mozilla engineers who regularly do code reviews for Firefox, mobile and for other programs.

These engineers aren’t scientists (though some may be interested in the underlying research).  Instead, they are skilled professionals who don’t have the domain expertise of the scientists, who in turn are not full-time software developers.

Over the course of the next month, each Mozilla engineer will review code from one or two papers. Their reviews, and their insights into the review process, will be captured in ReviewBoard, so that we can have a core collation point for the findings. Reviewers will have the ability to flip through the rest of the paper should they so choose, but this is really about the code snippet itself.

Once they’ve finished reviewing, we’ll analyze the findings, look for any patterns or interesting outliers, as well as examine how the actual commentary on the code itself was done, with the aim of using that as a model for code review in the future.

What we’re testing

This experiment is a means to explore the following:

– What does code review from a software engineer outside of academia look like? How do they approach the task?

– To what extent is domain knowledge needed to do a successful code review? Is the code parseable by someone outside of that discipline?

– What lessons can be learned about code review, possibly to influence and enhance traditional peer review?

– Does this process surface issues around best practice in writing software and code? If so, what are those issues?

– Following the review, how useful is the markup to the author? Does this feedback help them in their work? How can we change those norms?

We’ll be writing more about this experiment as the reviews continue, and hope to have our analysis completed next month. It’s just a first step, but we believe that this could help change the way researchers, administrators, publishers and funders think of code in scholarly research, and possibly unveil some deep-seated issues surrounding how code is recognized, maintained and shared in the scientific community.

 —

Have a question or would like to join in on the conversation? Drop us a line at sciencelab@mozillafoundation.org or add your comments here. We’d love to hear your thoughts. Also, have an idea you’d like explored. Let us know.

And special thanks to our Mozilla volunteers, Marian Petre from the Open University, Mike Hoye from Mozilla and Greg Wilson from Software Carpentry.

Advertisements