But the Greatest of These Is...

This post originally appeared on the Software Carpentry website.

In her EAGE keynote earlier today, Victoria Stodden talked about the central role of geophysics in the reproducible research movement. After discussing the problem, she identified five interlocking solutions:

1. Tools
2. Intellectual property barriers
3. Funding agency policy / federal regulation
4. Journal policy
5. Institutional expectations

I think these are all important, but I think they are all less important than something which doesn't appear anywhere in her slides:

0. Training

There are two reasons:

1. If people have the skills to do reproducible research, they can find ways around problems 1-5. (Proof: they're already doing so.)
2. If people don't have those skills, then providing tools, changing intellectual property rules and journal policies, and everything else won't matter.

The obvious response to the second claim is to say that if the tools make it easy, and the incentives are right, people will learn what they need to. But people have been saying that for the last twenty years (or possibly longer—I only started paying attention in the early 1990s), and it hasn't happened. I don't have data to back up this claim, but based on personal experience, and the experience of many other people, I believe that the average scientist is no more computationally literate today than he or she was in 1992. I think we need to accept that osmosis hasn't worked, doesn't work, and isn't likely to work, and that if we actually want to change the way people do science, our top priority has to be giving them the skills they need to implement those changes.