I've been thinking about the link between research reproducibility and the way students perform the work in their problem sets. The emphasis in problem sets and grading is the presentation of the correct answer. Many instructors demand some clarity in the methods used to provide the result but my suspicion is that they are in the minority. This values the result, usually a single number, above the methods and assumptions used to arrive at that number. Most of the time however, this numerical result is the basis of some decision, and the purpose of the calculation is to provide confidence in the decision.

This semester, I'm trying an approach where the problem sets are graded pass or resubmit and the central goal of the problem set document is to convince the reader that the problem set provides a reasonable estimate with defensible assumptions. It will be jarring for both the student and the instructor to return work with the correct answer as unacceptable, but I want to place the emphasis on reproducibility.

For instance, in the energy space, there have been several studies on the levelized cost of electricity for several technologies such as wind, solar, coal, nuclear that have provided divergent estimates for these costs. Some of the coverage of these studies have only mentioned the numerical results, without reporting the models and assumptions. My goal is for my students to reflexively ask about the underlying assumptions after they see a reported value by expecting the same of their classwork. The other goal is for the author to provide enough detail so that it is straightforward for the reader to reproduce the calculation. Many in the research community place and importance on this clarity and reproducibility and one way I can support that effort is by incorporating it into my class work. Toward the end of the semester I'll provide an update based on my experience.