Improving Computational Reproducibility

Computational Reproducibility of research findings may be just as important as their replicability. LMU Open Science Center members Susanne J. Adler, Mario Haim , Frauke Kreuter, Felix D. Schönbrodt and Stefan Feuerriegel contributed to an article in Nature Human Behaviour, where they highlight the significance of computational reproducibility as early validation of scientific claims, leading to a minimization of erroneous conclusions in the collective research field, e.g. in meta-analyses.

Eventually, they argue, reproducible workflows should lead to a stronger trust in science. At the same time individual researchers will become more efficient, by being able to retrace every step of their own work.

The authors put forward a number of recommendations for reproducible workflows focusing on the SEED method, which includes

• Scripting the entire workflow and data analysis in executable code,

• Encapsulating, using relative paths and capturing the working environment, while minimizing dependencies of packages,

• Explaining the script and the variables it contains to human readers,

• Depositing code and data in dedicated research repositories, while anonymizing data and making the data organization easily understandable to others.