The PRIMAD Model of Reproducibility
Norbert Fuhr
How published:
Dagstuhl Seminar 16111 - Rethinking Experimental Methods in Computing

For describing different degrees of reproducibility, the participants of Dagstuhl Seminar 16041 started from a model referring to the components of an experiment: R: the research goal, M: the method proposed for achieving this goal, I: the implementation of this method, P: the platform on which the implementation is run, D: the data (input + parameters) used in the experiment, and finally A: the actor performing the experiment. When a researcher tries to reproduce an experiment, he should specify which components are changed, i.e. 'primed': R -> R' repurpose for a new research goal, M -> M': a new method, I -> I': alternative implementation, P -> P': different platform, D -> D': other input/parameters. Finally, other important aspects of reproducibility are consistency of experimental results, and transparency, i.e. the ability to look into all necessary components to verify that the experiment does what it claims.

BibTeX entry

Fulltext as PDF (Presentation)