Posted: October 10, 2017

Writing Prose and Writing Code: Unrecognized Causes of Reproducibility Failures


The “crisis” in scientific reproducibility continues to vex, alarm and befuddle. Although the brute fact of failures to replicate expensive biomedical studies is well established, its causes are not. In addition to methodologic sloppiness or failures of rigor, frank misconduct, and what has been termed “secret sauce,” we hypothesize that two ubiquitous and practical features of scientific inquiry also have a role in failures to replicate studies. First, scientific prose is often of low quality, suffers from hedging, boasting, passive voice and other tricks, and is rarely among the skills included in the training of researchers. Second, investigators who write, modify and use computer software are untutored in the role of data analysis and its tools in the responsible conduct of research (RCR). We have therefore modified our institutional RCR training to include prose writing and code writing in addition to traditional components of the curriculum: data management, publication and authorship, human subjects protection, animal research, etc..  We have also promoted special events and presentations to address and advance the idea that prose and code are un- or under-recognized contributors to failures to replicate others’ results.


Curriculum, Training, Replication, Writing, Responsible Conduct of Research, Code, Reproducibility


Joanna Johnson, PhD, University of Miami

Kenneth W. Goodman, PhD, University of Miami