🔖 Statistical code in a high-impact medical journal
A journal started asking authors to submit code with their manuscripts. They then analysed the next 314 papers accepted
87% denied using code, even when publishing substantial statistical analysis
10% used code but refused to share it with the journal
For the few that provided code, none scored even moderately on basic quality criteria
Assel & Vickers
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6705117/
#MedMastodon
#OpenScience
#ResearchWaste
#doi:10.7326/M17-2863
This article should make all scientists feel bad.
But! There is an obvious fix!
Data work has to be scripted, reviewed, tested, & verison controlled. All of which is already done with free tools, and widely-tested methods in open source development.
The thorny part is that our bosses don't know how to do this. And our journals don't want their shoddy work exposed. So political activism on #ResearchAssessment is the hard bit.
Clear, funny overview by @rlmcelreath here
https://www.youtube.com/watch?v=zwRdO9_GGhY
@JessButler @rlmcelreath I think the other issue is the time investment. Writing good code, fully documenting it, making it reusable, takes a lot of time for (currently) little reward.
@nathanhuneke @JessButler @rlmcelreath https://royalsocietypublishing.org/doi/10.1098/rspb.2022.1113
This is an interesting analysis of the topic
