Clear and reproducible Bayesian statistical reports

The Bayesian analysis reporting guidelines (BARG) help make reports clear, cogent, and reproducible.
Published in Social Sciences
Clear and reproducible Bayesian statistical reports

Share this post

Choose a social network to share with, or copy the shortened URL to share elsewhere

This is a representation of how your post may appear on social media. The actual post will vary between social networks

Have you ever been frustrated when trying to read a statistical analysis in a scientific journal article? If you're like me, you responded with an emphatic "yes!" Some annoyances are merely stylistic, but other problems can seriously affect (i) the comprehensibility of the report, (ii) the reproducibility of the analysis, and (iii) the integrity of the conclusions.

To address rampant problems in reporting of statistical analyses, many professionals --journal editors, reviewers, and authors-- agree that it would be useful to have guidelines for reporting the analyses. The need for guidelines is especially true for Bayesian analyses, which are becoming more prevalent in the literature, yet remain relatively novel for many people.

Bayesian analysis reporting guidelines (BARG) are provided in an open access article in Nature Human Behaviour.  The key points of the BARG are summarized in a convenient six-step table, suitable for use as a checklist by authors and reviewers. The BARG are thoroughly explained in the article, and the BARG are extensively illustrated with a detailed example.

The goals of the BARG are to make reports of Bayesian analyses comprehensible and reproducible. By satisfying the points in the BARG, not only are the reports better, but the analyses themselves can be rescued from errors. Moreover, educators and students can use the BARG to teach and learn applied Bayesian analysis.

What kinds of problems can be alleviated by the BARG? Problems can arise in any of the six steps addressed in the BARG:

  • Image of Table 1 of the BARGExplaining the model.
  • Reporting details of the computation.
  • Reporting the parameter estimates, including posterior model probabilities in Bayesian null hypothesis testing.
  • Reporting decisions.
  • Reporting a sensitivity analysis for the prior distribution.
  • Making the analysis reproducible.

For each of the categories above, I've encountered many instances of problems in reports from other people and in write-ups of my own. When you apply the BARG, as an author or as a reviewer, you too will probably find many instances of problems that can be addressed prior to publication.

The BARG will help achieve our collective goals as a scientific community and your individual goals to make clear and impactful contributions to science. If you are an author or a reviewer, try using the BARG. Encourage editors of journals to endorse the BARG in their instructions to authors and reviewers. 

Please sign in or register for FREE

If you are a registered user on Research Communities by Springer Nature, please sign in