Tool for assessing the level of agreement between two raters following screening in ASReview. You can take a look at an example report on rpubs.
When knit
within R/RStudio, the .Rmd file will automatically produce a summary report (.html) on the level of agreement between two raters.
Step 1: Download and save the .csv outputs from ASReview into the 'data' folder.
Step 2: Open up the asreview_irr.Rproj
in RStudio.
Step 3: Open report_example.Rmd.
Step 4: Knit the report and when prompted pick the relevant data files.
Upon opening asreview_irr.Rproj
you might be prompted by warnings from renv. We would recommend matching the R version stated in the lockfile and executing renv::restore()
before building the report.
A number of descriptives are produced in the report, including the number of abstracts reviewed, the number flagged as relevant, the number left unreviewed, inter-rater reliability statistics (Kappa, Gwet's, Krippendorff's), and the number of abstracts for which researchers A and B disagreed (e.g., A said relevant, B said irrelevant). In the final case, tables are produced which detail the specific literature involved. Waffle graphics provide a visual overview of screening and rater agreement.
The report also identifies the number of cases in which there was disagreement 'implicitly'. This is when one researcher rates an abstract as relevant, but the other researcher never reviews that abstract. Ideally, this number is zero for both raters. The report is a work in progress. We would appreciate any checks, feedback, comparisons to your own findings, and suggestions for other features the report could include.
If you do use the tool in a paper, please cite it using the dropdown options above (under the About
tab).
Thanks to Roos Jansen, Amy Nivette, Amina op de Weegh, Stijn Ruiter, and the ASReview team for feedback on improving this report.