Redacted Dataset for Analyzing the Consistency of Retraction Indexing

CORRECTED 2026-07-10: This updated dataset removes items from Retraction Watch that are not in their "Retraction" category and includes MEDLINE data. Our previous dataset unintentionally included non-retracted items because Retraction Watch categorizes items as "Expression of concern", "Correction", or "Reinstatement" as well as "Retraction” in their publicly available Crossref data. Our previous dataset redacted MEDLINE data because we had been unsure about the rights for this public domain data retrieved via Web of Science.

This project investigates retraction indexing agreement among data sources: BCI, BIOABS, CCC, Compendex, Crossref, GEOBASE, MEDLINE, PubMed, Retraction Watch, Scopus, and Web of Science Core. Post-retraction citation may be partly due to authors’ and publishers' challenges in systematically identifying retracted publications. To investigate retraction indexing quality, we investigate the agreement in indexing retracted publications between 11 database sources, restricting to their coverage, resulting in a union list of 83,317 unique items (83,283 after redaction). We also discuss common errors in indexing retracted publications. Our results reveal low retraction indexing agreement scores, indicating that databases widely disagree on indexing retracted publications they cover, leading to a lack of consistency in what publications are identified as retracted. Our findings highlight the need for clear and standard practices in the curation and management of retracted publications.

Pipeline code to get the result files can be found in the GitHub repository
The iPython notebooks used in this project are available in the 'src' directory of the repository: https://github.com/infoqualitylab/retraction-indexing-agreement

The ‘redacted_unionlist_completed_ria_2024-07-09.csv ’ file has been redacted to remove proprietary data, as noted below in README.txt. Among our sources, data is openly available only for Crossref, Medline, PubMed, and Retraction Watch.

FILE FORMATS:
1) redacted_unionlist_completed_ria_2024-07-09.csv - UTF-8 CSV file
2) README.txt - text file