What Is Retraction Watch? A Guide to the World's Retraction Database

TL;DR

Retraction Watch is the world's largest database of retracted scientific publications. Founded in 2010 by Ivan Oransky and Adam Marcus. Contains 69,911 retraction records as of April 2026; 5,390 are misconduct-related. Now hosted publicly on GitLab via Crossref. The AMI's D6 dimension is built from this data.

What Retraction Watch is

Retraction Watch is two things:

1. A blog at retractionwatch.com, founded in 2010, that reports on individual retraction cases, broader trends in research integrity, and the operation of the scientific publishing system

1. A database that catalogues retracted scientific publications worldwide, with structured fields for author, journal, retraction reason, country attribution, and date

The database is the more relevant source for the AMI's D6 dimension.

Who runs it

• Ivan Oransky — co-founder, executive director of the Center for Scientific Integrity (the organisation behind Retraction Watch). Vice president of editorial at Medscape; longtime science journalist and medical editor.
• Adam Marcus — co-founder. Editor of Anesthesiology News; longtime medical journalist.

The Center for Scientific Integrity, the non-profit that operates Retraction Watch, is supported by foundation funding (Arnold Foundation, Helmsley Charitable Trust, MacArthur, others) [verify specific funders].

The database

As of April 2026, the Retraction Watch database contains:

• 69,911 total retraction records
• 5,390 records classified as misconduct-related (fabrication, falsification, fraud, image manipulation)
• Coverage from approximately 1990 to present

Each record includes:

• Article identifiers (DOI, PubMed ID where applicable)
• Authors and country attribution
• Journal and publisher
• Original publication date and retraction date
• Retraction reason (multiple categories)
• Notice text where available

The Crossref / GitLab partnership

For most of its history, the Retraction Watch database was a paid resource — accessible to academic institutions through subscription but not freely available to the public.

In 2023, Crossref entered a partnership with the Center for Scientific Integrity to make the database openly available. The data is now hosted on GitLab and updated periodically. The partnership has substantially expanded the database's accessibility for researchers, journalists, and instruments like the AMI.

How the AMI uses it

The AMI's D6 dimension is built directly from Retraction Watch data:

1. Filter — the full database is filtered to misconduct-linked retractions (fabrication, falsification, fraud, image manipulation, plagiarism in research context)
2. Country attribution — each retraction is attributed to a country based on author affiliation; multi-country papers are assigned proportionally
3. Normalisation — retraction counts are divided by total publications from OpenAlex for the same country and time period
4. Rate calculation — produces retractions per 10,000 publications per country
5. 0–100 scaling — the rates are rescaled across the 39-country set, with the highest-rate country (China) anchoring 100

The result is each country's D6 score on the 0–100 scale used in the AMI methodology.

Why this matters

Retraction Watch is the only globally comprehensive systematic record of research retractions. Without it, the AMI's D6 dimension would have no live data source. The Crossref partnership making the data public was a precondition for the AMI being able to build D6 from current data.

Cross-checking specific famous cases against the database is straightforward — the Hwang Woo-suk retractions, the STAP cell retractions, the Stapel cases, the Macchiarini case all appear in the Retraction Watch records.

Limitations of the data

The Retraction Watch data has known limitations:

• Detection-incidence confound: only detected and retracted papers are recorded. Actual fabrication that does not result in retraction is missing.
• Country attribution complexity: papers with authors from multiple countries are attributed proportionally; methodology choices affect specific country scores.
• Retraction lag: retractions often happen years after publication. Recent fabrication is under-represented.
• Reason coding inconsistency: retraction notices use widely varying language; classification into misconduct categories requires interpretation.

The AMI methodology applies a detection correction to partially address the first limitation but the fundamental challenge remains.

Sources

• Retraction Watch website
• Retraction Watch Database on GitLab
• Fang, Steen & Casadevall (2012), PNAS
• AMI v1.5 methodology document

Full methodology | Download dataset

Related

Read the full methodology