Data Fabrication Retraction Rates 2026: Country Rankings

TL;DR

D6 (data fabrication) rankings from AMI v1.5. Built from Retraction Watch (69,911 records, 5,390 misconduct-linked) normalised by publication volume from OpenAlex. China (D6=100) anchors the top; Nordic and Q1 countries score lowest. Detection-incidence confound applies.

Full D6 rankings

Top — highest D6 scores

Middle — moderate D6 scores

Lower — lower D6 scores

Bottom — lowest D6 scores

Note: Colombia, Argentina, and Greece show 0 because their absolute Retraction Watch presence is very low relative to publication volume. The 0 reflects the rescaling rather than literal zero retractions.

Methodology

The D6 dimension is built from Retraction Watch data:

Step 1 — Filter to misconduct-linked retractions

Retraction Watch records include retractions for many reasons (errors, duplicate publication, ethics issues, requests, misconduct). The AMI methodology filters to misconduct-linked retractions:

• Fabrication
• Falsification
• Image manipulation
• Plagiarism (in research context)
• Fraud
• Manipulation of peer review

The filter retains ~5,390 of the 69,911 total retractions.

Step 2 — Country attribution

Each retraction is attributed to a country based on author affiliations. Multi-country papers are attributed proportionally.

Step 3 — Normalise by publication volume

Retraction counts are divided by total publications from OpenAlex for the same country and time period (2016–2025 [verify exact window]). The result is retractions per 10,000 publications.

Step 4 — Rescale

The rates are rescaled to 0–100 across the 39-country set, with China's rate (the highest) anchoring 100.

Why China scores 100

China's D6 score of 100 reflects:

• Largest research output globally — China is now the world's largest producer of academic papers
• Documented paper mill industry — Fang, Steen & Casadevall (2012, PNAS) established systematic patterns
• Liang et al. (2024) — Nature study found 6.3–17.5% of Chinese papers contain detectable AI-generated content
• High absolute retraction count — even after normalising by publication volume, China leads

The score reflects both the scale of the problem and the relative effectiveness of detection. The actual misconduct rate may be higher still given undetected cases.

Why Russia scores 78

Russia's D6 score of 78 reflects:

• Dissernet documentation — 10,000+ plagiarised dissertations identified
• Retraction Watch entries — Russian-affiliated retractions per publication are elevated
• Limited institutional response — even when misconduct is identified, consequences are often limited

Why some low-output countries score 0

Colombia, Argentina, and Greece score D6=0. This does not mean zero misconduct — it means their absolute retraction count is very small relative to publication volume, and the rescaling produces 0 at the bottom of the distribution.

For these countries, the methodology limitation is that the Retraction Watch dataset is partly biased by:

• Higher-volume publishers being more represented (typically English-language)
• Sting operations and systematic detection efforts being concentrated on high-volume producers

Latin American and Greek research output is smaller in absolute terms; detection effort is correspondingly less concentrated. Future methodology iterations may address this through different normalisation approaches.

The detection-incidence confound

Countries with strong post-publication review culture (Netherlands, Sweden, Germany) catch more cases. The Macchiarini case (Karolinska, Sweden) demonstrated how a country can have very public misconduct that produces strong response — improving R-Score — while also having very low underlying base rate.

Countries with weak post-publication review may have undetected misconduct. Their low D6 scores partly reflect under-detection rather than genuine absence of misconduct.

The AMI methodology applies a detection correction in the dimension scoring but the fundamental confound remains. Cross-country comparison of D6 scores should weight this consideration.

Time trends

The Retraction Watch database has grown substantially over the past decade:

• 2010: ~5,000 records
• 2015: ~10,000 records
• 2020: ~30,000 records
• 2024: ~70,000 records [verify exact 2024 count]

Growth reflects both increasing actual misconduct detection and increasing database coverage. The acceleration in retractions in 2020–2024 was partly driven by systematic paper mill detection efforts following Crossref initiatives.

Sources

• Retraction Watch Database, Crossref/GitLab (April 2026)
• OpenAlex publication counts
• Fang, Steen & Casadevall (2012), PNAS
• Liang et al. (2024), Nature
• AMI v1.5 methodology

Full methodology | Download dataset

Related

Explore the full dataset