Methodology

How we compute these numbers.

Every figure on this site is a plagiarism-retraction rate: how many retractions for plagiarism a country, institution, or discipline accrued, divided by how much it published. This page documents exactly how each number is produced, the confidence we attach to it, and the biases you should keep in mind when citing it.

Step 1

How we count

The numerator comes from the Retraction Watch database (hosted by Crossref, ~70,000 records). We keep only records whose reason carries the strict code plagiarism of/in — an explicit plagiarism finding, not softer signals like “euphemisms for plagiarism” or duplication. Across 2015–2024 that is 1,879 distinct records.

The join that matters

Institution and discipline attribution joins on the paper DOI via OpenAlex — never on Retraction Watch’s free-text affiliation field. OpenAlex has already resolved each work’s affiliations to canonical ROR identifiers and tagged its discipline, so a DOI lookup returns the correct institution and field. Resolving the free-text affiliation string instead leads with the department and matches fragments to the wrong institution — false attribution we explicitly prohibit.

~91% of plagiarism records carry a DOI; ~90% of those resolve to a ROR institution, so roughly four in five records attribute to a canonical institution.

SourcesRetraction Watch (retraction events + reasons), OpenAlex (publications, affiliations, disciplines), Crossref (DOI metadata), ROR (institution identity). All free and openly licensed.

DenominatorOpenAlex publication counts for the same entity and window. Country counts use author-affiliation country; institution counts use the ROR identifier.

Step 2

How we normalize

Raw counts mislead: high-output research systems accumulate more retractions simply because they publish more. By raw count China leads with 564; by rate it sits at 7.0 per 100k. We therefore report rates, and the system is built so it cannot rank entities by absolute count.

Rate per 100,000 publicationsThe headline metric for every entity. Comparable across entities of very different size.

Wilson 95% confidence intervalsEvery rate carries an interval. Small numerators produce wide intervals — a signal to read the rate cautiously rather than as a point estimate.

Rolling windows, not single yearsRetractions lag the original paper by years, so a single year is too sparse to chart. The default unit is a multi-year window.

Volume floorEntities below 100,000 publications in the window are omitted — too few publications make the rate statistical noise.

Step 3

Source quality, on every number

Each statistic is tagged with how it was produced, so you can weigh it before citing.

Direct measurement

Counted from primary records (Retraction Watch).

Curated aggregate

Compiled from multiple primary sources.

Modeled / inferred

Statistical extrapolation with stated confidence.

Survey-based

Self-reported, with sampling limitations.

Read this first

Known biases & limits

We measure detected plagiarism, not committed plagiarismThese are retractions that were caught and recorded. The gap between detected and actual rates is real and unmeasured; we never present these numbers as total plagiarism rates.

English-language indexing biasRetraction Watch and OpenAlex over-represent English-language journals. A country with strong oversight may detect (and thus record) more, not necessarily commit more.

Uneven coverageSource coverage varies by region; we omit entities whose coverage is too thin to support a defensible rate.

Denominator is all works (for now)The current denominator counts all OpenAlex works, not only journal articles, which slightly understates rates. A tighter article-only denominator is a planned refinement.

For researchers

Reproduce it yourself

The whole pipeline is two free, keyless HTTP calls plus a join. No API keys, no paid tiers.

# 1. Retraction Watch database (~65MB CSV)
curl "https://api.labs.crossref.org/data/retractionwatch?mailto=YOU@example.com" \
  -o retraction_watch.csv

# 2. OpenAlex publications by country, 2015–2024 (the denominator)
curl "https://api.openalex.org/works?filter=publication_year:2015-2024\
&group_by=authorships.institutions.country_code&per-page=200&mailto=YOU@example.com"

# 3. Keep records with a strict 'plagiarism of/in' reason, attribute via DOI,
#    divide by publications, ×100,000. Wilson interval for the CI.

The reference implementation lives in the project’s stats/ pipeline.