Extended definition
Retraction is the formal removal of an article from the scientific record, declaring that it cannot be reliably used as a basis for subsequent research. It is not erasure: the retracted article remains visible with an explicit notice (“RETRACTED”) and active DOI, precisely so prior citations can be traced and future readers are not confused. Causes vary: honest fundamental error (calculation, sample contamination, undetected instrumental issue), substantive misconduct (data fabrication or falsification, plagiarism, image manipulation, fraudulent authorship), and increasingly — demonstrated irreproducibility in sensitive fields. Fang, Steen, and Casadevall (2012, PNAS) analyzed 2,047 PubMed retractions and showed that misconduct (not honest error) accounts for two thirds of the total — fraud (43%), plagiarism (10%), and duplicate publication (14%). Steen (2011, J Med Ethics) offered complementary analysis on intentionality. Retraction Watch (launched in 2010 by Ivan Oransky and Adam Marcus) is the blog/database documenting retractions in real time, today a reference resource. COPE offers a specific flowchart distinguishing retraction, expression of concern, and correction (erratum/corrigendum) — distinct editorial instruments with their own criteria.
When it applies
Retraction applies when the article’s findings are fundamentally unreliable or when there is substantive evidence of misconduct. COPE retraction criteria: clear evidence of unreliable findings (gross error or misconduct), duplicate publication, substantive plagiarism, unethical research conduct (e.g., serious ethical violation, data without consent). It applies to clinical trials with data manipulation — direct regulatory implications. It applies to systematic reviews based on retracted studies — in some cases the review must be republished with revised analysis after excluding retractions. It also applies to cases where authors themselves identify fundamental error (self-retraction) — growing and ethically positive practice. Coverage via Retraction Watch and databases like PubMed (with retraction flag) is modern standard practice.
When it does not apply
Retraction does not apply for minor correctable errors: for those, erratum/corrigendum is the appropriate instrument (typo correction, wrong figure attribution, incorrect patient ID). It does not apply in situations of uncertainty: expression of concern is the intermediate step, declaring substantive doubt while investigation is not yet concluded. It does not apply as a punitive instrument disconnected from evidence: COPE requires formal investigative workflow before retraction. It does not apply to suppress uncomfortable or contested results: scientific disagreement is resolved by published debate, not retraction. It does not apply as literal removal: a retracted article remains accessible, with notice, to preserve historical record.
Applications by field
— Health and biomedical sciences: highest retraction volume; fields with high replicability (cardiology, translational oncology) have more active oversight. — Basic biomedical sciences: the irreproducibility crisis in cancer (Begley & Ellis 2012) led to increased scrutiny and retractions. — Computer science: historically rare retractions; emergence of ML reproducibility problems led to growing cases at top-tier conferences. — Social sciences: retractions in social psychology rose after the replication crisis (2015+); cases like Stapel, Wansink are paradigmatic.
Common pitfalls
The first pitfall is confusing retraction with erratum/corrigendum: retraction removes integral reliability; erratum corrects specific details while preserving findings. The second is citing a retracted article without checking — Retraction Watch and PubMed flags show status; systematic reviews need workflows to handle retractions. The third is assuming absence of retraction = reliable article: many problematic articles are not retracted due to institutional inertia, absence of whistleblower, or lack of consolidated evidence. The fourth is reading retraction notices superficially without understanding the cause: retraction for honest error vs. misconduct has distinct epistemological implications. The fifth is treating retraction as punishment of authors rather than protection of the scientific record: punitive framing discourages honest self-retractions, a practice that should be encouraged.