A retraction is the formal withdrawal of a published paper — the system's mechanism for removing unreliable or dishonest work from the record. The paper is not deleted; it stays online with a watermarked retraction notice explaining who withdrew it and why, because citing scientists need to know their references failed. The scale has grown fast: 2023 set a record with more than 10,000 retractions across scholarly publishing, driven largely by paper-mill products at major publishers, according to data from the Retraction Watch database, which passed 50,000 total entries in 2024 and now operates as a public resource under the Crossref infrastructure.
This is an explainer on the mechanics of retraction, not guidance for any specific dispute; researchers facing a case should consult their institution and the Committee on Publication Ethics.
Who can retract a paper?
Three parties, with very different credibility. The authors themselves retract, most often after discovering an honest error — a mislabeled sample, a broken reagent, a coding mistake. Journals retract after investigation, usually prompted by a whistleblower, a failed replication, or a post-publication critique on the PubPeer commenting platform. Institutions retract — technically, they recommend retraction after an inquiry into misconduct; journals follow the finding. The chain matters: a journal cannot compel an institution to investigate, and slow institutional inquiries are a leading reason cases stall for years.
What triggers a retraction?
Reasons fall on a spectrum of blame. At the honest end: computational errors, contaminated cell lines, data-management mistakes. At the dishonest end: fabricated or manipulated data — duplicated Western blot bands being the classic molecular-biology tell — undisclosed conflicts of interest, plagiarism, and fake peer review. COPE's retraction guidelines, the field's standard, say retraction is for unreliable findings regardless of author intent, which is why honest-error retractions exist and should carry less stigma than misconduct cases. In practice, the stigma does not split so neatly, and many scientists treat any retraction as career damage — one reason authors contest notices and lawyers get involved.
Why does it take so long?
Due process is the official answer, and there is truth in it: institutions must give accused researchers a chance to respond, data must be examined, and multi-author papers require untangling who did what. But analyses of retraction timing find median delays stretching from months to years, and well-documented cases — including stem-cell papers cited hundreds of times — have taken more than a decade. Editors balance procedural fairness against the cost of leaving bad science in the citation stream the whole time. Meanwhile citations keep accruing: studies tracking retracted papers consistently find they continue to be cited as if valid, sometimes hundreds of times after the retraction notice appears.
Related stories: Preprints Changed Who Controls Scientific Priority · Predatory Journals: Nine Red Flags Before You Submit a Paper.
What is an expression of concern, and when is it used?
Between published and retracted sits the expression of concern, a notice editors attach when they have serious doubts but no completed investigation. It flags the paper as under question without withdrawing it, and it exists because institutional inquiries take months or years while the citation stream runs continuously. The device is controversial in both directions: researchers whose papers carry concerns describe them as lasting stigma without due process, while integrity researchers document cases where expressions of concern sat on papers for a decade before retraction, during which the papers accumulated citations as if healthy. Practice guidance from publishing-ethics bodies tells editors to set timelines and convert concerns into retractions or clearances rather than letting them linger, and databases now display concern notices prominently, but the middle state remains the system's least governed territory.
What do retractions cost co-authors?
Retraction is blunt about collective responsibility. A paper withdrawn for one author's fabrication stains every listed author, including those who contributed sound work and knew nothing, which is one reason multi-author retraction cases generate disputes and, occasionally, lawsuits. Authorship-order conventions complicate it further: the last and first positions carry credit in most systems, and integrity investigations typically focus on those roles, but mid-list authors still face database records flagging their participation. The practical response from some institutions is graduated — distinguishing authors who cooperated from those implicated — and major databases now allow records to note who was cleared. Whether that granularity reaches hiring committees is another matter, and early-career co-authors bear most of the exposure, a structural problem the retraction system acknowledges without having solved.
What changed recently?
Infrastructure. The Retraction Watch database became free and machine-readable in 2023 via Crossref, letting reference managers and publishers flag retracted citations automatically. Major publishers began mass-retracting mill-linked special issues — hundreds of papers at a time — rather than case by case. And funders started asking applicants to disclose retractions explicitly. The unresolved frontier is retraction at scale: when a paper mill produced thousands of manuscripts, the individual-paper process strained, and publishers wrote new batch procedures to cope.
What would make the system trustworthy?
Faster institution responses, standardized notice language that separates honest error from misconduct, and automatic citation flagging everywhere references are read. The measurable benchmark: shrink the gap between suspicion and notice, and stop post-retraction citations. Both are countable, and both remain open.
