Corrections, retractions, and expressions of concern
Published work that turns out to be wrong or doubtful is answered by a correction, a retraction or an expression of concern. Handling one turns on which notice fits the problem, who starts it, what an author does about their own paper and how a concern is investigated fairly. It also turns on what each notice has to say and who may read it, how a correction can itself go wrong, and how retractions are carried out in practice and where their procedures fall short.
Correction, retraction or expression of concern
Work published in a scientific or medical journal sometimes needs correction or, much worse, retraction. Most journals have established procedures for handling corrections and retractions, many of them based on the retraction guidelines of the Committee on Publication Ethics (COPE), which describe when and how corrections and retractions should be applied. The guidelines of the International Committee of Medical Journal Editors (ICMJE) should also be observed. COPE's retraction guidance also sets out when a journal should issue a correction or an expression of concern. When a correction or retraction becomes necessary, the author should liaise with the journal as early as possible. Online publication and literature databases have made a published article considerably easier to amend, but journals have to provide the systems and instructions that ensure readers are alerted to any change.
A correction addresses a mistake in the original publication that escaped the attention of the author and the editor. Corrections usually concern errors in calculations, chemical structures, doses, or the spelling of drugs and other chemical compounds, and a correction may also be needed when the list of authors or other contributors was incorrect or incomplete. A correction is permitted only where it does not alter the main findings and conclusions and the article is otherwise sound. Corrections should be avoided at all costs, because they affect the credibility of the scientific work and of the researchers responsible alike.
The ICMJE's recommendations note that errors serious enough to invalidate a paper's findings may require retraction. A title in which familial had been published as familiar, because the Swedish word familjär can carry both senses and the authors picked the wrong one, was by contrast a relatively trivial error: it did not lead to data being misinterpreted or erroneous conclusions being propagated, and the authors published an erratum and the title was corrected.
Retraction means, in the dictionary sense, to draw something back in. In publication terms, a retracted article is one that has been removed from Medline after a journal issues a retraction notice. The journal's editor decides on retraction on the ground either of an error by the author or the journal or, more seriously, of author misconduct.
A retraction is considerably more far-reaching than a correction. It implies that the main findings, conclusions or implications of a study are no longer valid, so it indicates that the work should not have been published and that its outcome must not be used as the basis of further research. The commonest reasons for retraction are scientific misconduct, including plagiarism and duplicate or concurrent publication, but in some cases a paper is retracted on the grounds of a genuine error that seriously affects the study outcome.
Journal editors may issue an expression of concern when they have reason to doubt the credibility of the research or the ethical conduct of the study, or suspect any form of publication misconduct. An expression of concern is particularly due where the authors' institution is not willing to investigate the alleged misconduct, or where there is good reason to think an investigation would not be objective or conclusive.
Like a retraction notice, an expression of concern should be clearly linked to the original article in electronic databases and should specify the reasons for the concern. If the work is subsequently shown to be credible and reliable, an exonerating statement is added to the expression of concern; if further evidence corroborates the concern, the expression of concern is replaced by a retraction notice.
Authors from The Lancet have argued that for papers that contain serious errors but are otherwise valuable, simple retraction is too punitive and an erratum inadequate. They propose instead a system of retraction and republication: the original version is retracted and a corrected version published in its stead, with annotated copies of both versions, errors highlighted, maintained in an online appendix.
The 11th edition of the AMA Manual of Style adds information about managing errors and corrections, because corrections are important to the integrity of the published literature. Before this edition there were only two kinds of error: minor, inconsequential ones corrected online without a correction notice, and substantive ones requiring a correction notice and a corrected article reciprocally linked to the original. A pervasive error is a new kind. It is serious but inadvertent and affects many of the data reported, as when a coding problem during data collection or a miscalculation causes extensive inaccuracies that need correcting in the abstract, text, tables, figures or supplementary material. Such a mistake, a coding error for example, occurs before the data are analysed and propagates incorrect numbers throughout the manuscript.
A pervasive error can be handled in three ways. A letter of explanation and correction serves when there are no major changes, no conclusion or interpretation is affected and there is no statistically significant change in the results. Retraction serves when the changes are substantial and the results, interpretations and conclusions change so that the science is no longer valid. Retraction and replacement, a new option for articles whose pervasive errors change the findings, serves when the changes are substantial and the direction or significance of the results, interpretations and conclusions changes but the science is still valid.
With retraction and replacement, the article is replaced with a corrected version that keeps the original DOI and carries no retraction watermark, a notice on the corrected article says that the original was replaced, the authors' complete explanation is published as a linked letter, and PDFs of the original with the errors highlighted and of the corrected article with the corrections highlighted are published in an online supplement. It does not carry the stigma of a full, "do not use" retraction, and it lets the literature be corrected while the science goes on benefiting the community.
Who starts a correction or retraction
A retraction may be initiated by the authors themselves or by their institution, or enforced by the journal's editors. A deputy editor of Science divides authors, where retraction is concerned, into the good, the bad and the ugly: good authors initiate the process themselves, usually by writing to the editor to request retraction because they have discovered an error; bad authors refuse to sign a retraction when one is appropriate; and the ugly refuse to retract even after their institution has found misconduct. At Science, as of 2012, the good outweighed the rest, but the reaction of the medical journal editors who heard him suggested that this might not be true of their journals.
What the author does depends on where the paper is. An author who discovers after submission that the content is incorrect or fraudulent, such as flawed evidence used during the research, must notify the editor immediately. Before acceptance or during review, withdrawal is usually possible and probably encouraged by the editor; once the article has been published, the author should prepare a letter on behalf of all contributing authors outlining the reasons for the retraction, which the journal will normally publish.
Investigating a concern about a published article
COPE sets out in flowcharts how an editor handles a concern about a published article.
For suspected plagiarism, thank the reader, secure documentary evidence and check the degree of copying. For minor copying without misattribution of data, contact the author, request a discussion of publishing a correction containing the missing reference, and inform the reader and the plagiarised author if different. For clear plagiarism, request an explanation from the corresponding author with the documentary evidence: an unsatisfactory answer or an admission of guilt leads to contacting all the authors and discussing retraction with the other editors involved, and a satisfactory explanation leads to considering a correction. If the authors do not respond, try to contact all of them and then the institution's research-governance officer, keep contacting the institution every 3–6 months if necessary, and consider other authorities. Inform the authors of your action, and the reader and the plagiarised authors of the outcome.
For suspected fabricated data, thank the reader, explain that you will investigate, consider a second reviewer, and ask the author for an explanation without making direct accusations. A satisfactory explanation, such as the detection of an honest error, leads to an apology and a correction if necessary, with the reader informed. An unsatisfactory answer or an admission of guilt leads to requesting an institutional investigation after informing the authors. If the authors do not respond, contact the other authors, then the institution's research governance, then a regulatory body if necessary. If the authors are cleared, apologise; if they are found guilty of fabrication, publish a retraction; with no response or an unsatisfactory response from the institution, publish an expression of concern. Inform the reader of the outcome.
For an undisclosed conflict of interest, thank the reader, investigate and contact the authors; supplying the journal's policy and definition may help. If the authors supply details, explain the seriousness of the omission and publish a correction to the competing-financial-interests statement if required. If they deny a conflict, explain the policy and obtain a signed statement about relevant conflicts if one was not obtained before. Inform the reader of the outcome.
At Science, as of 2012, the mean time from initiation of an investigation to retraction was 2.8 years, with a maximum of eight. An expression of concern, of which the journal had published eight, indicates that an investigation has been initiated or that the journal has worries. Alerts can reach the journal from an anonymous whistleblower, the corresponding author, co-authors, an identified or unidentified correspondent, an author's institution, or a reviewer.
Investigations have to be handled sensitively, bearing in mind language barriers, the involvement of multiple institutions and countries, and the human element of the co-authors. Such investigations may have provoked attempted suicide and hospitalisation. Intense media scrutiny can also lead a journal to act too fast.
What a correction notice contains
A 2008 AMWA Journal erratum gives the issue and page of a media review in which the title of a book had been spelled incorrectly, and gives the correct title.
A 2012 AMWA Journal erratum states that the state abbreviation for Kansas was incorrect in a figure on a named page of a recent issue and in a thumbnail on that issue's cover, and gives the correct abbreviation.
A journal can correct even a word in its own advertising. A 2010 AMWA Journal erratum records that the word "brats", short for Bratwurst, had been misspelled in the advertisement for the association's annual conference on a named page of the March issue, and that the journal regrets the error.
An erratum can correct a misdescription of another article as well as a misprint. EMWA's September 2024 issue corrects two errors in its June 2024 issue. The first, a production error, was in the guest editors' editorial, which had stated that an article on communication, conflict management and critical thinking included observations from a related survey its authors had conducted; the notice says that was incorrect, and that the text had been corrected in the online edition of the article, giving its DOI. The second error is an identifier: the print issue gave an incorrect DOI for the From the Editor column, and the notice gives the correct one and states that the online version carries it.
An erratum can correct a secondary report of clinical guidance. A 2008 AMWA Journal erratum corrected two sentences, on the blood pressure at which pharmacotherapy is indicated and on the recommended high-density lipoprotein cholesterol level, in the discussion of major risk factors in a summary of a conference session on cardiovascular disease prevention in women. The notice gives the page of the error and reprints both sentences as they should read, with italics indicating the correction.
A correction can supply an omitted byline. A 2007 AMWA Journal erratum for a profile whose byline had been inadvertently omitted names the author, her position and the office she held in the association profiled, and thanks her for the contribution.
Information missing from an article's acknowledgments can be corrected by updating its PDFs and saying so in the next issue. An editor's note in the Summer 2023 AMWA Journal records that the author of an article in the Spring 2023 issue on the grantsmanship hurdles of early-career clinician-scientists had noted that grant information was missing from its acknowledgments, and that the individual PDF of the article and the full-issue PDF galleys were updated on 7 April 2023 to add it.
A correction can be made in the online version of the journal. The brats notice records that the copy had been corrected in the online version, and a 2011 AMWA Journal erratum records that a text heading in an earlier article had been formatted incorrectly and had been corrected in the online version.
Errors made while correcting an error
A columnist can correct her own column in the next one. A 2009 usage column reports that its author's slogan, look it up, had come back to haunt her: in her previous column, which she identifies by volume and issue, she had given Yogi Berra's first name as Clayton, and she gave his birth name as Lawrence Peter Berra.
The correction was itself wrong, and the next column had to correct it. That column named the time-honoured tradition of making another mistake while correcting an initial one, and reported that Berra's birth name was Lorenzo Pietro, later anglicised to Lawrence Peter.
A series of classified advertisements that the advice columnist Dear Abby had used long before, reprinted in that column, shows such a cascade. The first advertisement offered a sewing machine for sale with a phrase in the wrong place, so that the reader was asked to telephone after 7 p.m. and ask for a woman "who lives with him cheap". The next day's correction moved "cheap" to the right sentence and "after 7 p.m." to the wrong one. The correction to that correction lost most of the time of day and introduced a typographical error changing "lives" to "loves". On the fourth day the advertiser published a notice of his own saying he had smashed the machine, that his telephone had been disconnected, and that he had not been carrying on with the woman, who until the day before had been his housekeeper and had quit.
Free access to correction notices
In 2014 a synthetic chemist blogging as See Arr Oh publicly objected to being asked by the American Chemical Society's publications arm for $35 to access an article correction, arguing that corrections are not publications and should be free; the publisher replied that corrections are considered additional materials, with thanks for the feedback. To one medical-writing columnist, charging for a correction is like buying a faulty product and then being asked to pay the same amount again for its repair. The columnist's rejoinder to the publisher is that a correction is information that should have been correctly presented in the original article, not a bonus or an upgrade.
In a Retraction Watch poll on the question, 386 of 463 voters (83%) thought at the time of reporting that all correction notices should be freely available; 11 voters (2.4%) did not agree; and a significant minority held that only corrections relating to significant errors, not spelling and the like, should be free. The columnist finds that last option puzzling, asking who could imagine paying to read corrections to spelling.
COPE recommends that all retraction notices be freely available to all readers, and as of 2014 had apparently issued no equivalent statement for correction notices. The columnist feels strongly that all such notices should be free to all, since publishers profiting from mistakes in their own journals is hard to swallow.
What a retraction notice states
A retraction should be published formally, and its notice should state the reason for the retraction explicitly, to help readers distinguish misconduct from genuine error. An apology from the authors for an honest error may help to limit the damage but will never undo it. Retractions themselves are occasionally incorrect or insincere, and may have been made for personal gain or under external pressure, so a reader still has to rely on their own judgment and common sense.
COPE's guidelines say that a notice of retraction should:
- be linked to the retracted article wherever possible, in all electronic versions
- clearly identify the retracted article, for example by including its title and authors in the retraction heading
- be clearly identified as a retraction, distinct from other types of correction or comment
- be published promptly, to minimise harmful effects from misleading publications
- be freely available to all readers, not behind an access barrier or available only to subscribers
- state who is retracting the article
- state the reason or reasons for retraction, so that misconduct can be distinguished from honest error
- avoid statements that are potentially defamatory or libellous
The wording of a retraction matters: it should be informative and state why the retraction is being made, rather than blandly asserting that the data are no longer reliable. A notice retracting a 1988 Science paper is an example of the bland kind: it reads, in substance, that the author had decided to retract the paper and that its data should no longer be considered reliable.
The Lancet's 2010 notice retracting the 1998 paper that suggested a link between the measles, mumps and rubella vaccine and developmental disorders in twelve children states that several elements of the paper are incorrect, contrary to the findings of an earlier investigation. It states that the claims that children were consecutively referred and that investigations were approved by the local ethics committee have been proven false, and that the paper is therefore fully retracted from the published record. A medical journalist criticised the journal for poor medical communication: it used language likely to be impenetrable to anyone not versed in the scientific literature, the retraction would have served public health better by stating its point a little more clearly and directly, and this was not a matter of an obscure scientific error. The retraction, he argued, should have been clear enough to play a role in the public debate about vaccines and autism.
How retractions are carried out
A retraction may cover part of a paper or all of it; partial retractions are very rare and often relate to interpretation. In one case at Science a partial retraction gave way to a full one. The journal published an expression of concern in July 2010 about a paper claiming to have found an infectious retrovirus, XMRV, in the blood of patients with chronic fatigue syndrome: three laboratories had contributed to the study, and at least ten others had been unable to detect the virus. In September the authors published a partial retraction of a figure, a supplemental figure and a table, all presenting data from contaminated samples. The journal then lost confidence in the paper altogether; most authors agreed to a full retraction, but the editor and all the authors could not reach consensus on its wording, so in December 2011 the editor took the rare step of retracting the article himself, stating that multiple laboratories including the original authors' had failed to detect the virus in these patients; there was also evidence of poor quality control of the experiments. A complicating element was pressure against retraction from patient groups who had welcomed the paper as allaying scepticism about the existence of the disease.
In several notorious cases of mass retraction, many articles have been linked to a single researcher. In one, 88 research articles by the German anaesthetist Joachim Boldt were recommended for retraction in 2011 because of ethics violations. A study published in 2014 sought to determine whether they were retracted as recommended and, if so, whether their retraction conformed to COPE guidelines and other recommended practices. It judged them against most of the COPE criteria for retraction and two others, free availability of the retracted article and preservation of its original content, and found that only five retractions, all from the same journal, fulfilled every one of its predefined criteria.
A commentator on the study, though, found the result less clear-cut on a closer look. No fewer than 25 of the articles were deemed inadequately retracted on the ground that the PDF was not adequately marked; in 14 of those cases, inadequate marking meant the retracted article carried an opaque "RETRACTED ARTICLE" watermark rather than a transparent one, while 10 articles whose watermarks were almost invisible were judged adequately marked. In the commentator's view, COPE's advice that retracted status should be indicated as clearly as possible would seem to be better fulfilled by an opaque watermark than by a faint one.
Whether the content of a retracted paper should be preserved is disputed. The editor of one journal that had deleted the content of the retracted Boldt articles disagreed that retracted articles should be preserved, on the ground that their data were false and therefore valueless. The same commentator doubts that data obtained unethically are automatically false and valueless, pointing out that the Boldt case is not one of data fabrication. The commentator also suggests that one could argue the data should perhaps be deleted because they are not false and valueless, since people could choose to ignore the apparent ethics breach and use them anyway.
Weaknesses of current retraction procedures
The authors of the 2014 study of the Boldt retractions named three problems with current retraction procedures: uncertainty as to which forms of misconduct warrant retraction; lack of clarity about who is responsible for retraction; and no oversight of whether articles have been retracted at all, or in the correct way. They proposed solutions that clearly apportion responsibility for executing and monitoring retraction and protect editors from litigation. The commentator finds these ideas sensible but holds that they leave open a question that applies to many efforts to improve publication and post-publication processes: how should they be implemented?